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;
326 down_read(&namespace_sem
);
327 list_for_each(p
, &n
->list
)
329 return list_entry(p
, struct vfsmount
, mnt_list
);
333 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
335 struct mnt_namespace
*n
= m
->private;
336 struct list_head
*p
= ((struct vfsmount
*)v
)->mnt_list
.next
;
338 return p
== &n
->list
? NULL
: list_entry(p
, struct vfsmount
, mnt_list
);
341 static void m_stop(struct seq_file
*m
, void *v
)
343 up_read(&namespace_sem
);
346 static inline void mangle(struct seq_file
*m
, const char *s
)
348 seq_escape(m
, s
, " \t\n\\");
351 static int show_vfsmnt(struct seq_file
*m
, void *v
)
353 struct vfsmount
*mnt
= v
;
355 static struct proc_fs_info
{
359 { MS_SYNCHRONOUS
, ",sync" },
360 { MS_DIRSYNC
, ",dirsync" },
361 { MS_MANDLOCK
, ",mand" },
364 static struct proc_fs_info mnt_info
[] = {
365 { MNT_NOSUID
, ",nosuid" },
366 { MNT_NODEV
, ",nodev" },
367 { MNT_NOEXEC
, ",noexec" },
368 { MNT_NOATIME
, ",noatime" },
369 { MNT_NODIRATIME
, ",nodiratime" },
370 { MNT_RELATIME
, ",relatime" },
373 struct proc_fs_info
*fs_infop
;
375 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
377 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
379 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
380 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
382 mangle(m
, mnt
->mnt_sb
->s_subtype
);
384 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
385 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
386 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
387 seq_puts(m
, fs_infop
->str
);
389 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
390 if (mnt
->mnt_flags
& fs_infop
->flag
)
391 seq_puts(m
, fs_infop
->str
);
393 if (mnt
->mnt_sb
->s_op
->show_options
)
394 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
395 seq_puts(m
, " 0 0\n");
399 struct seq_operations mounts_op
= {
406 static int show_vfsstat(struct seq_file
*m
, void *v
)
408 struct vfsmount
*mnt
= v
;
412 if (mnt
->mnt_devname
) {
413 seq_puts(m
, "device ");
414 mangle(m
, mnt
->mnt_devname
);
416 seq_puts(m
, "no device");
419 seq_puts(m
, " mounted on ");
420 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
423 /* file system type */
424 seq_puts(m
, "with fstype ");
425 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
427 /* optional statistics */
428 if (mnt
->mnt_sb
->s_op
->show_stats
) {
430 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
437 struct seq_operations mountstats_op
= {
441 .show
= show_vfsstat
,
445 * may_umount_tree - check if a mount tree is busy
446 * @mnt: root of mount tree
448 * This is called to check if a tree of mounts has any
449 * open files, pwds, chroots or sub mounts that are
452 int may_umount_tree(struct vfsmount
*mnt
)
455 int minimum_refs
= 0;
458 spin_lock(&vfsmount_lock
);
459 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
460 actual_refs
+= atomic_read(&p
->mnt_count
);
463 spin_unlock(&vfsmount_lock
);
465 if (actual_refs
> minimum_refs
)
471 EXPORT_SYMBOL(may_umount_tree
);
474 * may_umount - check if a mount point is busy
475 * @mnt: root of mount
477 * This is called to check if a mount point has any
478 * open files, pwds, chroots or sub mounts. If the
479 * mount has sub mounts this will return busy
480 * regardless of whether the sub mounts are busy.
482 * Doesn't take quota and stuff into account. IOW, in some cases it will
483 * give false negatives. The main reason why it's here is that we need
484 * a non-destructive way to look for easily umountable filesystems.
