blob: 34156260c9b6586939e932e3856cc0211e98cd97 [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>
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
28extern int __init init_rootfs(void);
29
30#ifdef CONFIG_SYSFS
31extern int __init sysfs_init(void);
32#else
33static 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
42static struct list_head *mount_hashtable;
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -070043static int hash_mask __read_mostly, hash_bits __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044static kmem_cache_t *mnt_cache;
45
46static 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
54struct 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);
Miklos Szeredi55e700b2005-07-07 17:57:30 -070064 INIT_LIST_HEAD(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 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
77void 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 */
87struct 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
109static inline int check_mnt(struct vfsmount *mnt)
110{
111 return mnt->mnt_namespace == current->namespace;
112}
113
114static 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
125static 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
134static 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
150static struct vfsmount *
151clone_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;
Miklos Szeredi68b47132005-08-07 09:42:25 -0700163 mnt->mnt_namespace = current->namespace;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 /* stick the duplicate mount on the same expiry list
166 * as the original if that was on one */
167 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700168 if (!list_empty(&old->mnt_expire))
169 list_add(&mnt->mnt_expire, &old->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 spin_unlock(&vfsmount_lock);
171 }
172 return mnt;
173}
174
175void __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
183EXPORT_SYMBOL(__mntput);
184
185/* iterator */
186static 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
199static 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
207static void m_stop(struct seq_file *m, void *v)
208{
209 struct namespace *n = m->private;
210 up_read(&n->sem);
211}
212
213static inline void mangle(struct seq_file *m, const char *s)
214{
215 seq_escape(m, s, " \t\n\\");
216}
217
218static 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
261struct 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 */
276int 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;
286repeat:
287 next = this_parent->mnt_mounts.next;
288resume:
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
316EXPORT_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 */
331int may_umount(struct vfsmount *mnt)
332{
333 if (atomic_read(&mnt->mnt_count) > 2)
334 return -EBUSY;
335 return 0;
336}
337
338EXPORT_SYMBOL(may_umount);
339
Adrian Bunk52c1da32005-06-23 22:05:33 -0700340static void umount_tree(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
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);
Miklos Szeredi202322e2005-07-07 17:57:22 -0700348 p->mnt_namespace = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 }
350
351 while (!list_empty(&kill)) {
352 mnt = list_entry(kill.next, struct vfsmount, mnt_list);
353 list_del_init(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700354 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 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
368static 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
469asmlinkage 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);
488dput_and_out:
489 path_release_on_umount(&nd);
490out:
491 return retval;
492}
493
494#ifdef __ARCH_WANT_SYS_OLDUMOUNT
495
496/*
497 * The 2.0 compatible umount. No flags.
498 */
499
500asmlinkage long sys_oldumount(char __user * name)
501{
502 return sys_umount(name,0);
503}
504
505#endif
506
507static 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
525static int
526lives_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
537static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
538{
539 struct vfsmount *res, *p, *q, *r, *s;
540 struct list_head *h;
541 struct nameidata nd;
542
543 res = q = clone_mnt(mnt, dentry);
544 if (!q)
545 goto Enomem;
546 q->mnt_mountpoint = mnt->mnt_mountpoint;
547
548 p = mnt;
549 for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) {
550 r = list_entry(h, struct vfsmount, mnt_child);
551 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
552 continue;
553
554 for (s = r; s; s = next_mnt(s, r)) {
555 while (p != s->mnt_parent) {
556 p = p->mnt_parent;
557 q = q->mnt_parent;
558 }
559 p = s;
560 nd.mnt = q;
561 nd.dentry = p->mnt_mountpoint;
562 q = clone_mnt(p, p->mnt_root);
563 if (!q)
564 goto Enomem;
565 spin_lock(&vfsmount_lock);
566 list_add_tail(&q->mnt_list, &res->mnt_list);
567 attach_mnt(q, &nd);
568 spin_unlock(&vfsmount_lock);
569 }
570 }
571 return res;
572 Enomem:
573 if (res) {
574 spin_lock(&vfsmount_lock);
575 umount_tree(res);
576 spin_unlock(&vfsmount_lock);
577 }
578 return NULL;
579}
580
581static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
582{
583 int err;
584 if (mnt->mnt_sb->s_flags & MS_NOUSER)
585 return -EINVAL;
586
587 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
588 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
589 return -ENOTDIR;
590
591 err = -ENOENT;
592 down(&nd->dentry->d_inode->i_sem);
593 if (IS_DEADDIR(nd->dentry->d_inode))
594 goto out_unlock;
595
596 err = security_sb_check_sb(mnt, nd);
597 if (err)
598 goto out_unlock;
599
600 err = -ENOENT;
601 spin_lock(&vfsmount_lock);
602 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
603 struct list_head head;
604
605 attach_mnt(mnt, nd);
606 list_add_tail(&head, &mnt->mnt_list);
607 list_splice(&head, current->namespace->list.prev);
608 mntget(mnt);
609 err = 0;
610 }
611 spin_unlock(&vfsmount_lock);
612out_unlock:
613 up(&nd->dentry->d_inode->i_sem);
614 if (!err)
615 security_sb_post_addmount(mnt, nd);
616 return err;
617}
618
619/*
620 * do loopback mount.
