blob: eef57635ee07c551a79ab2ad30b9438897379ba4 [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;
411
412 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
413 seq_putc(m, ' ');
414 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
415 seq_putc(m, ' ');
416 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700417 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
418 seq_putc(m, '.');
419 mangle(m, mnt->mnt_sb->s_subtype);
420 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
422 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
423 if (mnt->mnt_sb->s_flags & fs_infop->flag)
424 seq_puts(m, fs_infop->str);
425 }
426 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
427 if (mnt->mnt_flags & fs_infop->flag)
428 seq_puts(m, fs_infop->str);
429 }
430 if (mnt->mnt_sb->s_op->show_options)
431 err = mnt->mnt_sb->s_op->show_options(m, mnt);
432 seq_puts(m, " 0 0\n");
433 return err;
434}
435
436struct seq_operations mounts_op = {
437 .start = m_start,
438 .next = m_next,
439 .stop = m_stop,
440 .show = show_vfsmnt
441};
442
Chuck Leverb4629fe2006-03-20 13:44:12 -0500443static int show_vfsstat(struct seq_file *m, void *v)
444{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700445 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
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 ");
457 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
458 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500550 dput(dentry);
551 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 }
553 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500554 }
555}
556
Ram Paia05964f2005-11-07 17:20:17 -0500557void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500558{
559 struct vfsmount *p;
560
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700561 for (p = mnt; p; p = next_mnt(p, mnt))
562 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500563
Ram Paia05964f2005-11-07 17:20:17 -0500564 if (propagate)
565 propagate_umount(kill);
566
Ram Pai70fbcdf2005-11-07 17:17:04 -0500567 list_for_each_entry(p, kill, mnt_hash) {
568 list_del_init(&p->mnt_expire);
569 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800570 __touch_mnt_namespace(p->mnt_ns);
571 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500572 list_del_init(&p->mnt_child);
573 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500574 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500575 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 }
577}
578
579static int do_umount(struct vfsmount *mnt, int flags)
580{
Ram Paib58fed82005-11-07 17:16:09 -0500581 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500583 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 retval = security_sb_umount(mnt, flags);
586 if (retval)
587 return retval;
588
589 /*
590 * Allow userspace to request a mountpoint be expired rather than
591 * unmounting unconditionally. Unmount only happens if:
592 * (1) the mark is already set (the mark is cleared by mntput())
593 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
594 */
595 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -0800596 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 flags & (MNT_FORCE | MNT_DETACH))
598 return -EINVAL;
599
600 if (atomic_read(&mnt->mnt_count) != 2)
601 return -EBUSY;
602
603 if (!xchg(&mnt->mnt_expiry_mark, 1))
604 return -EAGAIN;
605 }
606
607 /*
608 * If we may have to abort operations to get out of this
609 * mount, and they will themselves hold resources we must
610 * allow the fs to do things. In the Unix tradition of
611 * 'Gee thats tricky lets do it in userspace' the umount_begin
612 * might fail to complete on the first run through as other tasks
613 * must return, and the like. Thats for the mount program to worry
614 * about for the moment.
615 */
616
617 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400618 if (sb->s_op->umount_begin)
619 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 unlock_kernel();
621
622 /*
623 * No sense to grab the lock for this test, but test itself looks
624 * somewhat bogus. Suggestions for better replacement?
625 * Ho-hum... In principle, we might treat that as umount + switch
626 * to rootfs. GC would eventually take care of the old vfsmount.
627 * Actually it makes sense, especially if rootfs would contain a
628 * /reboot - static binary that would close all descriptors and
629 * call reboot(9). Then init(8) could umount root and exec /reboot.
630 */
Jan Blunck6ac08c32008-02-14 19:34:38 -0800631 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 /*
633 * Special case for "unmounting" root ...
634 * we just try to remount it readonly.
635 */
636 down_write(&sb->s_umount);
637 if (!(sb->s_flags & MS_RDONLY)) {
638 lock_kernel();
639 DQUOT_OFF(sb);
640 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
641 unlock_kernel();
642 }
643 up_write(&sb->s_umount);
644 return retval;
645 }
646
Ram Pai390c6842005-11-07 17:17:51 -0500647 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500649 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500652 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500654 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 retval = 0;
656 }
657 spin_unlock(&vfsmount_lock);
658 if (retval)
659 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500660 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500661 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return retval;
663}
664
665/*
666 * Now umount can handle mount points as well as block devices.
667 * This is important for filesystems which use unnamed block devices.
