4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
32 /* spinlock for vfsmount related operations, inplace of dcache_lock */
33 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
37 static struct list_head
*mount_hashtable __read_mostly
;
38 static int hash_mask __read_mostly
, hash_bits __read_mostly
;
39 static struct kmem_cache
*mnt_cache __read_mostly
;
40 static struct rw_semaphore namespace_sem
;
43 decl_subsys(fs
, NULL
, NULL
);
44 EXPORT_SYMBOL_GPL(fs_subsys
);
46 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
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
;
54 struct vfsmount
*alloc_vfsmnt(const char *name
)
56 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
58 atomic_set(&mnt
->mnt_count
, 1);
59 INIT_LIST_HEAD(&mnt
->mnt_hash
);
60 INIT_LIST_HEAD(&mnt
->mnt_child
);
61 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
62 INIT_LIST_HEAD(&mnt
->mnt_list
);
63 INIT_LIST_HEAD(&mnt
->mnt_expire
);
64 INIT_LIST_HEAD(&mnt
->mnt_share
);
65 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
66 INIT_LIST_HEAD(&mnt
->mnt_slave
);
68 int size
= strlen(name
) + 1;
69 char *newname
= kmalloc(size
, GFP_KERNEL
);
71 memcpy(newname
, name
, size
);
72 mnt
->mnt_devname
= newname
;
79 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
82 mnt
->mnt_root
= dget(sb
->s_root
);
86 EXPORT_SYMBOL(simple_set_mnt
);
88 void free_vfsmnt(struct vfsmount
*mnt
)
90 kfree(mnt
->mnt_devname
);
91 kmem_cache_free(mnt_cache
, mnt
);
95 * find the first or last mount at @dentry on vfsmount @mnt depending on
96 * @dir. If @dir is set return the first mount else return the last mount.
98 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
101 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
102 struct list_head
*tmp
= head
;
103 struct vfsmount
*p
, *found
= NULL
;
106 tmp
= dir
? tmp
->next
: tmp
->prev
;
110 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
111 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
120 * lookup_mnt increments the ref count before returning
121 * the vfsmount struct.
123 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
125 struct vfsmount
*child_mnt
;
126 spin_lock(&vfsmount_lock
);
127 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
129 spin_unlock(&vfsmount_lock
);
133 static inline int check_mnt(struct vfsmount
*mnt
)
135 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
138 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
142 wake_up_interruptible(&ns
->poll
);
146 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
148 if (ns
&& ns
->event
!= event
) {
150 wake_up_interruptible(&ns
->poll
);
154 static void detach_mnt(struct vfsmount
*mnt
, struct nameidata
*old_nd
)
156 old_nd
->dentry
= mnt
->mnt_mountpoint
;
157 old_nd
->mnt
= mnt
->mnt_parent
;
158 mnt
->mnt_parent
= mnt
;
159 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
160 list_del_init(&mnt
->mnt_child
);
161 list_del_init(&mnt
->mnt_hash
);
162 old_nd
->dentry
->d_mounted
--;
165 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
166 struct vfsmount
*child_mnt
)
168 child_mnt
->mnt_parent
= mntget(mnt
);
169 child_mnt
->mnt_mountpoint
= dget(dentry
);
173 static void attach_mnt(struct vfsmount
*mnt
, struct nameidata
*nd
)
175 mnt_set_mountpoint(nd
->mnt
, nd
->dentry
, mnt
);
176 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
177 hash(nd
->mnt
, nd
->dentry
));
178 list_add_tail(&mnt
->mnt_child
, &nd
->mnt
->mnt_mounts
);
182 * the caller must hold vfsmount_lock
184 static void commit_tree(struct vfsmount
*mnt
)
186 struct vfsmount
*parent
= mnt
->mnt_parent
;
189 struct mnt_namespace
*n
= parent
->mnt_ns
;
191 BUG_ON(parent
== mnt
);
193 list_add_tail(&head
, &mnt
->mnt_list
);
194 list_for_each_entry(m
, &head
, mnt_list
)
196 list_splice(&head
, n
->list
.prev
);
198 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
199 hash(parent
, mnt
->mnt_mountpoint
));
200 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
201 touch_mnt_namespace(n
);
204 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
206 struct list_head
*next
= p
->mnt_mounts
.next
;
207 if (next
== &p
->mnt_mounts
) {
211 next
= p
->mnt_child
.next
;
212 if (next
!= &p
->mnt_parent
->mnt_mounts
)
217 return list_entry(next
, struct vfsmount
, mnt_child
);
220 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
222 struct list_head
*prev
= p
->mnt_mounts
.prev
;
223 while (prev
!= &p
->mnt_mounts
) {
224 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
225 prev
= p
->mnt_mounts
.prev
;
230 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
233 struct super_block
*sb
= old
->mnt_sb
;
234 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
237 mnt
->mnt_flags
= old
->mnt_flags
;
238 atomic_inc(&sb
->s_active
);
240 mnt
->mnt_root
= dget(root
);
241 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
242 mnt
->mnt_parent
= mnt
;
244 if (flag
& CL_SLAVE
) {
245 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
246 mnt
->mnt_master
