xref: /openbmc/linux/fs/namespace.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /*
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>
19 #include <linux/module.h>
20 #include <linux/seq_file.h>
21 #include <linux/namespace.h>
22 #include <linux/namei.h>
23 #include <linux/security.h>
24 #include <linux/mount.h>
25 #include <asm/uaccess.h>
26 #include <asm/unistd.h>
27 
28 extern int __init init_rootfs(void);
29 
30 #ifdef CONFIG_SYSFS
31 extern int __init sysfs_init(void);
32 #else
33 static inline int sysfs_init(void)
34 {
35 	return 0;
36 }
37 #endif
38 
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40  __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
41 
42 static struct list_head *mount_hashtable;
43 static int hash_mask __read_mostly, hash_bits __read_mostly;
44 static kmem_cache_t *mnt_cache;
45 
46 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
47 {
48 	unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
49 	tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
50 	tmp = tmp + (tmp >> hash_bits);
51 	return tmp & hash_mask;
52 }
53 
54 struct vfsmount *alloc_vfsmnt(const char *name)
55 {
56 	struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
57 	if (mnt) {
58 		memset(mnt, 0, sizeof(struct vfsmount));
59 		atomic_set(&mnt->mnt_count,1);
60 		INIT_LIST_HEAD(&mnt->mnt_hash);
61 		INIT_LIST_HEAD(&mnt->mnt_child);
62 		INIT_LIST_HEAD(&mnt->mnt_mounts);
63 		INIT_LIST_HEAD(&mnt->mnt_list);
64 		INIT_LIST_HEAD(&mnt->mnt_expire);
65 		if (name) {
66 			int size = strlen(name)+1;
67 			char *newname = kmalloc(size, GFP_KERNEL);
68 			if (newname) {
69 				memcpy(newname, name, size);
70 				mnt->mnt_devname = newname;
71 			}
72 		}
73 	}
74 	return mnt;
75 }
76 
77 void free_vfsmnt(struct vfsmount *mnt)
78 {
79 	kfree(mnt->mnt_devname);
80 	kmem_cache_free(mnt_cache, mnt);
81 }
82 
83 /*
84  * Now, lookup_mnt increments the ref count before returning
85  * the vfsmount struct.
86  */
87 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
88 {
89 	struct list_head * head = mount_hashtable + hash(mnt, dentry);
90 	struct list_head * tmp = head;
91 	struct vfsmount *p, *found = NULL;
92 
93 	spin_lock(&vfsmount_lock);
94 	for (;;) {
95 		tmp = tmp->next;
96 		p = NULL;
97 		if (tmp == head)
98 			break;
99 		p = list_entry(tmp, struct vfsmount, mnt_hash);
100 		if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
101 			found = mntget(p);
102 			break;
103 		}
104 	}
105 	spin_unlock(&vfsmount_lock);
106 	return found;
107 }
108 
109 static inline int check_mnt(struct vfsmount *mnt)
110 {
111 	return mnt->mnt_namespace == current->namespace;
112 }
113 
114 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
115 {
116 	old_nd->dentry = mnt->mnt_mountpoint;
117 	old_nd->mnt = mnt->mnt_parent;
118 	mnt->mnt_parent = mnt;
119 	mnt->mnt_mountpoint = mnt->mnt_root;
120 	list_del_init(&mnt->mnt_child);
121 	list_del_init(&mnt->mnt_hash);
122 	old_nd->dentry->d_mounted--;
123 }
124 
125 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
126 {
127 	mnt->mnt_parent = mntget(nd->mnt);
128 	mnt->mnt_mountpoint = dget(nd->dentry);
129 	list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
130 	list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
131 	nd->dentry->d_mounted++;
132 }
133 
134 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
135 {
136 	struct list_head *next = p->mnt_mounts.next;
137 	if (next == &p->mnt_mounts) {
138 		while (1) {
139 			if (p == root)
140 				return NULL;
141 			next = p->mnt_child.next;
142 			if (next != &p->mnt_parent->mnt_mounts)
143 				break;
144 			p = p->mnt_parent;
145 		}
146 	}
147 	return list_entry(next, struct vfsmount, mnt_child);
148 }
149 
150 static struct vfsmount *
151 clone_mnt(struct vfsmount *old, struct dentry *root)
152 {
153 	struct super_block *sb = old->mnt_sb;
154 	struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
155 
156 	if (mnt) {
157 		mnt->mnt_flags = old->mnt_flags;
158 		atomic_inc(&sb->s_active);
159 		mnt->mnt_sb = sb;
160 		mnt->mnt_root = dget(root);
161 		mnt->mnt_mountpoint = mnt->mnt_root;
162 		mnt->mnt_parent = mnt;
163 		mnt->mnt_namespace = current->namespace;
164 
165 		/* stick the duplicate mount on the same expiry list
166 		 * as the original if that was on one */
167 		spin_lock(&vfsmount_lock);
168 		if (!list_empty(&old->mnt_expire))
169 			list_add(&mnt->mnt_expire, &old->mnt_expire);
170 		spin_unlock(&vfsmount_lock);
171 	}
172 	return mnt;
173 }
174 
175 void __mntput(struct vfsmount *mnt)
176 {
177 	struct super_block *sb = mnt->mnt_sb;
178 	dput(mnt->mnt_root);
179 	free_vfsmnt(mnt);
180 	deactivate_super(sb);
181 }
182 
183 EXPORT_SYMBOL(__mntput);
184 
185 /* iterator */
186 static void *m_start(struct seq_file *m, loff_t *pos)
187 {
188 	struct namespace *n = m->private;
189 	struct list_head *p;
190 	loff_t l = *pos;
191 
192 	down_read(&n->sem);
193 	list_for_each(p, &n->list)
194 		if (!l--)
195 			return list_entry(p, struct vfsmount, mnt_list);
196 	return NULL;
197 }
198 
199 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
200 {
201 	struct namespace *n = m->private;
202 	struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
203 	(*pos)++;
204 	return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
205 }
206 
207 static void m_stop(struct seq_file *m, void *v)
208 {
209 	struct namespace *n = m->private;
210 	up_read(&n->sem);
211 }
212 
213 static inline void mangle(struct seq_file *m, const char *s)
