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/ramfs.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_fs.h>
28 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
29 #define HASH_SIZE (1UL << HASH_SHIFT)
32 static DEFINE_IDA(mnt_id_ida
);
33 static DEFINE_IDA(mnt_group_ida
);
34 static DEFINE_SPINLOCK(mnt_id_lock
);
35 static int mnt_id_start
= 0;
36 static int mnt_group_start
= 1;
38 static struct list_head
*mount_hashtable __read_mostly
;
39 static struct list_head
*mountpoint_hashtable __read_mostly
;
40 static struct kmem_cache
*mnt_cache __read_mostly
;
41 static struct rw_semaphore namespace_sem
;
44 struct kobject
*fs_kobj
;
45 EXPORT_SYMBOL_GPL(fs_kobj
);
48 * vfsmount lock may be taken for read to prevent changes to the
49 * vfsmount hash, ie. during mountpoint lookups or walking back
52 * It should be taken for write in all cases where the vfsmount
53 * tree or hash is modified or when a vfsmount structure is modified.
55 DEFINE_BRLOCK(vfsmount_lock
);
57 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
59 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
60 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
61 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
62 return tmp
& (HASH_SIZE
- 1);
65 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
68 * allocation is serialized by namespace_sem, but we need the spinlock to
69 * serialize with freeing.
71 static int mnt_alloc_id(struct mount
*mnt
)
76 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
77 spin_lock(&mnt_id_lock
);
78 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
80 mnt_id_start
= mnt
->mnt_id
+ 1;
81 spin_unlock(&mnt_id_lock
);
88 static void mnt_free_id(struct mount
*mnt
)
91 spin_lock(&mnt_id_lock
);
92 ida_remove(&mnt_id_ida
, id
);
93 if (mnt_id_start
> id
)
95 spin_unlock(&mnt_id_lock
);
99 * Allocate a new peer group ID
101 * mnt_group_ida is protected by namespace_sem
103 static int mnt_alloc_group_id(struct mount
*mnt
)
107 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
110 res
= ida_get_new_above(&mnt_group_ida
,
114 mnt_group_start
= mnt
->mnt_group_id
+ 1;
120 * Release a peer group ID
122 void mnt_release_group_id(struct mount
*mnt
)
124 int id
= mnt
->mnt_group_id
;
125 ida_remove(&mnt_group_ida
, id
);
126 if (mnt_group_start
> id
)
127 mnt_group_start
= id
;
128 mnt
->mnt_group_id
= 0;
132 * vfsmount lock must be held for read
134 static inline void mnt_add_count(struct mount
*mnt
, int n
)
137 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
146 * vfsmount lock must be held for write
148 unsigned int mnt_get_count(struct mount
*mnt
)
151 unsigned int count
= 0;
154 for_each_possible_cpu(cpu
) {
155 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
160 return mnt
->mnt_count
;
164 static struct mount
*alloc_vfsmnt(const char *name
)
166 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
170 err
= mnt_alloc_id(mnt
);
175 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
176 if (!mnt
->mnt_devname
)
181 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
183 goto out_free_devname
;
185 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
188 mnt
->mnt_writers
= 0;
191 INIT_LIST_HEAD(&mnt
->mnt_hash
);
192 INIT_LIST_HEAD(&mnt
->mnt_child
);
193 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
194 INIT_LIST_HEAD(&mnt
->mnt_list
);
195 INIT_LIST_HEAD(&mnt
->mnt_expire
);
196 INIT_LIST_HEAD(&mnt
->mnt_share
);
197 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
198 INIT_LIST_HEAD(&mnt
->mnt_slave
);
199 #ifdef CONFIG_FSNOTIFY
200 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
207 kfree(mnt
->mnt_devname
);
212 kmem_cache_free(mnt_cache
, mnt
);
217 * Most r/o checks on a fs are for operations that take
218 * discrete amounts of time, like a write() or unlink().
219 * We must keep track of when those operations start
220 * (for permission checks) and when they end, so that
221 * we can determine when writes are able to occur to
225 * __mnt_is_readonly: check whether a mount is read-only
226 * @mnt: the mount to check for its write status
228 * This shouldn't be used directly ouside of the VFS.
229 * It does not guarantee that the filesystem will stay
230 * r/w, just that it is right *now*. This can not and
231 * should not be used in place of IS_RDONLY(inode).
232 * mnt_want/drop_write() will _keep_ the filesystem
235 int __mnt_is_readonly(struct vfsmount
*mnt
)
237 if (mnt
->mnt_flags
& MNT_READONLY
)
239 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
243 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
245 static inline void mnt_inc_writers(struct mount
*mnt
)
248 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
254 static inline void mnt_dec_writers(struct mount
*mnt
)
257 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
263 static unsigned int mnt_get_writers(struct mount
*mnt
)
266 unsigned int count
= 0;
269 for_each_possible_cpu(cpu
) {
270 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
275 return mnt
->mnt_writers
;
279 static int mnt_is_readonly(struct vfsmount
*mnt
)
281 if (mnt
->mnt_sb
->s_readonly_remount
)
283 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
285 return __mnt_is_readonly(mnt
);
289 * Most r/o & frozen checks on a fs are for operations that take discrete
290 * amounts of time, like a write() or unlink(). We must keep track of when
291 * those operations start (for permission checks) and when they end, so that we
292 * can determine when writes are able to occur to a filesystem.
295 * __mnt_want_write - get write access to a mount without freeze protection
296 * @m: the mount on which to take a write
298 * This tells the low-level filesystem that a write is about to be performed to
299 * it, and makes sure that writes are allowed (mnt it read-write) before
300 * returning success. This operation does not protect against filesystem being
301 * frozen. When the write operation is finished, __mnt_drop_write() must be
302 * called. This is effectively a refcount.
