blob: 5585623f62522f33daffbae685958339c30e16ac [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
Ram Pai390c6842005-11-07 17:17:51 -0500324 down_read(&namespace_sem);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700325 return seq_list_start(&n->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326}
327
328static void *m_next(struct seq_file *m, void *v, loff_t *pos)
329{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800330 struct mnt_namespace *n = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700331
332 return seq_list_next(v, &n->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333}
334
335static void m_stop(struct seq_file *m, void *v)
336{
Ram Pai390c6842005-11-07 17:17:51 -0500337 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
340static inline void mangle(struct seq_file *m, const char *s)
341{
342 seq_escape(m, s, " \t\n\\");
343}
344
345static int show_vfsmnt(struct seq_file *m, void *v)
346{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700347 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 int err = 0;
349 static struct proc_fs_info {
350 int flag;
351 char *str;
352 } fs_info[] = {
353 { MS_SYNCHRONOUS, ",sync" },
354 { MS_DIRSYNC, ",dirsync" },
355 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 { 0, NULL }
357 };
358 static struct proc_fs_info mnt_info[] = {
359 { MNT_NOSUID, ",nosuid" },
360 { MNT_NODEV, ",nodev" },
361 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800362 { MNT_NOATIME, ",noatime" },
363 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800364 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 { 0, NULL }
366 };
367 struct proc_fs_info *fs_infop;
368
369 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
370 seq_putc(m, ' ');
371 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
372 seq_putc(m, ' ');
373 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700374 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
375 seq_putc(m, '.');
376 mangle(m, mnt->mnt_sb->s_subtype);
377 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
379 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
380 if (mnt->mnt_sb->s_flags & fs_infop->flag)
381 seq_puts(m, fs_infop->str);
382 }
383 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
384 if (mnt->mnt_flags & fs_infop->flag)
385 seq_puts(m, fs_infop->str);
386 }
387 if (mnt->mnt_sb->s_op->show_options)
388 err = mnt->mnt_sb->s_op->show_options(m, mnt);
389 seq_puts(m, " 0 0\n");
390 return err;
391}
392
393struct seq_operations mounts_op = {
394 .start = m_start,
395 .next = m_next,
396 .stop = m_stop,
397 .show = show_vfsmnt
398};
399
Chuck Leverb4629fe2006-03-20 13:44:12 -0500400static int show_vfsstat(struct seq_file *m, void *v)
401{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700402 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500403 int err = 0;
404
405 /* device */
406 if (mnt->mnt_devname) {
407 seq_puts(m, "device ");
408 mangle(m, mnt->mnt_devname);
409 } else
410 seq_puts(m, "no device");
411
412 /* mount point */
413 seq_puts(m, " mounted on ");
414 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
415 seq_putc(m, ' ');
416
417 /* file system type */
418 seq_puts(m, "with fstype ");
419 mangle(m, mnt->mnt_sb->s_type->name);
420
421 /* optional statistics */
422 if (mnt->mnt_sb->s_op->show_stats) {
423 seq_putc(m, ' ');
424 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
425 }
426
427 seq_putc(m, '\n');
428 return err;
429}
430
431struct seq_operations mountstats_op = {
432 .start = m_start,
433 .next = m_next,
434 .stop = m_stop,
435 .show = show_vfsstat,
436};
437
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438/**
439 * may_umount_tree - check if a mount tree is busy
440 * @mnt: root of mount tree
441 *
442 * This is called to check if a tree of mounts has any
443 * open files, pwds, chroots or sub mounts that are
444 * busy.
445 */
446int may_umount_tree(struct vfsmount *mnt)
447{
Ram Pai36341f62005-11-07 17:17:22 -0500448 int actual_refs = 0;
449 int minimum_refs = 0;
450 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500453 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 actual_refs += atomic_read(&p->mnt_count);
455 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 }
457 spin_unlock(&vfsmount_lock);
458
459 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800460 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Ian Kente3474a82006-03-27 01:14:51 -0800462 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464
465EXPORT_SYMBOL(may_umount_tree);
466
467/**
468 * may_umount - check if a mount point is busy
469 * @mnt: root of mount
470 *
471 * This is called to check if a mount point has any
472 * open files, pwds, chroots or sub mounts. If the
473 * mount has sub mounts this will return busy
474 * regardless of whether the sub mounts are busy.
475 *
476 * Doesn't take quota and stuff into account. IOW, in some cases it will
477 * give false negatives. The main reason why it's here is that we need
478 * a non-destructive way to look for easily umountable filesystems.