214 {
215 	seq_escape(m, s, " \t\n\\");
216 }
217 
218 static int show_vfsmnt(struct seq_file *m, void *v)
219 {
220 	struct vfsmount *mnt = v;
221 	int err = 0;
222 	static struct proc_fs_info {
223 		int flag;
224 		char *str;
225 	} fs_info[] = {
226 		{ MS_SYNCHRONOUS, ",sync" },
227 		{ MS_DIRSYNC, ",dirsync" },
228 		{ MS_MANDLOCK, ",mand" },
229 		{ MS_NOATIME, ",noatime" },
230 		{ MS_NODIRATIME, ",nodiratime" },
231 		{ 0, NULL }
232 	};
233 	static struct proc_fs_info mnt_info[] = {
234 		{ MNT_NOSUID, ",nosuid" },
235 		{ MNT_NODEV, ",nodev" },
236 		{ MNT_NOEXEC, ",noexec" },
237 		{ 0, NULL }
238 	};
239 	struct proc_fs_info *fs_infop;
240 
241 	mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
242 	seq_putc(m, ' ');
243 	seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
244 	seq_putc(m, ' ');
245 	mangle(m, mnt->mnt_sb->s_type->name);
246 	seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
247 	for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
248 		if (mnt->mnt_sb->s_flags & fs_infop->flag)
249 			seq_puts(m, fs_infop->str);
250 	}
251 	for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
252 		if (mnt->mnt_flags & fs_infop->flag)
253 			seq_puts(m, fs_infop->str);
254 	}
255 	if (mnt->mnt_sb->s_op->show_options)
256 		err = mnt->mnt_sb->s_op->show_options(m, mnt);
257 	seq_puts(m, " 0 0\n");
258 	return err;
259 }
260 
261 struct seq_operations mounts_op = {
262 	.start	= m_start,
263 	.next	= m_next,
264 	.stop	= m_stop,
265 	.show	= show_vfsmnt
266 };
267 
268 /**
269  * may_umount_tree - check if a mount tree is busy
270  * @mnt: root of mount tree
271  *
272  * This is called to check if a tree of mounts has any
273  * open files, pwds, chroots or sub mounts that are
274  * busy.
275  */
276 int may_umount_tree(struct vfsmount *mnt)
277 {
278 	struct list_head *next;
279 	struct vfsmount *this_parent = mnt;
280 	int actual_refs;
281 	int minimum_refs;
282 
283 	spin_lock(&vfsmount_lock);
284 	actual_refs = atomic_read(&mnt->mnt_count);
285 	minimum_refs = 2;
286 repeat:
287 	next = this_parent->mnt_mounts.next;
288 resume:
289 	while (next != &this_parent->mnt_mounts) {
290 		struct vfsmount *p = list_entry(next, struct vfsmount, mnt_child);
291 
292 		next = next->next;
293 
294 		actual_refs += atomic_read(&p->mnt_count);
295 		minimum_refs += 2;
296 
297 		if (!list_empty(&p->mnt_mounts)) {
298 			this_parent = p;
299 			goto repeat;
300 		}
301 	}
302 
303 	if (this_parent != mnt) {
304 		next = this_parent->mnt_child.next;
305 		this_parent = this_parent->mnt_parent;
306 		goto resume;
307 	}
308 	spin_unlock(&vfsmount_lock);
309 
310 	if (actual_refs > minimum_refs)
311 		return -EBUSY;
312 
313 	return 0;
314 }
315 
316 EXPORT_SYMBOL(may_umount_tree);
317 
318 /**
319  * may_umount - check if a mount point is busy
320  * @mnt: root of mount
321  *
322  * This is called to check if a mount point has any
323  * open files, pwds, chroots or sub mounts. If the
324  * mount has sub mounts this will return busy
325  * regardless of whether the sub mounts are busy.
326  *
327  * Doesn't take quota and stuff into account. IOW, in some cases it will
328  * give false negatives. The main reason why it's here is that we need
329  * a non-destructive way to look for easily umountable filesystems.
330  */
331 int may_umount(struct vfsmount *mnt)
332 {
333 	if (atomic_read(&mnt->mnt_count) > 2)
334 		return -EBUSY;
335 	return 0;
336 }
337 
338 EXPORT_SYMBOL(may_umount);
339 
340 static void umount_tree(struct vfsmount *mnt)
341 {
342 	struct vfsmount *p;
343 	LIST_HEAD(kill);
344 
345 	for (p = mnt; p; p = next_mnt(p, mnt)) {
346 		list_del(&p->mnt_list);
347 		list_add(&p->mnt_list, &kill);
348 		p->mnt_namespace = NULL;
349 	}
350 
351 	while (!list_empty(&kill)) {
352 		mnt = list_entry(kill.next, struct vfsmount, mnt_list);
353 		list_del_init(&mnt->mnt_list);
354 		list_del_init(&mnt->mnt_expire);
355 		if (mnt->mnt_parent == mnt) {
356 			spin_unlock(&vfsmount_lock);
357 		} else {
358 			struct nameidata old_nd;
359 			detach_mnt(mnt, &old_nd);
360 			spin_unlock(&vfsmount_lock);
361 			path_release(&old_nd);
362 		}
363 		mntput(mnt);
364 		spin_lock(&vfsmount_lock);
365 	}
366 }
367 
368 static int do_umount(struct vfsmount *mnt, int flags)
369 {
370 	struct super_block * sb = mnt->mnt_sb;
371 	int retval;
372 
373 	retval = security_sb_umount(mnt, flags);
374 	if (retval)
375 		return retval;
376 
377 	/*
378 	 * Allow userspace to request a mountpoint be expired rather than
379 	 * unmounting unconditionally. Unmount only happens if:
380 	 *  (1) the mark is already set (the mark is cleared by mntput())
381 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
382 	 */
383 	if (flags & MNT_EXPIRE) {
384 		if (mnt == current->fs->rootmnt ||
385 		    flags & (MNT_FORCE | MNT_DETACH))
386 			return -EINVAL;
387 
388 		if (atomic_read(&mnt->mnt_count) != 2)
389 			return -EBUSY;
390 
391 		if (!xchg(&mnt->mnt_expiry_mark, 1))
392 			return -EAGAIN;
393 	}
394 
395 	/*
396 	 * If we may have to abort operations to get out of this
397 	 * mount, and they will themselves hold resources we must
398 	 * allow the fs to do things. In the Unix tradition of
399 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
400 	 * might fail to complete on the first run through as other tasks
401 	 * must return, and the like. Thats for the mount program to worry
402 	 * about for the moment.
