blob: 058a44865bebdad95cb9416eefcedd1dd1473c25 [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
11#include <linux/config.h>
12#include <linux/syscalls.h>
13#include <linux/slab.h>
14#include <linux/sched.h>
15#include <linux/smp_lock.h>
16#include <linux/init.h>
17#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>
21#include <linux/seq_file.h>
22#include <linux/namespace.h>
23#include <linux/namei.h>
24#include <linux/security.h>
25#include <linux/mount.h>
26#include <asm/uaccess.h>
27#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050028#include "pnode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30extern int __init init_rootfs(void);
31
32#ifdef CONFIG_SYSFS
33extern int __init sysfs_init(void);
34#else
35static inline int sysfs_init(void)
36{
37 return 0;
38}
39#endif
40
41/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050042__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43
44static int event;
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46static struct list_head *mount_hashtable;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070047static int hash_mask __read_mostly, hash_bits __read_mostly;
Ram Paib58fed82005-11-07 17:16:09 -050048static kmem_cache_t *mnt_cache;
Ram Pai390c6842005-11-07 17:17:51 -050049static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080051/* /sys/fs */
52decl_subsys(fs, NULL, NULL);
53EXPORT_SYMBOL_GPL(fs_subsys);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56{
Ram Paib58fed82005-11-07 17:16:09 -050057 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
58 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059 tmp = tmp + (tmp >> hash_bits);
60 return tmp & hash_mask;
61}
62
63struct vfsmount *alloc_vfsmnt(const char *name)
64{
Ram Paib58fed82005-11-07 17:16:09 -050065 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 if (mnt) {
67 memset(mnt, 0, sizeof(struct vfsmount));
Ram Paib58fed82005-11-07 17:16:09 -050068 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 INIT_LIST_HEAD(&mnt->mnt_hash);
70 INIT_LIST_HEAD(&mnt->mnt_child);
71 INIT_LIST_HEAD(&mnt->mnt_mounts);
72 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070073 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -050074 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -050075 INIT_LIST_HEAD(&mnt->mnt_slave_list);
76 INIT_LIST_HEAD(&mnt->mnt_slave);
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 if (name) {
Ram Paib58fed82005-11-07 17:16:09 -050078 int size = strlen(name) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 char *newname = kmalloc(size, GFP_KERNEL);
80 if (newname) {
81 memcpy(newname, name, size);
82 mnt->mnt_devname = newname;
83 }
84 }
85 }
86 return mnt;
87}
88
89void free_vfsmnt(struct vfsmount *mnt)
90{
91 kfree(mnt->mnt_devname);
92 kmem_cache_free(mnt_cache, mnt);
93}
94
95/*
Ram Paia05964f2005-11-07 17:20:17 -050096 * find the first or last mount at @dentry on vfsmount @mnt depending on
97 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 */
Ram Paia05964f2005-11-07 17:20:17 -050099struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
100 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Ram Paib58fed82005-11-07 17:16:09 -0500102 struct list_head *head = mount_hashtable + hash(mnt, dentry);
103 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 struct vfsmount *p, *found = NULL;
105
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500107 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 p = NULL;
109 if (tmp == head)
110 break;
111 p = list_entry(tmp, struct vfsmount, mnt_hash);
112 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500113 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 break;
115 }
116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 return found;
118}
119
Ram Paia05964f2005-11-07 17:20:17 -0500120/*
121 * lookup_mnt increments the ref count before returning
122 * the vfsmount struct.
123 */
124struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
125{
126 struct vfsmount *child_mnt;
127 spin_lock(&vfsmount_lock);
128 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
129 mntget(child_mnt);
130 spin_unlock(&vfsmount_lock);
131 return child_mnt;
132}
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134static inline int check_mnt(struct vfsmount *mnt)
135{
136 return mnt->mnt_namespace == current->namespace;
137}
138
Al Viro5addc5d2005-11-07 17:15:49 -0500139static void touch_namespace(struct namespace *ns)
140{
141 if (ns) {
142 ns->event = ++event;
143 wake_up_interruptible(&ns->poll);
144 }
145}
146
147static void __touch_namespace(struct namespace *ns)
148{
149 if (ns && ns->event != event) {
150 ns->event = event;
151 wake_up_interruptible(&ns->poll);
152 }
153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
156{
157 old_nd->dentry = mnt->mnt_mountpoint;
158 old_nd->mnt = mnt->mnt_parent;
159 mnt->mnt_parent = mnt;
160 mnt->mnt_mountpoint = mnt->mnt_root;
161 list_del_init(&mnt->mnt_child);
162 list_del_init(&mnt->mnt_hash);
163 old_nd->dentry->d_mounted--;
164}
165
Ram Paib90fa9a2005-11-07 17:19:50 -0500166void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
167 struct vfsmount *child_mnt)
168{
169 child_mnt->mnt_parent = mntget(mnt);
170 child_mnt->mnt_mountpoint = dget(dentry);
171 dentry->d_mounted++;
172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
175{
Ram Paib90fa9a2005-11-07 17:19:50 -0500176 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
177 list_add_tail(&mnt->mnt_hash, mount_hashtable +
178 hash(nd->mnt, nd->dentry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500180}
181
182/*
183 * the caller must hold vfsmount_lock
184 */
185static void commit_tree(struct vfsmount *mnt)
186{
187 struct vfsmount *parent = mnt->mnt_parent;
188 struct vfsmount *m;
189 LIST_HEAD(head);
190 struct namespace *n = parent->mnt_namespace;
191
192 BUG_ON(parent == mnt);
193
194 list_add_tail(&head, &mnt->mnt_list);
195 list_for_each_entry(m, &head, mnt_list)
196 m->mnt_namespace = n;
197 list_splice(&head, n->list.prev);
198
199 list_add_tail(&mnt->mnt_hash, mount_hashtable +
200 hash(parent, mnt->mnt_mountpoint));
201 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
202 touch_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
206{
207 struct list_head *next = p->mnt_mounts.next;
208 if (next == &p->mnt_mounts) {
209 while (1) {
210 if (p == root)
211 return NULL;
212 next = p->mnt_child.next;
213 if (next != &p->mnt_parent->mnt_mounts)
214 break;
215 p = p->mnt_parent;
216 }
217 }
218 return list_entry(next, struct vfsmount, mnt_child);
219}
220