304 int __mnt_want_write(struct vfsmount
*m
)
306 struct mount
*mnt
= real_mount(m
);
310 mnt_inc_writers(mnt
);
312 * The store to mnt_inc_writers must be visible before we pass
313 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
314 * incremented count after it has set MNT_WRITE_HOLD.
317 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
320 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
321 * be set to match its requirements. So we must not load that until
322 * MNT_WRITE_HOLD is cleared.
325 if (mnt_is_readonly(m
)) {
326 mnt_dec_writers(mnt
);
335 * mnt_want_write - get write access to a mount
336 * @m: the mount on which to take a write
338 * This tells the low-level filesystem that a write is about to be performed to
339 * it, and makes sure that writes are allowed (mount is read-write, filesystem
340 * is not frozen) before returning success. When the write operation is
341 * finished, mnt_drop_write() must be called. This is effectively a refcount.
343 int mnt_want_write(struct vfsmount
*m
)
347 sb_start_write(m
->mnt_sb
);
348 ret
= __mnt_want_write(m
);
350 sb_end_write(m
->mnt_sb
);
353 EXPORT_SYMBOL_GPL(mnt_want_write
);
356 * mnt_clone_write - get write access to a mount
357 * @mnt: the mount on which to take a write
359 * This is effectively like mnt_want_write, except
360 * it must only be used to take an extra write reference
361 * on a mountpoint that we already know has a write reference
362 * on it. This allows some optimisation.
364 * After finished, mnt_drop_write must be called as usual to
365 * drop the reference.
367 int mnt_clone_write(struct vfsmount
*mnt
)
369 /* superblock may be r/o */
370 if (__mnt_is_readonly(mnt
))
373 mnt_inc_writers(real_mount(mnt
));
377 EXPORT_SYMBOL_GPL(mnt_clone_write
);
380 * __mnt_want_write_file - get write access to a file's mount
381 * @file: the file who's mount on which to take a write
383 * This is like __mnt_want_write, but it takes a file and can
384 * do some optimisations if the file is open for write already
386 int __mnt_want_write_file(struct file
*file
)
388 struct inode
*inode
= file_inode(file
);
390 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
391 return __mnt_want_write(file
->f_path
.mnt
);
393 return mnt_clone_write(file
->f_path
.mnt
);
397 * mnt_want_write_file - get write access to a file's mount
398 * @file: the file who's mount on which to take a write
400 * This is like mnt_want_write, but it takes a file and can
401 * do some optimisations if the file is open for write already
403 int mnt_want_write_file(struct file
*file
)
407 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
408 ret
= __mnt_want_write_file(file
);
410 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
413 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
416 * __mnt_drop_write - give up write access to a mount
417 * @mnt: the mount on which to give up write access
419 * Tells the low-level filesystem that we are done
420 * performing writes to it. Must be matched with
421 * __mnt_want_write() call above.
423 void __mnt_drop_write(struct vfsmount
*mnt
)
426 mnt_dec_writers(real_mount(mnt
));
431 * mnt_drop_write - give up write access to a mount
432 * @mnt: the mount on which to give up write access
434 * Tells the low-level filesystem that we are done performing writes to it and
435 * also allows filesystem to be frozen again. Must be matched with
436 * mnt_want_write() call above.
438 void mnt_drop_write(struct vfsmount
*mnt
)
440 __mnt_drop_write(mnt
);
441 sb_end_write(mnt
->mnt_sb
);
443 EXPORT_SYMBOL_GPL(mnt_drop_write
);
445 void __mnt_drop_write_file(struct file
*file
)
447 __mnt_drop_write(file
->f_path
.mnt
);
450 void mnt_drop_write_file(struct file
*file
)
452 mnt_drop_write(file
->f_path
.mnt
);
454 EXPORT_SYMBOL(mnt_drop_write_file
);
456 static int mnt_make_readonly(struct mount
*mnt
)
460 br_write_lock(&vfsmount_lock
);
461 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
463 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
464 * should be visible before we do.
469 * With writers on hold, if this value is zero, then there are
470 * definitely no active writers (although held writers may subsequently
471 * increment the count, they'll have to wait, and decrement it after
472 * seeing MNT_READONLY).
474 * It is OK to have counter incremented on one CPU and decremented on
475 * another: the sum will add up correctly. The danger would be when we
476 * sum up each counter, if we read a counter before it is incremented,
477 * but then read another CPU's count which it has been subsequently
478 * decremented from -- we would see more decrements than we should.
479 * MNT_WRITE_HOLD protects against this scenario, because
480 * mnt_want_write first increments count, then smp_mb, then spins on
481 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
482 * we're counting up here.
484 if (mnt_get_writers(mnt
) > 0)
487 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
489 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
490 * that become unheld will see MNT_READONLY.
493 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
494 br_write_unlock(&vfsmount_lock
);
498 static void __mnt_unmake_readonly(struct mount
*mnt
)
500 br_write_lock(&vfsmount_lock
);
501 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
502 br_write_unlock(&vfsmount_lock
);
505 int sb_prepare_remount_readonly(struct super_block
*sb
)
510 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
511 if (atomic_long_read(&sb
->s_remove_count
))
514 br_write_lock(&vfsmount_lock
);
515 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
516 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
517 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
519 if (mnt_get_writers(mnt
) > 0) {
525 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
529 sb
->s_readonly_remount
= 1;
532 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
533 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
534 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
536 br_write_unlock(&vfsmount_lock
);
541 static void free_vfsmnt(struct mount
*mnt
)
543 kfree(mnt
->mnt_devname
);
546 free_percpu(mnt
->mnt_pcp
);
548 kmem_cache_free(mnt_cache
, mnt
);
552 * find the first or last mount at @dentry on vfsmount @mnt depending on
553 * @dir. If @dir is set return the first mount else return the last mount.
554 * vfsmount_lock must be held for read or write.
