blob: c5b88100d914d8ae618261cb38cf2af652849340 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/uaccess.h>
29#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050030#include "pnode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050033__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
34
35static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Eric Dumazetfa3536c2006-03-26 01:37:24 -080037static struct list_head *mount_hashtable __read_mostly;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070038static int hash_mask __read_mostly, hash_bits __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080039static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050040static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080042/* /sys/fs */
43decl_subsys(fs, NULL, NULL);
44EXPORT_SYMBOL_GPL(fs_subsys);
45
Linus Torvalds1da177e2005-04-16 15:20:36 -070046static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
47{
Ram Paib58fed82005-11-07 17:16:09 -050048 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
49 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070050 tmp = tmp + (tmp >> hash_bits);
51 return tmp & hash_mask;
52}
53
54struct vfsmount *alloc_vfsmnt(const char *name)
55{
Robert P. J. Dayc3762222007-02-10 01:45:03 -080056 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057 if (mnt) {
Ram Paib58fed82005-11-07 17:16:09 -050058 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 INIT_LIST_HEAD(&mnt->mnt_hash);
60 INIT_LIST_HEAD(&mnt->mnt_child);
61 INIT_LIST_HEAD(&mnt->mnt_mounts);
62 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070063 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050064 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050065 INIT_LIST_HEAD(&mnt->mnt_slave_list);
66 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050068 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 char *newname = kmalloc(size, GFP_KERNEL);
70 if (newname) {
71 memcpy(newname, name, size);
72 mnt->mnt_devname = newname;
73 }
74 }
75 }
76 return mnt;
77}
78
David Howells454e2392006-06-23 02:02:57 -070079int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
80{
81 mnt->mnt_sb = sb;
82 mnt->mnt_root = dget(sb->s_root);
83 return 0;
84}
85
86EXPORT_SYMBOL(simple_set_mnt);
87
Linus Torvalds1da177e2005-04-16 15:20:36 -070088void free_vfsmnt(struct vfsmount *mnt)
89{
90 kfree(mnt->mnt_devname);
91 kmem_cache_free(mnt_cache, mnt);
92}
93
94/*
Ram Paia05964f2005-11-07 17:20:17 -050095 * find the first or last mount at @dentry on vfsmount @mnt depending on
96 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 */
Ram Paia05964f2005-11-07 17:20:17 -050098struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
99 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100{
Ram Paib58fed82005-11-07 17:16:09 -0500101 struct list_head *head = mount_hashtable + hash(mnt, dentry);
102 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 struct vfsmount *p, *found = NULL;
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500106 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 p = NULL;
108 if (tmp == head)
109 break;
110 p = list_entry(tmp, struct vfsmount, mnt_hash);
111 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500112 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 break;
114 }
115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 return found;
117}
118
Ram Paia05964f2005-11-07 17:20:17 -0500119/*
120 * lookup_mnt increments the ref count before returning
121 * the vfsmount struct.
122 */
123struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
124{
125 struct vfsmount *child_mnt;
126 spin_lock(&vfsmount_lock);
127 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
128 mntget(child_mnt);
129 spin_unlock(&vfsmount_lock);
130 return child_mnt;
131}
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133static inline int check_mnt(struct vfsmount *mnt)
134{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800135 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800138static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500139{
140 if (ns) {
141 ns->event = ++event;
142 wake_up_interruptible(&ns->poll);
143 }
144}
145
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800146static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500147{
148 if (ns && ns->event != event) {
149 ns->event = event;
150 wake_up_interruptible(&ns->poll);
151 }
152}
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
155{
156 old_nd->dentry = mnt->mnt_mountpoint;
157 old_nd->mnt = mnt->mnt_parent;
158 mnt->mnt_parent = mnt;
159 mnt->mnt_mountpoint = mnt->mnt_root;
160 list_del_init(&mnt->mnt_child);
161 list_del_init(&mnt->mnt_hash);
162 old_nd->dentry->d_mounted--;
163}
164
Ram Paib90fa9a2005-11-07 17:19:50 -0500165void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
166 struct vfsmount *child_mnt)
167{
168 child_mnt->mnt_parent = mntget(mnt);
169 child_mnt->mnt_mountpoint = dget(dentry);
170 dentry->d_mounted++;
171}
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
174{
Ram Paib90fa9a2005-11-07 17:19:50 -0500175 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
176 list_add_tail(&mnt->mnt_hash, mount_hashtable +
177 hash(nd->mnt, nd->dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500179}
180
181/*
182 * the caller must hold vfsmount_lock
183 */
184static void commit_tree(struct vfsmount *mnt)
185{
186 struct vfsmount *parent = mnt->mnt_parent;
187 struct vfsmount *m;
188 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800189 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500190
191 BUG_ON(parent == mnt);
192
193 list_add_tail(&head, &mnt->mnt_list);
194 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800195 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500196 list_splice(&head, n->list.prev);
197
198 list_add_tail(&mnt->mnt_hash, mount_hashtable +
199 hash(parent, mnt->mnt_mountpoint));
200 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800201 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202}
203
204static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
205{
206 struct list_head *next = p->mnt_mounts.next;
207 if (next == &p->mnt_mounts) {
208 while (1) {
209 if (p == root)
210 return NULL;
211 next = p->mnt_child.next;
212 if (next != &p->mnt_parent->mnt_mounts)
213 break;
214 p = p->mnt_parent;
215 }
216 }
217 return list_entry(next, struct vfsmount, mnt_child);
218}
219
Ram Pai9676f0c2005-11-07 17:21:20 -0500220static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
221{
222 struct list_head *prev = p->mnt_mounts.prev;
223 while (prev != &p->mnt_mounts) {
