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 *peer_mnt = mnt, *master = mnt->mnt_master;
71	struct mount *slave_mnt;
72
73	/*
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.
77	 */
78	while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
79	       peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
80
81	if (peer_mnt == mnt) {
82		peer_mnt = next_peer(mnt);
83		if (peer_mnt == mnt)
84			peer_mnt = NULL;
85	}
86	if (mnt->mnt_group_id && IS_MNT_SHARED(mnt) &&
87	    list_empty(&mnt->mnt_share))
88		mnt_release_group_id(mnt);
89
90	list_del_init(&mnt->mnt_share);
91	mnt->mnt_group_id = 0;
92
93	if (peer_mnt)
94		master = peer_mnt;
95
96	if (master) {
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);
102	} else {
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;
109		}
110	}
111	mnt->mnt_master = master;
112	CLEAR_MNT_SHARED(mnt);
113	return 0;
114}
115
116/*
117 * vfsmount lock must be held for write
118 */
119void change_mnt_propagation(struct mount *mnt, int type)
120{
121	if (type == MS_SHARED) {
122		set_mnt_shared(mnt);
123		return;
124	}
125	do_make_slave(mnt);
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;
131		else
132			mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
133	}
134}
135
136/*
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
140 *
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.
145 */
146static struct mount *propagation_next(struct mount *m,
147					 struct mount *origin)
148{
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);
152
153	while (1) {
154		struct mount *master = m->mnt_master;
155
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);
161
162		/* back at master */
163		m = master;
164	}
165}
166
167static struct mount *next_group(struct mount *m, struct mount *origin)
168{
169	while (1) {
170		while (1) {
171			struct mount *next;
172			if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
173				return first_slave(m);
174			next = next_peer(m);
175			if (m->mnt_group_id == origin->mnt_group_id) {
176				if (next == origin)
177					return NULL;
178			} else if (m->mnt_slave.next != &next->mnt_slave)
179				break;
180			m = next;
181		}
182		/* m is the last peer */
183		while (1) {
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)
189				break;
190			if (master->mnt_slave.next == &m->mnt_slave)
191				break;
192			m = master;
193		}
194		if (m == origin)
195			return NULL;
196	}
197}
198
199/* all accesses are serialized by namespace_sem */
200static struct user_namespace *user_ns;
201static struct mount *last_dest, *first_source, *last_source, *dest_master;
202static struct mountpoint *mp;
203static struct hlist_head *list;
204
205static inline bool peers(struct mount *m1, struct mount *m2)
206{
207	return m1->mnt_group_id == m2->mnt_group_id && m1->mnt_group_id;
208}
209
210static int propagate_one(struct mount *m)
211{
212	struct mount *child;
213	int type;
214	/* skip ones added by this propagate_mnt() */
215	if (IS_MNT_NEW(m))
216		return 0;
217	/* skip if mountpoint isn't covered by it */
218	if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
219		return 0;
220	if (peers(m, last_dest)) {
221		type = CL_MAKE_SHARED;
222	} else {
223		struct mount *n, *p;
224		bool done;
225		for (n = m; ; n = p) {
226			p = n->mnt_master;
227			if (p == dest_master || IS_MNT_MARKED(p))
228				break;
229		}
230		do {
231			struct mount *parent = last_source->mnt_parent;
232			if (last_source == first_source)
233				break;
234			done = parent->mnt_master == p;
235			if (done && peers(n, parent))
236				break;
237			last_source = last_source->mnt_master;
238		} while (!done);
239
240		type = CL_SLAVE;
241		/* beginning of peer group among the slaves? */
242		if (IS_MNT_SHARED(m))
243			type |= CL_MAKE_SHARED;
244	}
245
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);
250	if (IS_ERR(child))
251		return PTR_ERR(child);
252	child->mnt.mnt_flags &= ~MNT_LOCKED;
253	mnt_set_mountpoint(m, mp, child);
254	last_dest = m;
255	last_source = 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);
260	}
261	hlist_add_head(&child->mnt_hash, list);
262	return 0;
263}
264
265/*
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
272 *
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.