621 */
622static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
623{
624 struct nameidata old_nd;
625 struct vfsmount *mnt = NULL;
626 int err = mount_is_safe(nd);
627 if (err)
628 return err;
629 if (!old_name || !*old_name)
630 return -EINVAL;
631 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
632 if (err)
633 return err;
634
635 down_write(&current->namespace->sem);
636 err = -EINVAL;
637 if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
638 err = -ENOMEM;
639 if (recurse)
640 mnt = copy_tree(old_nd.mnt, old_nd.dentry);
641 else
642 mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
643 }
644
645 if (mnt) {
646 /* stop bind mounts from expiring */
647 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700648 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 spin_unlock(&vfsmount_lock);
650
651 err = graft_tree(mnt, nd);
652 if (err) {
653 spin_lock(&vfsmount_lock);
654 umount_tree(mnt);
655 spin_unlock(&vfsmount_lock);
656 } else
657 mntput(mnt);
658 }
659
660 up_write(&current->namespace->sem);
661 path_release(&old_nd);
662 return err;
663}
664
665/*
666 * change filesystem flags. dir should be a physical root of filesystem.
667 * If you've mounted a non-root directory somewhere and want to do remount
668 * on it - tough luck.
669 */
670
671static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
672 void *data)
673{
674 int err;
675 struct super_block * sb = nd->mnt->mnt_sb;
676
677 if (!capable(CAP_SYS_ADMIN))
678 return -EPERM;
679
680 if (!check_mnt(nd->mnt))
681 return -EINVAL;
682
683 if (nd->dentry != nd->mnt->mnt_root)
684 return -EINVAL;
685
686 down_write(&sb->s_umount);
687 err = do_remount_sb(sb, flags, data, 0);
688 if (!err)
689 nd->mnt->mnt_flags=mnt_flags;
690 up_write(&sb->s_umount);
691 if (!err)
692 security_sb_post_remount(nd->mnt, flags, data);
693 return err;
694}
695
696static int do_move_mount(struct nameidata *nd, char *old_name)
697{
698 struct nameidata old_nd, parent_nd;
699 struct vfsmount *p;
700 int err = 0;
701 if (!capable(CAP_SYS_ADMIN))
702 return -EPERM;
703 if (!old_name || !*old_name)
704 return -EINVAL;
705 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
706 if (err)
707 return err;
708
709 down_write(&current->namespace->sem);
710 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
711 ;
712 err = -EINVAL;
713 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
714 goto out;
715
716 err = -ENOENT;
717 down(&nd->dentry->d_inode->i_sem);
718 if (IS_DEADDIR(nd->dentry->d_inode))
719 goto out1;
720
721 spin_lock(&vfsmount_lock);
722 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
723 goto out2;
724
725 err = -EINVAL;
726 if (old_nd.dentry != old_nd.mnt->mnt_root)
727 goto out2;
728
729 if (old_nd.mnt == old_nd.mnt->mnt_parent)
730 goto out2;
731
732 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
733 S_ISDIR(old_nd.dentry->d_inode->i_mode))
734 goto out2;
735
736 err = -ELOOP;
737 for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
738 if (p == old_nd.mnt)
739 goto out2;
740 err = 0;
741
742 detach_mnt(old_nd.mnt, &parent_nd);
743 attach_mnt(old_nd.mnt, nd);
744
745 /* if the mount is moved, it should no longer be expire
746 * automatically */
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700747 list_del_init(&old_nd.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748out2:
749 spin_unlock(&vfsmount_lock);
750out1:
751 up(&nd->dentry->d_inode->i_sem);
752out:
753 up_write(&current->namespace->sem);
754 if (!err)
755 path_release(&parent_nd);
756 path_release(&old_nd);
757 return err;
758}
759
760/*
761 * create a new mount for userspace and request it to be added into the
762 * namespace's tree
763 */
764static int do_new_mount(struct nameidata *nd, char *type, int flags,
765 int mnt_flags, char *name, void *data)
766{
767 struct vfsmount *mnt;
