blob: 7953c96a20715a981032b4d7af18e428f39b3ca6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/quotaops.h>
18#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070021#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080023#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/namei.h>
25#include <linux/security.h>
26#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
30#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050031#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070032#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Eric Dumazet13f14b42008-02-06 01:37:57 -080034#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35#define HASH_SIZE (1UL << HASH_SHIFT)
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050038__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
39
40static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Eric Dumazetfa3536c2006-03-26 01:37:24 -080042static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080043static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050044static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080046/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060047struct kobject *fs_kobj;
48EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
51{
Ram Paib58fed82005-11-07 17:16:09 -050052 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
53 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054 tmp = tmp + (tmp >> HASH_SHIFT);
55 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056}
57
58struct vfsmount *alloc_vfsmnt(const char *name)
59{
Robert P. J. Dayc3762222007-02-10 01:45:03 -080060 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 if (mnt) {
Ram Paib58fed82005-11-07 17:16:09 -050062 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 INIT_LIST_HEAD(&mnt->mnt_hash);
64 INIT_LIST_HEAD(&mnt->mnt_child);
65 INIT_LIST_HEAD(&mnt->mnt_mounts);
66 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070067 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050068 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050069 INIT_LIST_HEAD(&mnt->mnt_slave_list);
70 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050072 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 char *newname = kmalloc(size, GFP_KERNEL);
74 if (newname) {
75 memcpy(newname, name, size);
76 mnt->mnt_devname = newname;
77 }
78 }
79 }
80 return mnt;
81}
82
David Howells454e2392006-06-23 02:02:57 -070083int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
84{
85 mnt->mnt_sb = sb;
86 mnt->mnt_root = dget(sb->s_root);
87 return 0;
88}
89
90EXPORT_SYMBOL(simple_set_mnt);
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092void free_vfsmnt(struct vfsmount *mnt)
93{
94 kfree(mnt->mnt_devname);
95 kmem_cache_free(mnt_cache, mnt);
96}
97
98/*
Ram Paia05964f2005-11-07 17:20:17 -050099 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Ram Paia05964f2005-11-07 17:20:17 -0500102struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
103 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Ram Paib58fed82005-11-07 17:16:09 -0500105 struct list_head *head = mount_hashtable + hash(mnt, dentry);
106 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct vfsmount *p, *found = NULL;
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500110 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 p = NULL;
112 if (tmp == head)
113 break;
114 p = list_entry(tmp, struct vfsmount, mnt_hash);
115 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500116 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 break;
118 }
119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return found;
121}
122
Ram Paia05964f2005-11-07 17:20:17 -0500123/*
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
126 */
127struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
128{
129 struct vfsmount *child_mnt;
130 spin_lock(&vfsmount_lock);
131 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 mntget(child_mnt);
133 spin_unlock(&vfsmount_lock);
134 return child_mnt;
135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137static inline int check_mnt(struct vfsmount *mnt)
138{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800139 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800142static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500143{
144 if (ns) {
145 ns->event = ++event;
146 wake_up_interruptible(&ns->poll);
147 }
148}
149
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800150static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500151{
152 if (ns && ns->event != event) {
153 ns->event = event;
154 wake_up_interruptible(&ns->poll);
155 }
156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
159{
Jan Blunck4ac91372008-02-14 19:34:32 -0800160 old_nd->path.dentry = mnt->mnt_mountpoint;
161 old_nd->path.mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 mnt->mnt_parent = mnt;
163 mnt->mnt_mountpoint = mnt->mnt_root;
164 list_del_init(&mnt->mnt_child);
165 list_del_init(&mnt->mnt_hash);
Jan Blunck4ac91372008-02-14 19:34:32 -0800166 old_nd->path.dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167}
168
Ram Paib90fa9a2005-11-07 17:19:50 -0500169void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
170 struct vfsmount *child_mnt)
171{
172 child_mnt->mnt_parent = mntget(mnt);
173 child_mnt->mnt_mountpoint = dget(dentry);
174 dentry->d_mounted++;
175}
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
178{
Jan Blunck4ac91372008-02-14 19:34:32 -0800179 mnt_set_mountpoint(nd->path.mnt, nd->path.dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Jan Blunck4ac91372008-02-14 19:34:32 -0800181 hash(nd->path.mnt, nd->path.dentry));
182 list_add_tail(&mnt->mnt_child, &nd->path.mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500183}
184
185/*
186 * the caller must hold vfsmount_lock
187 */
188static void commit_tree(struct vfsmount *mnt)
189{
190 struct vfsmount *parent = mnt->mnt_parent;
191 struct vfsmount *m;
192 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800193 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500194
195 BUG_ON(parent == mnt);
196
197 list_add_tail(&head, &mnt->mnt_list);
198 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800199 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500200 list_splice(&head, n->list.prev);
201
202 list_add_tail(&mnt->mnt_hash, mount_hashtable +
203 hash(parent, mnt->mnt_mountpoint));
204 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800205 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
208static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
209{
210 struct list_head *next = p->mnt_mounts.next;
211 if (next == &p->mnt_mounts) {
212 while (1) {
213 if (p == root)
214 return NULL;
215 next = p->mnt_child.next;
216 if (next != &p->mnt_parent->mnt_mounts)
217 break;
218 p = p->mnt_parent;
219 }
220 }
221 return list_entry(next, struct vfsmount, mnt_child);
222}
223
Ram Pai9676f0c2005-11-07 17:21:20 -0500224static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
225{
226 struct list_head *prev = p->mnt_mounts.prev;
227 while (prev != &p->mnt_mounts) {
228 p = list_entry(prev, struct vfsmount, mnt_child);
229 prev = p->mnt_mounts.prev;
230 }
231 return p;
232}
233
Ram Pai36341f62005-11-07 17:17:22 -0500234static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
235 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
237 struct super_block *sb = old->mnt_sb;
238 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
239
240 if (mnt) {
241 mnt->mnt_flags = old->mnt_flags;
242 atomic_inc(&sb->s_active);
243 mnt->mnt_sb = sb;
244 mnt->mnt_root = dget(root);
245 mnt->mnt_mountpoint = mnt->mnt_root;
246 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500247
Ram Pai5afe0022005-11-07 17:21:01 -0500248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400252 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
258 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500259 if (flag & CL_MAKE_SHARED)
260 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
271 return mnt;
272}
273
Al Viro7b7b1ac2005-11-07 17:13:39 -0500274static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct super_block *sb = mnt->mnt_sb;
277 dput(mnt->mnt_root);
278 free_vfsmnt(mnt);
279 deactivate_super(sb);
280}
281
Al Viro7b7b1ac2005-11-07 17:13:39 -0500282void mntput_no_expire(struct vfsmount *mnt)
283{
284repeat:
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
288 __mntput(mnt);
289 return;
290 }
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
292 mnt->mnt_pinned = 0;
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
296 goto repeat;
297 }
298}
299
300EXPORT_SYMBOL(mntput_no_expire);
301
302void mnt_pin(struct vfsmount *mnt)
303{
304 spin_lock(&vfsmount_lock);
305 mnt->mnt_pinned++;
306 spin_unlock(&vfsmount_lock);
307}
308
309EXPORT_SYMBOL(mnt_pin);
310
311void mnt_unpin(struct vfsmount *mnt)
312{
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
316 mnt->mnt_pinned--;
317 }
318 spin_unlock(&vfsmount_lock);
319}
320
321EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800323static inline void mangle(struct seq_file *m, const char *s)
324{
325 seq_escape(m, s, " \t\n\\");
326}
327
328/*
329 * Simple .show_options callback for filesystems which don't want to
330 * implement more complex mount option showing.