224 p = list_entry(prev, struct vfsmount, mnt_child);
225 prev = p->mnt_mounts.prev;
226 }
227 return p;
228}
229
Ram Pai36341f62005-11-07 17:17:22 -0500230static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
231 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
233 struct super_block *sb = old->mnt_sb;
234 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
235
236 if (mnt) {
237 mnt->mnt_flags = old->mnt_flags;
238 atomic_inc(&sb->s_active);
239 mnt->mnt_sb = sb;
240 mnt->mnt_root = dget(root);
241 mnt->mnt_mountpoint = mnt->mnt_root;
242 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500243
Ram Pai5afe0022005-11-07 17:21:01 -0500244 if (flag & CL_SLAVE) {
245 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
246 mnt->mnt_master = old;
247 CLEAR_MNT_SHARED(mnt);
248 } else {
249 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
250 list_add(&mnt->mnt_share, &old->mnt_share);
251 if (IS_MNT_SLAVE(old))
252 list_add(&mnt->mnt_slave, &old->mnt_slave);
253 mnt->mnt_master = old->mnt_master;
254 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500255 if (flag & CL_MAKE_SHARED)
256 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 /* stick the duplicate mount on the same expiry list
259 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500260 if (flag & CL_EXPIRE) {
261 spin_lock(&vfsmount_lock);
262 if (!list_empty(&old->mnt_expire))
263 list_add(&mnt->mnt_expire, &old->mnt_expire);
264 spin_unlock(&vfsmount_lock);
265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 }
267 return mnt;
268}
269
Al Viro7b7b1ac2005-11-07 17:13:39 -0500270static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 struct super_block *sb = mnt->mnt_sb;
273 dput(mnt->mnt_root);
274 free_vfsmnt(mnt);
275 deactivate_super(sb);
276}
277
Al Viro7b7b1ac2005-11-07 17:13:39 -0500278void mntput_no_expire(struct vfsmount *mnt)
279{
280repeat:
281 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
282 if (likely(!mnt->mnt_pinned)) {
283 spin_unlock(&vfsmount_lock);
284 __mntput(mnt);
285 return;
286 }
287 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
288 mnt->mnt_pinned = 0;
289 spin_unlock(&vfsmount_lock);
290 acct_auto_close_mnt(mnt);
291 security_sb_umount_close(mnt);
292 goto repeat;
293 }
294}
295
296EXPORT_SYMBOL(mntput_no_expire);
297
298void mnt_pin(struct vfsmount *mnt)
299{
300 spin_lock(&vfsmount_lock);
301 mnt->mnt_pinned++;
302 spin_unlock(&vfsmount_lock);
303}
304
305EXPORT_SYMBOL(mnt_pin);
306
307void mnt_unpin(struct vfsmount *mnt)
308{
309 spin_lock(&vfsmount_lock);
310 if (mnt->mnt_pinned) {
311 atomic_inc(&mnt->mnt_count);
312 mnt->mnt_pinned--;
313 }
314 spin_unlock(&vfsmount_lock);
315}
316
317EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319/* iterator */
320static void *m_start(struct seq_file *m, loff_t *pos)
321{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800322 struct mnt_namespace *n = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 struct list_head *p;
324 loff_t l = *pos;
325
Ram Pai390c6842005-11-07 17:17:51 -0500326 down_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 list_for_each(p, &n->list)
328 if (!l--)
329 return list_entry(p, struct vfsmount, mnt_list);
330 return NULL;
331}
332
333static void *m_next(struct seq_file *m, void *v, loff_t *pos)
334{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800335 struct mnt_namespace *n = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
337 (*pos)++;
Ram Paib58fed82005-11-07 17:16:09 -0500338 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339}
340
341static void m_stop(struct seq_file *m, void *v)
342{
Ram Pai390c6842005-11-07 17:17:51 -0500343 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346static inline void mangle(struct seq_file *m, const char *s)
347{
348 seq_escape(m, s, " \t\n\\");
349}
350
351static int show_vfsmnt(struct seq_file *m, void *v)
352{
353 struct vfsmount *mnt = v;
354 int err = 0;
355 static struct proc_fs_info {
356 int flag;
357 char *str;
358 } fs_info[] = {
359 { MS_SYNCHRONOUS, ",sync" },
360 { MS_DIRSYNC, ",dirsync" },
361 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 { 0, NULL }
363 };
364 static struct proc_fs_info mnt_info[] = {
365 { MNT_NOSUID, ",nosuid" },
366 { MNT_NODEV, ",nodev" },
367 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800368 { MNT_NOATIME, ",noatime" },
369 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800370 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 { 0, NULL }
372 };
373 struct proc_fs_info *fs_infop;
374
375 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
376 seq_putc(m, ' ');
377 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
378 seq_putc(m, ' ');
379 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700380 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
381 seq_putc(m, '.');
382 mangle(m, mnt->mnt_sb->s_subtype);
383 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
385 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
386 if (mnt->mnt_sb->s_flags & fs_infop->flag)
387 seq_puts(m, fs_infop->str);
388 }
389 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
390 if (mnt->mnt_flags & fs_infop->flag)
391 seq_puts(m, fs_infop->str);
392 }
393 if (mnt->mnt_sb->s_op->show_options)
394 err = mnt->mnt_sb->s_op->show_options(m, mnt);
395 seq_puts(m, " 0 0\n");
396 return err;
397}
398
399struct seq_operations mounts_op = {
400 .start = m_start,
401 .next = m_next,
402 .stop = m_stop,
403 .show = show_vfsmnt
404};
405
Chuck Leverb4629fe2006-03-20 13:44:12 -0500406static int show_vfsstat(struct seq_file *m, void *v)
407{
408 struct vfsmount *mnt = v;
409 int err = 0;
410
411 /* device */
412 if (mnt->mnt_devname) {
413 seq_puts(m, "device ");
414 mangle(m, mnt->mnt_devname);
415 } else
416 seq_puts(m, "no device");
417
418 /* mount point */
419 seq_puts(m, " mounted on ");
420 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
421 seq_putc(m, ' ');
422
423 /* file system type */
424 seq_puts(m, "with fstype ");
425 mangle(m, mnt->mnt_sb->s_type->name);
426
427 /* optional statistics */
428 if (mnt->mnt_sb->s_op->show_stats) {
429 seq_putc(m, ' ');
430 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
431 }
432
433 seq_putc(m, '\n');
434 return err;
435}
436
437struct seq_operations mountstats_op = {
438 .start = m_start,
439 .next = m_next,
440 .stop = m_stop,
441 .show = show_vfsstat,
442};
443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444/**
445 * may_umount_tree - check if a mount tree is busy
446 * @mnt: root of mount tree
447 *
448 * This is called to check if a tree of mounts has any
449 * open files, pwds, chroots or sub mounts that are
450 * busy.
