blob: 94f026ec990ae24548eb02754f221b844abe0c2b [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
Al Viro1a390682008-03-21 20:48:19 -0400158static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Al Viro1a390682008-03-21 20:48:19 -0400160 old_path->dentry = mnt->mnt_mountpoint;
161 old_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);
Al Viro1a390682008-03-21 20:48:19 -0400166 old_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
Al Viro1a390682008-03-21 20:48:19 -0400177static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Al Viro1a390682008-03-21 20:48:19 -0400179 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400181 hash(path->mnt, path->dentry));
182 list_add_tail(&mnt->mnt_child, &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) {
Ram Pai36341f62005-11-07 17:17:22 -0500265 if (!list_empty(&old->mnt_expire))
266 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 }
269 return mnt;
270}
271
Al Viro7b7b1ac2005-11-07 17:13:39 -0500272static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 struct super_block *sb = mnt->mnt_sb;
275 dput(mnt->mnt_root);
276 free_vfsmnt(mnt);
277 deactivate_super(sb);
278}
279
Al Viro7b7b1ac2005-11-07 17:13:39 -0500280void mntput_no_expire(struct vfsmount *mnt)
281{
282repeat:
283 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
284 if (likely(!mnt->mnt_pinned)) {
285 spin_unlock(&vfsmount_lock);
286 __mntput(mnt);
287 return;
288 }
289 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
290 mnt->mnt_pinned = 0;
291 spin_unlock(&vfsmount_lock);
292 acct_auto_close_mnt(mnt);
293 security_sb_umount_close(mnt);
294 goto repeat;
295 }
296}
297
298EXPORT_SYMBOL(mntput_no_expire);
299
300void mnt_pin(struct vfsmount *mnt)
301{
302 spin_lock(&vfsmount_lock);
303 mnt->mnt_pinned++;
304 spin_unlock(&vfsmount_lock);
305}
306
307EXPORT_SYMBOL(mnt_pin);
308
309void mnt_unpin(struct vfsmount *mnt)
310{
311 spin_lock(&vfsmount_lock);
312 if (mnt->mnt_pinned) {
313 atomic_inc(&mnt->mnt_count);
314 mnt->mnt_pinned--;
315 }
316 spin_unlock(&vfsmount_lock);
317}
318
319EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800321static inline void mangle(struct seq_file *m, const char *s)
322{
323 seq_escape(m, s, " \t\n\\");
324}
325
326/*
327 * Simple .show_options callback for filesystems which don't want to
328 * implement more complex mount option showing.
329 *
330 * See also save_mount_options().
331 */
332int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
333{
334 const char *options = mnt->mnt_sb->s_options;
335
336 if (options != NULL && options[0]) {
337 seq_putc(m, ',');
338 mangle(m, options);
339 }
340
341 return 0;
342}
343EXPORT_SYMBOL(generic_show_options);
344
345/*
346 * If filesystem uses generic_show_options(), this function should be
347 * called from the fill_super() callback.
348 *
349 * The .remount_fs callback usually needs to be handled in a special
350 * way, to make sure, that previous options are not overwritten if the
351 * remount fails.
352 *
353 * Also note, that if the filesystem's .remount_fs function doesn't
354 * reset all options to their default value, but changes only newly
355 * given options, then the displayed options will not reflect reality
356 * any more.
357 */
358void save_mount_options(struct super_block *sb, char *options)
359{
360 kfree(sb->s_options);
361 sb->s_options = kstrdup(options, GFP_KERNEL);
362}
363EXPORT_SYMBOL(save_mount_options);
364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365/* iterator */
366static void *m_start(struct seq_file *m, loff_t *pos)
367{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800368 struct mnt_namespace *n = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Ram Pai390c6842005-11-07 17:17:51 -0500370 down_read(&namespace_sem);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700371 return seq_list_start(&n->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
374static void *m_next(struct seq_file *m, void *v, loff_t *pos)
375{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800376 struct mnt_namespace *n = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700377
378 return seq_list_next(v, &n->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
381static void m_stop(struct seq_file *m, void *v)
382{
Ram Pai390c6842005-11-07 17:17:51 -0500383 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384}
385
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386static int show_vfsmnt(struct seq_file *m, void *v)
387{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700388 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 int err = 0;
390 static struct proc_fs_info {
391 int flag;
392 char *str;
393 } fs_info[] = {
394 { MS_SYNCHRONOUS, ",sync" },
395 { MS_DIRSYNC, ",dirsync" },
396 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 { 0, NULL }
398 };
399 static struct proc_fs_info mnt_info[] = {
400 { MNT_NOSUID, ",nosuid" },
401 { MNT_NODEV, ",nodev" },
402 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800403 { MNT_NOATIME, ",noatime" },
404 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800405 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 { 0, NULL }
407 };
408 struct proc_fs_info *fs_infop;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800409 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
412 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800413 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 seq_putc(m, ' ');
415 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700416 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
417 seq_putc(m, '.');
418 mangle(m, mnt->mnt_sb->s_subtype);
419 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
421 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
422 if (mnt->mnt_sb->s_flags & fs_infop->flag)
423 seq_puts(m, fs_infop->str);
424 }
425 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
426 if (mnt->mnt_flags & fs_infop->flag)
427 seq_puts(m, fs_infop->str);
428 }
429 if (mnt->mnt_sb->s_op->show_options)
430 err = mnt->mnt_sb->s_op->show_options(m, mnt);
431 seq_puts(m, " 0 0\n");
432 return err;
433}
434
435struct seq_operations mounts_op = {
436 .start = m_start,
437 .next = m_next,
438 .stop = m_stop,
439 .show = show_vfsmnt
440};
441
Chuck Leverb4629fe2006-03-20 13:44:12 -0500442static int show_vfsstat(struct seq_file *m, void *v)
443{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700444 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800445 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500446 int err = 0;
447
448 /* device */
449 if (mnt->mnt_devname) {
450 seq_puts(m, "device ");
451 mangle(m, mnt->mnt_devname);
452 } else
453 seq_puts(m, "no device");
454
455 /* mount point */
456 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800457 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500458 seq_putc(m, ' ');
459
460 /* file system type */
461 seq_puts(m, "with fstype ");
462 mangle(m, mnt->mnt_sb->s_type->name);
463
464 /* optional statistics */
465 if (mnt->mnt_sb->s_op->show_stats) {
466 seq_putc(m, ' ');
467 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
468 }
469
470 seq_putc(m, '\n');
471 return err;
472}
473
474struct seq_operations mountstats_op = {
475 .start = m_start,
476 .next = m_next,
477 .stop = m_stop,
478 .show = show_vfsstat,
479};
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481/**
482 * may_umount_tree - check if a mount tree is busy
483 * @mnt: root of mount tree
484 *
485 * This is called to check if a tree of mounts has any
486 * open files, pwds, chroots or sub mounts that are
487 * busy.