403 	 */
404 
405 	lock_kernel();
406 	if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
407 		sb->s_op->umount_begin(sb);
408 	unlock_kernel();
409 
410 	/*
411 	 * No sense to grab the lock for this test, but test itself looks
412 	 * somewhat bogus. Suggestions for better replacement?
413 	 * Ho-hum... In principle, we might treat that as umount + switch
414 	 * to rootfs. GC would eventually take care of the old vfsmount.
415 	 * Actually it makes sense, especially if rootfs would contain a
416 	 * /reboot - static binary that would close all descriptors and
417 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
418 	 */
419 	if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
420 		/*
421 		 * Special case for "unmounting" root ...
422 		 * we just try to remount it readonly.
423 		 */
424 		down_write(&sb->s_umount);
425 		if (!(sb->s_flags & MS_RDONLY)) {
426 			lock_kernel();
427 			DQUOT_OFF(sb);
428 			retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
429 			unlock_kernel();
430 		}
431 		up_write(&sb->s_umount);
432 		return retval;
433 	}
434 
435 	down_write(&current->namespace->sem);
436 	spin_lock(&vfsmount_lock);
437 
438 	if (atomic_read(&sb->s_active) == 1) {
439 		/* last instance - try to be smart */
440 		spin_unlock(&vfsmount_lock);
441 		lock_kernel();
442 		DQUOT_OFF(sb);
443 		acct_auto_close(sb);
444 		unlock_kernel();
445 		security_sb_umount_close(mnt);
446 		spin_lock(&vfsmount_lock);
447 	}
448 	retval = -EBUSY;
449 	if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
450 		if (!list_empty(&mnt->mnt_list))
451 			umount_tree(mnt);
452 		retval = 0;
453 	}
454 	spin_unlock(&vfsmount_lock);
455 	if (retval)
456 		security_sb_umount_busy(mnt);
457 	up_write(&current->namespace->sem);
458 	return retval;
459 }
460 
461 /*
462  * Now umount can handle mount points as well as block devices.
463  * This is important for filesystems which use unnamed block devices.
464  *
465  * We now support a flag for forced unmount like the other 'big iron'
466  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
467  */
468 
469 asmlinkage long sys_umount(char __user * name, int flags)
470 {
471 	struct nameidata nd;
472 	int retval;
473 
474 	retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
475 	if (retval)
476 		goto out;
477 	retval = -EINVAL;
478 	if (nd.dentry != nd.mnt->mnt_root)
479 		goto dput_and_out;
480 	if (!check_mnt(nd.mnt))
481 		goto dput_and_out;
482 
483 	retval = -EPERM;
484 	if (!capable(CAP_SYS_ADMIN))
485 		goto dput_and_out;
486 
487 	retval = do_umount(nd.mnt, flags);
488 dput_and_out:
489 	path_release_on_umount(&nd);
490 out:
491 	return retval;
492 }
493 
494 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
495 
496 /*
497  *	The 2.0 compatible umount. No flags.
498  */
499 
500 asmlinkage long sys_oldumount(char __user * name)
501 {
502 	return sys_umount(name,0);
503 }
504 
505 #endif
506 
507 static int mount_is_safe(struct nameidata *nd)
508 {
509 	if (capable(CAP_SYS_ADMIN))
510 		return 0;
511 	return -EPERM;
512 #ifdef notyet
513 	if (S_ISLNK(nd->dentry->d_inode->i_mode))
514 		return -EPERM;
515 	if (nd->dentry->d_inode->i_mode & S_ISVTX) {
516 		if (current->uid != nd->dentry->d_inode->i_uid)
517 			return -EPERM;
518 	}
519 	if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
520 		return -EPERM;
521 	return 0;
522 #endif
523 }
524 
525 static int
526 lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
527 {
528 	while (1) {
529 		if (d == dentry)
530 			return 1;
531 		if (d == NULL || d == d->d_parent)
532 			return 0;
533 		d = d->d_parent;
534 	}
535 }
536 
537 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
538 {
539 	struct vfsmount *res, *p, *q, *r, *s;
540 	struct nameidata nd;
541 
542 	res = q = clone_mnt(mnt, dentry);
543 	if (!q)
544 		goto Enomem;
545 	q->mnt_mountpoint = mnt->mnt_mountpoint;
546 
547 	p = mnt;
548 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
549 		if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
550 			continue;
551 
552 		for (s = r; s; s = next_mnt(s, r)) {
553 			while (p != s->mnt_parent) {
554 				p = p->mnt_parent;
555 				q = q->mnt_parent;
556 			}
557 			p = s;
558 			nd.mnt = q;
559 			nd.dentry = p->mnt_mountpoint;
560 			q = clone_mnt(p, p->mnt_root);
561 			if (!q)
562 				goto Enomem;
563 			spin_lock(&vfsmount_lock);
564 			list_add_tail(&q->mnt_list, &res->mnt_list);
565 			attach_mnt(q, &nd);
566 			spin_unlock(&vfsmount_lock);
567 		}
568 	}
569 	return res;
570  Enomem:
571 	if (res) {
572 		spin_lock(&vfsmount_lock);
573 		umount_tree(res);
574 		spin_unlock(&vfsmount_lock);
575 	}
576 	return NULL;
577 }
578 
579 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
580 {
581 	int err;
582 	if (mnt->mnt_sb->s_flags & MS_NOUSER)
583 		return -EINVAL;
584 
585 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
586 	      S_ISDIR(mnt->mnt_root->d_inode->i_mode))
587 		return -ENOTDIR;
