blob: 6519b3b4eb155b3689136bdbff1a269537849941 [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 */
Al Viroc9371352011-11-24 23:56:26 -050016static inline struct mount *next_peer(struct mount *p)
Ram Pai03e06e62005-11-07 17:19:33 -050017{
Al Viroc9371352011-11-24 23:56:26 -050018 return list_entry(p->mnt.mnt_share.next, struct mount, mnt.mnt_share);
Ram Pai03e06e62005-11-07 17:19:33 -050019}
20
Al Viroc9371352011-11-24 23:56:26 -050021static inline struct mount *first_slave(struct mount *p)
Ram Pai5afe0022005-11-07 17:21:01 -050022{
Al Viroc9371352011-11-24 23:56:26 -050023 return list_entry(p->mnt.mnt_slave_list.next, struct mount, mnt.mnt_slave);
Ram Pai5afe0022005-11-07 17:21:01 -050024}
25
Al Viroc9371352011-11-24 23:56:26 -050026static inline struct mount *next_slave(struct mount *p)
Ram Pai5afe0022005-11-07 17:21:01 -050027{
Al Viroc9371352011-11-24 23:56:26 -050028 return list_entry(p->mnt.mnt_slave.next, struct mount, mnt.mnt_slave);
Ram Pai5afe0022005-11-07 17:21:01 -050029}
30
Al Viro6fc78712011-11-24 23:35:54 -050031static struct mount *get_peer_under_root(struct mount *mnt,
32 struct mnt_namespace *ns,
33 const struct path *root)
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010034{
Al Viro6fc78712011-11-24 23:35:54 -050035 struct mount *m = mnt;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010036
37 do {
38 /* Check the namespace first for optimization */
Al Viro643822b2011-11-24 22:00:28 -050039 if (m->mnt.mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
Al Viro6fc78712011-11-24 23:35:54 -050040 return m;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010041
Al Viroc9371352011-11-24 23:56:26 -050042 m = next_peer(m);
Al Viro6fc78712011-11-24 23:35:54 -050043 } while (m != mnt);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010044
45 return NULL;
46}
47
48/*
49 * Get ID of closest dominating peer group having a representative
50 * under the given root.
51 *
52 * Caller must hold namespace_sem
53 */
Al Viro6fc78712011-11-24 23:35:54 -050054int get_dominating_id(struct mount *mnt, const struct path *root)
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010055{
Al Viro6fc78712011-11-24 23:35:54 -050056 struct mount *m;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010057
Al Viro6fc78712011-11-24 23:35:54 -050058 for (m = real_mount(mnt->mnt.mnt_master); m != NULL; m = real_mount(m->mnt.mnt_master)) {
59 struct mount *d = get_peer_under_root(m, mnt->mnt.mnt_ns, root);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010060 if (d)
Al Viro6fc78712011-11-24 23:35:54 -050061 return d->mnt.mnt_group_id;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010062 }
63
64 return 0;
65}
66
Al Viro6fc78712011-11-24 23:35:54 -050067static int do_make_slave(struct mount *mnt)
Ram Paia58b0eb2005-11-07 17:20:48 -050068{
Al Viro6fc78712011-11-24 23:35:54 -050069 struct mount *peer_mnt = mnt, *master = real_mount(mnt->mnt.mnt_master);
Ram Paia58b0eb2005-11-07 17:20:48 -050070 struct vfsmount *slave_mnt;
71
72 /*
73 * slave 'mnt' to a peer mount that has the
Al Viro796a6b52010-01-16 13:28:47 -050074 * same root dentry. If none is available then
Ram Paia58b0eb2005-11-07 17:20:48 -050075 * slave it to anything that is available.
