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>
34 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35 #define HASH_SIZE (1UL << HASH_SHIFT)
37 /* spinlock for vfsmount related operations, inplace of dcache_lock */
38 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
42 static struct list_head
*mount_hashtable __read_mostly
;
43 static struct kmem_cache
*mnt_cache __read_mostly
;
44 static struct rw_semaphore namespace_sem
;
47 struct kobject
*fs_kobj
;
48 EXPORT_SYMBOL_GPL(fs_kobj
);
50 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
52 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
53 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
54 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
55 return tmp
& (HASH_SIZE
- 1);
58 struct vfsmount
*alloc_vfsmnt(const char *name
)
60 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
62 atomic_set(&mnt
->mnt_count
, 1);
63 INIT_LIST_HEAD(&mnt
->mnt_hash
);
64 INIT_LIST_HEAD(&mnt
->mnt_child
);
65 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
66 INIT_LIST_HEAD(&mnt
->mnt_list
);
67 INIT_LIST_HEAD(&mnt
->mnt_expire
);
68 INIT_LIST_HEAD(&mnt
->mnt_share
);
69 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
70 INIT_LIST_HEAD(&mnt
->mnt_slave
);
72 int size
= strlen(name
) + 1;
73 char *newname
= kmalloc(size
, GFP_KERNEL
);
75 memcpy(newname
, name
, size
);
76 mnt
->mnt_devname
= newname
;
83 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
86 mnt
->mnt_root
= dget(sb
->s_root
);
90 EXPORT_SYMBOL(simple_set_mnt
);
92 void free_vfsmnt(struct vfsmount
*mnt
)
94 kfree(mnt
->mnt_devname
);
95 kmem_cache_free(mnt_cache
, mnt
);
99 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
102 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
105 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
106 struct list_head
*tmp
= head
;
107 struct vfsmount
*p
, *found
= NULL
;
110 tmp
= dir
? tmp
->next
: tmp
->prev
;
114 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
115 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
127 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
129 struct vfsmount
*child_mnt
;
130 spin_lock(&vfsmount_lock
);
131 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
133 spin_unlock(&vfsmount_lock
);
137 static inline int check_mnt(struct vfsmount
*mnt
)
139 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
142 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
146 wake_up_interruptible(&ns
->poll
);
150 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
152 if (ns
&& ns
->event
!= event
) {
154 wake_up_interruptible(&ns
->poll
);
158 static void detach_mnt(struct vfsmount
*mnt
, struct nameidata
*old_nd
)
160 old_nd
->dentry
= mnt
->mnt_mountpoint
;
161 old_nd
->mnt
= mnt
->mnt_parent
;
162 mnt
->mnt_parent
= mnt
;
163 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
164 list_del_init(&mnt
->mnt_child
);
165 list_del_init(&mnt
->mnt_hash
);
166 old_nd
->dentry
->d_mounted
--;
169 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
170 struct vfsmount
*child_mnt
)
172 child_mnt
->mnt_parent
= mntget(mnt
);
173 child_mnt
->mnt_mountpoint
= dget(dentry
);
177 static void attach_mnt(struct vfsmount
*mnt
, struct nameidata
*nd
)
179 mnt_set_mountpoint(nd
->mnt
, nd
->dentry
, mnt
);
180 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
181 hash(nd
->mnt
, nd
->dentry
));
182 list_add_tail(&mnt
->mnt_child
, &nd
->mnt
->mnt_mounts
);
186 * the caller must hold vfsmount_lock
188 static void commit_tree(struct vfsmount
*mnt
)
190 struct vfsmount
*parent
= mnt
->mnt_parent
;
193 struct mnt_namespace
*n
= parent
->mnt_ns
;
195 BUG_ON(parent
== mnt
);
197 list_add_tail(&head
, &mnt
->mnt_list
);
198 list_for_each_entry(m
, &head
, mnt_list
)
200 list_splice(&head
, n
->list
.prev
);
202 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
203 hash(parent
, mnt
->mnt_mountpoint
));
204 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
205 touch_mnt_namespace(n
);
208 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
210 struct list_head
*next
= p
->mnt_mounts
.next
;
211 if (next
== &p
->mnt_mounts
) {
215 next
= p
->mnt_child
.next
;
216 if (next
!= &p
->mnt_parent
->mnt_mounts
)
221 return list_entry(next
, struct vfsmount
, mnt_child
);
224 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
226 struct list_head
*prev
= p
->mnt_mounts
.prev
;
227 while (prev
!= &p
->mnt_mounts
) {
228 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
229 prev
= p
->mnt_mounts
.prev
;
234 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
237 struct super_block
*sb
= old
->mnt_sb
;
238 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
241 mnt
->mnt_flags
= old
->mnt_flags
;
242 atomic_inc(&sb
->s_active
);
244 mnt
->mnt_root
= dget(root
);
245 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
246 mnt
->mnt_parent
= mnt
;
248 if (flag
& CL_SLAVE
) {
249 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
250 mnt
->mnt_master
= old
;
251 CLEAR_MNT_SHARED(mnt
);
252 } else if (!(flag
& CL_PRIVATE
)) {
253 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
254 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
255 if (IS_MNT_SLAVE(old
))
256 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
257 mnt
->mnt_master
= old
->mnt_master
;
259 if (flag
& CL_MAKE_SHARED
)
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag
& CL_EXPIRE
) {
265 spin_lock(&vfsmount_lock
);
266 if (!list_empty(&old
->mnt_expire
))
267 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
268 spin_unlock(&vfsmount_lock
);
274 static inline void __mntput(struct vfsmount
*mnt
)
276 struct super_block
*sb
= mnt
->mnt_sb
;
279 deactivate_super(sb
);
282 void mntput_no_expire(struct vfsmount
*mnt
)
285 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
286 if (likely(!mnt
->mnt_pinned
)) {
287 spin_unlock(&vfsmount_lock
);
291 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
293 spin_unlock(&vfsmount_lock
);
294 acct_auto_close_mnt(mnt
);
295 security_sb_umount_close(mnt
);
300 EXPORT_SYMBOL(mntput_no_expire
);
302 void mnt_pin(struct vfsmount
*mnt
)
304 spin_lock(&vfsmount_lock
);
306 spin_unlock(&vfsmount_lock
);
309 EXPORT_SYMBOL(mnt_pin
);
311 void mnt_unpin(struct vfsmount
*mnt
)
313 spin_lock(&vfsmount_lock
);
314 if (mnt
->mnt_pinned
) {
315 atomic_inc(&mnt
->mnt_count
);
318 spin_unlock(&vfsmount_lock
);
321 EXPORT_SYMBOL(mnt_unpin
);
323 static inline void mangle(struct seq_file
*m
, const char *s
)
325 seq_escape(m
, s
, " \t\n\\");
329 * Simple .show_options callback for filesystems which don't want to
330 * implement more complex mount option showing.