588 
589 	err = -ENOENT;
590 	down(&nd->dentry->d_inode->i_sem);
591 	if (IS_DEADDIR(nd->dentry->d_inode))
592 		goto out_unlock;
593 
594 	err = security_sb_check_sb(mnt, nd);
595 	if (err)
596 		goto out_unlock;
597 
598 	err = -ENOENT;
599 	spin_lock(&vfsmount_lock);
600 	if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
601 		struct list_head head;
602 
603 		attach_mnt(mnt, nd);
604 		list_add_tail(&head, &mnt->mnt_list);
605 		list_splice(&head, current->namespace->list.prev);
606 		mntget(mnt);
607 		err = 0;
608 	}
609 	spin_unlock(&vfsmount_lock);
610 out_unlock:
611 	up(&nd->dentry->d_inode->i_sem);
612 	if (!err)
613 		security_sb_post_addmount(mnt, nd);
614 	return err;
615 }
616 
617 /*
618  * do loopback mount.
619  */
620 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
621 {
622 	struct nameidata old_nd;
623 	struct vfsmount *mnt = NULL;
624 	int err = mount_is_safe(nd);
625 	if (err)
626 		return err;
627 	if (!old_name || !*old_name)
628 		return -EINVAL;
629 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
630 	if (err)
631 		return err;
632 
633 	down_write(&current->namespace->sem);
634 	err = -EINVAL;
635 	if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
636 		err = -ENOMEM;
637 		if (recurse)
638 			mnt = copy_tree(old_nd.mnt, old_nd.dentry);
639 		else
640 			mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
641 	}
642 
643 	if (mnt) {
644 		/* stop bind mounts from expiring */
645 		spin_lock(&vfsmount_lock);
646 		list_del_init(&mnt->mnt_expire);
647 		spin_unlock(&vfsmount_lock);
648 
649 		err = graft_tree(mnt, nd);
650 		if (err) {
651 			spin_lock(&vfsmount_lock);
652 			umount_tree(mnt);
653 			spin_unlock(&vfsmount_lock);
654 		} else
655 			mntput(mnt);
656 	}
657 
658 	up_write(&current->namespace->sem);
659 	path_release(&old_nd);
660 	return err;
661 }
662 
663 /*
664  * change filesystem flags. dir should be a physical root of filesystem.
665  * If you've mounted a non-root directory somewhere and want to do remount
666  * on it - tough luck.
667  */
668 
669 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
670 		      void *data)
671 {
672 	int err;
673 	struct super_block * sb = nd->mnt->mnt_sb;
674 
675 	if (!capable(CAP_SYS_ADMIN))
676 		return -EPERM;
677 
678 	if (!check_mnt(nd->mnt))
679 		return -EINVAL;
680 
681 	if (nd->dentry != nd->mnt->mnt_root)
682 		return -EINVAL;
683 
684 	down_write(&sb->s_umount);
685 	err = do_remount_sb(sb, flags, data, 0);
686 	if (!err)
687 		nd->mnt->mnt_flags=mnt_flags;
688 	up_write(&sb->s_umount);
689 	if (!err)
690 		security_sb_post_remount(nd->mnt, flags, data);
691 	return err;
692 }
693 
694 static int do_move_mount(struct nameidata *nd, char *old_name)
695 {
696 	struct nameidata old_nd, parent_nd;
697 	struct vfsmount *p;
698 	int err = 0;
699 	if (!capable(CAP_SYS_ADMIN))
700 		return -EPERM;
701 	if (!old_name || !*old_name)
702 		return -EINVAL;
703 	err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
704 	if (err)
705 		return err;
706 
707 	down_write(&current->namespace->sem);
708 	while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
709 		;
710 	err = -EINVAL;
711 	if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
712 		goto out;
713 
714 	err = -ENOENT;
715 	down(&nd->dentry->d_inode->i_sem);
716 	if (IS_DEADDIR(nd->dentry->d_inode))
717 		goto out1;
718 
719 	spin_lock(&vfsmount_lock);
720 	if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
721 		goto out2;
722 
723 	err = -EINVAL;
724 	if (old_nd.dentry != old_nd.mnt->mnt_root)
725 		goto out2;
726 
727 	if (old_nd.mnt == old_nd.mnt->mnt_parent)
728 		goto out2;
729 
730 	if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
731 	      S_ISDIR(old_nd.dentry->d_inode->i_mode))
732 		goto out2;
733 
734 	err = -ELOOP;
735 	for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
736 		if (p == old_nd.mnt)
737 			goto out2;
738 	err = 0;
739 
740 	detach_mnt(old_nd.mnt, &parent_nd);
741 	attach_mnt(old_nd.mnt, nd);
742 
743 	/* if the mount is moved, it should no longer be expire
744 	 * automatically */
745 	list_del_init(&old_nd.mnt->mnt_expire);
746 out2:
747 	spin_unlock(&vfsmount_lock);
748 out1:
749 	up(&nd->dentry->d_inode->i_sem);
750 out:
751 	up_write(&current->namespace->sem);
752 	if (!err)
753 		path_release(&parent_nd);
754 	path_release(&old_nd);
755 	return err;
756 }
757 
758 /*
759  * create a new mount for userspace and request it to be added into the
760  * namespace's tree
761  */
762 static int do_new_mount(struct nameidata *nd, char *type, int flags,
763 			int mnt_flags, char *name, void *data)
764 {
765 	struct vfsmount *mnt;
766 
767 	if (!type || !memchr(type, 0, PAGE_SIZE))
768 		return -EINVAL;
769 
770 	/* we need capabilities... */
771 	if (!capable(CAP_SYS_ADMIN))
772 		return -EPERM;
773 
774 	mnt = do_kern_mount(type, flags, name, data);
775 	if (IS_ERR(mnt))
776 		return PTR_ERR(mnt);
777 