331 *
332 * See also save_mount_options().
333 */
334int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
335{
336 const char *options = mnt->mnt_sb->s_options;
337
338 if (options != NULL && options[0]) {
339 seq_putc(m, ',');
340 mangle(m, options);
341 }
342
343 return 0;
344}
345EXPORT_SYMBOL(generic_show_options);
346
347/*
348 * If filesystem uses generic_show_options(), this function should be
349 * called from the fill_super() callback.
350 *
351 * The .remount_fs callback usually needs to be handled in a special
352 * way, to make sure, that previous options are not overwritten if the
353 * remount fails.
354 *
355 * Also note, that if the filesystem's .remount_fs function doesn't
356 * reset all options to their default value, but changes only newly
357 * given options, then the displayed options will not reflect reality
358 * any more.
359 */
360void save_mount_options(struct super_block *sb, char *options)
361{
362 kfree(sb->s_options);
363 sb->s_options = kstrdup(options, GFP_KERNEL);
364}
365EXPORT_SYMBOL(save_mount_options);
366
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367/* iterator */
368static void *m_start(struct seq_file *m, loff_t *pos)
369{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800370 struct mnt_namespace *n = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371
Ram Pai390c6842005-11-07 17:17:51 -0500372 down_read(&namespace_sem);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700373 return seq_list_start(&n->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374}
375
376static void *m_next(struct seq_file *m, void *v, loff_t *pos)
377{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800378 struct mnt_namespace *n = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700379
380 return seq_list_next(v, &n->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381}
382
383static void m_stop(struct seq_file *m, void *v)
384{
Ram Pai390c6842005-11-07 17:17:51 -0500385 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386}
387
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388static int show_vfsmnt(struct seq_file *m, void *v)
389{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700390 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 int err = 0;
392 static struct proc_fs_info {
393 int flag;
394 char *str;
395 } fs_info[] = {
396 { MS_SYNCHRONOUS, ",sync" },
397 { MS_DIRSYNC, ",dirsync" },
398 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 { 0, NULL }
400 };
401 static struct proc_fs_info mnt_info[] = {
402 { MNT_NOSUID, ",nosuid" },
403 { MNT_NODEV, ",nodev" },
404 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800405 { MNT_NOATIME, ",noatime" },
406 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800407 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 { 0, NULL }
409 };
410 struct proc_fs_info *fs_infop;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800411 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
414 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800415 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 seq_putc(m, ' ');
417 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700418 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
419 seq_putc(m, '.');
420 mangle(m, mnt->mnt_sb->s_subtype);
421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
423 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
424 if (mnt->mnt_sb->s_flags & fs_infop->flag)
425 seq_puts(m, fs_infop->str);
426 }
427 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
428 if (mnt->mnt_flags & fs_infop->flag)
429 seq_puts(m, fs_infop->str);
430 }
431 if (mnt->mnt_sb->s_op->show_options)
432 err = mnt->mnt_sb->s_op->show_options(m, mnt);
433 seq_puts(m, " 0 0\n");
434 return err;
435}
436
437struct seq_operations mounts_op = {
438 .start = m_start,
439 .next = m_next,
440 .stop = m_stop,
441 .show = show_vfsmnt
442};
443
Chuck Leverb4629fe2006-03-20 13:44:12 -0500444static int show_vfsstat(struct seq_file *m, void *v)
445{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700446 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800447 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500448 int err = 0;
449
450 /* device */
451 if (mnt->mnt_devname) {
452 seq_puts(m, "device ");
453 mangle(m, mnt->mnt_devname);
454 } else
455 seq_puts(m, "no device");
456
457 /* mount point */
458 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800459 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500460 seq_putc(m, ' ');
461
462 /* file system type */
463 seq_puts(m, "with fstype ");
464 mangle(m, mnt->mnt_sb->s_type->name);
465
466 /* optional statistics */
467 if (mnt->mnt_sb->s_op->show_stats) {
468 seq_putc(m, ' ');
469 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
470 }
471
472 seq_putc(m, '\n');
473 return err;
474}
475
476struct seq_operations mountstats_op = {
477 .start = m_start,
478 .next = m_next,
479 .stop = m_stop,
480 .show = show_vfsstat,
481};
482
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483/**
484 * may_umount_tree - check if a mount tree is busy
485 * @mnt: root of mount tree
486 *
487 * This is called to check if a tree of mounts has any
488 * open files, pwds, chroots or sub mounts that are
489 * busy.
490 */
491int may_umount_tree(struct vfsmount *mnt)
492{
Ram Pai36341f62005-11-07 17:17:22 -0500493 int actual_refs = 0;
494 int minimum_refs = 0;
495 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496
497 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500498 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 actual_refs += atomic_read(&p->mnt_count);
500 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 }
502 spin_unlock(&vfsmount_lock);
503
504 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800505 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Ian Kente3474a82006-03-27 01:14:51 -0800507 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508}
509
510EXPORT_SYMBOL(may_umount_tree);
511
512/**
513 * may_umount - check if a mount point is busy
514 * @mnt: root of mount
515 *
516 * This is called to check if a mount point has any
517 * open files, pwds, chroots or sub mounts. If the
518 * mount has sub mounts this will return busy
519 * regardless of whether the sub mounts are busy.
520 *
521 * Doesn't take quota and stuff into account. IOW, in some cases it will
522 * give false negatives. The main reason why it's here is that we need
523 * a non-destructive way to look for easily umountable filesystems.