451 */
452int may_umount_tree(struct vfsmount *mnt)
453{
Ram Pai36341f62005-11-07 17:17:22 -0500454 int actual_refs = 0;
455 int minimum_refs = 0;
456 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500459 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 actual_refs += atomic_read(&p->mnt_count);
461 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463 spin_unlock(&vfsmount_lock);
464
465 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800466 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Ian Kente3474a82006-03-27 01:14:51 -0800468 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
471EXPORT_SYMBOL(may_umount_tree);
472
473/**
474 * may_umount - check if a mount point is busy
475 * @mnt: root of mount
476 *
477 * This is called to check if a mount point has any
478 * open files, pwds, chroots or sub mounts. If the
479 * mount has sub mounts this will return busy
480 * regardless of whether the sub mounts are busy.
481 *
482 * Doesn't take quota and stuff into account. IOW, in some cases it will
483 * give false negatives. The main reason why it's here is that we need
484 * a non-destructive way to look for easily umountable filesystems.
485 */
486int may_umount(struct vfsmount *mnt)
487{
Ian Kente3474a82006-03-27 01:14:51 -0800488 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500489 spin_lock(&vfsmount_lock);
490 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800491 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500492 spin_unlock(&vfsmount_lock);
493 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
496EXPORT_SYMBOL(may_umount);
497
Ram Paib90fa9a2005-11-07 17:19:50 -0500498void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500500 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800501 while (!list_empty(head)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500502 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
503 list_del_init(&mnt->mnt_hash);
504 if (mnt->mnt_parent != mnt) {
505 struct dentry *dentry;
506 struct vfsmount *m;
507 spin_lock(&vfsmount_lock);
508 dentry = mnt->mnt_mountpoint;
509 m = mnt->mnt_parent;
510 mnt->mnt_mountpoint = mnt->mnt_root;
511 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500513 dput(dentry);
514 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 }
516 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500517 }
518}
519
Ram Paia05964f2005-11-07 17:20:17 -0500520void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500521{
522 struct vfsmount *p;
523
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700524 for (p = mnt; p; p = next_mnt(p, mnt))
525 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500526
Ram Paia05964f2005-11-07 17:20:17 -0500527 if (propagate)
528 propagate_umount(kill);
529
Ram Pai70fbcdf2005-11-07 17:17:04 -0500530 list_for_each_entry(p, kill, mnt_hash) {
531 list_del_init(&p->mnt_expire);
532 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800533 __touch_mnt_namespace(p->mnt_ns);
534 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500535 list_del_init(&p->mnt_child);
536 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500537 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500538 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
540}
541
542static int do_umount(struct vfsmount *mnt, int flags)
543{
Ram Paib58fed82005-11-07 17:16:09 -0500544 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500546 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547
548 retval = security_sb_umount(mnt, flags);
549 if (retval)
550 return retval;
551
552 /*
553 * Allow userspace to request a mountpoint be expired rather than
554 * unmounting unconditionally. Unmount only happens if:
555 * (1) the mark is already set (the mark is cleared by mntput())
556 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
557 */
558 if (flags & MNT_EXPIRE) {
559 if (mnt == current->fs->rootmnt ||
560 flags & (MNT_FORCE | MNT_DETACH))
561 return -EINVAL;
562
563 if (atomic_read(&mnt->mnt_count) != 2)
564 return -EBUSY;
565
566 if (!xchg(&mnt->mnt_expiry_mark, 1))
567 return -EAGAIN;
568 }
569
570 /*
571 * If we may have to abort operations to get out of this
572 * mount, and they will themselves hold resources we must
573 * allow the fs to do things. In the Unix tradition of
574 * 'Gee thats tricky lets do it in userspace' the umount_begin
575 * might fail to complete on the first run through as other tasks
576 * must return, and the like. Thats for the mount program to worry
577 * about for the moment.
578 */
579
580 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400581 if (sb->s_op->umount_begin)
582 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 unlock_kernel();
584
585 /*
586 * No sense to grab the lock for this test, but test itself looks
587 * somewhat bogus. Suggestions for better replacement?
588 * Ho-hum... In principle, we might treat that as umount + switch
589 * to rootfs. GC would eventually take care of the old vfsmount.
590 * Actually it makes sense, especially if rootfs would contain a
591 * /reboot - static binary that would close all descriptors and
592 * call reboot(9). Then init(8) could umount root and exec /reboot.
593 */
594 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
595 /*
596 * Special case for "unmounting" root ...
597 * we just try to remount it readonly.
598 */
599 down_write(&sb->s_umount);
600 if (!(sb->s_flags & MS_RDONLY)) {
601 lock_kernel();
602 DQUOT_OFF(sb);
603 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
604 unlock_kernel();
605 }
606 up_write(&sb->s_umount);
607 return retval;
608 }
609
Ram Pai390c6842005-11-07 17:17:51 -0500610 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500612 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500615 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500617 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 retval = 0;
619 }
620 spin_unlock(&vfsmount_lock);
621 if (retval)
622 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500623 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500624 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 return retval;
626}
627
628/*
629 * Now umount can handle mount points as well as block devices.
630 * This is important for filesystems which use unnamed block devices.