778 	return do_add_mount(mnt, nd, mnt_flags, NULL);
779 }
780 
781 /*
782  * add a mount into a namespace's mount tree
783  * - provide the option of adding the new mount to an expiration list
784  */
785 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
786 		 int mnt_flags, struct list_head *fslist)
787 {
788 	int err;
789 
790 	down_write(&current->namespace->sem);
791 	/* Something was mounted here while we slept */
792 	while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
793 		;
794 	err = -EINVAL;
795 	if (!check_mnt(nd->mnt))
796 		goto unlock;
797 
798 	/* Refuse the same filesystem on the same mount point */
799 	err = -EBUSY;
800 	if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
801 	    nd->mnt->mnt_root == nd->dentry)
802 		goto unlock;
803 
804 	err = -EINVAL;
805 	if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
806 		goto unlock;
807 
808 	newmnt->mnt_flags = mnt_flags;
809 	newmnt->mnt_namespace = current->namespace;
810 	err = graft_tree(newmnt, nd);
811 
812 	if (err == 0 && fslist) {
813 		/* add to the specified expiration list */
814 		spin_lock(&vfsmount_lock);
815 		list_add_tail(&newmnt->mnt_expire, fslist);
816 		spin_unlock(&vfsmount_lock);
817 	}
818 
819 unlock:
820 	up_write(&current->namespace->sem);
821 	mntput(newmnt);
822 	return err;
823 }
824 
825 EXPORT_SYMBOL_GPL(do_add_mount);
826 
827 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts)
828 {
829 	spin_lock(&vfsmount_lock);
830 
831 	/*
832 	 * Check if mount is still attached, if not, let whoever holds it deal
833 	 * with the sucker
834 	 */
835 	if (mnt->mnt_parent == mnt) {
836 		spin_unlock(&vfsmount_lock);
837 		return;
838 	}
839 
840 	/*
841 	 * Check that it is still dead: the count should now be 2 - as
842 	 * contributed by the vfsmount parent and the mntget above
843 	 */
844 	if (atomic_read(&mnt->mnt_count) == 2) {
845 		struct nameidata old_nd;
846 
847 		/* delete from the namespace */
848 		list_del_init(&mnt->mnt_list);
849 		mnt->mnt_namespace = NULL;
850 		detach_mnt(mnt, &old_nd);
851 		spin_unlock(&vfsmount_lock);
852 		path_release(&old_nd);
853 
854 		/*
855 		 * Now lay it to rest if this was the last ref on the superblock
856 		 */
857 		if (atomic_read(&mnt->mnt_sb->s_active) == 1) {
858 			/* last instance - try to be smart */
859 			lock_kernel();
860 			DQUOT_OFF(mnt->mnt_sb);
861 			acct_auto_close(mnt->mnt_sb);
862 			unlock_kernel();
863 		}
864 		mntput(mnt);
865 	} else {
866 		/*
867 		 * Someone brought it back to life whilst we didn't have any
868 		 * locks held so return it to the expiration list
869 		 */
870 		list_add_tail(&mnt->mnt_expire, mounts);
871 		spin_unlock(&vfsmount_lock);
872 	}
873 }
874 
875 /*
876  * process a list of expirable mountpoints with the intent of discarding any
877  * mountpoints that aren't in use and haven't been touched since last we came
878  * here
879  */
880 void mark_mounts_for_expiry(struct list_head *mounts)
881 {
882 	struct namespace *namespace;
883 	struct vfsmount *mnt, *next;
884 	LIST_HEAD(graveyard);
885 
886 	if (list_empty(mounts))
887 		return;
888 
889 	spin_lock(&vfsmount_lock);
890 
891 	/* extract from the expiration list every vfsmount that matches the
892 	 * following criteria:
893 	 * - only referenced by its parent vfsmount
894 	 * - still marked for expiry (marked on the last call here; marks are
895 	 *   cleared by mntput())
896 	 */
897 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
898 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
899 		    atomic_read(&mnt->mnt_count) != 1)
900 			continue;
901 
902 		mntget(mnt);
903 		list_move(&mnt->mnt_expire, &graveyard);
904 	}
905 
906 	/*
907 	 * go through the vfsmounts we've just consigned to the graveyard to
908 	 * - check that they're still dead
909 	 * - delete the vfsmount from the appropriate namespace under lock
910 	 * - dispose of the corpse
911 	 */
912 	while (!list_empty(&graveyard)) {
913 		mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
914 		list_del_init(&mnt->mnt_expire);
915 
916 		/* don't do anything if the namespace is dead - all the
917 		 * vfsmounts from it are going away anyway */
918 		namespace = mnt->mnt_namespace;
919 		if (!namespace || !namespace->root)
920 			continue;
921 		get_namespace(namespace);
922 
923 		spin_unlock(&vfsmount_lock);
924 		down_write(&namespace->sem);
925 		expire_mount(mnt, mounts);
926 		up_write(&namespace->sem);
927 
928 		mntput(mnt);
929 		put_namespace(namespace);
930 
931 		spin_lock(&vfsmount_lock);
932 	}
933 
934 	spin_unlock(&vfsmount_lock);
935 }
936 
937 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
938 
939 /*
940  * Some copy_from_user() implementations do not return the exact number of
941  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
942  * Note that this function differs from copy_from_user() in that it will oops
943  * on bad values of `to', rather than returning a short copy.