524 */
525int may_umount(struct vfsmount *mnt)
526{
Ian Kente3474a82006-03-27 01:14:51 -0800527 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500528 spin_lock(&vfsmount_lock);
529 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800530 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500531 spin_unlock(&vfsmount_lock);
532 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533}
534
535EXPORT_SYMBOL(may_umount);
536
Ram Paib90fa9a2005-11-07 17:19:50 -0500537void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500539 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800540 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700541 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500542 list_del_init(&mnt->mnt_hash);
543 if (mnt->mnt_parent != mnt) {
544 struct dentry *dentry;
545 struct vfsmount *m;
546 spin_lock(&vfsmount_lock);
547 dentry = mnt->mnt_mountpoint;
548 m = mnt->mnt_parent;
549 mnt->mnt_mountpoint = mnt->mnt_root;
550 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500552 dput(dentry);
553 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 }
555 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500556 }
557}
558
Ram Paia05964f2005-11-07 17:20:17 -0500559void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500560{
561 struct vfsmount *p;
562
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700563 for (p = mnt; p; p = next_mnt(p, mnt))
564 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500565
Ram Paia05964f2005-11-07 17:20:17 -0500566 if (propagate)
567 propagate_umount(kill);
568
Ram Pai70fbcdf2005-11-07 17:17:04 -0500569 list_for_each_entry(p, kill, mnt_hash) {
570 list_del_init(&p->mnt_expire);
571 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800572 __touch_mnt_namespace(p->mnt_ns);
573 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500574 list_del_init(&p->mnt_child);
575 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500576 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500577 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
579}
580
581static int do_umount(struct vfsmount *mnt, int flags)
582{
Ram Paib58fed82005-11-07 17:16:09 -0500583 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500585 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 retval = security_sb_umount(mnt, flags);
588 if (retval)
589 return retval;
590
591 /*
592 * Allow userspace to request a mountpoint be expired rather than
593 * unmounting unconditionally. Unmount only happens if:
594 * (1) the mark is already set (the mark is cleared by mntput())
595 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
596 */
597 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -0800598 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 flags & (MNT_FORCE | MNT_DETACH))
600 return -EINVAL;
601
602 if (atomic_read(&mnt->mnt_count) != 2)
603 return -EBUSY;
604
605 if (!xchg(&mnt->mnt_expiry_mark, 1))
606 return -EAGAIN;
607 }
608
609 /*
610 * If we may have to abort operations to get out of this
611 * mount, and they will themselves hold resources we must
612 * allow the fs to do things. In the Unix tradition of
613 * 'Gee thats tricky lets do it in userspace' the umount_begin
614 * might fail to complete on the first run through as other tasks
615 * must return, and the like. Thats for the mount program to worry
616 * about for the moment.
617 */
618
619 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400620 if (sb->s_op->umount_begin)
621 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 unlock_kernel();
623
624 /*
625 * No sense to grab the lock for this test, but test itself looks
626 * somewhat bogus. Suggestions for better replacement?
627 * Ho-hum... In principle, we might treat that as umount + switch
628 * to rootfs. GC would eventually take care of the old vfsmount.
629 * Actually it makes sense, especially if rootfs would contain a
630 * /reboot - static binary that would close all descriptors and
631 * call reboot(9). Then init(8) could umount root and exec /reboot.
632 */
Jan Blunck6ac08c32008-02-14 19:34:38 -0800633 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 /*
635 * Special case for "unmounting" root ...
636 * we just try to remount it readonly.
637 */
638 down_write(&sb->s_umount);
639 if (!(sb->s_flags & MS_RDONLY)) {
640 lock_kernel();
641 DQUOT_OFF(sb);
642 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
643 unlock_kernel();
644 }
645 up_write(&sb->s_umount);
646 return retval;
647 }
648
Ram Pai390c6842005-11-07 17:17:51 -0500649 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500651 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500654 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500656 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 retval = 0;
658 }
659 spin_unlock(&vfsmount_lock);
660 if (retval)
661 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500662 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500663 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 return retval;
665}
666
667/*
668 * Now umount can handle mount points as well as block devices.
669 * This is important for filesystems which use unnamed block devices.
670 *
671 * We now support a flag for forced unmount like the other 'big iron'
672 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
673 */
674
675asmlinkage long sys_umount(char __user * name, int flags)
676{
677 struct nameidata nd;
678 int retval;
679
680 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
681 if (retval)
682 goto out;
683 retval = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800684 if (nd.path.dentry != nd.path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 goto dput_and_out;
Jan Blunck4ac91372008-02-14 19:34:32 -0800686 if (!check_mnt(nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 goto dput_and_out;
688
689 retval = -EPERM;
690 if (!capable(CAP_SYS_ADMIN))
691 goto dput_and_out;
692
Jan Blunck4ac91372008-02-14 19:34:32 -0800693 retval = do_umount(nd.path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -0800695 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Jan Blunck4ac91372008-02-14 19:34:32 -0800696 dput(nd.path.dentry);
697 mntput_no_expire(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698out:
699 return retval;
700}
701
702#ifdef __ARCH_WANT_SYS_OLDUMOUNT
703
704/*
Ram Paib58fed82005-11-07 17:16:09 -0500705 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707asmlinkage long sys_oldumount(char __user * name)
708{
Ram Paib58fed82005-11-07 17:16:09 -0500709 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710}
711
712#endif
713
714static int mount_is_safe(struct nameidata *nd)
715{
716 if (capable(CAP_SYS_ADMIN))
717 return 0;
718 return -EPERM;
719#ifdef notyet
Jan Blunck4ac91372008-02-14 19:34:32 -0800720 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 return -EPERM;
Jan Blunck4ac91372008-02-14 19:34:32 -0800722 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
723 if (current->uid != nd->path.dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 return -EPERM;
725 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800726 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return -EPERM;
728 return 0;
729#endif
730}
731
Ram Paib58fed82005-11-07 17:16:09 -0500732static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733{
734 while (1) {
735 if (d == dentry)
736 return 1;
737 if (d == NULL || d == d->d_parent)
738 return 0;
739 d = d->d_parent;
740 }
741}
742
Ram Paib90fa9a2005-11-07 17:19:50 -0500743struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500744 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
746 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 struct nameidata nd;
748
Ram Pai9676f0c2005-11-07 17:21:20 -0500749 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
750 return NULL;
751
Ram Pai36341f62005-11-07 17:17:22 -0500752 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (!q)
754 goto Enomem;
755 q->mnt_mountpoint = mnt->mnt_mountpoint;
756
757 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700758 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
760 continue;
761
762 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500763 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
764 s = skip_mnt_tree(s);
765 continue;
766 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 while (p != s->mnt_parent) {
768 p = p->mnt_parent;
769 q = q->mnt_parent;
770 }
771 p = s;
Jan Blunck4ac91372008-02-14 19:34:32 -0800772 nd.path.mnt = q;
773 nd.path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500774 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 if (!q)
776 goto Enomem;
777 spin_lock(&vfsmount_lock);
778 list_add_tail(&q->mnt_list, &res->mnt_list);
779 attach_mnt(q, &nd);
780 spin_unlock(&vfsmount_lock);
781 }
782 }
783 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500784Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500786 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500788 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500790 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 }
792 return NULL;
793}
794
Al Viro8aec0802007-06-07 12:20:32 -0400795struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
796{
797 struct vfsmount *tree;
798 down_read(&namespace_sem);
799 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
800 up_read(&namespace_sem);
801 return tree;
802}
803
804void drop_collected_mounts(struct vfsmount *mnt)
805{
806 LIST_HEAD(umount_list);
807 down_read(&namespace_sem);
808 spin_lock(&vfsmount_lock);
809 umount_tree(mnt, 0, &umount_list);
810 spin_unlock(&vfsmount_lock);
811 up_read(&namespace_sem);
812 release_mounts(&umount_list);
813}
814
Ram Paib90fa9a2005-11-07 17:19:50 -0500815/*
816 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500817 * @nd : place the mount tree @source_mnt is attached
818 * @parent_nd : if non-null, detach the source_mnt from its parent and
819 * store the parent mount and mountpoint dentry.
