Loading...
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 "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 *next_slave(struct mount *p)
28{
29 return list_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 int do_make_slave(struct mount *mnt)
69{
70 struct mount *master, *slave_mnt;
71
72 if (list_empty(&mnt->mnt_share)) {
73 if (IS_MNT_SHARED(mnt)) {
74 mnt_release_group_id(mnt);
75 CLEAR_MNT_SHARED(mnt);
76 }
77 master = mnt->mnt_master;
78 if (!master) {
79 struct list_head *p = &mnt->mnt_slave_list;
80 while (!list_empty(p)) {
81 slave_mnt = list_first_entry(p,
82 struct mount, mnt_slave);
83 list_del_init(&slave_mnt->mnt_slave);
84 slave_mnt->mnt_master = NULL;
85 }
86 return 0;
87 }
88 } else {
89 struct mount *m;
90 /*
91 * slave 'mnt' to a peer mount that has the
92 * same root dentry. If none is available then
93 * slave it to anything that is available.
94 */
95 for (m = master = next_peer(mnt); m != mnt; m = next_peer(m)) {
96 if (m->mnt.mnt_root == mnt->mnt.mnt_root) {
97 master = m;
98 break;
99 }
100 }
101 list_del_init(&mnt->mnt_share);
102 mnt->mnt_group_id = 0;
103 CLEAR_MNT_SHARED(mnt);
104 }
105 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
106 slave_mnt->mnt_master = master;
107 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
108 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
109 INIT_LIST_HEAD(&mnt->mnt_slave_list);
110 mnt->mnt_master = master;
111 return 0;
112}
113
114/*
115 * vfsmount lock must be held for write
116 */
117void change_mnt_propagation(struct mount *mnt, int type)
118{
119 if (type == MS_SHARED) {
120 set_mnt_shared(mnt);
121 return;
122 }
123 do_make_slave(mnt);
124 if (type != MS_SLAVE) {
125 list_del_init(&mnt->mnt_slave);
126 mnt->mnt_master = NULL;
127 if (type == MS_UNBINDABLE)
128 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
129 else
130 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
131 }
132}
133
134/*
135 * get the next mount in the propagation tree.
136 * @m: the mount seen last
137 * @origin: the original mount from where the tree walk initiated
138 *
139 * Note that peer groups form contiguous segments of slave lists.
140 * We rely on that in get_source() to be able to find out if
141 * vfsmount found while iterating with propagation_next() is
142 * a peer of one we'd found earlier.
143 */
144static struct mount *propagation_next(struct mount *m,
145 struct mount *origin)
146{
147 /* are there any slaves of this mount? */
148 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
149 return first_slave(m);
150
151 while (1) {
152 struct mount *master = m->mnt_master;
153
154 if (master == origin->mnt_master) {
155 struct mount *next = next_peer(m);
156 return (next == origin) ? NULL : next;
157 } else if (m->mnt_slave.next != &master->mnt_slave_list)
158 return next_slave(m);
159
160 /* back at master */
161 m = master;
162 }
163}
164
165static struct mount *next_group(struct mount *m, struct mount *origin)
166{
167 while (1) {
168 while (1) {
169 struct mount *next;
170 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
171 return first_slave(m);
172 next = next_peer(m);
173 if (m->mnt_group_id == origin->mnt_group_id) {
174 if (next == origin)
175 return NULL;
176 } else if (m->mnt_slave.next != &next->mnt_slave)
177 break;
178 m = next;
179 }
180 /* m is the last peer */
181 while (1) {
182 struct mount *master = m->mnt_master;
183 if (m->mnt_slave.next != &master->mnt_slave_list)
184 return next_slave(m);
185 m = next_peer(master);
186 if (master->mnt_group_id == origin->mnt_group_id)
187 break;
188 if (master->mnt_slave.next == &m->mnt_slave)
189 break;
190 m = master;
191 }
192 if (m == origin)
193 return NULL;
194 }
195}
196
197/* all accesses are serialized by namespace_sem */
198static struct user_namespace *user_ns;
199static struct mount *last_dest, *first_source, *last_source, *dest_master;
200static struct mountpoint *mp;
201static struct hlist_head *list;
202
203static inline bool peers(struct mount *m1, struct mount *m2)
204{