556 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
559 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
560 struct list_head
*tmp
= head
;
561 struct mount
*p
, *found
= NULL
;
564 tmp
= dir
? tmp
->next
: tmp
->prev
;
568 p
= list_entry(tmp
, struct mount
, mnt_hash
);
569 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
) {
578 * lookup_mnt - Return the first child mount mounted at path
580 * "First" means first mounted chronologically. If you create the
583 * mount /dev/sda1 /mnt
584 * mount /dev/sda2 /mnt
585 * mount /dev/sda3 /mnt
587 * Then lookup_mnt() on the base /mnt dentry in the root mount will
588 * return successively the root dentry and vfsmount of /dev/sda1, then
589 * /dev/sda2, then /dev/sda3, then NULL.
591 * lookup_mnt takes a reference to the found vfsmount.
593 struct vfsmount
*lookup_mnt(struct path
*path
)
595 struct mount
*child_mnt
;
597 br_read_lock(&vfsmount_lock
);
598 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
600 mnt_add_count(child_mnt
, 1);
601 br_read_unlock(&vfsmount_lock
);
602 return &child_mnt
->mnt
;
604 br_read_unlock(&vfsmount_lock
);
609 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
611 struct list_head
*chain
= mountpoint_hashtable
+ hash(NULL
, dentry
);
612 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 spin_lock(&dentry
->d_lock
);
629 if (d_unlinked(dentry
)) {
630 spin_unlock(&dentry
->d_lock
);
632 return ERR_PTR(-ENOENT
);
634 dentry
->d_flags
|= DCACHE_MOUNTED
;
635 spin_unlock(&dentry
->d_lock
);
636 mp
->m_dentry
= dentry
;
638 list_add(&mp
->m_hash
, chain
);
642 static void put_mountpoint(struct mountpoint
*mp
)
644 if (!--mp
->m_count
) {
645 struct dentry
*dentry
= mp
->m_dentry
;
646 spin_lock(&dentry
->d_lock
);
647 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
648 spin_unlock(&dentry
->d_lock
);
649 list_del(&mp
->m_hash
);
654 static inline int check_mnt(struct mount
*mnt
)
656 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
660 * vfsmount lock must be held for write
662 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
666 wake_up_interruptible(&ns
->poll
);
671 * vfsmount lock must be held for write
673 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
675 if (ns
&& ns
->event
!= event
) {
677 wake_up_interruptible(&ns
->poll
);
682 * vfsmount lock must be held for write
684 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
686 old_path
->dentry
= mnt
->mnt_mountpoint
;
687 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
688 mnt
->mnt_parent
= mnt
;
689 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
690 list_del_init(&mnt
->mnt_child
);
691 list_del_init(&mnt
->mnt_hash
);
692 put_mountpoint(mnt
->mnt_mp
);
697 * vfsmount lock must be held for write
699 void mnt_set_mountpoint(struct mount
*mnt
,
700 struct mountpoint
*mp
,
701 struct mount
*child_mnt
)
704 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
705 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
706 child_mnt
->mnt_parent
= mnt
;
707 child_mnt
->mnt_mp
= mp
;
711 * vfsmount lock must be held for write
713 static void attach_mnt(struct mount
*mnt
,
714 struct mount
*parent
,
715 struct mountpoint
*mp
)
717 mnt_set_mountpoint(parent
, mp
, mnt
);
718 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
719 hash(&parent
->mnt
, mp
->m_dentry
));
720 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
724 * vfsmount lock must be held for write
726 static void commit_tree(struct mount
*mnt
)
728 struct mount
*parent
= mnt
->mnt_parent
;
731 struct mnt_namespace
*n
= parent
->mnt_ns
;
733 BUG_ON(parent
== mnt
);
735 list_add_tail(&head
, &mnt
->mnt_list
);
736 list_for_each_entry(m
, &head
, mnt_list
)
739 list_splice(&head
, n
->list
.prev
);
741 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
742 hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
743 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
744 touch_mnt_namespace(n
);
747 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
749 struct list_head
*next
= p
->mnt_mounts
.next
;
750 if (next
== &p
->mnt_mounts
) {
754 next
= p
->mnt_child
.next
;
755 if (next
!= &p
->mnt_parent
->mnt_mounts
)
760 return list_entry(next
, struct mount
, mnt_child
);
763 static struct mount
*skip_mnt_tree(struct mount
*p
)
765 struct list_head
*prev
= p
->mnt_mounts
.prev
;
766 while (prev
!= &p
->mnt_mounts
) {
767 p
= list_entry(prev
, struct mount
, mnt_child
);
768 prev
= p
->mnt_mounts
.prev
;
774 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
780 return ERR_PTR(-ENODEV
);
782 mnt
= alloc_vfsmnt(name
);
784 return ERR_PTR(-ENOMEM
);
786 if (flags
& MS_KERNMOUNT
)
787 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
789 root
= mount_fs(type
, flags
, name
, data
);
792 return ERR_CAST(root
);
795 mnt
->mnt
.mnt_root
= root
;
796 mnt
->mnt
.mnt_sb
= root
->d_sb
;
797 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
798 mnt
->mnt_parent
= mnt
;
799 br_write_lock(&vfsmount_lock
);
800 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
801 br_write_unlock(&vfsmount_lock
);
804 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
806 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
809 struct super_block
*sb
= old
->mnt
.mnt_sb
;
813 mnt
= alloc_vfsmnt(old
->mnt_devname
);
815 return ERR_PTR(-ENOMEM
);
817 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
818 mnt
->mnt_group_id
= 0; /* not a peer of original */
820 mnt
->mnt_group_id
= old
->mnt_group_id
;
822 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
823 err
= mnt_alloc_group_id(mnt
);
828 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
829 atomic_inc(&sb
->s_active
);
830 mnt
->mnt
.mnt_sb
= sb
;
831 mnt
->mnt
.mnt_root
= dget(root
);
832 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
833 mnt
->mnt_parent
= mnt
;
834 br_write_lock(&vfsmount_lock
);
835 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
836 br_write_unlock(&vfsmount_lock
);
838 if ((flag
& CL_SLAVE
) ||
839 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
840 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
841 mnt
->mnt_master
= old
;
842 CLEAR_MNT_SHARED(mnt
);
843 } else if (!(flag
& CL_PRIVATE
)) {
844 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
845 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
846 if (IS_MNT_SLAVE(old
))
847 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
848 mnt
->mnt_master
= old
->mnt_master
;
850 if (flag
& CL_MAKE_SHARED
)
853 /* stick the duplicate mount on the same expiry list
854 * as the original if that was on one */
855 if (flag
& CL_EXPIRE
) {
856 if (!list_empty(&old
->mnt_expire
))
857 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
867 static inline void mntfree(struct mount
*mnt
)
869 struct vfsmount
*m
= &mnt
->mnt
;
870 struct super_block
*sb
= m
->mnt_sb
;
873 * This probably indicates that somebody messed
874 * up a mnt_want/drop_write() pair. If this
875 * happens, the filesystem was probably unable
876 * to make r/w->r/o transitions.