668 *
669 * We now support a flag for forced unmount like the other 'big iron'
670 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
671 */
672
673asmlinkage long sys_umount(char __user * name, int flags)
674{
675 struct nameidata nd;
676 int retval;
677
678 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
679 if (retval)
680 goto out;
681 retval = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800682 if (nd.path.dentry != nd.path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 goto dput_and_out;
Jan Blunck4ac91372008-02-14 19:34:32 -0800684 if (!check_mnt(nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 goto dput_and_out;
686
687 retval = -EPERM;
688 if (!capable(CAP_SYS_ADMIN))
689 goto dput_and_out;
690
Jan Blunck4ac91372008-02-14 19:34:32 -0800691 retval = do_umount(nd.path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -0800693 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Jan Blunck4ac91372008-02-14 19:34:32 -0800694 dput(nd.path.dentry);
695 mntput_no_expire(nd.path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696out:
697 return retval;
698}
699
700#ifdef __ARCH_WANT_SYS_OLDUMOUNT
701
702/*
Ram Paib58fed82005-11-07 17:16:09 -0500703 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705asmlinkage long sys_oldumount(char __user * name)
706{
Ram Paib58fed82005-11-07 17:16:09 -0500707 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708}
709
710#endif
711
712static int mount_is_safe(struct nameidata *nd)
713{
714 if (capable(CAP_SYS_ADMIN))
715 return 0;
716 return -EPERM;
717#ifdef notyet
Jan Blunck4ac91372008-02-14 19:34:32 -0800718 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 return -EPERM;
Jan Blunck4ac91372008-02-14 19:34:32 -0800720 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
721 if (current->uid != nd->path.dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 return -EPERM;
723 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800724 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return -EPERM;
726 return 0;
727#endif
728}
729
Ram Paib58fed82005-11-07 17:16:09 -0500730static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731{
732 while (1) {
733 if (d == dentry)
734 return 1;
735 if (d == NULL || d == d->d_parent)
736 return 0;
737 d = d->d_parent;
738 }
739}
740
Ram Paib90fa9a2005-11-07 17:19:50 -0500741struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500742 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
744 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 struct nameidata nd;
746
Ram Pai9676f0c2005-11-07 17:21:20 -0500747 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
748 return NULL;
749
Ram Pai36341f62005-11-07 17:17:22 -0500750 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 if (!q)
752 goto Enomem;
753 q->mnt_mountpoint = mnt->mnt_mountpoint;
754
755 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700756 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
758 continue;
759
760 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500761 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
762 s = skip_mnt_tree(s);
763 continue;
764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 while (p != s->mnt_parent) {
766 p = p->mnt_parent;
767 q = q->mnt_parent;
768 }
769 p = s;
Jan Blunck4ac91372008-02-14 19:34:32 -0800770 nd.path.mnt = q;
771 nd.path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500772 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 if (!q)
774 goto Enomem;
775 spin_lock(&vfsmount_lock);
776 list_add_tail(&q->mnt_list, &res->mnt_list);
777 attach_mnt(q, &nd);
778 spin_unlock(&vfsmount_lock);
779 }
780 }
781 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500782Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500784 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500786 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500788 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 }
790 return NULL;
791}
792
Al Viro8aec0802007-06-07 12:20:32 -0400793struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
794{
795 struct vfsmount *tree;
796 down_read(&namespace_sem);
797 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
798 up_read(&namespace_sem);
799 return tree;
800}
801
802void drop_collected_mounts(struct vfsmount *mnt)
803{
804 LIST_HEAD(umount_list);
805 down_read(&namespace_sem);
806 spin_lock(&vfsmount_lock);
807 umount_tree(mnt, 0, &umount_list);
808 spin_unlock(&vfsmount_lock);
809 up_read(&namespace_sem);
810 release_mounts(&umount_list);
811}
812
Ram Paib90fa9a2005-11-07 17:19:50 -0500813/*
814 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500815 * @nd : place the mount tree @source_mnt is attached
816 * @parent_nd : if non-null, detach the source_mnt from its parent and
817 * store the parent mount and mountpoint dentry.
818 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500819 *
820 * NOTE: in the table below explains the semantics when a source mount
821 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500822 * ---------------------------------------------------------------------------
823 * | BIND MOUNT OPERATION |
824 * |**************************************************************************
825 * | source-->| shared | private | slave | unbindable |
826 * | dest | | | | |
827 * | | | | | | |
828 * | v | | | | |
829 * |**************************************************************************
830 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
831 * | | | | | |
832 * |non-shared| shared (+) | private | slave (*) | invalid |
833 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500834 * A bind operation clones the source mount and mounts the clone on the
835 * destination mount.
836 *
837 * (++) the cloned mount is propagated to all the mounts in the propagation
838 * tree of the destination mount and the cloned mount is added to
839 * the peer group of the source mount.
840 * (+) the cloned mount is created under the destination mount and is marked
841 * as shared. The cloned mount is added to the peer group of the source
842 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500843 * (+++) the mount is propagated to all the mounts in the propagation tree
844 * of the destination mount and the cloned mount is made slave
845 * of the same master as that of the source mount. The cloned mount
846 * is marked as 'shared and slave'.
847 * (*) the cloned mount is made a slave of the same master as that of the
848 * source mount.
849 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500850 * ---------------------------------------------------------------------------
851 * | MOVE MOUNT OPERATION |
852 * |**************************************************************************
853 * | source-->| shared | private | slave | unbindable |
854 * | dest | | | | |
855 * | | | | | | |
856 * | v | | | | |
857 * |**************************************************************************
858 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
859 * | | | | | |
860 * |non-shared| shared (+*) | private | slave (*) | unbindable |
861 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500862 *
863 * (+) the mount is moved to the destination. And is then propagated to
864 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500865 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500866 * (+++) the mount is moved to the destination and is then propagated to
867 * all the mounts belonging to the destination mount's propagation tree.
