blob: 260ac8f898a43de82de925d9befad1613a7764fb [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>
Eric W. Biederman132c94e2013-03-22 04:08:05 -070012#include <linux/nsproxy.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040013#include "internal.h"
Ram Pai07b20882005-11-07 17:19:07 -050014#include "pnode.h"
15
Ram Pai03e06e62005-11-07 17:19:33 -050016/* return the next shared peer mount of @p */
Al Viroc9371352011-11-24 23:56:26 -050017static inline struct mount *next_peer(struct mount *p)
Ram Pai03e06e62005-11-07 17:19:33 -050018{
Al Viro6776db3d2011-11-25 00:22:05 -050019 return list_entry(p->mnt_share.next, struct mount, mnt_share);
Ram Pai03e06e62005-11-07 17:19:33 -050020}
21
Al Viroc9371352011-11-24 23:56:26 -050022static inline struct mount *first_slave(struct mount *p)
Ram Pai5afe0022005-11-07 17:21:01 -050023{
Al Viro6776db3d2011-11-25 00:22:05 -050024 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
Ram Pai5afe0022005-11-07 17:21:01 -050025}
26
Al Viroc9371352011-11-24 23:56:26 -050027static inline struct mount *next_slave(struct mount *p)
Ram Pai5afe0022005-11-07 17:21:01 -050028{
Al Viro6776db3d2011-11-25 00:22:05 -050029 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
Ram Pai5afe0022005-11-07 17:21:01 -050030}
31
Al Viro6fc78712011-11-24 23:35:54 -050032static struct mount *get_peer_under_root(struct mount *mnt,
33 struct mnt_namespace *ns,
34 const struct path *root)
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010035{
Al Viro6fc78712011-11-24 23:35:54 -050036 struct mount *m = mnt;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010037
38 do {
39 /* Check the namespace first for optimization */
Al Viro143c8c92011-11-25 00:46:35 -050040 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
Al Viro6fc78712011-11-24 23:35:54 -050041 return m;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010042
Al Viroc9371352011-11-24 23:56:26 -050043 m = next_peer(m);
Al Viro6fc78712011-11-24 23:35:54 -050044 } while (m != mnt);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010045
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 */
Al Viro6fc78712011-11-24 23:35:54 -050055int get_dominating_id(struct mount *mnt, const struct path *root)
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010056{
Al Viro6fc78712011-11-24 23:35:54 -050057 struct mount *m;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010058
Al Viro32301922011-11-25 00:10:28 -050059 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
Al Viro143c8c92011-11-25 00:46:35 -050060 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010061 if (d)
Al Viro15169fe2011-11-25 00:50:41 -050062 return d->mnt_group_id;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010063 }
64
65 return 0;
66}
67
Al Viro6fc78712011-11-24 23:35:54 -050068static int do_make_slave(struct mount *mnt)
Ram Paia58b0eb2005-11-07 17:20:48 -050069{
Al Viro32301922011-11-25 00:10:28 -050070 struct mount *peer_mnt = mnt, *master = mnt->mnt_master;
Al Virod10e8de2011-11-25 00:07:16 -050071 struct mount *slave_mnt;
Ram Paia58b0eb2005-11-07 17:20:48 -050072
73 /*
74 * slave 'mnt' to a peer mount that has the
Al Viro796a6b52010-01-16 13:28:47 -050075 * same root dentry. If none is available then
Ram Paia58b0eb2005-11-07 17:20:48 -050076 * slave it to anything that is available.
