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/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
29 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30 #define HASH_SIZE (1UL << HASH_SHIFT)
33 static DEFINE_IDA(mnt_id_ida
);
34 static DEFINE_IDA(mnt_group_ida
);
35 static DEFINE_SPINLOCK(mnt_id_lock
);
36 static int mnt_id_start
= 0;
37 static int mnt_group_start
= 1;
39 static struct list_head
*mount_hashtable __read_mostly
;
40 static struct list_head
*mountpoint_hashtable __read_mostly
;
41 static struct kmem_cache
*mnt_cache __read_mostly
;
42 static DECLARE_RWSEM(namespace_sem
);
45 struct kobject
*fs_kobj
;
46 EXPORT_SYMBOL_GPL(fs_kobj
);
49 * vfsmount lock may be taken for read to prevent changes to the
50 * vfsmount hash, ie. during mountpoint lookups or walking back
53 * It should be taken for write in all cases where the vfsmount
54 * tree or hash is modified or when a vfsmount structure is modified.
56 DEFINE_BRLOCK(vfsmount_lock
);
58 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
60 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
61 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
62 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
63 return tmp
& (HASH_SIZE
- 1);
67 * allocation is serialized by namespace_sem, but we need the spinlock to
68 * serialize with freeing.
70 static int mnt_alloc_id(struct mount
*mnt
)
75 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
76 spin_lock(&mnt_id_lock
);
77 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
79 mnt_id_start
= mnt
->mnt_id
+ 1;
80 spin_unlock(&mnt_id_lock
);
87 static void mnt_free_id(struct mount
*mnt
)
90 spin_lock(&mnt_id_lock
);
91 ida_remove(&mnt_id_ida
, id
);
92 if (mnt_id_start
> id
)
94 spin_unlock(&mnt_id_lock
);
98 * Allocate a new peer group ID
100 * mnt_group_ida is protected by namespace_sem
102 static int mnt_alloc_group_id(struct mount
*mnt
)
106 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
109 res
= ida_get_new_above(&mnt_group_ida
,
113 mnt_group_start
= mnt
->mnt_group_id
+ 1;
119 * Release a peer group ID
121 void mnt_release_group_id(struct mount
*mnt
)
123 int id
= mnt
->mnt_group_id
;
124 ida_remove(&mnt_group_ida
, id
);
125 if (mnt_group_start
> id
)
126 mnt_group_start
= id
;
127 mnt
->mnt_group_id
= 0;
131 * vfsmount lock must be held for read
133 static inline void mnt_add_count(struct mount
*mnt
, int n
)
136 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
145 * vfsmount lock must be held for write
147 unsigned int mnt_get_count(struct mount
*mnt
)
150 unsigned int count
= 0;
153 for_each_possible_cpu(cpu
) {
154 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
159 return mnt
->mnt_count
;
163 static struct mount
*alloc_vfsmnt(const char *name
)
165 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
169 err
= mnt_alloc_id(mnt
);
174 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
175 if (!mnt
->mnt_devname
)
180 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
182 goto out_free_devname
;
184 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
187 mnt
->mnt_writers
= 0;
190 INIT_LIST_HEAD(&mnt
->mnt_hash
);
191 INIT_LIST_HEAD(&mnt
->mnt_child
);
192 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
193 INIT_LIST_HEAD(&mnt
->mnt_list
);
194 INIT_LIST_HEAD(&mnt
->mnt_expire
);
195 INIT_LIST_HEAD(&mnt
->mnt_share
);
196 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
197 INIT_LIST_HEAD(&mnt
->mnt_slave
);
198 #ifdef CONFIG_FSNOTIFY
199 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
206 kfree(mnt
->mnt_devname
);
211 kmem_cache_free(mnt_cache
, mnt
);
216 * Most r/o checks on a fs are for operations that take
217 * discrete amounts of time, like a write() or unlink().
218 * We must keep track of when those operations start
219 * (for permission checks) and when they end, so that
220 * we can determine when writes are able to occur to
224 * __mnt_is_readonly: check whether a mount is read-only
225 * @mnt: the mount to check for its write status
227 * This shouldn't be used directly ouside of the VFS.
228 * It does not guarantee that the filesystem will stay
229 * r/w, just that it is right *now*. This can not and
230 * should not be used in place of IS_RDONLY(inode).
231 * mnt_want/drop_write() will _keep_ the filesystem
234 int __mnt_is_readonly(struct vfsmount
*mnt
)
236 if (mnt
->mnt_flags
& MNT_READONLY
)
238 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
242 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
244 static inline void mnt_inc_writers(struct mount
*mnt
)
247 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
253 static inline void mnt_dec_writers(struct mount
*mnt
)
256 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
262 static unsigned int mnt_get_writers(struct mount
*mnt
)
265 unsigned int count
= 0;
268 for_each_possible_cpu(cpu
) {
269 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
274 return mnt
->mnt_writers
;
278 static int mnt_is_readonly(struct vfsmount
*mnt
)
280 if (mnt
->mnt_sb
->s_readonly_remount
)
282 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
284 return __mnt_is_readonly(mnt
);
288 * Most r/o & frozen checks on a fs are for operations that take discrete
289 * amounts of time, like a write() or unlink(). We must keep track of when
290 * those operations start (for permission checks) and when they end, so that we
291 * can determine when writes are able to occur to a filesystem.
294 * __mnt_want_write - get write access to a mount without freeze protection
295 * @m: the mount on which to take a write
297 * This tells the low-level filesystem that a write is about to be performed to
298 * it, and makes sure that writes are allowed (mnt it read-write) before
299 * returning success. This operation does not protect against filesystem being
300 * frozen. When the write operation is finished, __mnt_drop_write() must be
301 * called. This is effectively a refcount.
303 int __mnt_want_write(struct vfsmount
*m
)
305 struct mount
*mnt
= real_mount(m
);
309 mnt_inc_writers(mnt
);
311 * The store to mnt_inc_writers must be visible before we pass
312 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
313 * incremented count after it has set MNT_WRITE_HOLD.
316 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
319 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
320 * be set to match its requirements. So we must not load that until
321 * MNT_WRITE_HOLD is cleared.
324 if (mnt_is_readonly(m
)) {
325 mnt_dec_writers(mnt
);
334 * mnt_want_write - get write access to a mount
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mount is read-write, filesystem
339 * is not frozen) before returning success. When the write operation is
340 * finished, mnt_drop_write() must be called. This is effectively a refcount.
342 int mnt_want_write(struct vfsmount
*m
)
346 sb_start_write(m
->mnt_sb
);
347 ret
= __mnt_want_write(m
);
349 sb_end_write(m
->mnt_sb
);
352 EXPORT_SYMBOL_GPL(mnt_want_write
);
355 * mnt_clone_write - get write access to a mount
356 * @mnt: the mount on which to take a write
358 * This is effectively like mnt_want_write, except
359 * it must only be used to take an extra write reference
360 * on a mountpoint that we already know has a write reference
361 * on it. This allows some optimisation.
363 * After finished, mnt_drop_write must be called as usual to
364 * drop the reference.