76 */
Al Viroc9371352011-11-24 23:56:26 -050077 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
Al Viro6fc78712011-11-24 23:35:54 -050078 peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
Ram Paia58b0eb2005-11-07 17:20:48 -050079
80 if (peer_mnt == mnt) {
Al Viroc9371352011-11-24 23:56:26 -050081 peer_mnt = next_peer(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -050082 if (peer_mnt == mnt)
83 peer_mnt = NULL;
84 }
Al Viro6fc78712011-11-24 23:35:54 -050085 if (IS_MNT_SHARED(&mnt->mnt) && list_empty(&mnt->mnt.mnt_share))
86 mnt_release_group_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010087
Al Viro6fc78712011-11-24 23:35:54 -050088 list_del_init(&mnt->mnt.mnt_share);
89 mnt->mnt.mnt_group_id = 0;
Ram Paia58b0eb2005-11-07 17:20:48 -050090
91 if (peer_mnt)
92 master = peer_mnt;
93
94 if (master) {
Al Viro6fc78712011-11-24 23:35:54 -050095 list_for_each_entry(slave_mnt, &mnt->mnt.mnt_slave_list, mnt_slave)
96 slave_mnt->mnt_master = &master->mnt;
97 list_move(&mnt->mnt.mnt_slave, &master->mnt.mnt_slave_list);
98 list_splice(&mnt->mnt.mnt_slave_list, master->mnt.mnt_slave_list.prev);
99 INIT_LIST_HEAD(&mnt->mnt.mnt_slave_list);
Ram Paia58b0eb2005-11-07 17:20:48 -0500100 } else {
Al Viro6fc78712011-11-24 23:35:54 -0500101 struct list_head *p = &mnt->mnt.mnt_slave_list;
Ram Paia58b0eb2005-11-07 17:20:48 -0500102 while (!list_empty(p)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700103 slave_mnt = list_first_entry(p,
Ram Paia58b0eb2005-11-07 17:20:48 -0500104 struct vfsmount, mnt_slave);
105 list_del_init(&slave_mnt->mnt_slave);
106 slave_mnt->mnt_master = NULL;
107 }
108 }
Al Viro6fc78712011-11-24 23:35:54 -0500109 mnt->mnt.mnt_master = &master->mnt;
110 CLEAR_MNT_SHARED(&mnt->mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500111 return 0;
112}
113
Nick Piggin99b7db72010-08-18 04:37:39 +1000114/*
115 * vfsmount lock must be held for write
116 */
Al Viro0f0afb12011-11-24 20:43:10 -0500117void change_mnt_propagation(struct mount *mnt, int type)
Ram Pai07b20882005-11-07 17:19:07 -0500118{
Ram Pai03e06e62005-11-07 17:19:33 -0500119 if (type == MS_SHARED) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500120 set_mnt_shared(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500121 return;
122 }
Al Viro6fc78712011-11-24 23:35:54 -0500123 do_make_slave(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500124 if (type != MS_SLAVE) {
Al Viro0f0afb12011-11-24 20:43:10 -0500125 list_del_init(&mnt->mnt.mnt_slave);
126 mnt->mnt.mnt_master = NULL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500127 if (type == MS_UNBINDABLE)
Al Viro0f0afb12011-11-24 20:43:10 -0500128 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
Andries E. Brouwer0b03cfb2008-02-06 01:36:32 -0800129 else
Al Viro0f0afb12011-11-24 20:43:10 -0500130 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
Ram Pai03e06e62005-11-07 17:19:33 -0500131 }
Ram Pai07b20882005-11-07 17:19:07 -0500132}
Ram Paib90fa9a2005-11-07 17:19:50 -0500133
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
Al Viro796a6b52010-01-16 13:28:47 -0500138 *
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.