77 */
Al Viroc9371352011-11-24 23:56:26 -050078 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
Al Viro6fc78712011-11-24 23:35:54 -050079 peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
Ram Paia58b0eb2005-11-07 17:20:48 -050080
81 if (peer_mnt == mnt) {
Al Viroc9371352011-11-24 23:56:26 -050082 peer_mnt = next_peer(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -050083 if (peer_mnt == mnt)
84 peer_mnt = NULL;
85 }
Takashi Iwai5d477b62013-05-10 14:04:11 +020086 if (mnt->mnt_group_id && IS_MNT_SHARED(mnt) &&
87 list_empty(&mnt->mnt_share))
Al Viro6fc78712011-11-24 23:35:54 -050088 mnt_release_group_id(mnt);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010089
Al Viro6776db3d2011-11-25 00:22:05 -050090 list_del_init(&mnt->mnt_share);
Al Viro15169fe2011-11-25 00:50:41 -050091 mnt->mnt_group_id = 0;
Ram Paia58b0eb2005-11-07 17:20:48 -050092
93 if (peer_mnt)
94 master = peer_mnt;
95
96 if (master) {
Al Viro6776db3d2011-11-25 00:22:05 -050097 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
Al Viro32301922011-11-25 00:10:28 -050098 slave_mnt->mnt_master = master;
Al Viro6776db3d2011-11-25 00:22:05 -050099 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);
Ram Paia58b0eb2005-11-07 17:20:48 -0500102 } else {
Al Viro6776db3d2011-11-25 00:22:05 -0500103 struct list_head *p = &mnt->mnt_slave_list;
Ram Paia58b0eb2005-11-07 17:20:48 -0500104 while (!list_empty(p)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700105 slave_mnt = list_first_entry(p,
Al Viro6776db3d2011-11-25 00:22:05 -0500106 struct mount, mnt_slave);
107 list_del_init(&slave_mnt->mnt_slave);
Ram Paia58b0eb2005-11-07 17:20:48 -0500108 slave_mnt->mnt_master = NULL;
109 }
110 }
Al Viro32301922011-11-25 00:10:28 -0500111 mnt->mnt_master = master;
Al Virofc7be132011-11-25 01:05:37 -0500112 CLEAR_MNT_SHARED(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500113 return 0;
114}
115
Nick Piggin99b7db72010-08-18 04:37:39 +1000116/*
117 * vfsmount lock must be held for write
118 */
Al Viro0f0afb12011-11-24 20:43:10 -0500119void change_mnt_propagation(struct mount *mnt, int type)
Ram Pai07b20882005-11-07 17:19:07 -0500120{
Ram Pai03e06e62005-11-07 17:19:33 -0500121 if (type == MS_SHARED) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500122 set_mnt_shared(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500123 return;
124 }
Al Viro6fc78712011-11-24 23:35:54 -0500125 do_make_slave(mnt);
Ram Paia58b0eb2005-11-07 17:20:48 -0500126 if (type != MS_SLAVE) {
Al Viro6776db3d2011-11-25 00:22:05 -0500127 list_del_init(&mnt->mnt_slave);
Al Virod10e8de2011-11-25 00:07:16 -0500128 mnt->mnt_master = NULL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500129 if (type == MS_UNBINDABLE)
Al Viro0f0afb12011-11-24 20:43:10 -0500130 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
Andries E. Brouwer0b03cfb2008-02-06 01:36:32 -0800131 else
Al Viro0f0afb12011-11-24 20:43:10 -0500132 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
Ram Pai03e06e62005-11-07 17:19:33 -0500133 }
Ram Pai07b20882005-11-07 17:19:07 -0500134}
Ram Paib90fa9a2005-11-07 17:19:50 -0500135
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
Al Viro796a6b52010-01-16 13:28:47 -0500140 *
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.