= old
;
247 CLEAR_MNT_SHARED(mnt
);
249 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
250 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
251 if (IS_MNT_SLAVE(old
))
252 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
253 mnt
->mnt_master
= old
->mnt_master
;
255 if (flag
& CL_MAKE_SHARED
)
258 /* stick the duplicate mount on the same expiry list
259 * as the original if that was on one */
260 if (flag
& CL_EXPIRE
) {
261 spin_lock(&vfsmount_lock
);
262 if (!list_empty(&old
->mnt_expire
))
263 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
264 spin_unlock(&vfsmount_lock
);
270 static inline void __mntput(struct vfsmount
*mnt
)
272 struct super_block
*sb
= mnt
->mnt_sb
;
275 deactivate_super(sb
);
278 void mntput_no_expire(struct vfsmount
*mnt
)
281 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
282 if (likely(!mnt
->mnt_pinned
)) {
283 spin_unlock(&vfsmount_lock
);
287 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
289 spin_unlock(&vfsmount_lock
);
290 acct_auto_close_mnt(mnt
);
291 security_sb_umount_close(mnt
);
296 EXPORT_SYMBOL(mntput_no_expire
);
298 void mnt_pin(struct vfsmount
*mnt
)
300 spin_lock(&vfsmount_lock
);
302 spin_unlock(&vfsmount_lock
);
305 EXPORT_SYMBOL(mnt_pin
);
307 void mnt_unpin(struct vfsmount
*mnt
)
309 spin_lock(&vfsmount_lock
);
310 if (mnt
->mnt_pinned
) {
311 atomic_inc(&mnt
->mnt_count
);
314 spin_unlock(&vfsmount_lock
);
317 EXPORT_SYMBOL(mnt_unpin
);
320 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
322 struct mnt_namespace
*n
= m
->private;
324 down_read(&namespace_sem
);
325 return seq_list_start(&n
->list
, *pos
);
328 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
330 struct mnt_namespace
*n
= m
->private;
332 return seq_list_next(v
, &n
->list
, pos
);
335 static void m_stop(struct seq_file
*m
, void *v
)
337 up_read(&namespace_sem
);
340 static inline void mangle(struct seq_file
*m
, const char *s
)
342 seq_escape(m
, s
, " \t\n\\");
345 static int show_vfsmnt(struct seq_file
*m
, void *v
)
347 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
349 static struct proc_fs_info
{
353 { MS_SYNCHRONOUS
, ",sync" },
354 { MS_DIRSYNC
, ",dirsync" },
355 { MS_MANDLOCK
, ",mand" },
358 static struct proc_fs_info mnt_info
[] = {
359 { MNT_NOSUID
, ",nosuid" },
360 { MNT_NODEV
, ",nodev" },
361 { MNT_NOEXEC
, ",noexec" },
362 { MNT_NOATIME
, ",noatime" },
363 { MNT_NODIRATIME
, ",nodiratime" },
364 { MNT_RELATIME
, ",relatime" },
367 struct proc_fs_info
*fs_infop
;
369 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
371 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
373 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
374 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
376 mangle(m
, mnt
->mnt_sb
->s_subtype
);
378 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
379 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
380 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
381 seq_puts(m
, fs_infop
->str
);
383 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
384 if (mnt
->mnt_flags
& fs_infop
->flag
)
385 seq_puts(m
, fs_infop
->str
);
387 if (mnt
->mnt_sb
->s_op
->show_options
)
388 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
389 seq_puts(m
, " 0 0\n");
393 struct seq_operations mounts_op
= {
400 static int show_vfsstat(struct seq_file
*m
, void *v
)
402 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
406 if (mnt
->mnt_devname
) {
407 seq_puts(m
, "device ");
408 mangle(m
, mnt
->mnt_devname
);
410 seq_puts(m
, "no device");
413 seq_puts(m
, " mounted on ");
414 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
417 /* file system type */
418 seq_puts(m
, "with fstype ");
419 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
421 /* optional statistics */
422 if (mnt
->mnt_sb
->s_op
->show_stats
) {
424 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
431 struct seq_operations mountstats_op
= {
435 .show
= show_vfsstat
,
439 * may_umount_tree - check if a mount tree is busy
440 * @mnt: root of mount tree
442 * This is called to check if a tree of mounts has any
443 * open files, pwds, chroots or sub mounts that are
446 int may_umount_tree(struct vfsmount
*mnt
)
449 int minimum_refs
= 0;
452 spin_lock(&vfsmount_lock
);
453 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
454 actual_refs
+= atomic_read(&p
->mnt_count
);
457 spin_unlock(&vfsmount_lock
);
459 if (actual_refs
> minimum_refs
)
465 EXPORT_SYMBOL(may_umount_tree
);
468 * may_umount - check if a mount point is busy
469 * @mnt: root of mount
471 * This is called to check if a mount point has any
472 * open files, pwds, chroots or sub mounts. If the
473 * mount has sub mounts this will return busy
474 * regardless of whether the sub mounts are busy.
476 * Doesn't take quota and stuff into account. IOW, in some cases it will
477 * give false negatives. The main reason why it's here is that we need
478 * a non-destructive way to look for easily umountable filesystems.