488 */
489int may_umount_tree(struct vfsmount *mnt)
490{
Ram Pai36341f62005-11-07 17:17:22 -0500491 int actual_refs = 0;
492 int minimum_refs = 0;
493 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500496 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 actual_refs += atomic_read(&p->mnt_count);
498 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500 spin_unlock(&vfsmount_lock);
501
502 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800503 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504
Ian Kente3474a82006-03-27 01:14:51 -0800505 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506}
507
508EXPORT_SYMBOL(may_umount_tree);
509
510/**
511 * may_umount - check if a mount point is busy
512 * @mnt: root of mount
513 *
514 * This is called to check if a mount point has any
515 * open files, pwds, chroots or sub mounts. If the
516 * mount has sub mounts this will return busy
517 * regardless of whether the sub mounts are busy.
518 *
519 * Doesn't take quota and stuff into account. IOW, in some cases it will
520 * give false negatives. The main reason why it's here is that we need
521 * a non-destructive way to look for easily umountable filesystems.
522 */
523int may_umount(struct vfsmount *mnt)
524{
Ian Kente3474a82006-03-27 01:14:51 -0800525 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500526 spin_lock(&vfsmount_lock);
527 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800528 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500529 spin_unlock(&vfsmount_lock);
530 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
533EXPORT_SYMBOL(may_umount);
534
Ram Paib90fa9a2005-11-07 17:19:50 -0500535void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500537 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800538 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700539 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500540 list_del_init(&mnt->mnt_hash);
541 if (mnt->mnt_parent != mnt) {
542 struct dentry *dentry;
543 struct vfsmount *m;
544 spin_lock(&vfsmount_lock);
545 dentry = mnt->mnt_mountpoint;
546 m = mnt->mnt_parent;
547 mnt->mnt_mountpoint = mnt->mnt_root;
548 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -0400549 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500551 dput(dentry);
552 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 }
554 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500555 }
556}
557
Ram Paia05964f2005-11-07 17:20:17 -0500558void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500559{
560 struct vfsmount *p;
561
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700562 for (p = mnt; p; p = next_mnt(p, mnt))
563 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500564
Ram Paia05964f2005-11-07 17:20:17 -0500565 if (propagate)
566 propagate_umount(kill);
567
Ram Pai70fbcdf2005-11-07 17:17:04 -0500568 list_for_each_entry(p, kill, mnt_hash) {
569 list_del_init(&p->mnt_expire);
570 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800571 __touch_mnt_namespace(p->mnt_ns);
572 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500573 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -0400574 if (p->mnt_parent != p) {
575 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -0500576 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -0400577 }
Ram Paia05964f2005-11-07 17:20:17 -0500578 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580}
581
Al Viroc35038b2008-03-22 00:46:23 -0400582static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584static int do_umount(struct vfsmount *mnt, int flags)
585{
Ram Paib58fed82005-11-07 17:16:09 -0500586 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500588 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 retval = security_sb_umount(mnt, flags);
591 if (retval)
592 return retval;
593
594 /*
595 * Allow userspace to request a mountpoint be expired rather than
596 * unmounting unconditionally. Unmount only happens if:
597 * (1) the mark is already set (the mark is cleared by mntput())
598 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
599 */
600 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -0800601 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 flags & (MNT_FORCE | MNT_DETACH))
603 return -EINVAL;
604
605 if (atomic_read(&mnt->mnt_count) != 2)
606 return -EBUSY;
607
608 if (!xchg(&mnt->mnt_expiry_mark, 1))
609 return -EAGAIN;
610 }
611
612 /*
613 * If we may have to abort operations to get out of this
614 * mount, and they will themselves hold resources we must
615 * allow the fs to do things. In the Unix tradition of
616 * 'Gee thats tricky lets do it in userspace' the umount_begin
617 * might fail to complete on the first run through as other tasks
618 * must return, and the like. Thats for the mount program to worry
619 * about for the moment.
620 */
621
622 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400623 if (sb->s_op->umount_begin)
624 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 unlock_kernel();
626
627 /*
628 * No sense to grab the lock for this test, but test itself looks
629 * somewhat bogus. Suggestions for better replacement?
630 * Ho-hum... In principle, we might treat that as umount + switch
631 * to rootfs. GC would eventually take care of the old vfsmount.
632 * Actually it makes sense, especially if rootfs would contain a
633 * /reboot - static binary that would close all descriptors and
634 * call reboot(9). Then init(8) could umount root and exec /reboot.
