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 <linux/log2.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
33 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
34 #define HASH_SIZE (1UL << HASH_SHIFT)
36 /* spinlock for vfsmount related operations, inplace of dcache_lock */
37 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
41 static struct list_head
*mount_hashtable __read_mostly
;
42 static struct kmem_cache
*mnt_cache __read_mostly
;
43 static struct rw_semaphore namespace_sem
;
46 decl_subsys(fs
, NULL
, NULL
);
47 EXPORT_SYMBOL_GPL(fs_subsys
);
49 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
51 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
52 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
53 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
54 return tmp
& (HASH_SIZE
- 1);
57 struct vfsmount
*alloc_vfsmnt(const char *name
)
59 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
61 atomic_set(&mnt
->mnt_count
, 1);
62 INIT_LIST_HEAD(&mnt
->mnt_hash
);
63 INIT_LIST_HEAD(&mnt
->mnt_child
);
64 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
65 INIT_LIST_HEAD(&mnt
->mnt_list
);
66 INIT_LIST_HEAD(&mnt
->mnt_expire
);
67 INIT_LIST_HEAD(&mnt
->mnt_share
);
68 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
69 INIT_LIST_HEAD(&mnt
->mnt_slave
);
71 int size
= strlen(name
) + 1;
72 char *newname
= kmalloc(size
, GFP_KERNEL
);
74 memcpy(newname
, name
, size
);
75 mnt
->mnt_devname
= newname
;
82 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
85 mnt
->mnt_root
= dget(sb
->s_root
);
89 EXPORT_SYMBOL(simple_set_mnt
);
91 void free_vfsmnt(struct vfsmount
*mnt
)
93 kfree(mnt
->mnt_devname
);
94 kmem_cache_free(mnt_cache
, mnt
);
98 * find the first or last mount at @dentry on vfsmount @mnt depending on
99 * @dir. If @dir is set return the first mount else return the last mount.
101 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
104 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
105 struct list_head
*tmp
= head
;
106 struct vfsmount
*p
, *found
= NULL
;
109 tmp
= dir
? tmp
->next
: tmp
->prev
;
113 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
114 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
123 * lookup_mnt increments the ref count before returning
124 * the vfsmount struct.
126 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
128 struct vfsmount
*child_mnt
;
129 spin_lock(&vfsmount_lock
);
130 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
132 spin_unlock(&vfsmount_lock
);
136 static inline int check_mnt(struct vfsmount
*mnt
)
138 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
141 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
145 wake_up_interruptible(&ns
->poll
);
149 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
151 if (ns
&& ns
->event
!= event
) {
153 wake_up_interruptible(&ns
->poll
);
157 static void detach_mnt(struct vfsmount
*mnt
, struct nameidata
*old_nd
)
159 old_nd
->dentry
= mnt
->mnt_mountpoint
;
160 old_nd
->mnt
= mnt
->mnt_parent
;
161 mnt
->mnt_parent
= mnt
;
162 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
163 list_del_init(&mnt
->mnt_child
);
164 list_del_init(&mnt
->mnt_hash
);
165 old_nd
->dentry
->d_mounted
--;
168 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
169 struct vfsmount
*child_mnt
)
171 child_mnt
->mnt_parent
= mntget(mnt
);
172 child_mnt
->mnt_mountpoint
= dget(dentry
);
176 static void attach_mnt(struct vfsmount
*mnt
, struct nameidata
*nd
)
178 mnt_set_mountpoint(nd
->mnt
, nd
->dentry
, mnt
);
179 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
180 hash(nd
->mnt
, nd
->dentry
));
181 list_add_tail(&mnt
->mnt_child
, &nd
->mnt
->mnt_mounts
);
185 * the caller must hold vfsmount_lock
187 static void commit_tree(struct vfsmount
*mnt
)
189 struct vfsmount
*parent
= mnt
->mnt_parent
;
192 struct mnt_namespace
*n
= parent
->mnt_ns
;
194 BUG_ON(parent
== mnt
);
196 list_add_tail(&head
, &mnt
->mnt_list
);
197 list_for_each_entry(m
, &head
, mnt_list
)
199 list_splice(&head
, n
->list
.prev
);
201 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
202 hash(parent
, mnt
->mnt_mountpoint
));
203 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
204 touch_mnt_namespace(n
);
207 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
209 struct list_head
*next
= p
->mnt_mounts
.next
;
210 if (next
== &p
->mnt_mounts
) {
214 next
= p
->mnt_child
.next
;
215 if (next
!= &p
->mnt_parent
->mnt_mounts
)
220 return list_entry(next
, struct vfsmount
, mnt_child
);
223 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
225 struct list_head
*prev
= p
->mnt_mounts
.prev
;
226 while (prev
!= &p
->mnt_mounts
) {
227 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
228 prev
= p
->mnt_mounts
.prev
;
233 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
236 struct super_block
*sb
= old
->mnt_sb
;
237 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
240 mnt
->mnt_flags
= old
->mnt_flags
;
241 atomic_inc(&sb
->s_active
);
243 mnt
->mnt_root
= dget(root
);
244 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
245 mnt
->mnt_parent
= mnt
;
247 if (flag
& CL_SLAVE
) {
