blob: 5cc564a83149a5fc311b57d87793febbdcfd4f27 [file] [log] [blame]
Ram Pai07b20882005-11-07 17:19:07 -05001/*
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 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08009#include <linux/mnt_namespace.h>
Ram Pai07b20882005-11-07 17:19:07 -050010#include <linux/mount.h>
11#include <linux/fs.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040012#include "internal.h"
Ram Pai07b20882005-11-07 17:19:07 -050013#include "pnode.h"
14
Ram Pai03e06e62005-11-07 17:19:33 -050015/* 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
Ram Pai5afe0022005-11-07 17:21:01 -050021static 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
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010031/*
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
Ram Paia58b0eb2005-11-07 17:20:48 -050082static 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
Al Viro796a6b52010-01-16 13:28:47 -050089 * same root dentry. If none is available then
Ram Paia58b0eb2005-11-07 17:20:48 -050090 * 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 }
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100100 if (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
101 mnt_release_group_id(mnt);
102
Ram Paia58b0eb2005-11-07 17:20:48 -0500103 list_del_init(&mnt->mnt_share);
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100104 mnt->mnt_group_id = 0;
Ram Paia58b0eb2005-11-07 17:20:48 -0500105
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;
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700112 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
Ram Paia58b0eb2005-11-07 17:20:48 -0500113 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)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700118 slave_mnt = list_first_entry(p,
Ram Paia58b0eb2005-11-07 17:20:48 -0500119 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);
Ram Paia58b0eb2005-11-07 17:20:48 -0500126 return 0;
127}
128
Ram Pai07b20882005-11-07 17:19:07 -0500129void change_mnt_propagation(struct vfsmount *mnt, int type)
130{
Ram Pai03e06e62005-11-07 17:19:33 -0500131 if (type == MS_SHARED) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500132 set_mnt_shared(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500133 return;
134 }
135 do_make_slave(mnt);
136 if (type != MS_SLAVE) {
137 list_del_init(&mnt->mnt_slave);
138 mnt->mnt_master = NULL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500139 if (type == MS_UNBINDABLE)
140 mnt->mnt_flags |= MNT_UNBINDABLE;
Andries E. Brouwer0b03cfb2008-02-06 01:36:32 -0800141 else
142 mnt->mnt_flags &= ~MNT_UNBINDABLE;
Ram Pai03e06e62005-11-07 17:19:33 -0500143 }
Ram Pai07b20882005-11-07 17:19:07 -0500144}
Ram Paib90fa9a2005-11-07 17:19:50 -0500145
146/*
147 * get the next mount in the propagation tree.
148 * @m: the mount seen last
149 * @origin: the original mount from where the tree walk initiated
Al Viro796a6b52010-01-16 13:28:47 -0500150 *
151 * Note that peer groups form contiguous segments of slave lists.
152 * We rely on that in get_source() to be able to find out if
153 * vfsmount found while iterating with propagation_next() is
154 * a peer of one we'd found earlier.
Ram Paib90fa9a2005-11-07 17:19:50 -0500155 */
156static struct vfsmount *propagation_next(struct vfsmount *m,
157 struct vfsmount *origin)
158{
Ram Pai5afe0022005-11-07 17:21:01 -0500159 /* are there any slaves of this mount? */
160 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
161 return first_slave(m);
162
163 while (1) {
164 struct vfsmount *next;
165 struct vfsmount *master = m->mnt_master;
166
Miklos Szeredibf066c72006-01-08 01:03:19 -0800167 if (master == origin->mnt_master) {
Ram Pai5afe0022005-11-07 17:21:01 -0500168 next = next_peer(m);
169 return ((next == origin) ? NULL : next);
170 } else if (m->mnt_slave.next != &master->mnt_slave_list)
171 return next_slave(m);
172
173 /* back at master */
174 m = master;
175 }
176}
177
178/*
179 * return the source mount to be used for cloning
180 *
181 * @dest the current destination mount
182 * @last_dest the last seen destination mount
183 * @last_src the last seen source mount
184 * @type return CL_SLAVE if the new mount has to be
185 * cloned as a slave.
186 */
187static struct vfsmount *get_source(struct vfsmount *dest,
188 struct vfsmount *last_dest,
189 struct vfsmount *last_src,
190 int *type)
191{
192 struct vfsmount *p_last_src = NULL;
193 struct vfsmount *p_last_dest = NULL;
Ram Pai5afe0022005-11-07 17:21:01 -0500194
195 while (last_dest != dest->mnt_master) {
196 p_last_dest = last_dest;
197 p_last_src = last_src;
198 last_dest = last_dest->mnt_master;
199 last_src = last_src->mnt_master;
200 }
201
202 if (p_last_dest) {
203 do {
204 p_last_dest = next_peer(p_last_dest);
205 } while (IS_MNT_NEW(p_last_dest));
Al Viro796a6b52010-01-16 13:28:47 -0500206 /* is that a peer of the earlier? */
207 if (dest == p_last_dest) {
208 *type = CL_MAKE_SHARED;
209 return p_last_src;
210 }
Ram Pai5afe0022005-11-07 17:21:01 -0500211 }
Al Viro796a6b52010-01-16 13:28:47 -0500212 /* slave of the earlier, then */
213 *type = CL_SLAVE;
214 /* beginning of peer group among the slaves? */
215 if (IS_MNT_SHARED(dest))
216 *type |= CL_MAKE_SHARED;
217 return last_src;
Ram Paib90fa9a2005-11-07 17:19:50 -0500218}
219
220/*
221 * mount 'source_mnt' under the destination 'dest_mnt' at
222 * dentry 'dest_dentry'. And propagate that mount to
223 * all the peer and slave mounts of 'dest_mnt'.