631 *
632 * We now support a flag for forced unmount like the other 'big iron'
633 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
634 */
635
636asmlinkage long sys_umount(char __user * name, int flags)
637{
638 struct nameidata nd;
639 int retval;
640
641 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
642 if (retval)
643 goto out;
644 retval = -EINVAL;
645 if (nd.dentry != nd.mnt->mnt_root)
646 goto dput_and_out;
647 if (!check_mnt(nd.mnt))
648 goto dput_and_out;
649
650 retval = -EPERM;
651 if (!capable(CAP_SYS_ADMIN))
652 goto dput_and_out;
653
654 retval = do_umount(nd.mnt, flags);
655dput_and_out:
656 path_release_on_umount(&nd);
657out:
658 return retval;
659}
660
661#ifdef __ARCH_WANT_SYS_OLDUMOUNT
662
663/*
Ram Paib58fed82005-11-07 17:16:09 -0500664 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666asmlinkage long sys_oldumount(char __user * name)
667{
Ram Paib58fed82005-11-07 17:16:09 -0500668 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669}
670
671#endif
672
673static int mount_is_safe(struct nameidata *nd)
674{
675 if (capable(CAP_SYS_ADMIN))
676 return 0;
677 return -EPERM;
678#ifdef notyet
679 if (S_ISLNK(nd->dentry->d_inode->i_mode))
680 return -EPERM;
681 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
682 if (current->uid != nd->dentry->d_inode->i_uid)
683 return -EPERM;
684 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800685 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 return -EPERM;
687 return 0;
688#endif
689}
690
Ram Paib58fed82005-11-07 17:16:09 -0500691static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
693 while (1) {
694 if (d == dentry)
695 return 1;
696 if (d == NULL || d == d->d_parent)
697 return 0;
698 d = d->d_parent;
699 }
700}
701
Ram Paib90fa9a2005-11-07 17:19:50 -0500702struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500703 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704{
705 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 struct nameidata nd;
707
Ram Pai9676f0c2005-11-07 17:21:20 -0500708 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
709 return NULL;
710
Ram Pai36341f62005-11-07 17:17:22 -0500711 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (!q)
713 goto Enomem;
714 q->mnt_mountpoint = mnt->mnt_mountpoint;
715
716 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700717 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
719 continue;
720
721 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500722 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
723 s = skip_mnt_tree(s);
724 continue;
725 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 while (p != s->mnt_parent) {
727 p = p->mnt_parent;
728 q = q->mnt_parent;
729 }
730 p = s;
731 nd.mnt = q;
732 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500733 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (!q)
735 goto Enomem;
736 spin_lock(&vfsmount_lock);
737 list_add_tail(&q->mnt_list, &res->mnt_list);
738 attach_mnt(q, &nd);
739 spin_unlock(&vfsmount_lock);
740 }
741 }
742 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500743Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500745 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500747 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500749 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751 return NULL;
752}
753
Ram Paib90fa9a2005-11-07 17:19:50 -0500754/*
755 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500756 * @nd : place the mount tree @source_mnt is attached
757 * @parent_nd : if non-null, detach the source_mnt from its parent and
758 * store the parent mount and mountpoint dentry.
759 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500760 *
761 * NOTE: in the table below explains the semantics when a source mount
762 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500763 * ---------------------------------------------------------------------------
764 * | BIND MOUNT OPERATION |
765 * |**************************************************************************
766 * | source-->| shared | private | slave | unbindable |
767 * | dest | | | | |
768 * | | | | | | |
769 * | v | | | | |
770 * |**************************************************************************
771 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
772 * | | | | | |
773 * |non-shared| shared (+) | private | slave (*) | invalid |
774 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500775 * A bind operation clones the source mount and mounts the clone on the
776 * destination mount.
777 *
778 * (++) the cloned mount is propagated to all the mounts in the propagation
779 * tree of the destination mount and the cloned mount is added to
780 * the peer group of the source mount.
781 * (+) the cloned mount is created under the destination mount and is marked
782 * as shared. The cloned mount is added to the peer group of the source
783 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500784 * (+++) the mount is propagated to all the mounts in the propagation tree
785 * of the destination mount and the cloned mount is made slave
786 * of the same master as that of the source mount. The cloned mount
787 * is marked as 'shared and slave'.
788 * (*) the cloned mount is made a slave of the same master as that of the
789 * source mount.
790 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500791 * ---------------------------------------------------------------------------
792 * | MOVE MOUNT OPERATION |
793 * |**************************************************************************
794 * | source-->| shared | private | slave | unbindable |
795 * | dest | | | | |
796 * | | | | | | |
797 * | v | | | | |
798 * |**************************************************************************
799 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
800 * | | | | | |
801 * |non-shared| shared (+*) | private | slave (*) | unbindable |
802 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500803 *
804 * (+) the mount is moved to the destination. And is then propagated to
805 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500806 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500807 * (+++) the mount is moved to the destination and is then propagated to
808 * all the mounts belonging to the destination mount's propagation tree.
809 * the mount is marked as 'shared and slave'.
810 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500811 *
812 * if the source mount is a tree, the operations explained above is
813 * applied to each mount in the tree.
814 * Must be called without spinlocks held, since this function can sleep
815 * in allocations.
816 */
817static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500818 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500819{
820 LIST_HEAD(tree_list);
821 struct vfsmount *dest_mnt = nd->mnt;
822 struct dentry *dest_dentry = nd->dentry;
823 struct vfsmount *child, *p;
824
825 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
826 return -EINVAL;
827
828 if (IS_MNT_SHARED(dest_mnt)) {
829 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
830 set_mnt_shared(p);
831 }
832
833 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500834 if (parent_nd) {
835 detach_mnt(source_mnt, parent_nd);
836 attach_mnt(source_mnt, nd);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800837 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500838 } else {
839 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
840 commit_tree(source_mnt);
841 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500842
843 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
844 list_del_init(&child->mnt_hash);
845 commit_tree(child);
846 }
847 spin_unlock(&vfsmount_lock);
848 return 0;
849}
850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
852{
853 int err;
854 if (mnt->mnt_sb->s_flags & MS_NOUSER)
855 return -EINVAL;
856
857 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
858 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
859 return -ENOTDIR;
860
861 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800862 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (IS_DEADDIR(nd->dentry->d_inode))
864 goto out_unlock;
865
866 err = security_sb_check_sb(mnt, nd);
867 if (err)
868 goto out_unlock;
869
870 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500871 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500872 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800874 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (!err)
876 security_sb_post_addmount(mnt, nd);
877 return err;
878}
879
880/*
Ram Pai07b20882005-11-07 17:19:07 -0500881 * recursively change the type of the mountpoint.
882 */
883static int do_change_type(struct nameidata *nd, int flag)
884{
885 struct vfsmount *m, *mnt = nd->mnt;
886 int recurse = flag & MS_REC;
887 int type = flag & ~MS_REC;
888
889 if (nd->dentry != nd->mnt->mnt_root)
890 return -EINVAL;
891
892 down_write(&namespace_sem);
893 spin_lock(&vfsmount_lock);
894 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
895 change_mnt_propagation(m, type);
896 spin_unlock(&vfsmount_lock);
897 up_write(&namespace_sem);
898 return 0;
899}
900
901/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 * do loopback mount.