486 int may_umount(struct vfsmount
*mnt
)
489 spin_lock(&vfsmount_lock
);
490 if (propagate_mount_busy(mnt
, 2))
492 spin_unlock(&vfsmount_lock
);
496 EXPORT_SYMBOL(may_umount
);
498 void release_mounts(struct list_head
*head
)
500 struct vfsmount
*mnt
;
501 while (!list_empty(head
)) {
502 mnt
= list_entry(head
->next
, struct vfsmount
, mnt_hash
);
503 list_del_init(&mnt
->mnt_hash
);
504 if (mnt
->mnt_parent
!= mnt
) {
505 struct dentry
*dentry
;
507 spin_lock(&vfsmount_lock
);
508 dentry
= mnt
->mnt_mountpoint
;
510 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
511 mnt
->mnt_parent
= mnt
;
512 spin_unlock(&vfsmount_lock
);
520 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
524 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
525 list_move(&p
->mnt_hash
, kill
);
528 propagate_umount(kill
);
530 list_for_each_entry(p
, kill
, mnt_hash
) {
531 list_del_init(&p
->mnt_expire
);
532 list_del_init(&p
->mnt_list
);
533 __touch_mnt_namespace(p
->mnt_ns
);
535 list_del_init(&p
->mnt_child
);
536 if (p
->mnt_parent
!= p
)
537 p
->mnt_mountpoint
->d_mounted
--;
538 change_mnt_propagation(p
, MS_PRIVATE
);
542 static int do_umount(struct vfsmount
*mnt
, int flags
)
544 struct super_block
*sb
= mnt
->mnt_sb
;
546 LIST_HEAD(umount_list
);
548 retval
= security_sb_umount(mnt
, flags
);
553 * Allow userspace to request a mountpoint be expired rather than
554 * unmounting unconditionally. Unmount only happens if:
555 * (1) the mark is already set (the mark is cleared by mntput())
556 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
558 if (flags
& MNT_EXPIRE
) {
559 if (mnt
== current
->fs
->rootmnt
||
560 flags
& (MNT_FORCE
| MNT_DETACH
))
563 if (atomic_read(&mnt
->mnt_count
) != 2)
566 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
571 * If we may have to abort operations to get out of this
572 * mount, and they will themselves hold resources we must
573 * allow the fs to do things. In the Unix tradition of
574 * 'Gee thats tricky lets do it in userspace' the umount_begin
575 * might fail to complete on the first run through as other tasks
576 * must return, and the like. Thats for the mount program to worry
577 * about for the moment.
581 if (sb
->s_op
->umount_begin
)
582 sb
->s_op
->umount_begin(mnt
, flags
);
586 * No sense to grab the lock for this test, but test itself looks
587 * somewhat bogus. Suggestions for better replacement?
588 * Ho-hum... In principle, we might treat that as umount + switch
589 * to rootfs. GC would eventually take care of the old vfsmount.
590 * Actually it makes sense, especially if rootfs would contain a
591 * /reboot - static binary that would close all descriptors and
592 * call reboot(9). Then init(8) could umount root and exec /reboot.
594 if (mnt
== current
->fs
->rootmnt
&& !(flags
& MNT_DETACH
)) {
596 * Special case for "unmounting" root ...
597 * we just try to remount it readonly.
599 down_write(&sb
->s_umount
);
600 if (!(sb
->s_flags
& MS_RDONLY
)) {
603 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
606 up_write(&sb
->s_umount
);
610 down_write(&namespace_sem
);
611 spin_lock(&vfsmount_lock
);
615 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
616 if (!list_empty(&mnt
->mnt_list
))
617 umount_tree(mnt
, 1, &umount_list
);
620 spin_unlock(&vfsmount_lock
);
622 security_sb_umount_busy(mnt
);
623 up_write(&namespace_sem
);
624 release_mounts(&umount_list
);
629 * Now umount can handle mount points as well as block devices.
630 * This is important for filesystems which use unnamed block devices.
632 * We now support a flag for forced unmount like the other 'big iron'
633 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
636 asmlinkage
long sys_umount(char __user
* name
, int flags
)
641 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
645 if (nd
.dentry
!= nd
.mnt
->mnt_root
)
647 if (!check_mnt(nd
.mnt
))
651 if (!capable(CAP_SYS_ADMIN
))
654 retval
= do_umount(nd
.mnt
, flags
);
656 path_release_on_umount(&nd
);
661 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
664 * The 2.0 compatible umount. No flags.
666 asmlinkage
long sys_oldumount(char __user
* name
)
668 return sys_umount(name
, 0);
673 static int mount_is_safe(struct nameidata
*nd
)
675 if (capable(CAP_SYS_ADMIN
))
679 if (S_ISLNK(nd
->dentry
->d_inode
->i_mode
))
681 if (nd
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
682 if (current
->uid
!= nd
->dentry
->d_inode
->i_uid
)
685 if (vfs_permission(nd
, MAY_WRITE
))
691 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
696 if (d
== NULL
|| d
== d
->d_parent
)
702 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
705 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
708 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
711 res
= q
= clone_mnt(mnt
, dentry
, flag
);
714 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
717 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
718 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
721 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
722 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
723 s
= skip_mnt_tree(s
);
726 while (p
!= s
->mnt_parent
) {
732 nd
.dentry
= p
->mnt_mountpoint
;
733 q
= clone_mnt(p
, p
->mnt_root
, flag
);
736 spin_lock(&vfsmount_lock
);
737 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
739 spin_unlock(&vfsmount_lock
);
745 LIST_HEAD(umount_list
);
746 spin_lock(&vfsmount_lock
);
747 umount_tree(res
, 0, &umount_list
);
748 spin_unlock(&vfsmount_lock
);
749 release_mounts(&umount_list
);
755 * @source_mnt : mount tree to be attached
756 * @nd : place the mount tree @source_mnt is attached
757 * @parent_nd : if non-null, detach the source_mnt from its parent and
758 * store the parent mount and mountpoint dentry.