224 * Link all the new mounts into a propagation tree headed at
225 * source_mnt. Also link all the new mounts using ->mnt_list
226 * headed at source_mnt's ->mnt_list
227 *
228 * @dest_mnt: destination mount.
229 * @dest_dentry: destination dentry.
230 * @source_mnt: source mount.
231 * @tree_list : list of heads of trees to be attached.
232 */
233int propagate_mnt(struct vfsmount *dest_mnt, struct dentry *dest_dentry,
234 struct vfsmount *source_mnt, struct list_head *tree_list)
235{
236 struct vfsmount *m, *child;
237 int ret = 0;
238 struct vfsmount *prev_dest_mnt = dest_mnt;
239 struct vfsmount *prev_src_mnt = source_mnt;
240 LIST_HEAD(tmp_list);
241 LIST_HEAD(umount_list);
242
243 for (m = propagation_next(dest_mnt, dest_mnt); m;
244 m = propagation_next(m, dest_mnt)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500245 int type;
246 struct vfsmount *source;
Ram Paib90fa9a2005-11-07 17:19:50 -0500247
248 if (IS_MNT_NEW(m))
249 continue;
250
Ram Pai5afe0022005-11-07 17:21:01 -0500251 source = get_source(m, prev_dest_mnt, prev_src_mnt, &type);
Ram Paib90fa9a2005-11-07 17:19:50 -0500252
Ram Pai5afe0022005-11-07 17:21:01 -0500253 if (!(child = copy_tree(source, source->mnt_root, type))) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500254 ret = -ENOMEM;
255 list_splice(tree_list, tmp_list.prev);
256 goto out;
257 }
258
259 if (is_subdir(dest_dentry, m->mnt_root)) {
260 mnt_set_mountpoint(m, dest_dentry, child);
261 list_add_tail(&child->mnt_hash, tree_list);
262 } else {
263 /*
264 * This can happen if the parent mount was bind mounted
265 * on some subdirectory of a shared/slave mount.
266 */
267 list_add_tail(&child->mnt_hash, &tmp_list);
268 }
269 prev_dest_mnt = m;
270 prev_src_mnt = child;
271 }
272out:
273 spin_lock(&vfsmount_lock);
274 while (!list_empty(&tmp_list)) {
Al Viro4e1b36f2008-03-24 00:16:03 -0400275 child = list_first_entry(&tmp_list, struct vfsmount, mnt_hash);
Ram Paia05964f2005-11-07 17:20:17 -0500276 umount_tree(child, 0, &umount_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500277 }
278 spin_unlock(&vfsmount_lock);
279 release_mounts(&umount_list);
280 return ret;
281}
Ram Paia05964f2005-11-07 17:20:17 -0500282
283/*
284 * return true if the refcount is greater than count
285 */
286static inline int do_refcount_check(struct vfsmount *mnt, int count)
287{
Al Viro7c4b93d2008-03-21 23:59:49 -0400288 int mycount = atomic_read(&mnt->mnt_count) - mnt->mnt_ghosts;
Ram Paia05964f2005-11-07 17:20:17 -0500289 return (mycount > count);
290}
291
292/*
293 * check if the mount 'mnt' can be unmounted successfully.
294 * @mnt: the mount to be checked for unmount
295 * NOTE: unmounting 'mnt' would naturally propagate to all
296 * other mounts its parent propagates to.
297 * Check if any of these mounts that **do not have submounts**
298 * have more references than 'refcnt'. If so return busy.
299 */
300int propagate_mount_busy(struct vfsmount *mnt, int refcnt)
301{
302 struct vfsmount *m, *child;
303 struct vfsmount *parent = mnt->mnt_parent;
304 int ret = 0;
305
306 if (mnt == parent)
307 return do_refcount_check(mnt, refcnt);
308
309 /*
310 * quickly check if the current mount can be unmounted.
311 * If not, we don't have to go checking for all other
312 * mounts
313 */
314 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
315 return 1;
316
317 for (m = propagation_next(parent, parent); m;
318 m = propagation_next(m, parent)) {
319 child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
320 if (child && list_empty(&child->mnt_mounts) &&
321 (ret = do_refcount_check(child, 1)))
322 break;
323 }
324 return ret;
325}
326
327/*
328 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
329 * parent propagates to.
330 */
331static void __propagate_umount(struct vfsmount *mnt)
332{
333 struct vfsmount *parent = mnt->mnt_parent;
334 struct vfsmount *m;
335
336 BUG_ON(parent == mnt);
337
338 for (m = propagation_next(parent, parent); m;
339 m = propagation_next(m, parent)) {
340
341 struct vfsmount *child = __lookup_mnt(m,
342 mnt->mnt_mountpoint, 0);
343 /*
344 * umount the child only if the child has no
345 * other children
346 */
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700347 if (child && list_empty(&child->mnt_mounts))
348 list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
Ram Paia05964f2005-11-07 17:20:17 -0500349 }
350}
351
352/*
353 * collect all mounts that receive propagation from the mount in @list,
354 * and return these additional mounts in the same list.
355 * @list: the list of mounts to be unmounted.
356 */
357int propagate_umount(struct list_head *list)
358{
359 struct vfsmount *mnt;
360
361 list_for_each_entry(mnt, list, mnt_hash)
362 __propagate_umount(mnt);
363 return 0;
364}