366 int mnt_clone_write(struct vfsmount
*mnt
)
368 /* superblock may be r/o */
369 if (__mnt_is_readonly(mnt
))
372 mnt_inc_writers(real_mount(mnt
));
376 EXPORT_SYMBOL_GPL(mnt_clone_write
);
379 * __mnt_want_write_file - get write access to a file's mount
380 * @file: the file who's mount on which to take a write
382 * This is like __mnt_want_write, but it takes a file and can
383 * do some optimisations if the file is open for write already
385 int __mnt_want_write_file(struct file
*file
)
387 struct inode
*inode
= file_inode(file
);
389 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
390 return __mnt_want_write(file
->f_path
.mnt
);
392 return mnt_clone_write(file
->f_path
.mnt
);
396 * mnt_want_write_file - get write access to a file's mount
397 * @file: the file who's mount on which to take a write
399 * This is like mnt_want_write, but it takes a file and can
400 * do some optimisations if the file is open for write already
402 int mnt_want_write_file(struct file
*file
)
406 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
407 ret
= __mnt_want_write_file(file
);
409 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
412 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
415 * __mnt_drop_write - give up write access to a mount
416 * @mnt: the mount on which to give up write access
418 * Tells the low-level filesystem that we are done
419 * performing writes to it. Must be matched with
420 * __mnt_want_write() call above.
422 void __mnt_drop_write(struct vfsmount
*mnt
)
425 mnt_dec_writers(real_mount(mnt
));
430 * mnt_drop_write - give up write access to a mount
431 * @mnt: the mount on which to give up write access
433 * Tells the low-level filesystem that we are done performing writes to it and
434 * also allows filesystem to be frozen again. Must be matched with
435 * mnt_want_write() call above.
437 void mnt_drop_write(struct vfsmount
*mnt
)
439 __mnt_drop_write(mnt
);
440 sb_end_write(mnt
->mnt_sb
);
442 EXPORT_SYMBOL_GPL(mnt_drop_write
);
444 void __mnt_drop_write_file(struct file
*file
)
446 __mnt_drop_write(file
->f_path
.mnt
);
449 void mnt_drop_write_file(struct file
*file
)
451 mnt_drop_write(file
->f_path
.mnt
);
453 EXPORT_SYMBOL(mnt_drop_write_file
);
455 static int mnt_make_readonly(struct mount
*mnt
)
459 br_write_lock(&vfsmount_lock
);
460 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
462 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
463 * should be visible before we do.
468 * With writers on hold, if this value is zero, then there are
469 * definitely no active writers (although held writers may subsequently
470 * increment the count, they'll have to wait, and decrement it after
471 * seeing MNT_READONLY).
473 * It is OK to have counter incremented on one CPU and decremented on
474 * another: the sum will add up correctly. The danger would be when we
475 * sum up each counter, if we read a counter before it is incremented,
476 * but then read another CPU's count which it has been subsequently
477 * decremented from -- we would see more decrements than we should.
478 * MNT_WRITE_HOLD protects against this scenario, because
479 * mnt_want_write first increments count, then smp_mb, then spins on
480 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
481 * we're counting up here.
483 if (mnt_get_writers(mnt
) > 0)
486 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
488 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
489 * that become unheld will see MNT_READONLY.
492 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
493 br_write_unlock(&vfsmount_lock
);
497 static void __mnt_unmake_readonly(struct mount
*mnt
)
499 br_write_lock(&vfsmount_lock
);
500 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
501 br_write_unlock(&vfsmount_lock
);
504 int sb_prepare_remount_readonly(struct super_block
*sb
)
509 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
510 if (atomic_long_read(&sb
->s_remove_count
))
513 br_write_lock(&vfsmount_lock
);
514 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
515 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
516 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
518 if (mnt_get_writers(mnt
) > 0) {
524 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
528 sb
->s_readonly_remount
= 1;
531 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
532 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
533 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
535 br_write_unlock(&vfsmount_lock
);
540 static void free_vfsmnt(struct mount
*mnt
)
542 kfree(mnt
->mnt_devname
);
545 free_percpu(mnt
->mnt_pcp
);
547 kmem_cache_free(mnt_cache
, mnt
);
551 * find the first or last mount at @dentry on vfsmount @mnt depending on
552 * @dir. If @dir is set return the first mount else return the last mount.
553 * vfsmount_lock must be held for read or write.
555 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
558 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
559 struct list_head
*tmp
= head
;
560 struct mount
*p
, *found
= NULL
;
563 tmp
= dir
? tmp
->next
: tmp
->prev
;
567 p
= list_entry(tmp
, struct mount
, mnt_hash
);
568 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
) {
577 * lookup_mnt - Return the first child mount mounted at path
579 * "First" means first mounted chronologically. If you create the
582 * mount /dev/sda1 /mnt
583 * mount /dev/sda2 /mnt
584 * mount /dev/sda3 /mnt
586 * Then lookup_mnt() on the base /mnt dentry in the root mount will
587 * return successively the root dentry and vfsmount of /dev/sda1, then
588 * /dev/sda2, then /dev/sda3, then NULL.
590 * lookup_mnt takes a reference to the found vfsmount.
592 struct vfsmount
*lookup_mnt(struct path
*path
)
594 struct mount
*child_mnt
;
596 br_read_lock(&vfsmount_lock
);
597 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
599 mnt_add_count(child_mnt
, 1);
600 br_read_unlock(&vfsmount_lock
);
601 return &child_mnt
->mnt
;
603 br_read_unlock(&vfsmount_lock
);
608 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
610 struct list_head
*chain
= mountpoint_hashtable
+ hash(NULL
, dentry
);
611 struct mountpoint
*mp
;
614 list_for_each_entry(mp
, chain
, m_hash
) {
615 if (mp
->m_dentry
== dentry
) {
616 /* might be worth a WARN_ON() */
617 if (d_unlinked(dentry
))
618 return ERR_PTR(-ENOENT
);
624 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
626 return ERR_PTR(-ENOMEM
);
628 ret
= d_set_mounted(dentry
);
634 mp
->m_dentry
= dentry
;
636 list_add(&mp
->m_hash
, chain
);
640 static void put_mountpoint(struct mountpoint
*mp
)
642 if (!--mp
->m_count
) {
643 struct dentry
*dentry
= mp
->m_dentry
;
644 spin_lock(&dentry
->d_lock
);
645 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
646 spin_unlock(&dentry
->d_lock
);
647 list_del(&mp
->m_hash
);
652 static inline int check_mnt(struct mount
*mnt
)
654 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
658 * vfsmount lock must be held for write
660 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
664 wake_up_interruptible(&ns
->poll
);
669 * vfsmount lock must be held for write
671 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
673 if (ns
&& ns
->event
!= event
) {
675 wake_up_interruptible(&ns
->poll
);
680 * vfsmount lock must be held for write