205 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
206}
207
208static int propagate_one(struct mount *m)
209{
210 struct mount *child;
211 int type;
212 /* skip ones added by this propagate_mnt() */
213 if (IS_MNT_NEW(m))
214 return 0;
215 /* skip if mountpoint isn't covered by it */
216 if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
217 return 0;
218 if (peers(m, last_dest)) {
219 type = CL_MAKE_SHARED;
220 } else {
221 struct mount *n, *p;
222 bool done;
223 for (n = m; ; n = p) {
224 p = n->mnt_master;
225 if (p == dest_master || IS_MNT_MARKED(p))
226 break;
227 }
228 do {
229 struct mount *parent = last_source->mnt_parent;
230 if (last_source == first_source)
231 break;
232 done = parent->mnt_master == p;
233 if (done && peers(n, parent))
234 break;
235 last_source = last_source->mnt_master;
236 } while (!done);
237
238 type = CL_SLAVE;
239 /* beginning of peer group among the slaves? */
240 if (IS_MNT_SHARED(m))
241 type |= CL_MAKE_SHARED;
242 }
243
244 /* Notice when we are propagating across user namespaces */
245 if (m->mnt_ns->user_ns != user_ns)
246 type |= CL_UNPRIVILEGED;
247 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
248 if (IS_ERR(child))
249 return PTR_ERR(child);
250 child->mnt.mnt_flags &= ~MNT_LOCKED;
251 mnt_set_mountpoint(m, mp, child);
252 last_dest = m;
253 last_source = child;
254 if (m->mnt_master != dest_master) {
255 read_seqlock_excl(&mount_lock);
256 SET_MNT_MARK(m->mnt_master);
257 read_sequnlock_excl(&mount_lock);
258 }
259 hlist_add_head(&child->mnt_hash, list);
260 return count_mounts(m->mnt_ns, child);
261}
262
263/*
264 * mount 'source_mnt' under the destination 'dest_mnt' at
265 * dentry 'dest_dentry'. And propagate that mount to
266 * all the peer and slave mounts of 'dest_mnt'.
267 * Link all the new mounts into a propagation tree headed at
268 * source_mnt. Also link all the new mounts using ->mnt_list
269 * headed at source_mnt's ->mnt_list
270 *
271 * @dest_mnt: destination mount.
272 * @dest_dentry: destination dentry.
273 * @source_mnt: source mount.
274 * @tree_list : list of heads of trees to be attached.
275 */
276int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
277 struct mount *source_mnt, struct hlist_head *tree_list)
278{
279 struct mount *m, *n;
280 int ret = 0;
281
282 /*
283 * we don't want to bother passing tons of arguments to
284 * propagate_one(); everything is serialized by namespace_sem,
285 * so globals will do just fine.
286 */
287 user_ns = current->nsproxy->mnt_ns->user_ns;
288 last_dest = dest_mnt;
289 first_source = source_mnt;
290 last_source = source_mnt;
291 mp = dest_mp;
292 list = tree_list;
293 dest_master = dest_mnt->mnt_master;
294
295 /* all peers of dest_mnt, except dest_mnt itself */
296 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
297 ret = propagate_one(n);
298 if (ret)
299 goto out;
300 }
301
302 /* all slave groups */
303 for (m = next_group(dest_mnt, dest_mnt); m;
304 m = next_group(m, dest_mnt)) {
305 /* everything in that slave group */
306 n = m;
307 do {
308 ret = propagate_one(n);
309 if (ret)
310 goto out;
311 n = next_peer(n);
312 } while (n != m);
313 }
314out:
315 read_seqlock_excl(&mount_lock);
316 hlist_for_each_entry(n, tree_list, mnt_hash) {
317 m = n->mnt_parent;
318 if (m->mnt_master != dest_mnt->mnt_master)
319 CLEAR_MNT_MARK(m->mnt_master);
320 }
321 read_sequnlock_excl(&mount_lock);
322 return ret;
323}
324
325static struct mount *find_topper(struct mount *mnt)
326{
327 /* If there is exactly one mount covering mnt completely return it. */
328 struct mount *child;
329
330 if (!list_is_singular(&mnt->mnt_mounts))
331 return NULL;
332
333 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
334 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
335 return NULL;
336
337 return child;
338}
339
340/*
341 * return true if the refcount is greater than count
342 */
343static inline int do_refcount_check(struct mount *mnt, int count)
344{
345 return mnt_get_count(mnt) > count;
346}
347
348/*
349 * check if the mount 'mnt' can be unmounted successfully.