248 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
249 mnt
->mnt_master
= old
;
250 CLEAR_MNT_SHARED(mnt
);
252 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
253 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
254 if (IS_MNT_SLAVE(old
))
255 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
256 mnt
->mnt_master
= old
->mnt_master
;
258 if (flag
& CL_MAKE_SHARED
)
261 /* stick the duplicate mount on the same expiry list
262 * as the original if that was on one */
263 if (flag
& CL_EXPIRE
) {
264 spin_lock(&vfsmount_lock
);
265 if (!list_empty(&old
->mnt_expire
))
266 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
267 spin_unlock(&vfsmount_lock
);
273 static inline void __mntput(struct vfsmount
*mnt
)
275 struct super_block
*sb
= mnt
->mnt_sb
;
278 deactivate_super(sb
);
281 void mntput_no_expire(struct vfsmount
*mnt
)
284 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
285 if (likely(!mnt
->mnt_pinned
)) {
286 spin_unlock(&vfsmount_lock
);
290 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
292 spin_unlock(&vfsmount_lock
);
293 acct_auto_close_mnt(mnt
);
294 security_sb_umount_close(mnt
);
299 EXPORT_SYMBOL(mntput_no_expire
);
301 void mnt_pin(struct vfsmount
*mnt
)
303 spin_lock(&vfsmount_lock
);
305 spin_unlock(&vfsmount_lock
);
308 EXPORT_SYMBOL(mnt_pin
);
310 void mnt_unpin(struct vfsmount
*mnt
)
312 spin_lock(&vfsmount_lock
);
313 if (mnt
->mnt_pinned
) {
314 atomic_inc(&mnt
->mnt_count
);
317 spin_unlock(&vfsmount_lock
);
320 EXPORT_SYMBOL(mnt_unpin
);
323 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
325 struct mnt_namespace
*n
= m
->private;
329 down_read(&namespace_sem
);
330 list_for_each(p
, &n
->list
)
332 return list_entry(p
, struct vfsmount
, mnt_list
);
336 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
338 struct mnt_namespace
*n
= m
->private;
339 struct list_head
*p
= ((struct vfsmount
*)v
)->mnt_list
.next
;
341 return p
== &n
->list
? NULL
: list_entry(p
, struct vfsmount
, mnt_list
);
344 static void m_stop(struct seq_file
*m
, void *v
)
346 up_read(&namespace_sem
);
349 static inline void mangle(struct seq_file
*m
, const char *s
)
351 seq_escape(m
, s
, " \t\n\\");
354 static int show_vfsmnt(struct seq_file
*m
, void *v
)
356 struct vfsmount
*mnt
= v
;
358 static struct proc_fs_info
{
362 { MS_SYNCHRONOUS
, ",sync" },
363 { MS_DIRSYNC
, ",dirsync" },
364 { MS_MANDLOCK
, ",mand" },
367 static struct proc_fs_info mnt_info
[] = {
368 { MNT_NOSUID
, ",nosuid" },
369 { MNT_NODEV
, ",nodev" },
370 { MNT_NOEXEC
, ",noexec" },
371 { MNT_NOATIME
, ",noatime" },
372 { MNT_NODIRATIME
, ",nodiratime" },
373 { MNT_RELATIME
, ",relatime" },
376 struct proc_fs_info
*fs_infop
;
378 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
380 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
382 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
383 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
385 mangle(m
, mnt
->mnt_sb
->s_subtype
);
387 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
388 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
389 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
390 seq_puts(m
, fs_infop
->str
);
392 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
393 if (mnt
->mnt_flags
& fs_infop
->flag
)
394 seq_puts(m
, fs_infop
->str
);
396 if (mnt
->mnt_sb
->s_op
->show_options
)
397 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
398 seq_puts(m
, " 0 0\n");
402 struct seq_operations mounts_op
= {
409 static int show_vfsstat(struct seq_file
*m
, void *v
)
411 struct vfsmount
*mnt
= v
;
415 if (mnt
->mnt_devname
) {
416 seq_puts(m
, "device ");
417 mangle(m
, mnt
->mnt_devname
);
419 seq_puts(m
, "no device");
422 seq_puts(m
, " mounted on ");
423 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
426 /* file system type */
427 seq_puts(m
, "with fstype ");
428 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
430 /* optional statistics */
431 if (mnt
->mnt_sb
->s_op
->show_stats
) {
433 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
440 struct seq_operations mountstats_op
= {
444 .show
= show_vfsstat
,
448 * may_umount_tree - check if a mount tree is busy
449 * @mnt: root of mount tree
451 * This is called to check if a tree of mounts has any
452 * open files, pwds, chroots or sub mounts that are
455 int may_umount_tree(struct vfsmount
*mnt
)
458 int minimum_refs
= 0;
461 spin_lock(&vfsmount_lock
);
462 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
463 actual_refs
+= atomic_read(&p
->mnt_count
);
466 spin_unlock(&vfsmount_lock
);
468 if (actual_refs
> minimum_refs
)
474 EXPORT_SYMBOL(may_umount_tree
);
477 * may_umount - check if a mount point is busy
478 * @mnt: root of mount
480 * This is called to check if a mount point has any
481 * open files, pwds, chroots or sub mounts. If the
482 * mount has sub mounts this will return busy
483 * regardless of whether the sub mounts are busy.
485 * Doesn't take quota and stuff into account. IOW, in some cases it will
486 * give false negatives. The main reason why it's here is that we need
487 * a non-destructive way to look for easily umountable filesystems.