Ram Pai9676f0c2005-11-07 17:21:20 -0500221static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
222{
223 struct list_head *prev = p->mnt_mounts.prev;
224 while (prev != &p->mnt_mounts) {
225 p = list_entry(prev, struct vfsmount, mnt_child);
226 prev = p->mnt_mounts.prev;
227 }
228 return p;
229}
230
Ram Pai36341f62005-11-07 17:17:22 -0500231static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
232 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
234 struct super_block *sb = old->mnt_sb;
235 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
236
237 if (mnt) {
238 mnt->mnt_flags = old->mnt_flags;
239 atomic_inc(&sb->s_active);
240 mnt->mnt_sb = sb;
241 mnt->mnt_root = dget(root);
242 mnt->mnt_mountpoint = mnt->mnt_root;
243 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500244
Ram Pai5afe0022005-11-07 17:21:01 -0500245 if (flag & CL_SLAVE) {
246 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
247 mnt->mnt_master = old;
248 CLEAR_MNT_SHARED(mnt);
249 } else {
250 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
251 list_add(&mnt->mnt_share, &old->mnt_share);
252 if (IS_MNT_SLAVE(old))
253 list_add(&mnt->mnt_slave, &old->mnt_slave);
254 mnt->mnt_master = old->mnt_master;
255 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500256 if (flag & CL_MAKE_SHARED)
257 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 /* stick the duplicate mount on the same expiry list
260 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500261 if (flag & CL_EXPIRE) {
262 spin_lock(&vfsmount_lock);
263 if (!list_empty(&old->mnt_expire))
264 list_add(&mnt->mnt_expire, &old->mnt_expire);
265 spin_unlock(&vfsmount_lock);
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 }
268 return mnt;
269}
270
Al Viro7b7b1ac2005-11-07 17:13:39 -0500271static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct super_block *sb = mnt->mnt_sb;
274 dput(mnt->mnt_root);
275 free_vfsmnt(mnt);
276 deactivate_super(sb);
277}
278
Al Viro7b7b1ac2005-11-07 17:13:39 -0500279void mntput_no_expire(struct vfsmount *mnt)
280{
281repeat:
282 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
283 if (likely(!mnt->mnt_pinned)) {
284 spin_unlock(&vfsmount_lock);
285 __mntput(mnt);
286 return;
287 }
288 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
289 mnt->mnt_pinned = 0;
290 spin_unlock(&vfsmount_lock);
291 acct_auto_close_mnt(mnt);
292 security_sb_umount_close(mnt);
293 goto repeat;
294 }
295}
296
297EXPORT_SYMBOL(mntput_no_expire);
298
299void mnt_pin(struct vfsmount *mnt)
300{
301 spin_lock(&vfsmount_lock);
302 mnt->mnt_pinned++;
303 spin_unlock(&vfsmount_lock);
304}
305
306EXPORT_SYMBOL(mnt_pin);
307
308void mnt_unpin(struct vfsmount *mnt)
309{
310 spin_lock(&vfsmount_lock);
311 if (mnt->mnt_pinned) {
312 atomic_inc(&mnt->mnt_count);
313 mnt->mnt_pinned--;
314 }
315 spin_unlock(&vfsmount_lock);
316}
317
318EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320/* iterator */
321static void *m_start(struct seq_file *m, loff_t *pos)
322{
323 struct namespace *n = m->private;
324 struct list_head *p;
325 loff_t l = *pos;
326
Ram Pai390c6842005-11-07 17:17:51 -0500327 down_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 list_for_each(p, &n->list)
329 if (!l--)
330 return list_entry(p, struct vfsmount, mnt_list);
331 return NULL;
332}
333
334static void *m_next(struct seq_file *m, void *v, loff_t *pos)
335{
336 struct namespace *n = m->private;
337 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
338 (*pos)++;
Ram Paib58fed82005-11-07 17:16:09 -0500339 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
341
342static void m_stop(struct seq_file *m, void *v)
343{
Ram Pai390c6842005-11-07 17:17:51 -0500344 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
347static inline void mangle(struct seq_file *m, const char *s)
348{
349 seq_escape(m, s, " \t\n\\");
350}
351
352static int show_vfsmnt(struct seq_file *m, void *v)
353{
354 struct vfsmount *mnt = v;
355 int err = 0;
356 static struct proc_fs_info {
357 int flag;
358 char *str;
359 } fs_info[] = {
360 { MS_SYNCHRONOUS, ",sync" },
361 { MS_DIRSYNC, ",dirsync" },
362 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 { 0, NULL }
364 };
365 static struct proc_fs_info mnt_info[] = {
366 { MNT_NOSUID, ",nosuid" },
367 { MNT_NODEV, ",nodev" },
368 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800369 { MNT_NOATIME, ",noatime" },
370 { MNT_NODIRATIME, ",nodiratime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 { 0, NULL }
372 };
373 struct proc_fs_info *fs_infop;
374
375 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
376 seq_putc(m, ' ');
377 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
378 seq_putc(m, ' ');
379 mangle(m, mnt->mnt_sb->s_type->name);
380 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
381 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
382 if (mnt->mnt_sb->s_flags & fs_infop->flag)
383 seq_puts(m, fs_infop->str);
384 }
385 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
386 if (mnt->mnt_flags & fs_infop->flag)
387 seq_puts(m, fs_infop->str);
388 }
389 if (mnt->mnt_sb->s_op->show_options)
390 err = mnt->mnt_sb->s_op->show_options(m, mnt);
391 seq_puts(m, " 0 0\n");
392 return err;
393}
394
395struct seq_operations mounts_op = {
396 .start = m_start,
397 .next = m_next,
398 .stop = m_stop,
399 .show = show_vfsmnt
400};
401
402/**
403 * may_umount_tree - check if a mount tree is busy
404 * @mnt: root of mount tree
405 *
406 * This is called to check if a tree of mounts has any
407 * open files, pwds, chroots or sub mounts that are
408 * busy.
409 */
410int may_umount_tree(struct vfsmount *mnt)
411{
Ram Pai36341f62005-11-07 17:17:22 -0500412 int actual_refs = 0;
413 int minimum_refs = 0;
414 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500417 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 actual_refs += atomic_read(&p->mnt_count);
419 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 }
421 spin_unlock(&vfsmount_lock);
422
423 if (actual_refs > minimum_refs)
424 return -EBUSY;
425
426 return 0;
427}
428
429EXPORT_SYMBOL(may_umount_tree);
430
431/**
432 * may_umount - check if a mount point is busy
433 * @mnt: root of mount
434 *
435 * This is called to check if a mount point has any
436 * open files, pwds, chroots or sub mounts. If the
437 * mount has sub mounts this will return busy
438 * regardless of whether the sub mounts are busy.
439 *
440 * Doesn't take quota and stuff into account. IOW, in some cases it will
441 * give false negatives. The main reason why it's here is that we need
442 * a non-destructive way to look for easily umountable filesystems.