350 * @mnt: the mount to be checked for unmount
351 * NOTE: unmounting 'mnt' would naturally propagate to all
352 * other mounts its parent propagates to.
353 * Check if any of these mounts that **do not have submounts**
354 * have more references than 'refcnt'. If so return busy.
355 *
356 * vfsmount lock must be held for write
357 */
358int propagate_mount_busy(struct mount *mnt, int refcnt)
359{
360 struct mount *m, *child, *topper;
361 struct mount *parent = mnt->mnt_parent;
362
363 if (mnt == parent)
364 return do_refcount_check(mnt, refcnt);
365
366 /*
367 * quickly check if the current mount can be unmounted.
368 * If not, we don't have to go checking for all other
369 * mounts
370 */
371 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
372 return 1;
373
374 for (m = propagation_next(parent, parent); m;
375 m = propagation_next(m, parent)) {
376 int count = 1;
377 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
378 if (!child)
379 continue;
380
381 /* Is there exactly one mount on the child that covers
382 * it completely whose reference should be ignored?
383 */
384 topper = find_topper(child);
385 if (topper)
386 count += 1;
387 else if (!list_empty(&child->mnt_mounts))
388 continue;
389
390 if (do_refcount_check(child, count))
391 return 1;
392 }
393 return 0;
394}
395
396/*
397 * Clear MNT_LOCKED when it can be shown to be safe.
398 *
399 * mount_lock lock must be held for write
400 */
401void propagate_mount_unlock(struct mount *mnt)
402{
403 struct mount *parent = mnt->mnt_parent;
404 struct mount *m, *child;
405
406 BUG_ON(parent == mnt);
407
408 for (m = propagation_next(parent, parent); m;
409 m = propagation_next(m, parent)) {
410 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
411 if (child)
412 child->mnt.mnt_flags &= ~MNT_LOCKED;
413 }
414}
415
416/*
417 * Mark all mounts that the MNT_LOCKED logic will allow to be unmounted.
418 */
419static void mark_umount_candidates(struct mount *mnt)
420{
421 struct mount *parent = mnt->mnt_parent;
422 struct mount *m;
423
424 BUG_ON(parent == mnt);
425
426 for (m = propagation_next(parent, parent); m;
427 m = propagation_next(m, parent)) {
428 struct mount *child = __lookup_mnt(&m->mnt,
429 mnt->mnt_mountpoint);
430 if (!child || (child->mnt.mnt_flags & MNT_UMOUNT))
431 continue;
432 if (!IS_MNT_LOCKED(child) || IS_MNT_MARKED(m)) {
433 SET_MNT_MARK(child);
434 }
435 }
436}
437
438/*
439 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
440 * parent propagates to.
441 */
442static void __propagate_umount(struct mount *mnt)
443{
444 struct mount *parent = mnt->mnt_parent;
445 struct mount *m;
446
447 BUG_ON(parent == mnt);
448
449 for (m = propagation_next(parent, parent); m;
450 m = propagation_next(m, parent)) {
451 struct mount *topper;
452 struct mount *child = __lookup_mnt(&m->mnt,
453 mnt->mnt_mountpoint);
454 /*
455 * umount the child only if the child has no children
456 * and the child is marked safe to unmount.
