blob: bd280200bd379f7825eff5ac1342251cf81a0663 [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 +010031static struct vfsmount *get_peer_under_root(struct vfsmount *mnt,
32 struct mnt_namespace *ns,
33 const struct path *root)
34{
Al Viro643822b2011-11-24 22:00:28 -050035 struct mount *m = real_mount(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))
40 return &m->mnt;
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +010041
Al Viro643822b2011-11-24 22:00:28 -050042 m = real_mount(next_peer(&m->mnt));
43 } while (&m->mnt != 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 */
54int get_dominating_id(struct vfsmount *mnt, const struct path *root)
55{
56 struct vfsmount *m;
57
58 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
59 struct vfsmount *d = get_peer_under_root(m, mnt->mnt_ns, root);
60 if (d)
61 return d->mnt_group_id;
62 }
63
64 return 0;
65}
66
Ram Paia58b0eb2005-11-07 17:20:48 -050067static int do_make_slave(struct vfsmount *mnt)
68{
69 struct vfsmount *peer_mnt = mnt, *master = mnt->mnt_master;
70 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 */
77 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
78 peer_mnt->mnt_root != mnt->mnt_root) ;
79
80 if (peer_mnt == mnt) {
81 peer_mnt = next_peer(mnt);
82 if (peer_mnt == mnt)
83 peer_mnt = NULL;
84 }
Miklos Szeredi719f5d72008-03-27 13:06:23 +010085 if (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
Al Viro4b8b21f2011-11-24 19:54:23 -050086 mnt_release_group_id(real_mount(mnt));
Miklos Szeredi719f5d72008-03-27 13:06:23 +010087
Ram Paia58b0eb2005-11-07 17:20:48 -050088 list_del_init(&mnt->mnt_share);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010089 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) {
95 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
96 slave_mnt->mnt_master = master;
Akinobu Mita1bfba4e2006-06-26 00:24:40 -070097 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
Ram Paia58b0eb2005-11-07 17:20:48 -050098 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
99 INIT_LIST_HEAD(&mnt->mnt_slave_list);
100 } else {
101 struct list_head *p = &mnt->mnt_slave_list;
102 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 }
109 mnt->mnt_master = master;
110 CLEAR_MNT_SHARED(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 Viro0f0afb12011-11-24 20:43:10 -0500123 do_make_slave(&mnt->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 */
144static struct vfsmount *propagation_next(struct vfsmount *m,
145 struct vfsmount *origin)
146{
Ram Pai5afe0022005-11-07 17:21:01 -0500147 /* are there any slaves of this mount? */
148 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
149 return first_slave(m);
150
151 while (1) {
152 struct vfsmount *next;
153 struct vfsmount *master = m->mnt_master;
154
Miklos Szeredibf066c72006-01-08 01:03:19 -0800155 if (master == origin->mnt_master) {
Ram Pai5afe0022005-11-07 17:21:01 -0500156 next = next_peer(m);
157 return ((next == origin) ? NULL : next);
158 } else if (m->mnt_slave.next != &master->mnt_slave_list)
159 return next_slave(m);
160
161 /* back at master */
162 m = master;
163 }
164}
165
166/*
167 * return the source mount to be used for cloning
168 *
169 * @dest the current destination mount
170 * @last_dest the last seen destination mount
171 * @last_src the last seen source mount
172 * @type return CL_SLAVE if the new mount has to be
173 * cloned as a slave.
174 */
175static struct vfsmount *get_source(struct vfsmount *dest,
176 struct vfsmount *last_dest,
177 struct vfsmount *last_src,
178 int *type)
179{
180 struct vfsmount *p_last_src = NULL;
181 struct vfsmount *p_last_dest = NULL;
Ram Pai5afe0022005-11-07 17:21:01 -0500182
183 while (last_dest != dest->mnt_master) {
184 p_last_dest = last_dest;
185 p_last_src = last_src;
186 last_dest = last_dest->mnt_master;
187 last_src = last_src->mnt_master;
188 }
189
190 if (p_last_dest) {
191 do {
192 p_last_dest = next_peer(p_last_dest);
193 } while (IS_MNT_NEW(p_last_dest));
Al Viro796a6b52010-01-16 13:28:47 -0500194 /* is that a peer of the earlier? */
195 if (dest == p_last_dest) {
196 *type = CL_MAKE_SHARED;
197 return p_last_src;
198 }
Ram Pai5afe0022005-11-07 17:21:01 -0500199 }
Al Viro796a6b52010-01-16 13:28:47 -0500200 /* slave of the earlier, then */
201 *type = CL_SLAVE;
202 /* beginning of peer group among the slaves? */
203 if (IS_MNT_SHARED(dest))
204 *type |= CL_MAKE_SHARED;
205 return last_src;
Ram Paib90fa9a2005-11-07 17:19:50 -0500206}
207
208/*
209 * mount 'source_mnt' under the destination 'dest_mnt' at
210 * dentry 'dest_dentry'. And propagate that mount to
211 * all the peer and slave mounts of 'dest_mnt'.
