mnt: release locks on error path in do_loopback
[linux-2.6-block.git] / fs / pnode.c
CommitLineData
07b20889
RP
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 */
6b3286ed 9#include <linux/mnt_namespace.h>
07b20889
RP
10#include <linux/mount.h>
11#include <linux/fs.h>
6d59e7f5 12#include "internal.h"
07b20889
RP
13#include "pnode.h"
14
03e06e68 15/* return the next shared peer mount of @p */
c937135d 16static inline struct mount *next_peer(struct mount *p)
03e06e68 17{
6776db3d 18 return list_entry(p->mnt_share.next, struct mount, mnt_share);
03e06e68
RP
19}
20
c937135d 21static inline struct mount *first_slave(struct mount *p)
5afe0022 22{
6776db3d 23 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
5afe0022
RP
24}
25
c937135d 26static inline struct mount *next_slave(struct mount *p)
5afe0022 27{
6776db3d 28 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
5afe0022
RP
29}
30
6fc7871f
AV
31static struct mount *get_peer_under_root(struct mount *mnt,
32 struct mnt_namespace *ns,
33 const struct path *root)
97e7e0f7 34{
6fc7871f 35 struct mount *m = mnt;
97e7e0f7
MS
36
37 do {
38 /* Check the namespace first for optimization */
143c8c91 39 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
6fc7871f 40 return m;
97e7e0f7 41
c937135d 42 m = next_peer(m);
6fc7871f 43 } while (m != mnt);
97e7e0f7
MS
44
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 */
6fc7871f 54int get_dominating_id(struct mount *mnt, const struct path *root)
97e7e0f7 55{
6fc7871f 56 struct mount *m;
97e7e0f7 57
32301920 58 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
143c8c91 59 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
97e7e0f7 60 if (d)
15169fe7 61 return d->mnt_group_id;
97e7e0f7
MS
62 }
63
64 return 0;
65}
66
6fc7871f 67static int do_make_slave(struct mount *mnt)
a58b0eb8 68{
32301920 69 struct mount *peer_mnt = mnt, *master = mnt->mnt_master;
d10e8def 70 struct mount *slave_mnt;
a58b0eb8
RP
71
72 /*
73 * slave 'mnt' to a peer mount that has the
796a6b52 74 * same root dentry. If none is available then
a58b0eb8
RP
75 * slave it to anything that is available.
76 */
c937135d 77 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
6fc7871f 78 peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
a58b0eb8
RP
79
80 if (peer_mnt == mnt) {
c937135d 81 peer_mnt = next_peer(mnt);
a58b0eb8
RP
82 if (peer_mnt == mnt)
83 peer_mnt = NULL;
84 }
fc7be130 85 if (IS_MNT_SHARED(mnt) && list_empty(&mnt->mnt_share))
6fc7871f 86 mnt_release_group_id(mnt);
719f5d7f 87
6776db3d 88 list_del_init(&mnt->mnt_share);
15169fe7 89 mnt->mnt_group_id = 0;
a58b0eb8
RP
90
91 if (peer_mnt)
92 master = peer_mnt;
93
94 if (master) {
6776db3d 95 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
32301920 96 slave_mnt->mnt_master = master;
6776db3d
AV
97 list_move(&mnt->mnt_slave, &master->mnt_slave_list);
98 list_splice(&mnt->mnt_slave_list, master->mnt_slave_list.prev);
99 INIT_LIST_HEAD(&mnt->mnt_slave_list);
a58b0eb8 100 } else {
6776db3d 101 struct list_head *p = &mnt->mnt_slave_list;
a58b0eb8 102 while (!list_empty(p)) {
b5e61818 103 slave_mnt = list_first_entry(p,
6776db3d
AV
104 struct mount, mnt_slave);
105 list_del_init(&slave_mnt->mnt_slave);
a58b0eb8
RP
106 slave_mnt->mnt_master = NULL;
107 }
108 }
32301920 109 mnt->mnt_master = master;
fc7be130 110 CLEAR_MNT_SHARED(mnt);
a58b0eb8
RP
111 return 0;
112}
113
99b7db7b
NP
114/*
115 * vfsmount lock must be held for write
116 */
0f0afb1d 117void change_mnt_propagation(struct mount *mnt, int type)
07b20889 118{
03e06e68 119 if (type == MS_SHARED) {
b90fa9ae 120 set_mnt_shared(mnt);
a58b0eb8
RP
121 return;
122 }
6fc7871f 123 do_make_slave(mnt);
a58b0eb8 124 if (type != MS_SLAVE) {
6776db3d 125 list_del_init(&mnt->mnt_slave);
d10e8def 126 mnt->mnt_master = NULL;
9676f0c6 127 if (type == MS_UNBINDABLE)
0f0afb1d 128 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
0b03cfb2 129 else
0f0afb1d 130 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
03e06e68 131 }
07b20889 132}
b90fa9ae
RP
133
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
796a6b52
AV
138 *
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.