903 */
Andrew Mortoneee391a2006-05-15 09:44:30 -0700904static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905{
906 struct nameidata old_nd;
907 struct vfsmount *mnt = NULL;
908 int err = mount_is_safe(nd);
909 if (err)
910 return err;
911 if (!old_name || !*old_name)
912 return -EINVAL;
913 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
914 if (err)
915 return err;
916
Ram Pai390c6842005-11-07 17:17:51 -0500917 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500919 if (IS_MNT_UNBINDABLE(old_nd.mnt))
920 goto out;
921
Al Viroccd48bc2005-11-07 17:15:04 -0500922 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
923 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924
Al Viroccd48bc2005-11-07 17:15:04 -0500925 err = -ENOMEM;
926 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500927 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500928 else
Ram Pai36341f62005-11-07 17:17:22 -0500929 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500930
931 if (!mnt)
932 goto out;
933
Al Viroccd48bc2005-11-07 17:15:04 -0500934 err = graft_tree(mnt, nd);
935 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500936 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500938 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500940 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -0500941 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Al Viroccd48bc2005-11-07 17:15:04 -0500943out:
Ram Pai390c6842005-11-07 17:17:51 -0500944 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 path_release(&old_nd);
946 return err;
947}
948
949/*
950 * change filesystem flags. dir should be a physical root of filesystem.
951 * If you've mounted a non-root directory somewhere and want to do remount
952 * on it - tough luck.
953 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
955 void *data)
956{
957 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500958 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 if (!capable(CAP_SYS_ADMIN))
961 return -EPERM;
962
963 if (!check_mnt(nd->mnt))
964 return -EINVAL;
965
966 if (nd->dentry != nd->mnt->mnt_root)
967 return -EINVAL;
968
969 down_write(&sb->s_umount);
970 err = do_remount_sb(sb, flags, data, 0);
971 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500972 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 up_write(&sb->s_umount);
974 if (!err)
975 security_sb_post_remount(nd->mnt, flags, data);
976 return err;
977}
978
Ram Pai9676f0c2005-11-07 17:21:20 -0500979static inline int tree_contains_unbindable(struct vfsmount *mnt)
980{
981 struct vfsmount *p;
982 for (p = mnt; p; p = next_mnt(p, mnt)) {
983 if (IS_MNT_UNBINDABLE(p))
984 return 1;
985 }
986 return 0;
987}
988
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989static int do_move_mount(struct nameidata *nd, char *old_name)
990{
991 struct nameidata old_nd, parent_nd;
992 struct vfsmount *p;
993 int err = 0;
994 if (!capable(CAP_SYS_ADMIN))
995 return -EPERM;
996 if (!old_name || !*old_name)
997 return -EINVAL;
998 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
999 if (err)
1000 return err;
1001
Ram Pai390c6842005-11-07 17:17:51 -05001002 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -05001003 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 ;
1005 err = -EINVAL;
1006 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1007 goto out;
1008
1009 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001010 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 if (IS_DEADDIR(nd->dentry->d_inode))
1012 goto out1;
1013
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001015 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
1017 err = -EINVAL;
1018 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001019 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001022 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023
1024 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1025 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001026 goto out1;
1027 /*
1028 * Don't move a mount residing in a shared parent.
1029 */
1030 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1031 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001032 /*
1033 * Don't move a mount tree containing unbindable mounts to a destination
1034 * mount which is shared.
1035 */
1036 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1037 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -05001039 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001041 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Ram Pai21444402005-11-07 17:20:03 -05001043 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1044 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Ram Pai21444402005-11-07 17:20:03 -05001046 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 /* if the mount is moved, it should no longer be expire
1048 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001049 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 spin_unlock(&vfsmount_lock);
1051out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001052 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053out:
Ram Pai390c6842005-11-07 17:17:51 -05001054 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (!err)
1056 path_release(&parent_nd);
1057 path_release(&old_nd);
1058 return err;
1059}
1060
1061/*
1062 * create a new mount for userspace and request it to be added into the
1063 * namespace's tree
1064 */
1065static int do_new_mount(struct nameidata *nd, char *type, int flags,
1066 int mnt_flags, char *name, void *data)
1067{
1068 struct vfsmount *mnt;
1069
1070 if (!type || !memchr(type, 0, PAGE_SIZE))
1071 return -EINVAL;
1072
1073 /* we need capabilities... */
1074 if (!capable(CAP_SYS_ADMIN))
1075 return -EPERM;
1076
1077 mnt = do_kern_mount(type, flags, name, data);
1078 if (IS_ERR(mnt))
1079 return PTR_ERR(mnt);
1080
1081 return do_add_mount(mnt, nd, mnt_flags, NULL);
1082}
1083
1084/*
1085 * add a mount into a namespace's mount tree
1086 * - provide the option of adding the new mount to an expiration list
1087 */
1088int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1089 int mnt_flags, struct list_head *fslist)
1090{
1091 int err;
1092
Ram Pai390c6842005-11-07 17:17:51 -05001093 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001095 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 ;
1097 err = -EINVAL;
1098 if (!check_mnt(nd->mnt))
1099 goto unlock;
1100
1101 /* Refuse the same filesystem on the same mount point */
1102 err = -EBUSY;
1103 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1104 nd->mnt->mnt_root == nd->dentry)
1105 goto unlock;
1106
1107 err = -EINVAL;
1108 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1109 goto unlock;
1110
1111 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001112 if ((err = graft_tree(newmnt, nd)))
1113 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001115 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 /* add to the specified expiration list */
1117 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001118 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 spin_unlock(&vfsmount_lock);
1120 }
Ram Pai390c6842005-11-07 17:17:51 -05001121 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
1124unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001125 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 mntput(newmnt);
1127 return err;
1128}
1129
1130EXPORT_SYMBOL_GPL(do_add_mount);
1131
Ram Pai70fbcdf2005-11-07 17:17:04 -05001132static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1133 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001134{
1135 spin_lock(&vfsmount_lock);
1136
1137 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001138 * Check if mount is still attached, if not, let whoever holds it deal
1139 * with the sucker
1140 */
1141 if (mnt->mnt_parent == mnt) {
1142 spin_unlock(&vfsmount_lock);
1143 return;
1144 }
1145