682 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
684 old_path
->dentry
= mnt
->mnt_mountpoint
;
685 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
686 mnt
->mnt_parent
= mnt
;
687 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
688 list_del_init(&mnt
->mnt_child
);
689 list_del_init(&mnt
->mnt_hash
);
690 put_mountpoint(mnt
->mnt_mp
);
695 * vfsmount lock must be held for write
697 void mnt_set_mountpoint(struct mount
*mnt
,
698 struct mountpoint
*mp
,
699 struct mount
*child_mnt
)
702 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
703 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
704 child_mnt
->mnt_parent
= mnt
;
705 child_mnt
->mnt_mp
= mp
;
709 * vfsmount lock must be held for write
711 static void attach_mnt(struct mount
*mnt
,
712 struct mount
*parent
,
713 struct mountpoint
*mp
)
715 mnt_set_mountpoint(parent
, mp
, mnt
);
716 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
717 hash(&parent
->mnt
, mp
->m_dentry
));
718 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
722 * vfsmount lock must be held for write
724 static void commit_tree(struct mount
*mnt
)
726 struct mount
*parent
= mnt
->mnt_parent
;
729 struct mnt_namespace
*n
= parent
->mnt_ns
;
731 BUG_ON(parent
== mnt
);
733 list_add_tail(&head
, &mnt
->mnt_list
);
734 list_for_each_entry(m
, &head
, mnt_list
)
737 list_splice(&head
, n
->list
.prev
);
739 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
740 hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
741 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
742 touch_mnt_namespace(n
);
745 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
747 struct list_head
*next
= p
->mnt_mounts
.next
;
748 if (next
== &p
->mnt_mounts
) {
752 next
= p
->mnt_child
.next
;
753 if (next
!= &p
->mnt_parent
->mnt_mounts
)
758 return list_entry(next
, struct mount
, mnt_child
);
761 static struct mount
*skip_mnt_tree(struct mount
*p
)
763 struct list_head
*prev
= p
->mnt_mounts
.prev
;
764 while (prev
!= &p
->mnt_mounts
) {
765 p
= list_entry(prev
, struct mount
, mnt_child
);
766 prev
= p
->mnt_mounts
.prev
;
772 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
778 return ERR_PTR(-ENODEV
);
780 mnt
= alloc_vfsmnt(name
);
782 return ERR_PTR(-ENOMEM
);
784 if (flags
& MS_KERNMOUNT
)
785 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
787 root
= mount_fs(type
, flags
, name
, data
);
790 return ERR_CAST(root
);
793 mnt
->mnt
.mnt_root
= root
;
794 mnt
->mnt
.mnt_sb
= root
->d_sb
;
795 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
796 mnt
->mnt_parent
= mnt
;
797 br_write_lock(&vfsmount_lock
);
798 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
799 br_write_unlock(&vfsmount_lock
);
802 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
804 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
807 struct super_block
*sb
= old
->mnt
.mnt_sb
;
811 mnt
= alloc_vfsmnt(old
->mnt_devname
);
813 return ERR_PTR(-ENOMEM
);
815 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
816 mnt
->mnt_group_id
= 0; /* not a peer of original */
818 mnt
->mnt_group_id
= old
->mnt_group_id
;
820 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
821 err
= mnt_alloc_group_id(mnt
);
826 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
827 /* Don't allow unprivileged users to change mount flags */
828 if ((flag
& CL_UNPRIVILEGED
) && (mnt
->mnt
.mnt_flags
& MNT_READONLY
))
829 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
831 /* Don't allow unprivileged users to reveal what is under a mount */
832 if ((flag
& CL_UNPRIVILEGED
) && list_empty(&old
->mnt_expire
))
833 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
835 atomic_inc(&sb
->s_active
);
836 mnt
->mnt
.mnt_sb
= sb
;
837 mnt
->mnt
.mnt_root
= dget(root
);
838 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
839 mnt
->mnt_parent
= mnt
;
840 br_write_lock(&vfsmount_lock
);
841 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
842 br_write_unlock(&vfsmount_lock
);
844 if ((flag
& CL_SLAVE
) ||
845 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
846 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
847 mnt
->mnt_master
= old
;
848 CLEAR_MNT_SHARED(mnt
);
849 } else if (!(flag
& CL_PRIVATE
)) {
850 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
851 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
852 if (IS_MNT_SLAVE(old
))
853 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
854 mnt
->mnt_master
= old
->mnt_master
;
856 if (flag
& CL_MAKE_SHARED
)
859 /* stick the duplicate mount on the same expiry list
860 * as the original if that was on one */
861 if (flag
& CL_EXPIRE
) {
862 if (!list_empty(&old
->mnt_expire
))
863 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
873 static void mntput_no_expire(struct mount
*mnt
)
877 br_read_lock(&vfsmount_lock
);
878 if (likely(mnt
->mnt_ns
)) {
879 /* shouldn't be the last one */
880 mnt_add_count(mnt
, -1);
881 br_read_unlock(&vfsmount_lock
);
884 br_read_unlock(&vfsmount_lock
);
886 br_write_lock(&vfsmount_lock
);
887 mnt_add_count(mnt
, -1);
888 if (mnt_get_count(mnt
)) {
889 br_write_unlock(&vfsmount_lock
);
893 mnt_add_count(mnt
, -1);
894 if (likely(mnt_get_count(mnt
)))
896 br_write_lock(&vfsmount_lock
);
898 if (unlikely(mnt
->mnt_pinned
)) {
899 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
901 br_write_unlock(&vfsmount_lock
);
902 acct_auto_close_mnt(&mnt
->mnt
);
906 list_del(&mnt
->mnt_instance
);
907 br_write_unlock(&vfsmount_lock
);
910 * This probably indicates that somebody messed
911 * up a mnt_want/drop_write() pair. If this
912 * happens, the filesystem was probably unable
913 * to make r/w->r/o transitions.
916 * The locking used to deal with mnt_count decrement provides barriers,
917 * so mnt_get_writers() below is safe.
919 WARN_ON(mnt_get_writers(mnt
));
920 fsnotify_vfsmount_delete(&mnt
->mnt
);
921 dput(mnt
->mnt
.mnt_root
);
922 deactivate_super(mnt
->mnt
.mnt_sb
);
926 void mntput(struct vfsmount
*mnt
)
929 struct mount
*m
= real_mount(mnt
);
930 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
931 if (unlikely(m
->mnt_expiry_mark
))
932 m
->mnt_expiry_mark
= 0;
936 EXPORT_SYMBOL(mntput
);
938 struct vfsmount
*mntget(struct vfsmount
*mnt
)
941 mnt_add_count(real_mount(mnt
), 1);
944 EXPORT_SYMBOL(mntget
);
946 void mnt_pin(struct vfsmount
*mnt
)
948 br_write_lock(&vfsmount_lock
);
949 real_mount(mnt
)->mnt_pinned
++;
950 br_write_unlock(&vfsmount_lock
);
952 EXPORT_SYMBOL(mnt_pin
);
954 void mnt_unpin(struct vfsmount
*m
)
956 struct mount
*mnt
= real_mount(m
);
957 br_write_lock(&vfsmount_lock
);
958 if (mnt
->mnt_pinned
) {
959 mnt_add_count(mnt
, 1);
962 br_write_unlock(&vfsmount_lock
);
964 EXPORT_SYMBOL(mnt_unpin
);
966 static inline void mangle(struct seq_file
*m
, const char *s
)
968 seq_escape(m
, s
, " \t\n\\");
972 * Simple .show_options callback for filesystems which don't want to
973 * implement more complex mount option showing.
975 * See also save_mount_options().
977 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
982 options
= rcu_dereference(root
->d_sb
->s_options
);
984 if (options
!= NULL
&& options
[0]) {
992 EXPORT_SYMBOL(generic_show_options
);
995 * If filesystem uses generic_show_options(), this function should be
996 * called from the fill_super() callback.