759 * (done when source_mnt is moved)
761 * NOTE: in the table below explains the semantics when a source mount
762 * of a given type is attached to a destination mount of a given type.
763 * ---------------------------------------------------------------------------
764 * | BIND MOUNT OPERATION |
765 * |**************************************************************************
766 * | source-->| shared | private | slave | unbindable |
770 * |**************************************************************************
771 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
773 * |non-shared| shared (+) | private | slave (*) | invalid |
774 * ***************************************************************************
775 * A bind operation clones the source mount and mounts the clone on the
778 * (++) the cloned mount is propagated to all the mounts in the propagation
779 * tree of the destination mount and the cloned mount is added to
780 * the peer group of the source mount.
781 * (+) the cloned mount is created under the destination mount and is marked
782 * as shared. The cloned mount is added to the peer group of the source
784 * (+++) the mount is propagated to all the mounts in the propagation tree
785 * of the destination mount and the cloned mount is made slave
786 * of the same master as that of the source mount. The cloned mount
787 * is marked as 'shared and slave'.
788 * (*) the cloned mount is made a slave of the same master as that of the
791 * ---------------------------------------------------------------------------
792 * | MOVE MOUNT OPERATION |
793 * |**************************************************************************
794 * | source-->| shared | private | slave | unbindable |
798 * |**************************************************************************
799 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
801 * |non-shared| shared (+*) | private | slave (*) | unbindable |
802 * ***************************************************************************
804 * (+) the mount is moved to the destination. And is then propagated to
805 * all the mounts in the propagation tree of the destination mount.
806 * (+*) the mount is moved to the destination.
807 * (+++) the mount is moved to the destination and is then propagated to
808 * all the mounts belonging to the destination mount's propagation tree.
809 * the mount is marked as 'shared and slave'.
810 * (*) the mount continues to be a slave at the new location.
812 * if the source mount is a tree, the operations explained above is
813 * applied to each mount in the tree.
814 * Must be called without spinlocks held, since this function can sleep
817 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
818 struct nameidata
*nd
, struct nameidata
*parent_nd
)
820 LIST_HEAD(tree_list
);
821 struct vfsmount
*dest_mnt
= nd
->mnt
;
822 struct dentry
*dest_dentry
= nd
->dentry
;
823 struct vfsmount
*child
, *p
;
825 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
828 if (IS_MNT_SHARED(dest_mnt
)) {
829 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
833 spin_lock(&vfsmount_lock
);
835 detach_mnt(source_mnt
, parent_nd
);
836 attach_mnt(source_mnt
, nd
);
837 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
839 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
840 commit_tree(source_mnt
);
843 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
844 list_del_init(&child
->mnt_hash
);
847 spin_unlock(&vfsmount_lock
);
851 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
854 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
857 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
858 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
862 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
863 if (IS_DEADDIR(nd
->dentry
->d_inode
))
866 err
= security_sb_check_sb(mnt
, nd
);
871 if (IS_ROOT(nd
->dentry
) || !d_unhashed(nd
->dentry
))
872 err
= attach_recursive_mnt(mnt
, nd
, NULL
);
874 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
876 security_sb_post_addmount(mnt
, nd
);
881 * recursively change the type of the mountpoint.