944  */
945 static long
946 exact_copy_from_user(void *to, const void __user *from, unsigned long n)
947 {
948 	char *t = to;
949 	const char __user *f = from;
950 	char c;
951 
952 	if (!access_ok(VERIFY_READ, from, n))
953 		return n;
954 
955 	while (n) {
956 		if (__get_user(c, f)) {
957 			memset(t, 0, n);
958 			break;
959 		}
960 		*t++ = c;
961 		f++;
962 		n--;
963 	}
964 	return n;
965 }
966 
967 int copy_mount_options(const void __user *data, unsigned long *where)
968 {
969 	int i;
970 	unsigned long page;
971 	unsigned long size;
972 
973 	*where = 0;
974 	if (!data)
975 		return 0;
976 
977 	if (!(page = __get_free_page(GFP_KERNEL)))
978 		return -ENOMEM;
979 
980 	/* We only care that *some* data at the address the user
981 	 * gave us is valid.  Just in case, we'll zero
982 	 * the remainder of the page.
983 	 */
984 	/* copy_from_user cannot cross TASK_SIZE ! */
985 	size = TASK_SIZE - (unsigned long)data;
986 	if (size > PAGE_SIZE)
987 		size = PAGE_SIZE;
988 
989 	i = size - exact_copy_from_user((void *)page, data, size);
990 	if (!i) {
991 		free_page(page);
992 		return -EFAULT;
993 	}
994 	if (i != PAGE_SIZE)
995 		memset((char *)page + i, 0, PAGE_SIZE - i);
996 	*where = page;
997 	return 0;
998 }
999 
1000 /*
1001  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1002  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1003  *
1004  * data is a (void *) that can point to any structure up to
1005  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1006  * information (or be NULL).
1007  *
1008  * Pre-0.97 versions of mount() didn't have a flags word.
1009  * When the flags word was introduced its top half was required
1010  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1011  * Therefore, if this magic number is present, it carries no information
1012  * and must be discarded.
1013  */
1014 long do_mount(char * dev_name, char * dir_name, char *type_page,
1015 		  unsigned long flags, void *data_page)
1016 {
1017 	struct nameidata nd;
1018 	int retval = 0;
1019 	int mnt_flags = 0;
1020 
1021 	/* Discard magic */
1022 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1023 		flags &= ~MS_MGC_MSK;
1024 
1025 	/* Basic sanity checks */
1026 
1027 	if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1028 		return -EINVAL;
1029 	if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1030 		return -EINVAL;
1031 
1032 	if (data_page)
1033 		((char *)data_page)[PAGE_SIZE - 1] = 0;
1034 
1035 	/* Separate the per-mountpoint flags */
1036 	if (flags & MS_NOSUID)
1037 		mnt_flags |= MNT_NOSUID;
1038 	if (flags & MS_NODEV)
1039 		mnt_flags |= MNT_NODEV;
1040 	if (flags & MS_NOEXEC)
1041 		mnt_flags |= MNT_NOEXEC;
1042 	flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_ACTIVE);
1043 
1044 	/* ... and get the mountpoint */
1045 	retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1046 	if (retval)
1047 		return retval;
1048 
1049 	retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1050 	if (retval)
1051 		goto dput_out;
1052 
1053 	if (flags & MS_REMOUNT)
1054 		retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1055 				    data_page);
1056 	else if (flags & MS_BIND)
1057 		retval = do_loopback(&nd, dev_name, flags & MS_REC);
1058 	else if (flags & MS_MOVE)
1059 		retval = do_move_mount(&nd, dev_name);
1060 	else
1061 		retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1062 				      dev_name, data_page);
1063 dput_out:
1064 	path_release(&nd);
1065 	return retval;
1066 }
1067 
1068 int copy_namespace(int flags, struct task_struct *tsk)
1069 {
1070 	struct namespace *namespace = tsk->namespace;
1071 	struct namespace *new_ns;
1072 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1073 	struct fs_struct *fs = tsk->fs;
1074 	struct vfsmount *p, *q;
1075 
1076 	if (!namespace)
1077 		return 0;
1078 
1079 	get_namespace(namespace);
1080 
1081 	if (!(flags & CLONE_NEWNS))
1082 		return 0;
1083 
1084 	if (!capable(CAP_SYS_ADMIN)) {
1085 		put_namespace(namespace);
1086 		return -EPERM;
1087 	}
1088 
1089 	new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1090 	if (!new_ns)
1091 		goto out;
1092 
1093 	atomic_set(&new_ns->count, 1);
1094 	init_rwsem(&new_ns->sem);
1095 	INIT_LIST_HEAD(&new_ns->list);
1096 
1097 	down_write(&tsk->namespace->sem);
1098 	/* First pass: copy the tree topology */
1099 	new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
1100 	if (!new_ns->root) {
1101 		up_write(&tsk->namespace->sem);
1102 		kfree(new_ns);
1103 		goto out;
1104 	}
1105 	spin_lock(&vfsmount_lock);
1106 	list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1107 	spin_unlock(&vfsmount_lock);
1108 
1109 	/*
1110 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1111 	 * as belonging to new namespace.  We have already acquired a private
1112 	 * fs_struct, so tsk->fs->lock is not needed.