489 int may_umount(struct vfsmount
*mnt
)
492 spin_lock(&vfsmount_lock
);
493 if (propagate_mount_busy(mnt
, 2))
495 spin_unlock(&vfsmount_lock
);
499 EXPORT_SYMBOL(may_umount
);
501 void release_mounts(struct list_head
*head
)
503 struct vfsmount
*mnt
;
504 while (!list_empty(head
)) {
505 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
506 list_del_init(&mnt
->mnt_hash
);
507 if (mnt
->mnt_parent
!= mnt
) {
508 struct dentry
*dentry
;
510 spin_lock(&vfsmount_lock
);
511 dentry
= mnt
->mnt_mountpoint
;
513 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
514 mnt
->mnt_parent
= mnt
;
515 spin_unlock(&vfsmount_lock
);
523 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
527 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
528 list_move(&p
->mnt_hash
, kill
);
531 propagate_umount(kill
);
533 list_for_each_entry(p
, kill
, mnt_hash
) {
534 list_del_init(&p
->mnt_expire
);
535 list_del_init(&p
->mnt_list
);
536 __touch_mnt_namespace(p
->mnt_ns
);
538 list_del_init(&p
->mnt_child
);
539 if (p
->mnt_parent
!= p
)
540 p
->mnt_mountpoint
->d_mounted
--;
541 change_mnt_propagation(p
, MS_PRIVATE
);
545 static int do_umount(struct vfsmount
*mnt
, int flags
)
547 struct super_block
*sb
= mnt
->mnt_sb
;
549 LIST_HEAD(umount_list
);
551 retval
= security_sb_umount(mnt
, flags
);
556 * Allow userspace to request a mountpoint be expired rather than
557 * unmounting unconditionally. Unmount only happens if:
558 * (1) the mark is already set (the mark is cleared by mntput())
559 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
561 if (flags
& MNT_EXPIRE
) {
562 if (mnt
== current
->fs
->rootmnt
||
563 flags
& (MNT_FORCE
| MNT_DETACH
))
566 if (atomic_read(&mnt
->mnt_count
) != 2)
569 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
574 * If we may have to abort operations to get out of this
575 * mount, and they will themselves hold resources we must
576 * allow the fs to do things. In the Unix tradition of
577 * 'Gee thats tricky lets do it in userspace' the umount_begin
578 * might fail to complete on the first run through as other tasks
579 * must return, and the like. Thats for the mount program to worry
580 * about for the moment.
584 if (sb
->s_op
->umount_begin
)
585 sb
->s_op
->umount_begin(mnt
, flags
);
589 * No sense to grab the lock for this test, but test itself looks
590 * somewhat bogus. Suggestions for better replacement?
591 * Ho-hum... In principle, we might treat that as umount + switch
592 * to rootfs. GC would eventually take care of the old vfsmount.
593 * Actually it makes sense, especially if rootfs would contain a
594 * /reboot - static binary that would close all descriptors and
595 * call reboot(9). Then init(8) could umount root and exec /reboot.
597 if (mnt
== current
->fs
->rootmnt
&& !(flags
& MNT_DETACH
)) {
599 * Special case for "unmounting" root ...
600 * we just try to remount it readonly.
602 down_write(&sb
->s_umount
);
603 if (!(sb
->s_flags
& MS_RDONLY
)) {
606 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
609 up_write(&sb
->s_umount
);
613 down_write(&namespace_sem
);
614 spin_lock(&vfsmount_lock
);
618 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
619 if (!list_empty(&mnt
->mnt_list
))
620 umount_tree(mnt
, 1, &umount_list
);
623 spin_unlock(&vfsmount_lock
);
625 security_sb_umount_busy(mnt
);
626 up_write(&namespace_sem
);
627 release_mounts(&umount_list
);
632 * Now umount can handle mount points as well as block devices.
633 * This is important for filesystems which use unnamed block devices.
635 * We now support a flag for forced unmount like the other 'big iron'
636 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
639 asmlinkage
long sys_umount(char __user
* name
, int flags
)
644 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
648 if (nd
.dentry
!= nd
.mnt
->mnt_root
)
650 if (!check_mnt(nd
.mnt
))
654 if (!capable(CAP_SYS_ADMIN
))
657 retval
= do_umount(nd
.mnt
, flags
);
659 path_release_on_umount(&nd
);
664 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
667 * The 2.0 compatible umount. No flags.
669 asmlinkage
long sys_oldumount(char __user
* name
)
671 return sys_umount(name
, 0);
676 static int mount_is_safe(struct nameidata
*nd
)
678 if (capable(CAP_SYS_ADMIN
))
682 if (S_ISLNK(nd
->dentry
->d_inode
->i_mode
))
684 if (nd
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
685 if (current
->uid
!= nd
->dentry
->d_inode
->i_uid
)
688 if (vfs_permission(nd
, MAY_WRITE
))
694 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
699 if (d
== NULL
|| d
== d
->d_parent
)
705 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
708 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
711 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
714 res
= q
= clone_mnt(mnt
, dentry
, flag
);
717 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
720 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
721 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
724 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
725 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
726 s
= skip_mnt_tree(s
);
729 while (p
!= s
->mnt_parent
) {
735 nd
.dentry
= p
->mnt_mountpoint
;
736 q
= clone_mnt(p
, p
->mnt_root
, flag
);
739 spin_lock(&vfsmount_lock
);
740 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
742 spin_unlock(&vfsmount_lock
);
748 LIST_HEAD(umount_list
);
749 spin_lock(&vfsmount_lock
);
750 umount_tree(res
, 0, &umount_list
);
751 spin_unlock(&vfsmount_lock
);
752 release_mounts(&umount_list
);
758 * @source_mnt : mount tree to be attached
759 * @nd : place the mount tree @source_mnt is attached
760 * @parent_nd : if non-null, detach the source_mnt from its parent and
761 * store the parent mount and mountpoint dentry.