868 * the mount is marked as 'shared and slave'.
869 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500870 *
871 * if the source mount is a tree, the operations explained above is
872 * applied to each mount in the tree.
873 * Must be called without spinlocks held, since this function can sleep
874 * in allocations.
875 */
876static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500877 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500878{
879 LIST_HEAD(tree_list);
Jan Blunck4ac91372008-02-14 19:34:32 -0800880 struct vfsmount *dest_mnt = nd->path.mnt;
881 struct dentry *dest_dentry = nd->path.dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -0500882 struct vfsmount *child, *p;
883
884 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
885 return -EINVAL;
886
887 if (IS_MNT_SHARED(dest_mnt)) {
888 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
889 set_mnt_shared(p);
890 }
891
892 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500893 if (parent_nd) {
894 detach_mnt(source_mnt, parent_nd);
895 attach_mnt(source_mnt, nd);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800896 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500897 } else {
898 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
899 commit_tree(source_mnt);
900 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500901
902 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
903 list_del_init(&child->mnt_hash);
904 commit_tree(child);
905 }
906 spin_unlock(&vfsmount_lock);
907 return 0;
908}
909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
911{
912 int err;
913 if (mnt->mnt_sb->s_flags & MS_NOUSER)
914 return -EINVAL;
915
Jan Blunck4ac91372008-02-14 19:34:32 -0800916 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
918 return -ENOTDIR;
919
920 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800921 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
922 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 goto out_unlock;
924
925 err = security_sb_check_sb(mnt, nd);
926 if (err)
927 goto out_unlock;
928
929 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -0800930 if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -0500931 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932out_unlock:
Jan Blunck4ac91372008-02-14 19:34:32 -0800933 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 if (!err)
935 security_sb_post_addmount(mnt, nd);
936 return err;
937}
938
939/*
Ram Pai07b20882005-11-07 17:19:07 -0500940 * recursively change the type of the mountpoint.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800941 * noinline this do_mount helper to save do_mount stack space.
Ram Pai07b20882005-11-07 17:19:07 -0500942 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800943static noinline int do_change_type(struct nameidata *nd, int flag)
Ram Pai07b20882005-11-07 17:19:07 -0500944{
Jan Blunck4ac91372008-02-14 19:34:32 -0800945 struct vfsmount *m, *mnt = nd->path.mnt;
Ram Pai07b20882005-11-07 17:19:07 -0500946 int recurse = flag & MS_REC;
947 int type = flag & ~MS_REC;
948
Miklos Szerediee6f9582007-05-08 00:30:40 -0700949 if (!capable(CAP_SYS_ADMIN))
950 return -EPERM;
951
Jan Blunck4ac91372008-02-14 19:34:32 -0800952 if (nd->path.dentry != nd->path.mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -0500953 return -EINVAL;
954
955 down_write(&namespace_sem);
956 spin_lock(&vfsmount_lock);
957 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
958 change_mnt_propagation(m, type);
959 spin_unlock(&vfsmount_lock);
960 up_write(&namespace_sem);
961 return 0;
962}
963
964/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 * do loopback mount.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800966 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -0800968static noinline int do_loopback(struct nameidata *nd, char *old_name,
969 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970{
971 struct nameidata old_nd;
972 struct vfsmount *mnt = NULL;
973 int err = mount_is_safe(nd);
974 if (err)
975 return err;
976 if (!old_name || !*old_name)
977 return -EINVAL;
978 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
979 if (err)
980 return err;
981
Ram Pai390c6842005-11-07 17:17:51 -0500982 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -0800984 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
985 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -0500986
Jan Blunck4ac91372008-02-14 19:34:32 -0800987 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -0500988 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989
Al Viroccd48bc2005-11-07 17:15:04 -0500990 err = -ENOMEM;
991 if (recurse)
Jan Blunck4ac91372008-02-14 19:34:32 -0800992 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500993 else
Jan Blunck4ac91372008-02-14 19:34:32 -0800994 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500995
996 if (!mnt)
997 goto out;
998
Al Viroccd48bc2005-11-07 17:15:04 -0500999 err = graft_tree(mnt, nd);
1000 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001001 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001003 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001005 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001006 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Al Viroccd48bc2005-11-07 17:15:04 -05001008out:
Ram Pai390c6842005-11-07 17:17:51 -05001009 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001010 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 return err;
1012}
1013
1014/*
1015 * change filesystem flags. dir should be a physical root of filesystem.
1016 * If you've mounted a non-root directory somewhere and want to do remount
1017 * on it - tough luck.