457 */
458 if (!child || !IS_MNT_MARKED(child))
459 continue;
460 CLEAR_MNT_MARK(child);
461
462 /* If there is exactly one mount covering all of child
463 * replace child with that mount.
464 */
465 topper = find_topper(child);
466 if (topper)
467 mnt_change_mountpoint(child->mnt_parent, child->mnt_mp,
468 topper);
469
470 if (list_empty(&child->mnt_mounts)) {
471 list_del_init(&child->mnt_child);
472 child->mnt.mnt_flags |= MNT_UMOUNT;
473 list_move_tail(&child->mnt_list, &mnt->mnt_list);
474 }
475 }
476}
477
478/*
479 * collect all mounts that receive propagation from the mount in @list,
480 * and return these additional mounts in the same list.
481 * @list: the list of mounts to be unmounted.
482 *
483 * vfsmount lock must be held for write
484 */
485int propagate_umount(struct list_head *list)
486{
487 struct mount *mnt;
488
489 list_for_each_entry_reverse(mnt, list, mnt_list)
490 mark_umount_candidates(mnt);
491
492 list_for_each_entry(mnt, list, mnt_list)
493 __propagate_umount(mnt);
494 return 0;
495}
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 mnt_set_mountpoint(m, mp, child);
265 last_dest = m;
266 last_source = child;
267 if (m->mnt_master != dest_master) {
268 read_seqlock_excl(&mount_lock);
269 SET_MNT_MARK(m->mnt_master);
270 read_sequnlock_excl(&mount_lock);
271 }
272 hlist_add_head(&child->mnt_hash, list);
273 return count_mounts(m->mnt_ns, child);
274}
275
276/*
277 * mount 'source_mnt' under the destination 'dest_mnt' at
278 * dentry 'dest_dentry'. And propagate that mount to
279 * all the peer and slave mounts of 'dest_mnt'.
280 * Link all the new mounts into a propagation tree headed at
281 * source_mnt. Also link all the new mounts using ->mnt_list
282 * headed at source_mnt's ->mnt_list
283 *
284 * @dest_mnt: destination mount.
285 * @dest_dentry: destination dentry.
286 * @source_mnt: source mount.
287 * @tree_list : list of heads of trees to be attached.
288 */
289int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
290 struct mount *source_mnt, struct hlist_head *tree_list)
291{
292 struct mount *m, *n;
293 int ret = 0;
294
295 /*
296 * we don't want to bother passing tons of arguments to
297 * propagate_one(); everything is serialized by namespace_sem,
298 * so globals will do just fine.
299 */
300 last_dest = dest_mnt;
301 first_source = source_mnt;
302 last_source = source_mnt;
303 mp = dest_mp;
304 list = tree_list;
305 dest_master = dest_mnt->mnt_master;
306
307 /* all peers of dest_mnt, except dest_mnt itself */
308 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
309 ret = propagate_one(n);
310 if (ret)
311 goto out;
312 }
313
314 /* all slave groups */
315 for (m = next_group(dest_mnt, dest_mnt); m;
316 m = next_group(m, dest_mnt)) {
317 /* everything in that slave group */
318 n = m;
319 do {
320 ret = propagate_one(n);
321 if (ret)
322 goto out;
323 n = next_peer(n);
324 } while (n != m);
325 }
326out:
327 read_seqlock_excl(&mount_lock);
328 hlist_for_each_entry(n, tree_list, mnt_hash) {
329 m = n->mnt_parent;
330 if (m->mnt_master != dest_mnt->mnt_master)
331 CLEAR_MNT_MARK(m->mnt_master);
332 }
333 read_sequnlock_excl(&mount_lock);
334 return ret;
335}
336
337static struct mount *find_topper(struct mount *mnt)
338{
339 /* If there is exactly one mount covering mnt completely return it. */
340 struct mount *child;
341
342 if (!list_is_singular(&mnt->mnt_mounts))
343 return NULL;
344
345 child = list_first_entry(&mnt->mnt_mounts, struct mount, mnt_child);
346 if (child->mnt_mountpoint != mnt->mnt.mnt_root)
347 return NULL;
348
349 return child;
350}
351
352/*
353 * return true if the refcount is greater than count
354 */
355static inline int do_refcount_check(struct mount *mnt, int count)
356{
357 return mnt_get_count(mnt) > count;
358}
359
360/*
361 * check if the mount 'mnt' can be unmounted successfully.