762 * (done when source_mnt is moved)
764 * NOTE: in the table below explains the semantics when a source mount
765 * of a given type is attached to a destination mount of a given type.
766 * ---------------------------------------------------------------------------
767 * | BIND MOUNT OPERATION |
768 * |**************************************************************************
769 * | source-->| shared | private | slave | unbindable |
773 * |**************************************************************************
774 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
776 * |non-shared| shared (+) | private | slave (*) | invalid |
777 * ***************************************************************************
778 * A bind operation clones the source mount and mounts the clone on the
781 * (++) the cloned mount is propagated to all the mounts in the propagation
782 * tree of the destination mount and the cloned mount is added to
783 * the peer group of the source mount.
784 * (+) the cloned mount is created under the destination mount and is marked
785 * as shared. The cloned mount is added to the peer group of the source
787 * (+++) the mount is propagated to all the mounts in the propagation tree
788 * of the destination mount and the cloned mount is made slave
789 * of the same master as that of the source mount. The cloned mount
790 * is marked as 'shared and slave'.
791 * (*) the cloned mount is made a slave of the same master as that of the
794 * ---------------------------------------------------------------------------
795 * | MOVE MOUNT OPERATION |
796 * |**************************************************************************
797 * | source-->| shared | private | slave | unbindable |
801 * |**************************************************************************
802 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
804 * |non-shared| shared (+*) | private | slave (*) | unbindable |
805 * ***************************************************************************
807 * (+) the mount is moved to the destination. And is then propagated to
808 * all the mounts in the propagation tree of the destination mount.
809 * (+*) the mount is moved to the destination.
810 * (+++) the mount is moved to the destination and is then propagated to
811 * all the mounts belonging to the destination mount's propagation tree.
812 * the mount is marked as 'shared and slave'.
813 * (*) the mount continues to be a slave at the new location.
815 * if the source mount is a tree, the operations explained above is
816 * applied to each mount in the tree.
817 * Must be called without spinlocks held, since this function can sleep
820 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
821 struct nameidata
*nd
, struct nameidata
*parent_nd
)
823 LIST_HEAD(tree_list
);
824 struct vfsmount
*dest_mnt
= nd
->mnt
;
825 struct dentry
*dest_dentry
= nd
->dentry
;
826 struct vfsmount
*child
, *p
;
828 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
831 if (IS_MNT_SHARED(dest_mnt
)) {
832 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
836 spin_lock(&vfsmount_lock
);
838 detach_mnt(source_mnt
, parent_nd
);
839 attach_mnt(source_mnt
, nd
);
840 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
842 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
843 commit_tree(source_mnt
);
846 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
847 list_del_init(&child
->mnt_hash
);
850 spin_unlock(&vfsmount_lock
);
854 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
857 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
860 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
861 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
865 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
866 if (IS_DEADDIR(nd
->dentry
->d_inode
))
869 err
= security_sb_check_sb(mnt
, nd
);
874 if (IS_ROOT(nd
->dentry
) || !d_unhashed(nd
->dentry
))
875 err
= attach_recursive_mnt(mnt
, nd
, NULL
);
877 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
879 security_sb_post_addmount(mnt
, nd
);
884 * recursively change the type of the mountpoint.
886 static int do_change_type(struct nameidata
*nd
, int flag
)
888 struct vfsmount
*m
, *mnt
= nd
->mnt
;
889 int recurse
= flag
& MS_REC
;
890 int type
= flag
& ~MS_REC
;
892 if (!capable(CAP_SYS_ADMIN
))
895 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
898 down_write(&namespace_sem
);
899 spin_lock(&vfsmount_lock
);
900 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
901 change_mnt_propagation(m
, type
);
902 spin_unlock(&vfsmount_lock
);
903 up_write(&namespace_sem
);
910 static int do_loopback(struct nameidata
*nd
, char *old_name
, int recurse
)
912 struct nameidata old_nd
;
913 struct vfsmount
*mnt
= NULL
;
914 int err
= mount_is_safe(nd
);
917 if (!old_name
|| !*old_name
)
919 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
923 down_write(&namespace_sem
);
925 if (IS_MNT_UNBINDABLE(old_nd
.mnt
))
928 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
933 mnt
= copy_tree(old_nd
.mnt
, old_nd
.dentry
, 0);
935 mnt
= clone_mnt(old_nd
.mnt
, old_nd
.dentry
, 0);
940 err
= graft_tree(mnt
, nd
);
942 LIST_HEAD(umount_list
);
943 spin_lock(&vfsmount_lock
);
944 umount_tree(mnt
, 0, &umount_list
);
945 spin_unlock(&vfsmount_lock
);
946 release_mounts(&umount_list
);
950 up_write(&namespace_sem
);
951 path_release(&old_nd
);
956 * change filesystem flags. dir should be a physical root of filesystem.