332 * See also save_mount_options().
334 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
336 const char *options
= mnt
->mnt_sb
->s_options
;
338 if (options
!= NULL
&& options
[0]) {
345 EXPORT_SYMBOL(generic_show_options
);
348 * If filesystem uses generic_show_options(), this function should be
349 * called from the fill_super() callback.
351 * The .remount_fs callback usually needs to be handled in a special
352 * way, to make sure, that previous options are not overwritten if the
355 * Also note, that if the filesystem's .remount_fs function doesn't
356 * reset all options to their default value, but changes only newly
357 * given options, then the displayed options will not reflect reality
360 void save_mount_options(struct super_block
*sb
, char *options
)
362 kfree(sb
->s_options
);
363 sb
->s_options
= kstrdup(options
, GFP_KERNEL
);
365 EXPORT_SYMBOL(save_mount_options
);
368 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
370 struct mnt_namespace
*n
= m
->private;
372 down_read(&namespace_sem
);
373 return seq_list_start(&n
->list
, *pos
);
376 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
378 struct mnt_namespace
*n
= m
->private;
380 return seq_list_next(v
, &n
->list
, pos
);
383 static void m_stop(struct seq_file
*m
, void *v
)
385 up_read(&namespace_sem
);
388 static int show_vfsmnt(struct seq_file
*m
, void *v
)
390 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
392 static struct proc_fs_info
{
396 { MS_SYNCHRONOUS
, ",sync" },
397 { MS_DIRSYNC
, ",dirsync" },
398 { MS_MANDLOCK
, ",mand" },
401 static struct proc_fs_info mnt_info
[] = {
402 { MNT_NOSUID
, ",nosuid" },
403 { MNT_NODEV
, ",nodev" },
404 { MNT_NOEXEC
, ",noexec" },
405 { MNT_NOATIME
, ",noatime" },
406 { MNT_NODIRATIME
, ",nodiratime" },
407 { MNT_RELATIME
, ",relatime" },
410 struct proc_fs_info
*fs_infop
;
412 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
414 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
416 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
417 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
419 mangle(m
, mnt
->mnt_sb
->s_subtype
);
421 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
422 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
423 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
424 seq_puts(m
, fs_infop
->str
);
426 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
427 if (mnt
->mnt_flags
& fs_infop
->flag
)
428 seq_puts(m
, fs_infop
->str
);
430 if (mnt
->mnt_sb
->s_op
->show_options
)
431 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
432 seq_puts(m
, " 0 0\n");
436 struct seq_operations mounts_op
= {
443 static int show_vfsstat(struct seq_file
*m
, void *v
)
445 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
449 if (mnt
->mnt_devname
) {
450 seq_puts(m
, "device ");
451 mangle(m
, mnt
->mnt_devname
);
453 seq_puts(m
, "no device");
456 seq_puts(m
, " mounted on ");
457 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
460 /* file system type */
461 seq_puts(m
, "with fstype ");
462 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
464 /* optional statistics */
465 if (mnt
->mnt_sb
->s_op
->show_stats
) {
467 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
474 struct seq_operations mountstats_op
= {
478 .show
= show_vfsstat
,
482 * may_umount_tree - check if a mount tree is busy
483 * @mnt: root of mount tree
485 * This is called to check if a tree of mounts has any
486 * open files, pwds, chroots or sub mounts that are
489 int may_umount_tree(struct vfsmount
*mnt
)
492 int minimum_refs
= 0;
495 spin_lock(&vfsmount_lock
);
496 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
497 actual_refs
+= atomic_read(&p
->mnt_count
);
500 spin_unlock(&vfsmount_lock
);
502 if (actual_refs
> minimum_refs
)
508 EXPORT_SYMBOL(may_umount_tree
);
511 * may_umount - check if a mount point is busy
512 * @mnt: root of mount
514 * This is called to check if a mount point has any
515 * open files, pwds, chroots or sub mounts. If the
516 * mount has sub mounts this will return busy
517 * regardless of whether the sub mounts are busy.
519 * Doesn't take quota and stuff into account. IOW, in some cases it will
520 * give false negatives. The main reason why it's here is that we need
521 * a non-destructive way to look for easily umountable filesystems.