635 */
Jan Blunck6ac08c32008-02-14 19:34:38 -0800636 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 /*
638 * Special case for "unmounting" root ...
639 * we just try to remount it readonly.
640 */
641 down_write(&sb->s_umount);
642 if (!(sb->s_flags & MS_RDONLY)) {
643 lock_kernel();
644 DQUOT_OFF(sb);
645 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
646 unlock_kernel();
647 }
648 up_write(&sb->s_umount);
649 return retval;
650 }
651
Ram Pai390c6842005-11-07 17:17:51 -0500652 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500654 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Al Viroc35038b2008-03-22 00:46:23 -0400656 if (!(flags & MNT_DETACH))
657 shrink_submounts(mnt, &umount_list);
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500660 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500662 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 retval = 0;
664 }
665 spin_unlock(&vfsmount_lock);
666 if (retval)
667 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500668 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500669 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 return retval;
671}
672
673/*
674 * Now umount can handle mount points as well as block devices.
675 * This is important for filesystems which use unnamed block devices.
676 *
677 * We now support a flag for forced unmount like the other 'big iron'
678 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
679 */
680
681asmlinkage long sys_umount(char __user * name, int flags)
682{
683 struct nameidata nd;
684 int retval;
685
686 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
687 if (retval)
688 goto out;
689 retval = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800690 if (nd.path.dentry != nd.path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 goto dput_and_out;
Jan Blunck4ac91372008-02-14 19:34:32 -0800692 if (!check_mnt(nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 goto dput_and_out;
694
695 retval = -EPERM;
696 if (!capable(CAP_SYS_ADMIN))
697 goto dput_and_out;
698
Jan Blunck4ac91372008-02-14 19:34:32 -0800699 retval = do_umount(nd.path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -0800701 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Jan Blunck4ac91372008-02-14 19:34:32 -0800702 dput(nd.path.dentry);
703 mntput_no_expire(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704out:
705 return retval;
706}
707
708#ifdef __ARCH_WANT_SYS_OLDUMOUNT
709
710/*
Ram Paib58fed82005-11-07 17:16:09 -0500711 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713asmlinkage long sys_oldumount(char __user * name)
714{
Ram Paib58fed82005-11-07 17:16:09 -0500715 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716}
717
718#endif
719
720static int mount_is_safe(struct nameidata *nd)
721{
722 if (capable(CAP_SYS_ADMIN))
723 return 0;
724 return -EPERM;
725#ifdef notyet
Jan Blunck4ac91372008-02-14 19:34:32 -0800726 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return -EPERM;
Jan Blunck4ac91372008-02-14 19:34:32 -0800728 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
729 if (current->uid != nd->path.dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 return -EPERM;
731 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800732 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 return -EPERM;
734 return 0;
735#endif
736}
737
Ram Paib58fed82005-11-07 17:16:09 -0500738static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739{
740 while (1) {
741 if (d == dentry)
742 return 1;
743 if (d == NULL || d == d->d_parent)
744 return 0;
745 d = d->d_parent;
746 }
747}
748
Ram Paib90fa9a2005-11-07 17:19:50 -0500749struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500750 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751{
752 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -0400753 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Ram Pai9676f0c2005-11-07 17:21:20 -0500755 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
756 return NULL;
757
Ram Pai36341f62005-11-07 17:17:22 -0500758 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 if (!q)
760 goto Enomem;
761 q->mnt_mountpoint = mnt->mnt_mountpoint;
762
763 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700764 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
766 continue;
767
768 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500769 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
770 s = skip_mnt_tree(s);
771 continue;
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 while (p != s->mnt_parent) {
774 p = p->mnt_parent;
775 q = q->mnt_parent;
776 }
777 p = s;
Al Viro1a390682008-03-21 20:48:19 -0400778 path.mnt = q;
779 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500780 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 if (!q)
782 goto Enomem;
783 spin_lock(&vfsmount_lock);
784 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -0400785 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_unlock(&vfsmount_lock);
787 }
788 }
789 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500790Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500792 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500794 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500796 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 }
798 return NULL;
799}
800
Al Viro8aec0802007-06-07 12:20:32 -0400801struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
802{
803 struct vfsmount *tree;
804 down_read(&namespace_sem);
805 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
806 up_read(&namespace_sem);
807 return tree;
808}
809
810void drop_collected_mounts(struct vfsmount *mnt)
811{
812 LIST_HEAD(umount_list);
813 down_read(&namespace_sem);
814 spin_lock(&vfsmount_lock);
815 umount_tree(mnt, 0, &umount_list);
816 spin_unlock(&vfsmount_lock);
817 up_read(&namespace_sem);
818 release_mounts(&umount_list);
819}
820
Ram Paib90fa9a2005-11-07 17:19:50 -0500821/*
822 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500823 * @nd : place the mount tree @source_mnt is attached
824 * @parent_nd : if non-null, detach the source_mnt from its parent and
825 * store the parent mount and mountpoint dentry.
826 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500827 *
828 * NOTE: in the table below explains the semantics when a source mount
829 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500830 * ---------------------------------------------------------------------------
831 * | BIND MOUNT OPERATION |
832 * |**************************************************************************
833 * | source-->| shared | private | slave | unbindable |
834 * | dest | | | | |
835 * | | | | | | |
836 * | v | | | | |
837 * |**************************************************************************
838 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
839 * | | | | | |
840 * |non-shared| shared (+) | private | slave (*) | invalid |
841 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500842 * A bind operation clones the source mount and mounts the clone on the
843 * destination mount.
844 *
845 * (++) the cloned mount is propagated to all the mounts in the propagation
846 * tree of the destination mount and the cloned mount is added to
847 * the peer group of the source mount.