1146 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001147 * Check that it is still dead: the count should now be 2 - as
1148 * contributed by the vfsmount parent and the mntget above
1149 */
Ram Paia05964f2005-11-07 17:20:17 -05001150 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001151 /* delete from the namespace */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001152 touch_mnt_namespace(mnt->mnt_ns);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001153 list_del_init(&mnt->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001154 mnt->mnt_ns = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001155 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001156 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001157 } else {
1158 /*
1159 * Someone brought it back to life whilst we didn't have any
1160 * locks held so return it to the expiration list
1161 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001162 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001163 spin_unlock(&vfsmount_lock);
1164 }
1165}
1166
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001168 * go through the vfsmounts we've just consigned to the graveyard to
1169 * - check that they're still dead
1170 * - delete the vfsmount from the appropriate namespace under lock
1171 * - dispose of the corpse
1172 */
1173static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1174{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001175 struct mnt_namespace *ns;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001176 struct vfsmount *mnt;
1177
1178 while (!list_empty(graveyard)) {
1179 LIST_HEAD(umounts);
1180 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1181 list_del_init(&mnt->mnt_expire);
1182
1183 /* don't do anything if the namespace is dead - all the
1184 * vfsmounts from it are going away anyway */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001185 ns = mnt->mnt_ns;
1186 if (!ns || !ns->root)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001187 continue;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001188 get_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001189
1190 spin_unlock(&vfsmount_lock);
1191 down_write(&namespace_sem);
1192 expire_mount(mnt, mounts, &umounts);
1193 up_write(&namespace_sem);
1194 release_mounts(&umounts);
1195 mntput(mnt);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001196 put_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001197 spin_lock(&vfsmount_lock);
1198 }
1199}
1200
1201/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 * process a list of expirable mountpoints with the intent of discarding any
1203 * mountpoints that aren't in use and haven't been touched since last we came
1204 * here
1205 */
1206void mark_mounts_for_expiry(struct list_head *mounts)
1207{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 struct vfsmount *mnt, *next;
1209 LIST_HEAD(graveyard);
1210
1211 if (list_empty(mounts))
1212 return;
1213
1214 spin_lock(&vfsmount_lock);
1215
1216 /* extract from the expiration list every vfsmount that matches the
1217 * following criteria:
1218 * - only referenced by its parent vfsmount
1219 * - still marked for expiry (marked on the last call here; marks are
1220 * cleared by mntput())
1221 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001222 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1224 atomic_read(&mnt->mnt_count) != 1)
1225 continue;
1226
1227 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001228 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
1230
Trond Myklebust5528f9112006-06-09 09:34:17 -04001231 expire_mount_list(&graveyard, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 spin_unlock(&vfsmount_lock);
1234}
1235
1236EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1237
1238/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001239 * Ripoff of 'select_parent()'
1240 *
1241 * search the list of submounts for a given mountpoint, and move any
1242 * shrinkable submounts to the 'graveyard' list.
1243 */
1244static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1245{
1246 struct vfsmount *this_parent = parent;
1247 struct list_head *next;
1248 int found = 0;
1249
1250repeat:
1251 next = this_parent->mnt_mounts.next;
1252resume:
1253 while (next != &this_parent->mnt_mounts) {
1254 struct list_head *tmp = next;
1255 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1256
1257 next = tmp->next;
1258 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1259 continue;
1260 /*
1261 * Descend a level if the d_mounts list is non-empty.
1262 */
1263 if (!list_empty(&mnt->mnt_mounts)) {
1264 this_parent = mnt;
1265 goto repeat;
1266 }
1267
1268 if (!propagate_mount_busy(mnt, 1)) {
1269 mntget(mnt);
1270 list_move_tail(&mnt->mnt_expire, graveyard);
1271 found++;
1272 }
1273 }
1274 /*
1275 * All done at this level ... ascend and resume the search
1276 */
1277 if (this_parent != parent) {
1278 next = this_parent->mnt_child.next;
1279 this_parent = this_parent->mnt_parent;
1280 goto resume;
1281 }
1282 return found;
1283}
1284
1285/*
1286 * process a list of expirable mountpoints with the intent of discarding any
1287 * submounts of a specific parent mountpoint
1288 */
1289void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1290{
1291 LIST_HEAD(graveyard);
1292 int found;
1293
1294 spin_lock(&vfsmount_lock);
1295
1296 /* extract submounts of 'mountpoint' from the expiration list */
1297 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1298 expire_mount_list(&graveyard, mounts);
1299
1300 spin_unlock(&vfsmount_lock);
1301}
1302
1303EXPORT_SYMBOL_GPL(shrink_submounts);
1304
1305/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 * Some copy_from_user() implementations do not return the exact number of
1307 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1308 * Note that this function differs from copy_from_user() in that it will oops
1309 * on bad values of `to', rather than returning a short copy.
1310 */
Ram Paib58fed82005-11-07 17:16:09 -05001311static long exact_copy_from_user(void *to, const void __user * from,
1312 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
1314 char *t = to;
1315 const char __user *f = from;
1316 char c;
1317
1318 if (!access_ok(VERIFY_READ, from, n))
1319 return n;
1320
1321 while (n) {
1322 if (__get_user(c, f)) {
1323 memset(t, 0, n);
1324 break;
1325 }
1326 *t++ = c;
1327 f++;
1328 n--;
1329 }
1330 return n;
1331}
1332
Ram Paib58fed82005-11-07 17:16:09 -05001333int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
1335 int i;
1336 unsigned long page;
1337 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 *where = 0;
1340 if (!data)
1341 return 0;
1342
1343 if (!(page = __get_free_page(GFP_KERNEL)))
1344 return -ENOMEM;
1345
1346 /* We only care that *some* data at the address the user
1347 * gave us is valid. Just in case, we'll zero
1348 * the remainder of the page.
1349 */
1350 /* copy_from_user cannot cross TASK_SIZE ! */
1351 size = TASK_SIZE - (unsigned long)data;
1352 if (size > PAGE_SIZE)
1353 size = PAGE_SIZE;
1354
1355 i = size - exact_copy_from_user((void *)page, data, size);
1356 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001357 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 return -EFAULT;
1359 }
1360 if (i != PAGE_SIZE)
1361 memset((char *)page + i, 0, PAGE_SIZE - i);
1362 *where = page;
1363 return 0;
1364}
1365
1366/*
1367 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1368 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1369 *
1370 * data is a (void *) that can point to any structure up to
1371 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1372 * information (or be NULL).
1373 *
1374 * Pre-0.97 versions of mount() didn't have a flags word.
1375 * When the flags word was introduced its top half was required
1376 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1377 * Therefore, if this magic number is present, it carries no information
1378 * and must be discarded.