523 int may_umount(struct vfsmount
*mnt
)
526 spin_lock(&vfsmount_lock
);
527 if (propagate_mount_busy(mnt
, 2))
529 spin_unlock(&vfsmount_lock
);
533 EXPORT_SYMBOL(may_umount
);
535 void release_mounts(struct list_head
*head
)
537 struct vfsmount
*mnt
;
538 while (!list_empty(head
)) {
539 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
540 list_del_init(&mnt
->mnt_hash
);
541 if (mnt
->mnt_parent
!= mnt
) {
542 struct dentry
*dentry
;
544 spin_lock(&vfsmount_lock
);
545 dentry
= mnt
->mnt_mountpoint
;
547 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
548 mnt
->mnt_parent
= mnt
;
549 spin_unlock(&vfsmount_lock
);
557 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
561 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
562 list_move(&p
->mnt_hash
, kill
);
565 propagate_umount(kill
);
567 list_for_each_entry(p
, kill
, mnt_hash
) {
568 list_del_init(&p
->mnt_expire
);
569 list_del_init(&p
->mnt_list
);
570 __touch_mnt_namespace(p
->mnt_ns
);
572 list_del_init(&p
->mnt_child
);
573 if (p
->mnt_parent
!= p
)
574 p
->mnt_mountpoint
->d_mounted
--;
575 change_mnt_propagation(p
, MS_PRIVATE
);
579 static int do_umount(struct vfsmount
*mnt
, int flags
)
581 struct super_block
*sb
= mnt
->mnt_sb
;
583 LIST_HEAD(umount_list
);
585 retval
= security_sb_umount(mnt
, flags
);
590 * Allow userspace to request a mountpoint be expired rather than
591 * unmounting unconditionally. Unmount only happens if:
592 * (1) the mark is already set (the mark is cleared by mntput())
593 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
595 if (flags
& MNT_EXPIRE
) {
596 if (mnt
== current
->fs
->rootmnt
||
597 flags
& (MNT_FORCE
| MNT_DETACH
))
600 if (atomic_read(&mnt
->mnt_count
) != 2)
603 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
608 * If we may have to abort operations to get out of this
609 * mount, and they will themselves hold resources we must
610 * allow the fs to do things. In the Unix tradition of
611 * 'Gee thats tricky lets do it in userspace' the umount_begin
612 * might fail to complete on the first run through as other tasks
613 * must return, and the like. Thats for the mount program to worry
614 * about for the moment.
618 if (sb
->s_op
->umount_begin
)
619 sb
->s_op
->umount_begin(mnt
, flags
);
623 * No sense to grab the lock for this test, but test itself looks
624 * somewhat bogus. Suggestions for better replacement?
625 * Ho-hum... In principle, we might treat that as umount + switch
626 * to rootfs. GC would eventually take care of the old vfsmount.
627 * Actually it makes sense, especially if rootfs would contain a
628 * /reboot - static binary that would close all descriptors and
629 * call reboot(9). Then init(8) could umount root and exec /reboot.
631 if (mnt
== current
->fs
->rootmnt
&& !(flags
& MNT_DETACH
)) {
633 * Special case for "unmounting" root ...
634 * we just try to remount it readonly.
636 down_write(&sb
->s_umount
);
637 if (!(sb
->s_flags
& MS_RDONLY
)) {
640 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
643 up_write(&sb
->s_umount
);
647 down_write(&namespace_sem
);
648 spin_lock(&vfsmount_lock
);
652 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
653 if (!list_empty(&mnt
->mnt_list
))
654 umount_tree(mnt
, 1, &umount_list
);
657 spin_unlock(&vfsmount_lock
);
659 security_sb_umount_busy(mnt
);
660 up_write(&namespace_sem
);
661 release_mounts(&umount_list
);
666 * Now umount can handle mount points as well as block devices.
667 * This is important for filesystems which use unnamed block devices.
669 * We now support a flag for forced unmount like the other 'big iron'
670 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
673 asmlinkage
long sys_umount(char __user
* name
, int flags
)
678 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
682 if (nd
.dentry
!= nd
.mnt
->mnt_root
)
684 if (!check_mnt(nd
.mnt
))
688 if (!capable(CAP_SYS_ADMIN
))
691 retval
= do_umount(nd
.mnt
, flags
);
693 path_release_on_umount(&nd
);
698 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
701 * The 2.0 compatible umount. No flags.
703 asmlinkage
long sys_oldumount(char __user
* name
)
705 return sys_umount(name
, 0);
710 static int mount_is_safe(struct nameidata
*nd
)
712 if (capable(CAP_SYS_ADMIN
))
716 if (S_ISLNK(nd
->dentry
->d_inode
->i_mode
))
718 if (nd
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
719 if (current
->uid
!= nd
->dentry
->d_inode
->i_uid
)
722 if (vfs_permission(nd
, MAY_WRITE
))
728 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
733 if (d
== NULL
|| d
== d
->d_parent
)
739 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
742 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
745 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
748 res
= q
= clone_mnt(mnt
, dentry
, flag
);
751 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
754 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
755 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
758 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
759 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
760 s
= skip_mnt_tree(s
);
763 while (p
!= s
->mnt_parent
) {
769 nd
.dentry
= p
->mnt_mountpoint
;
770 q
= clone_mnt(p
, p
->mnt_root
, flag
);
773 spin_lock(&vfsmount_lock
);
774 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
776 spin_unlock(&vfsmount_lock
);
782 LIST_HEAD(umount_list
);
783 spin_lock(&vfsmount_lock
);
784 umount_tree(res
, 0, &umount_list
);
785 spin_unlock(&vfsmount_lock
);
786 release_mounts(&umount_list
);
791 struct vfsmount
*collect_mounts(struct vfsmount
*mnt
, struct dentry
*dentry
)
793 struct vfsmount
*tree
;
794 down_read(&namespace_sem
);
795 tree
= copy_tree(mnt
, dentry
, CL_COPY_ALL
| CL_PRIVATE
);
796 up_read(&namespace_sem
);
800 void drop_collected_mounts(struct vfsmount
*mnt
)
802 LIST_HEAD(umount_list
);
803 down_read(&namespace_sem
);
804 spin_lock(&vfsmount_lock
);
805 umount_tree(mnt
, 0, &umount_list
);
806 spin_unlock(&vfsmount_lock
);
807 up_read(&namespace_sem
);
808 release_mounts(&umount_list
);
812 * @source_mnt : mount tree to be attached
813 * @nd : place the mount tree @source_mnt is attached
814 * @parent_nd : if non-null, detach the source_mnt from its parent and
815 * store the parent mount and mountpoint dentry.
