4 * (C) Copyright IBM Corporation 2005.
5 * Released under GPL v2.
6 * Author : Ram Pai (linuxram@us.ibm.com)
9 #include <linux/mnt_namespace.h>
10 #include <linux/mount.h>
12 #include <linux/nsproxy.h>
16 /* return the next shared peer mount of @p */
17 static inline struct mount *next_peer(struct mount *p)
19 return list_entry(p->mnt_share.next, struct mount, mnt_share);
22 static inline struct mount *first_slave(struct mount *p)
24 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
27 static inline struct mount *next_slave(struct mount *p)
29 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
32 static struct mount *get_peer_under_root(struct mount *mnt,
33 struct mnt_namespace *ns,
34 const struct path *root)
36 struct mount *m = mnt;
39 /* Check the namespace first for optimization */
40 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
50 * Get ID of closest dominating peer group having a representative
51 * under the given root.
53 * Caller must hold namespace_sem
55 int get_dominating_id(struct mount *mnt, const struct path *root)
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);
62 return d->mnt_group_id;
68 static int do_make_slave(struct mount *mnt)
70 struct mount *peer_mnt = mnt, *master = mnt->mnt_master;
71 struct mount *slave_mnt;
74 * slave 'mnt' to a peer mount that has the
75 * same root dentry. If none is available then
76 * slave it to anything that is available.
78 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
79 peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
81 if (peer_mnt == mnt) {
82 peer_mnt = next_peer(mnt);
86 if (mnt->mnt_group_id && IS_MNT_SHARED(mnt) &&
87 list_empty(&mnt->mnt_share))
88 mnt_release_group_id(mnt);
90 list_del_init(&mnt->mnt_share);
91 mnt->mnt_group_id = 0;
97 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
98 slave_mnt->mnt_master = master;
99 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
100 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
101 INIT_LIST_HEAD(&mnt->mnt_slave_list);
103 struct list_head *p = &mnt->mnt_slave_list;
104 while (!list_empty(p)) {
105 slave_mnt = list_first_entry(p,
106 struct mount, mnt_slave);
107 list_del_init(&slave_mnt->mnt_slave);
108 slave_mnt->mnt_master = NULL;
111 mnt->mnt_master = master;
112 CLEAR_MNT_SHARED(mnt);
117 * vfsmount lock must be held for write
119 void change_mnt_propagation(struct mount *mnt, int type)
121 if (type == MS_SHARED) {
126 if (type != MS_SLAVE) {
127 list_del_init(&mnt->mnt_slave);
128 mnt->mnt_master = NULL;
129 if (type == MS_UNBINDABLE)
130 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
132 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
137 * get the next mount in the propagation tree.
138 * @m: the mount seen last
139 * @origin: the original mount from where the tree walk initiated
141 * Note that peer groups form contiguous segments of slave lists.
142 * We rely on that in get_source() to be able to find out if
143 * vfsmount found while iterating with propagation_next() is
144 * a peer of one we'd found earlier.
146 static struct mount *propagation_next(struct mount *m,
147 struct mount *origin)
149 /* are there any slaves of this mount? */
150 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
151 return first_slave(m);
154 struct mount *master = m->mnt_master;
156 if (master == origin->mnt_master) {
157 struct mount *next = next_peer(m);
158 return (next == origin) ? NULL : next;
159 } else if (m->mnt_slave.next != &master->mnt_slave_list)
160 return next_slave(m);
167 static struct mount *next_group(struct mount *m, struct mount *origin)
172 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
173 return first_slave(m);
175 if (m->mnt_group_id == origin->mnt_group_id) {
178 } else if (m->mnt_slave.next != &next->mnt_slave)
182 /* m is the last peer */
184 struct mount *master = m->mnt_master;
185 if (m->mnt_slave.next != &master->mnt_slave_list)
186 return next_slave(m);
187 m = next_peer(master);
188 if (master->mnt_group_id == origin->mnt_group_id)
190 if (master->mnt_slave.next == &m->mnt_slave)
199 /* all accesses are serialized by namespace_sem */
200 static struct user_namespace *user_ns;
201 static struct mount *last_dest, *first_source, *last_source, *dest_master;
202 static struct mountpoint *mp;
203 static struct hlist_head *list;
205 static inline bool peers(struct mount *m1, struct mount *m2)
207 return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
210 static int propagate_one(struct mount *m)
214 /* skip ones added by this propagate_mnt() */
217 /* skip if mountpoint isn't covered by it */
218 if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
220 if (peers(m, last_dest)) {
221 type = CL_MAKE_SHARED;
225 for (n = m; ; n = p) {
227 if (p == dest_master || IS_MNT_MARKED(p))
231 struct mount *parent = last_source->mnt_parent;
232 if (last_source == first_source)
234 done = parent->mnt_master == p;
235 if (done && peers(n, parent))
237 last_source = last_source->mnt_master;
241 /* beginning of peer group among the slaves? */
242 if (IS_MNT_SHARED(m))
243 type |= CL_MAKE_SHARED;
246 /* Notice when we are propagating across user namespaces */
247 if (m->mnt_ns->user_ns != user_ns)
248 type |= CL_UNPRIVILEGED;
249 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
251 return PTR_ERR(child);
252 child->mnt.mnt_flags &= ~MNT_LOCKED;
253 mnt_set_mountpoint(m, mp, child);
256 if (m->mnt_master != dest_master) {
257 read_seqlock_excl(&mount_lock);
258 SET_MNT_MARK(m->mnt_master);
259 read_sequnlock_excl(&mount_lock);
261 hlist_add_head(&child->mnt_hash, list);
266 * mount 'source_mnt' under the destination 'dest_mnt' at
267 * dentry 'dest_dentry'. And propagate that mount to
268 * all the peer and slave mounts of 'dest_mnt'.