212 * Link all the new mounts into a propagation tree headed at
213 * source_mnt. Also link all the new mounts using ->mnt_list
214 * headed at source_mnt's ->mnt_list
215 *
216 * @dest_mnt: destination mount.
217 * @dest_dentry: destination dentry.
218 * @source_mnt: source mount.
219 * @tree_list : list of heads of trees to be attached.
220 */
221int propagate_mnt(struct vfsmount *dest_mnt, struct dentry *dest_dentry,
222 struct vfsmount *source_mnt, struct list_head *tree_list)
223{
Al Virocb338d02011-11-24 20:55:08 -0500224 struct vfsmount *m;
225 struct mount *child;
Ram Paib90fa9a2005-11-07 17:19:50 -0500226 int ret = 0;
227 struct vfsmount *prev_dest_mnt = dest_mnt;
228 struct vfsmount *prev_src_mnt = source_mnt;
229 LIST_HEAD(tmp_list);
230 LIST_HEAD(umount_list);
231
232 for (m = propagation_next(dest_mnt, dest_mnt); m;
233 m = propagation_next(m, dest_mnt)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500234 int type;
235 struct vfsmount *source;
Ram Paib90fa9a2005-11-07 17:19:50 -0500236
237 if (IS_MNT_NEW(m))
238 continue;
239
Ram Pai5afe0022005-11-07 17:21:01 -0500240 source = get_source(m, prev_dest_mnt, prev_src_mnt, &type);
Ram Paib90fa9a2005-11-07 17:19:50 -0500241
Al Viro87129cc2011-11-24 21:24:27 -0500242 if (!(child = copy_tree(real_mount(source), source->mnt_root, type))) {
Ram Paib90fa9a2005-11-07 17:19:50 -0500243 ret = -ENOMEM;
244 list_splice(tree_list, tmp_list.prev);
245 goto out;
246 }
247
248 if (is_subdir(dest_dentry, m->mnt_root)) {
Al Viro44d964d62011-11-24 21:28:22 -0500249 mnt_set_mountpoint(m, dest_dentry, child);
Al Viro1b8e5562011-11-24 21:01:32 -0500250 list_add_tail(&child->mnt_hash, tree_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500251 } else {
252 /*
253 * This can happen if the parent mount was bind mounted
254 * on some subdirectory of a shared/slave mount.
255 */
Al Viro1b8e5562011-11-24 21:01:32 -0500256 list_add_tail(&child->mnt_hash, &tmp_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500257 }
258 prev_dest_mnt = m;
Al Virocb338d02011-11-24 20:55:08 -0500259 prev_src_mnt = &child->mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500260 }
261out:
Nick Piggin99b7db72010-08-18 04:37:39 +1000262 br_write_lock(vfsmount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500263 while (!list_empty(&tmp_list)) {
Al Viro1b8e5562011-11-24 21:01:32 -0500264 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
Al Viro761d5c32011-11-24 21:07:43 -0500265 umount_tree(child, 0, &umount_list);
Ram Paib90fa9a2005-11-07 17:19:50 -0500266 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000267 br_write_unlock(vfsmount_lock);
Ram Paib90fa9a2005-11-07 17:19:50 -0500268 release_mounts(&umount_list);
269 return ret;
270}
Ram Paia05964f2005-11-07 17:20:17 -0500271
272/*
273 * return true if the refcount is greater than count
274 */
Al Viro1ab59732011-11-24 21:35:16 -0500275static inline int do_refcount_check(struct mount *mnt, int count)
Ram Paia05964f2005-11-07 17:20:17 -0500276{
Al Viro1ab59732011-11-24 21:35:16 -0500277 int mycount = mnt_get_count(&mnt->mnt) - mnt->mnt.mnt_ghosts;
Ram Paia05964f2005-11-07 17:20:17 -0500278 return (mycount > count);
279}
280
281/*
282 * check if the mount 'mnt' can be unmounted successfully.