816 * (done when source_mnt is moved)
818 * NOTE: in the table below explains the semantics when a source mount
819 * of a given type is attached to a destination mount of a given type.
820 * ---------------------------------------------------------------------------
821 * | BIND MOUNT OPERATION |
822 * |**************************************************************************
823 * | source-->| shared | private | slave | unbindable |
827 * |**************************************************************************
828 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
830 * |non-shared| shared (+) | private | slave (*) | invalid |
831 * ***************************************************************************
832 * A bind operation clones the source mount and mounts the clone on the
835 * (++) the cloned mount is propagated to all the mounts in the propagation
836 * tree of the destination mount and the cloned mount is added to
837 * the peer group of the source mount.
838 * (+) the cloned mount is created under the destination mount and is marked
839 * as shared. The cloned mount is added to the peer group of the source
841 * (+++) the mount is propagated to all the mounts in the propagation tree
842 * of the destination mount and the cloned mount is made slave
843 * of the same master as that of the source mount. The cloned mount
844 * is marked as 'shared and slave'.
845 * (*) the cloned mount is made a slave of the same master as that of the
848 * ---------------------------------------------------------------------------
849 * | MOVE MOUNT OPERATION |
850 * |**************************************************************************
851 * | source-->| shared | private | slave | unbindable |
855 * |**************************************************************************
856 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
858 * |non-shared| shared (+*) | private | slave (*) | unbindable |
859 * ***************************************************************************
861 * (+) the mount is moved to the destination. And is then propagated to
862 * all the mounts in the propagation tree of the destination mount.
863 * (+*) the mount is moved to the destination.
864 * (+++) the mount is moved to the destination and is then propagated to
865 * all the mounts belonging to the destination mount's propagation tree.
866 * the mount is marked as 'shared and slave'.
867 * (*) the mount continues to be a slave at the new location.
869 * if the source mount is a tree, the operations explained above is
870 * applied to each mount in the tree.
871 * Must be called without spinlocks held, since this function can sleep
874 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
875 struct nameidata
*nd
, struct nameidata
*parent_nd
)
877 LIST_HEAD(tree_list
);
878 struct vfsmount
*dest_mnt
= nd
->mnt
;
879 struct dentry
*dest_dentry
= nd
->dentry
;
880 struct vfsmount
*child
, *p
;
882 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
885 if (IS_MNT_SHARED(dest_mnt
)) {
886 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
890 spin_lock(&vfsmount_lock
);
892 detach_mnt(source_mnt
, parent_nd
);
893 attach_mnt(source_mnt
, nd
);
894 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
896 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
897 commit_tree(source_mnt
);
900 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
901 list_del_init(&child
->mnt_hash
);
904 spin_unlock(&vfsmount_lock
);
908 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
911 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
914 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
915 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
919 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
920 if (IS_DEADDIR(nd
->dentry
->d_inode
))
923 err
= security_sb_check_sb(mnt
, nd
);
928 if (IS_ROOT(nd
->dentry
) || !d_unhashed(nd
->dentry
))
929 err
= attach_recursive_mnt(mnt
, nd
, NULL
);
931 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
933 security_sb_post_addmount(mnt
, nd
);
938 * recursively change the type of the mountpoint.
939 * noinline this do_mount helper to save do_mount stack space.
941 static noinline
int do_change_type(struct nameidata
*nd
, int flag
)
943 struct vfsmount
*m
, *mnt
= nd
->mnt
;
944 int recurse
= flag
& MS_REC
;
945 int type
= flag
& ~MS_REC
;
947 if (!capable(CAP_SYS_ADMIN
))
950 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
953 down_write(&namespace_sem
);
954 spin_lock(&vfsmount_lock
);
955 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
956 change_mnt_propagation(m
, type
);
957 spin_unlock(&vfsmount_lock
);
958 up_write(&namespace_sem
);
964 * noinline this do_mount helper to save do_mount stack space.
966 static noinline
int do_loopback(struct nameidata
*nd
, char *old_name
,
969 struct nameidata old_nd
;
970 struct vfsmount
*mnt
= NULL
;
971 int err
= mount_is_safe(nd
);
974 if (!old_name
|| !*old_name
)
976 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
980 down_write(&namespace_sem
);
982 if (IS_MNT_UNBINDABLE(old_nd
.mnt
))
985 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
990 mnt
= copy_tree(old_nd
.mnt
, old_nd
.dentry
, 0);
992 mnt
= clone_mnt(old_nd
.mnt
, old_nd
.dentry
, 0);
997 err
= graft_tree(mnt
, nd
);
999 LIST_HEAD(umount_list
);
1000 spin_lock(&vfsmount_lock
);
1001 umount_tree(mnt
, 0, &umount_list
);
1002 spin_unlock(&vfsmount_lock
);
1003 release_mounts(&umount_list
);
1007 up_write(&namespace_sem
);
1008 path_release(&old_nd
);
1013 * change filesystem flags. dir should be a physical root of filesystem.