480 int may_umount(struct vfsmount
*mnt
)
483 spin_lock(&vfsmount_lock
);
484 if (propagate_mount_busy(mnt
, 2))
486 spin_unlock(&vfsmount_lock
);
490 EXPORT_SYMBOL(may_umount
);
492 void release_mounts(struct list_head
*head
)
494 struct vfsmount
*mnt
;
495 while (!list_empty(head
)) {
496 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
497 list_del_init(&mnt
->mnt_hash
);
498 if (mnt
->mnt_parent
!= mnt
) {
499 struct dentry
*dentry
;
501 spin_lock(&vfsmount_lock
);
502 dentry
= mnt
->mnt_mountpoint
;
504 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
505 mnt
->mnt_parent
= mnt
;
506 spin_unlock(&vfsmount_lock
);
514 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
518 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
519 list_move(&p
->mnt_hash
, kill
);
522 propagate_umount(kill
);
524 list_for_each_entry(p
, kill
, mnt_hash
) {
525 list_del_init(&p
->mnt_expire
);
526 list_del_init(&p
->mnt_list
);
527 __touch_mnt_namespace(p
->mnt_ns
);
529 list_del_init(&p
->mnt_child
);
530 if (p
->mnt_parent
!= p
)
531 p
->mnt_mountpoint
->d_mounted
--;
532 change_mnt_propagation(p
, MS_PRIVATE
);
536 static int do_umount(struct vfsmount
*mnt
, int flags
)
538 struct super_block
*sb
= mnt
->mnt_sb
;
540 LIST_HEAD(umount_list
);
542 retval
= security_sb_umount(mnt
, flags
);
547 * Allow userspace to request a mountpoint be expired rather than
548 * unmounting unconditionally. Unmount only happens if:
549 * (1) the mark is already set (the mark is cleared by mntput())
550 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
552 if (flags
& MNT_EXPIRE
) {
553 if (mnt
== current
->fs
->rootmnt
||
554 flags
& (MNT_FORCE
| MNT_DETACH
))
557 if (atomic_read(&mnt
->mnt_count
) != 2)
560 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
565 * If we may have to abort operations to get out of this
566 * mount, and they will themselves hold resources we must
567 * allow the fs to do things. In the Unix tradition of
568 * 'Gee thats tricky lets do it in userspace' the umount_begin
569 * might fail to complete on the first run through as other tasks
570 * must return, and the like. Thats for the mount program to worry
571 * about for the moment.
575 if (sb
->s_op
->umount_begin
)
576 sb
->s_op
->umount_begin(mnt
, flags
);
580 * No sense to grab the lock for this test, but test itself looks
581 * somewhat bogus. Suggestions for better replacement?
582 * Ho-hum... In principle, we might treat that as umount + switch
583 * to rootfs. GC would eventually take care of the old vfsmount.
584 * Actually it makes sense, especially if rootfs would contain a
585 * /reboot - static binary that would close all descriptors and
586 * call reboot(9). Then init(8) could umount root and exec /reboot.
588 if (mnt
== current
->fs
->rootmnt
&& !(flags
& MNT_DETACH
)) {
590 * Special case for "unmounting" root ...
591 * we just try to remount it readonly.
593 down_write(&sb
->s_umount
);
594 if (!(sb
->s_flags
& MS_RDONLY
)) {
597 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
600 up_write(&sb
->s_umount
);
604 down_write(&namespace_sem
);
605 spin_lock(&vfsmount_lock
);
609 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
610 if (!list_empty(&mnt
->mnt_list
))
611 umount_tree(mnt
, 1, &umount_list
);
614 spin_unlock(&vfsmount_lock
);
616 security_sb_umount_busy(mnt
);
617 up_write(&namespace_sem
);
618 release_mounts(&umount_list
);
623 * Now umount can handle mount points as well as block devices.
624 * This is important for filesystems which use unnamed block devices.
626 * We now support a flag for forced unmount like the other 'big iron'
627 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
630 asmlinkage
long sys_umount(char __user
* name
, int flags
)
635 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
639 if (nd
.dentry
!= nd
.mnt
->mnt_root
)
641 if (!check_mnt(nd
.mnt
))
645 if (!capable(CAP_SYS_ADMIN
))
648 retval
= do_umount(nd
.mnt
, flags
);
650 path_release_on_umount(&nd
);
655 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
658 * The 2.0 compatible umount. No flags.
660 asmlinkage
long sys_oldumount(char __user
* name
)
662 return sys_umount(name
, 0);
667 static int mount_is_safe(struct nameidata
*nd
)
669 if (capable(CAP_SYS_ADMIN
))
673 if (S_ISLNK(nd
->dentry
->d_inode
->i_mode
))
675 if (nd
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
676 if (current
->uid
!= nd
->dentry
->d_inode
->i_uid
)
679 if (vfs_permission(nd
, MAY_WRITE
))
685 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
690 if (d
== NULL
|| d
== d
->d_parent
)
696 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
699 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
702 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
705 res
= q
= clone_mnt(mnt
, dentry
, flag
);
708 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
711 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
712 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
715 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
716 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
717 s
= skip_mnt_tree(s
);
720 while (p
!= s
->mnt_parent
) {
726 nd
.dentry
= p
->mnt_mountpoint
;
727 q
= clone_mnt(p
, p
->mnt_root
, flag
);
730 spin_lock(&vfsmount_lock
);
731 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
733 spin_unlock(&vfsmount_lock
);
739 LIST_HEAD(umount_list
);
740 spin_lock(&vfsmount_lock
);
741 umount_tree(res
, 0, &umount_list
);
742 spin_unlock(&vfsmount_lock
);
743 release_mounts(&umount_list
);
749 * @source_mnt : mount tree to be attached
750 * @nd : place the mount tree @source_mnt is attached
751 * @parent_nd : if non-null, detach the source_mnt from its parent and
752 * store the parent mount and mountpoint dentry.