479 */
480int may_umount(struct vfsmount *mnt)
481{
Ian Kente3474a82006-03-27 01:14:51 -0800482 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500483 spin_lock(&vfsmount_lock);
484 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800485 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500486 spin_unlock(&vfsmount_lock);
487 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
490EXPORT_SYMBOL(may_umount);
491
Ram Paib90fa9a2005-11-07 17:19:50 -0500492void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500494 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800495 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700496 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500497 list_del_init(&mnt->mnt_hash);
498 if (mnt->mnt_parent != mnt) {
499 struct dentry *dentry;
500 struct vfsmount *m;
501 spin_lock(&vfsmount_lock);
502 dentry = mnt->mnt_mountpoint;
503 m = mnt->mnt_parent;
504 mnt->mnt_mountpoint = mnt->mnt_root;
505 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500507 dput(dentry);
508 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
510 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500511 }
512}
513
Ram Paia05964f2005-11-07 17:20:17 -0500514void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500515{
516 struct vfsmount *p;
517
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700518 for (p = mnt; p; p = next_mnt(p, mnt))
519 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500520
Ram Paia05964f2005-11-07 17:20:17 -0500521 if (propagate)
522 propagate_umount(kill);
523
Ram Pai70fbcdf2005-11-07 17:17:04 -0500524 list_for_each_entry(p, kill, mnt_hash) {
525 list_del_init(&p->mnt_expire);
526 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800527 __touch_mnt_namespace(p->mnt_ns);
528 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500529 list_del_init(&p->mnt_child);
530 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500531 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500532 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
534}
535
536static int do_umount(struct vfsmount *mnt, int flags)
537{
Ram Paib58fed82005-11-07 17:16:09 -0500538 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500540 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
542 retval = security_sb_umount(mnt, flags);
543 if (retval)
544 return retval;
545
546 /*
547 * Allow userspace to request a mountpoint be expired rather than
548 * unmounting unconditionally. Unmount only happens if:
549 * (1) the mark is already set (the mark is cleared by mntput())
550 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
551 */
552 if (flags & MNT_EXPIRE) {
553 if (mnt == current->fs->rootmnt ||
554 flags & (MNT_FORCE | MNT_DETACH))
555 return -EINVAL;
556
557 if (atomic_read(&mnt->mnt_count) != 2)
558 return -EBUSY;
559
560 if (!xchg(&mnt->mnt_expiry_mark, 1))
561 return -EAGAIN;
562 }
563
564 /*
565 * If we may have to abort operations to get out of this
566 * mount, and they will themselves hold resources we must
567 * allow the fs to do things. In the Unix tradition of
568 * 'Gee thats tricky lets do it in userspace' the umount_begin
569 * might fail to complete on the first run through as other tasks
570 * must return, and the like. Thats for the mount program to worry
571 * about for the moment.
572 */
573
574 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400575 if (sb->s_op->umount_begin)
576 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 unlock_kernel();
578
579 /*
580 * No sense to grab the lock for this test, but test itself looks
581 * somewhat bogus. Suggestions for better replacement?
582 * Ho-hum... In principle, we might treat that as umount + switch
583 * to rootfs. GC would eventually take care of the old vfsmount.
584 * Actually it makes sense, especially if rootfs would contain a
585 * /reboot - static binary that would close all descriptors and
586 * call reboot(9). Then init(8) could umount root and exec /reboot.
587 */
588 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
589 /*
590 * Special case for "unmounting" root ...
591 * we just try to remount it readonly.
592 */
593 down_write(&sb->s_umount);
594 if (!(sb->s_flags & MS_RDONLY)) {
595 lock_kernel();
596 DQUOT_OFF(sb);
597 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
598 unlock_kernel();
599 }
600 up_write(&sb->s_umount);
601 return retval;
602 }
603
Ram Pai390c6842005-11-07 17:17:51 -0500604 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500606 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500609 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500611 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 retval = 0;
613 }
614 spin_unlock(&vfsmount_lock);
615 if (retval)
616 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500617 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500618 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 return retval;
620}
621
622/*
623 * Now umount can handle mount points as well as block devices.
624 * This is important for filesystems which use unnamed block devices.
625 *
626 * We now support a flag for forced unmount like the other 'big iron'
627 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
628 */
629
630asmlinkage long sys_umount(char __user * name, int flags)
631{
632 struct nameidata nd;
633 int retval;
634
635 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
636 if (retval)
637 goto out;
638 retval = -EINVAL;
639 if (nd.dentry != nd.mnt->mnt_root)
640 goto dput_and_out;
641 if (!check_mnt(nd.mnt))
642 goto dput_and_out;
643
644 retval = -EPERM;
645 if (!capable(CAP_SYS_ADMIN))
646 goto dput_and_out;
647
648 retval = do_umount(nd.mnt, flags);
649dput_and_out:
650 path_release_on_umount(&nd);
651out:
652 return retval;
653}
654
655#ifdef __ARCH_WANT_SYS_OLDUMOUNT
656
657/*
Ram Paib58fed82005-11-07 17:16:09 -0500658 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660asmlinkage long sys_oldumount(char __user * name)
661{
Ram Paib58fed82005-11-07 17:16:09 -0500662 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
664
665#endif
666
667static int mount_is_safe(struct nameidata *nd)
668{
669 if (capable(CAP_SYS_ADMIN))
670 return 0;
671 return -EPERM;
672#ifdef notyet
673 if (S_ISLNK(nd->dentry->d_inode->i_mode))
674 return -EPERM;
675 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
676 if (current->uid != nd->dentry->d_inode->i_uid)
677 return -EPERM;
678 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800679 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 return -EPERM;
681 return 0;
682#endif
683}
684
Ram Paib58fed82005-11-07 17:16:09 -0500685static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 while (1) {
688 if (d == dentry)
689 return 1;
690 if (d == NULL || d == d->d_parent)
691 return 0;
692 d = d->d_parent;
693 }
694}
695
Ram Paib90fa9a2005-11-07 17:19:50 -0500696struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500697 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698{
699 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 struct nameidata nd;
701
Ram Pai9676f0c2005-11-07 17:21:20 -0500702 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
703 return NULL;
704
Ram Pai36341f62005-11-07 17:17:22 -0500705 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 if (!q)
707 goto Enomem;
708 q->mnt_mountpoint = mnt->mnt_mountpoint;
709
710 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700711 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
713 continue;
714
715 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500716 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
717 s = skip_mnt_tree(s);
718 continue;
719 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 while (p != s->mnt_parent) {
721 p = p->mnt_parent;
722 q = q->mnt_parent;
723 }
724 p = s;
725 nd.mnt = q;
726 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500727 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 if (!q)
729 goto Enomem;
730 spin_lock(&vfsmount_lock);
731 list_add_tail(&q->mnt_list, &res->mnt_list);
732 attach_mnt(q, &nd);
733 spin_unlock(&vfsmount_lock);
734 }
735 }
736 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500737Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500739 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500741 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500743 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745 return NULL;
746}
747
Ram Paib90fa9a2005-11-07 17:19:50 -0500748/*
749 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500750 * @nd : place the mount tree @source_mnt is attached
751 * @parent_nd : if non-null, detach the source_mnt from its parent and
752 * store the parent mount and mountpoint dentry.