269 * Link all the new mounts into a propagation tree headed at
270 * source_mnt. Also link all the new mounts using ->mnt_list
271 * headed at source_mnt's ->mnt_list
273 * @dest_mnt: destination mount.
274 * @dest_dentry: destination dentry.
275 * @source_mnt: source mount.
276 * @tree_list : list of heads of trees to be attached.
278 int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
279 struct mount *source_mnt, struct hlist_head *tree_list)
285 * we don't want to bother passing tons of arguments to
286 * propagate_one(); everything is serialized by namespace_sem,
287 * so globals will do just fine.
289 user_ns = current->nsproxy->mnt_ns->user_ns;
290 last_dest = dest_mnt;
291 first_source = source_mnt;
292 last_source = source_mnt;
295 dest_master = dest_mnt->mnt_master;
297 /* all peers of dest_mnt, except dest_mnt itself */
298 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
299 ret = propagate_one(n);
304 /* all slave groups */
305 for (m = next_group(dest_mnt, dest_mnt); m;
306 m = next_group(m, dest_mnt)) {
307 /* everything in that slave group */
310 ret = propagate_one(n);
317 read_seqlock_excl(&mount_lock);
318 hlist_for_each_entry(n, tree_list, mnt_hash) {
320 if (m->mnt_master != dest_mnt->mnt_master)
321 CLEAR_MNT_MARK(m->mnt_master);
323 read_sequnlock_excl(&mount_lock);
328 * return true if the refcount is greater than count
330 static inline int do_refcount_check(struct mount *mnt, int count)
332 return mnt_get_count(mnt) > count;
336 * check if the mount 'mnt' can be unmounted successfully.
337 * @mnt: the mount to be checked for unmount
338 * NOTE: unmounting 'mnt' would naturally propagate to all
339 * other mounts its parent propagates to.
340 * Check if any of these mounts that **do not have submounts**
341 * have more references than 'refcnt'. If so return busy.
343 * vfsmount lock must be held for write
345 int propagate_mount_busy(struct mount *mnt, int refcnt)
347 struct mount *m, *child;
348 struct mount *parent = mnt->mnt_parent;
352 return do_refcount_check(mnt, refcnt);
355 * quickly check if the current mount can be unmounted.
356 * If not, we don't have to go checking for all other
359 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
362 for (m = propagation_next(parent, parent); m;
363 m = propagation_next(m, parent)) {
364 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
365 if (child && list_empty(&child->mnt_mounts) &&
366 (ret = do_refcount_check(child, 1)))
373 * Clear MNT_LOCKED when it can be shown to be safe.
375 * mount_lock lock must be held for write
377 void propagate_mount_unlock(struct mount *mnt)
379 struct mount *parent = mnt->mnt_parent;
380 struct mount *m, *child;
382 BUG_ON(parent == mnt);
384 for (m = propagation_next(parent, parent); m;
385 m = propagation_next(m, parent)) {
386 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
388 child->mnt.mnt_flags &= ~MNT_LOCKED;
393 * Mark all mounts that the MNT_LOCKED logic will allow to be unmounted.
395 static void mark_umount_candidates(struct mount *mnt)
397 struct mount *parent = mnt->mnt_parent;
400 BUG_ON(parent == mnt);
402 for (m = propagation_next(parent, parent); m;
403 m = propagation_next(m, parent)) {
404 struct mount *child = __lookup_mnt_last(&m->mnt,
405 mnt->mnt_mountpoint);
406 if (child && (!IS_MNT_LOCKED(child) || IS_MNT_MARKED(m))) {
413 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
414 * parent propagates to.
416 static void __propagate_umount(struct mount *mnt)
418 struct mount *parent = mnt->mnt_parent;
421 BUG_ON(parent == mnt);
423 for (m = propagation_next(parent, parent); m;
424 m = propagation_next(m, parent)) {
426 struct mount *child = __lookup_mnt_last(&m->mnt,
427 mnt->mnt_mountpoint);
429 * umount the child only if the child has no children
430 * and the child is marked safe to unmount.
432 if (!child || !IS_MNT_MARKED(child))
434 CLEAR_MNT_MARK(child);
435 if (list_empty(&child->mnt_mounts)) {
436 list_del_init(&child->mnt_child);
437 child->mnt.mnt_flags |= MNT_UMOUNT;
438 list_move_tail(&child->mnt_list, &mnt->mnt_list);
444 * collect all mounts that receive propagation from the mount in @list,
445 * and return these additional mounts in the same list.
446 * @list: the list of mounts to be unmounted.
448 * vfsmount lock must be held for write
450 int propagate_umount(struct list_head *list)
454 list_for_each_entry_reverse(mnt, list, mnt_list)
455 mark_umount_candidates(mnt);
457 list_for_each_entry(mnt, list, mnt_list)
458 __propagate_umount(mnt);