blob: e9c10cd01e1325cff95869cbb0f23aa8d14c10ce [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/quotaops.h>
18#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080019#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070021#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080023#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/namei.h>
25#include <linux/security.h>
26#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#include <linux/log2.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <asm/uaccess.h>
30#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050031#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070032#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
Eric Dumazet13f14b42008-02-06 01:37:57 -080034#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35#define HASH_SIZE (1UL << HASH_SHIFT)
36
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050038__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
39
40static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Eric Dumazetfa3536c2006-03-26 01:37:24 -080042static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080043static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050044static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080046/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060047struct kobject *fs_kobj;
48EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
51{
Ram Paib58fed82005-11-07 17:16:09 -050052 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
53 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080054 tmp = tmp + (tmp >> HASH_SHIFT);
55 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056}
57
58struct vfsmount *alloc_vfsmnt(const char *name)
59{
Robert P. J. Dayc3762222007-02-10 01:45:03 -080060 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 if (mnt) {
Ram Paib58fed82005-11-07 17:16:09 -050062 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 INIT_LIST_HEAD(&mnt->mnt_hash);
64 INIT_LIST_HEAD(&mnt->mnt_child);
65 INIT_LIST_HEAD(&mnt->mnt_mounts);
66 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070067 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050068 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050069 INIT_LIST_HEAD(&mnt->mnt_slave_list);
70 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050072 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 char *newname = kmalloc(size, GFP_KERNEL);
74 if (newname) {
75 memcpy(newname, name, size);
76 mnt->mnt_devname = newname;
77 }
78 }
79 }
80 return mnt;
81}
82
David Howells454e2392006-06-23 02:02:57 -070083int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
84{
85 mnt->mnt_sb = sb;
86 mnt->mnt_root = dget(sb->s_root);
87 return 0;
88}
89
90EXPORT_SYMBOL(simple_set_mnt);
91
Linus Torvalds1da177e2005-04-16 15:20:36 -070092void free_vfsmnt(struct vfsmount *mnt)
93{
94 kfree(mnt->mnt_devname);
95 kmem_cache_free(mnt_cache, mnt);
96}
97
98/*
Ram Paia05964f2005-11-07 17:20:17 -050099 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 */
Ram Paia05964f2005-11-07 17:20:17 -0500102struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
103 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Ram Paib58fed82005-11-07 17:16:09 -0500105 struct list_head *head = mount_hashtable + hash(mnt, dentry);
106 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 struct vfsmount *p, *found = NULL;
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500110 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 p = NULL;
112 if (tmp == head)
113 break;
114 p = list_entry(tmp, struct vfsmount, mnt_hash);
115 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500116 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 break;
118 }
119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return found;
121}
122
Ram Paia05964f2005-11-07 17:20:17 -0500123/*
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
126 */
127struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
128{
129 struct vfsmount *child_mnt;
130 spin_lock(&vfsmount_lock);
131 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 mntget(child_mnt);
133 spin_unlock(&vfsmount_lock);
134 return child_mnt;
135}
136
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137static inline int check_mnt(struct vfsmount *mnt)
138{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800139 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800142static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500143{
144 if (ns) {
145 ns->event = ++event;
146 wake_up_interruptible(&ns->poll);
147 }
148}
149
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800150static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500151{
152 if (ns && ns->event != event) {
153 ns->event = event;
154 wake_up_interruptible(&ns->poll);
155 }
156}
157
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
159{
160 old_nd->dentry = mnt->mnt_mountpoint;
161 old_nd->mnt = mnt->mnt_parent;
162 mnt->mnt_parent = mnt;
163 mnt->mnt_mountpoint = mnt->mnt_root;
164 list_del_init(&mnt->mnt_child);
165 list_del_init(&mnt->mnt_hash);
166 old_nd->dentry->d_mounted--;
167}
168
Ram Paib90fa9a2005-11-07 17:19:50 -0500169void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
170 struct vfsmount *child_mnt)
171{
172 child_mnt->mnt_parent = mntget(mnt);
173 child_mnt->mnt_mountpoint = dget(dentry);
174 dentry->d_mounted++;
175}
176
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
178{
Ram Paib90fa9a2005-11-07 17:19:50 -0500179 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
181 hash(nd->mnt, nd->dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500183}
184
185/*
186 * the caller must hold vfsmount_lock
187 */
188static void commit_tree(struct vfsmount *mnt)
189{
190 struct vfsmount *parent = mnt->mnt_parent;
191 struct vfsmount *m;
192 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800193 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500194
195 BUG_ON(parent == mnt);
196
197 list_add_tail(&head, &mnt->mnt_list);
198 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800199 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500200 list_splice(&head, n->list.prev);
201
202 list_add_tail(&mnt->mnt_hash, mount_hashtable +
203 hash(parent, mnt->mnt_mountpoint));
204 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800205 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
208static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
209{
210 struct list_head *next = p->mnt_mounts.next;
211 if (next == &p->mnt_mounts) {
212 while (1) {
213 if (p == root)
214 return NULL;
215 next = p->mnt_child.next;
216 if (next != &p->mnt_parent->mnt_mounts)
217 break;
218 p = p->mnt_parent;
219 }
220 }
221 return list_entry(next, struct vfsmount, mnt_child);
222}
223
Ram Pai9676f0c2005-11-07 17:21:20 -0500224static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
225{
226 struct list_head *prev = p->mnt_mounts.prev;
227 while (prev != &p->mnt_mounts) {
228 p = list_entry(prev, struct vfsmount, mnt_child);
229 prev = p->mnt_mounts.prev;
230 }
231 return p;
232}
233