820 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500821 *
822 * NOTE: in the table below explains the semantics when a source mount
823 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500824 * ---------------------------------------------------------------------------
825 * | BIND MOUNT OPERATION |
826 * |**************************************************************************
827 * | source-->| shared | private | slave | unbindable |
828 * | dest | | | | |
829 * | | | | | | |
830 * | v | | | | |
831 * |**************************************************************************
832 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
833 * | | | | | |
834 * |non-shared| shared (+) | private | slave (*) | invalid |
835 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500836 * A bind operation clones the source mount and mounts the clone on the
837 * destination mount.
838 *
839 * (++) the cloned mount is propagated to all the mounts in the propagation
840 * tree of the destination mount and the cloned mount is added to
841 * the peer group of the source mount.
842 * (+) the cloned mount is created under the destination mount and is marked
843 * as shared. The cloned mount is added to the peer group of the source
844 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500845 * (+++) the mount is propagated to all the mounts in the propagation tree
846 * of the destination mount and the cloned mount is made slave
847 * of the same master as that of the source mount. The cloned mount
848 * is marked as 'shared and slave'.
849 * (*) the cloned mount is made a slave of the same master as that of the
850 * source mount.
851 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500852 * ---------------------------------------------------------------------------
853 * | MOVE MOUNT OPERATION |
854 * |**************************************************************************
855 * | source-->| shared | private | slave | unbindable |
856 * | dest | | | | |
857 * | | | | | | |
858 * | v | | | | |
859 * |**************************************************************************
860 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
861 * | | | | | |
862 * |non-shared| shared (+*) | private | slave (*) | unbindable |
863 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500864 *
865 * (+) the mount is moved to the destination. And is then propagated to
866 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500867 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500868 * (+++) the mount is moved to the destination and is then propagated to
869 * all the mounts belonging to the destination mount's propagation tree.
870 * the mount is marked as 'shared and slave'.
871 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500872 *
873 * if the source mount is a tree, the operations explained above is
874 * applied to each mount in the tree.
875 * Must be called without spinlocks held, since this function can sleep
876 * in allocations.
877 */
878static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500879 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500880{
881 LIST_HEAD(tree_list);
Jan Blunck4ac91372008-02-14 19:34:32 -0800882 struct vfsmount *dest_mnt = nd->path.mnt;
883 struct dentry *dest_dentry = nd->path.dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -0500884 struct vfsmount *child, *p;
885
886 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
887 return -EINVAL;
888
889 if (IS_MNT_SHARED(dest_mnt)) {
890 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
891 set_mnt_shared(p);
892 }
893
894 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500895 if (parent_nd) {
896 detach_mnt(source_mnt, parent_nd);
897 attach_mnt(source_mnt, nd);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800898 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500899 } else {
900 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
901 commit_tree(source_mnt);
902 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500903
904 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
905 list_del_init(&child->mnt_hash);
906 commit_tree(child);
907 }
908 spin_unlock(&vfsmount_lock);
909 return 0;
910}
911
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
913{
914 int err;
915 if (mnt->mnt_sb->s_flags & MS_NOUSER)
916 return -EINVAL;
917
Jan Blunck4ac91372008-02-14 19:34:32 -0800918 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
920 return -ENOTDIR;
921
922 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800923 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
924 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 goto out_unlock;
926
927 err = security_sb_check_sb(mnt, nd);
928 if (err)
929 goto out_unlock;
930
931 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800932 if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -0500933 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -0800935 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (!err)
937 security_sb_post_addmount(mnt, nd);
938 return err;
939}
940
941/*
Ram Pai07b20882005-11-07 17:19:07 -0500942 * recursively change the type of the mountpoint.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800943 * noinline this do_mount helper to save do_mount stack space.
Ram Pai07b20882005-11-07 17:19:07 -0500944 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800945static noinline int do_change_type(struct nameidata *nd, int flag)
Ram Pai07b20882005-11-07 17:19:07 -0500946{
Jan Blunck4ac91372008-02-14 19:34:32 -0800947 struct vfsmount *m, *mnt = nd->path.mnt;
Ram Pai07b20882005-11-07 17:19:07 -0500948 int recurse = flag & MS_REC;
949 int type = flag & ~MS_REC;
950
Miklos Szerediee6f9582007-05-08 00:30:40 -0700951 if (!capable(CAP_SYS_ADMIN))
952 return -EPERM;
953
Jan Blunck4ac91372008-02-14 19:34:32 -0800954 if (nd->path.dentry != nd->path.mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -0500955 return -EINVAL;
956
957 down_write(&namespace_sem);
958 spin_lock(&vfsmount_lock);
959 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
960 change_mnt_propagation(m, type);
961 spin_unlock(&vfsmount_lock);
962 up_write(&namespace_sem);
963 return 0;
964}
965
966/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 * do loopback mount.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800968 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800970static noinline int do_loopback(struct nameidata *nd, char *old_name,
971 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972{
973 struct nameidata old_nd;
974 struct vfsmount *mnt = NULL;
975 int err = mount_is_safe(nd);
976 if (err)
977 return err;
978 if (!old_name || !*old_name)
979 return -EINVAL;
980 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
981 if (err)
982 return err;
983
Ram Pai390c6842005-11-07 17:17:51 -0500984 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800986 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
987 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -0500988
Jan Blunck4ac91372008-02-14 19:34:32 -0800989 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -0500990 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991
Al Viroccd48bc2005-11-07 17:15:04 -0500992 err = -ENOMEM;
993 if (recurse)
Jan Blunck4ac91372008-02-14 19:34:32 -0800994 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500995 else
Jan Blunck4ac91372008-02-14 19:34:32 -0800996 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500997
998 if (!mnt)
999 goto out;
1000
Al Viroccd48bc2005-11-07 17:15:04 -05001001 err = graft_tree(mnt, nd);
1002 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001003 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001005 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001007 release_mounts(&umount_list);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Al Viroccd48bc2005-11-07 17:15:04 -05001010out:
Ram Pai390c6842005-11-07 17:17:51 -05001011 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001012 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return err;
1014}
1015
1016/*
1017 * change filesystem flags. dir should be a physical root of filesystem.