957 * If you've mounted a non-root directory somewhere and want to do remount
958 * on it - tough luck.
960 static int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
964 struct super_block
*sb
= nd
->mnt
->mnt_sb
;
966 if (!capable(CAP_SYS_ADMIN
))
969 if (!check_mnt(nd
->mnt
))
972 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
975 down_write(&sb
->s_umount
);
976 err
= do_remount_sb(sb
, flags
, data
, 0);
978 nd
->mnt
->mnt_flags
= mnt_flags
;
979 up_write(&sb
->s_umount
);
981 security_sb_post_remount(nd
->mnt
, flags
, data
);
985 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
988 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
989 if (IS_MNT_UNBINDABLE(p
))
995 static int do_move_mount(struct nameidata
*nd
, char *old_name
)
997 struct nameidata old_nd
, parent_nd
;
1000 if (!capable(CAP_SYS_ADMIN
))
1002 if (!old_name
|| !*old_name
)
1004 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1008 down_write(&namespace_sem
);
1009 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1012 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
1016 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
1017 if (IS_DEADDIR(nd
->dentry
->d_inode
))
1020 if (!IS_ROOT(nd
->dentry
) && d_unhashed(nd
->dentry
))
1024 if (old_nd
.dentry
!= old_nd
.mnt
->mnt_root
)
1027 if (old_nd
.mnt
== old_nd
.mnt
->mnt_parent
)
1030 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
1031 S_ISDIR(old_nd
.dentry
->d_inode
->i_mode
))
1034 * Don't move a mount residing in a shared parent.
1036 if (old_nd
.mnt
->mnt_parent
&& IS_MNT_SHARED(old_nd
.mnt
->mnt_parent
))
1039 * Don't move a mount tree containing unbindable mounts to a destination
1040 * mount which is shared.
1042 if (IS_MNT_SHARED(nd
->mnt
) && tree_contains_unbindable(old_nd
.mnt
))
1045 for (p
= nd
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1046 if (p
== old_nd
.mnt
)
1049 if ((err
= attach_recursive_mnt(old_nd
.mnt
, nd
, &parent_nd
)))
1052 spin_lock(&vfsmount_lock
);
1053 /* if the mount is moved, it should no longer be expire
1055 list_del_init(&old_nd
.mnt
->mnt_expire
);
1056 spin_unlock(&vfsmount_lock
);
1058 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
1060 up_write(&namespace_sem
);
1062 path_release(&parent_nd
);
1063 path_release(&old_nd
);
1068 * create a new mount for userspace and request it to be added into the
1071 static int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1072 int mnt_flags
, char *name
, void *data
)
1074 struct vfsmount
*mnt
;
1076 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1079 /* we need capabilities... */
1080 if (!capable(CAP_SYS_ADMIN
))
1083 mnt
= do_kern_mount(type
, flags
, name
, data
);
1085 return PTR_ERR(mnt
);
1087 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1091 * add a mount into a namespace's mount tree
1092 * - provide the option of adding the new mount to an expiration list
1094 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1095 int mnt_flags
, struct list_head
*fslist
)
1099 down_write(&namespace_sem
);
1100 /* Something was mounted here while we slept */
1101 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1104 if (!check_mnt(nd
->mnt
))
1107 /* Refuse the same filesystem on the same mount point */
1109 if (nd
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1110 nd
->mnt
->mnt_root
== nd
->dentry
)
1114 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1117 newmnt
->mnt_flags
= mnt_flags
;
1118 if ((err
= graft_tree(newmnt
, nd
)))
1122 /* add to the specified expiration list */
1123 spin_lock(&vfsmount_lock
);
1124 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1125 spin_unlock(&vfsmount_lock
);
1127 up_write(&namespace_sem
);
1131 up_write(&namespace_sem
);
1136 EXPORT_SYMBOL_GPL(do_add_mount
);
1138 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
1139 struct list_head
*umounts
)
1141 spin_lock(&vfsmount_lock
);
1144 * Check if mount is still attached, if not, let whoever holds it deal
1147 if (mnt
->mnt_parent
== mnt
) {
1148 spin_unlock(&vfsmount_lock
);
1153 * Check that it is still dead: the count should now be 2 - as
1154 * contributed by the vfsmount parent and the mntget above
1156 if (!propagate_mount_busy(mnt
, 2)) {
1157 /* delete from the namespace */
1158 touch_mnt_namespace(mnt
->mnt_ns
);
1159 list_del_init(&mnt
->mnt_list
);
1161 umount_tree(mnt
, 1, umounts
);
1162 spin_unlock(&vfsmount_lock
);
1165 * Someone brought it back to life whilst we didn't have any
1166 * locks held so return it to the expiration list
1168 list_add_tail(&mnt
->mnt_expire
, mounts
);
1169 spin_unlock(&vfsmount_lock
);
1174 * go through the vfsmounts we've just consigned to the graveyard to
1175 * - check that they're still dead
1176 * - delete the vfsmount from the appropriate namespace under lock
1177 * - dispose of the corpse
1179 static void expire_mount_list(struct list_head
*graveyard
, struct list_head
*mounts
)
1181 struct mnt_namespace
*ns
;
1182 struct vfsmount
*mnt
;
1184 while (!list_empty(graveyard
)) {
1186 mnt
= list_first_entry(graveyard
, struct vfsmount
, mnt_expire
);
1187 list_del_init(&mnt
->mnt_expire
);
1189 /* don't do anything if the namespace is dead - all the
1190 * vfsmounts from it are going away anyway */
1192 if (!ns
|| !ns
->root
)
1196 spin_unlock(&vfsmount_lock
);
1197 down_write(&namespace_sem
);
1198 expire_mount(mnt
, mounts
, &umounts
);
1199 up_write(&namespace_sem
);
1200 release_mounts(&umounts
);
1203 spin_lock(&vfsmount_lock
);
1208 * process a list of expirable mountpoints with the intent of discarding any
