]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/pnode.c
Merge branch 'for-linus' of git://git.kernel.dk/linux-block
[mirror_ubuntu-bionic-kernel.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>
132c94e3 12#include <linux/nsproxy.h>
6d59e7f5 13#include "internal.h"
07b20889
RP
14#include "pnode.h"
15
03e06e68 16/* return the next shared peer mount of @p */
c937135d 17static inline struct mount *next_peer(struct mount *p)
03e06e68 18{
6776db3d 19 return list_entry(p->mnt_share.next, struct mount, mnt_share);
03e06e68
RP
20}
21
c937135d 22static inline struct mount *first_slave(struct mount *p)
5afe0022 23{
6776db3d 24 return list_entry(p->mnt_slave_list.next, struct mount, mnt_slave);
5afe0022
RP
25}
26
c937135d 27static inline struct mount *next_slave(struct mount *p)
5afe0022 28{
6776db3d 29 return list_entry(p->mnt_slave.next, struct mount, mnt_slave);
5afe0022
RP
30}
31
6fc7871f
AV
32static struct mount *get_peer_under_root(struct mount *mnt,
33 struct mnt_namespace *ns,
34 const struct path *root)
97e7e0f7 35{
6fc7871f 36 struct mount *m = mnt;
97e7e0f7
MS
37
38 do {
39 /* Check the namespace first for optimization */
143c8c91 40 if (m->mnt_ns == ns && is_path_reachable(m, m->mnt.mnt_root, root))
6fc7871f 41 return m;
97e7e0f7 42
c937135d 43 m = next_peer(m);
6fc7871f 44 } while (m != mnt);
97e7e0f7
MS
45
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 */
6fc7871f 55int get_dominating_id(struct mount *mnt, const struct path *root)
97e7e0f7 56{
6fc7871f 57 struct mount *m;
97e7e0f7 58
32301920 59 for (m = mnt->mnt_master; m != NULL; m = m->mnt_master) {
143c8c91 60 struct mount *d = get_peer_under_root(m, mnt->mnt_ns, root);
97e7e0f7 61 if (d)
15169fe7 62 return d->mnt_group_id;
97e7e0f7
MS
63 }
64
65 return 0;
66}
67
6fc7871f 68static int do_make_slave(struct mount *mnt)
a58b0eb8 69{
32301920 70 struct mount *peer_mnt = mnt, *master = mnt->mnt_master;
d10e8def 71 struct mount *slave_mnt;
a58b0eb8
RP
72
73 /*
74 * slave 'mnt' to a peer mount that has the
796a6b52 75 * same root dentry. If none is available then
a58b0eb8
RP
76 * slave it to anything that is available.
77 */
c937135d 78 while ((peer_mnt = next_peer(peer_mnt)) != mnt &&
6fc7871f 79 peer_mnt->mnt.mnt_root != mnt->mnt.mnt_root) ;
a58b0eb8
RP
80
81 if (peer_mnt == mnt) {
c937135d 82 peer_mnt = next_peer(mnt);
a58b0eb8
RP
83 if (peer_mnt == mnt)
84 peer_mnt = NULL;
85 }
5d477b60
TI
86 if (mnt->mnt_group_id && IS_MNT_SHARED(mnt) &&
87 list_empty(&mnt->mnt_share))
6fc7871f 88 mnt_release_group_id(mnt);
719f5d7f 89
6776db3d 90 list_del_init(&mnt->mnt_share);
15169fe7 91 mnt->mnt_group_id = 0;
a58b0eb8
RP
92
93 if (peer_mnt)
94 master = peer_mnt;
95
96 if (master) {
6776db3d 97 list_for_each_entry(slave_mnt, &mnt->mnt_slave_list, mnt_slave)
32301920 98 slave_mnt->mnt_master = master;
6776db3d
AV
99 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);
a58b0eb8 102 } else {
6776db3d 103 struct list_head *p = &mnt->mnt_slave_list;
a58b0eb8 104 while (!list_empty(p)) {
b5e61818 105 slave_mnt = list_first_entry(p,
6776db3d
AV
106 struct mount, mnt_slave);
107 list_del_init(&slave_mnt->mnt_slave);
a58b0eb8
RP
108 slave_mnt->mnt_master = NULL;
109 }
110 }
32301920 111 mnt->mnt_master = master;
fc7be130 112 CLEAR_MNT_SHARED(mnt);
a58b0eb8
RP
113 return 0;
114}
115
99b7db7b
NP
116/*
117 * vfsmount lock must be held for write
118 */
0f0afb1d 119void change_mnt_propagation(struct mount *mnt, int type)
07b20889 120{
03e06e68 121 if (type == MS_SHARED) {
b90fa9ae 122 set_mnt_shared(mnt);
a58b0eb8
RP
123 return;
124 }
6fc7871f 125 do_make_slave(mnt);
a58b0eb8 126 if (type != MS_SLAVE) {
6776db3d 127 list_del_init(&mnt->mnt_slave);
d10e8def 128 mnt->mnt_master = NULL;
9676f0c6 129 if (type == MS_UNBINDABLE)
0f0afb1d 130 mnt->mnt.mnt_flags |= MNT_UNBINDABLE;
0b03cfb2 131 else
0f0afb1d 132 mnt->mnt.mnt_flags &= ~MNT_UNBINDABLE;
03e06e68 133 }
07b20889 134}
b90fa9ae
RP
135
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
796a6b52
AV
140 *
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.