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001018 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001020static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 void *data)
1022{
1023 int err;
Jan Blunck4ac91372008-02-14 19:34:32 -08001024 struct super_block *sb = nd->path.mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 if (!capable(CAP_SYS_ADMIN))
1027 return -EPERM;
1028
Jan Blunck4ac91372008-02-14 19:34:32 -08001029 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 return -EINVAL;
1031
Jan Blunck4ac91372008-02-14 19:34:32 -08001032 if (nd->path.dentry != nd->path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 return -EINVAL;
1034
1035 down_write(&sb->s_umount);
1036 err = do_remount_sb(sb, flags, data, 0);
1037 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001038 nd->path.mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 up_write(&sb->s_umount);
1040 if (!err)
Jan Blunck4ac91372008-02-14 19:34:32 -08001041 security_sb_post_remount(nd->path.mnt, flags, data);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 return err;
1043}
1044
Ram Pai9676f0c2005-11-07 17:21:20 -05001045static inline int tree_contains_unbindable(struct vfsmount *mnt)
1046{
1047 struct vfsmount *p;
1048 for (p = mnt; p; p = next_mnt(p, mnt)) {
1049 if (IS_MNT_UNBINDABLE(p))
1050 return 1;
1051 }
1052 return 0;
1053}
1054
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001055/*
1056 * noinline this do_mount helper to save do_mount stack space.
1057 */
1058static noinline int do_move_mount(struct nameidata *nd, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059{
1060 struct nameidata old_nd, parent_nd;
1061 struct vfsmount *p;
1062 int err = 0;
1063 if (!capable(CAP_SYS_ADMIN))
1064 return -EPERM;
1065 if (!old_name || !*old_name)
1066 return -EINVAL;
1067 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1068 if (err)
1069 return err;
1070
Ram Pai390c6842005-11-07 17:17:51 -05001071 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001072 while (d_mountpoint(nd->path.dentry) &&
1073 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 ;
1075 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001076 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 goto out;
1078
1079 err = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001080 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1081 if (IS_DEADDIR(nd->path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto out1;
1083
Jan Blunck4ac91372008-02-14 19:34:32 -08001084 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
Ram Pai21444402005-11-07 17:20:03 -05001085 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001088 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001089 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
Jan Blunck4ac91372008-02-14 19:34:32 -08001091 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001092 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
Jan Blunck4ac91372008-02-14 19:34:32 -08001094 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1095 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001096 goto out1;
1097 /*
1098 * Don't move a mount residing in a shared parent.
1099 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001100 if (old_nd.path.mnt->mnt_parent &&
1101 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001102 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001103 /*
1104 * Don't move a mount tree containing unbindable mounts to a destination
1105 * mount which is shared.
1106 */
Jan Blunck4ac91372008-02-14 19:34:32 -08001107 if (IS_MNT_SHARED(nd->path.mnt) &&
1108 tree_contains_unbindable(old_nd.path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001109 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 err = -ELOOP;
Jan Blunck4ac91372008-02-14 19:34:32 -08001111 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1112 if (p == old_nd.path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001113 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Jan Blunck4ac91372008-02-14 19:34:32 -08001115 err = attach_recursive_mnt(old_nd.path.mnt, nd, &parent_nd);
1116 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001117 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Ram Pai21444402005-11-07 17:20:03 -05001119 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 /* if the mount is moved, it should no longer be expire
1121 * automatically */
Jan Blunck4ac91372008-02-14 19:34:32 -08001122 list_del_init(&old_nd.path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 spin_unlock(&vfsmount_lock);
1124out1:
Jan Blunck4ac91372008-02-14 19:34:32 -08001125 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126out:
Ram Pai390c6842005-11-07 17:17:51 -05001127 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 if (!err)
Jan Blunck1d957f92008-02-14 19:34:35 -08001129 path_put(&parent_nd.path);
1130 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 return err;
1132}
1133
1134/*
1135 * create a new mount for userspace and request it to be added into the
1136 * namespace's tree
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001137 * noinline this do_mount helper to save do_mount stack space.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 */
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001139static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 int mnt_flags, char *name, void *data)
1141{
1142 struct vfsmount *mnt;
1143
1144 if (!type || !memchr(type, 0, PAGE_SIZE))
1145 return -EINVAL;
1146
1147 /* we need capabilities... */
1148 if (!capable(CAP_SYS_ADMIN))
1149 return -EPERM;
1150
1151 mnt = do_kern_mount(type, flags, name, data);
1152 if (IS_ERR(mnt))
1153 return PTR_ERR(mnt);
1154
1155 return do_add_mount(mnt, nd, mnt_flags, NULL);
1156}
1157
1158/*
1159 * add a mount into a namespace's mount tree
1160 * - provide the option of adding the new mount to an expiration list
1161 */
1162int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1163 int mnt_flags, struct list_head *fslist)
1164{
1165 int err;
1166
Ram Pai390c6842005-11-07 17:17:51 -05001167 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 /* Something was mounted here while we slept */
Jan Blunck4ac91372008-02-14 19:34:32 -08001169 while (d_mountpoint(nd->path.dentry) &&
1170 follow_down(&nd->path.mnt, &nd->path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 ;
1172 err = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001173 if (!check_mnt(nd->path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 goto unlock;
1175
1176 /* Refuse the same filesystem on the same mount point */
1177 err = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001178 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1179 nd->path.mnt->mnt_root == nd->path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 goto unlock;
1181
1182 err = -EINVAL;
1183 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1184 goto unlock;
1185
1186 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001187 if ((err = graft_tree(newmnt, nd)))