Ram Paib90fa9a2005-11-07 17:19:50 -0500143 */
Al Viroc9371352011-11-24 23:56:26 -0500144static struct mount *propagation_next(struct mount *m,
145 struct mount *origin)
Ram Paib90fa9a2005-11-07 17:19:50 -0500146{
Ram Pai5afe0022005-11-07 17:21:01 -0500147 /* are there any slaves of this mount? */
Al Viroc9371352011-11-24 23:56:26 -0500148 if (!IS_MNT_NEW(&m->mnt) && !list_empty(&m->mnt.mnt_slave_list))
Ram Pai5afe0022005-11-07 17:21:01 -0500149 return first_slave(m);
150
151 while (1) {
Al Viroc9371352011-11-24 23:56:26 -0500152 struct mount *master = real_mount(m->mnt.mnt_master);
Ram Pai5afe0022005-11-07 17:21:01 -0500153
Al Viroc9371352011-11-24 23:56:26 -0500154 if (&master->mnt == origin->mnt.mnt_master) {
155 struct mount *next = next_peer(m);
156 return (next == origin) ? NULL : next;
157 } else if (m->mnt.mnt_slave.next != &master->mnt.mnt_slave_list)
Ram Pai5afe0022005-11-07 17:21:01 -0500158 return next_slave(m);
159
160 /* back at master */
161 m = master;
162 }
163}
164
165/*
166 * return the source mount to be used for cloning
167 *
168 * @dest the current destination mount
169 * @last_dest the last seen destination mount
170 * @last_src the last seen source mount
171 * @type return CL_SLAVE if the new mount has to be
172 * cloned as a slave.
173 */
Al Viroc9371352011-11-24 23:56:26 -0500174static struct mount *get_source(struct mount *dest,
175 struct mount *last_dest,
176 struct mount *last_src,
177 int *type)
Ram Pai5afe0022005-11-07 17:21:01 -0500178{
Al Viroc9371352011-11-24 23:56:26 -0500179 struct mount *p_last_src = NULL;
180 struct mount *p_last_dest = NULL;
Ram Pai5afe0022005-11-07 17:21:01 -0500181
Al Viroc9371352011-11-24 23:56:26 -0500182 while (&last_dest->mnt != dest->mnt.mnt_master) {
Ram Pai5afe0022005-11-07 17:21:01 -0500183 p_last_dest = last_dest;
184 p_last_src = last_src;
Al Viroc9371352011-11-24 23:56:26 -0500185 last_dest = real_mount(last_dest->mnt.mnt_master);
186 last_src = real_mount(last_src->mnt.mnt_master);
Ram Pai5afe0022005-11-07 17:21:01 -0500187 }
188
189 if (p_last_dest) {
190 do {
191 p_last_dest = next_peer(p_last_dest);
Al Viroc9371352011-11-24 23:56:26 -0500192 } while (IS_MNT_NEW(&p_last_dest->mnt));
Al Viro796a6b52010-01-16 13:28:47 -0500193 /* is that a peer of the earlier? */
194 if (dest == p_last_dest) {
195 *type = CL_MAKE_SHARED;
196 return p_last_src;
197 }
Ram Pai5afe0022005-11-07 17:21:01 -0500198 }
Al Viro796a6b52010-01-16 13:28:47 -0500199 /* slave of the earlier, then */
200 *type = CL_SLAVE;
201 /* beginning of peer group among the slaves? */
Al Viroc9371352011-11-24 23:56:26 -0500202 if (IS_MNT_SHARED(&dest->mnt))
Al Viro796a6b52010-01-16 13:28:47 -0500203 *type |= CL_MAKE_SHARED;
204 return last_src;
Ram Paib90fa9a2005-11-07 17:19:50 -0500205}
206
207/*
208 * mount 'source_mnt' under the destination 'dest_mnt' at
209 * dentry 'dest_dentry'. And propagate that mount to
210 * all the peer and slave mounts of 'dest_mnt'.
211 * Link all the new mounts into a propagation tree headed at
212 * source_mnt. Also link all the new mounts using ->mnt_list
213 * headed at source_mnt's ->mnt_list
214 *
215 * @dest_mnt: destination mount.
216 * @dest_dentry: destination dentry.
217 * @source_mnt: source mount.
218 * @tree_list : list of heads of trees to be attached.
