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 "internal.h"
13#include "pnode.h"
14
15/* return the next shared peer mount of @p */
16static inline struct vfsmount *next_peer(struct vfsmount *p)
17{
18	return list_entry(p->mnt_share.next, struct vfsmount, mnt_share);
19}
20
21static inline struct vfsmount *first_slave(struct vfsmount *p)
22{
23	return list_entry(p->mnt_slave_list.next, struct vfsmount, mnt_slave);
24}
25
26static inline struct vfsmount *next_slave(struct vfsmount *p)
27{
28	return list_entry(p->mnt_slave.next, struct vfsmount, mnt_slave);
29}
30
31/*
32 * Return true if path is reachable from root
33 *
34 * namespace_sem is held, and mnt is attached
35 */
36static bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
37			 const struct path *root)
38{
39	while (mnt != root->mnt && mnt->mnt_parent != mnt) {
40		dentry = mnt->mnt_mountpoint;
41		mnt = mnt->mnt_parent;
42	}
43	return mnt == root->mnt && is_subdir(dentry, root->dentry);
44}
45
46static struct vfsmount *get_peer_under_root(struct vfsmount *mnt,
47					    struct mnt_namespace *ns,
48					    const struct path *root)
49{
50	struct vfsmount *m = mnt;
51
52	do {
53		/* Check the namespace first for optimization */
54		if (m->mnt_ns == ns && is_path_reachable(m, m->mnt_root, root))
55			return m;
56
57		m = next_peer(m);
58	} while (m != mnt);
59
60	return NULL;
61}
62
63/*
64 * Get ID of closest dominating peer group having a representative
65 * under the given root.
66 *
67 * Caller must hold namespace_sem
68 */
69int get_dominating_id(struct vfsmount *mnt, const struct path *root)
70{
71	struct vfsmount *m;
72
73	for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
74		struct vfsmount *d = get_peer_under_root(m, mnt->mnt_ns, root);
75		if (d)
76			return d->mnt_group_id;
77	}
78
79	return 0;
80}
81
82static int do_make_slave(struct vfsmount *mnt)
83{
84	struct vfsmount *peer_mnt = mnt, *master = mnt->mnt_master;
85	struct vfsmount *slave_mnt;
86
87	/*
88	 * slave 'mnt' to a peer mount that has the
89	 * same root dentry. If none is available then
90	 * slave it to anything that is available.
91	 */
92	while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
93	       peer_mnt->mnt_root != mnt->mnt_root) ;
94
95	if (peer_mnt == mnt) {
96		peer_mnt = next_peer(mnt);
97		if (peer_mnt == mnt)
98			peer_mnt = NULL;
99	}
100	if (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
101		mnt_release_group_id(mnt);
102
103	list_del_init(&mnt->mnt_share);
104	mnt->mnt_group_id = 0;
105
106	if (peer_mnt)
107		master = peer_mnt;
108
109	if (master) {
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	} else {
116		struct list_head *p = &mnt->mnt_slave_list;
117		while (!list_empty(p)) {
118                        slave_mnt = list_first_entry(p,
119					struct vfsmount, mnt_slave);
120			list_del_init(&slave_mnt->mnt_slave);
121			slave_mnt->mnt_master = NULL;
122		}
123	}
124	mnt->mnt_master = master;
125	CLEAR_MNT_SHARED(mnt);
126	return 0;
127}
128
129/*
130 * vfsmount lock must be held for write
131 */
132void change_mnt_propagation(struct vfsmount *mnt, int type)
133{
134	if (type == MS_SHARED) {
135		set_mnt_shared(mnt);
136		return;
137	}
138	do_make_slave(mnt);
139	if (type != MS_SLAVE) {
140		list_del_init(&mnt->mnt_slave);
141		mnt->mnt_master = NULL;
142		if (type == MS_UNBINDABLE)
143			mnt->mnt_flags |= MNT_UNBINDABLE;
144		else
145			mnt->mnt_flags &= ~MNT_UNBINDABLE;
146	}
147}
148
149/*
150 * get the next mount in the propagation tree.
151 * @m: the mount seen last
152 * @origin: the original mount from where the tree walk initiated
153 *
154 * Note that peer groups form contiguous segments of slave lists.
155 * We rely on that in get_source() to be able to find out if
156 * vfsmount found while iterating with propagation_next() is
157 * a peer of one we'd found earlier.
158 */
159static struct vfsmount *propagation_next(struct vfsmount *m,
160					 struct vfsmount *origin)
161{
162	/* are there any slaves of this mount? */
163	if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
164		return first_slave(m);
165
166	while (1) {
167		struct vfsmount *next;
168		struct vfsmount *master = m->mnt_master;
169
170		if (master == origin->mnt_master) {
171			next = next_peer(m);
172			return ((next == origin) ? NULL : next);
173		} else if (m->mnt_slave.next != &master->mnt_slave_list)
174			return next_slave(m);
175
176		/* back at master */
177		m = master;
178	}
179}
180
181/*
182 * return the source mount to be used for cloning
183 *
184 * @dest 	the current destination mount
185 * @last_dest  	the last seen destination mount
186 * @last_src  	the last seen source mount
187 * @type	return CL_SLAVE if the new mount has to be
188 * 		cloned as a slave.