768
769 if (!type || !memchr(type, 0, PAGE_SIZE))
770 return -EINVAL;
771
772 /* we need capabilities... */
773 if (!capable(CAP_SYS_ADMIN))
774 return -EPERM;
775
776 mnt = do_kern_mount(type, flags, name, data);
777 if (IS_ERR(mnt))
778 return PTR_ERR(mnt);
779
780 return do_add_mount(mnt, nd, mnt_flags, NULL);
781}
782
783/*
784 * add a mount into a namespace's mount tree
785 * - provide the option of adding the new mount to an expiration list
786 */
787int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
788 int mnt_flags, struct list_head *fslist)
789{
790 int err;
791
792 down_write(&current->namespace->sem);
793 /* Something was mounted here while we slept */
794 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
795 ;
796 err = -EINVAL;
797 if (!check_mnt(nd->mnt))
798 goto unlock;
799
800 /* Refuse the same filesystem on the same mount point */
801 err = -EBUSY;
802 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
803 nd->mnt->mnt_root == nd->dentry)
804 goto unlock;
805
806 err = -EINVAL;
807 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
808 goto unlock;
809
810 newmnt->mnt_flags = mnt_flags;
Miklos Szeredi484e3892005-07-07 17:57:28 -0700811 newmnt->mnt_namespace = current->namespace;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 err = graft_tree(newmnt, nd);
813
814 if (err == 0 && fslist) {
815 /* add to the specified expiration list */
816 spin_lock(&vfsmount_lock);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700817 list_add_tail(&newmnt->mnt_expire, fslist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 spin_unlock(&vfsmount_lock);
819 }
820
821unlock:
822 up_write(&current->namespace->sem);
823 mntput(newmnt);
824 return err;
825}
826
827EXPORT_SYMBOL_GPL(do_add_mount);
828
Miklos Szeredi24ca2af2005-07-07 17:57:25 -0700829static void expire_mount(struct vfsmount *mnt, struct list_head *mounts)
830{
831 spin_lock(&vfsmount_lock);
832
833 /*
Miklos Szeredied42c872005-07-07 17:57:26 -0700834 * Check if mount is still attached, if not, let whoever holds it deal
835 * with the sucker
836 */
837 if (mnt->mnt_parent == mnt) {
838 spin_unlock(&vfsmount_lock);
839 return;
840 }
841
842 /*
Miklos Szeredi24ca2af2005-07-07 17:57:25 -0700843 * Check that it is still dead: the count should now be 2 - as
844 * contributed by the vfsmount parent and the mntget above
845 */
846 if (atomic_read(&mnt->mnt_count) == 2) {
847 struct nameidata old_nd;
848
849 /* delete from the namespace */
850 list_del_init(&mnt->mnt_list);
Miklos Szerediac081152005-07-07 17:57:27 -0700851 mnt->mnt_namespace = NULL;
Miklos Szeredi24ca2af2005-07-07 17:57:25 -0700852 detach_mnt(mnt, &old_nd);
853 spin_unlock(&vfsmount_lock);
854 path_release(&old_nd);
855
856 /*
857 * Now lay it to rest if this was the last ref on the superblock
858 */
859 if (atomic_read(&mnt->mnt_sb->s_active) == 1) {
860 /* last instance - try to be smart */
861 lock_kernel();
862 DQUOT_OFF(mnt->mnt_sb);
863 acct_auto_close(mnt->mnt_sb);
864 unlock_kernel();
865 }
866 mntput(mnt);
867 } else {
868 /*
869 * Someone brought it back to life whilst we didn't have any
870 * locks held so return it to the expiration list
871 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700872 list_add_tail(&mnt->mnt_expire, mounts);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -0700873 spin_unlock(&vfsmount_lock);
874 }
875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877/*
878 * process a list of expirable mountpoints with the intent of discarding any
879 * mountpoints that aren't in use and haven't been touched since last we came
880 * here
881 */
882void mark_mounts_for_expiry(struct list_head *mounts)
883{
884 struct namespace *namespace;
885 struct vfsmount *mnt, *next;
886 LIST_HEAD(graveyard);
887
888 if (list_empty(mounts))
889 return;
890
891 spin_lock(&vfsmount_lock);