Ram Pai36341f62005-11-07 17:17:22 -0500234static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
235 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
237 struct super_block *sb = old->mnt_sb;
238 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
239
240 if (mnt) {
241 mnt->mnt_flags = old->mnt_flags;
242 atomic_inc(&sb->s_active);
243 mnt->mnt_sb = sb;
244 mnt->mnt_root = dget(root);
245 mnt->mnt_mountpoint = mnt->mnt_root;
246 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500247
Ram Pai5afe0022005-11-07 17:21:01 -0500248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400252 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
258 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500259 if (flag & CL_MAKE_SHARED)
260 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
271 return mnt;
272}
273
Al Viro7b7b1ac2005-11-07 17:13:39 -0500274static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 struct super_block *sb = mnt->mnt_sb;
277 dput(mnt->mnt_root);
278 free_vfsmnt(mnt);
279 deactivate_super(sb);
280}
281
Al Viro7b7b1ac2005-11-07 17:13:39 -0500282void mntput_no_expire(struct vfsmount *mnt)
283{
284repeat:
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
288 __mntput(mnt);
289 return;
290 }
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
292 mnt->mnt_pinned = 0;
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
296 goto repeat;
297 }
298}
299
300EXPORT_SYMBOL(mntput_no_expire);
301
302void mnt_pin(struct vfsmount *mnt)
303{
304 spin_lock(&vfsmount_lock);
305 mnt->mnt_pinned++;
306 spin_unlock(&vfsmount_lock);
307}
308
309EXPORT_SYMBOL(mnt_pin);
310
311void mnt_unpin(struct vfsmount *mnt)
312{
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
316 mnt->mnt_pinned--;
317 }
318 spin_unlock(&vfsmount_lock);
319}
320
321EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
323/* iterator */
324static void *m_start(struct seq_file *m, loff_t *pos)
325{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800326 struct mnt_namespace *n = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
Ram Pai390c6842005-11-07 17:17:51 -0500328 down_read(&namespace_sem);
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700329 return seq_list_start(&n->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330}
331
332static void *m_next(struct seq_file *m, void *v, loff_t *pos)
333{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800334 struct mnt_namespace *n = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700335
336 return seq_list_next(v, &n->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337}
338
339static void m_stop(struct seq_file *m, void *v)
340{
Ram Pai390c6842005-11-07 17:17:51 -0500341 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
344static inline void mangle(struct seq_file *m, const char *s)
345{
346 seq_escape(m, s, " \t\n\\");
347}
348
349static int show_vfsmnt(struct seq_file *m, void *v)
350{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700351 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 int err = 0;
353 static struct proc_fs_info {
354 int flag;
355 char *str;
356 } fs_info[] = {
357 { MS_SYNCHRONOUS, ",sync" },
358 { MS_DIRSYNC, ",dirsync" },
359 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 { 0, NULL }
361 };
362 static struct proc_fs_info mnt_info[] = {
363 { MNT_NOSUID, ",nosuid" },
364 { MNT_NODEV, ",nodev" },
365 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800366 { MNT_NOATIME, ",noatime" },
367 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800368 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 { 0, NULL }
370 };
371 struct proc_fs_info *fs_infop;
372
373 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
374 seq_putc(m, ' ');
375 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
376 seq_putc(m, ' ');
377 mangle(m, mnt->mnt_sb->s_type->name);
Miklos Szeredi79c0b2d2007-05-08 00:25:43 -0700378 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
379 seq_putc(m, '.');
380 mangle(m, mnt->mnt_sb->s_subtype);
381 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
383 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
384 if (mnt->mnt_sb->s_flags & fs_infop->flag)
385 seq_puts(m, fs_infop->str);
386 }
387 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
388 if (mnt->mnt_flags & fs_infop->flag)
389 seq_puts(m, fs_infop->str);
390 }
391 if (mnt->mnt_sb->s_op->show_options)
392 err = mnt->mnt_sb->s_op->show_options(m, mnt);
393 seq_puts(m, " 0 0\n");
394 return err;
395}
396
397struct seq_operations mounts_op = {
398 .start = m_start,
399 .next = m_next,
400 .stop = m_stop,
401 .show = show_vfsmnt
402};
403
Chuck Leverb4629fe2006-03-20 13:44:12 -0500404static int show_vfsstat(struct seq_file *m, void *v)
405{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700406 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500407 int err = 0;
408
409 /* device */
410 if (mnt->mnt_devname) {
411 seq_puts(m, "device ");
412 mangle(m, mnt->mnt_devname);
413 } else
414 seq_puts(m, "no device");
415
416 /* mount point */
417 seq_puts(m, " mounted on ");
418 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
419 seq_putc(m, ' ');
420
421 /* file system type */
422 seq_puts(m, "with fstype ");
423 mangle(m, mnt->mnt_sb->s_type->name);
424
425 /* optional statistics */
426 if (mnt->mnt_sb->s_op->show_stats) {
427 seq_putc(m, ' ');
428 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
429 }
430
431 seq_putc(m, '\n');
432 return err;
433}
434
435struct seq_operations mountstats_op = {
436 .start = m_start,
437 .next = m_next,
438 .stop = m_stop,
439 .show = show_vfsstat,
440};
441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442/**
443 * may_umount_tree - check if a mount tree is busy
444 * @mnt: root of mount tree
445 *
446 * This is called to check if a tree of mounts has any
447 * open files, pwds, chroots or sub mounts that are
448 * busy.
449 */
450int may_umount_tree(struct vfsmount *mnt)
451{
Ram Pai36341f62005-11-07 17:17:22 -0500452 int actual_refs = 0;
453 int minimum_refs = 0;
454 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
456 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500457 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 actual_refs += atomic_read(&p->mnt_count);
459 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 }
461 spin_unlock(&vfsmount_lock);
462
463 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800464 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Ian Kente3474a82006-03-27 01:14:51 -0800466 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467}
468
469EXPORT_SYMBOL(may_umount_tree);
470
471/**
472 * may_umount - check if a mount point is busy
473 * @mnt: root of mount
474 *
475 * This is called to check if a mount point has any
476 * open files, pwds, chroots or sub mounts. If the
477 * mount has sub mounts this will return busy
478 * regardless of whether the sub mounts are busy.
479 *
480 * Doesn't take quota and stuff into account. IOW, in some cases it will
481 * give false negatives. The main reason why it's here is that we need
482 * a non-destructive way to look for easily umountable filesystems.