1018 * If you've mounted a non-root directory somewhere and want to do remount
1019 * on it - tough luck.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001020 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001022static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 void *data)
1024{
1025 int err;
Jan Blunck4ac91372008-02-14 19:34:32 -08001026 struct super_block *sb = nd->path.mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028 if (!capable(CAP_SYS_ADMIN))
1029 return -EPERM;
1030
Jan Blunck4ac91372008-02-14 19:34:32 -08001031 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 return -EINVAL;
1033
Jan Blunck4ac91372008-02-14 19:34:32 -08001034 if (nd->path.dentry != nd->path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 return -EINVAL;
1036
1037 down_write(&sb->s_umount);
1038 err = do_remount_sb(sb, flags, data, 0);
1039 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001040 nd->path.mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 up_write(&sb->s_umount);
1042 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001043 security_sb_post_remount(nd->path.mnt, flags, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 return err;
1045}
1046
Ram Pai9676f0c2005-11-07 17:21:20 -05001047static inline int tree_contains_unbindable(struct vfsmount *mnt)
1048{
1049 struct vfsmount *p;
1050 for (p = mnt; p; p = next_mnt(p, mnt)) {
1051 if (IS_MNT_UNBINDABLE(p))
1052 return 1;
1053 }
1054 return 0;
1055}
1056
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001057/*
1058 * noinline this do_mount helper to save do_mount stack space.
1059 */
1060static noinline int do_move_mount(struct nameidata *nd, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
1062 struct nameidata old_nd, parent_nd;
1063 struct vfsmount *p;
1064 int err = 0;
1065 if (!capable(CAP_SYS_ADMIN))
1066 return -EPERM;
1067 if (!old_name || !*old_name)
1068 return -EINVAL;
1069 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1070 if (err)
1071 return err;
1072
Ram Pai390c6842005-11-07 17:17:51 -05001073 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001074 while (d_mountpoint(nd->path.dentry) &&
1075 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 ;
1077 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001078 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 goto out;
1080
1081 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001082 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1083 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 goto out1;
1085
Jan Blunck4ac91372008-02-14 19:34:32 -08001086 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05001087 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001090 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001091 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Jan Blunck4ac91372008-02-14 19:34:32 -08001093 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001094 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
Jan Blunck4ac91372008-02-14 19:34:32 -08001096 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1097 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001098 goto out1;
1099 /*
1100 * Don't move a mount residing in a shared parent.
1101 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001102 if (old_nd.path.mnt->mnt_parent &&
1103 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001104 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001105 /*
1106 * Don't move a mount tree containing unbindable mounts to a destination
1107 * mount which is shared.
1108 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001109 if (IS_MNT_SHARED(nd->path.mnt) &&
1110 tree_contains_unbindable(old_nd.path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001111 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 err = -ELOOP;
Jan Blunck4ac91372008-02-14 19:34:32 -08001113 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1114 if (p == old_nd.path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001115 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Jan Blunck4ac91372008-02-14 19:34:32 -08001117 err = attach_recursive_mnt(old_nd.path.mnt, nd, &parent_nd);
1118 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001119 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Ram Pai21444402005-11-07 17:20:03 -05001121 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 /* if the mount is moved, it should no longer be expire
1123 * automatically */
Jan Blunck4ac91372008-02-14 19:34:32 -08001124 list_del_init(&old_nd.path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 spin_unlock(&vfsmount_lock);
1126out1:
Jan Blunck4ac91372008-02-14 19:34:32 -08001127 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128out:
Ram Pai390c6842005-11-07 17:17:51 -05001129 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 if (!err)
Jan Blunck1d957f92008-02-14 19:34:35 -08001131 path_put(&parent_nd.path);
1132 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 return err;
1134}
1135
1136/*
1137 * create a new mount for userspace and request it to be added into the
1138 * namespace's tree
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001139 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001141static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 int mnt_flags, char *name, void *data)
1143{
1144 struct vfsmount *mnt;
1145
1146 if (!type || !memchr(type, 0, PAGE_SIZE))
1147 return -EINVAL;
1148
1149 /* we need capabilities... */
1150 if (!capable(CAP_SYS_ADMIN))
1151 return -EPERM;
1152
1153 mnt = do_kern_mount(type, flags, name, data);
1154 if (IS_ERR(mnt))
1155 return PTR_ERR(mnt);
1156
1157 return do_add_mount(mnt, nd, mnt_flags, NULL);
1158}
1159
1160/*
1161 * add a mount into a namespace's mount tree
1162 * - provide the option of adding the new mount to an expiration list
1163 */
1164int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1165 int mnt_flags, struct list_head *fslist)
1166{
1167 int err;
1168
Ram Pai390c6842005-11-07 17:17:51 -05001169 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /* Something was mounted here while we slept */
Jan Blunck4ac91372008-02-14 19:34:32 -08001171 while (d_mountpoint(nd->path.dentry) &&
1172 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 ;
1174 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001175 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 goto unlock;
1177
1178 /* Refuse the same filesystem on the same mount point */
1179 err = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001180 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1181 nd->path.mnt->mnt_root == nd->path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 goto unlock;
1183
1184 err = -EINVAL;
1185 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1186 goto unlock;
1187
1188 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c2005-11-07 17:16:29 -05001189 if ((err = graft_tree(newmnt, nd)))
1190 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Ram Pai5b83d2c2005-11-07 17:16:29 -05001192 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 /* add to the specified expiration list */
1194 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001195 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 spin_unlock(&vfsmount_lock);
1197 }
Ram Pai390c6842005-11-07 17:17:51 -05001198 up_write(&namespace_sem);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001199 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001202 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 mntput(newmnt);