443 */
444int may_umount(struct vfsmount *mnt)
445{
Ram Paia05964f2005-11-07 17:20:17 -0500446 int ret = 0;
447 spin_lock(&vfsmount_lock);
448 if (propagate_mount_busy(mnt, 2))
449 ret = -EBUSY;
450 spin_unlock(&vfsmount_lock);
451 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452}
453
454EXPORT_SYMBOL(may_umount);
455
Ram Paib90fa9a2005-11-07 17:19:50 -0500456void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500458 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800459 while (!list_empty(head)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500460 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
461 list_del_init(&mnt->mnt_hash);
462 if (mnt->mnt_parent != mnt) {
463 struct dentry *dentry;
464 struct vfsmount *m;
465 spin_lock(&vfsmount_lock);
466 dentry = mnt->mnt_mountpoint;
467 m = mnt->mnt_parent;
468 mnt->mnt_mountpoint = mnt->mnt_root;
469 mnt->mnt_parent = mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500471 dput(dentry);
472 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 }
474 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500475 }
476}
477
Ram Paia05964f2005-11-07 17:20:17 -0500478void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -0500479{
480 struct vfsmount *p;
481
482 for (p = mnt; p; p = next_mnt(p, mnt)) {
483 list_del(&p->mnt_hash);
484 list_add(&p->mnt_hash, kill);
485 }
486
Ram Paia05964f2005-11-07 17:20:17 -0500487 if (propagate)
488 propagate_umount(kill);
489
Ram Pai70fbcdf2005-11-07 17:17:04 -0500490 list_for_each_entry(p, kill, mnt_hash) {
491 list_del_init(&p->mnt_expire);
492 list_del_init(&p->mnt_list);
493 __touch_namespace(p->mnt_namespace);
494 p->mnt_namespace = NULL;
495 list_del_init(&p->mnt_child);
496 if (p->mnt_parent != p)
Al Virof30ac312006-02-01 07:53:21 -0500497 p->mnt_mountpoint->d_mounted--;
Ram Paia05964f2005-11-07 17:20:17 -0500498 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 }
500}
501
502static int do_umount(struct vfsmount *mnt, int flags)
503{
Ram Paib58fed82005-11-07 17:16:09 -0500504 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -0500506 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508 retval = security_sb_umount(mnt, flags);
509 if (retval)
510 return retval;
511
512 /*
513 * Allow userspace to request a mountpoint be expired rather than
514 * unmounting unconditionally. Unmount only happens if:
515 * (1) the mark is already set (the mark is cleared by mntput())
516 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
517 */
518 if (flags & MNT_EXPIRE) {
519 if (mnt == current->fs->rootmnt ||
520 flags & (MNT_FORCE | MNT_DETACH))
521 return -EINVAL;
522
523 if (atomic_read(&mnt->mnt_count) != 2)
524 return -EBUSY;
525
526 if (!xchg(&mnt->mnt_expiry_mark, 1))
527 return -EAGAIN;
528 }
529
530 /*
531 * If we may have to abort operations to get out of this
532 * mount, and they will themselves hold resources we must
533 * allow the fs to do things. In the Unix tradition of
534 * 'Gee thats tricky lets do it in userspace' the umount_begin
535 * might fail to complete on the first run through as other tasks
536 * must return, and the like. Thats for the mount program to worry
537 * about for the moment.
538 */
539
540 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -0500541 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 sb->s_op->umount_begin(sb);
543 unlock_kernel();
544
545 /*
546 * No sense to grab the lock for this test, but test itself looks
547 * somewhat bogus. Suggestions for better replacement?
548 * Ho-hum... In principle, we might treat that as umount + switch
549 * to rootfs. GC would eventually take care of the old vfsmount.
550 * Actually it makes sense, especially if rootfs would contain a
551 * /reboot - static binary that would close all descriptors and
552 * call reboot(9). Then init(8) could umount root and exec /reboot.
553 */
554 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
555 /*
556 * Special case for "unmounting" root ...
557 * we just try to remount it readonly.
558 */
559 down_write(&sb->s_umount);
560 if (!(sb->s_flags & MS_RDONLY)) {
561 lock_kernel();
562 DQUOT_OFF(sb);
563 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
564 unlock_kernel();
565 }
566 up_write(&sb->s_umount);
567 return retval;
568 }
569
Ram Pai390c6842005-11-07 17:17:51 -0500570 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -0500572 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -0500575 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -0500577 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 retval = 0;
579 }
580 spin_unlock(&vfsmount_lock);
581 if (retval)
582 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -0500583 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500584 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 return retval;
586}
587
588/*
589 * Now umount can handle mount points as well as block devices.
590 * This is important for filesystems which use unnamed block devices.
591 *
592 * We now support a flag for forced unmount like the other 'big iron'
593 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
594 */
595
596asmlinkage long sys_umount(char __user * name, int flags)
597{
598 struct nameidata nd;
599 int retval;
600
601 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
602 if (retval)
603 goto out;
604 retval = -EINVAL;
605 if (nd.dentry != nd.mnt->mnt_root)
606 goto dput_and_out;
607 if (!check_mnt(nd.mnt))
608 goto dput_and_out;
609
610 retval = -EPERM;
611 if (!capable(CAP_SYS_ADMIN))
612 goto dput_and_out;
613
614 retval = do_umount(nd.mnt, flags);
615dput_and_out:
616 path_release_on_umount(&nd);
617out:
618 return retval;
619}
620
621#ifdef __ARCH_WANT_SYS_OLDUMOUNT
622
623/*
Ram Paib58fed82005-11-07 17:16:09 -0500624 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626asmlinkage long sys_oldumount(char __user * name)
627{
Ram Paib58fed82005-11-07 17:16:09 -0500628 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629}
630
631#endif
632
633static int mount_is_safe(struct nameidata *nd)
634{
635 if (capable(CAP_SYS_ADMIN))
636 return 0;
637 return -EPERM;
638#ifdef notyet
639 if (S_ISLNK(nd->dentry->d_inode->i_mode))
640 return -EPERM;
641 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
642 if (current->uid != nd->dentry->d_inode->i_uid)
643 return -EPERM;
644 }
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800645 if (vfs_permission(nd, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 return -EPERM;
647 return 0;
648#endif
649}
650
Ram Paib58fed82005-11-07 17:16:09 -0500651static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
653 while (1) {
654 if (d == dentry)
655 return 1;
656 if (d == NULL || d == d->d_parent)
657 return 0;
658 d = d->d_parent;
659 }
660}
661
Ram Paib90fa9a2005-11-07 17:19:50 -0500662struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -0500663 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
665 struct vfsmount *res, *p, *q, *r, *s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 struct nameidata nd;
667
Ram Pai9676f0c2005-11-07 17:21:20 -0500668 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
669 return NULL;
670
Ram Pai36341f62005-11-07 17:17:22 -0500671 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 if (!q)
673 goto Enomem;
674 q->mnt_mountpoint = mnt->mnt_mountpoint;
675
676 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -0700677 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
679 continue;
680
681 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -0500682 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
683 s = skip_mnt_tree(s);
684 continue;
685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 while (p != s->mnt_parent) {
687 p = p->mnt_parent;
688 q = q->mnt_parent;
689 }
690 p = s;
691 nd.mnt = q;
692 nd.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -0500693 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 if (!q)
695 goto Enomem;
696 spin_lock(&vfsmount_lock);
697 list_add_tail(&q->mnt_list, &res->mnt_list);
698 attach_mnt(q, &nd);
699 spin_unlock(&vfsmount_lock);
700 }
701 }
702 return res;
Ram Paib58fed82005-11-07 17:16:09 -0500703Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500705 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500707 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500709 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 }
711 return NULL;
712}
713
Ram Paib90fa9a2005-11-07 17:19:50 -0500714/*
715 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -0500716 * @nd : place the mount tree @source_mnt is attached
717 * @parent_nd : if non-null, detach the source_mnt from its parent and
718 * store the parent mount and mountpoint dentry.