1379 */
Ram Paib58fed82005-11-07 17:16:09 -05001380long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 unsigned long flags, void *data_page)
1382{
1383 struct nameidata nd;
1384 int retval = 0;
1385 int mnt_flags = 0;
1386
1387 /* Discard magic */
1388 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1389 flags &= ~MS_MGC_MSK;
1390
1391 /* Basic sanity checks */
1392
1393 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1394 return -EINVAL;
1395 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1396 return -EINVAL;
1397
1398 if (data_page)
1399 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1400
1401 /* Separate the per-mountpoint flags */
1402 if (flags & MS_NOSUID)
1403 mnt_flags |= MNT_NOSUID;
1404 if (flags & MS_NODEV)
1405 mnt_flags |= MNT_NODEV;
1406 if (flags & MS_NOEXEC)
1407 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001408 if (flags & MS_NOATIME)
1409 mnt_flags |= MNT_NOATIME;
1410 if (flags & MS_NODIRATIME)
1411 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001412 if (flags & MS_RELATIME)
1413 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001414
1415 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Valerie Henson47ae32d2006-12-13 00:34:34 -08001416 MS_NOATIME | MS_NODIRATIME | MS_RELATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
1418 /* ... and get the mountpoint */
1419 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1420 if (retval)
1421 return retval;
1422
1423 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1424 if (retval)
1425 goto dput_out;
1426
1427 if (flags & MS_REMOUNT)
1428 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1429 data_page);
1430 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001431 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001432 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001433 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 else if (flags & MS_MOVE)
1435 retval = do_move_mount(&nd, dev_name);
1436 else
1437 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1438 dev_name, data_page);
1439dput_out:
1440 path_release(&nd);
1441 return retval;
1442}
1443
JANAK DESAI741a2952006-02-07 12:59:00 -08001444/*
1445 * Allocate a new namespace structure and populate it with contents
1446 * copied from the namespace of the passed in task structure.
1447 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001448static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001449 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001451 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 struct vfsmount *p, *q;
1454
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001455 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 if (!new_ns)
Adrian Bunkf13b8352006-03-15 17:37:32 +01001457 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458
1459 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001461 init_waitqueue_head(&new_ns->poll);
1462 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
Ram Pai390c6842005-11-07 17:17:51 -05001464 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001466 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001467 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001469 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 kfree(new_ns);
Adrian Bunkf13b8352006-03-15 17:37:32 +01001471 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 }
1473 spin_lock(&vfsmount_lock);
1474 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1475 spin_unlock(&vfsmount_lock);
1476
1477 /*
1478 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1479 * as belonging to new namespace. We have already acquired a private
1480 * fs_struct, so tsk->fs->lock is not needed.
1481 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001482 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 q = new_ns->root;
1484 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001485 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (fs) {
1487 if (p == fs->rootmnt) {
1488 rootmnt = p;
1489 fs->rootmnt = mntget(q);
1490 }
1491 if (p == fs->pwdmnt) {
1492 pwdmnt = p;
1493 fs->pwdmnt = mntget(q);
1494 }
1495 if (p == fs->altrootmnt) {
1496 altrootmnt = p;
1497 fs->altrootmnt = mntget(q);
1498 }
1499 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001500 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 q = next_mnt(q, new_ns->root);
1502 }
Ram Pai390c6842005-11-07 17:17:51 -05001503 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 if (rootmnt)
1506 mntput(rootmnt);
1507 if (pwdmnt)
1508 mntput(pwdmnt);
1509 if (altrootmnt)
1510 mntput(altrootmnt);
1511
JANAK DESAI741a2952006-02-07 12:59:00 -08001512 return new_ns;
1513}
1514
Badari Pulavartye3222c42007-05-08 00:25:21 -07001515struct mnt_namespace *copy_mnt_ns(int flags, struct mnt_namespace *ns,
1516 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001517{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001518 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001519
Badari Pulavartye3222c42007-05-08 00:25:21 -07001520 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001521 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001522
1523 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001524 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001525
Badari Pulavartye3222c42007-05-08 00:25:21 -07001526 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001527
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001528 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001529 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530}
1531
1532asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1533 char __user * type, unsigned long flags,
1534 void __user * data)
1535{
1536 int retval;
1537 unsigned long data_page;
1538 unsigned long type_page;
1539 unsigned long dev_page;
1540 char *dir_page;
1541
Ram Paib58fed82005-11-07 17:16:09 -05001542 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (retval < 0)
1544 return retval;
1545
1546 dir_page = getname(dir_name);
1547 retval = PTR_ERR(dir_page);
1548 if (IS_ERR(dir_page))
1549 goto out1;
1550
Ram Paib58fed82005-11-07 17:16:09 -05001551 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (retval < 0)
1553 goto out2;
1554
Ram Paib58fed82005-11-07 17:16:09 -05001555 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 if (retval < 0)
1557 goto out3;
1558
1559 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001560 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1561 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 unlock_kernel();
1563 free_page(data_page);
1564
1565out3:
1566 free_page(dev_page);
1567out2:
1568 putname(dir_page);
1569out1:
1570 free_page(type_page);
1571 return retval;
1572}
1573
1574/*
1575 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1576 * It can block. Requires the big lock held.
1577 */
1578void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1579 struct dentry *dentry)
1580{
1581 struct dentry *old_root;
1582 struct vfsmount *old_rootmnt;
1583 write_lock(&fs->lock);
1584 old_root = fs->root;
1585 old_rootmnt = fs->rootmnt;
1586 fs->rootmnt = mntget(mnt);
1587 fs->root = dget(dentry);
1588 write_unlock(&fs->lock);
1589 if (old_root) {
1590 dput(old_root);
1591 mntput(old_rootmnt);
1592 }
1593}
1594
1595/*
1596 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1597 * It can block. Requires the big lock held.
1598 */
1599void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1600 struct dentry *dentry)
1601{
1602 struct dentry *old_pwd;
1603 struct vfsmount *old_pwdmnt;
1604
1605 write_lock(&fs->lock);
1606 old_pwd = fs->pwd;
1607 old_pwdmnt = fs->pwdmnt;
1608 fs->pwdmnt = mntget(mnt);
1609 fs->pwd = dget(dentry);
1610 write_unlock(&fs->lock);
1611
1612 if (old_pwd) {
1613 dput(old_pwd);
1614 mntput(old_pwdmnt);
1615 }
1616}
1617
1618static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1619{
1620 struct task_struct *g, *p;
1621 struct fs_struct *fs;
1622
1623 read_lock(&tasklist_lock);
1624 do_each_thread(g, p) {
1625 task_lock(p);
1626 fs = p->fs;
1627 if (fs) {
1628 atomic_inc(&fs->count);
1629 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001630 if (fs->root == old_nd->dentry
1631 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001633 if (fs->pwd == old_nd->dentry
1634 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1636 put_fs_struct(fs);
1637 } else
1638 task_unlock(p);
1639 } while_each_thread(g, p);
1640 read_unlock(&tasklist_lock);
1641}
1642
1643/*
1644 * pivot_root Semantics:
1645 * Moves the root file system of the current process to the directory put_old,
1646 * makes new_root as the new root file system of the current process, and sets
1647 * root/cwd of all processes which had them on the current root to new_root.