277 */
278int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
279		    struct mount *source_mnt, struct hlist_head *tree_list)
280{
281	struct mount *m, *n;
282	int ret = 0;
283
284	/*
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.
288	 */
289	user_ns = current->nsproxy->mnt_ns->user_ns;
290	last_dest = dest_mnt;
291	first_source = source_mnt;
292	last_source = source_mnt;
293	mp = dest_mp;
294	list = tree_list;
295	dest_master = dest_mnt->mnt_master;
296
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);
300		if (ret)
301			goto out;
302	}
303
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 */
308		n = m;
309		do {
310			ret = propagate_one(n);
311			if (ret)
312				goto out;
313			n = next_peer(n);
314		} while (n != m);
315	}
316out:
317	read_seqlock_excl(&mount_lock);
318	hlist_for_each_entry(n, tree_list, mnt_hash) {
319		m = n->mnt_parent;
320		if (m->mnt_master != dest_mnt->mnt_master)
321			CLEAR_MNT_MARK(m->mnt_master);
322	}
323	read_sequnlock_excl(&mount_lock);
324	return ret;
325}
326
327/*
328 * return true if the refcount is greater than count
329 */
330static inline int do_refcount_check(struct mount *mnt, int count)
331{
332	return mnt_get_count(mnt) > count;
333}
334
335/*
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.
342 *
343 * vfsmount lock must be held for write
344 */
345int propagate_mount_busy(struct mount *mnt, int refcnt)
346{
347	struct mount *m, *child;
348	struct mount *parent = mnt->mnt_parent;
349	int ret = 0;
350
351	if (mnt == parent)
352		return do_refcount_check(mnt, refcnt);
353
354	/*
355	 * quickly check if the current mount can be unmounted.
356	 * If not, we don't have to go checking for all other
357	 * mounts
358	 */
359	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
360		return 1;
361
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)))
367			break;
368	}
369	return ret;
370}
371
372/*
373 * Clear MNT_LOCKED when it can be shown to be safe.
374 *
375 * mount_lock lock must be held for write
376 */
377void propagate_mount_unlock(struct mount *mnt)
378{
379	struct mount *parent = mnt->mnt_parent;
380	struct mount *m, *child;
381
382	BUG_ON(parent == mnt);
383
384	for (m = propagation_next(parent, parent); m;
385			m = propagation_next(m, parent)) {
386		child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
387		if (child)
388			child->mnt.mnt_flags &= ~MNT_LOCKED;
389	}
390}
391
392/*
393 * Mark all mounts that the MNT_LOCKED logic will allow to be unmounted.
394 */
395static void mark_umount_candidates(struct mount *mnt)
396{
397	struct mount *parent = mnt->mnt_parent;
398	struct mount *m;
399
400	BUG_ON(parent == mnt);
401
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))) {
407			SET_MNT_MARK(child);
408		}
409	}
410}
411
412/*
413 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
414 * parent propagates to.
415 */
416static void __propagate_umount(struct mount *mnt)
417{
418	struct mount *parent = mnt->mnt_parent;
419	struct mount *m;
420
421	BUG_ON(parent == mnt);
422
423	for (m = propagation_next(parent, parent); m;
424			m = propagation_next(m, parent)) {
425
426		struct mount *child = __lookup_mnt_last(&m->mnt,
427						mnt->mnt_mountpoint);
428		/*
429		 * umount the child only if the child has no children
430		 * and the child is marked safe to unmount.
431		 */
432		if (!child || !IS_MNT_MARKED(child))
433			continue;
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);
439		}
440	}
441}
442
443/*
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.
447 *
448 * vfsmount lock must be held for write
449 */
450int propagate_umount(struct list_head *list)
451{
452	struct mount *mnt;
453
454	list_for_each_entry_reverse(mnt, list, mnt_list)
455		mark_umount_candidates(mnt);
456
457	list_for_each_entry(mnt, list, mnt_list)
458		__propagate_umount(mnt);
459	return 0;
460}
461