1209 * mountpoints that aren't in use and haven't been touched since last we came
1212 void mark_mounts_for_expiry(struct list_head
*mounts
)
1214 struct vfsmount
*mnt
, *next
;
1215 LIST_HEAD(graveyard
);
1217 if (list_empty(mounts
))
1220 spin_lock(&vfsmount_lock
);
1222 /* extract from the expiration list every vfsmount that matches the
1223 * following criteria:
1224 * - only referenced by its parent vfsmount
1225 * - still marked for expiry (marked on the last call here; marks are
1226 * cleared by mntput())
1228 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1229 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1230 atomic_read(&mnt
->mnt_count
) != 1)
1234 list_move(&mnt
->mnt_expire
, &graveyard
);
1237 expire_mount_list(&graveyard
, mounts
);
1239 spin_unlock(&vfsmount_lock
);
1242 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1245 * Ripoff of 'select_parent()'
1247 * search the list of submounts for a given mountpoint, and move any
1248 * shrinkable submounts to the 'graveyard' list.
1250 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1252 struct vfsmount
*this_parent
= parent
;
1253 struct list_head
*next
;
1257 next
= this_parent
->mnt_mounts
.next
;
1259 while (next
!= &this_parent
->mnt_mounts
) {
1260 struct list_head
*tmp
= next
;
1261 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1264 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1267 * Descend a level if the d_mounts list is non-empty.
1269 if (!list_empty(&mnt
->mnt_mounts
)) {
1274 if (!propagate_mount_busy(mnt
, 1)) {
1276 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1281 * All done at this level ... ascend and resume the search
1283 if (this_parent
!= parent
) {
1284 next
= this_parent
->mnt_child
.next
;
1285 this_parent
= this_parent
->mnt_parent
;
1292 * process a list of expirable mountpoints with the intent of discarding any
1293 * submounts of a specific parent mountpoint
1295 void shrink_submounts(struct vfsmount
*mountpoint
, struct list_head
*mounts
)
1297 LIST_HEAD(graveyard
);
1300 spin_lock(&vfsmount_lock
);
1302 /* extract submounts of 'mountpoint' from the expiration list */
1303 while ((found
= select_submounts(mountpoint
, &graveyard
)) != 0)
1304 expire_mount_list(&graveyard
, mounts
);
1306 spin_unlock(&vfsmount_lock
);
1309 EXPORT_SYMBOL_GPL(shrink_submounts
);
1312 * Some copy_from_user() implementations do not return the exact number of
1313 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1314 * Note that this function differs from copy_from_user() in that it will oops
1315 * on bad values of `to', rather than returning a short copy.
1317 static long exact_copy_from_user(void *to
, const void __user
* from
,
1321 const char __user
*f
= from
;
1324 if (!access_ok(VERIFY_READ
, from
, n
))
1328 if (__get_user(c
, f
)) {
1339 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1349 if (!(page
= __get_free_page(GFP_KERNEL
)))
1352 /* We only care that *some* data at the address the user
1353 * gave us is valid. Just in case, we'll zero
1354 * the remainder of the page.
1356 /* copy_from_user cannot cross TASK_SIZE ! */
1357 size
= TASK_SIZE
- (unsigned long)data
;
1358 if (size
> PAGE_SIZE
)
1361 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1367 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1373 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1374 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1376 * data is a (void *) that can point to any structure up to
1377 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1378 * information (or be NULL).
1380 * Pre-0.97 versions of mount() didn't have a flags word.
1381 * When the flags word was introduced its top half was required
1382 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1383 * Therefore, if this magic number is present, it carries no information
1384 * and must be discarded.
1386 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1387 unsigned long flags
, void *data_page
)
1389 struct nameidata nd
;
1394 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1395 flags
&= ~MS_MGC_MSK
;
1397 /* Basic sanity checks */
1399 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1401 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1405 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1407 /* Separate the per-mountpoint flags */
1408 if (flags
& MS_NOSUID
)
1409 mnt_flags
|= MNT_NOSUID
;
1410 if (flags
& MS_NODEV
)
1411 mnt_flags
|= MNT_NODEV
;
1412 if (flags
& MS_NOEXEC
)
1413 mnt_flags
|= MNT_NOEXEC
;
1414 if (flags
& MS_NOATIME
)
1415 mnt_flags
|= MNT_NOATIME
;
1416 if (flags
& MS_NODIRATIME
)
1417 mnt_flags
|= MNT_NODIRATIME
;
1418 if (flags
& MS_RELATIME
)
1419 mnt_flags
|= MNT_RELATIME
;
1421 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1422 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
);
1424 /* ... and get the mountpoint */
1425 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1429 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1433 if (flags
& MS_REMOUNT
)
1434 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1436 else if (flags
& MS_BIND
)
1437 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1438 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1439 retval
= do_change_type(&nd
, flags
);
1440 else if (flags
& MS_MOVE
)
1441 retval
= do_move_mount(&nd
, dev_name
);
1443 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1444 dev_name
, data_page
);
1451 * Allocate a new namespace structure and populate it with contents
1452 * copied from the namespace of the passed in task structure.