1014 * If you've mounted a non-root directory somewhere and want to do remount
1015 * on it - tough luck.
1016 * noinline this do_mount helper to save do_mount stack space.
1018 static noinline
int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
1022 struct super_block
*sb
= nd
->mnt
->mnt_sb
;
1024 if (!capable(CAP_SYS_ADMIN
))
1027 if (!check_mnt(nd
->mnt
))
1030 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
1033 down_write(&sb
->s_umount
);
1034 err
= do_remount_sb(sb
, flags
, data
, 0);
1036 nd
->mnt
->mnt_flags
= mnt_flags
;
1037 up_write(&sb
->s_umount
);
1039 security_sb_post_remount(nd
->mnt
, flags
, data
);
1043 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1046 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1047 if (IS_MNT_UNBINDABLE(p
))
1054 * noinline this do_mount helper to save do_mount stack space.
1056 static noinline
int do_move_mount(struct nameidata
*nd
, char *old_name
)
1058 struct nameidata old_nd
, parent_nd
;
1061 if (!capable(CAP_SYS_ADMIN
))
1063 if (!old_name
|| !*old_name
)
1065 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1069 down_write(&namespace_sem
);
1070 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1073 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
1077 mutex_lock(&nd
->dentry
->d_inode
->i_mutex
);
1078 if (IS_DEADDIR(nd
->dentry
->d_inode
))
1081 if (!IS_ROOT(nd
->dentry
) && d_unhashed(nd
->dentry
))
1085 if (old_nd
.dentry
!= old_nd
.mnt
->mnt_root
)
1088 if (old_nd
.mnt
== old_nd
.mnt
->mnt_parent
)
1091 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
1092 S_ISDIR(old_nd
.dentry
->d_inode
->i_mode
))
1095 * Don't move a mount residing in a shared parent.
1097 if (old_nd
.mnt
->mnt_parent
&& IS_MNT_SHARED(old_nd
.mnt
->mnt_parent
))
1100 * Don't move a mount tree containing unbindable mounts to a destination
1101 * mount which is shared.
1103 if (IS_MNT_SHARED(nd
->mnt
) && tree_contains_unbindable(old_nd
.mnt
))
1106 for (p
= nd
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1107 if (p
== old_nd
.mnt
)
1110 if ((err
= attach_recursive_mnt(old_nd
.mnt
, nd
, &parent_nd
)))
1113 spin_lock(&vfsmount_lock
);
1114 /* if the mount is moved, it should no longer be expire
1116 list_del_init(&old_nd
.mnt
->mnt_expire
);
1117 spin_unlock(&vfsmount_lock
);
1119 mutex_unlock(&nd
->dentry
->d_inode
->i_mutex
);
1121 up_write(&namespace_sem
);
1123 path_release(&parent_nd
);
1124 path_release(&old_nd
);
1129 * create a new mount for userspace and request it to be added into the
1131 * noinline this do_mount helper to save do_mount stack space.
1133 static noinline
int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1134 int mnt_flags
, char *name
, void *data
)
1136 struct vfsmount
*mnt
;
1138 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1141 /* we need capabilities... */
1142 if (!capable(CAP_SYS_ADMIN
))
1145 mnt
= do_kern_mount(type
, flags
, name
, data
);
1147 return PTR_ERR(mnt
);
1149 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1153 * add a mount into a namespace's mount tree
1154 * - provide the option of adding the new mount to an expiration list
1156 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1157 int mnt_flags
, struct list_head
*fslist
)
1161 down_write(&namespace_sem
);
1162 /* Something was mounted here while we slept */
1163 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
1166 if (!check_mnt(nd
->mnt
))
1169 /* Refuse the same filesystem on the same mount point */
1171 if (nd
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1172 nd
->mnt
->mnt_root
== nd
->dentry
)
1176 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1179 newmnt
->mnt_flags
= mnt_flags
;
1180 if ((err
= graft_tree(newmnt
, nd
)))
1184 /* add to the specified expiration list */
1185 spin_lock(&vfsmount_lock
);
1186 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1187 spin_unlock(&vfsmount_lock
);
1189 up_write(&namespace_sem
);
1193 up_write(&namespace_sem
);
1198 EXPORT_SYMBOL_GPL(do_add_mount
);
1200 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
1201 struct list_head
*umounts
)
1203 spin_lock(&vfsmount_lock
);
1206 * Check if mount is still attached, if not, let whoever holds it deal
1209 if (mnt
->mnt_parent
== mnt
) {
1210 spin_unlock(&vfsmount_lock
);
1215 * Check that it is still dead: the count should now be 2 - as
1216 * contributed by the vfsmount parent and the mntget above
1218 if (!propagate_mount_busy(mnt
, 2)) {
1219 /* delete from the namespace */
1220 touch_mnt_namespace(mnt
->mnt_ns
);
1221 list_del_init(&mnt
->mnt_list
);
1223 umount_tree(mnt
, 1, umounts
);
1224 spin_unlock(&vfsmount_lock
);
1227 * Someone brought it back to life whilst we didn't have any
1228 * locks held so return it to the expiration list
1230 list_add_tail(&mnt
->mnt_expire
, mounts
);
1231 spin_unlock(&vfsmount_lock
);
1236 * go through the vfsmounts we've just consigned to the graveyard to
1237 * - check that they're still dead
1238 * - delete the vfsmount from the appropriate namespace under lock
1239 * - dispose of the corpse
1241 static void expire_mount_list(struct list_head
*graveyard
, struct list_head
*mounts
)
1243 struct mnt_namespace
*ns
;
1244 struct vfsmount
*mnt
;
1246 while (!list_empty(graveyard
)) {
1248 mnt
= list_first_entry(graveyard
, struct vfsmount
, mnt_expire
);
1249 list_del_init(&mnt
->mnt_expire
);
1251 /* don't do anything if the namespace is dead - all the
1252 * vfsmounts from it are going away anyway */
1254 if (!ns
|| !ns
->root
)
1258 spin_unlock(&vfsmount_lock
);
1259 down_write(&namespace_sem
);
1260 expire_mount(mnt
, mounts
, &umounts
);
1261 up_write(&namespace_sem
);
1262 release_mounts(&umounts
);