892
893 /* extract from the expiration list every vfsmount that matches the
894 * following criteria:
895 * - only referenced by its parent vfsmount
896 * - still marked for expiry (marked on the last call here; marks are
897 * cleared by mntput())
898 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700899 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
901 atomic_read(&mnt->mnt_count) != 1)
902 continue;
903
904 mntget(mnt);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700905 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 }
907
908 /*
909 * go through the vfsmounts we've just consigned to the graveyard to
910 * - check that they're still dead
911 * - delete the vfsmount from the appropriate namespace under lock
912 * - dispose of the corpse
913 */
914 while (!list_empty(&graveyard)) {
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700915 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
916 list_del_init(&mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
918 /* don't do anything if the namespace is dead - all the
919 * vfsmounts from it are going away anyway */
920 namespace = mnt->mnt_namespace;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -0700921 if (!namespace || !namespace->root)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 continue;
923 get_namespace(namespace);
924
925 spin_unlock(&vfsmount_lock);
926 down_write(&namespace->sem);
Miklos Szeredi24ca2af2005-07-07 17:57:25 -0700927 expire_mount(mnt, mounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 up_write(&namespace->sem);
929
930 mntput(mnt);
931 put_namespace(namespace);
932
933 spin_lock(&vfsmount_lock);
934 }
935
936 spin_unlock(&vfsmount_lock);
937}
938
939EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
940
941/*
942 * Some copy_from_user() implementations do not return the exact number of
943 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
944 * Note that this function differs from copy_from_user() in that it will oops
945 * on bad values of `to', rather than returning a short copy.
946 */
947static long
948exact_copy_from_user(void *to, const void __user *from, unsigned long n)
949{
950 char *t = to;
951 const char __user *f = from;
952 char c;
953
954 if (!access_ok(VERIFY_READ, from, n))
955 return n;
956
957 while (n) {
958 if (__get_user(c, f)) {
959 memset(t, 0, n);
960 break;
961 }
962 *t++ = c;
963 f++;
964 n--;
965 }
966 return n;
967}
968
969int copy_mount_options(const void __user *data, unsigned long *where)
970{
971 int i;
972 unsigned long page;
973 unsigned long size;
974
975 *where = 0;
976 if (!data)
977 return 0;
978
979 if (!(page = __get_free_page(GFP_KERNEL)))
980 return -ENOMEM;
981
982 /* We only care that *some* data at the address the user
983 * gave us is valid. Just in case, we'll zero
984 * the remainder of the page.
985 */
986 /* copy_from_user cannot cross TASK_SIZE ! */
987 size = TASK_SIZE - (unsigned long)data;
988 if (size > PAGE_SIZE)
989 size = PAGE_SIZE;
990
991 i = size - exact_copy_from_user((void *)page, data, size);
992 if (!i) {
993 free_page(page);
994 return -EFAULT;
995 }
996 if (i != PAGE_SIZE)
997 memset((char *)page + i, 0, PAGE_SIZE - i);
998 *where = page;
999 return 0;
1000}
1001
1002/*
1003 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1004 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1005 *
1006 * data is a (void *) that can point to any structure up to
1007 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1008 * information (or be NULL).
1009 *
1010 * Pre-0.97 versions of mount() didn't have a flags word.
1011 * When the flags word was introduced its top half was required
1012 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1013 * Therefore, if this magic number is present, it carries no information
1014 * and must be discarded.