Ram Paib90fa9a2005-11-07 17:19:50 -0500145 */
Al Viroc9371352011-11-24 23:56:26 -0500146static struct mount *propagation_next(struct mount *m,
147 struct mount *origin)
Ram Paib90fa9a2005-11-07 17:19:50 -0500148{
Ram Pai5afe0022005-11-07 17:21:01 -0500149 /* are there any slaves of this mount? */
Al Viro143c8c92011-11-25 00:46:35 -0500150 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
Ram Pai5afe0022005-11-07 17:21:01 -0500151 return first_slave(m);
152
153 while (1) {
Al Viro32301922011-11-25 00:10:28 -0500154 struct mount *master = m->mnt_master;
Ram Pai5afe0022005-11-07 17:21:01 -0500155
Al Viro32301922011-11-25 00:10:28 -0500156 if (master == origin->mnt_master) {
Al Viroc9371352011-11-24 23:56:26 -0500157 struct mount *next = next_peer(m);
158 return (next == origin) ? NULL : next;
Al Viro6776db3d2011-11-25 00:22:05 -0500159 } else if (m->mnt_slave.next != &master->mnt_slave_list)
Ram Pai5afe0022005-11-07 17:21:01 -0500160 return next_slave(m);
161
162 /* back at master */
163 m = master;
164 }
165}
166
Al Virof2ebb3a2014-02-27 09:35:45 -0500167static struct mount *next_group(struct mount *m, struct mount *origin)
Ram Pai5afe0022005-11-07 17:21:01 -0500168{
Al Virof2ebb3a2014-02-27 09:35:45 -0500169 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;
Al Viro796a6b52010-01-16 13:28:47 -0500181 }
Al Virof2ebb3a2014-02-27 09:35:45 -0500182 /* 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;
Ram Pai5afe0022005-11-07 17:21:01 -0500196 }
Al Virof2ebb3a2014-02-27 09:35:45 -0500197}
198
199/* all accesses are serialized by namespace_sem */
200static struct user_namespace *user_ns;
201static struct mount *last_dest, *last_source, *dest_master;
202static struct mountpoint *mp;
203static struct hlist_head *list;
204
205static int propagate_one(struct mount *m)
206{
207 struct mount *child;
208 int type;
209 /* skip ones added by this propagate_mnt() */
210 if (IS_MNT_NEW(m))
211 return 0;
212 /* skip if mountpoint isn't covered by it */
213 if (!is_subdir(mp->m_dentry, m->mnt.mnt_root))
214 return 0;
215 if (m->mnt_group_id == last_dest->mnt_group_id) {
216 type = CL_MAKE_SHARED;
217 } else {
218 struct mount *n, *p;
219 for (n = m; ; n = p) {
220 p = n->mnt_master;
221 if (p == dest_master || IS_MNT_MARKED(p)) {
222 while (last_dest->mnt_master != p) {
223 last_source = last_source->mnt_master;
224 last_dest = last_source->mnt_parent;
225 }
226 if (n->mnt_group_id != last_dest->mnt_group_id) {
227 last_source = last_source->mnt_master;
228 last_dest = last_source->mnt_parent;
229 }
230 break;
231 }
232 }
233 type = CL_SLAVE;
234 /* beginning of peer group among the slaves? */
235 if (IS_MNT_SHARED(m))
236 type |= CL_MAKE_SHARED;
237 }
238
239 /* Notice when we are propagating across user namespaces */
240 if (m->mnt_ns->user_ns != user_ns)
241 type |= CL_UNPRIVILEGED;
242 child = copy_tree(last_source, last_source->mnt.mnt_root, type);
243 if (IS_ERR(child))
244 return PTR_ERR(child);
Eric W. Biederman8486a782014-10-07 16:22:52 -0700245 child->mnt.mnt_flags &= ~MNT_LOCKED;
Al Virof2ebb3a2014-02-27 09:35:45 -0500246 mnt_set_mountpoint(m, mp, child);
247 last_dest = m;
248 last_source = child;
249 if (m->mnt_master != dest_master) {
250 read_seqlock_excl(&mount_lock);
251 SET_MNT_MARK(m->mnt_master);
252 read_sequnlock_excl(&mount_lock);
253 }
254 hlist_add_head(&child->mnt_hash, list);
255 return 0;
Ram Paib90fa9a2005-11-07 17:19:50 -0500256}
257
258/*
259 * mount 'source_mnt' under the destination 'dest_mnt' at
260 * dentry 'dest_dentry'. And propagate that mount to
261 * all the peer and slave mounts of 'dest_mnt'.
262 * Link all the new mounts into a propagation tree headed at
263 * source_mnt. Also link all the new mounts using ->mnt_list
264 * headed at source_mnt's ->mnt_list
265 *
266 * @dest_mnt: destination mount.