1113 	 */
1114 	p = namespace->root;
1115 	q = new_ns->root;
1116 	while (p) {
1117 		q->mnt_namespace = new_ns;
1118 		if (fs) {
1119 			if (p == fs->rootmnt) {
1120 				rootmnt = p;
1121 				fs->rootmnt = mntget(q);
1122 			}
1123 			if (p == fs->pwdmnt) {
1124 				pwdmnt = p;
1125 				fs->pwdmnt = mntget(q);
1126 			}
1127 			if (p == fs->altrootmnt) {
1128 				altrootmnt = p;
1129 				fs->altrootmnt = mntget(q);
1130 			}
1131 		}
1132 		p = next_mnt(p, namespace->root);
1133 		q = next_mnt(q, new_ns->root);
1134 	}
1135 	up_write(&tsk->namespace->sem);
1136 
1137 	tsk->namespace = new_ns;
1138 
1139 	if (rootmnt)
1140 		mntput(rootmnt);
1141 	if (pwdmnt)
1142 		mntput(pwdmnt);
1143 	if (altrootmnt)
1144 		mntput(altrootmnt);
1145 
1146 	put_namespace(namespace);
1147 	return 0;
1148 
1149 out:
1150 	put_namespace(namespace);
1151 	return -ENOMEM;
1152 }
1153 
1154 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1155 			  char __user * type, unsigned long flags,
1156 			  void __user * data)
1157 {
1158 	int retval;
1159 	unsigned long data_page;
1160 	unsigned long type_page;
1161 	unsigned long dev_page;
1162 	char *dir_page;
1163 
1164 	retval = copy_mount_options (type, &type_page);
1165 	if (retval < 0)
1166 		return retval;
1167 
1168 	dir_page = getname(dir_name);
1169 	retval = PTR_ERR(dir_page);
1170 	if (IS_ERR(dir_page))
1171 		goto out1;
1172 
1173 	retval = copy_mount_options (dev_name, &dev_page);
1174 	if (retval < 0)
1175 		goto out2;
1176 
1177 	retval = copy_mount_options (data, &data_page);
1178 	if (retval < 0)
1179 		goto out3;
1180 
1181 	lock_kernel();
1182 	retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
1183 			  flags, (void*)data_page);
1184 	unlock_kernel();
1185 	free_page(data_page);
1186 
1187 out3:
1188 	free_page(dev_page);
1189 out2:
1190 	putname(dir_page);
1191 out1:
1192 	free_page(type_page);
1193 	return retval;
1194 }
1195 
1196 /*
1197  * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1198  * It can block. Requires the big lock held.
1199  */
1200 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1201 		 struct dentry *dentry)
1202 {
1203 	struct dentry *old_root;
1204 	struct vfsmount *old_rootmnt;
1205 	write_lock(&fs->lock);
1206 	old_root = fs->root;
1207 	old_rootmnt = fs->rootmnt;
1208 	fs->rootmnt = mntget(mnt);
1209 	fs->root = dget(dentry);
1210 	write_unlock(&fs->lock);
1211 	if (old_root) {
1212 		dput(old_root);
1213 		mntput(old_rootmnt);
1214 	}
1215 }
1216 
1217 /*
1218  * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1219  * It can block. Requires the big lock held.
1220  */
1221 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1222 		struct dentry *dentry)
1223 {
1224 	struct dentry *old_pwd;
1225 	struct vfsmount *old_pwdmnt;
1226 
1227 	write_lock(&fs->lock);
1228 	old_pwd = fs->pwd;
1229 	old_pwdmnt = fs->pwdmnt;
1230 	fs->pwdmnt = mntget(mnt);
1231 	fs->pwd = dget(dentry);
1232 	write_unlock(&fs->lock);
1233 
1234 	if (old_pwd) {
1235 		dput(old_pwd);
1236 		mntput(old_pwdmnt);
1237 	}
1238 }
1239 
1240 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1241 {
1242 	struct task_struct *g, *p;
1243 	struct fs_struct *fs;
1244 
1245 	read_lock(&tasklist_lock);
1246 	do_each_thread(g, p) {
1247 		task_lock(p);
1248 		fs = p->fs;
1249 		if (fs) {
1250 			atomic_inc(&fs->count);
1251 			task_unlock(p);
1252 			if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
1253 				set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1254 			if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
1255 				set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1256 			put_fs_struct(fs);
1257 		} else
1258 			task_unlock(p);
1259 	} while_each_thread(g, p);
1260 	read_unlock(&tasklist_lock);
1261 }
1262 
1263 /*
1264  * pivot_root Semantics:
1265  * Moves the root file system of the current process to the directory put_old,
1266  * makes new_root as the new root file system of the current process, and sets
1267  * root/cwd of all processes which had them on the current root to new_root.
1268  *
1269  * Restrictions:
1270  * The new_root and put_old must be directories, and  must not be on the
1271  * same file  system as the current process root. The put_old  must  be
1272  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
1273  * pointed to by put_old must yield the same directory as new_root. No other
1274  * file system may be mounted on put_old. After all, new_root is a mountpoint.
1275  *
1276  * Notes:
1277  *  - we don't move root/cwd if they are not at the root (reason: if something
1278  *    cared enough to change them, it's probably wrong to force them elsewhere)
1279  *  - it's okay to pick a root that isn't the root of a file system, e.g.
1280  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1281  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1282  *    first.