753 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500754 *
755 * NOTE: in the table below explains the semantics when a source mount
756 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500757 * ---------------------------------------------------------------------------
758 * | BIND MOUNT OPERATION |
759 * |**************************************************************************
760 * | source-->| shared | private | slave | unbindable |
761 * | dest | | | | |
762 * | | | | | | |
763 * | v | | | | |
764 * |**************************************************************************
765 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
766 * | | | | | |
767 * |non-shared| shared (+) | private | slave (*) | invalid |
768 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500769 * A bind operation clones the source mount and mounts the clone on the
770 * destination mount.
771 *
772 * (++) the cloned mount is propagated to all the mounts in the propagation
773 * tree of the destination mount and the cloned mount is added to
774 * the peer group of the source mount.
775 * (+) the cloned mount is created under the destination mount and is marked
776 * as shared. The cloned mount is added to the peer group of the source
777 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500778 * (+++) the mount is propagated to all the mounts in the propagation tree
779 * of the destination mount and the cloned mount is made slave
780 * of the same master as that of the source mount. The cloned mount
781 * is marked as 'shared and slave'.
782 * (*) the cloned mount is made a slave of the same master as that of the
783 * source mount.
784 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500785 * ---------------------------------------------------------------------------
786 * | MOVE MOUNT OPERATION |
787 * |**************************************************************************
788 * | source-->| shared | private | slave | unbindable |
789 * | dest | | | | |
790 * | | | | | | |
791 * | v | | | | |
792 * |**************************************************************************
793 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
794 * | | | | | |
795 * |non-shared| shared (+*) | private | slave (*) | unbindable |
796 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500797 *
798 * (+) the mount is moved to the destination. And is then propagated to
799 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500800 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500801 * (+++) the mount is moved to the destination and is then propagated to
802 * all the mounts belonging to the destination mount's propagation tree.
803 * the mount is marked as 'shared and slave'.
804 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500805 *
806 * if the source mount is a tree, the operations explained above is
807 * applied to each mount in the tree.
808 * Must be called without spinlocks held, since this function can sleep
809 * in allocations.
810 */
811static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500812 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500813{
814 LIST_HEAD(tree_list);
815 struct vfsmount *dest_mnt = nd->mnt;
816 struct dentry *dest_dentry = nd->dentry;
817 struct vfsmount *child, *p;
818
819 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
820 return -EINVAL;
821
822 if (IS_MNT_SHARED(dest_mnt)) {
823 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
824 set_mnt_shared(p);
825 }
826
827 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500828 if (parent_nd) {
829 detach_mnt(source_mnt, parent_nd);
830 attach_mnt(source_mnt, nd);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800831 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500832 } else {
833 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
834 commit_tree(source_mnt);
835 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500836
837 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
838 list_del_init(&child->mnt_hash);
839 commit_tree(child);
840 }
841 spin_unlock(&vfsmount_lock);
842 return 0;
843}
844
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
846{
847 int err;
848 if (mnt->mnt_sb->s_flags & MS_NOUSER)
849 return -EINVAL;
850
851 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
852 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
853 return -ENOTDIR;
854
855 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800856 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (IS_DEADDIR(nd->dentry->d_inode))
858 goto out_unlock;
859
860 err = security_sb_check_sb(mnt, nd);
861 if (err)
862 goto out_unlock;
863
864 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500865 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500866 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800868 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (!err)
870 security_sb_post_addmount(mnt, nd);
871 return err;
872}
873
874/*
Ram Pai07b20882005-11-07 17:19:07 -0500875 * recursively change the type of the mountpoint.
876 */
877static int do_change_type(struct nameidata *nd, int flag)
878{
879 struct vfsmount *m, *mnt = nd->mnt;
880 int recurse = flag & MS_REC;
881 int type = flag & ~MS_REC;
882
Miklos Szerediee6f9582007-05-08 00:30:40 -0700883 if (!capable(CAP_SYS_ADMIN))
884 return -EPERM;
885
Ram Pai07b20882005-11-07 17:19:07 -0500886 if (nd->dentry != nd->mnt->mnt_root)
887 return -EINVAL;
888
889 down_write(&namespace_sem);
890 spin_lock(&vfsmount_lock);
891 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
892 change_mnt_propagation(m, type);
893 spin_unlock(&vfsmount_lock);
894 up_write(&namespace_sem);
895 return 0;
896}
897
898/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 * do loopback mount.