267 * @dest_dentry: destination dentry.
268 * @source_mnt: source mount.
269 * @tree_list : list of heads of trees to be attached.
270 */
Al Viro84d17192013-03-15 10:53:28 -0400271int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
Al Viro38129a12014-03-20 21:10:51 -0400272 struct mount *source_mnt, struct hlist_head *tree_list)
Ram Paib90fa9a2005-11-07 17:19:50 -0500273{
Al Virof2ebb3a2014-02-27 09:35:45 -0500274 struct mount *m, *n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500275 int ret = 0;
Ram Paib90fa9a2005-11-07 17:19:50 -0500276
Al Virof2ebb3a2014-02-27 09:35:45 -0500277 /*
278 * we don't want to bother passing tons of arguments to
279 * propagate_one(); everything is serialized by namespace_sem,
280 * so globals will do just fine.
281 */
282 user_ns = current->nsproxy->mnt_ns->user_ns;
283 last_dest = dest_mnt;
284 last_source = source_mnt;
285 mp = dest_mp;
286 list = tree_list;
287 dest_master = dest_mnt->mnt_master;
Ram Paib90fa9a2005-11-07 17:19:50 -0500288
Al Virof2ebb3a2014-02-27 09:35:45 -0500289 /* all peers of dest_mnt, except dest_mnt itself */
290 for (n = next_peer(dest_mnt); n != dest_mnt; n = next_peer(n)) {
291 ret = propagate_one(n);
292 if (ret)
Ram Paib90fa9a2005-11-07 17:19:50 -0500293 goto out;
Al Virof2ebb3a2014-02-27 09:35:45 -0500294 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500295
Al Virof2ebb3a2014-02-27 09:35:45 -0500296 /* all slave groups */
297 for (m = next_group(dest_mnt, dest_mnt); m;
298 m = next_group(m, dest_mnt)) {
299 /* everything in that slave group */
300 n = m;
301 do {
302 ret = propagate_one(n);
303 if (ret)
304 goto out;
305 n = next_peer(n);
306 } while (n != m);
Ram Paib90fa9a2005-11-07 17:19:50 -0500307 }
308out:
Al Virof2ebb3a2014-02-27 09:35:45 -0500309 read_seqlock_excl(&mount_lock);
310 hlist_for_each_entry(n, tree_list, mnt_hash) {
311 m = n->mnt_parent;
312 if (m->mnt_master != dest_mnt->mnt_master)
313 CLEAR_MNT_MARK(m->mnt_master);
Ram Paib90fa9a2005-11-07 17:19:50 -0500314 }
Al Virof2ebb3a2014-02-27 09:35:45 -0500315 read_sequnlock_excl(&mount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500316 return ret;
317}
Ram Paia05964f2005-11-07 17:20:17 -0500318
319/*
320 * return true if the refcount is greater than count
321 */
Al Viro1ab59732011-11-24 21:35:16 -0500322static inline int do_refcount_check(struct mount *mnt, int count)
Ram Paia05964f2005-11-07 17:20:17 -0500323{
Al Viroaba809cf2013-09-28 23:10:55 -0400324 return mnt_get_count(mnt) > count;
Ram Paia05964f2005-11-07 17:20:17 -0500325}
326
327/*
328 * check if the mount 'mnt' can be unmounted successfully.
329 * @mnt: the mount to be checked for unmount
330 * NOTE: unmounting 'mnt' would naturally propagate to all
331 * other mounts its parent propagates to.
332 * Check if any of these mounts that **do not have submounts**
333 * have more references than 'refcnt'. If so return busy.