753 * (done when source_mnt is moved)
755 * NOTE: in the table below explains the semantics when a source mount
756 * of a given type is attached to a destination mount of a given type.
757 * ---------------------------------------------------------------------------
758 * | BIND MOUNT OPERATION |
759 * |**************************************************************************
760 * | source-->| shared | private | slave | unbindable |
764 * |**************************************************************************
765 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
767 * |non-shared| shared (+) | private | slave (*) | invalid |
768 * ***************************************************************************
769 * A bind operation clones the source mount and mounts the clone on the
772 * (++) the cloned mount is propagated to all the mounts in the propagation
773 * tree of the destination mount and the cloned mount is added to
774 * the peer group of the source mount.
775 * (+) the cloned mount is created under the destination mount and is marked
776 * as shared. The cloned mount is added to the peer group of the source
778 * (+++) the mount is propagated to all the mounts in the propagation tree
779 * of the destination mount and the cloned mount is made slave
780 * of the same master as that of the source mount. The cloned mount
781 * is marked as 'shared and slave'.
782 * (*) the cloned mount is made a slave of the same master as that of the
785 * ---------------------------------------------------------------------------
786 * | MOVE MOUNT OPERATION |
787 * |**************************************************************************
788 * | source-->| shared | private | slave | unbindable |
792 * |**************************************************************************
793 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
795 * |non-shared| shared (+*) | private | slave (*) | unbindable |
796 * ***************************************************************************
798 * (+) the mount is moved to the destination. And is then propagated to
799 * all the mounts in the propagation tree of the destination mount.
800 * (+*) the mount is moved to the destination.
801 * (+++) the mount is moved to the destination and is then propagated to
802 * all the mounts belonging to the destination mount's propagation tree.
803 * the mount is marked as 'shared and slave'.
804 * (*) the mount continues to be a slave at the new location.
806 * if the source mount is a tree, the operations explained above is
807 * applied to each mount in the tree.
808 * Must be called without spinlocks held, since this function can sleep
811 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
812 struct nameidata
*nd
, struct nameidata
*parent_nd
)
814 LIST_HEAD(tree_list
);
815 struct vfsmount
*dest_mnt
= nd
->mnt
;
816 struct dentry
*dest_dentry
= nd
->dentry
;
817 struct vfsmount
*child
, *p
;
819 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
822 if (IS_MNT_SHARED(dest_mnt
)) {
823 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
827 spin_lock(&vfsmount_lock
);
829 detach_mnt(source_mnt
, parent_nd
);
830 attach_mnt(source_mnt
, nd
);
831 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
833 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
834 commit_tree(source_mnt
);
837 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
838 list_del_init(&child
->mnt_hash
);
841 spin_unlock(&vfsmount_lock
);
845 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
848 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
851 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
852 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
856 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
857 if (IS_DEADDIR(nd
->dentry
->d_inode
))
860 err
= security_sb_check_sb(mnt
, nd
);
865 if (IS_ROOT(nd
->dentry
) || !d_unhashed(nd
->dentry
))
866 err
= attach_recursive_mnt(mnt
, nd
, NULL
);
868 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
870 security_sb_post_addmount(mnt
, nd
);
875 * recursively change the type of the mountpoint.
877 static int do_change_type(struct nameidata
*nd
, int flag
)
879 struct vfsmount
*m
, *mnt
= nd
->mnt
;
880 int recurse
= flag
& MS_REC
;
881 int type
= flag
& ~MS_REC
;
883 if (!capable(CAP_SYS_ADMIN
))
886 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
889 down_write(&namespace_sem
);
890 spin_lock(&vfsmount_lock
);
891 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
892 change_mnt_propagation(m
, type
);
893 spin_unlock(&vfsmount_lock
);
894 up_write(&namespace_sem
);
901 static int do_loopback(struct nameidata
*nd
, char *old_name
, int recurse
)
903 struct nameidata old_nd
;
904 struct vfsmount
*mnt
= NULL
;
905 int err
= mount_is_safe(nd
);
908 if (!old_name
|| !*old_name
)
910 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
914 down_write(&namespace_sem
);
916 if (IS_MNT_UNBINDABLE(old_nd
.mnt
))
919 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
924 mnt
= copy_tree(old_nd
.mnt
, old_nd
.dentry
, 0);
926 mnt
= clone_mnt(old_nd
.mnt
, old_nd
.dentry
, 0);
931 err
= graft_tree(mnt
, nd
);
933 LIST_HEAD(umount_list
);
934 spin_lock(&vfsmount_lock
);
935 umount_tree(mnt
, 0, &umount_list
);
936 spin_unlock(&vfsmount_lock
);
937 release_mounts(&umount_list
);
941 up_write(&namespace_sem
);
942 path_release(&old_nd
);
947 * change filesystem flags. dir should be a physical root of filesystem.
948 * If you've mounted a non-root directory somewhere and want to do remount
949 * on it - tough luck.