1454 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1455 struct fs_struct
*fs
)
1457 struct mnt_namespace
*new_ns
;
1458 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1459 struct vfsmount
*p
, *q
;
1461 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1465 atomic_set(&new_ns
->count
, 1);
1466 INIT_LIST_HEAD(&new_ns
->list
);
1467 init_waitqueue_head(&new_ns
->poll
);
1470 down_write(&namespace_sem
);
1471 /* First pass: copy the tree topology */
1472 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1473 CL_COPY_ALL
| CL_EXPIRE
);
1474 if (!new_ns
->root
) {
1475 up_write(&namespace_sem
);
1479 spin_lock(&vfsmount_lock
);
1480 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1481 spin_unlock(&vfsmount_lock
);
1484 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1485 * as belonging to new namespace. We have already acquired a private
1486 * fs_struct, so tsk->fs->lock is not needed.
1493 if (p
== fs
->rootmnt
) {
1495 fs
->rootmnt
= mntget(q
);
1497 if (p
== fs
->pwdmnt
) {
1499 fs
->pwdmnt
= mntget(q
);
1501 if (p
== fs
->altrootmnt
) {
1503 fs
->altrootmnt
= mntget(q
);
1506 p
= next_mnt(p
, mnt_ns
->root
);
1507 q
= next_mnt(q
, new_ns
->root
);
1509 up_write(&namespace_sem
);
1521 struct mnt_namespace
*copy_mnt_ns(int flags
, struct mnt_namespace
*ns
,
1522 struct fs_struct
*new_fs
)
1524 struct mnt_namespace
*new_ns
;
1529 if (!(flags
& CLONE_NEWNS
))
1532 new_ns
= dup_mnt_ns(ns
, new_fs
);
1538 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1539 char __user
* type
, unsigned long flags
,
1543 unsigned long data_page
;
1544 unsigned long type_page
;
1545 unsigned long dev_page
;
1548 retval
= copy_mount_options(type
, &type_page
);
1552 dir_page
= getname(dir_name
);
1553 retval
= PTR_ERR(dir_page
);
1554 if (IS_ERR(dir_page
))
1557 retval
= copy_mount_options(dev_name
, &dev_page
);
1561 retval
= copy_mount_options(data
, &data_page
);
1566 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1567 flags
, (void *)data_page
);
1569 free_page(data_page
);
1572 free_page(dev_page
);
1576 free_page(type_page
);
1581 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1582 * It can block. Requires the big lock held.
1584 void set_fs_root(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1585 struct dentry
*dentry
)
1587 struct dentry
*old_root
;
1588 struct vfsmount
*old_rootmnt
;
1589 write_lock(&fs
->lock
);
1590 old_root
= fs
->root
;
1591 old_rootmnt
= fs
->rootmnt
;
1592 fs
->rootmnt
= mntget(mnt
);
1593 fs
->root
= dget(dentry
);
1594 write_unlock(&fs
->lock
);
1597 mntput(old_rootmnt
);
1602 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1603 * It can block. Requires the big lock held.
1605 void set_fs_pwd(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1606 struct dentry
*dentry
)
1608 struct dentry
*old_pwd
;
1609 struct vfsmount
*old_pwdmnt
;
1611 write_lock(&fs
->lock
);
1613 old_pwdmnt
= fs
->pwdmnt
;
1614 fs
->pwdmnt
= mntget(mnt
);
1615 fs
->pwd
= dget(dentry
);
1616 write_unlock(&fs
->lock
);
1624 static void chroot_fs_refs(struct nameidata
*old_nd
, struct nameidata
*new_nd
)
1626 struct task_struct
*g
, *p
;
1627 struct fs_struct
*fs
;
1629 read_lock(&tasklist_lock
);
1630 do_each_thread(g
, p
) {
1634 atomic_inc(&fs
->count
);
1636 if (fs
->root
== old_nd
->dentry
1637 && fs
->rootmnt
== old_nd
->mnt
)
1638 set_fs_root(fs
, new_nd
->mnt
, new_nd
->dentry
);
1639 if (fs
->pwd
== old_nd
->dentry
1640 && fs
->pwdmnt
== old_nd
->mnt
)
1641 set_fs_pwd(fs
, new_nd
->mnt
, new_nd
->dentry
);
1645 } while_each_thread(g
, p
);
1646 read_unlock(&tasklist_lock
);
1650 * pivot_root Semantics:
1651 * Moves the root file system of the current process to the directory put_old,
1652 * makes new_root as the new root file system of the current process, and sets
1653 * root/cwd of all processes which had them on the current root to new_root.
1656 * The new_root and put_old must be directories, and must not be on the
1657 * same file system as the current process root. The put_old must be
1658 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1659 * pointed to by put_old must yield the same directory as new_root. No other
1660 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1662 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1663 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1664 * in this situation.
1667 * - we don't move root/cwd if they are not at the root (reason: if something
1668 * cared enough to change them, it's probably wrong to force them elsewhere)
1669 * - it's okay to pick a root that isn't the root of a file system, e.g.