719 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -0500720 *
721 * NOTE: in the table below explains the semantics when a source mount
722 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -0500723 * ---------------------------------------------------------------------------
724 * | BIND MOUNT OPERATION |
725 * |**************************************************************************
726 * | source-->| shared | private | slave | unbindable |
727 * | dest | | | | |
728 * | | | | | | |
729 * | v | | | | |
730 * |**************************************************************************
731 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
732 * | | | | | |
733 * |non-shared| shared (+) | private | slave (*) | invalid |
734 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -0500735 * A bind operation clones the source mount and mounts the clone on the
736 * destination mount.
737 *
738 * (++) the cloned mount is propagated to all the mounts in the propagation
739 * tree of the destination mount and the cloned mount is added to
740 * the peer group of the source mount.
741 * (+) the cloned mount is created under the destination mount and is marked
742 * as shared. The cloned mount is added to the peer group of the source
743 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -0500744 * (+++) the mount is propagated to all the mounts in the propagation tree
745 * of the destination mount and the cloned mount is made slave
746 * of the same master as that of the source mount. The cloned mount
747 * is marked as 'shared and slave'.
748 * (*) the cloned mount is made a slave of the same master as that of the
749 * source mount.
750 *
Ram Pai9676f0c2005-11-07 17:21:20 -0500751 * ---------------------------------------------------------------------------
752 * | MOVE MOUNT OPERATION |
753 * |**************************************************************************
754 * | source-->| shared | private | slave | unbindable |
755 * | dest | | | | |
756 * | | | | | | |
757 * | v | | | | |
758 * |**************************************************************************
759 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
760 * | | | | | |
761 * |non-shared| shared (+*) | private | slave (*) | unbindable |
762 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -0500763 *
764 * (+) the mount is moved to the destination. And is then propagated to
765 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -0500766 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -0500767 * (+++) the mount is moved to the destination and is then propagated to
768 * all the mounts belonging to the destination mount's propagation tree.
769 * the mount is marked as 'shared and slave'.
770 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -0500771 *
772 * if the source mount is a tree, the operations explained above is
773 * applied to each mount in the tree.
774 * Must be called without spinlocks held, since this function can sleep
775 * in allocations.
776 */
777static int attach_recursive_mnt(struct vfsmount *source_mnt,
Ram Pai21444402005-11-07 17:20:03 -0500778 struct nameidata *nd, struct nameidata *parent_nd)
Ram Paib90fa9a2005-11-07 17:19:50 -0500779{
780 LIST_HEAD(tree_list);
781 struct vfsmount *dest_mnt = nd->mnt;
782 struct dentry *dest_dentry = nd->dentry;
783 struct vfsmount *child, *p;
784
785 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
786 return -EINVAL;
787
788 if (IS_MNT_SHARED(dest_mnt)) {
789 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
790 set_mnt_shared(p);
791 }
792
793 spin_lock(&vfsmount_lock);
Ram Pai21444402005-11-07 17:20:03 -0500794 if (parent_nd) {
795 detach_mnt(source_mnt, parent_nd);
796 attach_mnt(source_mnt, nd);
797 touch_namespace(current->namespace);
798 } else {
799 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
800 commit_tree(source_mnt);
801 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500802
803 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
804 list_del_init(&child->mnt_hash);
805 commit_tree(child);
806 }
807 spin_unlock(&vfsmount_lock);
808 return 0;
809}
810
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
812{
813 int err;
814 if (mnt->mnt_sb->s_flags & MS_NOUSER)
815 return -EINVAL;
816
817 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
818 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
819 return -ENOTDIR;
820
821 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800822 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 if (IS_DEADDIR(nd->dentry->d_inode))
824 goto out_unlock;
825
826 err = security_sb_check_sb(mnt, nd);
827 if (err)
828 goto out_unlock;
829
830 err = -ENOENT;
Ram Paib90fa9a2005-11-07 17:19:50 -0500831 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500832 err = attach_recursive_mnt(mnt, nd, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800834 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 if (!err)
836 security_sb_post_addmount(mnt, nd);
837 return err;
838}
839
840/*
Ram Pai07b20882005-11-07 17:19:07 -0500841 * recursively change the type of the mountpoint.
842 */
843static int do_change_type(struct nameidata *nd, int flag)
844{
845 struct vfsmount *m, *mnt = nd->mnt;
846 int recurse = flag & MS_REC;
847 int type = flag & ~MS_REC;
848
849 if (nd->dentry != nd->mnt->mnt_root)
850 return -EINVAL;
851
852 down_write(&namespace_sem);
853 spin_lock(&vfsmount_lock);
854 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
855 change_mnt_propagation(m, type);
856 spin_unlock(&vfsmount_lock);
857 up_write(&namespace_sem);
858 return 0;
859}
860
861/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 * do loopback mount.