900 */
Andrew Mortoneee391a2006-05-15 09:44:30 -0700901static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902{
903 struct nameidata old_nd;
904 struct vfsmount *mnt = NULL;
905 int err = mount_is_safe(nd);
906 if (err)
907 return err;
908 if (!old_name || !*old_name)
909 return -EINVAL;
910 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
911 if (err)
912 return err;
913
Ram Pai390c6842005-11-07 17:17:51 -0500914 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500916 if (IS_MNT_UNBINDABLE(old_nd.mnt))
917 goto out;
918
Al Viroccd48bc2005-11-07 17:15:04 -0500919 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
920 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
Al Viroccd48bc2005-11-07 17:15:04 -0500922 err = -ENOMEM;
923 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500924 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500925 else
Ram Pai36341f62005-11-07 17:17:22 -0500926 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500927
928 if (!mnt)
929 goto out;
930
Al Viroccd48bc2005-11-07 17:15:04 -0500931 err = graft_tree(mnt, nd);
932 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500933 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500935 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500937 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -0500938 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Al Viroccd48bc2005-11-07 17:15:04 -0500940out:
Ram Pai390c6842005-11-07 17:17:51 -0500941 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 path_release(&old_nd);
943 return err;
944}
945
946/*
947 * change filesystem flags. dir should be a physical root of filesystem.
948 * If you've mounted a non-root directory somewhere and want to do remount
949 * on it - tough luck.
950 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
952 void *data)
953{
954 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500955 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 if (!capable(CAP_SYS_ADMIN))
958 return -EPERM;
959
960 if (!check_mnt(nd->mnt))
961 return -EINVAL;
962
963 if (nd->dentry != nd->mnt->mnt_root)
964 return -EINVAL;
965
966 down_write(&sb->s_umount);
967 err = do_remount_sb(sb, flags, data, 0);
968 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500969 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 up_write(&sb->s_umount);
971 if (!err)
972 security_sb_post_remount(nd->mnt, flags, data);
973 return err;
974}
975
Ram Pai9676f0c2005-11-07 17:21:20 -0500976static inline int tree_contains_unbindable(struct vfsmount *mnt)
977{
978 struct vfsmount *p;
979 for (p = mnt; p; p = next_mnt(p, mnt)) {
980 if (IS_MNT_UNBINDABLE(p))
981 return 1;
982 }
983 return 0;
984}
985
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986static int do_move_mount(struct nameidata *nd, char *old_name)
987{
988 struct nameidata old_nd, parent_nd;
989 struct vfsmount *p;
990 int err = 0;
991 if (!capable(CAP_SYS_ADMIN))
992 return -EPERM;
993 if (!old_name || !*old_name)
994 return -EINVAL;
995 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
996 if (err)
997 return err;
998
Ram Pai390c6842005-11-07 17:17:51 -0500999 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -05001000 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 ;
1002 err = -EINVAL;
1003 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1004 goto out;
1005
1006 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001007 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (IS_DEADDIR(nd->dentry->d_inode))
1009 goto out1;
1010
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001012 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
1014 err = -EINVAL;
1015 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001016 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001019 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1022 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001023 goto out1;
1024 /*
1025 * Don't move a mount residing in a shared parent.
1026 */
1027 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1028 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001029 /*
1030 * Don't move a mount tree containing unbindable mounts to a destination
1031 * mount which is shared.
1032 */
1033 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1034 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -05001036 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001038 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
Ram Pai21444402005-11-07 17:20:03 -05001040 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1041 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Ram Pai21444402005-11-07 17:20:03 -05001043 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 /* if the mount is moved, it should no longer be expire
1045 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001046 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 spin_unlock(&vfsmount_lock);
1048out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001049 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050out:
Ram Pai390c6842005-11-07 17:17:51 -05001051 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 if (!err)
1053 path_release(&parent_nd);
1054 path_release(&old_nd);
1055 return err;
1056}
1057
1058/*
1059 * create a new mount for userspace and request it to be added into the
1060 * namespace's tree
1061 */
1062static int do_new_mount(struct nameidata *nd, char *type, int flags,
1063 int mnt_flags, char *name, void *data)
1064{
1065 struct vfsmount *mnt;
1066
1067 if (!type || !memchr(type, 0, PAGE_SIZE))
1068 return -EINVAL;
1069
1070 /* we need capabilities... */
1071 if (!capable(CAP_SYS_ADMIN))
1072 return -EPERM;
1073
1074 mnt = do_kern_mount(type, flags, name, data);
1075 if (IS_ERR(mnt))
1076 return PTR_ERR(mnt);
1077
1078 return do_add_mount(mnt, nd, mnt_flags, NULL);
1079}
1080
1081/*
1082 * add a mount into a namespace's mount tree
1083 * - provide the option of adding the new mount to an expiration list
1084 */
1085int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1086 int mnt_flags, struct list_head *fslist)
1087{
1088 int err;
1089
Ram Pai390c6842005-11-07 17:17:51 -05001090 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001092 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 ;