1188 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001190 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 /* add to the specified expiration list */
1192 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001193 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 spin_unlock(&vfsmount_lock);
1195 }
Ram Pai390c6842005-11-07 17:17:51 -05001196 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001197 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001200 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 mntput(newmnt);
1202 return err;
1203}
1204
1205EXPORT_SYMBOL_GPL(do_add_mount);
1206
Ram Pai70fbcdf2005-11-07 17:17:04 -05001207static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1208 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001209{
1210 spin_lock(&vfsmount_lock);
1211
1212 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001213 * Check if mount is still attached, if not, let whoever holds it deal
1214 * with the sucker
1215 */
1216 if (mnt->mnt_parent == mnt) {
1217 spin_unlock(&vfsmount_lock);
1218 return;
1219 }
1220
1221 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001222 * Check that it is still dead: the count should now be 2 - as
1223 * contributed by the vfsmount parent and the mntget above
1224 */
Ram Paia05964f2005-11-07 17:20:17 -05001225 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001226 /* delete from the namespace */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001227 touch_mnt_namespace(mnt->mnt_ns);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001228 list_del_init(&mnt->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001229 mnt->mnt_ns = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001230 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001231 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001232 } else {
1233 /*
1234 * Someone brought it back to life whilst we didn't have any
1235 * locks held so return it to the expiration list
1236 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001237 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001238 spin_unlock(&vfsmount_lock);
1239 }
1240}
1241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001243 * go through the vfsmounts we've just consigned to the graveyard to
1244 * - check that they're still dead
1245 * - delete the vfsmount from the appropriate namespace under lock
1246 * - dispose of the corpse
1247 */
1248static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1249{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001250 struct mnt_namespace *ns;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001251 struct vfsmount *mnt;
1252
1253 while (!list_empty(graveyard)) {
1254 LIST_HEAD(umounts);
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001255 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001256 list_del_init(&mnt->mnt_expire);
1257
1258 /* don't do anything if the namespace is dead - all the
1259 * vfsmounts from it are going away anyway */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001260 ns = mnt->mnt_ns;
1261 if (!ns || !ns->root)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001262 continue;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001263 get_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001264
1265 spin_unlock(&vfsmount_lock);
1266 down_write(&namespace_sem);
1267 expire_mount(mnt, mounts, &umounts);
1268 up_write(&namespace_sem);
1269 release_mounts(&umounts);
1270 mntput(mnt);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001271 put_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001272 spin_lock(&vfsmount_lock);
1273 }
1274}
1275
1276/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 * process a list of expirable mountpoints with the intent of discarding any
1278 * mountpoints that aren't in use and haven't been touched since last we came
1279 * here
1280 */
1281void mark_mounts_for_expiry(struct list_head *mounts)
1282{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 struct vfsmount *mnt, *next;
1284 LIST_HEAD(graveyard);
1285
1286 if (list_empty(mounts))
1287 return;
1288
1289 spin_lock(&vfsmount_lock);
1290
1291 /* extract from the expiration list every vfsmount that matches the
1292 * following criteria:
1293 * - only referenced by its parent vfsmount
1294 * - still marked for expiry (marked on the last call here; marks are
1295 * cleared by mntput())
1296 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001297 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1299 atomic_read(&mnt->mnt_count) != 1)
1300 continue;
1301
1302 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001303 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 }
1305
Trond Myklebust5528f9112006-06-09 09:34:17 -04001306 expire_mount_list(&graveyard, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 spin_unlock(&vfsmount_lock);
1309}
1310
1311EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1312
1313/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001314 * Ripoff of 'select_parent()'
1315 *
1316 * search the list of submounts for a given mountpoint, and move any
1317 * shrinkable submounts to the 'graveyard' list.
1318 */
1319static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1320{
1321 struct vfsmount *this_parent = parent;
1322 struct list_head *next;
1323 int found = 0;
1324
1325repeat:
1326 next = this_parent->mnt_mounts.next;
1327resume:
1328 while (next != &this_parent->mnt_mounts) {
1329 struct list_head *tmp = next;
1330 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1331
1332 next = tmp->next;
1333 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1334 continue;
1335 /*
1336 * Descend a level if the d_mounts list is non-empty.
1337 */
1338 if (!list_empty(&mnt->mnt_mounts)) {
1339 this_parent = mnt;
1340 goto repeat;
1341 }
1342
1343 if (!propagate_mount_busy(mnt, 1)) {
1344 mntget(mnt);
1345 list_move_tail(&mnt->mnt_expire, graveyard);
1346 found++;
1347 }
1348 }
1349 /*
1350 * All done at this level ... ascend and resume the search
1351 */
1352 if (this_parent != parent) {
1353 next = this_parent->mnt_child.next;
1354 this_parent = this_parent->mnt_parent;
1355 goto resume;
1356 }
1357 return found;
1358}
1359
1360/*
1361 * process a list of expirable mountpoints with the intent of discarding any
1362 * submounts of a specific parent mountpoint
1363 */
1364void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1365{
1366 LIST_HEAD(graveyard);
1367 int found;
1368
1369 spin_lock(&vfsmount_lock);
1370
1371 /* extract submounts of 'mountpoint' from the expiration list */
1372 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1373 expire_mount_list(&graveyard, mounts);
1374
1375 spin_unlock(&vfsmount_lock);
1376}
1377
1378EXPORT_SYMBOL_GPL(shrink_submounts);
1379
1380/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 * Some copy_from_user() implementations do not return the exact number of
1382 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1383 * Note that this function differs from copy_from_user() in that it will oops
1384 * on bad values of `to', rather than returning a short copy.