1648 *
1649 * Restrictions:
1650 * The new_root and put_old must be directories, and must not be on the
1651 * same file system as the current process root. The put_old must be
1652 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1653 * pointed to by put_old must yield the same directory as new_root. No other
1654 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1655 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001656 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1657 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1658 * in this situation.
1659 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 * Notes:
1661 * - we don't move root/cwd if they are not at the root (reason: if something
1662 * cared enough to change them, it's probably wrong to force them elsewhere)
1663 * - it's okay to pick a root that isn't the root of a file system, e.g.
1664 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1665 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1666 * first.
1667 */
Ram Paib58fed82005-11-07 17:16:09 -05001668asmlinkage long sys_pivot_root(const char __user * new_root,
1669 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670{
1671 struct vfsmount *tmp;
1672 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1673 int error;
1674
1675 if (!capable(CAP_SYS_ADMIN))
1676 return -EPERM;
1677
1678 lock_kernel();
1679
Ram Paib58fed82005-11-07 17:16:09 -05001680 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1681 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (error)
1683 goto out0;
1684 error = -EINVAL;
1685 if (!check_mnt(new_nd.mnt))
1686 goto out1;
1687
Ram Paib58fed82005-11-07 17:16:09 -05001688 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 if (error)
1690 goto out1;
1691
1692 error = security_sb_pivotroot(&old_nd, &new_nd);
1693 if (error) {
1694 path_release(&old_nd);
1695 goto out1;
1696 }
1697
1698 read_lock(&current->fs->lock);
1699 user_nd.mnt = mntget(current->fs->rootmnt);
1700 user_nd.dentry = dget(current->fs->root);
1701 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001702 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001703 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001705 if (IS_MNT_SHARED(old_nd.mnt) ||
1706 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1707 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1708 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (!check_mnt(user_nd.mnt))
1710 goto out2;
1711 error = -ENOENT;
1712 if (IS_DEADDIR(new_nd.dentry->d_inode))
1713 goto out2;
1714 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1715 goto out2;
1716 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1717 goto out2;
1718 error = -EBUSY;
1719 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1720 goto out2; /* loop, on the same file system */
1721 error = -EINVAL;
1722 if (user_nd.mnt->mnt_root != user_nd.dentry)
1723 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001724 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1725 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (new_nd.mnt->mnt_root != new_nd.dentry)
1727 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001728 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1729 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1731 spin_lock(&vfsmount_lock);
1732 if (tmp != new_nd.mnt) {
1733 for (;;) {
1734 if (tmp->mnt_parent == tmp)
1735 goto out3; /* already mounted on put_old */
1736 if (tmp->mnt_parent == new_nd.mnt)
1737 break;
1738 tmp = tmp->mnt_parent;
1739 }
1740 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1741 goto out3;
1742 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1743 goto out3;
1744 detach_mnt(new_nd.mnt, &parent_nd);
1745 detach_mnt(user_nd.mnt, &root_parent);
1746 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1747 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001748 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 spin_unlock(&vfsmount_lock);
1750 chroot_fs_refs(&user_nd, &new_nd);
1751 security_sb_post_pivotroot(&user_nd, &new_nd);
1752 error = 0;
1753 path_release(&root_parent);
1754 path_release(&parent_nd);
1755out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001756 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001757 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 path_release(&user_nd);
1759 path_release(&old_nd);
1760out1:
1761 path_release(&new_nd);
1762out0:
1763 unlock_kernel();
1764 return error;
1765out3:
1766 spin_unlock(&vfsmount_lock);
1767 goto out2;
1768}
1769
1770static void __init init_mount_tree(void)
1771{
1772 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001773 struct mnt_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1776 if (IS_ERR(mnt))
1777 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001778 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1779 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001781 atomic_set(&ns->count, 1);
1782 INIT_LIST_HEAD(&ns->list);
1783 init_waitqueue_head(&ns->poll);
1784 ns->event = 0;
1785 list_add(&mnt->mnt_list, &ns->list);
1786 ns->root = mnt;
1787 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001789 init_task.nsproxy->mnt_ns = ns;
1790 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001792 set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1793 set_fs_root(current->fs, ns->root, ns->root->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794}
1795
1796void __init mnt_init(unsigned long mempages)
1797{
1798 struct list_head *d;
1799 unsigned int nr_hash;
1800 int i;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001801 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Ram Pai390c6842005-11-07 17:17:51 -05001803 init_rwsem(&namespace_sem);
1804
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Ram Paib58fed82005-11-07 17:16:09 -05001806 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Ram Paib58fed82005-11-07 17:16:09 -05001808 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
1810 if (!mount_hashtable)
1811 panic("Failed to allocate mount hash table\n");
1812
1813 /*
1814 * Find the power-of-two list-heads that can fit into the allocation..
1815 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1816 * a power-of-two.
1817 */
1818 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1819 hash_bits = 0;
1820 do {
1821 hash_bits++;
1822 } while ((nr_hash >> hash_bits) != 0);
1823 hash_bits--;
1824
1825 /*
1826 * Re-calculate the actual number of entries and the mask
1827 * from the number of bits we can fit.
1828 */
1829 nr_hash = 1UL << hash_bits;
Ram Paib58fed82005-11-07 17:16:09 -05001830 hash_mask = nr_hash - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
1832 printk("Mount-cache hash table entries: %d\n", nr_hash);
1833
1834 /* And initialize the newly allocated array */
1835 d = mount_hashtable;
1836 i = nr_hash;
1837 do {
1838 INIT_LIST_HEAD(d);
1839 d++;
1840 i--;
1841 } while (i);
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001842 err = sysfs_init();
1843 if (err)
1844 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1845 __FUNCTION__, err);
1846 err = subsystem_register(&fs_subsys);
1847 if (err)
1848 printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1849 __FUNCTION__, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 init_rootfs();
1851 init_mount_tree();
1852}
1853
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001854void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001856 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001857 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001858 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001859 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001860 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001862 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001864 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001865 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001866 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867}