998 * The .remount_fs callback usually needs to be handled in a special
999 * way, to make sure, that previous options are not overwritten if the
1002 * Also note, that if the filesystem's .remount_fs function doesn't
1003 * reset all options to their default value, but changes only newly
1004 * given options, then the displayed options will not reflect reality
1007 void save_mount_options(struct super_block
*sb
, char *options
)
1009 BUG_ON(sb
->s_options
);
1010 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1012 EXPORT_SYMBOL(save_mount_options
);
1014 void replace_mount_options(struct super_block
*sb
, char *options
)
1016 char *old
= sb
->s_options
;
1017 rcu_assign_pointer(sb
->s_options
, options
);
1023 EXPORT_SYMBOL(replace_mount_options
);
1025 #ifdef CONFIG_PROC_FS
1026 /* iterator; we want it to have access to namespace_sem, thus here... */
1027 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1029 struct proc_mounts
*p
= proc_mounts(m
);
1031 down_read(&namespace_sem
);
1032 return seq_list_start(&p
->ns
->list
, *pos
);
1035 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1037 struct proc_mounts
*p
= proc_mounts(m
);
1039 return seq_list_next(v
, &p
->ns
->list
, pos
);
1042 static void m_stop(struct seq_file
*m
, void *v
)
1044 up_read(&namespace_sem
);
1047 static int m_show(struct seq_file
*m
, void *v
)
1049 struct proc_mounts
*p
= proc_mounts(m
);
1050 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1051 return p
->show(m
, &r
->mnt
);
1054 const struct seq_operations mounts_op
= {
1060 #endif /* CONFIG_PROC_FS */
1063 * may_umount_tree - check if a mount tree is busy
1064 * @mnt: root of mount tree
1066 * This is called to check if a tree of mounts has any
1067 * open files, pwds, chroots or sub mounts that are
1070 int may_umount_tree(struct vfsmount
*m
)
1072 struct mount
*mnt
= real_mount(m
);
1073 int actual_refs
= 0;
1074 int minimum_refs
= 0;
1078 /* write lock needed for mnt_get_count */
1079 br_write_lock(&vfsmount_lock
);
1080 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1081 actual_refs
+= mnt_get_count(p
);
1084 br_write_unlock(&vfsmount_lock
);
1086 if (actual_refs
> minimum_refs
)
1092 EXPORT_SYMBOL(may_umount_tree
);
1095 * may_umount - check if a mount point is busy
1096 * @mnt: root of mount
1098 * This is called to check if a mount point has any
1099 * open files, pwds, chroots or sub mounts. If the
1100 * mount has sub mounts this will return busy
1101 * regardless of whether the sub mounts are busy.
1103 * Doesn't take quota and stuff into account. IOW, in some cases it will
1104 * give false negatives. The main reason why it's here is that we need
1105 * a non-destructive way to look for easily umountable filesystems.
1107 int may_umount(struct vfsmount
*mnt
)
1110 down_read(&namespace_sem
);
1111 br_write_lock(&vfsmount_lock
);
1112 if (propagate_mount_busy(real_mount(mnt
), 2))
1114 br_write_unlock(&vfsmount_lock
);
1115 up_read(&namespace_sem
);
1119 EXPORT_SYMBOL(may_umount
);
1121 static LIST_HEAD(unmounted
); /* protected by namespace_sem */
1123 static void namespace_unlock(void)
1128 if (likely(list_empty(&unmounted
))) {
1129 up_write(&namespace_sem
);
1133 list_splice_init(&unmounted
, &head
);
1134 up_write(&namespace_sem
);
1136 while (!list_empty(&head
)) {
1137 mnt
= list_first_entry(&head
, struct mount
, mnt_hash
);
1138 list_del_init(&mnt
->mnt_hash
);
1139 if (mnt_has_parent(mnt
)) {
1140 struct dentry
*dentry
;
1143 br_write_lock(&vfsmount_lock
);
1144 dentry
= mnt
->mnt_mountpoint
;
1145 m
= mnt
->mnt_parent
;
1146 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1147 mnt
->mnt_parent
= mnt
;
1149 br_write_unlock(&vfsmount_lock
);
1157 static inline void namespace_lock(void)
1159 down_write(&namespace_sem
);
1163 * vfsmount lock must be held for write
1164 * namespace_sem must be held for write
1166 void umount_tree(struct mount
*mnt
, int propagate
)
1168 LIST_HEAD(tmp_list
);
1171 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1172 list_move(&p
->mnt_hash
, &tmp_list
);
1175 propagate_umount(&tmp_list
);
1177 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1178 list_del_init(&p
->mnt_expire
);
1179 list_del_init(&p
->mnt_list
);
1180 __touch_mnt_namespace(p
->mnt_ns
);
1182 list_del_init(&p
->mnt_child
);
1183 if (mnt_has_parent(p
)) {
1184 p
->mnt_parent
->mnt_ghosts
++;
1185 put_mountpoint(p
->mnt_mp
);
1188 change_mnt_propagation(p
, MS_PRIVATE
);
1190 list_splice(&tmp_list
, &unmounted
);
1193 static void shrink_submounts(struct mount
*mnt
);
1195 static int do_umount(struct mount
*mnt
, int flags
)
1197 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1200 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1205 * Allow userspace to request a mountpoint be expired rather than
1206 * unmounting unconditionally. Unmount only happens if:
1207 * (1) the mark is already set (the mark is cleared by mntput())
1208 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1210 if (flags
& MNT_EXPIRE
) {
1211 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1212 flags
& (MNT_FORCE
| MNT_DETACH
))
1216 * probably don't strictly need the lock here if we examined
1217 * all race cases, but it's a slowpath.
1219 br_write_lock(&vfsmount_lock
);
1220 if (mnt_get_count(mnt
) != 2) {
1221 br_write_unlock(&vfsmount_lock
);
1224 br_write_unlock(&vfsmount_lock
);
1226 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1231 * If we may have to abort operations to get out of this
1232 * mount, and they will themselves hold resources we must
1233 * allow the fs to do things. In the Unix tradition of
1234 * 'Gee thats tricky lets do it in userspace' the umount_begin
1235 * might fail to complete on the first run through as other tasks
1236 * must return, and the like. Thats for the mount program to worry
1237 * about for the moment.
1240 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1241 sb
->s_op
->umount_begin(sb
);
1245 * No sense to grab the lock for this test, but test itself looks
1246 * somewhat bogus. Suggestions for better replacement?
1247 * Ho-hum... In principle, we might treat that as umount + switch
1248 * to rootfs. GC would eventually take care of the old vfsmount.
1249 * Actually it makes sense, especially if rootfs would contain a
1250 * /reboot - static binary that would close all descriptors and
1251 * call reboot(9). Then init(8) could umount root and exec /reboot.
1253 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1255 * Special case for "unmounting" root ...
1256 * we just try to remount it readonly.
1258 down_write(&sb
->s_umount
);
1259 if (!(sb
->s_flags
& MS_RDONLY
))
1260 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1261 up_write(&sb
->s_umount
);
1266 br_write_lock(&vfsmount_lock
);
1269 if (!(flags
& MNT_DETACH
))
1270 shrink_submounts(mnt
);
1273 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1274 if (!list_empty(&mnt
->mnt_list
))
1275 umount_tree(mnt
, 1);
1278 br_write_unlock(&vfsmount_lock
);
1284 * Is the caller allowed to modify his namespace?
1286 static inline bool may_mount(void)
1288 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1292 * Now umount can handle mount points as well as block devices.
1293 * This is important for filesystems which use unnamed block devices.
1295 * We now support a flag for forced unmount like the other 'big iron'
1296 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1299 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1304 int lookup_flags
= 0;
1306 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1312 if (!(flags
& UMOUNT_NOFOLLOW
))
1313 lookup_flags
|= LOOKUP_FOLLOW
;
1315 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1318 mnt
= real_mount(path
.mnt
);
1320 if (path
.dentry
!= path
.mnt
->mnt_root
)
1322 if (!check_mnt(mnt
))
1324 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1327 retval
= do_umount(mnt
, flags
);
1329 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1331 mntput_no_expire(mnt
);
1336 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1339 * The 2.0 compatible umount. No flags.
1341 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1343 return sys_umount(name
, 0);
1348 static bool is_mnt_ns_file(struct dentry
*dentry
)
1350 /* Is this a proxy for a mount namespace? */
1351 struct inode
*inode
= dentry
->d_inode
;
1354 if (!proc_ns_inode(inode
))
1357 ei
= get_proc_ns(inode
);
1358 if (ei
->ns_ops
!= &mntns_operations
)
1364 static bool mnt_ns_loop(struct dentry
*dentry
)
1366 /* Could bind mounting the mount namespace inode cause a
1367 * mount namespace loop?