883 static int do_change_type(struct nameidata
*nd
, int flag
)
885 struct vfsmount
*m
, *mnt
= nd
->mnt
;
886 int recurse
= flag
& MS_REC
;
887 int type
= flag
& ~MS_REC
;
889 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
892 down_write(&namespace_sem
);
893 spin_lock(&vfsmount_lock
);
894 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
895 change_mnt_propagation(m
, type
);
896 spin_unlock(&vfsmount_lock
);
897 up_write(&namespace_sem
);
904 static int do_loopback(struct nameidata
*nd
, char *old_name
, int recurse
)
906 struct nameidata old_nd
;
907 struct vfsmount
*mnt
= NULL
;
908 int err
= mount_is_safe(nd
);
911 if (!old_name
|| !*old_name
)
913 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
917 down_write(&namespace_sem
);
919 if (IS_MNT_UNBINDABLE(old_nd
.mnt
))
922 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
927 mnt
= copy_tree(old_nd
.mnt
, old_nd
.dentry
, 0);
929 mnt
= clone_mnt(old_nd
.mnt
, old_nd
.dentry
, 0);
934 err
= graft_tree(mnt
, nd
);
936 LIST_HEAD(umount_list
);
937 spin_lock(&vfsmount_lock
);
938 umount_tree(mnt
, 0, &umount_list
);
939 spin_unlock(&vfsmount_lock
);
940 release_mounts(&umount_list
);
944 up_write(&namespace_sem
);
945 path_release(&old_nd
);
950 * change filesystem flags. dir should be a physical root of filesystem.
951 * If you've mounted a non-root directory somewhere and want to do remount
952 * on it - tough luck.
954 static int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
958 struct super_block
*sb
= nd
->mnt
->mnt_sb
;
960 if (!capable(CAP_SYS_ADMIN
))
963 if (!check_mnt(nd
->mnt
))
966 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
969 down_write(&sb
->s_umount
);
970 err
= do_remount_sb(sb
, flags
, data
, 0);
972 nd
->mnt
->mnt_flags
= mnt_flags
;
973 up_write(&sb
->s_umount
);
975 security_sb_post_remount(nd
->mnt
, flags
, data
);
979 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
982 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
983 if (IS_MNT_UNBINDABLE(p
))
989 static int do_move_mount(struct nameidata
*nd
, char *old_name
)
991 struct nameidata old_nd
, parent_nd
;
994 if (!capable(CAP_SYS_ADMIN
))
996 if (!old_name
|| !*old_name
)
998 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1002 down_write(&namespace_sem
);
1003 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1006 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
1010 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
1011 if (IS_DEADDIR(nd
->dentry
->d_inode
))
1014 if (!IS_ROOT(nd
->dentry
) && d_unhashed(nd
->dentry
))
1018 if (old_nd
.dentry
!= old_nd
.mnt
->mnt_root
)
1021 if (old_nd
.mnt
== old_nd
.mnt
->mnt_parent
)
1024 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
1025 S_ISDIR(old_nd
.dentry
->d_inode
->i_mode
))
1028 * Don't move a mount residing in a shared parent.
1030 if (old_nd
.mnt
->mnt_parent
&& IS_MNT_SHARED(old_nd
.mnt
->mnt_parent
))
1033 * Don't move a mount tree containing unbindable mounts to a destination
1034 * mount which is shared.
1036 if (IS_MNT_SHARED(nd
->mnt
) && tree_contains_unbindable(old_nd
.mnt
))
1039 for (p
= nd
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1040 if (p
== old_nd
.mnt
)
1043 if ((err
= attach_recursive_mnt(old_nd
.mnt
, nd
, &parent_nd
)))
1046 spin_lock(&vfsmount_lock
);
1047 /* if the mount is moved, it should no longer be expire
1049 list_del_init(&old_nd
.mnt
->mnt_expire
);
1050 spin_unlock(&vfsmount_lock
);
1052 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
1054 up_write(&namespace_sem
);
1056 path_release(&parent_nd
);
1057 path_release(&old_nd
);
1062 * create a new mount for userspace and request it to be added into the
1065 static int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1066 int mnt_flags
, char *name
, void *data
)
1068 struct vfsmount
*mnt
;
1070 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1073 /* we need capabilities... */
1074 if (!capable(CAP_SYS_ADMIN
))
1077 mnt
= do_kern_mount(type
, flags
, name
, data
);
1079 return PTR_ERR(mnt
);
1081 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1085 * add a mount into a namespace's mount tree
1086 * - provide the option of adding the new mount to an expiration list
1088 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1089 int mnt_flags
, struct list_head
*fslist
)
1093 down_write(&namespace_sem
);
1094 /* Something was mounted here while we slept */
1095 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1098 if (!check_mnt(nd
->mnt
))
1101 /* Refuse the same filesystem on the same mount point */
1103 if (nd
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1104 nd
->mnt
->mnt_root
== nd
->dentry
)
1108 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1111 newmnt
->mnt_flags