848 * (+) the cloned mount is created under the destination mount and is marked
849 * as shared. The cloned mount is added to the peer group of the source
850 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500851 * (+++) the mount is propagated to all the mounts in the propagation tree
852 * of the destination mount and the cloned mount is made slave
853 * of the same master as that of the source mount. The cloned mount
854 * is marked as 'shared and slave'.
855 * (*) the cloned mount is made a slave of the same master as that of the
856 * source mount.
857 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500858 * ---------------------------------------------------------------------------
859 * | MOVE MOUNT OPERATION |
860 * |**************************************************************************
861 * | source-->| shared | private | slave | unbindable |
862 * | dest | | | | |
863 * | | | | | | |
864 * | v | | | | |
865 * |**************************************************************************
866 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
867 * | | | | | |
868 * |non-shared| shared (+*) | private | slave (*) | unbindable |
869 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500870 *
871 * (+) the mount is moved to the destination. And is then propagated to
872 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500873 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500874 * (+++) the mount is moved to the destination and is then propagated to
875 * all the mounts belonging to the destination mount's propagation tree.
876 * the mount is marked as 'shared and slave'.
877 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500878 *
879 * if the source mount is a tree, the operations explained above is
880 * applied to each mount in the tree.
881 * Must be called without spinlocks held, since this function can sleep
882 * in allocations.
883 */
884static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -0400885 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -0500886{
887 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -0400888 struct vfsmount *dest_mnt = path->mnt;
889 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -0500890 struct vfsmount *child, *p;
891
892 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
893 return -EINVAL;
894
895 if (IS_MNT_SHARED(dest_mnt)) {
896 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
897 set_mnt_shared(p);
898 }
899
900 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -0400901 if (parent_path) {
902 detach_mnt(source_mnt, parent_path);
903 attach_mnt(source_mnt, path);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800904 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500905 } else {
906 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
907 commit_tree(source_mnt);
908 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500909
910 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
911 list_del_init(&child->mnt_hash);
912 commit_tree(child);
913 }
914 spin_unlock(&vfsmount_lock);
915 return 0;
916}
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
919{
920 int err;
921 if (mnt->mnt_sb->s_flags & MS_NOUSER)
922 return -EINVAL;
923
Jan Blunck4ac91372008-02-14 19:34:32 -0800924 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
926 return -ENOTDIR;
927
928 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800929 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
930 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 goto out_unlock;
932
933 err = security_sb_check_sb(mnt, nd);
934 if (err)
935 goto out_unlock;
936
937 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800938 if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry))
Al Viro1a390682008-03-21 20:48:19 -0400939 err = attach_recursive_mnt(mnt, &nd->path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -0800941 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 if (!err)
943 security_sb_post_addmount(mnt, nd);
944 return err;
945}
946
947/*
Ram Pai07b20882005-11-07 17:19:07 -0500948 * recursively change the type of the mountpoint.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800949 * noinline this do_mount helper to save do_mount stack space.
Ram Pai07b20882005-11-07 17:19:07 -0500950 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800951static noinline int do_change_type(struct nameidata *nd, int flag)
Ram Pai07b20882005-11-07 17:19:07 -0500952{
Jan Blunck4ac91372008-02-14 19:34:32 -0800953 struct vfsmount *m, *mnt = nd->path.mnt;
Ram Pai07b20882005-11-07 17:19:07 -0500954 int recurse = flag & MS_REC;
955 int type = flag & ~MS_REC;
956
Miklos Szerediee6f9582007-05-08 00:30:40 -0700957 if (!capable(CAP_SYS_ADMIN))
958 return -EPERM;
959
Jan Blunck4ac91372008-02-14 19:34:32 -0800960 if (nd->path.dentry != nd->path.mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -0500961 return -EINVAL;
962
963 down_write(&namespace_sem);
964 spin_lock(&vfsmount_lock);
965 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
966 change_mnt_propagation(m, type);
967 spin_unlock(&vfsmount_lock);
968 up_write(&namespace_sem);
969 return 0;
970}
971
972/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 * do loopback mount.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800974 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800976static noinline int do_loopback(struct nameidata *nd, char *old_name,
977 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978{
979 struct nameidata old_nd;
980 struct vfsmount *mnt = NULL;
981 int err = mount_is_safe(nd);
982 if (err)
983 return err;
984 if (!old_name || !*old_name)
985 return -EINVAL;
986 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
987 if (err)
988 return err;
989
Ram Pai390c6842005-11-07 17:17:51 -0500990 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800992 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
993 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -0500994
Jan Blunck4ac91372008-02-14 19:34:32 -0800995 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -0500996 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Al Viroccd48bc2005-11-07 17:15:04 -0500998 err = -ENOMEM;
999 if (recurse)
Jan Blunck4ac91372008-02-14 19:34:32 -08001000 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001001 else
Jan Blunck4ac91372008-02-14 19:34:32 -08001002 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001003
1004 if (!mnt)
1005 goto out;
1006
Al Viroccd48bc2005-11-07 17:15:04 -05001007 err = graft_tree(mnt, nd);
1008 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001009 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001011 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001013 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001014 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Al Viroccd48bc2005-11-07 17:15:04 -05001016out:
Ram Pai390c6842005-11-07 17:17:51 -05001017 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001018 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 return err;
1020}
1021
1022/*
1023 * change filesystem flags. dir should be a physical root of filesystem.