483 */
484int may_umount(struct vfsmount *mnt)
485{
Ian Kente3474a82006-03-27 01:14:51 -0800486 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500487 spin_lock(&vfsmount_lock);
488 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800489 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500490 spin_unlock(&vfsmount_lock);
491 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492}
493
494EXPORT_SYMBOL(may_umount);
495
Ram Paib90fa9a2005-11-07 17:19:50 -0500496void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500498 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800499 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700500 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500501 list_del_init(&mnt->mnt_hash);
502 if (mnt->mnt_parent != mnt) {
503 struct dentry *dentry;
504 struct vfsmount *m;
505 spin_lock(&vfsmount_lock);
506 dentry = mnt->mnt_mountpoint;
507 m = mnt->mnt_parent;
508 mnt->mnt_mountpoint = mnt->mnt_root;
509 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500511 dput(dentry);
512 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
514 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500515 }
516}
517
Ram Paia05964f2005-11-07 17:20:17 -0500518void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500519{
520 struct vfsmount *p;
521
Akinobu Mita1bfba4e2006-06-26 00:24:40 -0700522 for (p = mnt; p; p = next_mnt(p, mnt))
523 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500524
Ram Paia05964f2005-11-07 17:20:17 -0500525 if (propagate)
526 propagate_umount(kill);
527
Ram Pai70fbcdf2005-11-07 17:17:04 -0500528 list_for_each_entry(p, kill, mnt_hash) {
529 list_del_init(&p->mnt_expire);
530 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800531 __touch_mnt_namespace(p->mnt_ns);
532 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500533 list_del_init(&p->mnt_child);
534 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500535 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500536 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 }
538}
539
540static int do_umount(struct vfsmount *mnt, int flags)
541{
Ram Paib58fed82005-11-07 17:16:09 -0500542 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500544 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
546 retval = security_sb_umount(mnt, flags);
547 if (retval)
548 return retval;
549
550 /*
551 * Allow userspace to request a mountpoint be expired rather than
552 * unmounting unconditionally. Unmount only happens if:
553 * (1) the mark is already set (the mark is cleared by mntput())
554 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
555 */
556 if (flags & MNT_EXPIRE) {
557 if (mnt == current->fs->rootmnt ||
558 flags & (MNT_FORCE | MNT_DETACH))
559 return -EINVAL;
560
561 if (atomic_read(&mnt->mnt_count) != 2)
562 return -EBUSY;
563
564 if (!xchg(&mnt->mnt_expiry_mark, 1))
565 return -EAGAIN;
566 }
567
568 /*
569 * If we may have to abort operations to get out of this
570 * mount, and they will themselves hold resources we must
571 * allow the fs to do things. In the Unix tradition of
572 * 'Gee thats tricky lets do it in userspace' the umount_begin
573 * might fail to complete on the first run through as other tasks
574 * must return, and the like. Thats for the mount program to worry
575 * about for the moment.
576 */
577
578 lock_kernel();
Trond Myklebust8b512d92006-06-09 09:34:18 -0400579 if (sb->s_op->umount_begin)
580 sb->s_op->umount_begin(mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 unlock_kernel();
582
583 /*
584 * No sense to grab the lock for this test, but test itself looks
585 * somewhat bogus. Suggestions for better replacement?
586 * Ho-hum... In principle, we might treat that as umount + switch
587 * to rootfs. GC would eventually take care of the old vfsmount.
588 * Actually it makes sense, especially if rootfs would contain a
589 * /reboot - static binary that would close all descriptors and
590 * call reboot(9). Then init(8) could umount root and exec /reboot.
591 */
592 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
593 /*
594 * Special case for "unmounting" root ...
595 * we just try to remount it readonly.
596 */
597 down_write(&sb->s_umount);
598 if (!(sb->s_flags & MS_RDONLY)) {
599 lock_kernel();
600 DQUOT_OFF(sb);
601 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
602 unlock_kernel();
603 }
604 up_write(&sb->s_umount);
605 return retval;
606 }
607
Ram Pai390c6842005-11-07 17:17:51 -0500608 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500610 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500613 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500615 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 retval = 0;
617 }
618 spin_unlock(&vfsmount_lock);
619 if (retval)
620 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500621 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500622 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 return retval;
624}
625
626/*
627 * Now umount can handle mount points as well as block devices.
628 * This is important for filesystems which use unnamed block devices.
629 *
630 * We now support a flag for forced unmount like the other 'big iron'
631 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
632 */
633
634asmlinkage long sys_umount(char __user * name, int flags)
635{
636 struct nameidata nd;
637 int retval;
638
639 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
640 if (retval)
641 goto out;
642 retval = -EINVAL;
643 if (nd.dentry != nd.mnt->mnt_root)
644 goto dput_and_out;
645 if (!check_mnt(nd.mnt))
646 goto dput_and_out;
647
648 retval = -EPERM;
649 if (!capable(CAP_SYS_ADMIN))
650 goto dput_and_out;
651
652 retval = do_umount(nd.mnt, flags);
653dput_and_out:
654 path_release_on_umount(&nd);
655out:
656 return retval;
657}
658
659#ifdef __ARCH_WANT_SYS_OLDUMOUNT
660
661/*
Ram Paib58fed82005-11-07 17:16:09 -0500662 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664asmlinkage long sys_oldumount(char __user * name)
665{
Ram Paib58fed82005-11-07 17:16:09 -0500666 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
669#endif
670
671static int mount_is_safe(struct nameidata *nd)
672{
673 if (capable(CAP_SYS_ADMIN))
674 return 0;
675 return -EPERM;
676#ifdef notyet
677 if (S_ISLNK(nd->dentry->d_inode->i_mode))
678 return -EPERM;
679 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
680 if (current->uid != nd->dentry->d_inode->i_uid)
681 return -EPERM;
682 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800683 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 return -EPERM;
685 return 0;
686#endif
687}
688
Ram Paib58fed82005-11-07 17:16:09 -0500689static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 while (1) {
692 if (d == dentry)
693 return 1;
694 if (d == NULL || d == d->d_parent)
695 return 0;
696 d = d->d_parent;
697 }
698}
699
Ram Paib90fa9a2005-11-07 17:19:50 -0500700struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500701 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702{
703 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 struct nameidata nd;
705
Ram Pai9676f0c2005-11-07 17:21:20 -0500706 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
707 return NULL;
708
Ram Pai36341f62005-11-07 17:17:22 -0500709 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 if (!q)
711 goto Enomem;
712 q->mnt_mountpoint = mnt->mnt_mountpoint;
713
714 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700715 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
717 continue;
718
719 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500720 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
721 s = skip_mnt_tree(s);
722 continue;
723 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 while (p != s->mnt_parent) {
725 p = p->mnt_parent;
726 q = q->mnt_parent;
727 }
728 p = s;
729 nd.mnt = q;
730 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500731 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 if (!q)
733 goto Enomem;
734 spin_lock(&vfsmount_lock);
735 list_add_tail(&q->mnt_list, &res->mnt_list);
736 attach_mnt(q, &nd);
737 spin_unlock(&vfsmount_lock);
738 }
739 }
740 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500741Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500743 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500745 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500747 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
749 return NULL;
750}
751
Al Viro8aec0802007-06-07 12:20:32 -0400752struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
753{
754 struct vfsmount *tree;
755 down_read(&namespace_sem);
756 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
757 up_read(&namespace_sem);
758 return tree;
759}
760
761void drop_collected_mounts(struct vfsmount *mnt)
762{
763 LIST_HEAD(umount_list);
764 down_read(&namespace_sem);
765 spin_lock(&vfsmount_lock);
766 umount_tree(mnt, 0, &umount_list);
767 spin_unlock(&vfsmount_lock);
768 up_read(&namespace_sem);
769 release_mounts(&umount_list);
770}
771
Ram Paib90fa9a2005-11-07 17:19:50 -0500772/*
773 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500774 * @nd : place the mount tree @source_mnt is attached
775 * @parent_nd : if non-null, detach the source_mnt from its parent and
776 * store the parent mount and mountpoint dentry.