= mnt_flags
;
1112 if ((err
= graft_tree(newmnt
, nd
)))
1116 /* add to the specified expiration list */
1117 spin_lock(&vfsmount_lock
);
1118 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1119 spin_unlock(&vfsmount_lock
);
1121 up_write(&namespace_sem
);
1125 up_write(&namespace_sem
);
1130 EXPORT_SYMBOL_GPL(do_add_mount
);
1132 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
1133 struct list_head
*umounts
)
1135 spin_lock(&vfsmount_lock
);
1138 * Check if mount is still attached, if not, let whoever holds it deal
1141 if (mnt
->mnt_parent
== mnt
) {
1142 spin_unlock(&vfsmount_lock
);
1147 * Check that it is still dead: the count should now be 2 - as
1148 * contributed by the vfsmount parent and the mntget above
1150 if (!propagate_mount_busy(mnt
, 2)) {
1151 /* delete from the namespace */
1152 touch_mnt_namespace(mnt
->mnt_ns
);
1153 list_del_init(&mnt
->mnt_list
);
1155 umount_tree(mnt
, 1, umounts
);
1156 spin_unlock(&vfsmount_lock
);
1159 * Someone brought it back to life whilst we didn't have any
1160 * locks held so return it to the expiration list
1162 list_add_tail(&mnt
->mnt_expire
, mounts
);
1163 spin_unlock(&vfsmount_lock
);
1168 * go through the vfsmounts we've just consigned to the graveyard to
1169 * - check that they're still dead
1170 * - delete the vfsmount from the appropriate namespace under lock
1171 * - dispose of the corpse
1173 static void expire_mount_list(struct list_head
*graveyard
, struct list_head
*mounts
)
1175 struct mnt_namespace
*ns
;
1176 struct vfsmount
*mnt
;
1178 while (!list_empty(graveyard
)) {
1180 mnt
= list_entry(graveyard
->next
, struct vfsmount
, mnt_expire
);
1181 list_del_init(&mnt
->mnt_expire
);
1183 /* don't do anything if the namespace is dead - all the
1184 * vfsmounts from it are going away anyway */
1186 if (!ns
|| !ns
->root
)
1190 spin_unlock(&vfsmount_lock
);
1191 down_write(&namespace_sem
);
1192 expire_mount(mnt
, mounts
, &umounts
);
1193 up_write(&namespace_sem
);
1194 release_mounts(&umounts
);
1197 spin_lock(&vfsmount_lock
);
1202 * process a list of expirable mountpoints with the intent of discarding any
1203 * mountpoints that aren't in use and haven't been touched since last we came
1206 void mark_mounts_for_expiry(struct list_head
*mounts
)
1208 struct vfsmount
*mnt
, *next
;
1209 LIST_HEAD(graveyard
);
1211 if (list_empty(mounts
))
1214 spin_lock(&vfsmount_lock
);
1216 /* extract from the expiration list every vfsmount that matches the
1217 * following criteria:
1218 * - only referenced by its parent vfsmount
1219 * - still marked for expiry (marked on the last call here; marks are
1220 * cleared by mntput())
1222 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1223 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1224 atomic_read(&mnt
->mnt_count
) != 1)
1228 list_move(&mnt
->mnt_expire
, &graveyard
);
1231 expire_mount_list(&graveyard
, mounts
);
1233 spin_unlock(&vfsmount_lock
);
1236 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1239 * Ripoff of 'select_parent()'
1241 * search the list of submounts for a given mountpoint, and move any
1242 * shrinkable submounts to the 'graveyard' list.
1244 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1246 struct vfsmount
*this_parent
= parent
;
1247 struct list_head
*next
;
1251 next
= this_parent
->mnt_mounts
.next
;
1253 while (next
!= &this_parent
->mnt_mounts
) {
1254 struct list_head
*tmp
= next
;
1255 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1258 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1261 * Descend a level if the d_mounts list is non-empty.
1263 if (!list_empty(&mnt
->mnt_mounts
)) {
1268 if (!propagate_mount_busy(mnt
, 1)) {
1270 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1275 * All done at this level ... ascend and resume the search
1277 if (this_parent
!= parent
) {
1278 next
= this_parent
->mnt_child
.next
;
1279 this_parent
= this_parent
->mnt_parent
;
1286 * process a list of expirable mountpoints with the intent of discarding any
1287 * submounts of a specific parent mountpoint
1289 void shrink_submounts(struct vfsmount
*mountpoint
, struct list_head
*mounts
)
1291 LIST_HEAD(graveyard
);
1294 spin_lock(&vfsmount_lock
);
1296 /* extract submounts of 'mountpoint' from the expiration list */
1297 while ((found
= select_submounts(mountpoint
, &graveyard
)) != 0)
1298 expire_mount_list(&graveyard
, mounts
);
1300 spin_unlock(&vfsmount_lock
);
1303 EXPORT_SYMBOL_GPL(shrink_submounts
);
1306 * Some copy_from_user() implementations do not return the exact number of
1307 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1308 * Note that this function differs from copy_from_user() in that it will oops
1309 * on bad values of `to', rather than returning a short copy.