b90fa9ae 145 */
c937135d
AV
146static struct mount *propagation_next(struct mount *m,
147 struct mount *origin)
b90fa9ae 148{
5afe0022 149 /* are there any slaves of this mount? */
143c8c91 150 if (!IS_MNT_NEW(m) && !list_empty(&m->mnt_slave_list))
5afe0022
RP
151 return first_slave(m);
152
153 while (1) {
32301920 154 struct mount *master = m->mnt_master;
5afe0022 155
32301920 156 if (master == origin->mnt_master) {
c937135d
AV
157 struct mount *next = next_peer(m);
158 return (next == origin) ? NULL : next;
6776db3d 159 } else if (m->mnt_slave.next != &master->mnt_slave_list)
5afe0022
RP
160 return next_slave(m);
161
162 /* back at master */
163 m = master;
164 }
165}
166
f2ebb3a9 167static struct mount *next_group(struct mount *m, struct mount *origin)
5afe0022 168{
f2ebb3a9
AV
169 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;
181 }
182 /* 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;
5afe0022 196 }
f2ebb3a9 197}
5afe0022 198
f2ebb3a9
AV
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 }
796a6b52 232 }
f2ebb3a9
AV
233 type = CL_SLAVE;
234 /* beginning of peer group among the slaves? */
235 if (IS_MNT_SHARED(m))
236 type |= CL_MAKE_SHARED;
5afe0022 237 }
f2ebb3a9
AV
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);
8486a788 245 child->mnt.mnt_flags &= ~MNT_LOCKED;
f2ebb3a9
AV
246 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;
b90fa9ae
RP
256}
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 */
84d17192 271int propagate_mnt(struct mount *dest_mnt, struct mountpoint *dest_mp,
38129a13 272 struct mount *source_mnt, struct hlist_head *tree_list)
b90fa9ae 273{
f2ebb3a9 274 struct mount *m, *n;
b90fa9ae 275 int ret = 0;
132c94e3 276
f2ebb3a9
AV
277 /*
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;
288
289 /* 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)
b90fa9ae 293 goto out;
f2ebb3a9 294 }
b90fa9ae 295
f2ebb3a9
AV
296 /* 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);
b90fa9ae
RP
307 }
308out:
f2ebb3a9
AV
309 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);
b90fa9ae 314 }
f2ebb3a9 315 read_sequnlock_excl(&mount_lock);
b90fa9ae
RP
316 return ret;
317}
a05964f3
RP
318
319/*
320 * return true if the refcount is greater than count
321 */
1ab59738 322static inline int do_refcount_check(struct mount *mnt, int count)
a05964f3 323{
aba809cf 324 return mnt_get_count(mnt) > count;
a05964f3
RP
325}
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.
99b7db7b 334 *
b3e19d92 335 * vfsmount lock must be held for write
a05964f3 336 */
1ab59738 337int propagate_mount_busy(struct mount *mnt, int refcnt)
a05964f3 338{
c937135d 339 struct mount *m, *child;
0714a533 340 struct mount *parent = mnt->mnt_parent;
a05964f3
RP
341 int ret = 0;
342
0714a533 343 if (mnt == parent)
a05964f3
RP
344 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 */
6b41d536 351 if (!list_empty(&mnt->mnt_mounts) || do_refcount_check(mnt, refcnt))
a05964f3
RP
352 return 1;
353
c937135d
AV
354 for (m = propagation_next(parent, parent); m;
355 m = propagation_next(m, parent)) {
474279dc 356 child = __lookup_mnt_last(&m->mnt, mnt->mnt_mountpoint);
6b41d536 357 if (child && list_empty(&child->mnt_mounts) &&
1ab59738 358 (ret = do_refcount_check(child, 1)))
a05964f3
RP
359 break;
360 }
361 return ret;
362}
363
364/*
365 * NOTE: unmounting 'mnt' naturally propagates to all other mounts its
366 * parent propagates to.
367 */
61ef47b1 368static void __propagate_umount(struct mount *mnt)
a05964f3 369{
0714a533 370 struct mount *parent = mnt->mnt_parent;
c937135d 371 struct mount *m;
a05964f3 372
0714a533 373 BUG_ON(parent == mnt);
a05964f3 374
c937135d
AV
375 for (m = propagation_next(parent, parent); m;
376 m = propagation_next(m, parent)) {
a05964f3 377
474279dc
AV
378 struct mount *child = __lookup_mnt_last(&m->mnt,
379 mnt->mnt_mountpoint);
a05964f3
RP
380 /*
381 * umount the child only if the child has no
382 * other children
383 */
38129a13 384 if (child && list_empty(&child->mnt_mounts)) {
88b368f2 385 list_del_init(&child->mnt_child);
38129a13
AV
386 hlist_del_init_rcu(&child->mnt_hash);
387 hlist_add_before_rcu(&child->mnt_hash, &mnt->mnt_hash);
388 }
a05964f3
RP
389 }
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.
99b7db7b
NP
396 *
397 * vfsmount lock must be held for write
a05964f3 398 */
38129a13 399int propagate_umount(struct hlist_head *list)
a05964f3 400{
61ef47b1 401 struct mount *mnt;
a05964f3 402
38129a13 403 hlist_for_each_entry(mnt, list, mnt_hash)
a05964f3
RP
404 __propagate_umount(mnt);
405 return 0;
406}