951 static int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
955 struct super_block
*sb
= nd
->mnt
->mnt_sb
;
957 if (!capable(CAP_SYS_ADMIN
))
960 if (!check_mnt(nd
->mnt
))
963 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
966 down_write(&sb
->s_umount
);
967 err
= do_remount_sb(sb
, flags
, data
, 0);
969 nd
->mnt
->mnt_flags
= mnt_flags
;
970 up_write(&sb
->s_umount
);
972 security_sb_post_remount(nd
->mnt
, flags
, data
);
976 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
979 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
980 if (IS_MNT_UNBINDABLE(p
))
986 static int do_move_mount(struct nameidata
*nd
, char *old_name
)
988 struct nameidata old_nd
, parent_nd
;
991 if (!capable(CAP_SYS_ADMIN
))
993 if (!old_name
|| !*old_name
)
995 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
999 down_write(&namespace_sem
);
1000 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1003 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
1007 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
1008 if (IS_DEADDIR(nd
->dentry
->d_inode
))
1011 if (!IS_ROOT(nd
->dentry
) && d_unhashed(nd
->dentry
))
1015 if (old_nd
.dentry
!= old_nd
.mnt
->mnt_root
)
1018 if (old_nd
.mnt
== old_nd
.mnt
->mnt_parent
)
1021 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
1022 S_ISDIR(old_nd
.dentry
->d_inode
->i_mode
))
1025 * Don't move a mount residing in a shared parent.
1027 if (old_nd
.mnt
->mnt_parent
&& IS_MNT_SHARED(old_nd
.mnt
->mnt_parent
))
1030 * Don't move a mount tree containing unbindable mounts to a destination
1031 * mount which is shared.
1033 if (IS_MNT_SHARED(nd
->mnt
) && tree_contains_unbindable(old_nd
.mnt
))
1036 for (p
= nd
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1037 if (p
== old_nd
.mnt
)
1040 if ((err
= attach_recursive_mnt(old_nd
.mnt
, nd
, &parent_nd
)))
1043 spin_lock(&vfsmount_lock
);
1044 /* if the mount is moved, it should no longer be expire
1046 list_del_init(&old_nd
.mnt
->mnt_expire
);
1047 spin_unlock(&vfsmount_lock
);
1049 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
1051 up_write(&namespace_sem
);
1053 path_release(&parent_nd
);
1054 path_release(&old_nd
);
1059 * create a new mount for userspace and request it to be added into the
1062 static int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1063 int mnt_flags
, char *name
, void *data
)
1065 struct vfsmount
*mnt
;
1067 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1070 /* we need capabilities... */
1071 if (!capable(CAP_SYS_ADMIN
))
1074 mnt
= do_kern_mount(type
, flags
, name
, data
);
1076 return PTR_ERR(mnt
);
1078 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1082 * add a mount into a namespace's mount tree
1083 * - provide the option of adding the new mount to an expiration list
1085 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1086 int mnt_flags
, struct list_head
*fslist
)
1090 down_write(&namespace_sem
);
1091 /* Something was mounted here while we slept */
1092 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1095 if (!check_mnt(nd
->mnt
))
1098 /* Refuse the same filesystem on the same mount point */
1100 if (nd
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1101 nd
->mnt
->mnt_root
== nd
->dentry
)
1105 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1108 newmnt
->mnt_flags
= mnt_flags
;
1109 if ((err
= graft_tree(newmnt
, nd
)))
1113 /* add to the specified expiration list */
1114 spin_lock(&vfsmount_lock
);
1115 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1116 spin_unlock(&vfsmount_lock
);
1118 up_write(&namespace_sem
);
1122 up_write(&namespace_sem
);
1127 EXPORT_SYMBOL_GPL(do_add_mount
);
1129 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
1130 struct list_head
*umounts
)
1132 spin_lock(&vfsmount_lock
);
1135 * Check if mount is still attached, if not, let whoever holds it deal
1138 if (mnt
->mnt_parent
== mnt
) {
1139 spin_unlock(&vfsmount_lock
);
1144 * Check that it is still dead: the count should now be 2 - as
1145 * contributed by the vfsmount parent and the mntget above
1147 if (!propagate_mount_busy(mnt
, 2)) {
1148 /* delete from the namespace */
1149 touch_mnt_namespace(mnt
->mnt_ns
);
1150 list_del_init(&mnt
->mnt_list
);
1152 umount_tree(mnt
, 1, umounts
);
1153 spin_unlock(&vfsmount_lock
);
1156 * Someone brought it back to life whilst we didn't have any
1157 * locks held so return it to the expiration list
1159 list_add_tail(&mnt
->mnt_expire
, mounts
);
1160 spin_unlock(&vfsmount_lock
);
1165 * go through the vfsmounts we've just consigned to the graveyard to
1166 * - check that they're still dead
1167 * - delete the vfsmount from the appropriate namespace under lock
1168 * - dispose of the corpse
1170 static void expire_mount_list(struct list_head
*graveyard
, struct list_head
*mounts
)
1172 struct mnt_namespace
*ns
;
1173 struct vfsmount
*mnt
;
1175 while (!list_empty(graveyard
)) {
1177 mnt
= list_first_entry(graveyard
, struct vfsmount
, mnt_expire
);
1178 list_del_init(&mnt
->mnt_expire
);
1180 /* don't do anything if the namespace is dead - all the
1181 * vfsmounts from it are going away anyway */
1183 if (!ns
|| !ns
->root
)
1187 spin_unlock(&vfsmount_lock
);
1188 down_write(&namespace_sem
);
1189 expire_mount(mnt
, mounts
, &umounts
);
1190 up_write(&namespace_sem
);
1191 release_mounts(&umounts
);
1194 spin_lock(&vfsmount_lock
);
1199 * process a list of expirable mountpoints with the intent of discarding any
1200 * mountpoints that aren't in use and haven't been touched since last we came
1203 void mark_mounts_for_expiry(struct list_head
*mounts
)
1205 struct vfsmount
*mnt
, *next
;
1206 LIST_HEAD(graveyard
);
1208 if (list_empty(mounts
))
1211 spin_lock(&vfsmount_lock
);
1213 /* extract from the expiration list every vfsmount that matches the
1214 * following criteria:
1215 * - only referenced by its parent vfsmount
1216 * - still marked for expiry (marked on the last call here; marks are
1217 * cleared by mntput())
1219 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1220 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1221 atomic_read(&mnt
->mnt_count
) != 1)
1225 list_move(&mnt
->mnt_expire
, &graveyard
);
1228 expire_mount_list(&graveyard
, mounts
);
1230 spin_unlock(&vfsmount_lock
);
1233 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1236 * Ripoff of 'select_parent()'
1238 * search the list of submounts for a given mountpoint, and move any
1239 * shrinkable submounts to the 'graveyard' list.