1265 spin_lock(&vfsmount_lock
);
1270 * process a list of expirable mountpoints with the intent of discarding any
1271 * mountpoints that aren't in use and haven't been touched since last we came
1274 void mark_mounts_for_expiry(struct list_head
*mounts
)
1276 struct vfsmount
*mnt
, *next
;
1277 LIST_HEAD(graveyard
);
1279 if (list_empty(mounts
))
1282 spin_lock(&vfsmount_lock
);
1284 /* extract from the expiration list every vfsmount that matches the
1285 * following criteria:
1286 * - only referenced by its parent vfsmount
1287 * - still marked for expiry (marked on the last call here; marks are
1288 * cleared by mntput())
1290 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1291 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1292 atomic_read(&mnt
->mnt_count
) != 1)
1296 list_move(&mnt
->mnt_expire
, &graveyard
);
1299 expire_mount_list(&graveyard
, mounts
);
1301 spin_unlock(&vfsmount_lock
);
1304 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1307 * Ripoff of 'select_parent()'
1309 * search the list of submounts for a given mountpoint, and move any
1310 * shrinkable submounts to the 'graveyard' list.
1312 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1314 struct vfsmount
*this_parent
= parent
;
1315 struct list_head
*next
;
1319 next
= this_parent
->mnt_mounts
.next
;
1321 while (next
!= &this_parent
->mnt_mounts
) {
1322 struct list_head
*tmp
= next
;
1323 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1326 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1329 * Descend a level if the d_mounts list is non-empty.
1331 if (!list_empty(&mnt
->mnt_mounts
)) {
1336 if (!propagate_mount_busy(mnt
, 1)) {
1338 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1343 * All done at this level ... ascend and resume the search
1345 if (this_parent
!= parent
) {
1346 next
= this_parent
->mnt_child
.next
;
1347 this_parent
= this_parent
->mnt_parent
;
1354 * process a list of expirable mountpoints with the intent of discarding any
1355 * submounts of a specific parent mountpoint
1357 void shrink_submounts(struct vfsmount
*mountpoint
, struct list_head
*mounts
)
1359 LIST_HEAD(graveyard
);
1362 spin_lock(&vfsmount_lock
);
1364 /* extract submounts of 'mountpoint' from the expiration list */
1365 while ((found
= select_submounts(mountpoint
, &graveyard
)) != 0)
1366 expire_mount_list(&graveyard
, mounts
);
1368 spin_unlock(&vfsmount_lock
);
1371 EXPORT_SYMBOL_GPL(shrink_submounts
);
1374 * Some copy_from_user() implementations do not return the exact number of
1375 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1376 * Note that this function differs from copy_from_user() in that it will oops
1377 * on bad values of `to', rather than returning a short copy.
1379 static long exact_copy_from_user(void *to
, const void __user
* from
,
1383 const char __user
*f
= from
;
1386 if (!access_ok(VERIFY_READ
, from
, n
))
1390 if (__get_user(c
, f
)) {
1401 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1411 if (!(page
= __get_free_page(GFP_KERNEL
)))
1414 /* We only care that *some* data at the address the user
1415 * gave us is valid. Just in case, we'll zero
1416 * the remainder of the page.
1418 /* copy_from_user cannot cross TASK_SIZE ! */
1419 size
= TASK_SIZE
- (unsigned long)data
;
1420 if (size
> PAGE_SIZE
)
1423 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1429 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1435 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1436 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1438 * data is a (void *) that can point to any structure up to
1439 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1440 * information (or be NULL).
1442 * Pre-0.97 versions of mount() didn't have a flags word.
1443 * When the flags word was introduced its top half was required
1444 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1445 * Therefore, if this magic number is present, it carries no information
1446 * and must be discarded.
1448 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1449 unsigned long flags
, void *data_page
)
1451 struct nameidata nd
;
1456 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1457 flags
&= ~MS_MGC_MSK
;
1459 /* Basic sanity checks */
1461 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1463 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1467 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1469 /* Separate the per-mountpoint flags */
1470 if (flags
& MS_NOSUID
)
1471 mnt_flags
|= MNT_NOSUID
;
1472 if (flags
& MS_NODEV
)
1473 mnt_flags
|= MNT_NODEV
;
1474 if (flags
& MS_NOEXEC
)
1475 mnt_flags
|= MNT_NOEXEC
;
1476 if (flags
& MS_NOATIME
)
1477 mnt_flags
|= MNT_NOATIME
;
1478 if (flags
& MS_NODIRATIME
)
1479 mnt_flags
|= MNT_NODIRATIME
;
1480 if (flags
& MS_RELATIME
)
1481 mnt_flags
|= MNT_RELATIME
;
1483 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1484 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
);
1486 /* ... and get the mountpoint */
1487 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1491 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1495 if (flags
& MS_REMOUNT
)
1496 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1498 else if (flags
& MS_BIND
)
1499 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1500 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1501 retval
= do_change_type(&nd
, flags
);
1502 else if (flags
& MS_MOVE
)
1503 retval
= do_move_mount(&nd
, dev_name
);
1505 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1506 dev_name
, data_page
);
1513 * Allocate a new namespace structure and populate it with contents
1514 * copied from the namespace of the passed in task structure.