219 */
Al Viroa8d56d82011-11-24 23:59:29 -0500220int propagate_mnt(struct mount *dest_mnt, struct dentry *dest_dentry,
221 struct mount *source_mnt, struct list_head *tree_list)
Ram Paib90fa9a2005-11-07 17:19:50 -0500222{
Al Viroc9371352011-11-24 23:56:26 -0500223 struct mount *m, *child;
Ram Paib90fa9a2005-11-07 17:19:50 -0500224 int ret = 0;
Al Viroa8d56d82011-11-24 23:59:29 -0500225 struct mount *prev_dest_mnt = dest_mnt;
226 struct mount *prev_src_mnt = source_mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500227 LIST_HEAD(tmp_list);
228 LIST_HEAD(umount_list);
229
Al Viroa8d56d82011-11-24 23:59:29 -0500230 for (m = propagation_next(dest_mnt, dest_mnt); m;
231 m = propagation_next(m, dest_mnt)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500232 int type;
Al Viroc9371352011-11-24 23:56:26 -0500233 struct mount *source;
Ram Paib90fa9a2005-11-07 17:19:50 -0500234
Al Viroc9371352011-11-24 23:56:26 -0500235 if (IS_MNT_NEW(&m->mnt))
Ram Paib90fa9a2005-11-07 17:19:50 -0500236 continue;
237
Ram Pai5afe0022005-11-07 17:21:01 -0500238 source = get_source(m, prev_dest_mnt, prev_src_mnt, &type);
Ram Paib90fa9a2005-11-07 17:19:50 -0500239
Al Viroc9371352011-11-24 23:56:26 -0500240 if (!(child = copy_tree(source, source->mnt.mnt_root, type))) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500241 ret = -ENOMEM;
242 list_splice(tree_list, tmp_list.prev);
243 goto out;
244 }
245
Al Viroc9371352011-11-24 23:56:26 -0500246 if (is_subdir(dest_dentry, m->mnt.mnt_root)) {
247 mnt_set_mountpoint(&m->mnt, dest_dentry, child);
Al Viro1b8e5562011-11-24 21:01:32 -0500248 list_add_tail(&child->mnt_hash, tree_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500249 } else {
250 /*
251 * This can happen if the parent mount was bind mounted
252 * on some subdirectory of a shared/slave mount.
253 */
Al Viro1b8e5562011-11-24 21:01:32 -0500254 list_add_tail(&child->mnt_hash, &tmp_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500255 }
256 prev_dest_mnt = m;
Al Viroc9371352011-11-24 23:56:26 -0500257 prev_src_mnt = child;
Ram Paib90fa9a2005-11-07 17:19:50 -0500258 }
259out:
Nick Piggin99b7db72010-08-18 04:37:39 +1000260 br_write_lock(vfsmount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500261 while (!list_empty(&tmp_list)) {
Al Viro1b8e5562011-11-24 21:01:32 -0500262 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
Al Viro761d5c32011-11-24 21:07:43 -0500263 umount_tree(child, 0, &umount_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500264 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000265 br_write_unlock(vfsmount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500266 release_mounts(&umount_list);
267 return ret;
268}
Ram Paia05964f2005-11-07 17:20:17 -0500269
270/*
271 * return true if the refcount is greater than count
272 */
Al Viro1ab59732011-11-24 21:35:16 -0500273static inline int do_refcount_check(struct mount *mnt, int count)
Ram Paia05964f2005-11-07 17:20:17 -0500274{
Al Viro83adc752011-11-24 22:37:54 -0500275 int mycount = mnt_get_count(mnt) - mnt->mnt.mnt_ghosts;
Ram Paia05964f2005-11-07 17:20:17 -0500276 return (mycount > count);
277}
278
279/*
280 * check if the mount 'mnt' can be unmounted successfully.
281 * @mnt: the mount to be checked for unmount
282 * NOTE: unmounting 'mnt' would naturally propagate to all
283 * other mounts its parent propagates to.
284 * Check if any of these mounts that **do not have submounts**
285 * have more references than 'refcnt'. If so return busy.