1241 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1243 struct vfsmount
*this_parent
= parent
;
1244 struct list_head
*next
;
1248 next
= this_parent
->mnt_mounts
.next
;
1250 while (next
!= &this_parent
->mnt_mounts
) {
1251 struct list_head
*tmp
= next
;
1252 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1255 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1258 * Descend a level if the d_mounts list is non-empty.
1260 if (!list_empty(&mnt
->mnt_mounts
)) {
1265 if (!propagate_mount_busy(mnt
, 1)) {
1267 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1272 * All done at this level ... ascend and resume the search
1274 if (this_parent
!= parent
) {
1275 next
= this_parent
->mnt_child
.next
;
1276 this_parent
= this_parent
->mnt_parent
;
1283 * process a list of expirable mountpoints with the intent of discarding any
1284 * submounts of a specific parent mountpoint
1286 void shrink_submounts(struct vfsmount
*mountpoint
, struct list_head
*mounts
)
1288 LIST_HEAD(graveyard
);
1291 spin_lock(&vfsmount_lock
);
1293 /* extract submounts of 'mountpoint' from the expiration list */
1294 while ((found
= select_submounts(mountpoint
, &graveyard
)) != 0)
1295 expire_mount_list(&graveyard
, mounts
);
1297 spin_unlock(&vfsmount_lock
);
1300 EXPORT_SYMBOL_GPL(shrink_submounts
);
1303 * Some copy_from_user() implementations do not return the exact number of
1304 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1305 * Note that this function differs from copy_from_user() in that it will oops
1306 * on bad values of `to', rather than returning a short copy.
1308 static long exact_copy_from_user(void *to
, const void __user
* from
,
1312 const char __user
*f
= from
;
1315 if (!access_ok(VERIFY_READ
, from
, n
))
1319 if (__get_user(c
, f
)) {
1330 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1340 if (!(page
= __get_free_page(GFP_KERNEL
)))
1343 /* We only care that *some* data at the address the user
1344 * gave us is valid. Just in case, we'll zero
1345 * the remainder of the page.
1347 /* copy_from_user cannot cross TASK_SIZE ! */
1348 size
= TASK_SIZE
- (unsigned long)data
;
1349 if (size
> PAGE_SIZE
)
1352 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1358 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1364 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1365 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1367 * data is a (void *) that can point to any structure up to
1368 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1369 * information (or be NULL).
1371 * Pre-0.97 versions of mount() didn't have a flags word.
1372 * When the flags word was introduced its top half was required
1373 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1374 * Therefore, if this magic number is present, it carries no information
1375 * and must be discarded.
1377 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1378 unsigned long flags
, void *data_page
)
1380 struct nameidata nd
;
1385 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1386 flags
&= ~MS_MGC_MSK
;
1388 /* Basic sanity checks */
1390 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1392 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1396 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1398 /* Separate the per-mountpoint flags */
1399 if (flags
& MS_NOSUID
)
1400 mnt_flags
|= MNT_NOSUID
;
1401 if (flags
& MS_NODEV
)
1402 mnt_flags
|= MNT_NODEV
;
1403 if (flags
& MS_NOEXEC
)
1404 mnt_flags
|= MNT_NOEXEC
;
1405 if (flags
& MS_NOATIME
)
1406 mnt_flags
|= MNT_NOATIME
;
1407 if (flags
& MS_NODIRATIME
)
1408 mnt_flags
|= MNT_NODIRATIME
;
1409 if (flags
& MS_RELATIME
)
1410 mnt_flags
|= MNT_RELATIME
;
1412 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1413 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
);
1415 /* ... and get the mountpoint */
1416 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1420 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1424 if (flags
& MS_REMOUNT
)
1425 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1427 else if (flags
& MS_BIND
)
1428 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1429 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1430 retval
= do_change_type(&nd
, flags
);
1431 else if (flags
& MS_MOVE
)
1432 retval
= do_move_mount(&nd
, dev_name
);
1434 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1435 dev_name
, data_page
);
1442 * Allocate a new namespace structure and populate it with contents
1443 * copied from the namespace of the passed in task structure.