1311 static long exact_copy_from_user(void *to
, const void __user
* from
,
1315 const char __user
*f
= from
;
1318 if (!access_ok(VERIFY_READ
, from
, n
))
1322 if (__get_user(c
, f
)) {
1333 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1343 if (!(page
= __get_free_page(GFP_KERNEL
)))
1346 /* We only care that *some* data at the address the user
1347 * gave us is valid. Just in case, we'll zero
1348 * the remainder of the page.
1350 /* copy_from_user cannot cross TASK_SIZE ! */
1351 size
= TASK_SIZE
- (unsigned long)data
;
1352 if (size
> PAGE_SIZE
)
1355 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1361 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1367 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1368 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1370 * data is a (void *) that can point to any structure up to
1371 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1372 * information (or be NULL).
1374 * Pre-0.97 versions of mount() didn't have a flags word.
1375 * When the flags word was introduced its top half was required
1376 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1377 * Therefore, if this magic number is present, it carries no information
1378 * and must be discarded.
1380 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1381 unsigned long flags
, void *data_page
)
1383 struct nameidata nd
;
1388 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1389 flags
&= ~MS_MGC_MSK
;
1391 /* Basic sanity checks */
1393 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1395 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1399 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1401 /* Separate the per-mountpoint flags */
1402 if (flags
& MS_NOSUID
)
1403 mnt_flags
|= MNT_NOSUID
;
1404 if (flags
& MS_NODEV
)
1405 mnt_flags
|= MNT_NODEV
;
1406 if (flags
& MS_NOEXEC
)
1407 mnt_flags
|= MNT_NOEXEC
;
1408 if (flags
& MS_NOATIME
)
1409 mnt_flags
|= MNT_NOATIME
;
1410 if (flags
& MS_NODIRATIME
)
1411 mnt_flags
|= MNT_NODIRATIME
;
1412 if (flags
& MS_RELATIME
)
1413 mnt_flags
|= MNT_RELATIME
;
1415 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1416 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
);
1418 /* ... and get the mountpoint */
1419 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1423 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1427 if (flags
& MS_REMOUNT
)
1428 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1430 else if (flags
& MS_BIND
)
1431 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1432 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1433 retval
= do_change_type(&nd
, flags
);
1434 else if (flags
& MS_MOVE
)
1435 retval
= do_move_mount(&nd
, dev_name
);
1437 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1438 dev_name
, data_page
);
1445 * Allocate a new namespace structure and populate it with contents
1446 * copied from the namespace of the passed in task structure.
1448 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1449 struct fs_struct
*fs
)
1451 struct mnt_namespace
*new_ns
;
1452 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1453 struct vfsmount
*p
, *q
;
1455 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1459 atomic_set(&new_ns
->count
, 1);
1460 INIT_LIST_HEAD(&new_ns
->list
);
1461 init_waitqueue_head(&new_ns
->poll
);
1464 down_write(&namespace_sem
);
1465 /* First pass: copy the tree topology */
1466 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1467 CL_COPY_ALL
| CL_EXPIRE
);
1468 if (!new_ns
->root
) {
1469 up_write(&namespace_sem
);
1473 spin_lock(&vfsmount_lock
);
1474 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1475 spin_unlock(&vfsmount_lock
);
1478 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1479 * as belonging to new namespace. We have already acquired a private
1480 * fs_struct, so tsk->fs->lock is not needed.
1487 if (p
== fs
->rootmnt
) {
1489 fs
->rootmnt
= mntget(q
);
1491 if (p
== fs
->pwdmnt
) {
1493 fs
->pwdmnt
= mntget(q
);
1495 if (p
== fs
->altrootmnt
) {
1497 fs
->altrootmnt
= mntget(q
);
1500 p
= next_mnt(p
, mnt_ns
->root
);
1501 q
= next_mnt(q
, new_ns
->root
);
1503 up_write(&namespace_sem
);
1515 struct mnt_namespace
*copy_mnt_ns(int flags
, struct mnt_namespace
*ns
,
1516 struct fs_struct
*new_fs
)
1518 struct mnt_namespace
*new_ns
;
1523 if (!(flags
& CLONE_NEWNS
))
1526 new_ns
= dup_mnt_ns(ns
, new_fs
);
1532 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1533 char __user
* type
, unsigned long flags
,
1537 unsigned long data_page
;
1538 unsigned long type_page
;
1539 unsigned long dev_page
;
1542 retval
= copy_mount_options(type
, &type_page
);
1546 dir_page
= getname(dir_name
);
1547 retval
= PTR_ERR(dir_page
);
1548 if (IS_ERR(dir_page
))
1551 retval
= copy_mount_options(dev_name
, &dev_page
);
1555 retval
= copy_mount_options(data
, &data_page
);
1560 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1561 flags
, (void *)data_page
);
1563 free_page(data_page
);
1566 free_page(dev_page
);
1570 free_page(type_page
);
1575 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1576 * It can block. Requires the big lock held.