1094 err = -EINVAL;
1095 if (!check_mnt(nd->mnt))
1096 goto unlock;
1097
1098 /* Refuse the same filesystem on the same mount point */
1099 err = -EBUSY;
1100 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1101 nd->mnt->mnt_root == nd->dentry)
1102 goto unlock;
1103
1104 err = -EINVAL;
1105 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1106 goto unlock;
1107
1108 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001109 if ((err = graft_tree(newmnt, nd)))
1110 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001112 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 /* add to the specified expiration list */
1114 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001115 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 spin_unlock(&vfsmount_lock);
1117 }
Ram Pai390c6842005-11-07 17:17:51 -05001118 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001119 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
1121unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001122 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 mntput(newmnt);
1124 return err;
1125}
1126
1127EXPORT_SYMBOL_GPL(do_add_mount);
1128
Ram Pai70fbcdf2005-11-07 17:17:04 -05001129static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1130 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001131{
1132 spin_lock(&vfsmount_lock);
1133
1134 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001135 * Check if mount is still attached, if not, let whoever holds it deal
1136 * with the sucker
1137 */
1138 if (mnt->mnt_parent == mnt) {
1139 spin_unlock(&vfsmount_lock);
1140 return;
1141 }
1142
1143 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001144 * Check that it is still dead: the count should now be 2 - as
1145 * contributed by the vfsmount parent and the mntget above
1146 */
Ram Paia05964f2005-11-07 17:20:17 -05001147 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001148 /* delete from the namespace */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001149 touch_mnt_namespace(mnt->mnt_ns);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001150 list_del_init(&mnt->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001151 mnt->mnt_ns = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001152 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001153 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001154 } else {
1155 /*
1156 * Someone brought it back to life whilst we didn't have any
1157 * locks held so return it to the expiration list
1158 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001159 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001160 spin_unlock(&vfsmount_lock);
1161 }
1162}
1163
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001165 * go through the vfsmounts we've just consigned to the graveyard to
1166 * - check that they're still dead
1167 * - delete the vfsmount from the appropriate namespace under lock
1168 * - dispose of the corpse
1169 */
1170static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1171{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001172 struct mnt_namespace *ns;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001173 struct vfsmount *mnt;
1174
1175 while (!list_empty(graveyard)) {
1176 LIST_HEAD(umounts);
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001177 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001178 list_del_init(&mnt->mnt_expire);
1179
1180 /* don't do anything if the namespace is dead - all the
1181 * vfsmounts from it are going away anyway */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001182 ns = mnt->mnt_ns;
1183 if (!ns || !ns->root)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001184 continue;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001185 get_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001186
1187 spin_unlock(&vfsmount_lock);
1188 down_write(&namespace_sem);
1189 expire_mount(mnt, mounts, &umounts);
1190 up_write(&namespace_sem);
1191 release_mounts(&umounts);
1192 mntput(mnt);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001193 put_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001194 spin_lock(&vfsmount_lock);
1195 }
1196}
1197
1198/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 * process a list of expirable mountpoints with the intent of discarding any
1200 * mountpoints that aren't in use and haven't been touched since last we came
1201 * here
1202 */
1203void mark_mounts_for_expiry(struct list_head *mounts)
1204{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 struct vfsmount *mnt, *next;
1206 LIST_HEAD(graveyard);
1207
1208 if (list_empty(mounts))
1209 return;
1210
1211 spin_lock(&vfsmount_lock);
1212
1213 /* extract from the expiration list every vfsmount that matches the
1214 * following criteria:
1215 * - only referenced by its parent vfsmount
1216 * - still marked for expiry (marked on the last call here; marks are
1217 * cleared by mntput())
1218 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001219 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1221 atomic_read(&mnt->mnt_count) != 1)
1222 continue;
1223
1224 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001225 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 }
1227
Trond Myklebust5528f9112006-06-09 09:34:17 -04001228 expire_mount_list(&graveyard, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230 spin_unlock(&vfsmount_lock);
1231}
1232
1233EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1234
1235/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001236 * Ripoff of 'select_parent()'
1237 *
1238 * search the list of submounts for a given mountpoint, and move any
1239 * shrinkable submounts to the 'graveyard' list.
1240 */
1241static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1242{
1243 struct vfsmount *this_parent = parent;
1244 struct list_head *next;
1245 int found = 0;
1246
1247repeat:
1248 next = this_parent->mnt_mounts.next;
1249resume:
1250 while (next != &this_parent->mnt_mounts) {
1251 struct list_head *tmp = next;
1252 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1253
1254 next = tmp->next;
1255 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1256 continue;
1257 /*
1258 * Descend a level if the d_mounts list is non-empty.