1024 * If you've mounted a non-root directory somewhere and want to do remount
1025 * on it - tough luck.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001026 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001028static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 void *data)
1030{
1031 int err;
Jan Blunck4ac91372008-02-14 19:34:32 -08001032 struct super_block *sb = nd->path.mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 if (!capable(CAP_SYS_ADMIN))
1035 return -EPERM;
1036
Jan Blunck4ac91372008-02-14 19:34:32 -08001037 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 return -EINVAL;
1039
Jan Blunck4ac91372008-02-14 19:34:32 -08001040 if (nd->path.dentry != nd->path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 return -EINVAL;
1042
1043 down_write(&sb->s_umount);
1044 err = do_remount_sb(sb, flags, data, 0);
1045 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001046 nd->path.mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 up_write(&sb->s_umount);
1048 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001049 security_sb_post_remount(nd->path.mnt, flags, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 return err;
1051}
1052
Ram Pai9676f0c2005-11-07 17:21:20 -05001053static inline int tree_contains_unbindable(struct vfsmount *mnt)
1054{
1055 struct vfsmount *p;
1056 for (p = mnt; p; p = next_mnt(p, mnt)) {
1057 if (IS_MNT_UNBINDABLE(p))
1058 return 1;
1059 }
1060 return 0;
1061}
1062
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001063/*
1064 * noinline this do_mount helper to save do_mount stack space.
1065 */
1066static noinline int do_move_mount(struct nameidata *nd, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
Al Viro1a390682008-03-21 20:48:19 -04001068 struct nameidata old_nd;
1069 struct path parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 struct vfsmount *p;
1071 int err = 0;
1072 if (!capable(CAP_SYS_ADMIN))
1073 return -EPERM;
1074 if (!old_name || !*old_name)
1075 return -EINVAL;
1076 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1077 if (err)
1078 return err;
1079
Ram Pai390c6842005-11-07 17:17:51 -05001080 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001081 while (d_mountpoint(nd->path.dentry) &&
1082 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 ;
1084 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001085 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 goto out;
1087
1088 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001089 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1090 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 goto out1;
1092
Jan Blunck4ac91372008-02-14 19:34:32 -08001093 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05001094 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095
1096 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001097 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001098 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
Jan Blunck4ac91372008-02-14 19:34:32 -08001100 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001101 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Jan Blunck4ac91372008-02-14 19:34:32 -08001103 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1104 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001105 goto out1;
1106 /*
1107 * Don't move a mount residing in a shared parent.
1108 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001109 if (old_nd.path.mnt->mnt_parent &&
1110 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001111 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001112 /*
1113 * Don't move a mount tree containing unbindable mounts to a destination
1114 * mount which is shared.
1115 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001116 if (IS_MNT_SHARED(nd->path.mnt) &&
1117 tree_contains_unbindable(old_nd.path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001118 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 err = -ELOOP;
Jan Blunck4ac91372008-02-14 19:34:32 -08001120 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1121 if (p == old_nd.path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001122 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Al Viro1a390682008-03-21 20:48:19 -04001124 err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001125 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001126 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128 /* if the mount is moved, it should no longer be expire
1129 * automatically */
Jan Blunck4ac91372008-02-14 19:34:32 -08001130 list_del_init(&old_nd.path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131out1:
Jan Blunck4ac91372008-02-14 19:34:32 -08001132 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133out:
Ram Pai390c6842005-11-07 17:17:51 -05001134 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001136 path_put(&parent_path);
Jan Blunck1d957f92008-02-14 19:34:35 -08001137 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 return err;
1139}
1140
1141/*
1142 * create a new mount for userspace and request it to be added into the
1143 * namespace's tree
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001144 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001146static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 int mnt_flags, char *name, void *data)
1148{
1149 struct vfsmount *mnt;
1150
1151 if (!type || !memchr(type, 0, PAGE_SIZE))
1152 return -EINVAL;
1153
1154 /* we need capabilities... */
1155 if (!capable(CAP_SYS_ADMIN))
1156 return -EPERM;
1157
1158 mnt = do_kern_mount(type, flags, name, data);
1159 if (IS_ERR(mnt))
1160 return PTR_ERR(mnt);
1161
1162 return do_add_mount(mnt, nd, mnt_flags, NULL);
1163}
1164
1165/*
1166 * add a mount into a namespace's mount tree
1167 * - provide the option of adding the new mount to an expiration list
1168 */
1169int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1170 int mnt_flags, struct list_head *fslist)
1171{
1172 int err;
1173
Ram Pai390c6842005-11-07 17:17:51 -05001174 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 /* Something was mounted here while we slept */
Jan Blunck4ac91372008-02-14 19:34:32 -08001176 while (d_mountpoint(nd->path.dentry) &&
1177 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 ;
1179 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001180 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 goto unlock;
1182
1183 /* Refuse the same filesystem on the same mount point */
1184 err = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001185 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1186 nd->path.mnt->mnt_root == nd->path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 goto unlock;
1188
1189 err = -EINVAL;
1190 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1191 goto unlock;
1192
1193 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001194 if ((err = graft_tree(newmnt, nd)))
1195 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Al Viro6758f952008-03-22 16:14:30 -04001197 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001198 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001199
Ram Pai390c6842005-11-07 17:17:51 -05001200 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001201 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
1203unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001204 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 mntput(newmnt);
1206 return err;
1207}
1208
1209EXPORT_SYMBOL_GPL(do_add_mount);
1210
Trond Myklebust5528f9112006-06-09 09:34:17 -04001211/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 * process a list of expirable mountpoints with the intent of discarding any
1213 * mountpoints that aren't in use and haven't been touched since last we came
1214 * here
1215 */
1216void mark_mounts_for_expiry(struct list_head *mounts)
1217{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 struct vfsmount *mnt, *next;
1219 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001220 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222 if (list_empty(mounts))
1223 return;
1224
Al Virobcc5c7d2008-03-22 00:21:53 -04001225 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 spin_lock(&vfsmount_lock);
1227
1228 /* extract from the expiration list every vfsmount that matches the
1229 * following criteria:
1230 * - only referenced by its parent vfsmount
1231 * - still marked for expiry (marked on the last call here; marks are
1232 * cleared by mntput())
1233 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001234 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001236 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001238 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001240 while (!list_empty(&graveyard)) {
1241 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1242 touch_mnt_namespace(mnt->mnt_ns);
1243 umount_tree(mnt, 1, &umounts);
1244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001246 up_write(&namespace_sem);
1247
1248 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249}
1250
1251EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1252
1253/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001254 * Ripoff of 'select_parent()'
1255 *
1256 * search the list of submounts for a given mountpoint, and move any
1257 * shrinkable submounts to the 'graveyard' list.