1204 return err;
1205}
1206
1207EXPORT_SYMBOL_GPL(do_add_mount);
1208
Ram Pai70fbcdf2005-11-07 17:17:04 -05001209static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1210 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001211{
1212 spin_lock(&vfsmount_lock);
1213
1214 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001215 * Check if mount is still attached, if not, let whoever holds it deal
1216 * with the sucker
1217 */
1218 if (mnt->mnt_parent == mnt) {
1219 spin_unlock(&vfsmount_lock);
1220 return;
1221 }
1222
1223 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001224 * Check that it is still dead: the count should now be 2 - as
1225 * contributed by the vfsmount parent and the mntget above
1226 */
Ram Paia05964f2005-11-07 17:20:17 -05001227 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001228 /* delete from the namespace */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001229 touch_mnt_namespace(mnt->mnt_ns);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001230 list_del_init(&mnt->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001231 mnt->mnt_ns = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001232 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001233 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001234 } else {
1235 /*
1236 * Someone brought it back to life whilst we didn't have any
1237 * locks held so return it to the expiration list
1238 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001239 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001240 spin_unlock(&vfsmount_lock);
1241 }
1242}
1243
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001245 * go through the vfsmounts we've just consigned to the graveyard to
1246 * - check that they're still dead
1247 * - delete the vfsmount from the appropriate namespace under lock
1248 * - dispose of the corpse
1249 */
1250static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1251{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001252 struct mnt_namespace *ns;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001253 struct vfsmount *mnt;
1254
1255 while (!list_empty(graveyard)) {
1256 LIST_HEAD(umounts);
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001257 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001258 list_del_init(&mnt->mnt_expire);
1259
1260 /* don't do anything if the namespace is dead - all the
1261 * vfsmounts from it are going away anyway */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001262 ns = mnt->mnt_ns;
1263 if (!ns || !ns->root)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001264 continue;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001265 get_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001266
1267 spin_unlock(&vfsmount_lock);
1268 down_write(&namespace_sem);
1269 expire_mount(mnt, mounts, &umounts);
1270 up_write(&namespace_sem);
1271 release_mounts(&umounts);
1272 mntput(mnt);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001273 put_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001274 spin_lock(&vfsmount_lock);
1275 }
1276}
1277
1278/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 * process a list of expirable mountpoints with the intent of discarding any
1280 * mountpoints that aren't in use and haven't been touched since last we came
1281 * here
1282 */
1283void mark_mounts_for_expiry(struct list_head *mounts)
1284{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 struct vfsmount *mnt, *next;
1286 LIST_HEAD(graveyard);
1287
1288 if (list_empty(mounts))
1289 return;
1290
1291 spin_lock(&vfsmount_lock);
1292
1293 /* extract from the expiration list every vfsmount that matches the
1294 * following criteria:
1295 * - only referenced by its parent vfsmount
1296 * - still marked for expiry (marked on the last call here; marks are
1297 * cleared by mntput())
1298 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001299 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1301 atomic_read(&mnt->mnt_count) != 1)
1302 continue;
1303
1304 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001305 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 }
1307
Trond Myklebust5528f9112006-06-09 09:34:17 -04001308 expire_mount_list(&graveyard, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309
1310 spin_unlock(&vfsmount_lock);
1311}
1312
1313EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1314
1315/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001316 * Ripoff of 'select_parent()'
1317 *
1318 * search the list of submounts for a given mountpoint, and move any
1319 * shrinkable submounts to the 'graveyard' list.
1320 */
1321static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1322{
1323 struct vfsmount *this_parent = parent;
1324 struct list_head *next;
1325 int found = 0;
1326
1327repeat:
1328 next = this_parent->mnt_mounts.next;
1329resume:
1330 while (next != &this_parent->mnt_mounts) {
1331 struct list_head *tmp = next;
1332 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1333
1334 next = tmp->next;
1335 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1336 continue;
1337 /*
1338 * Descend a level if the d_mounts list is non-empty.
1339 */
1340 if (!list_empty(&mnt->mnt_mounts)) {
1341 this_parent = mnt;
1342 goto repeat;
1343 }
1344
1345 if (!propagate_mount_busy(mnt, 1)) {
1346 mntget(mnt);
1347 list_move_tail(&mnt->mnt_expire, graveyard);
1348 found++;
1349 }
1350 }
1351 /*
1352 * All done at this level ... ascend and resume the search
1353 */
1354 if (this_parent != parent) {
1355 next = this_parent->mnt_child.next;
1356 this_parent = this_parent->mnt_parent;
1357 goto resume;
1358 }
1359 return found;
1360}
1361
1362/*
1363 * process a list of expirable mountpoints with the intent of discarding any
1364 * submounts of a specific parent mountpoint
1365 */
1366void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1367{
1368 LIST_HEAD(graveyard);
1369 int found;
1370
1371 spin_lock(&vfsmount_lock);
1372
1373 /* extract submounts of 'mountpoint' from the expiration list */
1374 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1375 expire_mount_list(&graveyard, mounts);
1376
1377 spin_unlock(&vfsmount_lock);
1378}
1379
1380EXPORT_SYMBOL_GPL(shrink_submounts);
1381
1382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 * Some copy_from_user() implementations do not return the exact number of
1384 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1385 * Note that this function differs from copy_from_user() in that it will oops
1386 * on bad values of `to', rather than returning a short copy.
1387 */
Ram Paib58fed82005-11-07 17:16:09 -05001388static long exact_copy_from_user(void *to, const void __user * from,
1389 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390{
1391 char *t = to;
1392 const char __user *f = from;
1393 char c;
1394
1395 if (!access_ok(VERIFY_READ, from, n))
1396 return n;
1397
1398 while (n) {
1399 if (__get_user(c, f)) {
1400 memset(t, 0, n);
1401 break;
1402 }
1403 *t++ = c;
1404 f++;
1405 n--;
1406 }
1407 return n;
1408}
1409
Ram Paib58fed82005-11-07 17:16:09 -05001410int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
1412 int i;
1413 unsigned long page;
1414 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 *where = 0;
1417 if (!data)
1418 return 0;
1419
1420 if (!(page = __get_free_page(GFP_KERNEL)))
1421 return -ENOMEM;
1422
1423 /* We only care that *some* data at the address the user
1424 * gave us is valid. Just in case, we'll zero
1425 * the remainder of the page.