1385 */
Ram Paib58fed82005-11-07 17:16:09 -05001386static long exact_copy_from_user(void *to, const void __user * from,
1387 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 char *t = to;
1390 const char __user *f = from;
1391 char c;
1392
1393 if (!access_ok(VERIFY_READ, from, n))
1394 return n;
1395
1396 while (n) {
1397 if (__get_user(c, f)) {
1398 memset(t, 0, n);
1399 break;
1400 }
1401 *t++ = c;
1402 f++;
1403 n--;
1404 }
1405 return n;
1406}
1407
Ram Paib58fed82005-11-07 17:16:09 -05001408int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
1410 int i;
1411 unsigned long page;
1412 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001413
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 *where = 0;
1415 if (!data)
1416 return 0;
1417
1418 if (!(page = __get_free_page(GFP_KERNEL)))
1419 return -ENOMEM;
1420
1421 /* We only care that *some* data at the address the user
1422 * gave us is valid. Just in case, we'll zero
1423 * the remainder of the page.
1424 */
1425 /* copy_from_user cannot cross TASK_SIZE ! */
1426 size = TASK_SIZE - (unsigned long)data;
1427 if (size > PAGE_SIZE)
1428 size = PAGE_SIZE;
1429
1430 i = size - exact_copy_from_user((void *)page, data, size);
1431 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001432 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 return -EFAULT;
1434 }
1435 if (i != PAGE_SIZE)
1436 memset((char *)page + i, 0, PAGE_SIZE - i);
1437 *where = page;
1438 return 0;
1439}
1440
1441/*
1442 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1443 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1444 *
1445 * data is a (void *) that can point to any structure up to
1446 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1447 * information (or be NULL).
1448 *
1449 * Pre-0.97 versions of mount() didn't have a flags word.
1450 * When the flags word was introduced its top half was required
1451 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1452 * Therefore, if this magic number is present, it carries no information
1453 * and must be discarded.
1454 */
Ram Paib58fed82005-11-07 17:16:09 -05001455long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 unsigned long flags, void *data_page)
1457{
1458 struct nameidata nd;
1459 int retval = 0;
1460 int mnt_flags = 0;
1461
1462 /* Discard magic */
1463 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1464 flags &= ~MS_MGC_MSK;
1465
1466 /* Basic sanity checks */
1467
1468 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1469 return -EINVAL;
1470 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1471 return -EINVAL;
1472
1473 if (data_page)
1474 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1475
1476 /* Separate the per-mountpoint flags */
1477 if (flags & MS_NOSUID)
1478 mnt_flags |= MNT_NOSUID;
1479 if (flags & MS_NODEV)
1480 mnt_flags |= MNT_NODEV;
1481 if (flags & MS_NOEXEC)
1482 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001483 if (flags & MS_NOATIME)
1484 mnt_flags |= MNT_NOATIME;
1485 if (flags & MS_NODIRATIME)
1486 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001487 if (flags & MS_RELATIME)
1488 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001489
1490 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001491 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
1493 /* ... and get the mountpoint */
1494 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1495 if (retval)
1496 return retval;
1497
1498 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1499 if (retval)
1500 goto dput_out;
1501
1502 if (flags & MS_REMOUNT)
1503 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1504 data_page);
1505 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001506 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001507 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001508 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 else if (flags & MS_MOVE)
1510 retval = do_move_mount(&nd, dev_name);
1511 else
1512 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1513 dev_name, data_page);
1514dput_out:
Jan Blunck1d957f92008-02-14 19:34:35 -08001515 path_put(&nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return retval;
1517}
1518
JANAK DESAI741a2952006-02-07 12:59:00 -08001519/*
1520 * Allocate a new namespace structure and populate it with contents
1521 * copied from the namespace of the passed in task structure.
1522 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001523static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001524 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001526 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 struct vfsmount *p, *q;
1529
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001530 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001532 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
1534 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001536 init_waitqueue_head(&new_ns->poll);
1537 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Ram Pai390c6842005-11-07 17:17:51 -05001539 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001541 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001542 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001544 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001546 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
1548 spin_lock(&vfsmount_lock);
1549 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1550 spin_unlock(&vfsmount_lock);
1551
1552 /*
1553 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1554 * as belonging to new namespace. We have already acquired a private
1555 * fs_struct, so tsk->fs->lock is not needed.