1258 */
1259static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1260{
1261 struct vfsmount *this_parent = parent;
1262 struct list_head *next;
1263 int found = 0;
1264
1265repeat:
1266 next = this_parent->mnt_mounts.next;
1267resume:
1268 while (next != &this_parent->mnt_mounts) {
1269 struct list_head *tmp = next;
1270 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1271
1272 next = tmp->next;
1273 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1274 continue;
1275 /*
1276 * Descend a level if the d_mounts list is non-empty.
1277 */
1278 if (!list_empty(&mnt->mnt_mounts)) {
1279 this_parent = mnt;
1280 goto repeat;
1281 }
1282
1283 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001284 list_move_tail(&mnt->mnt_expire, graveyard);
1285 found++;
1286 }
1287 }
1288 /*
1289 * All done at this level ... ascend and resume the search
1290 */
1291 if (this_parent != parent) {
1292 next = this_parent->mnt_child.next;
1293 this_parent = this_parent->mnt_parent;
1294 goto resume;
1295 }
1296 return found;
1297}
1298
1299/*
1300 * process a list of expirable mountpoints with the intent of discarding any
1301 * submounts of a specific parent mountpoint
1302 */
Al Viroc35038b2008-03-22 00:46:23 -04001303static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001304{
1305 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001306 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001307
Trond Myklebust5528f9112006-06-09 09:34:17 -04001308 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001309 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001310 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001311 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001312 mnt_expire);
1313 touch_mnt_namespace(mnt->mnt_ns);
Al Viroc35038b2008-03-22 00:46:23 -04001314 umount_tree(mnt, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001315 }
1316 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001317}
1318
Trond Myklebust5528f9112006-06-09 09:34:17 -04001319/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 * Some copy_from_user() implementations do not return the exact number of
1321 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1322 * Note that this function differs from copy_from_user() in that it will oops
1323 * on bad values of `to', rather than returning a short copy.
1324 */
Ram Paib58fed82005-11-07 17:16:09 -05001325static long exact_copy_from_user(void *to, const void __user * from,
1326 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327{
1328 char *t = to;
1329 const char __user *f = from;
1330 char c;
1331
1332 if (!access_ok(VERIFY_READ, from, n))
1333 return n;
1334
1335 while (n) {
1336 if (__get_user(c, f)) {
1337 memset(t, 0, n);
1338 break;
1339 }
1340 *t++ = c;
1341 f++;
1342 n--;
1343 }
1344 return n;
1345}
1346
Ram Paib58fed82005-11-07 17:16:09 -05001347int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
1349 int i;
1350 unsigned long page;
1351 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 *where = 0;
1354 if (!data)
1355 return 0;
1356
1357 if (!(page = __get_free_page(GFP_KERNEL)))
1358 return -ENOMEM;
1359
1360 /* We only care that *some* data at the address the user
1361 * gave us is valid. Just in case, we'll zero
1362 * the remainder of the page.
1363 */
1364 /* copy_from_user cannot cross TASK_SIZE ! */
1365 size = TASK_SIZE - (unsigned long)data;
1366 if (size > PAGE_SIZE)
1367 size = PAGE_SIZE;
1368
1369 i = size - exact_copy_from_user((void *)page, data, size);
1370 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001371 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 return -EFAULT;
1373 }
1374 if (i != PAGE_SIZE)
1375 memset((char *)page + i, 0, PAGE_SIZE - i);
1376 *where = page;
1377 return 0;
1378}
1379
1380/*
1381 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1382 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1383 *
1384 * data is a (void *) that can point to any structure up to
1385 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1386 * information (or be NULL).
1387 *
1388 * Pre-0.97 versions of mount() didn't have a flags word.
1389 * When the flags word was introduced its top half was required
1390 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1391 * Therefore, if this magic number is present, it carries no information
1392 * and must be discarded.
1393 */
Ram Paib58fed82005-11-07 17:16:09 -05001394long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 unsigned long flags, void *data_page)
1396{
1397 struct nameidata nd;
1398 int retval = 0;
1399 int mnt_flags = 0;
1400
1401 /* Discard magic */
1402 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1403 flags &= ~MS_MGC_MSK;
1404
1405 /* Basic sanity checks */
1406
1407 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1408 return -EINVAL;
1409 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1410 return -EINVAL;
1411
1412 if (data_page)
1413 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1414
1415 /* Separate the per-mountpoint flags */
1416 if (flags & MS_NOSUID)
1417 mnt_flags |= MNT_NOSUID;
1418 if (flags & MS_NODEV)
1419 mnt_flags |= MNT_NODEV;
1420 if (flags & MS_NOEXEC)
1421 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001422 if (flags & MS_NOATIME)
1423 mnt_flags |= MNT_NOATIME;
1424 if (flags & MS_NODIRATIME)
1425 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001426 if (flags & MS_RELATIME)
1427 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001428
1429 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001430 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
1432 /* ... and get the mountpoint */
1433 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1434 if (retval)
1435 return retval;
1436
1437 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1438 if (retval)
1439 goto dput_out;
1440
1441 if (flags & MS_REMOUNT)
1442 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1443 data_page);
1444 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001445 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001446 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001447 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 else if (flags & MS_MOVE)
1449 retval = do_move_mount(&nd, dev_name);
1450 else
1451 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1452 dev_name, data_page);
1453dput_out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001454 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 return retval;
1456}
1457
JANAK DESAI741a2952006-02-07 12:59:00 -08001458/*
1459 * Allocate a new namespace structure and populate it with contents
1460 * copied from the namespace of the passed in task structure.