1369 struct mnt_namespace
*mnt_ns
;
1370 if (!is_mnt_ns_file(dentry
))
1373 mnt_ns
= get_proc_ns(dentry
->d_inode
)->ns
;
1374 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1377 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1380 struct mount
*res
, *p
, *q
, *r
, *parent
;
1382 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1383 return ERR_PTR(-EINVAL
);
1385 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1386 return ERR_PTR(-EINVAL
);
1388 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1392 q
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1393 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1396 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1398 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1401 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1402 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1403 IS_MNT_UNBINDABLE(s
)) {
1404 s
= skip_mnt_tree(s
);
1407 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1408 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1409 s
= skip_mnt_tree(s
);
1412 while (p
!= s
->mnt_parent
) {
1418 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1421 br_write_lock(&vfsmount_lock
);
1422 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1423 attach_mnt(q
, parent
, p
->mnt_mp
);
1424 br_write_unlock(&vfsmount_lock
);
1430 br_write_lock(&vfsmount_lock
);
1431 umount_tree(res
, 0);
1432 br_write_unlock(&vfsmount_lock
);
1437 /* Caller should check returned pointer for errors */
1439 struct vfsmount
*collect_mounts(struct path
*path
)
1443 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1444 CL_COPY_ALL
| CL_PRIVATE
);
1447 return ERR_CAST(tree
);
1451 void drop_collected_mounts(struct vfsmount
*mnt
)
1454 br_write_lock(&vfsmount_lock
);
1455 umount_tree(real_mount(mnt
), 0);
1456 br_write_unlock(&vfsmount_lock
);
1460 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1461 struct vfsmount
*root
)
1464 int res
= f(root
, arg
);
1467 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1468 res
= f(&mnt
->mnt
, arg
);
1475 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1479 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1480 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1481 mnt_release_group_id(p
);
1485 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1489 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1490 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1491 int err
= mnt_alloc_group_id(p
);
1493 cleanup_group_ids(mnt
, p
);
1503 * @source_mnt : mount tree to be attached
1504 * @nd : place the mount tree @source_mnt is attached
1505 * @parent_nd : if non-null, detach the source_mnt from its parent and
1506 * store the parent mount and mountpoint dentry.
1507 * (done when source_mnt is moved)
1509 * NOTE: in the table below explains the semantics when a source mount
1510 * of a given type is attached to a destination mount of a given type.
1511 * ---------------------------------------------------------------------------
1512 * | BIND MOUNT OPERATION |
1513 * |**************************************************************************
1514 * | source-->| shared | private | slave | unbindable |
1518 * |**************************************************************************
1519 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1521 * |non-shared| shared (+) | private | slave (*) | invalid |
1522 * ***************************************************************************
1523 * A bind operation clones the source mount and mounts the clone on the
1524 * destination mount.
1526 * (++) the cloned mount is propagated to all the mounts in the propagation
1527 * tree of the destination mount and the cloned mount is added to
1528 * the peer group of the source mount.
1529 * (+) the cloned mount is created under the destination mount and is marked
1530 * as shared. The cloned mount is added to the peer group of the source
1532 * (+++) the mount is propagated to all the mounts in the propagation tree
1533 * of the destination mount and the cloned mount is made slave
1534 * of the same master as that of the source mount. The cloned mount
1535 * is marked as 'shared and slave'.
1536 * (*) the cloned mount is made a slave of the same master as that of the
1539 * ---------------------------------------------------------------------------
1540 * | MOVE MOUNT OPERATION |
1541 * |**************************************************************************
1542 * | source-->| shared | private | slave | unbindable |
1546 * |**************************************************************************
1547 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1549 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1550 * ***************************************************************************
1552 * (+) the mount is moved to the destination. And is then propagated to
1553 * all the mounts in the propagation tree of the destination mount.
1554 * (+*) the mount is moved to the destination.
1555 * (+++) the mount is moved to the destination and is then propagated to
1556 * all the mounts belonging to the destination mount's propagation tree.
1557 * the mount is marked as 'shared and slave'.
1558 * (*) the mount continues to be a slave at the new location.
1560 * if the source mount is a tree, the operations explained above is
1561 * applied to each mount in the tree.
1562 * Must be called without spinlocks held, since this function can sleep
1565 static int attach_recursive_mnt(struct mount
*source_mnt
,
1566 struct mount
*dest_mnt
,
1567 struct mountpoint
*dest_mp
,
1568 struct path
*parent_path
)
1570 LIST_HEAD(tree_list
);
1571 struct mount
*child
, *p
;
1574 if (IS_MNT_SHARED(dest_mnt
)) {
1575 err
= invent_group_ids(source_mnt
, true);
1579 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1581 goto out_cleanup_ids
;
1583 br_write_lock(&vfsmount_lock
);
1585 if (IS_MNT_SHARED(dest_mnt
)) {
1586 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1590 detach_mnt(source_mnt
, parent_path
);
1591 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1592 touch_mnt_namespace(source_mnt
->mnt_ns
);
1594 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1595 commit_tree(source_mnt
);
1598 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1599 list_del_init(&child
->mnt_hash
);
1602 br_write_unlock(&vfsmount_lock
);
1607 if (IS_MNT_SHARED(dest_mnt
))
1608 cleanup_group_ids(source_mnt
, NULL
);
1613 static struct mountpoint
*lock_mount(struct path
*path
)
1615 struct vfsmount
*mnt
;
1616 struct dentry
*dentry
= path
->dentry
;
1618 mutex_lock(&dentry
->d_inode
->i_mutex
);
1619 if (unlikely(cant_mount(dentry
))) {
1620 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1621 return ERR_PTR(-ENOENT
);
1624 mnt
= lookup_mnt(path
);
1626 struct mountpoint
*mp
= new_mountpoint(dentry
);
1629 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1635 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1638 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1642 static void unlock_mount(struct mountpoint
*where
)
1644 struct dentry
*dentry
= where
->m_dentry
;
1645 put_mountpoint(where
);
1647 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1650 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1652 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1655 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1656 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1659 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1663 * Sanity check the flags to change_mnt_propagation.
1666 static int flags_to_propagation_type(int flags
)
1668 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1670 /* Fail if any non-propagation flags are set */
1671 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1673 /* Only one propagation flag should be set */
1674 if (!is_power_of_2(type
))
1680 * recursively change the type of the mountpoint.
1682 static int do_change_type(struct path
*path
, int flag
)
1685 struct mount
*mnt
= real_mount(path
->mnt
);
1686 int recurse
= flag
& MS_REC
;
1690 if (path
->dentry
!= path
->mnt
->mnt_root
)
1693 type
= flags_to_propagation_type(flag
);
1698 if (type
== MS_SHARED
) {
1699 err
= invent_group_ids(mnt
, recurse
);
1704 br_write_lock(&vfsmount_lock
);
1705 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1706 change_mnt_propagation(m
, type
);
1707 br_write_unlock(&vfsmount_lock
);
1714 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
1716 struct mount
*child
;
1717 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
1718 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
1721 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
1728 * do loopback mount.