1015 */
1016long do_mount(char * dev_name, char * dir_name, char *type_page,
1017 unsigned long flags, void *data_page)
1018{
1019 struct nameidata nd;
1020 int retval = 0;
1021 int mnt_flags = 0;
1022
1023 /* Discard magic */
1024 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1025 flags &= ~MS_MGC_MSK;
1026
1027 /* Basic sanity checks */
1028
1029 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1030 return -EINVAL;
1031 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1032 return -EINVAL;
1033
1034 if (data_page)
1035 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1036
1037 /* Separate the per-mountpoint flags */
1038 if (flags & MS_NOSUID)
1039 mnt_flags |= MNT_NOSUID;
1040 if (flags & MS_NODEV)
1041 mnt_flags |= MNT_NODEV;
1042 if (flags & MS_NOEXEC)
1043 mnt_flags |= MNT_NOEXEC;
1044 flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_ACTIVE);
1045
1046 /* ... and get the mountpoint */
1047 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1048 if (retval)
1049 return retval;
1050
1051 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1052 if (retval)
1053 goto dput_out;
1054
1055 if (flags & MS_REMOUNT)
1056 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1057 data_page);
1058 else if (flags & MS_BIND)
1059 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1060 else if (flags & MS_MOVE)
1061 retval = do_move_mount(&nd, dev_name);
1062 else
1063 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1064 dev_name, data_page);
1065dput_out:
1066 path_release(&nd);
1067 return retval;
1068}
1069
1070int copy_namespace(int flags, struct task_struct *tsk)
1071{
1072 struct namespace *namespace = tsk->namespace;
1073 struct namespace *new_ns;
1074 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1075 struct fs_struct *fs = tsk->fs;
1076 struct vfsmount *p, *q;
1077
1078 if (!namespace)
1079 return 0;
1080
1081 get_namespace(namespace);
1082
1083 if (!(flags & CLONE_NEWNS))
1084 return 0;
1085
1086 if (!capable(CAP_SYS_ADMIN)) {
1087 put_namespace(namespace);
1088 return -EPERM;
1089 }
1090
1091 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1092 if (!new_ns)
1093 goto out;
1094
1095 atomic_set(&new_ns->count, 1);
1096 init_rwsem(&new_ns->sem);
1097 INIT_LIST_HEAD(&new_ns->list);
1098
1099 down_write(&tsk->namespace->sem);
1100 /* First pass: copy the tree topology */
1101 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
1102 if (!new_ns->root) {
1103 up_write(&tsk->namespace->sem);
1104 kfree(new_ns);
1105 goto out;
1106 }
1107 spin_lock(&vfsmount_lock);
1108 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1109 spin_unlock(&vfsmount_lock);
1110
1111 /*
1112 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1113 * as belonging to new namespace. We have already acquired a private
1114 * fs_struct, so tsk->fs->lock is not needed.
1115 */
1116 p = namespace->root;
1117 q = new_ns->root;
1118 while (p) {
1119 q->mnt_namespace = new_ns;
1120 if (fs) {
1121 if (p == fs->rootmnt) {
1122 rootmnt = p;
1123 fs->rootmnt = mntget(q);
1124 }
1125 if (p == fs->pwdmnt) {
1126 pwdmnt = p;
1127 fs->pwdmnt = mntget(q);
1128 }
1129 if (p == fs->altrootmnt) {
1130 altrootmnt = p;
1131 fs->altrootmnt = mntget(q);
1132 }
1133 }
1134 p = next_mnt(p, namespace->root);
1135 q = next_mnt(q, new_ns->root);
1136 }
1137 up_write(&tsk->namespace->sem);
1138
1139 tsk->namespace = new_ns;
1140
1141 if (rootmnt)
1142 mntput(rootmnt);
1143 if (pwdmnt)
1144 mntput(pwdmnt);
1145 if (altrootmnt)
1146 mntput(altrootmnt);
1147
1148 put_namespace(namespace);
1149 return 0;
1150
1151out:
1152 put_namespace(namespace);
1153 return -ENOMEM;
1154}
1155
1156asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1157 char __user * type, unsigned long flags,
1158 void __user * data)
1159{
1160 int retval;
1161 unsigned long data_page;
1162 unsigned long type_page;
1163 unsigned long dev_page;
1164 char *dir_page;
1165
1166 retval = copy_mount_options (type, &type_page);
1167 if (retval < 0)
1168 return retval;
1169
1170 dir_page = getname(dir_name);
1171 retval = PTR_ERR(dir_page);
1172 if (IS_ERR(dir_page))
1173 goto out1;
1174
1175 retval = copy_mount_options (dev_name, &dev_page);
1176 if (retval < 0)
1177 goto out2;
1178
1179 retval = copy_mount_options (data, &data_page);
1180 if (retval < 0)
1181 goto out3;
1182
1183 lock_kernel();
1184 retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
1185 flags, (void*)data_page);
1186 unlock_kernel();
1187 free_page(data_page);
1188
1189out3:
1190 free_page(dev_page);
1191out2:
1192 putname(dir_page);
1193out1:
1194 free_page(type_page);
1195 return retval;
1196}
1197
1198/*
1199 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1200 * It can block. Requires the big lock held.