879 * The locking used to deal with mnt_count decrement provides barriers,
880 * so mnt_get_writers() below is safe.
882 WARN_ON(mnt_get_writers(mnt
));
883 fsnotify_vfsmount_delete(m
);
886 deactivate_super(sb
);
889 static void mntput_no_expire(struct mount
*mnt
)
893 br_read_lock(&vfsmount_lock
);
894 if (likely(mnt
->mnt_ns
)) {
895 /* shouldn't be the last one */
896 mnt_add_count(mnt
, -1);
897 br_read_unlock(&vfsmount_lock
);
900 br_read_unlock(&vfsmount_lock
);
902 br_write_lock(&vfsmount_lock
);
903 mnt_add_count(mnt
, -1);
904 if (mnt_get_count(mnt
)) {
905 br_write_unlock(&vfsmount_lock
);
909 mnt_add_count(mnt
, -1);
910 if (likely(mnt_get_count(mnt
)))
912 br_write_lock(&vfsmount_lock
);
914 if (unlikely(mnt
->mnt_pinned
)) {
915 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
917 br_write_unlock(&vfsmount_lock
);
918 acct_auto_close_mnt(&mnt
->mnt
);
922 list_del(&mnt
->mnt_instance
);
923 br_write_unlock(&vfsmount_lock
);
927 void mntput(struct vfsmount
*mnt
)
930 struct mount
*m
= real_mount(mnt
);
931 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
932 if (unlikely(m
->mnt_expiry_mark
))
933 m
->mnt_expiry_mark
= 0;
937 EXPORT_SYMBOL(mntput
);
939 struct vfsmount
*mntget(struct vfsmount
*mnt
)
942 mnt_add_count(real_mount(mnt
), 1);
945 EXPORT_SYMBOL(mntget
);
947 void mnt_pin(struct vfsmount
*mnt
)
949 br_write_lock(&vfsmount_lock
);
950 real_mount(mnt
)->mnt_pinned
++;
951 br_write_unlock(&vfsmount_lock
);
953 EXPORT_SYMBOL(mnt_pin
);
955 void mnt_unpin(struct vfsmount
*m
)
957 struct mount
*mnt
= real_mount(m
);
958 br_write_lock(&vfsmount_lock
);
959 if (mnt
->mnt_pinned
) {
960 mnt_add_count(mnt
, 1);
963 br_write_unlock(&vfsmount_lock
);
965 EXPORT_SYMBOL(mnt_unpin
);
967 static inline void mangle(struct seq_file
*m
, const char *s
)
969 seq_escape(m
, s
, " \t\n\\");
973 * Simple .show_options callback for filesystems which don't want to
974 * implement more complex mount option showing.
976 * See also save_mount_options().
978 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
983 options
= rcu_dereference(root
->d_sb
->s_options
);
985 if (options
!= NULL
&& options
[0]) {
993 EXPORT_SYMBOL(generic_show_options
);
996 * If filesystem uses generic_show_options(), this function should be
997 * called from the fill_super() callback.
999 * The .remount_fs callback usually needs to be handled in a special
1000 * way, to make sure, that previous options are not overwritten if the
1003 * Also note, that if the filesystem's .remount_fs function doesn't
1004 * reset all options to their default value, but changes only newly
1005 * given options, then the displayed options will not reflect reality
1008 void save_mount_options(struct super_block
*sb
, char *options
)
1010 BUG_ON(sb
->s_options
);
1011 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1013 EXPORT_SYMBOL(save_mount_options
);
1015 void replace_mount_options(struct super_block
*sb
, char *options
)
1017 char *old
= sb
->s_options
;
1018 rcu_assign_pointer(sb
->s_options
, options
);
1024 EXPORT_SYMBOL(replace_mount_options
);
1026 #ifdef CONFIG_PROC_FS
1027 /* iterator; we want it to have access to namespace_sem, thus here... */
1028 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1030 struct proc_mounts
*p
= proc_mounts(m
);
1032 down_read(&namespace_sem
);
1033 return seq_list_start(&p
->ns
->list
, *pos
);
1036 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1038 struct proc_mounts
*p
= proc_mounts(m
);
1040 return seq_list_next(v
, &p
->ns
->list
, pos
);
1043 static void m_stop(struct seq_file
*m
, void *v
)
1045 up_read(&namespace_sem
);
1048 static int m_show(struct seq_file
*m
, void *v
)
1050 struct proc_mounts
*p
= proc_mounts(m
);
1051 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1052 return p
->show(m
, &r
->mnt
);
1055 const struct seq_operations mounts_op
= {
1061 #endif /* CONFIG_PROC_FS */
1064 * may_umount_tree - check if a mount tree is busy
1065 * @mnt: root of mount tree
1067 * This is called to check if a tree of mounts has any
1068 * open files, pwds, chroots or sub mounts that are
1071 int may_umount_tree(struct vfsmount
*m
)
1073 struct mount
*mnt
= real_mount(m
);
1074 int actual_refs
= 0;
1075 int minimum_refs
= 0;
1079 /* write lock needed for mnt_get_count */
1080 br_write_lock(&vfsmount_lock
);
1081 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1082 actual_refs
+= mnt_get_count(p
);
1085 br_write_unlock(&vfsmount_lock
);
1087 if (actual_refs
> minimum_refs
)
1093 EXPORT_SYMBOL(may_umount_tree
);
1096 * may_umount - check if a mount point is busy
1097 * @mnt: root of mount
1099 * This is called to check if a mount point has any
1100 * open files, pwds, chroots or sub mounts. If the
1101 * mount has sub mounts this will return busy
1102 * regardless of whether the sub mounts are busy.