1730 static int do_loopback(struct path
*path
, const char *old_name
,
1733 struct path old_path
;
1734 struct mount
*mnt
= NULL
, *old
, *parent
;
1735 struct mountpoint
*mp
;
1737 if (!old_name
|| !*old_name
)
1739 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1744 if (mnt_ns_loop(old_path
.dentry
))
1747 mp
= lock_mount(path
);
1752 old
= real_mount(old_path
.mnt
);
1753 parent
= real_mount(path
->mnt
);
1756 if (IS_MNT_UNBINDABLE(old
))
1759 if (!check_mnt(parent
) || !check_mnt(old
))
1762 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
1766 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
1768 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1775 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1777 err
= graft_tree(mnt
, parent
, mp
);
1779 br_write_lock(&vfsmount_lock
);
1780 umount_tree(mnt
, 0);
1781 br_write_unlock(&vfsmount_lock
);
1786 path_put(&old_path
);
1790 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1793 int readonly_request
= 0;
1795 if (ms_flags
& MS_RDONLY
)
1796 readonly_request
= 1;
1797 if (readonly_request
== __mnt_is_readonly(mnt
))
1800 if (mnt
->mnt_flags
& MNT_LOCK_READONLY
)
1803 if (readonly_request
)
1804 error
= mnt_make_readonly(real_mount(mnt
));
1806 __mnt_unmake_readonly(real_mount(mnt
));
1811 * change filesystem flags. dir should be a physical root of filesystem.
1812 * If you've mounted a non-root directory somewhere and want to do remount
1813 * on it - tough luck.
1815 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1819 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1820 struct mount
*mnt
= real_mount(path
->mnt
);
1822 if (!check_mnt(mnt
))
1825 if (path
->dentry
!= path
->mnt
->mnt_root
)
1828 err
= security_sb_remount(sb
, data
);
1832 down_write(&sb
->s_umount
);
1833 if (flags
& MS_BIND
)
1834 err
= change_mount_flags(path
->mnt
, flags
);
1835 else if (!capable(CAP_SYS_ADMIN
))
1838 err
= do_remount_sb(sb
, flags
, data
, 0);
1840 br_write_lock(&vfsmount_lock
);
1841 mnt_flags
|= mnt
->mnt
.mnt_flags
& MNT_PROPAGATION_MASK
;
1842 mnt
->mnt
.mnt_flags
= mnt_flags
;
1843 touch_mnt_namespace(mnt
->mnt_ns
);
1844 br_write_unlock(&vfsmount_lock
);
1846 up_write(&sb
->s_umount
);
1850 static inline int tree_contains_unbindable(struct mount
*mnt
)
1853 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1854 if (IS_MNT_UNBINDABLE(p
))
1860 static int do_move_mount(struct path
*path
, const char *old_name
)
1862 struct path old_path
, parent_path
;
1865 struct mountpoint
*mp
;
1867 if (!old_name
|| !*old_name
)
1869 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1873 mp
= lock_mount(path
);
1878 old
= real_mount(old_path
.mnt
);
1879 p
= real_mount(path
->mnt
);
1882 if (!check_mnt(p
) || !check_mnt(old
))
1885 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
1889 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1892 if (!mnt_has_parent(old
))
1895 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1896 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1899 * Don't move a mount residing in a shared parent.
1901 if (IS_MNT_SHARED(old
->mnt_parent
))
1904 * Don't move a mount tree containing unbindable mounts to a destination
1905 * mount which is shared.
1907 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
1910 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
1914 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
1918 /* if the mount is moved, it should no longer be expire
1920 list_del_init(&old
->mnt_expire
);
1925 path_put(&parent_path
);
1926 path_put(&old_path
);
1930 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
1933 const char *subtype
= strchr(fstype
, '.');
1942 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
1944 if (!mnt
->mnt_sb
->s_subtype
)
1950 return ERR_PTR(err
);
1954 * add a mount into a namespace's mount tree
1956 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
1958 struct mountpoint
*mp
;
1959 struct mount
*parent
;
1962 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
1964 mp
= lock_mount(path
);
1968 parent
= real_mount(path
->mnt
);
1970 if (unlikely(!check_mnt(parent
))) {
1971 /* that's acceptable only for automounts done in private ns */
1972 if (!(mnt_flags
& MNT_SHRINKABLE
))
1974 /* ... and for those we'd better have mountpoint still alive */
1975 if (!parent
->mnt_ns
)
1979 /* Refuse the same filesystem on the same mount point */
1981 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
1982 path
->mnt
->mnt_root
== path
->dentry
)
1986 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1989 newmnt
->mnt
.mnt_flags
= mnt_flags
;
1990 err
= graft_tree(newmnt
, parent
, mp
);
1998 * create a new mount for userspace and request it to be added into the
2001 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2002 int mnt_flags
, const char *name
, void *data
)
2004 struct file_system_type
*type
;
2005 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2006 struct vfsmount
*mnt
;
2012 type
= get_fs_type(fstype
);
2016 if (user_ns
!= &init_user_ns
) {
2017 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2018 put_filesystem(type
);
2021 /* Only in special cases allow devices from mounts
2022 * created outside the initial user namespace.
2024 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2026 mnt_flags
|= MNT_NODEV
;
2030 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2031 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2032 !mnt
->mnt_sb
->s_subtype
)
2033 mnt
= fs_set_subtype(mnt
, fstype
);
2035 put_filesystem(type
);
2037 return PTR_ERR(mnt
);
2039 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2045 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2047 struct mount
*mnt
= real_mount(m
);
2049 /* The new mount record should have at least 2 refs to prevent it being
2050 * expired before we get a chance to add it
2052 BUG_ON(mnt_get_count(mnt
) < 2);
2054 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2055 m
->mnt_root
== path
->dentry
) {
2060 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2064 /* remove m from any expiration list it may be on */
2065 if (!list_empty(&mnt
->mnt_expire
)) {
2067 list_del_init(&mnt
->mnt_expire
);
2076 * mnt_set_expiry - Put a mount on an expiration list
2077 * @mnt: The mount to list.
2078 * @expiry_list: The list to add the mount to.
2080 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2084 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2088 EXPORT_SYMBOL(mnt_set_expiry
);
2091 * process a list of expirable mountpoints with the intent of discarding any
2092 * mountpoints that aren't in use and haven't been touched since last we came
2095 void mark_mounts_for_expiry(struct list_head
*mounts
)
2097 struct mount
*mnt
, *next
;
2098 LIST_HEAD(graveyard
);
2100 if (list_empty(mounts
))
2104 br_write_lock(&vfsmount_lock
);
2106 /* extract from the expiration list every vfsmount that matches the
2107 * following criteria:
2108 * - only referenced by its parent vfsmount
2109 * - still marked for expiry (marked on the last call here; marks are
2110 * cleared by mntput())
2112 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2113 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2114 propagate_mount_busy(mnt
, 1))
2116 list_move(&mnt
->mnt_expire
, &graveyard
);
2118 while (!list_empty(&graveyard
)) {
2119 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2120 touch_mnt_namespace(mnt
->mnt_ns
);
2121 umount_tree(mnt
, 1);
2123 br_write_unlock(&vfsmount_lock
);
2127 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2130 * Ripoff of 'select_parent()'
2132 * search the list of submounts for a given mountpoint, and move any
2133 * shrinkable submounts to the 'graveyard' list.
2135 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2137 struct mount
*this_parent
= parent
;
2138 struct list_head
*next
;
2142 next
= this_parent
->mnt_mounts
.next
;
2144 while (next
!= &this_parent
->mnt_mounts
) {
2145 struct list_head
*tmp
= next
;
2146 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2149 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2152 * Descend a level if the d_mounts list is non-empty.