1578 void set_fs_root(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1579 struct dentry
*dentry
)
1581 struct dentry
*old_root
;
1582 struct vfsmount
*old_rootmnt
;
1583 write_lock(&fs
->lock
);
1584 old_root
= fs
->root
;
1585 old_rootmnt
= fs
->rootmnt
;
1586 fs
->rootmnt
= mntget(mnt
);
1587 fs
->root
= dget(dentry
);
1588 write_unlock(&fs
->lock
);
1591 mntput(old_rootmnt
);
1596 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1597 * It can block. Requires the big lock held.
1599 void set_fs_pwd(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1600 struct dentry
*dentry
)
1602 struct dentry
*old_pwd
;
1603 struct vfsmount
*old_pwdmnt
;
1605 write_lock(&fs
->lock
);
1607 old_pwdmnt
= fs
->pwdmnt
;
1608 fs
->pwdmnt
= mntget(mnt
);
1609 fs
->pwd
= dget(dentry
);
1610 write_unlock(&fs
->lock
);
1618 static void chroot_fs_refs(struct nameidata
*old_nd
, struct nameidata
*new_nd
)
1620 struct task_struct
*g
, *p
;
1621 struct fs_struct
*fs
;
1623 read_lock(&tasklist_lock
);
1624 do_each_thread(g
, p
) {
1628 atomic_inc(&fs
->count
);
1630 if (fs
->root
== old_nd
->dentry
1631 && fs
->rootmnt
== old_nd
->mnt
)
1632 set_fs_root(fs
, new_nd
->mnt
, new_nd
->dentry
);
1633 if (fs
->pwd
== old_nd
->dentry
1634 && fs
->pwdmnt
== old_nd
->mnt
)
1635 set_fs_pwd(fs
, new_nd
->mnt
, new_nd
->dentry
);
1639 } while_each_thread(g
, p
);
1640 read_unlock(&tasklist_lock
);
1644 * pivot_root Semantics:
1645 * Moves the root file system of the current process to the directory put_old,
1646 * makes new_root as the new root file system of the current process, and sets
1647 * root/cwd of all processes which had them on the current root to new_root.
1650 * The new_root and put_old must be directories, and must not be on the
1651 * same file system as the current process root. The put_old must be
1652 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1653 * pointed to by put_old must yield the same directory as new_root. No other
1654 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1656 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1657 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1658 * in this situation.
1661 * - we don't move root/cwd if they are not at the root (reason: if something
1662 * cared enough to change them, it's probably wrong to force them elsewhere)
1663 * - it's okay to pick a root that isn't the root of a file system, e.g.