863 */
864static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
865{
866 struct nameidata old_nd;
867 struct vfsmount *mnt = NULL;
868 int err = mount_is_safe(nd);
869 if (err)
870 return err;
871 if (!old_name || !*old_name)
872 return -EINVAL;
873 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
874 if (err)
875 return err;
876
Ram Pai390c6842005-11-07 17:17:51 -0500877 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 err = -EINVAL;
Ram Pai9676f0c2005-11-07 17:21:20 -0500879 if (IS_MNT_UNBINDABLE(old_nd.mnt))
880 goto out;
881
Al Viroccd48bc2005-11-07 17:15:04 -0500882 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
883 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
Al Viroccd48bc2005-11-07 17:15:04 -0500885 err = -ENOMEM;
886 if (recurse)
Ram Pai36341f62005-11-07 17:17:22 -0500887 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500888 else
Ram Pai36341f62005-11-07 17:17:22 -0500889 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -0500890
891 if (!mnt)
892 goto out;
893
Al Viroccd48bc2005-11-07 17:15:04 -0500894 err = graft_tree(mnt, nd);
895 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -0500896 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500898 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500900 release_mounts(&umount_list);
Ram Pai5b83d2c2005-11-07 17:16:29 -0500901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Al Viroccd48bc2005-11-07 17:15:04 -0500903out:
Ram Pai390c6842005-11-07 17:17:51 -0500904 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 path_release(&old_nd);
906 return err;
907}
908
909/*
910 * change filesystem flags. dir should be a physical root of filesystem.
911 * If you've mounted a non-root directory somewhere and want to do remount
912 * on it - tough luck.
913 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
915 void *data)
916{
917 int err;
Ram Paib58fed82005-11-07 17:16:09 -0500918 struct super_block *sb = nd->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 if (!capable(CAP_SYS_ADMIN))
921 return -EPERM;
922
923 if (!check_mnt(nd->mnt))
924 return -EINVAL;
925
926 if (nd->dentry != nd->mnt->mnt_root)
927 return -EINVAL;
928
929 down_write(&sb->s_umount);
930 err = do_remount_sb(sb, flags, data, 0);
931 if (!err)
Ram Paib58fed82005-11-07 17:16:09 -0500932 nd->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 up_write(&sb->s_umount);
934 if (!err)
935 security_sb_post_remount(nd->mnt, flags, data);
936 return err;
937}
938
Ram Pai9676f0c2005-11-07 17:21:20 -0500939static inline int tree_contains_unbindable(struct vfsmount *mnt)
940{
941 struct vfsmount *p;
942 for (p = mnt; p; p = next_mnt(p, mnt)) {
943 if (IS_MNT_UNBINDABLE(p))
944 return 1;
945 }
946 return 0;
947}
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949static int do_move_mount(struct nameidata *nd, char *old_name)
950{
951 struct nameidata old_nd, parent_nd;
952 struct vfsmount *p;
953 int err = 0;
954 if (!capable(CAP_SYS_ADMIN))
955 return -EPERM;
956 if (!old_name || !*old_name)
957 return -EINVAL;
958 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
959 if (err)
960 return err;
961
Ram Pai390c6842005-11-07 17:17:51 -0500962 down_write(&namespace_sem);
Ram Paib58fed82005-11-07 17:16:09 -0500963 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 ;
965 err = -EINVAL;
966 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
967 goto out;
968
969 err = -ENOENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800970 mutex_lock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (IS_DEADDIR(nd->dentry->d_inode))
972 goto out1;
973
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
Ram Pai21444402005-11-07 17:20:03 -0500975 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976
977 err = -EINVAL;
978 if (old_nd.dentry != old_nd.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -0500979 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
981 if (old_nd.mnt == old_nd.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -0500982 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983
984 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
985 S_ISDIR(old_nd.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -0500986 goto out1;
987 /*
988 * Don't move a mount residing in a shared parent.
989 */
990 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
991 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -0500992 /*
993 * Don't move a mount tree containing unbindable mounts to a destination
994 * mount which is shared.
995 */
996 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
997 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 err = -ELOOP;
Ram Paib58fed82005-11-07 17:16:09 -0500999 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (p == old_nd.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001001 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002
Ram Pai21444402005-11-07 17:20:03 -05001003 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1004 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
Ram Pai21444402005-11-07 17:20:03 -05001006 spin_lock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 /* if the mount is moved, it should no longer be expire
1008 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001009 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 spin_unlock(&vfsmount_lock);
1011out1:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001012 mutex_unlock(&nd->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013out:
Ram Pai390c6842005-11-07 17:17:51 -05001014 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (!err)
1016 path_release(&parent_nd);
1017 path_release(&old_nd);
1018 return err;
1019}
1020
1021/*
1022 * create a new mount for userspace and request it to be added into the
1023 * namespace's tree
1024 */
1025static int do_new_mount(struct nameidata *nd, char *type, int flags,
1026 int mnt_flags, char *name, void *data)
1027{
1028 struct vfsmount *mnt;
1029
1030 if (!type || !memchr(type, 0, PAGE_SIZE))
1031 return -EINVAL;
1032
1033 /* we need capabilities... */
1034 if (!capable(CAP_SYS_ADMIN))
1035 return -EPERM;
1036
1037 mnt = do_kern_mount(type, flags, name, data);
1038 if (IS_ERR(mnt))
1039 return PTR_ERR(mnt);
1040
1041 return do_add_mount(mnt, nd, mnt_flags, NULL);
1042}
1043
1044/*
1045 * add a mount into a namespace's mount tree
1046 * - provide the option of adding the new mount to an expiration list
1047 */
1048int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1049 int mnt_flags, struct list_head *fslist)
1050{
1051 int err;
1052
Ram Pai390c6842005-11-07 17:17:51 -05001053 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 /* Something was mounted here while we slept */
Ram Paib58fed82005-11-07 17:16:09 -05001055 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 ;
1057 err = -EINVAL;
1058 if (!check_mnt(nd->mnt))
1059 goto unlock;
1060
1061 /* Refuse the same filesystem on the same mount point */
1062 err = -EBUSY;
1063 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1064 nd->mnt->mnt_root == nd->dentry)
1065 goto unlock;
1066
1067 err = -EINVAL;
1068 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1069 goto unlock;
1070
1071 newmnt->mnt_flags = mnt_flags;
Ram Pai5b83d2c2005-11-07 17:16:29 -05001072 if ((err = graft_tree(newmnt, nd)))
1073 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074
Ram Pai5b83d2c2005-11-07 17:16:29 -05001075 if (fslist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 /* add to the specified expiration list */
1077 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001078 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 spin_unlock(&vfsmount_lock);
1080 }
Ram Pai390c6842005-11-07 17:17:51 -05001081 up_write(&namespace_sem);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001082 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
1084unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001085 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 mntput(newmnt);
1087 return err;
1088}
1089
1090EXPORT_SYMBOL_GPL(do_add_mount);
1091
Ram Pai70fbcdf2005-11-07 17:17:04 -05001092static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1093 struct list_head *umounts)
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001094{
1095 spin_lock(&vfsmount_lock);
1096
1097 /*
Miklos Szeredied42c872005-07-07 17:57:26 -07001098 * Check if mount is still attached, if not, let whoever holds it deal