777 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500778 *
779 * NOTE: in the table below explains the semantics when a source mount
780 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500781 * ---------------------------------------------------------------------------
782 * | BIND MOUNT OPERATION |
783 * |**************************************************************************
784 * | source-->| shared | private | slave | unbindable |
785 * | dest | | | | |
786 * | | | | | | |
787 * | v | | | | |
788 * |**************************************************************************
789 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
790 * | | | | | |
791 * |non-shared| shared (+) | private | slave (*) | invalid |
792 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500793 * A bind operation clones the source mount and mounts the clone on the
794 * destination mount.
795 *
796 * (++) the cloned mount is propagated to all the mounts in the propagation
797 * tree of the destination mount and the cloned mount is added to
798 * the peer group of the source mount.
799 * (+) the cloned mount is created under the destination mount and is marked
800 * as shared. The cloned mount is added to the peer group of the source
801 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500802 * (+++) the mount is propagated to all the mounts in the propagation tree
803 * of the destination mount and the cloned mount is made slave
804 * of the same master as that of the source mount. The cloned mount
805 * is marked as 'shared and slave'.
806 * (*) the cloned mount is made a slave of the same master as that of the
807 * source mount.
808 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500809 * ---------------------------------------------------------------------------
810 * | MOVE MOUNT OPERATION |
811 * |**************************************************************************
812 * | source-->| shared | private | slave | unbindable |
813 * | dest | | | | |
814 * | | | | | | |
815 * | v | | | | |
816 * |**************************************************************************
817 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
818 * | | | | | |
819 * |non-shared| shared (+*) | private | slave (*) | unbindable |
820 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500821 *
822 * (+) the mount is moved to the destination. And is then propagated to
823 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500824 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500825 * (+++) the mount is moved to the destination and is then propagated to
826 * all the mounts belonging to the destination mount's propagation tree.
827 * the mount is marked as 'shared and slave'.
828 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500829 *
830 * if the source mount is a tree, the operations explained above is
831 * applied to each mount in the tree.
832 * Must be called without spinlocks held, since this function can sleep
833 * in allocations.
834 */
835static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500836 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500837{
838 LIST_HEAD(tree_list);
839 struct vfsmount *dest_mnt = nd->mnt;
840 struct dentry *dest_dentry = nd->dentry;
841 struct vfsmount *child, *p;
842
843 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
844 return -EINVAL;
845
846 if (IS_MNT_SHARED(dest_mnt)) {
847 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
848 set_mnt_shared(p);
849 }
850
851 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500852 if (parent_nd) {
853 detach_mnt(source_mnt, parent_nd);
854 attach_mnt(source_mnt, nd);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800855 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -0500856 } else {
857 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
858 commit_tree(source_mnt);
859 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500860
861 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
862 list_del_init(&child->mnt_hash);
863 commit_tree(child);
864 }
865 spin_unlock(&vfsmount_lock);
866 return 0;
867}
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
870{
871 int err;
872 if (mnt->mnt_sb->s_flags & MS_NOUSER)
873 return -EINVAL;
874
875 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
876 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
877 return -ENOTDIR;
878
879 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800880 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 if (IS_DEADDIR(nd->dentry->d_inode))
882 goto out_unlock;
883
884 err = security_sb_check_sb(mnt, nd);
885 if (err)
886 goto out_unlock;
887
888 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500889 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500890 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800892 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (!err)
894 security_sb_post_addmount(mnt, nd);
895 return err;
896}
897
898/*
Ram Pai07b20882005-11-07 17:19:07 -0500899 * recursively change the type of the mountpoint.
900 */
901static int do_change_type(struct nameidata *nd, int flag)
902{
903 struct vfsmount *m, *mnt = nd->mnt;
904 int recurse = flag & MS_REC;
905 int type = flag & ~MS_REC;
906
Miklos Szerediee6f9582007-05-08 00:30:40 -0700907 if (!capable(CAP_SYS_ADMIN))
908 return -EPERM;
909
Ram Pai07b20882005-11-07 17:19:07 -0500910 if (nd->dentry != nd->mnt->mnt_root)
911 return -EINVAL;
912
913 down_write(&namespace_sem);
914 spin_lock(&vfsmount_lock);
915 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
916 change_mnt_propagation(m, type);
917 spin_unlock(&vfsmount_lock);
918 up_write(&namespace_sem);
919 return 0;
920}
921
922/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 * do loopback mount.