1283  */
1284 
1285 asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
1286 {
1287 	struct vfsmount *tmp;
1288 	struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1289 	int error;
1290 
1291 	if (!capable(CAP_SYS_ADMIN))
1292 		return -EPERM;
1293 
1294 	lock_kernel();
1295 
1296 	error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
1297 	if (error)
1298 		goto out0;
1299 	error = -EINVAL;
1300 	if (!check_mnt(new_nd.mnt))
1301 		goto out1;
1302 
1303 	error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
1304 	if (error)
1305 		goto out1;
1306 
1307 	error = security_sb_pivotroot(&old_nd, &new_nd);
1308 	if (error) {
1309 		path_release(&old_nd);
1310 		goto out1;
1311 	}
1312 
1313 	read_lock(&current->fs->lock);
1314 	user_nd.mnt = mntget(current->fs->rootmnt);
1315 	user_nd.dentry = dget(current->fs->root);
1316 	read_unlock(&current->fs->lock);
1317 	down_write(&current->namespace->sem);
1318 	down(&old_nd.dentry->d_inode->i_sem);
1319 	error = -EINVAL;
1320 	if (!check_mnt(user_nd.mnt))
1321 		goto out2;
1322 	error = -ENOENT;
1323 	if (IS_DEADDIR(new_nd.dentry->d_inode))
1324 		goto out2;
1325 	if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1326 		goto out2;
1327 	if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1328 		goto out2;
1329 	error = -EBUSY;
1330 	if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1331 		goto out2; /* loop, on the same file system  */
1332 	error = -EINVAL;
1333 	if (user_nd.mnt->mnt_root != user_nd.dentry)
1334 		goto out2; /* not a mountpoint */
1335 	if (user_nd.mnt->mnt_parent == user_nd.mnt)
1336 		goto out2; /* not attached */
1337 	if (new_nd.mnt->mnt_root != new_nd.dentry)
1338 		goto out2; /* not a mountpoint */
1339 	if (new_nd.mnt->mnt_parent == new_nd.mnt)
1340 		goto out2; /* not attached */
1341 	tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1342 	spin_lock(&vfsmount_lock);
1343 	if (tmp != new_nd.mnt) {
1344 		for (;;) {
1345 			if (tmp->mnt_parent == tmp)
1346 				goto out3; /* already mounted on put_old */
1347 			if (tmp->mnt_parent == new_nd.mnt)
1348 				break;
1349 			tmp = tmp->mnt_parent;
1350 		}
1351 		if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1352 			goto out3;
1353 	} else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1354 		goto out3;
1355 	detach_mnt(new_nd.mnt, &parent_nd);
1356 	detach_mnt(user_nd.mnt, &root_parent);
1357 	attach_mnt(user_nd.mnt, &old_nd);     /* mount old root on put_old */
1358 	attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1359 	spin_unlock(&vfsmount_lock);
1360 	chroot_fs_refs(&user_nd, &new_nd);
1361 	security_sb_post_pivotroot(&user_nd, &new_nd);
1362 	error = 0;
1363 	path_release(&root_parent);
1364 	path_release(&parent_nd);
1365 out2:
1366 	up(&old_nd.dentry->d_inode->i_sem);
1367 	up_write(&current->namespace->sem);
1368 	path_release(&user_nd);
1369 	path_release(&old_nd);
1370 out1:
1371 	path_release(&new_nd);
1372 out0:
1373 	unlock_kernel();
1374 	return error;
1375 out3:
1376 	spin_unlock(&vfsmount_lock);
1377 	goto out2;
1378 }
1379 
1380 static void __init init_mount_tree(void)
1381 {
1382 	struct vfsmount *mnt;
1383 	struct namespace *namespace;
1384 	struct task_struct *g, *p;
1385 
1386 	mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1387 	if (IS_ERR(mnt))
1388 		panic("Can't create rootfs");
1389 	namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1390 	if (!namespace)
1391 		panic("Can't allocate initial namespace");
1392 	atomic_set(&namespace->count, 1);
1393 	INIT_LIST_HEAD(&namespace->list);
1394 	init_rwsem(&namespace->sem);
1395 	list_add(&mnt->mnt_list, &namespace->list);
1396 	namespace->root = mnt;
1397 	mnt->mnt_namespace = namespace;
1398 
1399 	init_task.namespace = namespace;
1400 	read_lock(&tasklist_lock);
1401 	do_each_thread(g, p) {
1402 		get_namespace(namespace);
1403 		p->namespace = namespace;
1404 	} while_each_thread(g, p);
1405 	read_unlock(&tasklist_lock);
1406 
1407 	set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1408 	set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1409 }
1410 
1411 void __init mnt_init(unsigned long mempages)
1412 {
1413 	struct list_head *d;
1414 	unsigned int nr_hash;
1415 	int i;
1416 
1417 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1418 			0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1419 
1420 	mount_hashtable = (struct list_head *)
1421 		__get_free_page(GFP_ATOMIC);
1422 
1423 	if (!mount_hashtable)
1424 		panic("Failed to allocate mount hash table\n");
1425 
1426 	/*
1427 	 * Find the power-of-two list-heads that can fit into the allocation..
1428 	 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1429 	 * a power-of-two.
1430 	 */
1431 	nr_hash = PAGE_SIZE / sizeof(struct list_head);
1432 	hash_bits = 0;
1433 	do {
1434 		hash_bits++;
1435 	} while ((nr_hash >> hash_bits) != 0);
1436 	hash_bits--;
1437 
1438 	/*
1439 	 * Re-calculate the actual number of entries and the mask
1440 	 * from the number of bits we can fit.
1441 	 */
1442 	nr_hash = 1UL << hash_bits;
1443 	hash_mask = nr_hash-1;
1444 
1445 	printk("Mount-cache hash table entries: %d\n", nr_hash);
1446 
1447 	/* And initialize the newly allocated array */
1448 	d = mount_hashtable;
1449 	i = nr_hash;
1450 	do {
1451 		INIT_LIST_HEAD(d);
1452 		d++;
1453 		i--;
1454 	} while (i);
1455 	sysfs_init();
1456 	init_rootfs();
1457 	init_mount_tree();
1458 }
1459 
1460 void __put_namespace(struct namespace *namespace)
1461 {
1462 	struct vfsmount *root = namespace->root;
1463 	namespace->root = NULL;
1464 	spin_unlock(&vfsmount_lock);
1465 	down_write(&namespace->sem);
1466 	spin_lock(&vfsmount_lock);
1467 	umount_tree(root);
1468 	spin_unlock(&vfsmount_lock);
1469 	up_write(&namespace->sem);
1470 	kfree(namespace);
1471 }
1472