1099 * with the sucker
1100 */
1101 if (mnt->mnt_parent == mnt) {
1102 spin_unlock(&vfsmount_lock);
1103 return;
1104 }
1105
1106 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001107 * Check that it is still dead: the count should now be 2 - as
1108 * contributed by the vfsmount parent and the mntget above
1109 */
Ram Paia05964f2005-11-07 17:20:17 -05001110 if (!propagate_mount_busy(mnt, 2)) {
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001111 /* delete from the namespace */
Al Viro5addc5d2005-11-07 17:15:49 -05001112 touch_namespace(mnt->mnt_namespace);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001113 list_del_init(&mnt->mnt_list);
Miklos Szerediac081152005-07-07 17:57:27 -07001114 mnt->mnt_namespace = NULL;
Ram Paia05964f2005-11-07 17:20:17 -05001115 umount_tree(mnt, 1, umounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001116 spin_unlock(&vfsmount_lock);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001117 } else {
1118 /*
1119 * Someone brought it back to life whilst we didn't have any
1120 * locks held so return it to the expiration list
1121 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001122 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -07001123 spin_unlock(&vfsmount_lock);
1124 }
1125}
1126
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127/*
1128 * process a list of expirable mountpoints with the intent of discarding any
1129 * mountpoints that aren't in use and haven't been touched since last we came
1130 * here
1131 */
1132void mark_mounts_for_expiry(struct list_head *mounts)
1133{
1134 struct namespace *namespace;
1135 struct vfsmount *mnt, *next;
1136 LIST_HEAD(graveyard);
1137
1138 if (list_empty(mounts))
1139 return;
1140
1141 spin_lock(&vfsmount_lock);
1142
1143 /* extract from the expiration list every vfsmount that matches the
1144 * following criteria:
1145 * - only referenced by its parent vfsmount
1146 * - still marked for expiry (marked on the last call here; marks are
1147 * cleared by mntput())
1148 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001149 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1151 atomic_read(&mnt->mnt_count) != 1)
1152 continue;
1153
1154 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001155 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
1157
1158 /*
1159 * go through the vfsmounts we've just consigned to the graveyard to
1160 * - check that they're still dead
1161 * - delete the vfsmount from the appropriate namespace under lock
1162 * - dispose of the corpse
1163 */
1164 while (!list_empty(&graveyard)) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001165 LIST_HEAD(umounts);
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001166 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1167 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169 /* don't do anything if the namespace is dead - all the
1170 * vfsmounts from it are going away anyway */
1171 namespace = mnt->mnt_namespace;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001172 if (!namespace || !namespace->root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 continue;
1174 get_namespace(namespace);
1175
1176 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001177 down_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001178 expire_mount(mnt, mounts, &umounts);
Ram Pai390c6842005-11-07 17:17:51 -05001179 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001180 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 mntput(mnt);
1182 put_namespace(namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 spin_lock(&vfsmount_lock);
1184 }
1185
1186 spin_unlock(&vfsmount_lock);
1187}
1188
1189EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1190
1191/*
1192 * Some copy_from_user() implementations do not return the exact number of
1193 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1194 * Note that this function differs from copy_from_user() in that it will oops
1195 * on bad values of `to', rather than returning a short copy.
1196 */
Ram Paib58fed82005-11-07 17:16:09 -05001197static long exact_copy_from_user(void *to, const void __user * from,
1198 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199{
1200 char *t = to;
1201 const char __user *f = from;
1202 char c;
1203
1204 if (!access_ok(VERIFY_READ, from, n))
1205 return n;
1206
1207 while (n) {
1208 if (__get_user(c, f)) {
1209 memset(t, 0, n);
1210 break;
1211 }
1212 *t++ = c;
1213 f++;
1214 n--;
1215 }
1216 return n;
1217}
1218
Ram Paib58fed82005-11-07 17:16:09 -05001219int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 int i;
1222 unsigned long page;
1223 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001224
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 *where = 0;
1226 if (!data)
1227 return 0;
1228
1229 if (!(page = __get_free_page(GFP_KERNEL)))
1230 return -ENOMEM;
1231
1232 /* We only care that *some* data at the address the user
1233 * gave us is valid. Just in case, we'll zero
1234 * the remainder of the page.
1235 */
1236 /* copy_from_user cannot cross TASK_SIZE ! */
1237 size = TASK_SIZE - (unsigned long)data;
1238 if (size > PAGE_SIZE)
1239 size = PAGE_SIZE;
1240
1241 i = size - exact_copy_from_user((void *)page, data, size);
1242 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001243 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 return -EFAULT;
1245 }
1246 if (i != PAGE_SIZE)
1247 memset((char *)page + i, 0, PAGE_SIZE - i);
1248 *where = page;
1249 return 0;
1250}
1251
1252/*
1253 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1254 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1255 *
1256 * data is a (void *) that can point to any structure up to
1257 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1258 * information (or be NULL).
1259 *
1260 * Pre-0.97 versions of mount() didn't have a flags word.
1261 * When the flags word was introduced its top half was required
1262 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1263 * Therefore, if this magic number is present, it carries no information
1264 * and must be discarded.
1265 */
Ram Paib58fed82005-11-07 17:16:09 -05001266long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 unsigned long flags, void *data_page)
1268{
1269 struct nameidata nd;
1270 int retval = 0;
1271 int mnt_flags = 0;
1272
1273 /* Discard magic */
1274 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1275 flags &= ~MS_MGC_MSK;
1276
1277 /* Basic sanity checks */
1278
1279 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1280 return -EINVAL;
1281 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1282 return -EINVAL;
1283
1284 if (data_page)
1285 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1286
1287 /* Separate the per-mountpoint flags */
1288 if (flags & MS_NOSUID)
1289 mnt_flags |= MNT_NOSUID;
1290 if (flags & MS_NODEV)
1291 mnt_flags |= MNT_NODEV;
1292 if (flags & MS_NOEXEC)
1293 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001294 if (flags & MS_NOATIME)
1295 mnt_flags |= MNT_NOATIME;
1296 if (flags & MS_NODIRATIME)
1297 mnt_flags |= MNT_NODIRATIME;
1298
1299 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1300 MS_NOATIME | MS_NODIRATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
1302 /* ... and get the mountpoint */
1303 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1304 if (retval)
1305 return retval;
1306
1307 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1308 if (retval)
1309 goto dput_out;
1310
1311 if (flags & MS_REMOUNT)
1312 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1313 data_page);
1314 else if (flags & MS_BIND)
1315 retval = do_loopback(&nd, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001316 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Ram Pai07b20882005-11-07 17:19:07 -05001317 retval = do_change_type(&nd, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 else if (flags & MS_MOVE)
1319 retval = do_move_mount(&nd, dev_name);
1320 else
1321 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1322 dev_name, data_page);
1323dput_out:
1324 path_release(&nd);
1325 return retval;
1326}
1327
JANAK DESAI741a2952006-02-07 12:59:00 -08001328/*
1329 * Allocate a new namespace structure and populate it with contents
1330 * copied from the namespace of the passed in task structure.