924 */
Andrew Mortoneee391a2006-05-15 09:44:30 -0700925static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926{
927 struct nameidata old_nd;
928 struct vfsmount *mnt = NULL;
929 int err = mount_is_safe(nd);
930 if (err)
931 return err;
932 if (!old_name || !*old_name)
933 return -EINVAL;
934 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
935 if (err)
936 return err;
937
Ram Pai390c6842005-11-07 17:17:51 -0500938 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500940 if (IS_MNT_UNBINDABLE(old_nd.mnt))
941 goto out;
942
Al Viroccd48bc2005-11-07 17:15:04 -0500943 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
944 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945
Al Viroccd48bc2005-11-07 17:15:04 -0500946 err = -ENOMEM;
947 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500948 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500949 else
Ram Pai36341f62005-11-07 17:17:22 -0500950 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500951
952 if (!mnt)
953 goto out;
954
Al Viroccd48bc2005-11-07 17:15:04 -0500955 err = graft_tree(mnt, nd);
956 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500957 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500959 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500961 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -0500962 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963
Al Viroccd48bc2005-11-07 17:15:04 -0500964out:
Ram Pai390c6842005-11-07 17:17:51 -0500965 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 path_release(&old_nd);
967 return err;
968}
969
970/*
971 * change filesystem flags. dir should be a physical root of filesystem.
972 * If you've mounted a non-root directory somewhere and want to do remount
973 * on it - tough luck.
974 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
976 void *data)
977{
978 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500979 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if (!capable(CAP_SYS_ADMIN))
982 return -EPERM;
983
984 if (!check_mnt(nd->mnt))
985 return -EINVAL;
986
987 if (nd->dentry != nd->mnt->mnt_root)
988 return -EINVAL;
989
990 down_write(&sb->s_umount);
991 err = do_remount_sb(sb, flags, data, 0);
992 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500993 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 up_write(&sb->s_umount);
995 if (!err)
996 security_sb_post_remount(nd->mnt, flags, data);
997 return err;
998}
999
Ram Pai9676f0c2005-11-07 17:21:20 -05001000static inline int tree_contains_unbindable(struct vfsmount *mnt)
1001{
1002 struct vfsmount *p;
1003 for (p = mnt; p; p = next_mnt(p, mnt)) {
1004 if (IS_MNT_UNBINDABLE(p))
1005 return 1;
1006 }
1007 return 0;
1008}
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010static int do_move_mount(struct nameidata *nd, char *old_name)
1011{
1012 struct nameidata old_nd, parent_nd;
1013 struct vfsmount *p;
1014 int err = 0;
1015 if (!capable(CAP_SYS_ADMIN))
1016 return -EPERM;
1017 if (!old_name || !*old_name)
1018 return -EINVAL;
1019 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1020 if (err)
1021 return err;
1022
Ram Pai390c6842005-11-07 17:17:51 -05001023 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -05001024 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 ;
1026 err = -EINVAL;
1027 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1028 goto out;
1029
1030 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001031 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 if (IS_DEADDIR(nd->dentry->d_inode))
1033 goto out1;
1034
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001036 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
1038 err = -EINVAL;
1039 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001040 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001043 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1046 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001047 goto out1;
1048 /*
1049 * Don't move a mount residing in a shared parent.
1050 */
1051 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1052 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001053 /*
1054 * Don't move a mount tree containing unbindable mounts to a destination
1055 * mount which is shared.
1056 */
1057 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1058 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -05001060 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001062 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Ram Pai21444402005-11-07 17:20:03 -05001064 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1065 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
Ram Pai21444402005-11-07 17:20:03 -05001067 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 /* if the mount is moved, it should no longer be expire
1069 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001070 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 spin_unlock(&vfsmount_lock);
1072out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001073 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074out:
Ram Pai390c6842005-11-07 17:17:51 -05001075 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (!err)
1077 path_release(&parent_nd);
1078 path_release(&old_nd);
1079 return err;
1080}
1081
1082/*
1083 * create a new mount for userspace and request it to be added into the
1084 * namespace's tree
1085 */
1086static int do_new_mount(struct nameidata *nd, char *type, int flags,
1087 int mnt_flags, char *name, void *data)
1088{
1089 struct vfsmount *mnt;
1090
1091 if (!type || !memchr(type, 0, PAGE_SIZE))
1092 return -EINVAL;
1093
1094 /* we need capabilities... */
1095 if (!capable(CAP_SYS_ADMIN))
1096 return -EPERM;
1097
1098 mnt = do_kern_mount(type, flags, name, data);
1099 if (IS_ERR(mnt))
1100 return PTR_ERR(mnt);
1101
1102 return do_add_mount(mnt, nd, mnt_flags, NULL);
1103}
1104
1105/*
1106 * add a mount into a namespace's mount tree
1107 * - provide the option of adding the new mount to an expiration list
1108 */
1109int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1110 int mnt_flags, struct list_head *fslist)
1111{
1112 int err;
1113
Ram Pai390c6842005-11-07 17:17:51 -05001114 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001116 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 ;
1118 err = -EINVAL;
1119 if (!check_mnt(nd->mnt))
1120 goto unlock;
1121
1122 /* Refuse the same filesystem on the same mount point */
1123 err = -EBUSY;
1124 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1125 nd->mnt->mnt_root == nd->dentry)
1126 goto unlock;
1127
1128 err = -EINVAL;
1129 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1130 goto unlock;
1131
1132 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001133 if ((err = graft_tree(newmnt, nd)))
1134 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001136 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /* add to the specified expiration list */
1138 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001139 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 spin_unlock(&vfsmount_lock);
1141 }
Ram Pai390c6842005-11-07 17:17:51 -05001142 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001143 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144
1145unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001146 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 mntput(newmnt);
1148 return err;
1149}
1150
1151EXPORT_SYMBOL_GPL(do_add_mount);
1152
Ram Pai70fbcdf2005-11-07 17:17:04 -05001153static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1154 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001155{
1156 spin_lock(&vfsmount_lock);
1157
1158 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001159 * Check if mount is still attached, if not, let whoever holds it deal
1160 * with the sucker
1161 */
1162 if (mnt->mnt_parent == mnt) {
1163 spin_unlock(&vfsmount_lock);
1164 return;
1165 }
1166