b90fa9ae 143 */
c937135d
AV
144static struct mount *propagation_next(struct mount *m,
145 struct mount *origin)
b90fa9ae 146{
5afe0022 147 /* are there any slaves of this mount? */
143c8c91 148 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
5afe0022
RP
149 return first_slave(m);
150
151 while (1) {
32301920 152 struct mount *master = m->mnt_master;
5afe0022 153
32301920 154 if (master == origin->mnt_master) {
c937135d
AV
155 struct mount *next = next_peer(m);
156 return (next == origin) ? NULL : next;
6776db3d 157 } else if (m->mnt_slave.next != &master->mnt_slave_list)
5afe0022
RP
158 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 */
c937135d
AV
174static struct mount *get_source(struct mount *dest,
175 struct mount *last_dest,
176 struct mount *last_src,
177 int *type)
5afe0022 178{
c937135d
AV
179 struct mount *p_last_src = NULL;
180 struct mount *p_last_dest = NULL;
5afe0022 181
32301920 182 while (last_dest != dest->mnt_master) {
5afe0022
RP
183 p_last_dest = last_dest;
184 p_last_src = last_src;
32301920
AV
185 last_dest = last_dest->mnt_master;
186 last_src = last_src->mnt_master;
5afe0022
RP
187 }
188
189 if (p_last_dest) {
190 do {
191 p_last_dest = next_peer(p_last_dest);
143c8c91 192 } while (IS_MNT_NEW(p_last_dest));
796a6b52
AV
193 /* is that a peer of the earlier? */
194 if (dest == p_last_dest) {
195 *type = CL_MAKE_SHARED;
196 return p_last_src;
197 }
5afe0022 198 }
796a6b52
AV
199 /* slave of the earlier, then */
200 *type = CL_SLAVE;
201 /* beginning of peer group among the slaves? */
fc7be130 202 if (IS_MNT_SHARED(dest))
796a6b52
AV
203 *type |= CL_MAKE_SHARED;
204 return last_src;
b90fa9ae
RP
205}
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 */
a8d56d8e
AV
220int propagate_mnt(struct mount *dest_mnt, struct dentry *dest_dentry,
221 struct mount *source_mnt, struct list_head *tree_list)
b90fa9ae 222{
c937135d 223 struct mount *m, *child;
b90fa9ae 224 int ret = 0;
a8d56d8e
AV
225 struct mount *prev_dest_mnt = dest_mnt;
226 struct mount *prev_src_mnt = source_mnt;
b90fa9ae
RP
227 LIST_HEAD(tmp_list);
228 LIST_HEAD(umount_list);
229
a8d56d8e
AV
230 for (m = propagation_next(dest_mnt, dest_mnt); m;
231 m = propagation_next(m, dest_mnt)) {
5afe0022 232 int type;
c937135d 233 struct mount *source;
b90fa9ae 234
143c8c91 235 if (IS_MNT_NEW(m))
b90fa9ae
RP
236 continue;
237
5afe0022 238 source = get_source(m, prev_dest_mnt, prev_src_mnt, &type);
b90fa9ae 239
be34d1a3
DH
240 child = copy_tree(source, source->mnt.mnt_root, type);
241 if (IS_ERR(child)) {
242 ret = PTR_ERR(child);
b90fa9ae
RP
243 list_splice(tree_list, tmp_list.prev);
244 goto out;
245 }
246
c937135d 247 if (is_subdir(dest_dentry, m->mnt.mnt_root)) {
14cf1fa8 248 mnt_set_mountpoint(m, dest_dentry, child);
1b8e5564 249 list_add_tail(&child->mnt_hash, tree_list);
b90fa9ae
RP
250 } else {
251 /*
252 * This can happen if the parent mount was bind mounted
253 * on some subdirectory of a shared/slave mount.