1201 */
1202void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1203 struct dentry *dentry)
1204{
1205 struct dentry *old_root;
1206 struct vfsmount *old_rootmnt;
1207 write_lock(&fs->lock);
1208 old_root = fs->root;
1209 old_rootmnt = fs->rootmnt;
1210 fs->rootmnt = mntget(mnt);
1211 fs->root = dget(dentry);
1212 write_unlock(&fs->lock);
1213 if (old_root) {
1214 dput(old_root);
1215 mntput(old_rootmnt);
1216 }
1217}
1218
1219/*
1220 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1221 * It can block. Requires the big lock held.
1222 */
1223void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1224 struct dentry *dentry)
1225{
1226 struct dentry *old_pwd;
1227 struct vfsmount *old_pwdmnt;
1228
1229 write_lock(&fs->lock);
1230 old_pwd = fs->pwd;
1231 old_pwdmnt = fs->pwdmnt;
1232 fs->pwdmnt = mntget(mnt);
1233 fs->pwd = dget(dentry);
1234 write_unlock(&fs->lock);
1235
1236 if (old_pwd) {
1237 dput(old_pwd);
1238 mntput(old_pwdmnt);
1239 }
1240}
1241
1242static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1243{
1244 struct task_struct *g, *p;
1245 struct fs_struct *fs;
1246
1247 read_lock(&tasklist_lock);
1248 do_each_thread(g, p) {
1249 task_lock(p);
1250 fs = p->fs;
1251 if (fs) {
1252 atomic_inc(&fs->count);
1253 task_unlock(p);
1254 if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
1255 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1256 if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
1257 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1258 put_fs_struct(fs);
1259 } else
1260 task_unlock(p);
1261 } while_each_thread(g, p);
1262 read_unlock(&tasklist_lock);
1263}
1264
1265/*
1266 * pivot_root Semantics:
1267 * Moves the root file system of the current process to the directory put_old,
1268 * makes new_root as the new root file system of the current process, and sets
1269 * root/cwd of all processes which had them on the current root to new_root.
1270 *
1271 * Restrictions:
1272 * The new_root and put_old must be directories, and must not be on the
1273 * same file system as the current process root. The put_old must be
1274 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1275 * pointed to by put_old must yield the same directory as new_root. No other
1276 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1277 *
1278 * Notes:
1279 * - we don't move root/cwd if they are not at the root (reason: if something
1280 * cared enough to change them, it's probably wrong to force them elsewhere)
1281 * - it's okay to pick a root that isn't the root of a file system, e.g.
1282 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1283 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1284 * first.
1285 */
1286
1287asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
1288{
1289 struct vfsmount *tmp;
1290 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1291 int error;
1292
1293 if (!capable(CAP_SYS_ADMIN))
1294 return -EPERM;
1295
1296 lock_kernel();
1297
1298 error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
1299 if (error)
1300 goto out0;
1301 error = -EINVAL;
1302 if (!check_mnt(new_nd.mnt))
1303 goto out1;
1304
1305 error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
1306 if (error)
1307 goto out1;
1308
1309 error = security_sb_pivotroot(&old_nd, &new_nd);
1310 if (error) {
1311 path_release(&old_nd);
1312 goto out1;
1313 }
1314
1315 read_lock(&current->fs->lock);
1316 user_nd.mnt = mntget(current->fs->rootmnt);
1317 user_nd.dentry = dget(current->fs->root);
1318 read_unlock(&current->fs->lock);
1319 down_write(&current->namespace->sem);
1320 down(&old_nd.dentry->d_inode->i_sem);
1321 error = -EINVAL;
1322 if (!check_mnt(user_nd.mnt))
1323 goto out2;
1324 error = -ENOENT;
1325 if (IS_DEADDIR(new_nd.dentry->d_inode))
1326 goto out2;
1327 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1328 goto out2;
1329 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1330 goto out2;
1331 error = -EBUSY;
1332 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1333 goto out2; /* loop, on the same file system */
1334 error = -EINVAL;
1335 if (user_nd.mnt->mnt_root != user_nd.dentry)
1336 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001337 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1338 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 if (new_nd.mnt->mnt_root != new_nd.dentry)