1167 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001168 * Check that it is still dead: the count should now be 2 - as
1169 * contributed by the vfsmount parent and the mntget above
1170 */
Ram Paia05964f2005-11-07 17:20:17 -05001171 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001172 /* delete from the namespace */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001173 touch_mnt_namespace(mnt->mnt_ns);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001174 list_del_init(&mnt->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001175 mnt->mnt_ns = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001176 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001177 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001178 } else {
1179 /*
1180 * Someone brought it back to life whilst we didn't have any
1181 * locks held so return it to the expiration list
1182 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001183 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001184 spin_unlock(&vfsmount_lock);
1185 }
1186}
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001189 * go through the vfsmounts we've just consigned to the graveyard to
1190 * - check that they're still dead
1191 * - delete the vfsmount from the appropriate namespace under lock
1192 * - dispose of the corpse
1193 */
1194static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1195{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001196 struct mnt_namespace *ns;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001197 struct vfsmount *mnt;
1198
1199 while (!list_empty(graveyard)) {
1200 LIST_HEAD(umounts);
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001201 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001202 list_del_init(&mnt->mnt_expire);
1203
1204 /* don't do anything if the namespace is dead - all the
1205 * vfsmounts from it are going away anyway */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001206 ns = mnt->mnt_ns;
1207 if (!ns || !ns->root)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001208 continue;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001209 get_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001210
1211 spin_unlock(&vfsmount_lock);
1212 down_write(&namespace_sem);
1213 expire_mount(mnt, mounts, &umounts);
1214 up_write(&namespace_sem);
1215 release_mounts(&umounts);
1216 mntput(mnt);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001217 put_mnt_ns(ns);
Trond Myklebust5528f9112006-06-09 09:34:17 -04001218 spin_lock(&vfsmount_lock);
1219 }
1220}
1221
1222/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 * process a list of expirable mountpoints with the intent of discarding any
1224 * mountpoints that aren't in use and haven't been touched since last we came
1225 * here
1226 */
1227void mark_mounts_for_expiry(struct list_head *mounts)
1228{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 struct vfsmount *mnt, *next;
1230 LIST_HEAD(graveyard);
1231
1232 if (list_empty(mounts))
1233 return;
1234
1235 spin_lock(&vfsmount_lock);
1236
1237 /* extract from the expiration list every vfsmount that matches the
1238 * following criteria:
1239 * - only referenced by its parent vfsmount
1240 * - still marked for expiry (marked on the last call here; marks are
1241 * cleared by mntput())
1242 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001243 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1245 atomic_read(&mnt->mnt_count) != 1)
1246 continue;
1247
1248 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001249 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 }
1251
Trond Myklebust5528f9112006-06-09 09:34:17 -04001252 expire_mount_list(&graveyard, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 spin_unlock(&vfsmount_lock);
1255}
1256
1257EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1258
1259/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001260 * Ripoff of 'select_parent()'
1261 *
1262 * search the list of submounts for a given mountpoint, and move any
1263 * shrinkable submounts to the 'graveyard' list.
1264 */
1265static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1266{
1267 struct vfsmount *this_parent = parent;
1268 struct list_head *next;
1269 int found = 0;
1270
1271repeat:
1272 next = this_parent->mnt_mounts.next;
1273resume:
1274 while (next != &this_parent->mnt_mounts) {
1275 struct list_head *tmp = next;
1276 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1277
1278 next = tmp->next;
1279 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1280 continue;
1281 /*
1282 * Descend a level if the d_mounts list is non-empty.
1283 */
1284 if (!list_empty(&mnt->mnt_mounts)) {
1285 this_parent = mnt;
1286 goto repeat;
1287 }
1288
1289 if (!propagate_mount_busy(mnt, 1)) {
1290 mntget(mnt);
1291 list_move_tail(&mnt->mnt_expire, graveyard);
1292 found++;
1293 }
1294 }
1295 /*
1296 * All done at this level ... ascend and resume the search
1297 */
1298 if (this_parent != parent) {
1299 next = this_parent->mnt_child.next;
1300 this_parent = this_parent->mnt_parent;
1301 goto resume;
1302 }
1303 return found;
1304}
1305
1306/*
1307 * process a list of expirable mountpoints with the intent of discarding any
1308 * submounts of a specific parent mountpoint
1309 */
1310void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1311{
1312 LIST_HEAD(graveyard);
1313 int found;
1314
1315 spin_lock(&vfsmount_lock);
1316
1317 /* extract submounts of 'mountpoint' from the expiration list */
1318 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1319 expire_mount_list(&graveyard, mounts);
1320
1321 spin_unlock(&vfsmount_lock);
1322}
1323
1324EXPORT_SYMBOL_GPL(shrink_submounts);
1325
1326/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 * Some copy_from_user() implementations do not return the exact number of
1328 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1329 * Note that this function differs from copy_from_user() in that it will oops
1330 * on bad values of `to', rather than returning a short copy.
1331 */
Ram Paib58fed82005-11-07 17:16:09 -05001332static long exact_copy_from_user(void *to, const void __user * from,
1333 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
1335 char *t = to;
1336 const char __user *f = from;
1337 char c;
1338
1339 if (!access_ok(VERIFY_READ, from, n))
1340 return n;
1341
1342 while (n) {
1343 if (__get_user(c, f)) {
1344 memset(t, 0, n);
1345 break;
1346 }
1347 *t++ = c;
1348 f++;
1349 n--;
1350 }
1351 return n;
1352}
1353
Ram Paib58fed82005-11-07 17:16:09 -05001354int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
1356 int i;
1357 unsigned long page;
1358 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 *where = 0;
1361 if (!data)
1362 return 0;
1363
1364 if (!(page = __get_free_page(GFP_KERNEL)))
1365 return -ENOMEM;
1366
1367 /* We only care that *some* data at the address the user
1368 * gave us is valid. Just in case, we'll zero
1369 * the remainder of the page.
1370 */
1371 /* copy_from_user cannot cross TASK_SIZE ! */
1372 size = TASK_SIZE - (unsigned long)data;
1373 if (size > PAGE_SIZE)
1374 size = PAGE_SIZE;
1375
1376 i = size - exact_copy_from_user((void *)page, data, size);
1377 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001378 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 return -EFAULT;
1380 }
1381 if (i != PAGE_SIZE)
1382 memset((char *)page + i, 0, PAGE_SIZE - i);
1383 *where = page;
1384 return 0;
1385}
1386
1387/*
1388 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1389 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1390 *
1391 * data is a (void *) that can point to any structure up to
1392 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1393 * information (or be NULL).
1394 *
1395 * Pre-0.97 versions of mount() didn't have a flags word.
1396 * When the flags word was introduced its top half was required
1397 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1398 * Therefore, if this magic number is present, it carries no information
1399 * and must be discarded.