283 * @mnt: the mount to be checked for unmount
284 * NOTE: unmounting 'mnt' would naturally propagate to all
285 * other mounts its parent propagates to.
286 * Check if any of these mounts that **do not have submounts**
287 * have more references than 'refcnt'. If so return busy.
Nick Piggin99b7db72010-08-18 04:37:39 +1000288 *
Nick Pigginb3e19d92011-01-07 17:50:11 +1100289 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500290 */
Al Viro1ab59732011-11-24 21:35:16 -0500291int propagate_mount_busy(struct mount *mnt, int refcnt)
Ram Paia05964f2005-11-07 17:20:17 -0500292{
Al Viroc7105362011-11-24 18:22:03 -0500293 struct vfsmount *m;
294 struct mount *child;
Al Viro0714a532011-11-24 22:19:58 -0500295 struct mount *parent = mnt->mnt_parent;
Ram Paia05964f2005-11-07 17:20:17 -0500296 int ret = 0;
297
Al Viro0714a532011-11-24 22:19:58 -0500298 if (mnt == parent)
Ram Paia05964f2005-11-07 17:20:17 -0500299 return do_refcount_check(mnt, refcnt);
300
301 /*
302 * quickly check if the current mount can be unmounted.
303 * If not, we don't have to go checking for all other
304 * mounts
305 */
Al Viro1ab59732011-11-24 21:35:16 -0500306 if (!list_empty(&mnt->mnt.mnt_mounts) || do_refcount_check(mnt, refcnt))
Ram Paia05964f2005-11-07 17:20:17 -0500307 return 1;
308
Al Viro0714a532011-11-24 22:19:58 -0500309 for (m = propagation_next(&parent->mnt, &parent->mnt); m;
310 m = propagation_next(m, &parent->mnt)) {
Al Viroa73324d2011-11-24 22:25:07 -0500311 child = __lookup_mnt(m, mnt->mnt_mountpoint, 0);
Al Viroc7105362011-11-24 18:22:03 -0500312 if (child && list_empty(&child->mnt.mnt_mounts) &&
Al Viro1ab59732011-11-24 21:35:16 -0500313 (ret = do_refcount_check(child, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500314 break;
315 }
316 return ret;
317}
318
319/*
320 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
321 * parent propagates to.
322 */
Al Viro61ef47b2011-11-24 18:25:28 -0500323static void __propagate_umount(struct mount *mnt)
Ram Paia05964f2005-11-07 17:20:17 -0500324{
Al Viro0714a532011-11-24 22:19:58 -0500325 struct mount *parent = mnt->mnt_parent;
Ram Paia05964f2005-11-07 17:20:17 -0500326 struct vfsmount *m;
327
Al Viro0714a532011-11-24 22:19:58 -0500328 BUG_ON(parent == mnt);
Ram Paia05964f2005-11-07 17:20:17 -0500329
Al Viro0714a532011-11-24 22:19:58 -0500330 for (m = propagation_next(&parent->mnt, &parent->mnt); m;
331 m = propagation_next(m, &parent->mnt)) {
Ram Paia05964f2005-11-07 17:20:17 -0500332
Al Viroc7105362011-11-24 18:22:03 -0500333 struct mount *child = __lookup_mnt(m,
Al Viroa73324d2011-11-24 22:25:07 -0500334 mnt->mnt_mountpoint, 0);
Ram Paia05964f2005-11-07 17:20:17 -0500335 /*
336 * umount the child only if the child has no
337 * other children
338 */
Al Viroc7105362011-11-24 18:22:03 -0500339 if (child && list_empty(&child->mnt.mnt_mounts))
Al Viro1b8e5562011-11-24 21:01:32 -0500340 list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
Ram Paia05964f2005-11-07 17:20:17 -0500341 }
342}
343
344/*
345 * collect all mounts that receive propagation from the mount in @list,
346 * and return these additional mounts in the same list.
347 * @list: the list of mounts to be unmounted.
Nick Piggin99b7db72010-08-18 04:37:39 +1000348 *
349 * vfsmount lock must be held for write
Ram Paia05964f2005-11-07 17:20:17 -0500350 */
351int propagate_umount(struct list_head *list)
352{
Al Viro61ef47b2011-11-24 18:25:28 -0500353 struct mount *mnt;
Ram Paia05964f2005-11-07 17:20:17 -0500354
Al Viro1b8e5562011-11-24 21:01:32 -0500355 list_for_each_entry(mnt, list, mnt_hash)
Ram Paia05964f2005-11-07 17:20:17 -0500356 __propagate_umount(mnt);
357 return 0;
358}