1104 * Doesn't take quota and stuff into account. IOW, in some cases it will
1105 * give false negatives. The main reason why it's here is that we need
1106 * a non-destructive way to look for easily umountable filesystems.
1108 int may_umount(struct vfsmount
*mnt
)
1111 down_read(&namespace_sem
);
1112 br_write_lock(&vfsmount_lock
);
1113 if (propagate_mount_busy(real_mount(mnt
), 2))
1115 br_write_unlock(&vfsmount_lock
);
1116 up_read(&namespace_sem
);
1120 EXPORT_SYMBOL(may_umount
);
1122 static LIST_HEAD(unmounted
); /* protected by namespace_sem */
1124 static void namespace_unlock(void)
1129 if (likely(list_empty(&unmounted
))) {
1130 up_write(&namespace_sem
);
1134 list_splice_init(&unmounted
, &head
);
1135 up_write(&namespace_sem
);
1137 while (!list_empty(&head
)) {
1138 mnt
= list_first_entry(&head
, struct mount
, mnt_hash
);
1139 list_del_init(&mnt
->mnt_hash
);
1140 if (mnt_has_parent(mnt
)) {
1141 struct dentry
*dentry
;
1144 br_write_lock(&vfsmount_lock
);
1145 dentry
= mnt
->mnt_mountpoint
;
1146 m
= mnt
->mnt_parent
;
1147 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1148 mnt
->mnt_parent
= mnt
;
1150 br_write_unlock(&vfsmount_lock
);
1158 static inline void namespace_lock(void)
1160 down_write(&namespace_sem
);
1164 * vfsmount lock must be held for write
1165 * namespace_sem must be held for write
1167 void umount_tree(struct mount
*mnt
, int propagate
)
1169 LIST_HEAD(tmp_list
);
1172 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1173 list_move(&p
->mnt_hash
, &tmp_list
);
1176 propagate_umount(&tmp_list
);
1178 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1179 list_del_init(&p
->mnt_expire
);
1180 list_del_init(&p
->mnt_list
);
1181 __touch_mnt_namespace(p
->mnt_ns
);
1183 list_del_init(&p
->mnt_child
);
1184 if (mnt_has_parent(p
)) {
1185 p
->mnt_parent
->mnt_ghosts
++;
1186 put_mountpoint(p
->mnt_mp
);
1189 change_mnt_propagation(p
, MS_PRIVATE
);
1191 list_splice(&tmp_list
, &unmounted
);
1194 static void shrink_submounts(struct mount
*mnt
);
1196 static int do_umount(struct mount
*mnt
, int flags
)
1198 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1201 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1206 * Allow userspace to request a mountpoint be expired rather than
1207 * unmounting unconditionally. Unmount only happens if:
1208 * (1) the mark is already set (the mark is cleared by mntput())
1209 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1211 if (flags
& MNT_EXPIRE
) {
1212 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1213 flags
& (MNT_FORCE
| MNT_DETACH
))
1217 * probably don't strictly need the lock here if we examined
1218 * all race cases, but it's a slowpath.
1220 br_write_lock(&vfsmount_lock
);
1221 if (mnt_get_count(mnt
) != 2) {
1222 br_write_unlock(&vfsmount_lock
);
1225 br_write_unlock(&vfsmount_lock
);
1227 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1232 * If we may have to abort operations to get out of this
1233 * mount, and they will themselves hold resources we must
1234 * allow the fs to do things. In the Unix tradition of
1235 * 'Gee thats tricky lets do it in userspace' the umount_begin
1236 * might fail to complete on the first run through as other tasks
1237 * must return, and the like. Thats for the mount program to worry
1238 * about for the moment.
1241 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1242 sb
->s_op
->umount_begin(sb
);
1246 * No sense to grab the lock for this test, but test itself looks
1247 * somewhat bogus. Suggestions for better replacement?
1248 * Ho-hum... In principle, we might treat that as umount + switch
1249 * to rootfs. GC would eventually take care of the old vfsmount.
1250 * Actually it makes sense, especially if rootfs would contain a
1251 * /reboot - static binary that would close all descriptors and
1252 * call reboot(9). Then init(8) could umount root and exec /reboot.
1254 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1256 * Special case for "unmounting" root ...
1257 * we just try to remount it readonly.
1259 down_write(&sb
->s_umount
);
1260 if (!(sb
->s_flags
& MS_RDONLY
))
1261 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1262 up_write(&sb
->s_umount
);
1267 br_write_lock(&vfsmount_lock
);
1270 if (!(flags
& MNT_DETACH
))
1271 shrink_submounts(mnt
);
1274 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1275 if (!list_empty(&mnt
->mnt_list
))
1276 umount_tree(mnt
, 1);
1279 br_write_unlock(&vfsmount_lock
);
1285 * Is the caller allowed to modify his namespace?
1287 static inline bool may_mount(void)
1289 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1293 * Now umount can handle mount points as well as block devices.
1294 * This is important for filesystems which use unnamed block devices.