1331 */
1332struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
1334 struct namespace *namespace = tsk->namespace;
1335 struct namespace *new_ns;
1336 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 struct vfsmount *p, *q;
1338
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1340 if (!new_ns)
1341 goto out;
1342
1343 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001345 init_waitqueue_head(&new_ns->poll);
1346 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
Ram Pai390c6842005-11-07 17:17:51 -05001348 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 /* First pass: copy the tree topology */
Ram Pai36341f62005-11-07 17:17:22 -05001350 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001351 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001353 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 kfree(new_ns);
1355 goto out;
1356 }
1357 spin_lock(&vfsmount_lock);
1358 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1359 spin_unlock(&vfsmount_lock);
1360
1361 /*
1362 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1363 * as belonging to new namespace. We have already acquired a private
1364 * fs_struct, so tsk->fs->lock is not needed.
1365 */
1366 p = namespace->root;
1367 q = new_ns->root;
1368 while (p) {
1369 q->mnt_namespace = new_ns;
1370 if (fs) {
1371 if (p == fs->rootmnt) {
1372 rootmnt = p;
1373 fs->rootmnt = mntget(q);
1374 }
1375 if (p == fs->pwdmnt) {
1376 pwdmnt = p;
1377 fs->pwdmnt = mntget(q);
1378 }
1379 if (p == fs->altrootmnt) {
1380 altrootmnt = p;
1381 fs->altrootmnt = mntget(q);
1382 }
1383 }
1384 p = next_mnt(p, namespace->root);
1385 q = next_mnt(q, new_ns->root);
1386 }
Ram Pai390c6842005-11-07 17:17:51 -05001387 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (rootmnt)
1390 mntput(rootmnt);
1391 if (pwdmnt)
1392 mntput(pwdmnt);
1393 if (altrootmnt)
1394 mntput(altrootmnt);
1395
JANAK DESAI741a2952006-02-07 12:59:00 -08001396out:
1397 return new_ns;
1398}
1399
1400int copy_namespace(int flags, struct task_struct *tsk)
1401{
1402 struct namespace *namespace = tsk->namespace;
1403 struct namespace *new_ns;
1404 int err = 0;
1405
1406 if (!namespace)
1407 return 0;
1408
1409 get_namespace(namespace);
1410
1411 if (!(flags & CLONE_NEWNS))
1412 return 0;
1413
1414 if (!capable(CAP_SYS_ADMIN)) {
1415 err = -EPERM;
1416 goto out;
1417 }
1418
1419 new_ns = dup_namespace(tsk, tsk->fs);
1420 if (!new_ns) {
1421 err = -ENOMEM;
1422 goto out;
1423 }
1424
1425 tsk->namespace = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
1427out:
1428 put_namespace(namespace);
JANAK DESAI741a2952006-02-07 12:59:00 -08001429 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430}
1431
1432asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1433 char __user * type, unsigned long flags,
1434 void __user * data)
1435{
1436 int retval;
1437 unsigned long data_page;
1438 unsigned long type_page;
1439 unsigned long dev_page;
1440 char *dir_page;
1441
Ram Paib58fed82005-11-07 17:16:09 -05001442 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (retval < 0)
1444 return retval;
1445
1446 dir_page = getname(dir_name);
1447 retval = PTR_ERR(dir_page);
1448 if (IS_ERR(dir_page))
1449 goto out1;
1450
Ram Paib58fed82005-11-07 17:16:09 -05001451 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 if (retval < 0)
1453 goto out2;
1454
Ram Paib58fed82005-11-07 17:16:09 -05001455 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 if (retval < 0)
1457 goto out3;
1458
1459 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05001460 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1461 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 unlock_kernel();
1463 free_page(data_page);
1464
1465out3:
1466 free_page(dev_page);
1467out2:
1468 putname(dir_page);
1469out1:
1470 free_page(type_page);
1471 return retval;
1472}
1473
1474/*
1475 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1476 * It can block. Requires the big lock held.
1477 */
1478void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1479 struct dentry *dentry)
1480{
1481 struct dentry *old_root;
1482 struct vfsmount *old_rootmnt;
1483 write_lock(&fs->lock);
1484 old_root = fs->root;
1485 old_rootmnt = fs->rootmnt;
1486 fs->rootmnt = mntget(mnt);
1487 fs->root = dget(dentry);
1488 write_unlock(&fs->lock);
1489 if (old_root) {
1490 dput(old_root);
1491 mntput(old_rootmnt);
1492 }
1493}
1494
1495/*
1496 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1497 * It can block. Requires the big lock held.
1498 */
1499void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1500 struct dentry *dentry)
1501{
1502 struct dentry *old_pwd;
1503 struct vfsmount *old_pwdmnt;
1504
1505 write_lock(&fs->lock);
1506 old_pwd = fs->pwd;
1507 old_pwdmnt = fs->pwdmnt;
1508 fs->pwdmnt = mntget(mnt);
1509 fs->pwd = dget(dentry);
1510 write_unlock(&fs->lock);
1511
1512 if (old_pwd) {
1513 dput(old_pwd);
1514 mntput(old_pwdmnt);
1515 }
1516}
1517
1518static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1519{
1520 struct task_struct *g, *p;
1521 struct fs_struct *fs;
1522
1523 read_lock(&tasklist_lock);
1524 do_each_thread(g, p) {
1525 task_lock(p);
1526 fs = p->fs;
1527 if (fs) {
1528 atomic_inc(&fs->count);
1529 task_unlock(p);
Ram Paib58fed82005-11-07 17:16:09 -05001530 if (fs->root == old_nd->dentry
1531 && fs->rootmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
Ram Paib58fed82005-11-07 17:16:09 -05001533 if (fs->pwd == old_nd->dentry
1534 && fs->pwdmnt == old_nd->mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1536 put_fs_struct(fs);
1537 } else
1538 task_unlock(p);
1539 } while_each_thread(g, p);
1540 read_unlock(&tasklist_lock);
1541}
1542
1543/*
1544 * pivot_root Semantics:
1545 * Moves the root file system of the current process to the directory put_old,
1546 * makes new_root as the new root file system of the current process, and sets
1547 * root/cwd of all processes which had them on the current root to new_root.