189 */
190static struct vfsmount *get_source(struct vfsmount *dest,
191					struct vfsmount *last_dest,
192					struct vfsmount *last_src,
193					int *type)
194{
195	struct vfsmount *p_last_src = NULL;
196	struct vfsmount *p_last_dest = NULL;
197
198	while (last_dest != dest->mnt_master) {
199		p_last_dest = last_dest;
200		p_last_src = last_src;
201		last_dest = last_dest->mnt_master;
202		last_src = last_src->mnt_master;
203	}
204
205	if (p_last_dest) {
206		do {
207			p_last_dest = next_peer(p_last_dest);
208		} while (IS_MNT_NEW(p_last_dest));
209		/* is that a peer of the earlier? */
210		if (dest == p_last_dest) {
211			*type = CL_MAKE_SHARED;
212			return p_last_src;
213		}
214	}
215	/* slave of the earlier, then */
216	*type = CL_SLAVE;
217	/* beginning of peer group among the slaves? */
218	if (IS_MNT_SHARED(dest))
219		*type |= CL_MAKE_SHARED;
220	return last_src;
221}
222
223/*
224 * mount 'source_mnt' under the destination 'dest_mnt' at
225 * dentry 'dest_dentry'. And propagate that mount to
226 * all the peer and slave mounts of 'dest_mnt'.
227 * Link all the new mounts into a propagation tree headed at
228 * source_mnt. Also link all the new mounts using ->mnt_list
229 * headed at source_mnt's ->mnt_list
230 *
231 * @dest_mnt: destination mount.
232 * @dest_dentry: destination dentry.
233 * @source_mnt: source mount.
234 * @tree_list : list of heads of trees to be attached.
235 */
236int propagate_mnt(struct vfsmount *dest_mnt, struct dentry *dest_dentry,
237		    struct vfsmount *source_mnt, struct list_head *tree_list)
238{
239	struct vfsmount *m, *child;
240	int ret = 0;
241	struct vfsmount *prev_dest_mnt = dest_mnt;
242	struct vfsmount *prev_src_mnt  = source_mnt;
243	LIST_HEAD(tmp_list);
244	LIST_HEAD(umount_list);
245
246	for (m = propagation_next(dest_mnt, dest_mnt); m;
247			m = propagation_next(m, dest_mnt)) {
248		int type;
249		struct vfsmount *source;
250
251		if (IS_MNT_NEW(m))
252			continue;
253
254		source =  get_source(m, prev_dest_mnt, prev_src_mnt, &type);
255
256		if (!(child = copy_tree(source, source->mnt_root, type))) {
257			ret = -ENOMEM;
258			list_splice(tree_list, tmp_list.prev);
259			goto out;
260		}
261
262		if (is_subdir(dest_dentry, m->mnt_root)) {
263			mnt_set_mountpoint(m, dest_dentry, child);
264			list_add_tail(&child->mnt_hash, tree_list);
265		} else {
266			/*
267			 * This can happen if the parent mount was bind mounted
268			 * on some subdirectory of a shared/slave mount.
269			 */
270			list_add_tail(&child->mnt_hash, &tmp_list);
271		}
272		prev_dest_mnt = m;
273		prev_src_mnt  = child;
274	}
275out:
276	br_write_lock(vfsmount_lock);
277	while (!list_empty(&tmp_list)) {
278		child = list_first_entry(&tmp_list, struct vfsmount, mnt_hash);
279		umount_tree(child, 0, &umount_list);
280	}
281	br_write_unlock(vfsmount_lock);
282	release_mounts(&umount_list);
283	return ret;
284}
285
286/*
287 * return true if the refcount is greater than count
288 */
289static inline int do_refcount_check(struct vfsmount *mnt, int count)
290{
291	int mycount = atomic_read(&mnt->mnt_count) - mnt->mnt_ghosts;
292	return (mycount > count);
293}
294
295/*
296 * check if the mount 'mnt' can be unmounted successfully.
297 * @mnt: the mount to be checked for unmount
298 * NOTE: unmounting 'mnt' would naturally propagate to all
299 * other mounts its parent propagates to.
300 * Check if any of these mounts that **do not have submounts**
301 * have more references than 'refcnt'. If so return busy.
302 *
303 * vfsmount lock must be held for read or write
304 */
305int propagate_mount_busy(struct vfsmount *mnt, int refcnt)
306{
307	struct vfsmount *m, *child;
308	struct vfsmount *parent = mnt->mnt_parent;
309	int ret = 0;
310
311	if (mnt == parent)
312		return do_refcount_check(mnt, refcnt);
313
314	/*
315	 * quickly check if the current mount can be unmounted.
316	 * If not, we don't have to go checking for all other
317	 * mounts
318	 */
319	if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
320		return 1;
321
322	for (m = propagation_next(parent, parent); m;
323	     		m = propagation_next(m, parent)) {
324		child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
325		if (child && list_empty(&child->mnt_mounts) &&
326		    (ret = do_refcount_check(child, 1)))
327			break;
328	}
329	return ret;
330}
331
332/*
333 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
334 * parent propagates to.
335 */
336static void __propagate_umount(struct vfsmount *mnt)
337{
338	struct vfsmount *parent = mnt->mnt_parent;
339	struct vfsmount *m;
340
341	BUG_ON(parent == mnt);
342
343	for (m = propagation_next(parent, parent); m;
344			m = propagation_next(m, parent)) {
345
346		struct vfsmount *child = __lookup_mnt(m,
347					mnt->mnt_mountpoint, 0);
348		/*
349		 * umount the child only if the child has no
350		 * other children
351		 */
352		if (child && list_empty(&child->mnt_mounts))
353			list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
354	}
355}
356
357/*
358 * collect all mounts that receive propagation from the mount in @list,
359 * and return these additional mounts in the same list.
360 * @list: the list of mounts to be unmounted.
361 *
362 * vfsmount lock must be held for write
363 */
364int propagate_umount(struct list_head *list)
365{
366	struct vfsmount *mnt;
367
368	list_for_each_entry(mnt, list, mnt_hash)
369		__propagate_umount(mnt);
370	return 0;
371}
372