1296 * We now support a flag for forced unmount like the other 'big iron'
1297 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1300 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1305 int lookup_flags
= 0;
1307 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1313 if (!(flags
& UMOUNT_NOFOLLOW
))
1314 lookup_flags
|= LOOKUP_FOLLOW
;
1316 retval
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1319 mnt
= real_mount(path
.mnt
);
1321 if (path
.dentry
!= path
.mnt
->mnt_root
)
1323 if (!check_mnt(mnt
))
1326 retval
= do_umount(mnt
, flags
);
1328 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1330 mntput_no_expire(mnt
);
1335 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1338 * The 2.0 compatible umount. No flags.
1340 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1342 return sys_umount(name
, 0);
1347 static bool mnt_ns_loop(struct path
*path
)
1349 /* Could bind mounting the mount namespace inode cause a
1350 * mount namespace loop?
1352 struct inode
*inode
= path
->dentry
->d_inode
;
1353 struct proc_inode
*ei
;
1354 struct mnt_namespace
*mnt_ns
;
1356 if (!proc_ns_inode(inode
))
1360 if (ei
->ns_ops
!= &mntns_operations
)
1364 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1367 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1370 struct mount
*res
, *p
, *q
, *r
, *parent
;
1372 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1373 return ERR_PTR(-EINVAL
);
1375 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1379 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1382 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1384 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1387 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1388 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
1389 s
= skip_mnt_tree(s
);
1392 while (p
!= s
->mnt_parent
) {
1398 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1401 br_write_lock(&vfsmount_lock
);
1402 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1403 attach_mnt(q
, parent
, p
->mnt_mp
);
1404 br_write_unlock(&vfsmount_lock
);
1410 br_write_lock(&vfsmount_lock
);
1411 umount_tree(res
, 0);
1412 br_write_unlock(&vfsmount_lock
);
1417 /* Caller should check returned pointer for errors */
1419 struct vfsmount
*collect_mounts(struct path
*path
)
1423 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1424 CL_COPY_ALL
| CL_PRIVATE
);
1431 void drop_collected_mounts(struct vfsmount
*mnt
)
1434 br_write_lock(&vfsmount_lock
);
1435 umount_tree(real_mount(mnt
), 0);
1436 br_write_unlock(&vfsmount_lock
);
1440 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1441 struct vfsmount
*root
)
1444 int res
= f(root
, arg
);
1447 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1448 res
= f(&mnt
->mnt
, arg
);
1455 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1459 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1460 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1461 mnt_release_group_id(p
);
1465 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1469 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1470 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1471 int err
= mnt_alloc_group_id(p
);
1473 cleanup_group_ids(mnt
, p
);
1483 * @source_mnt : mount tree to be attached
1484 * @nd : place the mount tree @source_mnt is attached
1485 * @parent_nd : if non-null, detach the source_mnt from its parent and
1486 * store the parent mount and mountpoint dentry.
1487 * (done when source_mnt is moved)
1489 * NOTE: in the table below explains the semantics when a source mount
1490 * of a given type is attached to a destination mount of a given type.
1491 * ---------------------------------------------------------------------------
1492 * | BIND MOUNT OPERATION |
1493 * |**************************************************************************
1494 * | source-->| shared | private | slave | unbindable |
1498 * |**************************************************************************
1499 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1501 * |non-shared| shared (+) | private | slave (*) | invalid |
1502 * ***************************************************************************
1503 * A bind operation clones the source mount and mounts the clone on the
1504 * destination mount.
1506 * (++) the cloned mount is propagated to all the mounts in the propagation
1507 * tree of the destination mount and the cloned mount is added to
1508 * the peer group of the source mount.
1509 * (+) the cloned mount is created under the destination mount and is marked
1510 * as shared. The cloned mount is added to the peer group of the source
1512 * (+++) the mount is propagated to all the mounts in the propagation tree
1513 * of the destination mount and the cloned mount is made slave
1514 * of the same master as that of the source mount. The cloned mount
1515 * is marked as 'shared and slave'.
1516 * (*) the cloned mount is made a slave of the same master as that of the
1519 * ---------------------------------------------------------------------------
1520 * | MOVE MOUNT OPERATION |
1521 * |**************************************************************************
1522 * | source-->| shared | private | slave | unbindable |
1526 * |**************************************************************************
1527 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1529 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1530 * ***************************************************************************
1532 * (+) the mount is moved to the destination. And is then propagated to
1533 * all the mounts in the propagation tree of the destination mount.
1534 * (+*) the mount is moved to the destination.
1535 * (+++) the mount is moved to the destination and is then propagated to
1536 * all the mounts belonging to the destination mount's propagation tree.
1537 * the mount is marked as 'shared and slave'.
1538 * (*) the mount continues to be a slave at the new location.
1540 * if the source mount is a tree, the operations explained above is
1541 * applied to each mount in the tree.
1542 * Must be called without spinlocks held, since this function can sleep
1545 static int attach_recursive_mnt(struct mount
*source_mnt
,
1546 struct mount
*dest_mnt
,
1547 struct mountpoint
*dest_mp
,
1548 struct path
*parent_path
)
1550 LIST_HEAD(tree_list
);
1551 struct mount
*child
, *p
;
1554 if (IS_MNT_SHARED(dest_mnt
)) {
1555 err
= invent_group_ids(source_mnt
, true);
1559 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1561 goto out_cleanup_ids
;
1563 br_write_lock(&vfsmount_lock
);
1565 if (IS_MNT_SHARED(dest_mnt
)) {
1566 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1570 detach_mnt(source_mnt
, parent_path
);
1571 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1572 touch_mnt_namespace(source_mnt
->mnt_ns
);
1574 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1575 commit_tree(source_mnt
);
1578 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1579 list_del_init(&child
->mnt_hash
);
1582 br_write_unlock(&vfsmount_lock
);
1587 if (IS_MNT_SHARED(dest_mnt
))
1588 cleanup_group_ids(source_mnt
, NULL
);
1593 static struct mountpoint
*lock_mount(struct path
*path
)
1595 struct vfsmount
*mnt
;
1596 struct dentry
*dentry
= path
->dentry
;
1598 mutex_lock(&dentry
->d_inode
->i_mutex
);
1599 if (unlikely(cant_mount(dentry
))) {
1600 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1601 return ERR_PTR(-ENOENT
);
1604 mnt
= lookup_mnt(path
);
1606 struct mountpoint
*mp
= new_mountpoint(dentry
);
1609 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1615 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1618 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1622 static void unlock_mount(struct mountpoint
*where
)
1624 struct dentry
*dentry
= where
->m_dentry
;
1625 put_mountpoint(where
);
1627 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1630 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1632 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1635 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1636 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1639 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1643 * Sanity check the flags to change_mnt_propagation.