1548 *
1549 * Restrictions:
1550 * The new_root and put_old must be directories, and must not be on the
1551 * same file system as the current process root. The put_old must be
1552 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1553 * pointed to by put_old must yield the same directory as new_root. No other
1554 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1555 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08001556 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1557 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1558 * in this situation.
1559 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 * Notes:
1561 * - we don't move root/cwd if they are not at the root (reason: if something
1562 * cared enough to change them, it's probably wrong to force them elsewhere)
1563 * - it's okay to pick a root that isn't the root of a file system, e.g.
1564 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1565 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1566 * first.
1567 */
Ram Paib58fed82005-11-07 17:16:09 -05001568asmlinkage long sys_pivot_root(const char __user * new_root,
1569 const char __user * put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
1571 struct vfsmount *tmp;
1572 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1573 int error;
1574
1575 if (!capable(CAP_SYS_ADMIN))
1576 return -EPERM;
1577
1578 lock_kernel();
1579
Ram Paib58fed82005-11-07 17:16:09 -05001580 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1581 &new_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 if (error)
1583 goto out0;
1584 error = -EINVAL;
1585 if (!check_mnt(new_nd.mnt))
1586 goto out1;
1587
Ram Paib58fed82005-11-07 17:16:09 -05001588 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 if (error)
1590 goto out1;
1591
1592 error = security_sb_pivotroot(&old_nd, &new_nd);
1593 if (error) {
1594 path_release(&old_nd);
1595 goto out1;
1596 }
1597
1598 read_lock(&current->fs->lock);
1599 user_nd.mnt = mntget(current->fs->rootmnt);
1600 user_nd.dentry = dget(current->fs->root);
1601 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05001602 down_write(&namespace_sem);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001603 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 error = -EINVAL;
Ram Pai21444402005-11-07 17:20:03 -05001605 if (IS_MNT_SHARED(old_nd.mnt) ||
1606 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1607 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1608 goto out2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 if (!check_mnt(user_nd.mnt))
1610 goto out2;
1611 error = -ENOENT;
1612 if (IS_DEADDIR(new_nd.dentry->d_inode))
1613 goto out2;
1614 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1615 goto out2;
1616 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1617 goto out2;
1618 error = -EBUSY;
1619 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1620 goto out2; /* loop, on the same file system */
1621 error = -EINVAL;
1622 if (user_nd.mnt->mnt_root != user_nd.dentry)
1623 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001624 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1625 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 if (new_nd.mnt->mnt_root != new_nd.dentry)
1627 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001628 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1629 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1631 spin_lock(&vfsmount_lock);
1632 if (tmp != new_nd.mnt) {
1633 for (;;) {
1634 if (tmp->mnt_parent == tmp)
1635 goto out3; /* already mounted on put_old */
1636 if (tmp->mnt_parent == new_nd.mnt)
1637 break;
1638 tmp = tmp->mnt_parent;
1639 }
1640 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1641 goto out3;
1642 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1643 goto out3;
1644 detach_mnt(new_nd.mnt, &parent_nd);
1645 detach_mnt(user_nd.mnt, &root_parent);
1646 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1647 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
Al Viro5addc5d2005-11-07 17:15:49 -05001648 touch_namespace(current->namespace);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 spin_unlock(&vfsmount_lock);
1650 chroot_fs_refs(&user_nd, &new_nd);
1651 security_sb_post_pivotroot(&user_nd, &new_nd);
1652 error = 0;
1653 path_release(&root_parent);
1654 path_release(&parent_nd);
1655out2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001656 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05001657 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 path_release(&user_nd);
1659 path_release(&old_nd);
1660out1:
1661 path_release(&new_nd);
1662out0:
1663 unlock_kernel();
1664 return error;
1665out3:
1666 spin_unlock(&vfsmount_lock);
1667 goto out2;
1668}
1669
1670static void __init init_mount_tree(void)
1671{
1672 struct vfsmount *mnt;
1673 struct namespace *namespace;
1674 struct task_struct *g, *p;
1675
1676 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1677 if (IS_ERR(mnt))
1678 panic("Can't create rootfs");
1679 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1680 if (!namespace)
1681 panic("Can't allocate initial namespace");
1682 atomic_set(&namespace->count, 1);
1683 INIT_LIST_HEAD(&namespace->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001684 init_waitqueue_head(&namespace->poll);
1685 namespace->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 list_add(&mnt->mnt_list, &namespace->list);
1687 namespace->root = mnt;
1688 mnt->mnt_namespace = namespace;
1689
1690 init_task.namespace = namespace;
1691 read_lock(&tasklist_lock);
1692 do_each_thread(g, p) {
1693 get_namespace(namespace);
1694 p->namespace = namespace;
1695 } while_each_thread(g, p);
1696 read_unlock(&tasklist_lock);
1697
1698 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1699 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1700}
1701
1702void __init mnt_init(unsigned long mempages)
1703{
1704 struct list_head *d;
1705 unsigned int nr_hash;
1706 int i;
1707
Ram Pai390c6842005-11-07 17:17:51 -05001708 init_rwsem(&namespace_sem);
1709
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Ram Paib58fed82005-11-07 17:16:09 -05001711 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
Ram Paib58fed82005-11-07 17:16:09 -05001713 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
1715 if (!mount_hashtable)
1716 panic("Failed to allocate mount hash table\n");
1717
1718 /*
1719 * Find the power-of-two list-heads that can fit into the allocation..
1720 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1721 * a power-of-two.
1722 */
1723 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1724 hash_bits = 0;
1725 do {
1726 hash_bits++;
1727 } while ((nr_hash >> hash_bits) != 0);
1728 hash_bits--;
1729
1730 /*
1731 * Re-calculate the actual number of entries and the mask
1732 * from the number of bits we can fit.
1733 */
1734 nr_hash = 1UL << hash_bits;
Ram Paib58fed82005-11-07 17:16:09 -05001735 hash_mask = nr_hash - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
1737 printk("Mount-cache hash table entries: %d\n", nr_hash);
1738
1739 /* And initialize the newly allocated array */
1740 d = mount_hashtable;
1741 i = nr_hash;
1742 do {
1743 INIT_LIST_HEAD(d);
1744 d++;
1745 i--;
1746 } while (i);
1747 sysfs_init();
Miklos Szeredif87fd4c2006-01-16 22:14:23 -08001748 subsystem_register(&fs_subsys);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 init_rootfs();
1750 init_mount_tree();
1751}
1752
1753void __put_namespace(struct namespace *namespace)
1754{
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001755 struct vfsmount *root = namespace->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001756 LIST_HEAD(umount_list);
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001757 namespace->root = NULL;
1758 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001759 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001761 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001763 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001764 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 kfree(namespace);
1766}