362 * @mnt: the mount to be checked for unmount
363 * NOTE: unmounting 'mnt' would naturally propagate to all
364 * other mounts its parent propagates to.
365 * Check if any of these mounts that **do not have submounts**
366 * have more references than 'refcnt'. If so return busy.
367 *
368 * vfsmount lock must be held for write
369 */
370int propagate_mount_busy(struct mount *mnt, int refcnt)
371{
372 struct mount *m, *child, *topper;
373 struct mount *parent = mnt->mnt_parent;
374
375 if (mnt == parent)
376 return do_refcount_check(mnt, refcnt);
377
378 /*
379 * quickly check if the current mount can be unmounted.
380 * If not, we don't have to go checking for all other
381 * mounts
382 */
383 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
384 return 1;
385
386 for (m = propagation_next(parent, parent); m;
387 m = propagation_next(m, parent)) {
388 int count = 1;
389 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
390 if (!child)
391 continue;
392
393 /* Is there exactly one mount on the child that covers
394 * it completely whose reference should be ignored?
395 */
396 topper = find_topper(child);
397 if (topper)
398 count += 1;
399 else if (!list_empty(&child->mnt_mounts))
400 continue;
401
402 if (do_refcount_check(child, count))
403 return 1;
404 }
405 return 0;
406}
407
408/*
409 * Clear MNT_LOCKED when it can be shown to be safe.
410 *
411 * mount_lock lock must be held for write
412 */
413void propagate_mount_unlock(struct mount *mnt)
414{
415 struct mount *parent = mnt->mnt_parent;
416 struct mount *m, *child;
417
418 BUG_ON(parent == mnt);
419
420 for (m = propagation_next(parent, parent); m;
421 m = propagation_next(m, parent)) {
422 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint);
423 if (child)
424 child->mnt.mnt_flags &= ~MNT_LOCKED;
425 }
426}
427
428static void umount_one(struct mount *mnt, struct list_head *to_umount)
429{
430 CLEAR_MNT_MARK(mnt);
431 mnt->mnt.mnt_flags |= MNT_UMOUNT;
432 list_del_init(&mnt->mnt_child);
433 list_del_init(&mnt->mnt_umounting);
434 list_move_tail(&mnt->mnt_list, to_umount);
435}
436
437/*
438 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
439 * parent propagates to.
440 */
441static bool __propagate_umount(struct mount *mnt,
442 struct list_head *to_umount,
443 struct list_head *to_restore)
444{
445 bool progress = false;
446 struct mount *child;
447
448 /*
449 * The state of the parent won't change if this mount is
450 * already unmounted or marked as without children.
451 */
452 if (mnt->mnt.mnt_flags & (MNT_UMOUNT | MNT_MARKED))
453 goto out;
454
455 /* Verify topper is the only grandchild that has not been
456 * speculatively unmounted.