1445 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1446 struct fs_struct
*fs
)
1448 struct mnt_namespace
*new_ns
;
1449 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1450 struct vfsmount
*p
, *q
;
1452 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1454 return ERR_PTR(-ENOMEM
);
1456 atomic_set(&new_ns
->count
, 1);
1457 INIT_LIST_HEAD(&new_ns
->list
);
1458 init_waitqueue_head(&new_ns
->poll
);
1461 down_write(&namespace_sem
);
1462 /* First pass: copy the tree topology */
1463 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1464 CL_COPY_ALL
| CL_EXPIRE
);
1465 if (!new_ns
->root
) {
1466 up_write(&namespace_sem
);
1468 return ERR_PTR(-ENOMEM
);;
1470 spin_lock(&vfsmount_lock
);
1471 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1472 spin_unlock(&vfsmount_lock
);
1475 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1476 * as belonging to new namespace. We have already acquired a private
1477 * fs_struct, so tsk->fs->lock is not needed.
1484 if (p
== fs
->rootmnt
) {
1486 fs
->rootmnt
= mntget(q
);
1488 if (p
== fs
->pwdmnt
) {
1490 fs
->pwdmnt
= mntget(q
);
1492 if (p
== fs
->altrootmnt
) {
1494 fs
->altrootmnt
= mntget(q
);
1497 p
= next_mnt(p
, mnt_ns
->root
);
1498 q
= next_mnt(q
, new_ns
->root
);
1500 up_write(&namespace_sem
);
1512 struct mnt_namespace
*copy_mnt_ns(int flags
, struct mnt_namespace
*ns
,
1513 struct fs_struct
*new_fs
)
1515 struct mnt_namespace
*new_ns
;
1520 if (!(flags
& CLONE_NEWNS
))
1523 new_ns
= dup_mnt_ns(ns
, new_fs
);
1529 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1530 char __user
* type
, unsigned long flags
,
1534 unsigned long data_page
;
1535 unsigned long type_page
;
1536 unsigned long dev_page
;
1539 retval
= copy_mount_options(type
, &type_page
);
1543 dir_page
= getname(dir_name
);
1544 retval
= PTR_ERR(dir_page
);
1545 if (IS_ERR(dir_page
))
1548 retval
= copy_mount_options(dev_name
, &dev_page
);
1552 retval
= copy_mount_options(data
, &data_page
);
1557 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1558 flags
, (void *)data_page
);
1560 free_page(data_page
);
1563 free_page(dev_page
);
1567 free_page(type_page
);
1572 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1573 * It can block. Requires the big lock held.
1575 void set_fs_root(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1576 struct dentry
*dentry
)
1578 struct dentry
*old_root
;
1579 struct vfsmount
*old_rootmnt
;
1580 write_lock(&fs
->lock
);
1581 old_root
= fs
->root
;
1582 old_rootmnt
= fs
->rootmnt
;
1583 fs
->rootmnt
= mntget(mnt
);
1584 fs
->root
= dget(dentry
);
1585 write_unlock(&fs
->lock
);
1588 mntput(old_rootmnt
);
1593 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1594 * It can block. Requires the big lock held.
1596 void set_fs_pwd(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1597 struct dentry
*dentry
)
1599 struct dentry
*old_pwd
;
1600 struct vfsmount
*old_pwdmnt
;
1602 write_lock(&fs
->lock
);
1604 old_pwdmnt
= fs
->pwdmnt
;
1605 fs
->pwdmnt
= mntget(mnt
);
1606 fs
->pwd
= dget(dentry
);
1607 write_unlock(&fs
->lock
);
1615 static void chroot_fs_refs(struct nameidata
*old_nd
, struct nameidata
*new_nd
)
1617 struct task_struct
*g
, *p
;
1618 struct fs_struct
*fs
;
1620 read_lock(&tasklist_lock
);
1621 do_each_thread(g
, p
) {
1625 atomic_inc(&fs
->count
);
1627 if (fs
->root
== old_nd
->dentry
1628 && fs
->rootmnt
== old_nd
->mnt
)
1629 set_fs_root(fs
, new_nd
->mnt
, new_nd
->dentry
);
1630 if (fs
->pwd
== old_nd
->dentry
1631 && fs
->pwdmnt
== old_nd
->mnt
)
1632 set_fs_pwd(fs
, new_nd
->mnt
, new_nd
->dentry
);
1636 } while_each_thread(g
, p
);
1637 read_unlock(&tasklist_lock
);
1641 * pivot_root Semantics:
1642 * Moves the root file system of the current process to the directory put_old,
1643 * makes new_root as the new root file system of the current process, and sets
1644 * root/cwd of all processes which had them on the current root to new_root.
1647 * The new_root and put_old must be directories, and must not be on the
1648 * same file system as the current process root. The put_old must be
1649 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1650 * pointed to by put_old must yield the same directory as new_root. No other
1651 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1653 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1654 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1655 * in this situation.
1658 * - we don't move root/cwd if they are not at the root (reason: if something
1659 * cared enough to change them, it's probably wrong to force them elsewhere)
1660 * - it's okay to pick a root that isn't the root of a file system, e.g.