1426 */
1427 /* copy_from_user cannot cross TASK_SIZE ! */
1428 size = TASK_SIZE - (unsigned long)data;
1429 if (size > PAGE_SIZE)
1430 size = PAGE_SIZE;
1431
1432 i = size - exact_copy_from_user((void *)page, data, size);
1433 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001434 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 return -EFAULT;
1436 }
1437 if (i != PAGE_SIZE)
1438 memset((char *)page + i, 0, PAGE_SIZE - i);
1439 *where = page;
1440 return 0;
1441}
1442
1443/*
1444 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1445 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1446 *
1447 * data is a (void *) that can point to any structure up to
1448 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1449 * information (or be NULL).
1450 *
1451 * Pre-0.97 versions of mount() didn't have a flags word.
1452 * When the flags word was introduced its top half was required
1453 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1454 * Therefore, if this magic number is present, it carries no information
1455 * and must be discarded.
1456 */
Ram Paib58fed82005-11-07 17:16:09 -05001457long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 unsigned long flags, void *data_page)
1459{
1460 struct nameidata nd;
1461 int retval = 0;
1462 int mnt_flags = 0;
1463
1464 /* Discard magic */
1465 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1466 flags &= ~MS_MGC_MSK;
1467
1468 /* Basic sanity checks */
1469
1470 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1471 return -EINVAL;
1472 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1473 return -EINVAL;
1474
1475 if (data_page)
1476 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1477
1478 /* Separate the per-mountpoint flags */
1479 if (flags & MS_NOSUID)
1480 mnt_flags |= MNT_NOSUID;
1481 if (flags & MS_NODEV)
1482 mnt_flags |= MNT_NODEV;
1483 if (flags & MS_NOEXEC)
1484 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001485 if (flags & MS_NOATIME)
1486 mnt_flags |= MNT_NOATIME;
1487 if (flags & MS_NODIRATIME)
1488 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001489 if (flags & MS_RELATIME)
1490 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001491
1492 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001493 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495 /* ... and get the mountpoint */
1496 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1497 if (retval)
1498 return retval;
1499
1500 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1501 if (retval)
1502 goto dput_out;
1503
1504 if (flags & MS_REMOUNT)
1505 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1506 data_page);
1507 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001508 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001509 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001510 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 else if (flags & MS_MOVE)
1512 retval = do_move_mount(&nd, dev_name);
1513 else
1514 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1515 dev_name, data_page);
1516dput_out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001517 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 return retval;
1519}
1520
JANAK DESAI741a2952006-02-07 12:59:00 -08001521/*
1522 * Allocate a new namespace structure and populate it with contents
1523 * copied from the namespace of the passed in task structure.
1524 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001525static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001526 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001528 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 struct vfsmount *p, *q;
1531
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001532 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001534 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001538 init_waitqueue_head(&new_ns->poll);
1539 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540
Ram Pai390c6842005-11-07 17:17:51 -05001541 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001543 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001544 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001546 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001548 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 }
1550 spin_lock(&vfsmount_lock);
1551 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1552 spin_unlock(&vfsmount_lock);
1553
1554 /*
1555 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1556 * as belonging to new namespace. We have already acquired a private
1557 * fs_struct, so tsk->fs->lock is not needed.
1558 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001559 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 q = new_ns->root;
1561 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001562 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001564 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001566 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001568 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001570 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001572 if (p == fs->altroot.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 altrootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001574 fs->altroot.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 }
1576 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001577 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 q = next_mnt(q, new_ns->root);
1579 }
Ram Pai390c6842005-11-07 17:17:51 -05001580 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 if (rootmnt)
1583 mntput(rootmnt);
1584 if (pwdmnt)
1585 mntput(pwdmnt);
1586 if (altrootmnt)
1587 mntput(altrootmnt);
1588
JANAK DESAI741a2952006-02-07 12:59:00 -08001589 return new_ns;
1590}
1591
Eric W. Biederman213dd262007-07-15 23:41:15 -07001592struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07001593 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001594{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001595 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001596
Badari Pulavartye3222c42007-05-08 00:25:21 -07001597 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001598 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001599
1600 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001601 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001602
Badari Pulavartye3222c42007-05-08 00:25:21 -07001603 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001604
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001605 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001606 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607}
1608
1609asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1610 char __user * type, unsigned long flags,
1611 void __user * data)
1612{
1613 int retval;
1614 unsigned long data_page;
1615 unsigned long type_page;
1616 unsigned long dev_page;
1617 char *dir_page;
1618
Ram Paib58fed82005-11-07 17:16:09 -05001619 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 if (retval < 0)
1621 return retval;
1622
1623 dir_page = getname(dir_name);
1624 retval = PTR_ERR(dir_page);
1625 if (IS_ERR(dir_page))
1626 goto out1;
1627
Ram Paib58fed82005-11-07 17:16:09 -05001628 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 if (retval < 0)
1630 goto out2;
1631
Ram Paib58fed82005-11-07 17:16:09 -05001632 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 if (retval < 0)
1634 goto out3;
1635
1636 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001637 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1638 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 unlock_kernel();
1640 free_page(data_page);
1641
1642out3:
1643 free_page(dev_page);
1644out2:
1645 putname(dir_page);
1646out1:
1647 free_page(type_page);
1648 return retval;
1649}
1650
1651/*
1652 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1653 * It can block. Requires the big lock held.
1654 */
Jan Blunckac748a02008-02-14 19:34:39 -08001655void set_fs_root(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001657 struct path old_root;
1658
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 write_lock(&fs->lock);
1660 old_root = fs->root;
Jan Blunckac748a02008-02-14 19:34:39 -08001661 fs->root = *path;
1662 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 write_unlock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001664 if (old_root.dentry)
1665 path_put(&old_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666}
1667
1668/*
1669 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1670 * It can block. Requires the big lock held.