1259 */
1260 if (!list_empty(&mnt->mnt_mounts)) {
1261 this_parent = mnt;
1262 goto repeat;
1263 }
1264
1265 if (!propagate_mount_busy(mnt, 1)) {
1266 mntget(mnt);
1267 list_move_tail(&mnt->mnt_expire, graveyard);
1268 found++;
1269 }
1270 }
1271 /*
1272 * All done at this level ... ascend and resume the search
1273 */
1274 if (this_parent != parent) {
1275 next = this_parent->mnt_child.next;
1276 this_parent = this_parent->mnt_parent;
1277 goto resume;
1278 }
1279 return found;
1280}
1281
1282/*
1283 * process a list of expirable mountpoints with the intent of discarding any
1284 * submounts of a specific parent mountpoint
1285 */
1286void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1287{
1288 LIST_HEAD(graveyard);
1289 int found;
1290
1291 spin_lock(&vfsmount_lock);
1292
1293 /* extract submounts of 'mountpoint' from the expiration list */
1294 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1295 expire_mount_list(&graveyard, mounts);
1296
1297 spin_unlock(&vfsmount_lock);
1298}
1299
1300EXPORT_SYMBOL_GPL(shrink_submounts);
1301
1302/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 * Some copy_from_user() implementations do not return the exact number of
1304 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1305 * Note that this function differs from copy_from_user() in that it will oops
1306 * on bad values of `to', rather than returning a short copy.
1307 */
Ram Paib58fed82005-11-07 17:16:09 -05001308static long exact_copy_from_user(void *to, const void __user * from,
1309 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
1311 char *t = to;
1312 const char __user *f = from;
1313 char c;
1314
1315 if (!access_ok(VERIFY_READ, from, n))
1316 return n;
1317
1318 while (n) {
1319 if (__get_user(c, f)) {
1320 memset(t, 0, n);
1321 break;
1322 }
1323 *t++ = c;
1324 f++;
1325 n--;
1326 }
1327 return n;
1328}
1329
Ram Paib58fed82005-11-07 17:16:09 -05001330int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
1332 int i;
1333 unsigned long page;
1334 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001335
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 *where = 0;
1337 if (!data)
1338 return 0;
1339
1340 if (!(page = __get_free_page(GFP_KERNEL)))
1341 return -ENOMEM;
1342
1343 /* We only care that *some* data at the address the user
1344 * gave us is valid. Just in case, we'll zero
1345 * the remainder of the page.
1346 */
1347 /* copy_from_user cannot cross TASK_SIZE ! */
1348 size = TASK_SIZE - (unsigned long)data;
1349 if (size > PAGE_SIZE)
1350 size = PAGE_SIZE;
1351
1352 i = size - exact_copy_from_user((void *)page, data, size);
1353 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001354 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 return -EFAULT;
1356 }
1357 if (i != PAGE_SIZE)
1358 memset((char *)page + i, 0, PAGE_SIZE - i);
1359 *where = page;
1360 return 0;
1361}
1362
1363/*
1364 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1365 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1366 *
1367 * data is a (void *) that can point to any structure up to
1368 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1369 * information (or be NULL).
1370 *
1371 * Pre-0.97 versions of mount() didn't have a flags word.
1372 * When the flags word was introduced its top half was required
1373 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1374 * Therefore, if this magic number is present, it carries no information
1375 * and must be discarded.
1376 */
Ram Paib58fed82005-11-07 17:16:09 -05001377long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 unsigned long flags, void *data_page)
1379{
1380 struct nameidata nd;
1381 int retval = 0;
1382 int mnt_flags = 0;
1383
1384 /* Discard magic */
1385 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1386 flags &= ~MS_MGC_MSK;
1387
1388 /* Basic sanity checks */
1389
1390 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1391 return -EINVAL;
1392 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1393 return -EINVAL;
1394
1395 if (data_page)
1396 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1397
1398 /* Separate the per-mountpoint flags */
1399 if (flags & MS_NOSUID)
1400 mnt_flags |= MNT_NOSUID;
1401 if (flags & MS_NODEV)
1402 mnt_flags |= MNT_NODEV;
1403 if (flags & MS_NOEXEC)
1404 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001405 if (flags & MS_NOATIME)
1406 mnt_flags |= MNT_NOATIME;
1407 if (flags & MS_NODIRATIME)
1408 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001409 if (flags & MS_RELATIME)
1410 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001411
1412 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Valerie Henson47ae32d2006-12-13 00:34:34 -08001413 MS_NOATIME | MS_NODIRATIME | MS_RELATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
1415 /* ... and get the mountpoint */
1416 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1417 if (retval)
1418 return retval;
1419
1420 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1421 if (retval)
1422 goto dput_out;
1423
1424 if (flags & MS_REMOUNT)
1425 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1426 data_page);
1427 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001428 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001429 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001430 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 else if (flags & MS_MOVE)
1432 retval = do_move_mount(&nd, dev_name);
1433 else
1434 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1435 dev_name, data_page);
1436dput_out:
1437 path_release(&nd);
1438 return retval;
1439}
1440
JANAK DESAI741a2952006-02-07 12:59:00 -08001441/*
1442 * Allocate a new namespace structure and populate it with contents
1443 * copied from the namespace of the passed in task structure.
1444 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001445static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001446 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001448 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 struct vfsmount *p, *q;
1451
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001452 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001454 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001458 init_waitqueue_head(&new_ns->poll);
1459 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Ram Pai390c6842005-11-07 17:17:51 -05001461 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001463 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001464 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001466 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001468 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
1470 spin_lock(&vfsmount_lock);
1471 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1472 spin_unlock(&vfsmount_lock);
1473
1474 /*
1475 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1476 * as belonging to new namespace. We have already acquired a private
1477 * fs_struct, so tsk->fs->lock is not needed.