1556 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001557 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 q = new_ns->root;
1559 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001560 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001562 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001564 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001566 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001568 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08001570 if (p == fs->altroot.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 altrootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08001572 fs->altroot.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 }
1574 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001575 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 q = next_mnt(q, new_ns->root);
1577 }
Ram Pai390c6842005-11-07 17:17:51 -05001578 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 if (rootmnt)
1581 mntput(rootmnt);
1582 if (pwdmnt)
1583 mntput(pwdmnt);
1584 if (altrootmnt)
1585 mntput(altrootmnt);
1586
JANAK DESAI741a2952006-02-07 12:59:00 -08001587 return new_ns;
1588}
1589
Eric W. Biederman213dd262007-07-15 23:41:15 -07001590struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07001591 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001592{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001593 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001594
Badari Pulavartye3222c42007-05-08 00:25:21 -07001595 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001596 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001597
1598 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001599 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001600
Badari Pulavartye3222c42007-05-08 00:25:21 -07001601 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001602
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001603 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001604 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605}
1606
1607asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1608 char __user * type, unsigned long flags,
1609 void __user * data)
1610{
1611 int retval;
1612 unsigned long data_page;
1613 unsigned long type_page;
1614 unsigned long dev_page;
1615 char *dir_page;
1616
Ram Paib58fed82005-11-07 17:16:09 -05001617 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 if (retval < 0)
1619 return retval;
1620
1621 dir_page = getname(dir_name);
1622 retval = PTR_ERR(dir_page);
1623 if (IS_ERR(dir_page))
1624 goto out1;
1625
Ram Paib58fed82005-11-07 17:16:09 -05001626 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 if (retval < 0)
1628 goto out2;
1629
Ram Paib58fed82005-11-07 17:16:09 -05001630 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 if (retval < 0)
1632 goto out3;
1633
1634 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001635 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1636 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 unlock_kernel();
1638 free_page(data_page);
1639
1640out3:
1641 free_page(dev_page);
1642out2:
1643 putname(dir_page);
1644out1:
1645 free_page(type_page);
1646 return retval;
1647}
1648
1649/*
1650 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1651 * It can block. Requires the big lock held.
1652 */
Jan Blunckac748a02008-02-14 19:34:39 -08001653void set_fs_root(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001655 struct path old_root;
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 write_lock(&fs->lock);
1658 old_root = fs->root;
Jan Blunckac748a02008-02-14 19:34:39 -08001659 fs->root = *path;
1660 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 write_unlock(&fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001662 if (old_root.dentry)
1663 path_put(&old_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664}
1665
1666/*
1667 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1668 * It can block. Requires the big lock held.
1669 */
Jan Blunckac748a02008-02-14 19:34:39 -08001670void set_fs_pwd(struct fs_struct *fs, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
Jan Blunck6ac08c32008-02-14 19:34:38 -08001672 struct path old_pwd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
1674 write_lock(&fs->lock);
1675 old_pwd = fs->pwd;
Jan Blunckac748a02008-02-14 19:34:39 -08001676 fs->pwd = *path;
1677 path_get(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 write_unlock(&fs->lock);
1679
Jan Blunck6ac08c32008-02-14 19:34:38 -08001680 if (old_pwd.dentry)
1681 path_put(&old_pwd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682}
1683
1684static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1685{
1686 struct task_struct *g, *p;
1687 struct fs_struct *fs;
1688
1689 read_lock(&tasklist_lock);
1690 do_each_thread(g, p) {
1691 task_lock(p);
1692 fs = p->fs;
1693 if (fs) {
1694 atomic_inc(&fs->count);
1695 task_unlock(p);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001696 if (fs->root.dentry == old_nd->path.dentry
1697 && fs->root.mnt == old_nd->path.mnt)
Jan Blunckac748a02008-02-14 19:34:39 -08001698 set_fs_root(fs, &new_nd->path);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001699 if (fs->pwd.dentry == old_nd->path.dentry
1700 && fs->pwd.mnt == old_nd->path.mnt)
Jan Blunckac748a02008-02-14 19:34:39 -08001701 set_fs_pwd(fs, &new_nd->path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 put_fs_struct(fs);
1703 } else
1704 task_unlock(p);
1705 } while_each_thread(g, p);
1706 read_unlock(&tasklist_lock);
1707}
1708
1709/*
1710 * pivot_root Semantics:
1711 * Moves the root file system of the current process to the directory put_old,
1712 * makes new_root as the new root file system of the current process, and sets
1713 * root/cwd of all processes which had them on the current root to new_root.
1714 *
1715 * Restrictions:
1716 * The new_root and put_old must be directories, and must not be on the
1717 * same file system as the current process root. The put_old must be
1718 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1719 * pointed to by put_old must yield the same directory as new_root. No other
1720 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1721 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001722 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1723 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1724 * in this situation.
1725 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 * Notes:
1727 * - we don't move root/cwd if they are not at the root (reason: if something
1728 * cared enough to change them, it's probably wrong to force them elsewhere)
1729 * - it's okay to pick a root that isn't the root of a file system, e.g.
1730 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1731 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1732 * first.