1646 static int flags_to_propagation_type(int flags
)
1648 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1650 /* Fail if any non-propagation flags are set */
1651 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1653 /* Only one propagation flag should be set */
1654 if (!is_power_of_2(type
))
1660 * recursively change the type of the mountpoint.
1662 static int do_change_type(struct path
*path
, int flag
)
1665 struct mount
*mnt
= real_mount(path
->mnt
);
1666 int recurse
= flag
& MS_REC
;
1670 if (path
->dentry
!= path
->mnt
->mnt_root
)
1673 type
= flags_to_propagation_type(flag
);
1678 if (type
== MS_SHARED
) {
1679 err
= invent_group_ids(mnt
, recurse
);
1684 br_write_lock(&vfsmount_lock
);
1685 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1686 change_mnt_propagation(m
, type
);
1687 br_write_unlock(&vfsmount_lock
);
1695 * do loopback mount.
1697 static int do_loopback(struct path
*path
, const char *old_name
,
1700 struct path old_path
;
1701 struct mount
*mnt
= NULL
, *old
, *parent
;
1702 struct mountpoint
*mp
;
1704 if (!old_name
|| !*old_name
)
1706 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1711 if (mnt_ns_loop(&old_path
))
1714 mp
= lock_mount(path
);
1719 old
= real_mount(old_path
.mnt
);
1720 parent
= real_mount(path
->mnt
);
1723 if (IS_MNT_UNBINDABLE(old
))
1726 if (!check_mnt(parent
) || !check_mnt(old
))
1730 mnt
= copy_tree(old
, old_path
.dentry
, 0);
1732 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1739 err
= graft_tree(mnt
, parent
, mp
);
1741 br_write_lock(&vfsmount_lock
);
1742 umount_tree(mnt
, 0);
1743 br_write_unlock(&vfsmount_lock
);
1748 path_put(&old_path
);
1752 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1755 int readonly_request
= 0;
1757 if (ms_flags
& MS_RDONLY
)
1758 readonly_request
= 1;
1759 if (readonly_request
== __mnt_is_readonly(mnt
))
1762 if (readonly_request
)
1763 error
= mnt_make_readonly(real_mount(mnt
));
1765 __mnt_unmake_readonly(real_mount(mnt
));
1770 * change filesystem flags. dir should be a physical root of filesystem.
1771 * If you've mounted a non-root directory somewhere and want to do remount
1772 * on it - tough luck.
1774 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1778 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1779 struct mount
*mnt
= real_mount(path
->mnt
);
1781 if (!check_mnt(mnt
))
1784 if (path
->dentry
!= path
->mnt
->mnt_root
)
1787 err
= security_sb_remount(sb
, data
);
1791 down_write(&sb
->s_umount
);
1792 if (flags
& MS_BIND
)
1793 err
= change_mount_flags(path
->mnt
, flags
);
1794 else if (!capable(CAP_SYS_ADMIN
))
1797 err
= do_remount_sb(sb
, flags
, data
, 0);
1799 br_write_lock(&vfsmount_lock
);
1800 mnt_flags
|= mnt
->mnt
.mnt_flags
& MNT_PROPAGATION_MASK
;
1801 mnt
->mnt
.mnt_flags
= mnt_flags
;
1802 br_write_unlock(&vfsmount_lock
);
1804 up_write(&sb
->s_umount
);
1806 br_write_lock(&vfsmount_lock
);
1807 touch_mnt_namespace(mnt
->mnt_ns
);
1808 br_write_unlock(&vfsmount_lock
);
1813 static inline int tree_contains_unbindable(struct mount
*mnt
)
1816 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1817 if (IS_MNT_UNBINDABLE(p
))
1823 static int do_move_mount(struct path
*path
, const char *old_name
)
1825 struct path old_path
, parent_path
;
1828 struct mountpoint
*mp
;
1830 if (!old_name
|| !*old_name
)
1832 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1836 mp
= lock_mount(path
);
1841 old
= real_mount(old_path
.mnt
);
1842 p
= real_mount(path
->mnt
);
1845 if (!check_mnt(p
) || !check_mnt(old
))
1849 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1852 if (!mnt_has_parent(old
))
1855 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1856 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1859 * Don't move a mount residing in a shared parent.
1861 if (IS_MNT_SHARED(old
->mnt_parent
))
1864 * Don't move a mount tree containing unbindable mounts to a destination
1865 * mount which is shared.