1478 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001479 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 q = new_ns->root;
1481 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001482 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 if (fs) {
1484 if (p == fs->rootmnt) {
1485 rootmnt = p;
1486 fs->rootmnt = mntget(q);
1487 }
1488 if (p == fs->pwdmnt) {
1489 pwdmnt = p;
1490 fs->pwdmnt = mntget(q);
1491 }
1492 if (p == fs->altrootmnt) {
1493 altrootmnt = p;
1494 fs->altrootmnt = mntget(q);
1495 }
1496 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001497 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 q = next_mnt(q, new_ns->root);
1499 }
Ram Pai390c6842005-11-07 17:17:51 -05001500 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (rootmnt)
1503 mntput(rootmnt);
1504 if (pwdmnt)
1505 mntput(pwdmnt);
1506 if (altrootmnt)
1507 mntput(altrootmnt);
1508
JANAK DESAI741a2952006-02-07 12:59:00 -08001509 return new_ns;
1510}
1511
Badari Pulavartye3222c42007-05-08 00:25:21 -07001512struct mnt_namespace *copy_mnt_ns(int flags, struct mnt_namespace *ns,
1513 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001514{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001515 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001516
Badari Pulavartye3222c42007-05-08 00:25:21 -07001517 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001518 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001519
1520 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001521 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001522
Badari Pulavartye3222c42007-05-08 00:25:21 -07001523 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001524
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001525 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001526 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
1528
1529asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1530 char __user * type, unsigned long flags,
1531 void __user * data)
1532{
1533 int retval;
1534 unsigned long data_page;
1535 unsigned long type_page;
1536 unsigned long dev_page;
1537 char *dir_page;
1538
Ram Paib58fed82005-11-07 17:16:09 -05001539 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 if (retval < 0)
1541 return retval;
1542
1543 dir_page = getname(dir_name);
1544 retval = PTR_ERR(dir_page);
1545 if (IS_ERR(dir_page))
1546 goto out1;
1547
Ram Paib58fed82005-11-07 17:16:09 -05001548 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 if (retval < 0)
1550 goto out2;
1551
Ram Paib58fed82005-11-07 17:16:09 -05001552 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (retval < 0)
1554 goto out3;
1555
1556 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001557 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1558 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 unlock_kernel();
1560 free_page(data_page);
1561
1562out3:
1563 free_page(dev_page);
1564out2:
1565 putname(dir_page);
1566out1:
1567 free_page(type_page);
1568 return retval;
1569}
1570
1571/*
1572 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1573 * It can block. Requires the big lock held.
1574 */
1575void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1576 struct dentry *dentry)
1577{
1578 struct dentry *old_root;
1579 struct vfsmount *old_rootmnt;
1580 write_lock(&fs->lock);
1581 old_root = fs->root;
1582 old_rootmnt = fs->rootmnt;
1583 fs->rootmnt = mntget(mnt);
1584 fs->root = dget(dentry);
1585 write_unlock(&fs->lock);
1586 if (old_root) {
1587 dput(old_root);
1588 mntput(old_rootmnt);
1589 }
1590}
1591
1592/*
1593 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1594 * It can block. Requires the big lock held.
1595 */
1596void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1597 struct dentry *dentry)
1598{
1599 struct dentry *old_pwd;
1600 struct vfsmount *old_pwdmnt;
1601
1602 write_lock(&fs->lock);
1603 old_pwd = fs->pwd;
1604 old_pwdmnt = fs->pwdmnt;
1605 fs->pwdmnt = mntget(mnt);
1606 fs->pwd = dget(dentry);
1607 write_unlock(&fs->lock);
1608
1609 if (old_pwd) {
1610 dput(old_pwd);
1611 mntput(old_pwdmnt);
1612 }
1613}
1614
1615static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1616{
1617 struct task_struct *g, *p;
1618 struct fs_struct *fs;
1619
1620 read_lock(&tasklist_lock);
1621 do_each_thread(g, p) {
1622 task_lock(p);
1623 fs = p->fs;
1624 if (fs) {
1625 atomic_inc(&fs->count);
1626 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001627 if (fs->root == old_nd->dentry
1628 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001630 if (fs->pwd == old_nd->dentry
1631 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1633 put_fs_struct(fs);
1634 } else
1635 task_unlock(p);
1636 } while_each_thread(g, p);
1637 read_unlock(&tasklist_lock);
1638}
1639
1640/*
1641 * pivot_root Semantics:
1642 * Moves the root file system of the current process to the directory put_old,
1643 * makes new_root as the new root file system of the current process, and sets
1644 * root/cwd of all processes which had them on the current root to new_root.
1645 *
1646 * Restrictions:
1647 * The new_root and put_old must be directories, and must not be on the
1648 * same file system as the current process root. The put_old must be
1649 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1650 * pointed to by put_old must yield the same directory as new_root. No other
1651 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1652 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001653 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1654 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1655 * in this situation.
1656 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 * Notes:
1658 * - we don't move root/cwd if they are not at the root (reason: if something
1659 * cared enough to change them, it's probably wrong to force them elsewhere)
1660 * - it's okay to pick a root that isn't the root of a file system, e.g.
1661 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1662 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1663 * first.