254 */
1b8e5564 255 list_add_tail(&child->mnt_hash, &tmp_list);
b90fa9ae
RP
256 }
257 prev_dest_mnt = m;
c937135d 258 prev_src_mnt = child;
b90fa9ae
RP
259 }
260out:
962830df 261 br_write_lock(&vfsmount_lock);
b90fa9ae 262 while (!list_empty(&tmp_list)) {
1b8e5564 263 child = list_first_entry(&tmp_list, struct mount, mnt_hash);
761d5c38 264 umount_tree(child, 0, &umount_list);
b90fa9ae 265 }
962830df 266 br_write_unlock(&vfsmount_lock);
b90fa9ae
RP
267 release_mounts(&umount_list);
268 return ret;
269}
a05964f3
RP
270
271/*
272 * return true if the refcount is greater than count
273 */
1ab59738 274static inline int do_refcount_check(struct mount *mnt, int count)
a05964f3 275{
863d684f 276 int mycount = mnt_get_count(mnt) - mnt->mnt_ghosts;
a05964f3
RP
277 return (mycount > count);
278}
279
280/*
281 * check if the mount 'mnt' can be unmounted successfully.
282 * @mnt: the mount to be checked for unmount
283 * NOTE: unmounting 'mnt' would naturally propagate to all
284 * other mounts its parent propagates to.
285 * Check if any of these mounts that **do not have submounts**
286 * have more references than 'refcnt'. If so return busy.
99b7db7b 287 *
b3e19d92 288 * vfsmount lock must be held for write
a05964f3 289 */
1ab59738 290int propagate_mount_busy(struct mount *mnt, int refcnt)
a05964f3 291{
c937135d 292 struct mount *m, *child;
0714a533 293 struct mount *parent = mnt->mnt_parent;
a05964f3
RP
294 int ret = 0;
295
0714a533 296 if (mnt == parent)
a05964f3
RP
297 return do_refcount_check(mnt, refcnt);
298
299 /*
300 * quickly check if the current mount can be unmounted.
301 * If not, we don't have to go checking for all other
302 * mounts
303 */
6b41d536 304 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
a05964f3
RP
305 return 1;
306
c937135d
AV
307 for (m = propagation_next(parent, parent); m;
308 m = propagation_next(m, parent)) {
309 child = __lookup_mnt(&m->mnt, mnt->mnt_mountpoint, 0);
6b41d536 310 if (child && list_empty(&child->mnt_mounts) &&
1ab59738 311 (ret = do_refcount_check(child, 1)))
a05964f3
RP
312 break;
313 }
314 return ret;
315}
316
317/*
318 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
319 * parent propagates to.
320 */
61ef47b1 321static void __propagate_umount(struct mount *mnt)
a05964f3 322{
0714a533 323 struct mount *parent = mnt->mnt_parent;
c937135d 324 struct mount *m;
a05964f3 325
0714a533 326 BUG_ON(parent == mnt);
a05964f3 327
c937135d
AV
328 for (m = propagation_next(parent, parent); m;
329 m = propagation_next(m, parent)) {
a05964f3 330
c937135d 331 struct mount *child = __lookup_mnt(&m->mnt,
a73324da 332 mnt->mnt_mountpoint, 0);
a05964f3
RP
333 /*
334 * umount the child only if the child has no
335 * other children
336 */
6b41d536 337 if (child && list_empty(&child->mnt_mounts))
1b8e5564 338 list_move_tail(&child->mnt_hash, &mnt->mnt_hash);
a05964f3
RP
339 }
340}
341
342/*
343 * collect all mounts that receive propagation from the mount in @list,
344 * and return these additional mounts in the same list.
345 * @list: the list of mounts to be unmounted.
99b7db7b
NP
346 *
347 * vfsmount lock must be held for write
a05964f3
RP
348 */
349int propagate_umount(struct list_head *list)
350{
61ef47b1 351 struct mount *mnt;
a05964f3 352
1b8e5564 353 list_for_each_entry(mnt, list, mnt_hash)
a05964f3
RP
354 __propagate_umount(mnt);
355 return 0;
356}