2154 if (!list_empty(&mnt
->mnt_mounts
)) {
2159 if (!propagate_mount_busy(mnt
, 1)) {
2160 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2165 * All done at this level ... ascend and resume the search
2167 if (this_parent
!= parent
) {
2168 next
= this_parent
->mnt_child
.next
;
2169 this_parent
= this_parent
->mnt_parent
;
2176 * process a list of expirable mountpoints with the intent of discarding any
2177 * submounts of a specific parent mountpoint
2179 * vfsmount_lock must be held for write
2181 static void shrink_submounts(struct mount
*mnt
)
2183 LIST_HEAD(graveyard
);
2186 /* extract submounts of 'mountpoint' from the expiration list */
2187 while (select_submounts(mnt
, &graveyard
)) {
2188 while (!list_empty(&graveyard
)) {
2189 m
= list_first_entry(&graveyard
, struct mount
,
2191 touch_mnt_namespace(m
->mnt_ns
);
2198 * Some copy_from_user() implementations do not return the exact number of
2199 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2200 * Note that this function differs from copy_from_user() in that it will oops
2201 * on bad values of `to', rather than returning a short copy.
2203 static long exact_copy_from_user(void *to
, const void __user
* from
,
2207 const char __user
*f
= from
;
2210 if (!access_ok(VERIFY_READ
, from
, n
))
2214 if (__get_user(c
, f
)) {
2225 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2235 if (!(page
= __get_free_page(GFP_KERNEL
)))
2238 /* We only care that *some* data at the address the user
2239 * gave us is valid. Just in case, we'll zero
2240 * the remainder of the page.
2242 /* copy_from_user cannot cross TASK_SIZE ! */
2243 size
= TASK_SIZE
- (unsigned long)data
;
2244 if (size
> PAGE_SIZE
)
2247 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2253 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2258 int copy_mount_string(const void __user
*data
, char **where
)
2267 tmp
= strndup_user(data
, PAGE_SIZE
);
2269 return PTR_ERR(tmp
);
2276 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2277 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2279 * data is a (void *) that can point to any structure up to
2280 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2281 * information (or be NULL).
2283 * Pre-0.97 versions of mount() didn't have a flags word.
2284 * When the flags word was introduced its top half was required
2285 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2286 * Therefore, if this magic number is present, it carries no information
2287 * and must be discarded.
2289 long do_mount(const char *dev_name
, const char *dir_name
,
2290 const char *type_page
, unsigned long flags
, void *data_page
)
2297 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2298 flags
&= ~MS_MGC_MSK
;
2300 /* Basic sanity checks */
2302 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2306 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2308 /* ... and get the mountpoint */
2309 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2313 retval
= security_sb_mount(dev_name
, &path
,
2314 type_page
, flags
, data_page
);
2315 if (!retval
&& !may_mount())
2320 /* Default to relatime unless overriden */
2321 if (!(flags
& MS_NOATIME
))
2322 mnt_flags
|= MNT_RELATIME
;
2324 /* Separate the per-mountpoint flags */
2325 if (flags
& MS_NOSUID
)
2326 mnt_flags
|= MNT_NOSUID
;
2327 if (flags
& MS_NODEV
)
2328 mnt_flags
|= MNT_NODEV
;
2329 if (flags
& MS_NOEXEC
)
2330 mnt_flags
|= MNT_NOEXEC
;
2331 if (flags
& MS_NOATIME
)
2332 mnt_flags
|= MNT_NOATIME
;
2333 if (flags
& MS_NODIRATIME
)
2334 mnt_flags
|= MNT_NODIRATIME
;
2335 if (flags
& MS_STRICTATIME
)
2336 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2337 if (flags
& MS_RDONLY
)
2338 mnt_flags
|= MNT_READONLY
;
2340 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2341 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2344 if (flags
& MS_REMOUNT
)
2345 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2347 else if (flags
& MS_BIND
)
2348 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2349 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2350 retval
= do_change_type(&path
, flags
);
2351 else if (flags
& MS_MOVE
)
2352 retval
= do_move_mount(&path
, dev_name
);
2354 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2355 dev_name
, data_page
);
2361 static void free_mnt_ns(struct mnt_namespace
*ns
)
2363 proc_free_inum(ns
->proc_inum
);
2364 put_user_ns(ns
->user_ns
);
2369 * Assign a sequence number so we can detect when we attempt to bind
2370 * mount a reference to an older mount namespace into the current
2371 * mount namespace, preventing reference counting loops. A 64bit
2372 * number incrementing at 10Ghz will take 12,427 years to wrap which
2373 * is effectively never, so we can ignore the possibility.
2375 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2377 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2379 struct mnt_namespace
*new_ns
;
2382 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2384 return ERR_PTR(-ENOMEM
);
2385 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2388 return ERR_PTR(ret
);
2390 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2391 atomic_set(&new_ns
->count
, 1);
2392 new_ns
->root
= NULL
;
2393 INIT_LIST_HEAD(&new_ns
->list
);
2394 init_waitqueue_head(&new_ns
->poll
);
2396 new_ns
->user_ns
= get_user_ns(user_ns
);
2401 * Allocate a new namespace structure and populate it with contents
2402 * copied from the namespace of the passed in task structure.
2404 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2405 struct user_namespace
*user_ns
, struct fs_struct
*fs
)
2407 struct mnt_namespace
*new_ns
;
2408 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2409 struct mount
*p
, *q
;
2410 struct mount
*old
= mnt_ns
->root
;
2414 new_ns
= alloc_mnt_ns(user_ns
);
2419 /* First pass: copy the tree topology */
2420 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2421 if (user_ns
!= mnt_ns
->user_ns
)
2422 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2423 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2426 free_mnt_ns(new_ns
);
2427 return ERR_CAST(new);
2430 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2433 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2434 * as belonging to new namespace. We have already acquired a private
2435 * fs_struct, so tsk->fs->lock is not needed.
2442 if (&p
->mnt
== fs
->root
.mnt
) {
2443 fs
->root
.mnt
= mntget(&q
->mnt
);
2446 if (&p
->mnt
== fs
->pwd
.mnt
) {
2447 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2451 p
= next_mnt(p
, old
);
2452 q
= next_mnt(q
, new);
2455 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2456 p
= next_mnt(p
, old
);
2468 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2469 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2471 struct mnt_namespace
*new_ns
;
2476 if (!(flags
& CLONE_NEWNS
))
2479 new_ns
= dup_mnt_ns(ns
, user_ns
, new_fs
);
2486 * create_mnt_ns - creates a private namespace and adds a root filesystem
2487 * @mnt: pointer to the new root filesystem mountpoint
2489 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2491 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2492 if (!IS_ERR(new_ns
)) {
2493 struct mount
*mnt
= real_mount(m
);
2494 mnt
->mnt_ns
= new_ns
;
2496 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2503 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2505 struct mnt_namespace
*ns
;
2506 struct super_block
*s
;
2510 ns
= create_mnt_ns(mnt
);
2512 return ERR_CAST(ns
);
2514 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2515 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2520 return ERR_PTR(err
);
2522 /* trade a vfsmount reference for active sb one */
2523 s
= path
.mnt
->mnt_sb
;
2524 atomic_inc(&s
->s_active
);
2526 /* lock the sucker */
2527 down_write(&s
->s_umount
);
2528 /* ... and return the root of (sub)tree on it */
2531 EXPORT_SYMBOL(mount_subtree
);
2533 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2534 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2538 struct filename
*kernel_dir
;
2540 unsigned long data_page
;
2542 ret
= copy_mount_string(type
, &kernel_type
);
2546 kernel_dir
= getname(dir_name
);
2547 if (IS_ERR(kernel_dir
)) {
2548 ret
= PTR_ERR(kernel_dir
);
2552 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2556 ret
= copy_mount_options(data
, &data_page
);
2560 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2561 (void *) data_page
);
2563 free_page(data_page
);
2567 putname(kernel_dir
);
2575 * Return true if path is reachable from root
2577 * namespace_sem or vfsmount_lock is held
2579 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2580 const struct path
*root
)
2582 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2583 dentry
= mnt
->mnt_mountpoint
;
2584 mnt
= mnt
->mnt_parent
;
2586 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2589 int path_is_under(struct path
*path1
, struct path
*path2
)
2592 br_read_lock(&vfsmount_lock
);
2593 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2594 br_read_unlock(&vfsmount_lock
);
2597 EXPORT_SYMBOL(path_is_under
);
2600 * pivot_root Semantics:
2601 * Moves the root file system of the current process to the directory put_old,
2602 * makes new_root as the new root file system of the current process, and sets
2603 * root/cwd of all processes which had them on the current root to new_root.