Nick Piggin99b7db72010-08-18 04:37:39 +1000286 *
Nick Pigginb3e19d92011-01-07 17:50:11 +1100287 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500288 */
Al Viro1ab59732011-11-24 21:35:16 -0500289int propagate_mount_busy(struct mount *mnt, int refcnt)
Ram Paia05964f2005-11-07 17:20:17 -0500290{
Al Viroc9371352011-11-24 23:56:26 -0500291 struct mount *m, *child;
Al Viro0714a532011-11-24 22:19:58 -0500292 struct mount *parent = mnt->mnt_parent;
Ram Paia05964f2005-11-07 17:20:17 -0500293 int ret = 0;
294
Al Viro0714a532011-11-24 22:19:58 -0500295 if (mnt == parent)
Ram Paia05964f2005-11-07 17:20:17 -0500296 return do_refcount_check(mnt, refcnt);
297
298 /*
299 * quickly check if the current mount can be unmounted.
300 * If not, we don't have to go checking for all other
301 * mounts
302 */
Al Viro6b41d532011-11-24 23:24:33 -0500303 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
Ram Paia05964f2005-11-07 17:20:17 -0500304 return 1;
305
Al Viroc9371352011-11-24 23:56:26 -0500306 for (m = propagation_next(parent, parent); m;
307 m = propagation_next(m, parent)) {
308 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint, 0);
Al Viro6b41d532011-11-24 23:24:33 -0500309 if (child && list_empty(&child->mnt_mounts) &&
Al Viro1ab59732011-11-24 21:35:16 -0500310 (ret = do_refcount_check(child, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500311 break;
312 }
313 return ret;
314}
315
316/*
317 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
318 * parent propagates to.
319 */
Al Viro61ef47b2011-11-24 18:25:28 -0500320static void __propagate_umount(struct mount *mnt)
Ram Paia05964f2005-11-07 17:20:17 -0500321{
Al Viro0714a532011-11-24 22:19:58 -0500322 struct mount *parent = mnt->mnt_parent;
Al Viroc9371352011-11-24 23:56:26 -0500323 struct mount *m;
Ram Paia05964f2005-11-07 17:20:17 -0500324
Al Viro0714a532011-11-24 22:19:58 -0500325 BUG_ON(parent == mnt);
Ram Paia05964f2005-11-07 17:20:17 -0500326
Al Viroc9371352011-11-24 23:56:26 -0500327 for (m = propagation_next(parent, parent); m;
328 m = propagation_next(m, parent)) {
Ram Paia05964f2005-11-07 17:20:17 -0500329
Al Viroc9371352011-11-24 23:56:26 -0500330 struct mount *child = __lookup_mnt(&m->mnt,
Al Viroa73324d2011-11-24 22:25:07 -0500331 mnt->mnt_mountpoint, 0);
Ram Paia05964f2005-11-07 17:20:17 -0500332 /*
333 * umount the child only if the child has no
334 * other children
335 */
Al Viro6b41d532011-11-24 23:24:33 -0500336 if (child && list_empty(&child->mnt_mounts))
Al Viro1b8e5562011-11-24 21:01:32 -0500337 list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
Ram Paia05964f2005-11-07 17:20:17 -0500338 }
339}
340
341/*
342 * collect all mounts that receive propagation from the mount in @list,
343 * and return these additional mounts in the same list.
344 * @list: the list of mounts to be unmounted.
Nick Piggin99b7db72010-08-18 04:37:39 +1000345 *
346 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500347 */
348int propagate_umount(struct list_head *list)
349{
Al Viro61ef47b2011-11-24 18:25:28 -0500350 struct mount *mnt;
Ram Paia05964f2005-11-07 17:20:17 -0500351
Al Viro1b8e5562011-11-24 21:01:32 -0500352 list_for_each_entry(mnt, list, mnt_hash)
Ram Paia05964f2005-11-07 17:20:17 -0500353 __propagate_umount(mnt);
354 return 0;
355}