1340 goto out2; /* not a mountpoint */
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07001341 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1342 goto out2; /* not attached */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1344 spin_lock(&vfsmount_lock);
1345 if (tmp != new_nd.mnt) {
1346 for (;;) {
1347 if (tmp->mnt_parent == tmp)
1348 goto out3; /* already mounted on put_old */
1349 if (tmp->mnt_parent == new_nd.mnt)
1350 break;
1351 tmp = tmp->mnt_parent;
1352 }
1353 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1354 goto out3;
1355 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1356 goto out3;
1357 detach_mnt(new_nd.mnt, &parent_nd);
1358 detach_mnt(user_nd.mnt, &root_parent);
1359 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1360 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1361 spin_unlock(&vfsmount_lock);
1362 chroot_fs_refs(&user_nd, &new_nd);
1363 security_sb_post_pivotroot(&user_nd, &new_nd);
1364 error = 0;
1365 path_release(&root_parent);
1366 path_release(&parent_nd);
1367out2:
1368 up(&old_nd.dentry->d_inode->i_sem);
1369 up_write(&current->namespace->sem);
1370 path_release(&user_nd);
1371 path_release(&old_nd);
1372out1:
1373 path_release(&new_nd);
1374out0:
1375 unlock_kernel();
1376 return error;
1377out3:
1378 spin_unlock(&vfsmount_lock);
1379 goto out2;
1380}
1381
1382static void __init init_mount_tree(void)
1383{
1384 struct vfsmount *mnt;
1385 struct namespace *namespace;
1386 struct task_struct *g, *p;
1387
1388 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1389 if (IS_ERR(mnt))
1390 panic("Can't create rootfs");
1391 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1392 if (!namespace)
1393 panic("Can't allocate initial namespace");
1394 atomic_set(&namespace->count, 1);
1395 INIT_LIST_HEAD(&namespace->list);
1396 init_rwsem(&namespace->sem);
1397 list_add(&mnt->mnt_list, &namespace->list);
1398 namespace->root = mnt;
1399 mnt->mnt_namespace = namespace;
1400
1401 init_task.namespace = namespace;
1402 read_lock(&tasklist_lock);
1403 do_each_thread(g, p) {
1404 get_namespace(namespace);
1405 p->namespace = namespace;
1406 } while_each_thread(g, p);
1407 read_unlock(&tasklist_lock);
1408
1409 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1410 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1411}
1412
1413void __init mnt_init(unsigned long mempages)
1414{
1415 struct list_head *d;
1416 unsigned int nr_hash;
1417 int i;
1418
1419 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1420 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1421
1422 mount_hashtable = (struct list_head *)
1423 __get_free_page(GFP_ATOMIC);
1424
1425 if (!mount_hashtable)
1426 panic("Failed to allocate mount hash table\n");
1427
1428 /*
1429 * Find the power-of-two list-heads that can fit into the allocation..
1430 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1431 * a power-of-two.
1432 */
1433 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1434 hash_bits = 0;
1435 do {
1436 hash_bits++;
1437 } while ((nr_hash >> hash_bits) != 0);
1438 hash_bits--;
1439
1440 /*
1441 * Re-calculate the actual number of entries and the mask
1442 * from the number of bits we can fit.
1443 */
1444 nr_hash = 1UL << hash_bits;
1445 hash_mask = nr_hash-1;
1446
1447 printk("Mount-cache hash table entries: %d\n", nr_hash);
1448
1449 /* And initialize the newly allocated array */
1450 d = mount_hashtable;
1451 i = nr_hash;
1452 do {
1453 INIT_LIST_HEAD(d);
1454 d++;
1455 i--;
1456 } while (i);
1457 sysfs_init();
1458 init_rootfs();
1459 init_mount_tree();
1460}
1461
1462void __put_namespace(struct namespace *namespace)
1463{
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001464 struct vfsmount *root = namespace->root;
1465 namespace->root = NULL;
1466 spin_unlock(&vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 down_write(&namespace->sem);
1468 spin_lock(&vfsmount_lock);
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07001469 umount_tree(root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 spin_unlock(&vfsmount_lock);
1471 up_write(&namespace->sem);
1472 kfree(namespace);
1473}