2606 * The new_root and put_old must be directories, and must not be on the
2607 * same file system as the current process root. The put_old must be
2608 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2609 * pointed to by put_old must yield the same directory as new_root. No other
2610 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2612 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2613 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2614 * in this situation.
2617 * - we don't move root/cwd if they are not at the root (reason: if something
2618 * cared enough to change them, it's probably wrong to force them elsewhere)
2619 * - it's okay to pick a root that isn't the root of a file system, e.g.
2620 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2621 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2624 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2625 const char __user
*, put_old
)
2627 struct path
new, old
, parent_path
, root_parent
, root
;
2628 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2629 struct mountpoint
*old_mp
, *root_mp
;
2635 error
= user_path_dir(new_root
, &new);
2639 error
= user_path_dir(put_old
, &old
);
2643 error
= security_sb_pivotroot(&old
, &new);
2647 get_fs_root(current
->fs
, &root
);
2648 old_mp
= lock_mount(&old
);
2649 error
= PTR_ERR(old_mp
);
2654 new_mnt
= real_mount(new.mnt
);
2655 root_mnt
= real_mount(root
.mnt
);
2656 old_mnt
= real_mount(old
.mnt
);
2657 if (IS_MNT_SHARED(old_mnt
) ||
2658 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2659 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2661 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2663 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
2666 if (d_unlinked(new.dentry
))
2669 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2670 goto out4
; /* loop, on the same file system */
2672 if (root
.mnt
->mnt_root
!= root
.dentry
)
2673 goto out4
; /* not a mountpoint */
2674 if (!mnt_has_parent(root_mnt
))
2675 goto out4
; /* not attached */
2676 root_mp
= root_mnt
->mnt_mp
;
2677 if (new.mnt
->mnt_root
!= new.dentry
)
2678 goto out4
; /* not a mountpoint */
2679 if (!mnt_has_parent(new_mnt
))
2680 goto out4
; /* not attached */
2681 /* make sure we can reach put_old from new_root */
2682 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2684 root_mp
->m_count
++; /* pin it so it won't go away */
2685 br_write_lock(&vfsmount_lock
);
2686 detach_mnt(new_mnt
, &parent_path
);
2687 detach_mnt(root_mnt
, &root_parent
);
2688 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
2689 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
2690 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2692 /* mount old root on put_old */
2693 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2694 /* mount new_root on / */
2695 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2696 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2697 br_write_unlock(&vfsmount_lock
);
2698 chroot_fs_refs(&root
, &new);
2699 put_mountpoint(root_mp
);
2702 unlock_mount(old_mp
);
2704 path_put(&root_parent
);
2705 path_put(&parent_path
);
2717 static void __init
init_mount_tree(void)
2719 struct vfsmount
*mnt
;
2720 struct mnt_namespace
*ns
;
2722 struct file_system_type
*type
;
2724 type
= get_fs_type("rootfs");
2726 panic("Can't find rootfs type");
2727 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2728 put_filesystem(type
);
2730 panic("Can't create rootfs");
2732 ns
= create_mnt_ns(mnt
);
2734 panic("Can't allocate initial namespace");
2736 init_task
.nsproxy
->mnt_ns
= ns
;
2740 root
.dentry
= mnt
->mnt_root
;
2742 set_fs_pwd(current
->fs
, &root
);
2743 set_fs_root(current
->fs
, &root
);
2746 void __init
mnt_init(void)
2751 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2752 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2754 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2755 mountpoint_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2757 if (!mount_hashtable
|| !mountpoint_hashtable
)
2758 panic("Failed to allocate mount hash table\n");
2760 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2762 for (u
= 0; u
< HASH_SIZE
; u
++)
2763 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2764 for (u
= 0; u
< HASH_SIZE
; u
++)
2765 INIT_LIST_HEAD(&mountpoint_hashtable
[u
]);
2767 br_lock_init(&vfsmount_lock
);
2771 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2773 fs_kobj
= kobject_create_and_add("fs", NULL
);
2775 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2780 void put_mnt_ns(struct mnt_namespace
*ns
)
2782 if (!atomic_dec_and_test(&ns
->count
))
2784 drop_collected_mounts(&ns
->root
->mnt
);
2788 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2790 struct vfsmount
*mnt
;
2791 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2794 * it is a longterm mount, don't release mnt until
2795 * we unmount before file sys is unregistered
2797 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2801 EXPORT_SYMBOL_GPL(kern_mount_data
);
2803 void kern_unmount(struct vfsmount
*mnt
)
2805 /* release long term mount so mount point can be released */
2806 if (!IS_ERR_OR_NULL(mnt
)) {
2807 br_write_lock(&vfsmount_lock
);
2808 real_mount(mnt
)->mnt_ns
= NULL
;
2809 br_write_unlock(&vfsmount_lock
);
2813 EXPORT_SYMBOL(kern_unmount
);
2815 bool our_mnt(struct vfsmount
*mnt
)
2817 return check_mnt(real_mount(mnt
));
2820 bool current_chrooted(void)
2822 /* Does the current process have a non-standard root */
2823 struct path ns_root
;
2824 struct path fs_root
;
2827 /* Find the namespace root */
2828 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
2829 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
2831 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
2834 get_fs_root(current
->fs
, &fs_root
);
2836 chrooted
= !path_equal(&fs_root
, &ns_root
);
2844 bool fs_fully_visible(struct file_system_type
*type
)
2846 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
2848 bool visible
= false;
2853 down_read(&namespace_sem
);
2854 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
2855 struct mount
*child
;
2856 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
2859 /* This mount is not fully visible if there are any child mounts
2860 * that cover anything except for empty directories.
2862 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2863 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
2864 if (!S_ISDIR(inode
->i_mode
))
2866 if (inode
->i_nlink
!= 2)
2874 up_read(&namespace_sem
);
2878 static void *mntns_get(struct task_struct
*task
)
2880 struct mnt_namespace
*ns
= NULL
;
2881 struct nsproxy
*nsproxy
;
2884 nsproxy
= task_nsproxy(task
);
2886 ns
= nsproxy
->mnt_ns
;
2894 static void mntns_put(void *ns
)
2899 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
2901 struct fs_struct
*fs
= current
->fs
;
2902 struct mnt_namespace
*mnt_ns
= ns
;
2905 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
2906 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
2907 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
2914 put_mnt_ns(nsproxy
->mnt_ns
);
2915 nsproxy
->mnt_ns
= mnt_ns
;
2918 root
.mnt
= &mnt_ns
->root
->mnt
;
2919 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
2921 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
2924 /* Update the pwd and root */
2925 set_fs_pwd(fs
, &root
);
2926 set_fs_root(fs
, &root
);
2932 static unsigned int mntns_inum(void *ns
)
2934 struct mnt_namespace
*mnt_ns
= ns
;
2935 return mnt_ns
->proc_inum
;
2938 const struct proc_ns_operations mntns_operations
= {
2940 .type
= CLONE_NEWNS
,
2943 .install
= mntns_install
,