1461 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001462static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001463 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001465 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 struct vfsmount *p, *q;
1468
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001469 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001471 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
1473 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001475 init_waitqueue_head(&new_ns->poll);
1476 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
Ram Pai390c6842005-11-07 17:17:51 -05001478 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001480 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001481 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001483 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001485 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 }
1487 spin_lock(&vfsmount_lock);
1488 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1489 spin_unlock(&vfsmount_lock);
1490
1491 /*
1492 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1493 * as belonging to new namespace. We have already acquired a private
1494 * fs_struct, so tsk->fs->lock is not needed.
1495 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001496 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 q = new_ns->root;
1498 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001499 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001501 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001503 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001505 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001507 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001509 if (p == fs->altroot.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 altrootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001511 fs->altroot.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 }
1513 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001514 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 q = next_mnt(q, new_ns->root);
1516 }
Ram Pai390c6842005-11-07 17:17:51 -05001517 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 if (rootmnt)
1520 mntput(rootmnt);
1521 if (pwdmnt)
1522 mntput(pwdmnt);
1523 if (altrootmnt)
1524 mntput(altrootmnt);
1525
JANAK DESAI741a2952006-02-07 12:59:00 -08001526 return new_ns;
1527}
1528
Eric W. Biederman213dd262007-07-15 23:41:15 -07001529struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07001530 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001531{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001532 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001533
Badari Pulavartye3222c42007-05-08 00:25:21 -07001534 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001535 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001536
1537 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001538 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001539
Badari Pulavartye3222c42007-05-08 00:25:21 -07001540 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001541
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001542 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001543 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544}
1545
1546asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1547 char __user * type, unsigned long flags,
1548 void __user * data)
1549{
1550 int retval;
1551 unsigned long data_page;
1552 unsigned long type_page;
1553 unsigned long dev_page;
1554 char *dir_page;
1555
Ram Paib58fed82005-11-07 17:16:09 -05001556 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (retval < 0)
1558 return retval;
1559
1560 dir_page = getname(dir_name);
1561 retval = PTR_ERR(dir_page);
1562 if (IS_ERR(dir_page))
1563 goto out1;
1564
Ram Paib58fed82005-11-07 17:16:09 -05001565 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 if (retval < 0)
1567 goto out2;
1568
Ram Paib58fed82005-11-07 17:16:09 -05001569 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 if (retval < 0)
1571 goto out3;
1572
1573 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001574 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1575 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 unlock_kernel();
1577 free_page(data_page);
1578
1579out3:
1580 free_page(dev_page);
1581out2:
1582 putname(dir_page);
1583out1:
1584 free_page(type_page);
1585 return retval;
1586}
1587
1588/*
1589 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1590 * It can block. Requires the big lock held.
1591 */
Jan Blunckac748a02008-02-14 19:34:39 -08001592void set_fs_root(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001594 struct path old_root;
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 write_lock(&fs->lock);
1597 old_root = fs->root;
Jan Blunckac748a02008-02-14 19:34:39 -08001598 fs->root = *path;
1599 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 write_unlock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001601 if (old_root.dentry)
1602 path_put(&old_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603}
1604
1605/*
1606 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1607 * It can block. Requires the big lock held.
1608 */
Jan Blunckac748a02008-02-14 19:34:39 -08001609void set_fs_pwd(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001611 struct path old_pwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 write_lock(&fs->lock);
1614 old_pwd = fs->pwd;
Jan Blunckac748a02008-02-14 19:34:39 -08001615 fs->pwd = *path;
1616 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 write_unlock(&fs->lock);
1618
Jan Blunck6ac08c32008-02-14 19:34:38 -08001619 if (old_pwd.dentry)
1620 path_put(&old_pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621}
1622
Al Viro1a390682008-03-21 20:48:19 -04001623static void chroot_fs_refs(struct path *old_root, struct path *new_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624{
1625 struct task_struct *g, *p;
1626 struct fs_struct *fs;
1627
1628 read_lock(&tasklist_lock);
1629 do_each_thread(g, p) {
1630 task_lock(p);
1631 fs = p->fs;
1632 if (fs) {
1633 atomic_inc(&fs->count);
1634 task_unlock(p);
Al Viro1a390682008-03-21 20:48:19 -04001635 if (fs->root.dentry == old_root->dentry
1636 && fs->root.mnt == old_root->mnt)
1637 set_fs_root(fs, new_root);
1638 if (fs->pwd.dentry == old_root->dentry
1639 && fs->pwd.mnt == old_root->mnt)
1640 set_fs_pwd(fs, new_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 put_fs_struct(fs);
1642 } else
1643 task_unlock(p);
1644 } while_each_thread(g, p);
1645 read_unlock(&tasklist_lock);
1646}
1647
1648/*
1649 * pivot_root Semantics:
1650 * Moves the root file system of the current process to the directory put_old,
1651 * makes new_root as the new root file system of the current process, and sets
1652 * root/cwd of all processes which had them on the current root to new_root.