1661 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1662 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1665 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1666 const char __user
* put_old
)
1668 struct vfsmount
*tmp
;
1669 struct nameidata new_nd
, old_nd
, parent_nd
, root_parent
, user_nd
;
1672 if (!capable(CAP_SYS_ADMIN
))
1677 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1682 if (!check_mnt(new_nd
.mnt
))
1685 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1689 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1691 path_release(&old_nd
);
1695 read_lock(¤t
->fs
->lock
);
1696 user_nd
.mnt
= mntget(current
->fs
->rootmnt
);
1697 user_nd
.dentry
= dget(current
->fs
->root
);
1698 read_unlock(¤t
->fs
->lock
);
1699 down_write(&namespace_sem
);
1700 mutex_lock(&old_nd
.dentry
->d_inode
->i_mutex
);
1702 if (IS_MNT_SHARED(old_nd
.mnt
) ||
1703 IS_MNT_SHARED(new_nd
.mnt
->mnt_parent
) ||
1704 IS_MNT_SHARED(user_nd
.mnt
->mnt_parent
))
1706 if (!check_mnt(user_nd
.mnt
))
1709 if (IS_DEADDIR(new_nd
.dentry
->d_inode
))
1711 if (d_unhashed(new_nd
.dentry
) && !IS_ROOT(new_nd
.dentry
))
1713 if (d_unhashed(old_nd
.dentry
) && !IS_ROOT(old_nd
.dentry
))
1716 if (new_nd
.mnt
== user_nd
.mnt
|| old_nd
.mnt
== user_nd
.mnt
)
1717 goto out2
; /* loop, on the same file system */
1719 if (user_nd
.mnt
->mnt_root
!= user_nd
.dentry
)
1720 goto out2
; /* not a mountpoint */
1721 if (user_nd
.mnt
->mnt_parent
== user_nd
.mnt
)
1722 goto out2
; /* not attached */
1723 if (new_nd
.mnt
->mnt_root
!= new_nd
.dentry
)
1724 goto out2
; /* not a mountpoint */
1725 if (new_nd
.mnt
->mnt_parent
== new_nd
.mnt
)
1726 goto out2
; /* not attached */
1727 tmp
= old_nd
.mnt
; /* make sure we can reach put_old from new_root */
1728 spin_lock(&vfsmount_lock
);
1729 if (tmp
!= new_nd
.mnt
) {
1731 if (tmp
->mnt_parent
== tmp
)
1732 goto out3
; /* already mounted on put_old */
1733 if (tmp
->mnt_parent
== new_nd
.mnt
)
1735 tmp
= tmp
->mnt_parent
;
1737 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.dentry
))
1739 } else if (!is_subdir(old_nd
.dentry
, new_nd
.dentry
))
1741 detach_mnt(new_nd
.mnt
, &parent_nd
);
1742 detach_mnt(user_nd
.mnt
, &root_parent
);
1743 attach_mnt(user_nd
.mnt
, &old_nd
); /* mount old root on put_old */
1744 attach_mnt(new_nd
.mnt
, &root_parent
); /* mount new_root on / */
1745 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1746 spin_unlock(&vfsmount_lock
);
1747 chroot_fs_refs(&user_nd
, &new_nd
);
1748 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1750 path_release(&root_parent
);
1751 path_release(&parent_nd
);
1753 mutex_unlock(&old_nd
.dentry
->d_inode
->i_mutex
);
1754 up_write(&namespace_sem
);
1755 path_release(&user_nd
);
1756 path_release(&old_nd
);
1758 path_release(&new_nd
);
1763 spin_unlock(&vfsmount_lock
);
1767 static void __init
init_mount_tree(void)
1769 struct vfsmount
*mnt
;
1770 struct mnt_namespace
*ns
;
1772 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1774 panic("Can't create rootfs");
1775 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1777 panic("Can't allocate initial namespace");
1778 atomic_set(&ns
->count
, 1);
1779 INIT_LIST_HEAD(&ns
->list
);
1780 init_waitqueue_head(&ns
->poll
);
1782 list_add(&mnt
->mnt_list
, &ns
->list
);
1786 init_task
.nsproxy
->mnt_ns
= ns
;
1789 set_fs_pwd(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1790 set_fs_root(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1793 void __init
mnt_init(unsigned long mempages
)
1795 struct list_head
*d
;
1796 unsigned int nr_hash
;
1800 init_rwsem(&namespace_sem
);
1802 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1803 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
, NULL
);
1805 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1807 if (!mount_hashtable
)
1808 panic("Failed to allocate mount hash table\n");
1811 * Find the power-of-two list-heads that can fit into the allocation..
1812 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1815 nr_hash
= PAGE_SIZE
/ sizeof(struct list_head
);
1819 } while ((nr_hash
>> hash_bits
) != 0);
1823 * Re-calculate the actual number of entries and the mask
1824 * from the number of bits we can fit.
1826 nr_hash
= 1UL << hash_bits
;
1827 hash_mask
= nr_hash
- 1;
1829 printk("Mount-cache hash table entries: %d\n", nr_hash
);
1831 /* And initialize the newly allocated array */
1832 d
= mount_hashtable
;
1841 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1843 err
= subsystem_register(&fs_subsys
);
1845 printk(KERN_WARNING
"%s: subsystem_register error: %d\n",
1851 void __put_mnt_ns(struct mnt_namespace
*ns
)
1853 struct vfsmount
*root
= ns
->root
;
1854 LIST_HEAD(umount_list
);
1856 spin_unlock(&vfsmount_lock
);
1857 down_write(&namespace_sem
);
1858 spin_lock(&vfsmount_lock
);
1859 umount_tree(root
, 0, &umount_list
);
1860 spin_unlock(&vfsmount_lock
);
1861 up_write(&namespace_sem
);
1862 release_mounts(&umount_list
);