1400 */
Ram Paib58fed82005-11-07 17:16:09 -05001401long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 unsigned long flags, void *data_page)
1403{
1404 struct nameidata nd;
1405 int retval = 0;
1406 int mnt_flags = 0;
1407
1408 /* Discard magic */
1409 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1410 flags &= ~MS_MGC_MSK;
1411
1412 /* Basic sanity checks */
1413
1414 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1415 return -EINVAL;
1416 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1417 return -EINVAL;
1418
1419 if (data_page)
1420 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1421
1422 /* Separate the per-mountpoint flags */
1423 if (flags & MS_NOSUID)
1424 mnt_flags |= MNT_NOSUID;
1425 if (flags & MS_NODEV)
1426 mnt_flags |= MNT_NODEV;
1427 if (flags & MS_NOEXEC)
1428 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001429 if (flags & MS_NOATIME)
1430 mnt_flags |= MNT_NOATIME;
1431 if (flags & MS_NODIRATIME)
1432 mnt_flags |= MNT_NODIRATIME;
Valerie Henson47ae32d2006-12-13 00:34:34 -08001433 if (flags & MS_RELATIME)
1434 mnt_flags |= MNT_RELATIME;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001435
1436 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Pavel Emelyanov8bf97252007-10-18 23:40:02 -07001437 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439 /* ... and get the mountpoint */
1440 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1441 if (retval)
1442 return retval;
1443
1444 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1445 if (retval)
1446 goto dput_out;
1447
1448 if (flags & MS_REMOUNT)
1449 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1450 data_page);
1451 else if (flags & MS_BIND)
Andrew Mortoneee391a2006-05-15 09:44:30 -07001452 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001453 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001454 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 else if (flags & MS_MOVE)
1456 retval = do_move_mount(&nd, dev_name);
1457 else
1458 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1459 dev_name, data_page);
1460dput_out:
1461 path_release(&nd);
1462 return retval;
1463}
1464
JANAK DESAI741a2952006-02-07 12:59:00 -08001465/*
1466 * Allocate a new namespace structure and populate it with contents
1467 * copied from the namespace of the passed in task structure.
1468 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001469static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001470 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001472 struct mnt_namespace *new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 struct vfsmount *p, *q;
1475
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001476 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001478 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
1480 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001482 init_waitqueue_head(&new_ns->poll);
1483 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Ram Pai390c6842005-11-07 17:17:51 -05001485 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001487 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001488 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001490 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 kfree(new_ns);
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001492 return ERR_PTR(-ENOMEM);;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 }
1494 spin_lock(&vfsmount_lock);
1495 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1496 spin_unlock(&vfsmount_lock);
1497
1498 /*
1499 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1500 * as belonging to new namespace. We have already acquired a private
1501 * fs_struct, so tsk->fs->lock is not needed.
1502 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001503 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 q = new_ns->root;
1505 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001506 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 if (fs) {
1508 if (p == fs->rootmnt) {
1509 rootmnt = p;
1510 fs->rootmnt = mntget(q);
1511 }
1512 if (p == fs->pwdmnt) {
1513 pwdmnt = p;
1514 fs->pwdmnt = mntget(q);
1515 }
1516 if (p == fs->altrootmnt) {
1517 altrootmnt = p;
1518 fs->altrootmnt = mntget(q);
1519 }
1520 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001521 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 q = next_mnt(q, new_ns->root);
1523 }
Ram Pai390c6842005-11-07 17:17:51 -05001524 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 if (rootmnt)
1527 mntput(rootmnt);
1528 if (pwdmnt)
1529 mntput(pwdmnt);
1530 if (altrootmnt)
1531 mntput(altrootmnt);
1532
JANAK DESAI741a2952006-02-07 12:59:00 -08001533 return new_ns;
1534}
1535
Eric W. Biederman213dd262007-07-15 23:41:15 -07001536struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07001537 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08001538{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001539 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001540
Badari Pulavartye3222c42007-05-08 00:25:21 -07001541 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001542 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08001543
1544 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07001545 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08001546
Badari Pulavartye3222c42007-05-08 00:25:21 -07001547 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08001548
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001549 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07001550 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551}
1552
1553asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1554 char __user * type, unsigned long flags,
1555 void __user * data)
1556{
1557 int retval;
1558 unsigned long data_page;
1559 unsigned long type_page;
1560 unsigned long dev_page;
1561 char *dir_page;
1562
Ram Paib58fed82005-11-07 17:16:09 -05001563 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (retval < 0)
1565 return retval;
1566
1567 dir_page = getname(dir_name);
1568 retval = PTR_ERR(dir_page);
1569 if (IS_ERR(dir_page))
1570 goto out1;
1571
Ram Paib58fed82005-11-07 17:16:09 -05001572 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 if (retval < 0)
1574 goto out2;
1575
Ram Paib58fed82005-11-07 17:16:09 -05001576 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 if (retval < 0)
1578 goto out3;
1579
1580 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001581 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1582 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 unlock_kernel();
1584 free_page(data_page);
1585
1586out3:
1587 free_page(dev_page);
1588out2:
1589 putname(dir_page);
1590out1:
1591 free_page(type_page);
1592 return retval;
1593}
1594
1595/*
1596 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1597 * It can block. Requires the big lock held.
1598 */
1599void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1600 struct dentry *dentry)
1601{
1602 struct dentry *old_root;
1603 struct vfsmount *old_rootmnt;
1604 write_lock(&fs->lock);
1605 old_root = fs->root;
1606 old_rootmnt = fs->rootmnt;
1607 fs->rootmnt = mntget(mnt);
1608 fs->root = dget(dentry);
1609 write_unlock(&fs->lock);
1610 if (old_root) {
1611 dput(old_root);
1612 mntput(old_rootmnt);
1613 }
1614}
1615
1616/*
1617 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1618 * It can block. Requires the big lock held.