1670 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1671 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1674 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1675 const char __user
* put_old
)
1677 struct vfsmount
*tmp
;
1678 struct nameidata new_nd
, old_nd
, parent_nd
, root_parent
, user_nd
;
1681 if (!capable(CAP_SYS_ADMIN
))
1686 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1691 if (!check_mnt(new_nd
.mnt
))
1694 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1698 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1700 path_release(&old_nd
);
1704 read_lock(¤t
->fs
->lock
);
1705 user_nd
.mnt
= mntget(current
->fs
->rootmnt
);
1706 user_nd
.dentry
= dget(current
->fs
->root
);
1707 read_unlock(¤t
->fs
->lock
);
1708 down_write(&namespace_sem
);
1709 mutex_lock(&old_nd
.dentry
->d_inode
->i_mutex
);
1711 if (IS_MNT_SHARED(old_nd
.mnt
) ||
1712 IS_MNT_SHARED(new_nd
.mnt
->mnt_parent
) ||
1713 IS_MNT_SHARED(user_nd
.mnt
->mnt_parent
))
1715 if (!check_mnt(user_nd
.mnt
))
1718 if (IS_DEADDIR(new_nd
.dentry
->d_inode
))
1720 if (d_unhashed(new_nd
.dentry
) && !IS_ROOT(new_nd
.dentry
))
1722 if (d_unhashed(old_nd
.dentry
) && !IS_ROOT(old_nd
.dentry
))
1725 if (new_nd
.mnt
== user_nd
.mnt
|| old_nd
.mnt
== user_nd
.mnt
)
1726 goto out2
; /* loop, on the same file system */
1728 if (user_nd
.mnt
->mnt_root
!= user_nd
.dentry
)
1729 goto out2
; /* not a mountpoint */
1730 if (user_nd
.mnt
->mnt_parent
== user_nd
.mnt
)
1731 goto out2
; /* not attached */
1732 if (new_nd
.mnt
->mnt_root
!= new_nd
.dentry
)
1733 goto out2
; /* not a mountpoint */
1734 if (new_nd
.mnt
->mnt_parent
== new_nd
.mnt
)
1735 goto out2
; /* not attached */
1736 tmp
= old_nd
.mnt
; /* make sure we can reach put_old from new_root */
1737 spin_lock(&vfsmount_lock
);
1738 if (tmp
!= new_nd
.mnt
) {
1740 if (tmp
->mnt_parent
== tmp
)
1741 goto out3
; /* already mounted on put_old */
1742 if (tmp
->mnt_parent
== new_nd
.mnt
)
1744 tmp
= tmp
->mnt_parent
;
1746 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.dentry
))
1748 } else if (!is_subdir(old_nd
.dentry
, new_nd
.dentry
))
1750 detach_mnt(new_nd
.mnt
, &parent_nd
);
1751 detach_mnt(user_nd
.mnt
, &root_parent
);
1752 attach_mnt(user_nd
.mnt
, &old_nd
); /* mount old root on put_old */
1753 attach_mnt(new_nd
.mnt
, &root_parent
); /* mount new_root on / */
1754 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1755 spin_unlock(&vfsmount_lock
);
1756 chroot_fs_refs(&user_nd
, &new_nd
);
1757 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1759 path_release(&root_parent
);
1760 path_release(&parent_nd
);
1762 mutex_unlock(&old_nd
.dentry
->d_inode
->i_mutex
);
1763 up_write(&namespace_sem
);
1764 path_release(&user_nd
);
1765 path_release(&old_nd
);
1767 path_release(&new_nd
);
1772 spin_unlock(&vfsmount_lock
);
1776 static void __init
init_mount_tree(void)
1778 struct vfsmount
*mnt
;
1779 struct mnt_namespace
*ns
;
1781 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1783 panic("Can't create rootfs");
1784 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1786 panic("Can't allocate initial namespace");
1787 atomic_set(&ns
->count
, 1);
1788 INIT_LIST_HEAD(&ns
->list
);
1789 init_waitqueue_head(&ns
->poll
);
1791 list_add(&mnt
->mnt_list
, &ns
->list
);
1795 init_task
.nsproxy
->mnt_ns
= ns
;
1798 set_fs_pwd(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1799 set_fs_root(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1802 void __init
mnt_init(unsigned long mempages
)
1807 init_rwsem(&namespace_sem
);
1809 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1810 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
, NULL
);
1812 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1814 if (!mount_hashtable
)
1815 panic("Failed to allocate mount hash table\n");
1817 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
1819 for (u
= 0; u
< HASH_SIZE
; u
++)
1820 INIT_LIST_HEAD(&mount_hashtable
[u
]);
1824 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1826 err
= subsystem_register(&fs_subsys
);
1828 printk(KERN_WARNING
"%s: subsystem_register error: %d\n",
1834 void __put_mnt_ns(struct mnt_namespace
*ns
)
1836 struct vfsmount
*root
= ns
->root
;
1837 LIST_HEAD(umount_list
);
1839 spin_unlock(&vfsmount_lock
);
1840 down_write(&namespace_sem
);
1841 spin_lock(&vfsmount_lock
);
1842 umount_tree(root
, 0, &umount_list
);
1843 spin_unlock(&vfsmount_lock
);
1844 up_write(&namespace_sem
);
1845 release_mounts(&umount_list
);