1664 */
Ram Paib58fed82005-11-07 17:16:09 -05001665asmlinkage long sys_pivot_root(const char __user * new_root,
1666 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667{
1668 struct vfsmount *tmp;
1669 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1670 int error;
1671
1672 if (!capable(CAP_SYS_ADMIN))
1673 return -EPERM;
1674
1675 lock_kernel();
1676
Ram Paib58fed82005-11-07 17:16:09 -05001677 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1678 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 if (error)
1680 goto out0;
1681 error = -EINVAL;
1682 if (!check_mnt(new_nd.mnt))
1683 goto out1;
1684
Ram Paib58fed82005-11-07 17:16:09 -05001685 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (error)
1687 goto out1;
1688
1689 error = security_sb_pivotroot(&old_nd, &new_nd);
1690 if (error) {
1691 path_release(&old_nd);
1692 goto out1;
1693 }
1694
1695 read_lock(&current->fs->lock);
1696 user_nd.mnt = mntget(current->fs->rootmnt);
1697 user_nd.dentry = dget(current->fs->root);
1698 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001699 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001700 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001702 if (IS_MNT_SHARED(old_nd.mnt) ||
1703 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1704 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1705 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 if (!check_mnt(user_nd.mnt))
1707 goto out2;
1708 error = -ENOENT;
1709 if (IS_DEADDIR(new_nd.dentry->d_inode))
1710 goto out2;
1711 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1712 goto out2;
1713 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1714 goto out2;
1715 error = -EBUSY;
1716 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1717 goto out2; /* loop, on the same file system */
1718 error = -EINVAL;
1719 if (user_nd.mnt->mnt_root != user_nd.dentry)
1720 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001721 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1722 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 if (new_nd.mnt->mnt_root != new_nd.dentry)
1724 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001725 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1726 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1728 spin_lock(&vfsmount_lock);
1729 if (tmp != new_nd.mnt) {
1730 for (;;) {
1731 if (tmp->mnt_parent == tmp)
1732 goto out3; /* already mounted on put_old */
1733 if (tmp->mnt_parent == new_nd.mnt)
1734 break;
1735 tmp = tmp->mnt_parent;
1736 }
1737 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1738 goto out3;
1739 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1740 goto out3;
1741 detach_mnt(new_nd.mnt, &parent_nd);
1742 detach_mnt(user_nd.mnt, &root_parent);
1743 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1744 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001745 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 spin_unlock(&vfsmount_lock);
1747 chroot_fs_refs(&user_nd, &new_nd);
1748 security_sb_post_pivotroot(&user_nd, &new_nd);
1749 error = 0;
1750 path_release(&root_parent);
1751 path_release(&parent_nd);
1752out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001753 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001754 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 path_release(&user_nd);
1756 path_release(&old_nd);
1757out1:
1758 path_release(&new_nd);
1759out0:
1760 unlock_kernel();
1761 return error;
1762out3:
1763 spin_unlock(&vfsmount_lock);
1764 goto out2;
1765}
1766
1767static void __init init_mount_tree(void)
1768{
1769 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001770 struct mnt_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1773 if (IS_ERR(mnt))
1774 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001775 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1776 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001778 atomic_set(&ns->count, 1);
1779 INIT_LIST_HEAD(&ns->list);
1780 init_waitqueue_head(&ns->poll);
1781 ns->event = 0;
1782 list_add(&mnt->mnt_list, &ns->list);
1783 ns->root = mnt;
1784 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001786 init_task.nsproxy->mnt_ns = ns;
1787 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001789 set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1790 set_fs_root(current->fs, ns->root, ns->root->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791}
1792
1793void __init mnt_init(unsigned long mempages)
1794{
1795 struct list_head *d;
1796 unsigned int nr_hash;
1797 int i;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001798 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
Ram Pai390c6842005-11-07 17:17:51 -05001800 init_rwsem(&namespace_sem);
1801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Ram Paib58fed82005-11-07 17:16:09 -05001803 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Ram Paib58fed82005-11-07 17:16:09 -05001805 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 if (!mount_hashtable)
1808 panic("Failed to allocate mount hash table\n");
1809
1810 /*
1811 * Find the power-of-two list-heads that can fit into the allocation..
1812 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1813 * a power-of-two.
1814 */
1815 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1816 hash_bits = 0;
1817 do {
1818 hash_bits++;
1819 } while ((nr_hash >> hash_bits) != 0);
1820 hash_bits--;
1821
1822 /*
1823 * Re-calculate the actual number of entries and the mask
1824 * from the number of bits we can fit.
1825 */
1826 nr_hash = 1UL << hash_bits;
Ram Paib58fed82005-11-07 17:16:09 -05001827 hash_mask = nr_hash - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 printk("Mount-cache hash table entries: %d\n", nr_hash);
1830
1831 /* And initialize the newly allocated array */
1832 d = mount_hashtable;
1833 i = nr_hash;
1834 do {
1835 INIT_LIST_HEAD(d);
1836 d++;
1837 i--;
1838 } while (i);
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001839 err = sysfs_init();
1840 if (err)
1841 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1842 __FUNCTION__, err);
1843 err = subsystem_register(&fs_subsys);
1844 if (err)
1845 printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1846 __FUNCTION__, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 init_rootfs();
1848 init_mount_tree();
1849}
1850
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001851void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001853 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001854 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001855 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001856 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001857 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001859 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001861 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001862 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001863 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864}