1619 */
1620void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1621 struct dentry *dentry)
1622{
1623 struct dentry *old_pwd;
1624 struct vfsmount *old_pwdmnt;
1625
1626 write_lock(&fs->lock);
1627 old_pwd = fs->pwd;
1628 old_pwdmnt = fs->pwdmnt;
1629 fs->pwdmnt = mntget(mnt);
1630 fs->pwd = dget(dentry);
1631 write_unlock(&fs->lock);
1632
1633 if (old_pwd) {
1634 dput(old_pwd);
1635 mntput(old_pwdmnt);
1636 }
1637}
1638
1639static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1640{
1641 struct task_struct *g, *p;
1642 struct fs_struct *fs;
1643
1644 read_lock(&tasklist_lock);
1645 do_each_thread(g, p) {
1646 task_lock(p);
1647 fs = p->fs;
1648 if (fs) {
1649 atomic_inc(&fs->count);
1650 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001651 if (fs->root == old_nd->dentry
1652 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001654 if (fs->pwd == old_nd->dentry
1655 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1657 put_fs_struct(fs);
1658 } else
1659 task_unlock(p);
1660 } while_each_thread(g, p);
1661 read_unlock(&tasklist_lock);
1662}
1663
1664/*
1665 * pivot_root Semantics:
1666 * Moves the root file system of the current process to the directory put_old,
1667 * makes new_root as the new root file system of the current process, and sets
1668 * root/cwd of all processes which had them on the current root to new_root.
1669 *
1670 * Restrictions:
1671 * The new_root and put_old must be directories, and must not be on the
1672 * same file system as the current process root. The put_old must be
1673 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1674 * pointed to by put_old must yield the same directory as new_root. No other
1675 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1676 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001677 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1678 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1679 * in this situation.
1680 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 * Notes:
1682 * - we don't move root/cwd if they are not at the root (reason: if something
1683 * cared enough to change them, it's probably wrong to force them elsewhere)
1684 * - it's okay to pick a root that isn't the root of a file system, e.g.
1685 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1686 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1687 * first.
1688 */
Ram Paib58fed82005-11-07 17:16:09 -05001689asmlinkage long sys_pivot_root(const char __user * new_root,
1690 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691{
1692 struct vfsmount *tmp;
1693 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1694 int error;
1695
1696 if (!capable(CAP_SYS_ADMIN))
1697 return -EPERM;
1698
1699 lock_kernel();
1700
Ram Paib58fed82005-11-07 17:16:09 -05001701 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1702 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (error)
1704 goto out0;
1705 error = -EINVAL;
1706 if (!check_mnt(new_nd.mnt))
1707 goto out1;
1708
Ram Paib58fed82005-11-07 17:16:09 -05001709 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 if (error)
1711 goto out1;
1712
1713 error = security_sb_pivotroot(&old_nd, &new_nd);
1714 if (error) {
1715 path_release(&old_nd);
1716 goto out1;
1717 }
1718
1719 read_lock(&current->fs->lock);
1720 user_nd.mnt = mntget(current->fs->rootmnt);
1721 user_nd.dentry = dget(current->fs->root);
1722 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001723 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001724 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001726 if (IS_MNT_SHARED(old_nd.mnt) ||
1727 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1728 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1729 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 if (!check_mnt(user_nd.mnt))
1731 goto out2;
1732 error = -ENOENT;
1733 if (IS_DEADDIR(new_nd.dentry->d_inode))
1734 goto out2;
1735 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1736 goto out2;
1737 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1738 goto out2;
1739 error = -EBUSY;
1740 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1741 goto out2; /* loop, on the same file system */
1742 error = -EINVAL;
1743 if (user_nd.mnt->mnt_root != user_nd.dentry)
1744 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001745 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1746 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 if (new_nd.mnt->mnt_root != new_nd.dentry)
1748 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001749 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1750 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1752 spin_lock(&vfsmount_lock);
1753 if (tmp != new_nd.mnt) {
1754 for (;;) {
1755 if (tmp->mnt_parent == tmp)
1756 goto out3; /* already mounted on put_old */
1757 if (tmp->mnt_parent == new_nd.mnt)
1758 break;
1759 tmp = tmp->mnt_parent;
1760 }
1761 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1762 goto out3;
1763 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1764 goto out3;
1765 detach_mnt(new_nd.mnt, &parent_nd);
1766 detach_mnt(user_nd.mnt, &root_parent);
1767 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1768 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001769 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 spin_unlock(&vfsmount_lock);
1771 chroot_fs_refs(&user_nd, &new_nd);
1772 security_sb_post_pivotroot(&user_nd, &new_nd);
1773 error = 0;
1774 path_release(&root_parent);
1775 path_release(&parent_nd);
1776out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001777 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001778 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 path_release(&user_nd);
1780 path_release(&old_nd);
1781out1:
1782 path_release(&new_nd);
1783out0:
1784 unlock_kernel();
1785 return error;
1786out3:
1787 spin_unlock(&vfsmount_lock);
1788 goto out2;
1789}
1790
1791static void __init init_mount_tree(void)
1792{
1793 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001794 struct mnt_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1797 if (IS_ERR(mnt))
1798 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001799 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1800 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001802 atomic_set(&ns->count, 1);
1803 INIT_LIST_HEAD(&ns->list);
1804 init_waitqueue_head(&ns->poll);
1805 ns->event = 0;
1806 list_add(&mnt->mnt_list, &ns->list);
1807 ns->root = mnt;
1808 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001810 init_task.nsproxy->mnt_ns = ns;
1811 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001813 set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1814 set_fs_root(current->fs, ns->root, ns->root->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815}
1816
Denis Cheng74bf17c2007-10-16 23:26:30 -07001817void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818{
Eric Dumazet13f14b42008-02-06 01:37:57 -08001819 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001820 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821
Ram Pai390c6842005-11-07 17:17:51 -05001822 init_rwsem(&namespace_sem);
1823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09001825 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826
Ram Paib58fed82005-11-07 17:16:09 -05001827 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 if (!mount_hashtable)
1830 panic("Failed to allocate mount hash table\n");
1831
Eric Dumazet13f14b42008-02-06 01:37:57 -08001832 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Eric Dumazet13f14b42008-02-06 01:37:57 -08001834 for (u = 0; u < HASH_SIZE; u++)
1835 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836
Randy Dunlap15a67dd2006-09-29 01:58:57 -07001837 err = sysfs_init();
1838 if (err)
1839 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1840 __FUNCTION__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06001841 fs_kobj = kobject_create_and_add("fs", NULL);
1842 if (!fs_kobj)
1843 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 init_rootfs();
1845 init_mount_tree();
1846}
1847
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001848void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001850 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001851 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001852 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001853 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001854 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001856 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001858 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001859 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001860 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861}