1664 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1665 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1668 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1669 const char __user
* put_old
)
1671 struct vfsmount
*tmp
;
1672 struct nameidata new_nd
, old_nd
, parent_nd
, root_parent
, user_nd
;
1675 if (!capable(CAP_SYS_ADMIN
))
1680 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1685 if (!check_mnt(new_nd
.mnt
))
1688 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1692 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1694 path_release(&old_nd
);
1698 read_lock(¤t
->fs
->lock
);
1699 user_nd
.mnt
= mntget(current
->fs
->rootmnt
);
1700 user_nd
.dentry
= dget(current
->fs
->root
);
1701 read_unlock(¤t
->fs
->lock
);
1702 down_write(&namespace_sem
);
1703 mutex_lock(&old_nd
.dentry
->d_inode
->i_mutex
);
1705 if (IS_MNT_SHARED(old_nd
.mnt
) ||
1706 IS_MNT_SHARED(new_nd
.mnt
->mnt_parent
) ||
1707 IS_MNT_SHARED(user_nd
.mnt
->mnt_parent
))
1709 if (!check_mnt(user_nd
.mnt
))
1712 if (IS_DEADDIR(new_nd
.dentry
->d_inode
))
1714 if (d_unhashed(new_nd
.dentry
) && !IS_ROOT(new_nd
.dentry
))
1716 if (d_unhashed(old_nd
.dentry
) && !IS_ROOT(old_nd
.dentry
))
1719 if (new_nd
.mnt
== user_nd
.mnt
|| old_nd
.mnt
== user_nd
.mnt
)
1720 goto out2
; /* loop, on the same file system */
1722 if (user_nd
.mnt
->mnt_root
!= user_nd
.dentry
)
1723 goto out2
; /* not a mountpoint */
1724 if (user_nd
.mnt
->mnt_parent
== user_nd
.mnt
)
1725 goto out2
; /* not attached */
1726 if (new_nd
.mnt
->mnt_root
!= new_nd
.dentry
)
1727 goto out2
; /* not a mountpoint */
1728 if (new_nd
.mnt
->mnt_parent
== new_nd
.mnt
)
1729 goto out2
; /* not attached */
1730 tmp
= old_nd
.mnt
; /* make sure we can reach put_old from new_root */
1731 spin_lock(&vfsmount_lock
);
1732 if (tmp
!= new_nd
.mnt
) {
1734 if (tmp
->mnt_parent
== tmp
)
1735 goto out3
; /* already mounted on put_old */
1736 if (tmp
->mnt_parent
== new_nd
.mnt
)
1738 tmp
= tmp
->mnt_parent
;
1740 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.dentry
))
1742 } else if (!is_subdir(old_nd
.dentry
, new_nd
.dentry
))
1744 detach_mnt(new_nd
.mnt
, &parent_nd
);
1745 detach_mnt(user_nd
.mnt
, &root_parent
);
1746 attach_mnt(user_nd
.mnt
, &old_nd
); /* mount old root on put_old */
1747 attach_mnt(new_nd
.mnt
, &root_parent
); /* mount new_root on / */
1748 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1749 spin_unlock(&vfsmount_lock
);
1750 chroot_fs_refs(&user_nd
, &new_nd
);
1751 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1753 path_release(&root_parent
);
1754 path_release(&parent_nd
);
1756 mutex_unlock(&old_nd
.dentry
->d_inode
->i_mutex
);
1757 up_write(&namespace_sem
);
1758 path_release(&user_nd
);
1759 path_release(&old_nd
);
1761 path_release(&new_nd
);
1766 spin_unlock(&vfsmount_lock
);
1770 static void __init
init_mount_tree(void)
1772 struct vfsmount
*mnt
;
1773 struct mnt_namespace
*ns
;
1775 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1777 panic("Can't create rootfs");
1778 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1780 panic("Can't allocate initial namespace");
1781 atomic_set(&ns
->count
, 1);
1782 INIT_LIST_HEAD(&ns
->list
);
1783 init_waitqueue_head(&ns
->poll
);
1785 list_add(&mnt
->mnt_list
, &ns
->list
);
1789 init_task
.nsproxy
->mnt_ns
= ns
;
1792 set_fs_pwd(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1793 set_fs_root(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1796 void __init
mnt_init(unsigned long mempages
)
1798 struct list_head
*d
;
1799 unsigned int nr_hash
;
1803 init_rwsem(&namespace_sem
);
1805 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1806 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
, NULL
);
1808 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1810 if (!mount_hashtable
)
1811 panic("Failed to allocate mount hash table\n");
1814 * Find the power-of-two list-heads that can fit into the allocation..
1815 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1818 nr_hash
= PAGE_SIZE
/ sizeof(struct list_head
);
1822 } while ((nr_hash
>> hash_bits
) != 0);
1826 * Re-calculate the actual number of entries and the mask
1827 * from the number of bits we can fit.
1829 nr_hash
= 1UL << hash_bits
;
1830 hash_mask
= nr_hash
- 1;
1832 printk("Mount-cache hash table entries: %d\n", nr_hash
);
1834 /* And initialize the newly allocated array */
1835 d
= mount_hashtable
;
1844 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1846 err
= subsystem_register(&fs_subsys
);
1848 printk(KERN_WARNING
"%s: subsystem_register error: %d\n",
1854 void __put_mnt_ns(struct mnt_namespace
*ns
)
1856 struct vfsmount
*root
= ns
->root
;
1857 LIST_HEAD(umount_list
);
1859 spin_unlock(&vfsmount_lock
);
1860 down_write(&namespace_sem
);
1861 spin_lock(&vfsmount_lock
);
1862 umount_tree(root
, 0, &umount_list
);
1863 spin_unlock(&vfsmount_lock
);
1864 up_write(&namespace_sem
);
1865 release_mounts(&umount_list
);