1671 */
Jan Blunckac748a02008-02-14 19:34:39 -08001672void set_fs_pwd(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001674 struct path old_pwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 write_lock(&fs->lock);
1677 old_pwd = fs->pwd;
Jan Blunckac748a02008-02-14 19:34:39 -08001678 fs->pwd = *path;
1679 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 write_unlock(&fs->lock);
1681
Jan Blunck6ac08c32008-02-14 19:34:38 -08001682 if (old_pwd.dentry)
1683 path_put(&old_pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684}
1685
1686static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1687{
1688 struct task_struct *g, *p;
1689 struct fs_struct *fs;
1690
1691 read_lock(&tasklist_lock);
1692 do_each_thread(g, p) {
1693 task_lock(p);
1694 fs = p->fs;
1695 if (fs) {
1696 atomic_inc(&fs->count);
1697 task_unlock(p);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001698 if (fs->root.dentry == old_nd->path.dentry
1699 && fs->root.mnt == old_nd->path.mnt)
Jan Blunckac748a02008-02-14 19:34:39 -08001700 set_fs_root(fs, &new_nd->path);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001701 if (fs->pwd.dentry == old_nd->path.dentry
1702 && fs->pwd.mnt == old_nd->path.mnt)
Jan Blunckac748a02008-02-14 19:34:39 -08001703 set_fs_pwd(fs, &new_nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 put_fs_struct(fs);
1705 } else
1706 task_unlock(p);
1707 } while_each_thread(g, p);
1708 read_unlock(&tasklist_lock);
1709}
1710
1711/*
1712 * pivot_root Semantics:
1713 * Moves the root file system of the current process to the directory put_old,
1714 * makes new_root as the new root file system of the current process, and sets
1715 * root/cwd of all processes which had them on the current root to new_root.
1716 *
1717 * Restrictions:
1718 * The new_root and put_old must be directories, and must not be on the
1719 * same file system as the current process root. The put_old must be
1720 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1721 * pointed to by put_old must yield the same directory as new_root. No other
1722 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1723 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001724 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1725 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1726 * in this situation.
1727 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 * Notes:
1729 * - we don't move root/cwd if they are not at the root (reason: if something
1730 * cared enough to change them, it's probably wrong to force them elsewhere)
1731 * - it's okay to pick a root that isn't the root of a file system, e.g.
1732 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1733 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1734 * first.
1735 */
Ram Paib58fed82005-11-07 17:16:09 -05001736asmlinkage long sys_pivot_root(const char __user * new_root,
1737 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738{
1739 struct vfsmount *tmp;
1740 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1741 int error;
1742
1743 if (!capable(CAP_SYS_ADMIN))
1744 return -EPERM;
1745
1746 lock_kernel();
1747
Ram Paib58fed82005-11-07 17:16:09 -05001748 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1749 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 if (error)
1751 goto out0;
1752 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001753 if (!check_mnt(new_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 goto out1;
1755
Ram Paib58fed82005-11-07 17:16:09 -05001756 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (error)
1758 goto out1;
1759
1760 error = security_sb_pivotroot(&old_nd, &new_nd);
1761 if (error) {
Jan Blunck1d957f92008-02-14 19:34:35 -08001762 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 goto out1;
1764 }
1765
1766 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001767 user_nd.path = current->fs->root;
1768 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001770 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001771 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001773 if (IS_MNT_SHARED(old_nd.path.mnt) ||
1774 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
1775 IS_MNT_SHARED(user_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001776 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001777 if (!check_mnt(user_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 goto out2;
1779 error = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001780 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001782 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001784 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 goto out2;
1786 error = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001787 if (new_nd.path.mnt == user_nd.path.mnt ||
1788 old_nd.path.mnt == user_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 goto out2; /* loop, on the same file system */
1790 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001791 if (user_nd.path.mnt->mnt_root != user_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001793 if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001794 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001795 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001797 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001798 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001799 /* make sure we can reach put_old from new_root */
1800 tmp = old_nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -08001802 if (tmp != new_nd.path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 for (;;) {
1804 if (tmp->mnt_parent == tmp)
1805 goto out3; /* already mounted on put_old */
Jan Blunck4ac91372008-02-14 19:34:32 -08001806 if (tmp->mnt_parent == new_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 break;
1808 tmp = tmp->mnt_parent;
1809 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001810 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08001812 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08001814 detach_mnt(new_nd.path.mnt, &parent_nd);
1815 detach_mnt(user_nd.path.mnt, &root_parent);
1816 /* mount old root on put_old */
1817 attach_mnt(user_nd.path.mnt, &old_nd);
1818 /* mount new_root on / */
1819 attach_mnt(new_nd.path.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001820 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 spin_unlock(&vfsmount_lock);
1822 chroot_fs_refs(&user_nd, &new_nd);
1823 security_sb_post_pivotroot(&user_nd, &new_nd);
1824 error = 0;
Jan Blunck1d957f92008-02-14 19:34:35 -08001825 path_put(&root_parent.path);
1826 path_put(&parent_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827out2:
Jan Blunck4ac91372008-02-14 19:34:32 -08001828 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001829 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001830 path_put(&user_nd.path);
1831 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832out1:
Jan Blunck1d957f92008-02-14 19:34:35 -08001833 path_put(&new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834out0:
1835 unlock_kernel();
1836 return error;
1837out3:
1838 spin_unlock(&vfsmount_lock);
1839 goto out2;
1840}
1841
1842static void __init init_mount_tree(void)
1843{
1844 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001845 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08001846 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1849 if (IS_ERR(mnt))
1850 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001851 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1852 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001854 atomic_set(&ns->count, 1);
1855 INIT_LIST_HEAD(&ns->list);
1856 init_waitqueue_head(&ns->poll);
1857 ns->event = 0;
1858 list_add(&mnt->mnt_list, &ns->list);
1859 ns->root = mnt;
1860 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001862 init_task.nsproxy->mnt_ns = ns;
1863 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
Jan Blunckac748a02008-02-14 19:34:39 -08001865 root.mnt = ns->root;
1866 root.dentry = ns->root->mnt_root;
1867
1868 set_fs_pwd(current->fs, &root);
1869 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870}
1871
Denis Cheng74bf17c2007-10-16 23:26:30 -07001872void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873{
Eric Dumazet13f14b42008-02-06 01:37:57 -08001874 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001875 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876
Ram Pai390c6842005-11-07 17:17:51 -05001877 init_rwsem(&namespace_sem);
1878
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09001880 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
Ram Paib58fed82005-11-07 17:16:09 -05001882 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 if (!mount_hashtable)
1885 panic("Failed to allocate mount hash table\n");
1886
Eric Dumazet13f14b42008-02-06 01:37:57 -08001887 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
Eric Dumazet13f14b42008-02-06 01:37:57 -08001889 for (u = 0; u < HASH_SIZE; u++)
1890 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001892 err = sysfs_init();
1893 if (err)
1894 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1895 __FUNCTION__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06001896 fs_kobj = kobject_create_and_add("fs", NULL);
1897 if (!fs_kobj)
1898 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 init_rootfs();
1900 init_mount_tree();
1901}
1902
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001903void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001905 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001906 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001907 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001908 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001909 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001911 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001913 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001914 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001915 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916}