1516 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1517 struct fs_struct
*fs
)
1519 struct mnt_namespace
*new_ns
;
1520 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1521 struct vfsmount
*p
, *q
;
1523 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1525 return ERR_PTR(-ENOMEM
);
1527 atomic_set(&new_ns
->count
, 1);
1528 INIT_LIST_HEAD(&new_ns
->list
);
1529 init_waitqueue_head(&new_ns
->poll
);
1532 down_write(&namespace_sem
);
1533 /* First pass: copy the tree topology */
1534 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1535 CL_COPY_ALL
| CL_EXPIRE
);
1536 if (!new_ns
->root
) {
1537 up_write(&namespace_sem
);
1539 return ERR_PTR(-ENOMEM
);;
1541 spin_lock(&vfsmount_lock
);
1542 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1543 spin_unlock(&vfsmount_lock
);
1546 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1547 * as belonging to new namespace. We have already acquired a private
1548 * fs_struct, so tsk->fs->lock is not needed.
1555 if (p
== fs
->rootmnt
) {
1557 fs
->rootmnt
= mntget(q
);
1559 if (p
== fs
->pwdmnt
) {
1561 fs
->pwdmnt
= mntget(q
);
1563 if (p
== fs
->altrootmnt
) {
1565 fs
->altrootmnt
= mntget(q
);
1568 p
= next_mnt(p
, mnt_ns
->root
);
1569 q
= next_mnt(q
, new_ns
->root
);
1571 up_write(&namespace_sem
);
1583 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
1584 struct fs_struct
*new_fs
)
1586 struct mnt_namespace
*new_ns
;
1591 if (!(flags
& CLONE_NEWNS
))
1594 new_ns
= dup_mnt_ns(ns
, new_fs
);
1600 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1601 char __user
* type
, unsigned long flags
,
1605 unsigned long data_page
;
1606 unsigned long type_page
;
1607 unsigned long dev_page
;
1610 retval
= copy_mount_options(type
, &type_page
);
1614 dir_page
= getname(dir_name
);
1615 retval
= PTR_ERR(dir_page
);
1616 if (IS_ERR(dir_page
))
1619 retval
= copy_mount_options(dev_name
, &dev_page
);
1623 retval
= copy_mount_options(data
, &data_page
);
1628 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1629 flags
, (void *)data_page
);
1631 free_page(data_page
);
1634 free_page(dev_page
);
1638 free_page(type_page
);
1643 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1644 * It can block. Requires the big lock held.
1646 void set_fs_root(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1647 struct dentry
*dentry
)
1649 struct dentry
*old_root
;
1650 struct vfsmount
*old_rootmnt
;
1651 write_lock(&fs
->lock
);
1652 old_root
= fs
->root
;
1653 old_rootmnt
= fs
->rootmnt
;
1654 fs
->rootmnt
= mntget(mnt
);
1655 fs
->root
= dget(dentry
);
1656 write_unlock(&fs
->lock
);
1659 mntput(old_rootmnt
);
1664 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1665 * It can block. Requires the big lock held.
1667 void set_fs_pwd(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1668 struct dentry
*dentry
)
1670 struct dentry
*old_pwd
;
1671 struct vfsmount
*old_pwdmnt
;
1673 write_lock(&fs
->lock
);
1675 old_pwdmnt
= fs
->pwdmnt
;
1676 fs
->pwdmnt
= mntget(mnt
);
1677 fs
->pwd
= dget(dentry
);
1678 write_unlock(&fs
->lock
);
1686 static void chroot_fs_refs(struct nameidata
*old_nd
, struct nameidata
*new_nd
)
1688 struct task_struct
*g
, *p
;
1689 struct fs_struct
*fs
;
1691 read_lock(&tasklist_lock
);
1692 do_each_thread(g
, p
) {
1696 atomic_inc(&fs
->count
);
1698 if (fs
->root
== old_nd
->dentry
1699 && fs
->rootmnt
== old_nd
->mnt
)
1700 set_fs_root(fs
, new_nd
->mnt
, new_nd
->dentry
);
1701 if (fs
->pwd
== old_nd
->dentry
1702 && fs
->pwdmnt
== old_nd
->mnt
)
1703 set_fs_pwd(fs
, new_nd
->mnt
, new_nd
->dentry
);
1707 } while_each_thread(g
, p
);
1708 read_unlock(&tasklist_lock
);
1712 * pivot_root Semantics:
1713 * Moves the root file system of the current process to the directory put_old,
1714 * makes new_root as the new root file system of the current process, and sets
1715 * root/cwd of all processes which had them on the current root to new_root.
1718 * The new_root and put_old must be directories, and must not be on the
1719 * same file system as the current process root. The put_old must be
1720 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1721 * pointed to by put_old must yield the same directory as new_root. No other
1722 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1724 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1725 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1726 * in this situation.