1653 *
1654 * Restrictions:
1655 * The new_root and put_old must be directories, and must not be on the
1656 * same file system as the current process root. The put_old must be
1657 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1658 * pointed to by put_old must yield the same directory as new_root. No other
1659 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1660 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001661 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1662 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1663 * in this situation.
1664 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 * Notes:
1666 * - we don't move root/cwd if they are not at the root (reason: if something
1667 * cared enough to change them, it's probably wrong to force them elsewhere)
1668 * - it's okay to pick a root that isn't the root of a file system, e.g.
1669 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1670 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1671 * first.
1672 */
Ram Paib58fed82005-11-07 17:16:09 -05001673asmlinkage long sys_pivot_root(const char __user * new_root,
1674 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675{
1676 struct vfsmount *tmp;
Al Viro1a390682008-03-21 20:48:19 -04001677 struct nameidata new_nd, old_nd, user_nd;
1678 struct path parent_path, root_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 int error;
1680
1681 if (!capable(CAP_SYS_ADMIN))
1682 return -EPERM;
1683
1684 lock_kernel();
1685
Ram Paib58fed82005-11-07 17:16:09 -05001686 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1687 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 if (error)
1689 goto out0;
1690 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001691 if (!check_mnt(new_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 goto out1;
1693
Ram Paib58fed82005-11-07 17:16:09 -05001694 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 if (error)
1696 goto out1;
1697
1698 error = security_sb_pivotroot(&old_nd, &new_nd);
1699 if (error) {
Jan Blunck1d957f92008-02-14 19:34:35 -08001700 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 goto out1;
1702 }
1703
1704 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001705 user_nd.path = current->fs->root;
1706 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001708 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001709 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001711 if (IS_MNT_SHARED(old_nd.path.mnt) ||
1712 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
1713 IS_MNT_SHARED(user_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001714 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001715 if (!check_mnt(user_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 goto out2;
1717 error = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001718 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001720 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001722 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 goto out2;
1724 error = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001725 if (new_nd.path.mnt == user_nd.path.mnt ||
1726 old_nd.path.mnt == user_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 goto out2; /* loop, on the same file system */
1728 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001729 if (user_nd.path.mnt->mnt_root != user_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001731 if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001732 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001733 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001735 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001736 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001737 /* make sure we can reach put_old from new_root */
1738 tmp = old_nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -08001740 if (tmp != new_nd.path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 for (;;) {
1742 if (tmp->mnt_parent == tmp)
1743 goto out3; /* already mounted on put_old */
Jan Blunck4ac91372008-02-14 19:34:32 -08001744 if (tmp->mnt_parent == new_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 break;
1746 tmp = tmp->mnt_parent;
1747 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001748 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08001750 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 goto out3;
Al Viro1a390682008-03-21 20:48:19 -04001752 detach_mnt(new_nd.path.mnt, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001753 detach_mnt(user_nd.path.mnt, &root_parent);
1754 /* mount old root on put_old */
Al Viro1a390682008-03-21 20:48:19 -04001755 attach_mnt(user_nd.path.mnt, &old_nd.path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001756 /* mount new_root on / */
1757 attach_mnt(new_nd.path.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001758 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 spin_unlock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001760 chroot_fs_refs(&user_nd.path, &new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 security_sb_post_pivotroot(&user_nd, &new_nd);
1762 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04001763 path_put(&root_parent);
1764 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765out2:
Jan Blunck4ac91372008-02-14 19:34:32 -08001766 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001767 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001768 path_put(&user_nd.path);
1769 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770out1:
Jan Blunck1d957f92008-02-14 19:34:35 -08001771 path_put(&new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772out0:
1773 unlock_kernel();
1774 return error;
1775out3:
1776 spin_unlock(&vfsmount_lock);
1777 goto out2;
1778}
1779
1780static void __init init_mount_tree(void)
1781{
1782 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001783 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08001784 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
1786 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1787 if (IS_ERR(mnt))
1788 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001789 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1790 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001792 atomic_set(&ns->count, 1);
1793 INIT_LIST_HEAD(&ns->list);
1794 init_waitqueue_head(&ns->poll);
1795 ns->event = 0;
1796 list_add(&mnt->mnt_list, &ns->list);
1797 ns->root = mnt;
1798 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001800 init_task.nsproxy->mnt_ns = ns;
1801 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Jan Blunckac748a02008-02-14 19:34:39 -08001803 root.mnt = ns->root;
1804 root.dentry = ns->root->mnt_root;
1805
1806 set_fs_pwd(current->fs, &root);
1807 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808}
1809
Denis Cheng74bf17c2007-10-16 23:26:30 -07001810void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811{
Eric Dumazet13f14b42008-02-06 01:37:57 -08001812 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001813 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
Ram Pai390c6842005-11-07 17:17:51 -05001815 init_rwsem(&namespace_sem);
1816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09001818 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819
Ram Paib58fed82005-11-07 17:16:09 -05001820 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
1822 if (!mount_hashtable)
1823 panic("Failed to allocate mount hash table\n");
1824
Eric Dumazet13f14b42008-02-06 01:37:57 -08001825 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Eric Dumazet13f14b42008-02-06 01:37:57 -08001827 for (u = 0; u < HASH_SIZE; u++)
1828 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001830 err = sysfs_init();
1831 if (err)
1832 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1833 __FUNCTION__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06001834 fs_kobj = kobject_create_and_add("fs", NULL);
1835 if (!fs_kobj)
1836 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 init_rootfs();
1838 init_mount_tree();
1839}
1840
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001841void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001843 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001844 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001845 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001846 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001847 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001849 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001851 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001852 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001853 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854}