1733 */
Ram Paib58fed82005-11-07 17:16:09 -05001734asmlinkage long sys_pivot_root(const char __user * new_root,
1735 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736{
1737 struct vfsmount *tmp;
1738 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1739 int error;
1740
1741 if (!capable(CAP_SYS_ADMIN))
1742 return -EPERM;
1743
1744 lock_kernel();
1745
Ram Paib58fed82005-11-07 17:16:09 -05001746 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1747 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 if (error)
1749 goto out0;
1750 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001751 if (!check_mnt(new_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 goto out1;
1753
Ram Paib58fed82005-11-07 17:16:09 -05001754 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 if (error)
1756 goto out1;
1757
1758 error = security_sb_pivotroot(&old_nd, &new_nd);
1759 if (error) {
Jan Blunck1d957f92008-02-14 19:34:35 -08001760 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 goto out1;
1762 }
1763
1764 read_lock(&current->fs->lock);
Jan Blunck6ac08c32008-02-14 19:34:38 -08001765 user_nd.path = current->fs->root;
1766 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001768 down_write(&namespace_sem);
Jan Blunck4ac91372008-02-14 19:34:32 -08001769 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001771 if (IS_MNT_SHARED(old_nd.path.mnt) ||
1772 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
1773 IS_MNT_SHARED(user_nd.path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001774 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001775 if (!check_mnt(user_nd.path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 goto out2;
1777 error = -ENOENT;
Jan Blunck4ac91372008-02-14 19:34:32 -08001778 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001780 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781 goto out2;
Jan Blunck4ac91372008-02-14 19:34:32 -08001782 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto out2;
1784 error = -EBUSY;
Jan Blunck4ac91372008-02-14 19:34:32 -08001785 if (new_nd.path.mnt == user_nd.path.mnt ||
1786 old_nd.path.mnt == user_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 goto out2; /* loop, on the same file system */
1788 error = -EINVAL;
Jan Blunck4ac91372008-02-14 19:34:32 -08001789 if (user_nd.path.mnt->mnt_root != user_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001791 if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001792 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001793 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794 goto out2; /* not a mountpoint */
Jan Blunck4ac91372008-02-14 19:34:32 -08001795 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001796 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08001797 /* make sure we can reach put_old from new_root */
1798 tmp = old_nd.path.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 spin_lock(&vfsmount_lock);
Jan Blunck4ac91372008-02-14 19:34:32 -08001800 if (tmp != new_nd.path.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 for (;;) {
1802 if (tmp->mnt_parent == tmp)
1803 goto out3; /* already mounted on put_old */
Jan Blunck4ac91372008-02-14 19:34:32 -08001804 if (tmp->mnt_parent == new_nd.path.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 break;
1806 tmp = tmp->mnt_parent;
1807 }
Jan Blunck4ac91372008-02-14 19:34:32 -08001808 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08001810 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto out3;
Jan Blunck4ac91372008-02-14 19:34:32 -08001812 detach_mnt(new_nd.path.mnt, &parent_nd);
1813 detach_mnt(user_nd.path.mnt, &root_parent);
1814 /* mount old root on put_old */
1815 attach_mnt(user_nd.path.mnt, &old_nd);
1816 /* mount new_root on / */
1817 attach_mnt(new_nd.path.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001818 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 spin_unlock(&vfsmount_lock);
1820 chroot_fs_refs(&user_nd, &new_nd);
1821 security_sb_post_pivotroot(&user_nd, &new_nd);
1822 error = 0;
Jan Blunck1d957f92008-02-14 19:34:35 -08001823 path_put(&root_parent.path);
1824 path_put(&parent_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825out2:
Jan Blunck4ac91372008-02-14 19:34:32 -08001826 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001827 up_write(&namespace_sem);
Jan Blunck1d957f92008-02-14 19:34:35 -08001828 path_put(&user_nd.path);
1829 path_put(&old_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830out1:
Jan Blunck1d957f92008-02-14 19:34:35 -08001831 path_put(&new_nd.path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832out0:
1833 unlock_kernel();
1834 return error;
1835out3:
1836 spin_unlock(&vfsmount_lock);
1837 goto out2;
1838}
1839
1840static void __init init_mount_tree(void)
1841{
1842 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001843 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08001844 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845
1846 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1847 if (IS_ERR(mnt))
1848 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001849 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1850 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001852 atomic_set(&ns->count, 1);
1853 INIT_LIST_HEAD(&ns->list);
1854 init_waitqueue_head(&ns->poll);
1855 ns->event = 0;
1856 list_add(&mnt->mnt_list, &ns->list);
1857 ns->root = mnt;
1858 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001860 init_task.nsproxy->mnt_ns = ns;
1861 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Jan Blunckac748a02008-02-14 19:34:39 -08001863 root.mnt = ns->root;
1864 root.dentry = ns->root->mnt_root;
1865
1866 set_fs_pwd(current->fs, &root);
1867 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868}
1869
Denis Cheng74bf17c2007-10-16 23:26:30 -07001870void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
Eric Dumazet13f14b42008-02-06 01:37:57 -08001872 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001873 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
Ram Pai390c6842005-11-07 17:17:51 -05001875 init_rwsem(&namespace_sem);
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09001878 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879
Ram Paib58fed82005-11-07 17:16:09 -05001880 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881
1882 if (!mount_hashtable)
1883 panic("Failed to allocate mount hash table\n");
1884
Eric Dumazet13f14b42008-02-06 01:37:57 -08001885 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
Eric Dumazet13f14b42008-02-06 01:37:57 -08001887 for (u = 0; u < HASH_SIZE; u++)
1888 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001890 err = sysfs_init();
1891 if (err)
1892 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1893 __FUNCTION__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06001894 fs_kobj = kobject_create_and_add("fs", NULL);
1895 if (!fs_kobj)
1896 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 init_rootfs();
1898 init_mount_tree();
1899}
1900
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001901void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001903 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001904 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001905 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001906 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001907 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001909 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001911 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001912 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001913 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914}