457 */
458 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
459 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
460 continue;
461 if (!list_empty(&child->mnt_umounting) && IS_MNT_MARKED(child))
462 continue;
463 /* Found a mounted child */
464 goto children;
465 }
466
467 /* Mark mounts that can be unmounted if not locked */
468 SET_MNT_MARK(mnt);
469 progress = true;
470
471 /* If a mount is without children and not locked umount it. */
472 if (!IS_MNT_LOCKED(mnt)) {
473 umount_one(mnt, to_umount);
474 } else {
475children:
476 list_move_tail(&mnt->mnt_umounting, to_restore);
477 }
478out:
479 return progress;
480}
481
482static void umount_list(struct list_head *to_umount,
483 struct list_head *to_restore)
484{
485 struct mount *mnt, *child, *tmp;
486 list_for_each_entry(mnt, to_umount, mnt_list) {
487 list_for_each_entry_safe(child, tmp, &mnt->mnt_mounts, mnt_child) {
488 /* topper? */
489 if (child->mnt_mountpoint == mnt->mnt.mnt_root)
490 list_move_tail(&child->mnt_umounting, to_restore);
491 else
492 umount_one(child, to_umount);
493 }
494 }
495}
496
497static void restore_mounts(struct list_head *to_restore)
498{
499 /* Restore mounts to a clean working state */
500 while (!list_empty(to_restore)) {
501 struct mount *mnt, *parent;
502 struct mountpoint *mp;
503
504 mnt = list_first_entry(to_restore, struct mount, mnt_umounting);
505 CLEAR_MNT_MARK(mnt);
506 list_del_init(&mnt->mnt_umounting);
507
508 /* Should this mount be reparented? */
509 mp = mnt->mnt_mp;
510 parent = mnt->mnt_parent;
511 while (parent->mnt.mnt_flags & MNT_UMOUNT) {
512 mp = parent->mnt_mp;
513 parent = parent->mnt_parent;
514 }
515 if (parent != mnt->mnt_parent)
516 mnt_change_mountpoint(parent, mp, mnt);
517 }
518}
519
520static void cleanup_umount_visitations(struct list_head *visited)
521{
522 while (!list_empty(visited)) {
523 struct mount *mnt =
524 list_first_entry(visited, struct mount, mnt_umounting);
525 list_del_init(&mnt->mnt_umounting);
526 }
527}
528
529/*
530 * collect all mounts that receive propagation from the mount in @list,
531 * and return these additional mounts in the same list.
532 * @list: the list of mounts to be unmounted.
533 *
534 * vfsmount lock must be held for write
535 */
536int propagate_umount(struct list_head *list)
537{
538 struct mount *mnt;
539 LIST_HEAD(to_restore);
540 LIST_HEAD(to_umount);
541 LIST_HEAD(visited);
542
543 /* Find candidates for unmounting */
544 list_for_each_entry_reverse(mnt, list, mnt_list) {
545 struct mount *parent = mnt->mnt_parent;
546 struct mount *m;
547
548 /*
549 * If this mount has already been visited it is known that it's
550 * entire peer group and all of their slaves in the propagation
551 * tree for the mountpoint has already been visited and there is
552 * no need to visit them again.
553 */
554 if (!list_empty(&mnt->mnt_umounting))
555 continue;
556
557 list_add_tail(&mnt->mnt_umounting, &visited);
558 for (m = propagation_next(parent, parent); m;
559 m = propagation_next(m, parent)) {
560 struct mount *child = __lookup_mnt(&m->mnt,
561 mnt->mnt_mountpoint);
562 if (!child)
563 continue;
564
565 if (!list_empty(&child->mnt_umounting)) {
566 /*
567 * If the child has already been visited it is
568 * know that it's entire peer group and all of
569 * their slaves in the propgation tree for the
570 * mountpoint has already been visited and there
571 * is no need to visit this subtree again.
572 */
573 m = skip_propagation_subtree(m, parent);
574 continue;
575 } else if (child->mnt.mnt_flags & MNT_UMOUNT) {
576 /*
577 * We have come accross an partially unmounted
578 * mount in list that has not been visited yet.
579 * Remember it has been visited and continue
580 * about our merry way.
581 */
582 list_add_tail(&child->mnt_umounting, &visited);
583 continue;
584 }
585
586 /* Check the child and parents while progress is made */
587 while (__propagate_umount(child,
588 &to_umount, &to_restore)) {
589 /* Is the parent a umount candidate? */
590 child = child->mnt_parent;
591 if (list_empty(&child->mnt_umounting))
592 break;
593 }
594 }
595 }
596
597 umount_list(&to_umount, &to_restore);
598 restore_mounts(&to_restore);
599 cleanup_umount_visitations(&visited);
600 list_splice_tail(&to_umount, list);
601
602 return 0;
603}