1729 * - we don't move root/cwd if they are not at the root (reason: if something
1730 * cared enough to change them, it's probably wrong to force them elsewhere)
1731 * - it's okay to pick a root that isn't the root of a file system, e.g.
1732 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1733 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1736 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1737 const char __user
* put_old
)
1739 struct vfsmount
*tmp
;
1740 struct nameidata new_nd
, old_nd
, parent_nd
, root_parent
, user_nd
;
1743 if (!capable(CAP_SYS_ADMIN
))
1748 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1753 if (!check_mnt(new_nd
.mnt
))
1756 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1760 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1762 path_release(&old_nd
);
1766 read_lock(¤t
->fs
->lock
);
1767 user_nd
.mnt
= mntget(current
->fs
->rootmnt
);
1768 user_nd
.dentry
= dget(current
->fs
->root
);
1769 read_unlock(¤t
->fs
->lock
);
1770 down_write(&namespace_sem
);
1771 mutex_lock(&old_nd
.dentry
->d_inode
->i_mutex
);
1773 if (IS_MNT_SHARED(old_nd
.mnt
) ||
1774 IS_MNT_SHARED(new_nd
.mnt
->mnt_parent
) ||
1775 IS_MNT_SHARED(user_nd
.mnt
->mnt_parent
))
1777 if (!check_mnt(user_nd
.mnt
))
1780 if (IS_DEADDIR(new_nd
.dentry
->d_inode
))
1782 if (d_unhashed(new_nd
.dentry
) && !IS_ROOT(new_nd
.dentry
))
1784 if (d_unhashed(old_nd
.dentry
) && !IS_ROOT(old_nd
.dentry
))
1787 if (new_nd
.mnt
== user_nd
.mnt
|| old_nd
.mnt
== user_nd
.mnt
)
1788 goto out2
; /* loop, on the same file system */
1790 if (user_nd
.mnt
->mnt_root
!= user_nd
.dentry
)
1791 goto out2
; /* not a mountpoint */
1792 if (user_nd
.mnt
->mnt_parent
== user_nd
.mnt
)
1793 goto out2
; /* not attached */
1794 if (new_nd
.mnt
->mnt_root
!= new_nd
.dentry
)
1795 goto out2
; /* not a mountpoint */
1796 if (new_nd
.mnt
->mnt_parent
== new_nd
.mnt
)
1797 goto out2
; /* not attached */
1798 tmp
= old_nd
.mnt
; /* make sure we can reach put_old from new_root */
1799 spin_lock(&vfsmount_lock
);
1800 if (tmp
!= new_nd
.mnt
) {
1802 if (tmp
->mnt_parent
== tmp
)
1803 goto out3
; /* already mounted on put_old */
1804 if (tmp
->mnt_parent
== new_nd
.mnt
)
1806 tmp
= tmp
->mnt_parent
;
1808 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.dentry
))
1810 } else if (!is_subdir(old_nd
.dentry
, new_nd
.dentry
))
1812 detach_mnt(new_nd
.mnt
, &parent_nd
);
1813 detach_mnt(user_nd
.mnt
, &root_parent
);
1814 attach_mnt(user_nd
.mnt
, &old_nd
); /* mount old root on put_old */
1815 attach_mnt(new_nd
.mnt
, &root_parent
); /* mount new_root on / */
1816 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1817 spin_unlock(&vfsmount_lock
);
1818 chroot_fs_refs(&user_nd
, &new_nd
);
1819 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1821 path_release(&root_parent
);
1822 path_release(&parent_nd
);
1824 mutex_unlock(&old_nd
.dentry
->d_inode
->i_mutex
);
1825 up_write(&namespace_sem
);
1826 path_release(&user_nd
);
1827 path_release(&old_nd
);
1829 path_release(&new_nd
);
1834 spin_unlock(&vfsmount_lock
);
1838 static void __init
init_mount_tree(void)
1840 struct vfsmount
*mnt
;
1841 struct mnt_namespace
*ns
;
1843 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1845 panic("Can't create rootfs");
1846 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1848 panic("Can't allocate initial namespace");
1849 atomic_set(&ns
->count
, 1);
1850 INIT_LIST_HEAD(&ns
->list
);
1851 init_waitqueue_head(&ns
->poll
);
1853 list_add(&mnt
->mnt_list
, &ns
->list
);
1857 init_task
.nsproxy
->mnt_ns
= ns
;
1860 set_fs_pwd(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1861 set_fs_root(current
->fs
, ns
->root
, ns
->root
->mnt_root
);
1864 void __init
mnt_init(void)
1869 init_rwsem(&namespace_sem
);
1871 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1872 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1874 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1876 if (!mount_hashtable
)
1877 panic("Failed to allocate mount hash table\n");
1879 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
1881 for (u
= 0; u
< HASH_SIZE
; u
++)
1882 INIT_LIST_HEAD(&mount_hashtable
[u
]);
1886 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1888 fs_kobj
= kobject_create_and_add("fs", NULL
);
1890 printk(KERN_WARNING
"%s: kobj create error\n", __FUNCTION__
);
1895 void __put_mnt_ns(struct mnt_namespace
*ns
)
1897 struct vfsmount
*root
= ns
->root
;
1898 LIST_HEAD(umount_list
);
1900 spin_unlock(&vfsmount_lock
);
1901 down_write(&namespace_sem
);
1902 spin_lock(&vfsmount_lock
);
1903 umount_tree(root
, 0, &umount_list
);
1904 spin_unlock(&vfsmount_lock
);
1905 up_write(&namespace_sem
);
1906 release_mounts(&umount_list
);