Nick Piggin99b7db72010-08-18 04:37:39 +1000334 *
Nick Pigginb3e19d92011-01-07 17:50:11 +1100335 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500336 */
Al Viro1ab59732011-11-24 21:35:16 -0500337int propagate_mount_busy(struct mount *mnt, int refcnt)
Ram Paia05964f2005-11-07 17:20:17 -0500338{
Al Viroc9371352011-11-24 23:56:26 -0500339 struct mount *m, *child;
Al Viro0714a532011-11-24 22:19:58 -0500340 struct mount *parent = mnt->mnt_parent;
Ram Paia05964f2005-11-07 17:20:17 -0500341 int ret = 0;
342
Al Viro0714a532011-11-24 22:19:58 -0500343 if (mnt == parent)
Ram Paia05964f2005-11-07 17:20:17 -0500344 return do_refcount_check(mnt, refcnt);
345
346 /*
347 * quickly check if the current mount can be unmounted.
348 * If not, we don't have to go checking for all other
349 * mounts
350 */
Al Viro6b41d532011-11-24 23:24:33 -0500351 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
Ram Paia05964f2005-11-07 17:20:17 -0500352 return 1;
353
Al Viroc9371352011-11-24 23:56:26 -0500354 for (m = propagation_next(parent, parent); m;
355 m = propagation_next(m, parent)) {
Al Viro474279d2013-10-01 16:11:26 -0400356 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
Al Viro6b41d532011-11-24 23:24:33 -0500357 if (child && list_empty(&child->mnt_mounts) &&
Al Viro1ab59732011-11-24 21:35:16 -0500358 (ret = do_refcount_check(child, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500359 break;
360 }
361 return ret;
362}
363
364/*
365 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
366 * parent propagates to.
367 */
Al Viro61ef47b2011-11-24 18:25:28 -0500368static void __propagate_umount(struct mount *mnt)
Ram Paia05964f2005-11-07 17:20:17 -0500369{
Al Viro0714a532011-11-24 22:19:58 -0500370 struct mount *parent = mnt->mnt_parent;
Al Viroc9371352011-11-24 23:56:26 -0500371 struct mount *m;
Ram Paia05964f2005-11-07 17:20:17 -0500372
Al Viro0714a532011-11-24 22:19:58 -0500373 BUG_ON(parent == mnt);
Ram Paia05964f2005-11-07 17:20:17 -0500374
Al Viroc9371352011-11-24 23:56:26 -0500375 for (m = propagation_next(parent, parent); m;
376 m = propagation_next(m, parent)) {
Ram Paia05964f2005-11-07 17:20:17 -0500377
Al Viro474279d2013-10-01 16:11:26 -0400378 struct mount *child = __lookup_mnt_last(&m->mnt,
379 mnt->mnt_mountpoint);
Ram Paia05964f2005-11-07 17:20:17 -0500380 /*
381 * umount the child only if the child has no
382 * other children
383 */
Al Viro38129a12014-03-20 21:10:51 -0400384 if (child && list_empty(&child->mnt_mounts)) {
Al Viro88b368f2014-08-18 15:09:26 -0400385 list_del_init(&child->mnt_child);
Al Viro38129a12014-03-20 21:10:51 -0400386 hlist_del_init_rcu(&child->mnt_hash);
387 hlist_add_before_rcu(&child->mnt_hash, &mnt->mnt_hash);
388 }
Ram Paia05964f2005-11-07 17:20:17 -0500389 }
390}
391
392/*
393 * collect all mounts that receive propagation from the mount in @list,
394 * and return these additional mounts in the same list.
395 * @list: the list of mounts to be unmounted.
Nick Piggin99b7db72010-08-18 04:37:39 +1000396 *
397 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500398 */
Al Viro38129a12014-03-20 21:10:51 -0400399int propagate_umount(struct hlist_head *list)
Ram Paia05964f2005-11-07 17:20:17 -0500400{
Al Viro61ef47b2011-11-24 18:25:28 -0500401 struct mount *mnt;
Ram Paia05964f2005-11-07 17:20:17 -0500402
Al Viro38129a12014-03-20 21:10:51 -0400403 hlist_for_each_entry(mnt, list, mnt_hash)
Ram Paia05964f2005-11-07 17:20:17 -0500404 __propagate_umount(mnt);
405 return 0;
406}