1867 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
1870 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
1874 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
1878 /* if the mount is moved, it should no longer be expire
1880 list_del_init(&old
->mnt_expire
);
1885 path_put(&parent_path
);
1886 path_put(&old_path
);
1890 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
1893 const char *subtype
= strchr(fstype
, '.');
1902 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
1904 if (!mnt
->mnt_sb
->s_subtype
)
1910 return ERR_PTR(err
);
1914 * add a mount into a namespace's mount tree
1916 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
1918 struct mountpoint
*mp
;
1919 struct mount
*parent
;
1922 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
1924 mp
= lock_mount(path
);
1928 parent
= real_mount(path
->mnt
);
1930 if (unlikely(!check_mnt(parent
))) {
1931 /* that's acceptable only for automounts done in private ns */
1932 if (!(mnt_flags
& MNT_SHRINKABLE
))
1934 /* ... and for those we'd better have mountpoint still alive */
1935 if (!parent
->mnt_ns
)
1939 /* Refuse the same filesystem on the same mount point */
1941 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
1942 path
->mnt
->mnt_root
== path
->dentry
)
1946 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1949 newmnt
->mnt
.mnt_flags
= mnt_flags
;
1950 err
= graft_tree(newmnt
, parent
, mp
);
1958 * create a new mount for userspace and request it to be added into the
1961 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
1962 int mnt_flags
, const char *name
, void *data
)
1964 struct file_system_type
*type
;
1965 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
1966 struct vfsmount
*mnt
;
1972 type
= get_fs_type(fstype
);
1976 if (user_ns
!= &init_user_ns
) {
1977 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
1978 put_filesystem(type
);
1981 /* Only in special cases allow devices from mounts
1982 * created outside the initial user namespace.
1984 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
1986 mnt_flags
|= MNT_NODEV
;
1990 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
1991 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
1992 !mnt
->mnt_sb
->s_subtype
)
1993 mnt
= fs_set_subtype(mnt
, fstype
);
1995 put_filesystem(type
);
1997 return PTR_ERR(mnt
);
1999 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2005 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2007 struct mount
*mnt
= real_mount(m
);
2009 /* The new mount record should have at least 2 refs to prevent it being
2010 * expired before we get a chance to add it
2012 BUG_ON(mnt_get_count(mnt
) < 2);
2014 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2015 m
->mnt_root
== path
->dentry
) {
2020 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2024 /* remove m from any expiration list it may be on */
2025 if (!list_empty(&mnt
->mnt_expire
)) {
2027 br_write_lock(&vfsmount_lock
);
2028 list_del_init(&mnt
->mnt_expire
);
2029 br_write_unlock(&vfsmount_lock
);
2038 * mnt_set_expiry - Put a mount on an expiration list
2039 * @mnt: The mount to list.
2040 * @expiry_list: The list to add the mount to.
2042 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2045 br_write_lock(&vfsmount_lock
);
2047 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2049 br_write_unlock(&vfsmount_lock
);
2052 EXPORT_SYMBOL(mnt_set_expiry
);
2055 * process a list of expirable mountpoints with the intent of discarding any
2056 * mountpoints that aren't in use and haven't been touched since last we came
2059 void mark_mounts_for_expiry(struct list_head
*mounts
)
2061 struct mount
*mnt
, *next
;
2062 LIST_HEAD(graveyard
);
2064 if (list_empty(mounts
))
2068 br_write_lock(&vfsmount_lock
);
2070 /* extract from the expiration list every vfsmount that matches the
2071 * following criteria:
2072 * - only referenced by its parent vfsmount
2073 * - still marked for expiry (marked on the last call here; marks are
2074 * cleared by mntput())
2076 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2077 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2078 propagate_mount_busy(mnt
, 1))
2080 list_move(&mnt
->mnt_expire
, &graveyard
);
2082 while (!list_empty(&graveyard
)) {
2083 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2084 touch_mnt_namespace(mnt
->mnt_ns
);
2085 umount_tree(mnt
, 1);
2087 br_write_unlock(&vfsmount_lock
);
2091 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2094 * Ripoff of 'select_parent()'
2096 * search the list of submounts for a given mountpoint, and move any
2097 * shrinkable submounts to the 'graveyard' list.
2099 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2101 struct mount
*this_parent
= parent
;
2102 struct list_head
*next
;
2106 next
= this_parent
->mnt_mounts
.next
;
2108 while (next
!= &this_parent
->mnt_mounts
) {
2109 struct list_head
*tmp
= next
;
2110 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2113 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2116 * Descend a level if the d_mounts list is non-empty.
2118 if (!list_empty(&mnt
->mnt_mounts
)) {
2123 if (!propagate_mount_busy(mnt
, 1)) {
2124 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2129 * All done at this level ... ascend and resume the search
2131 if (this_parent
!= parent
) {
2132 next
= this_parent
->mnt_child
.next
;
2133 this_parent
= this_parent
->mnt_parent
;
2140 * process a list of expirable mountpoints with the intent of discarding any
2141 * submounts of a specific parent mountpoint
2143 * vfsmount_lock must be held for write
2145 static void shrink_submounts(struct mount
*mnt
)
2147 LIST_HEAD(graveyard
);
2150 /* extract submounts of 'mountpoint' from the expiration list */
2151 while (select_submounts(mnt
, &graveyard
)) {
2152 while (!list_empty(&graveyard
)) {
2153 m
= list_first_entry(&graveyard
, struct mount
,
2155 touch_mnt_namespace(m
->mnt_ns
);
2162 * Some copy_from_user() implementations do not return the exact number of
2163 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2164 * Note that this function differs from copy_from_user() in that it will oops
2165 * on bad values of `to', rather than returning a short copy.
2167 static long exact_copy_from_user(void *to
, const void __user
* from
,
2171 const char __user
*f
= from
;
2174 if (!access_ok(VERIFY_READ
, from
, n
))
2178 if (__get_user(c
, f
)) {
2189 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2199 if (!(page
= __get_free_page(GFP_KERNEL
)))
2202 /* We only care that *some* data at the address the user
2203 * gave us is valid. Just in case, we'll zero
2204 * the remainder of the page.
2206 /* copy_from_user cannot cross TASK_SIZE ! */
2207 size
= TASK_SIZE
- (unsigned long)data
;
2208 if (size
> PAGE_SIZE
)
2211 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2217 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2222 int copy_mount_string(const void __user
*data
, char **where
)
2231 tmp
= strndup_user(data
, PAGE_SIZE
);
2233 return PTR_ERR(tmp
);
2240 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2241 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2243 * data is a (void *) that can point to any structure up to
2244 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2245 * information (or be NULL).
2247 * Pre-0.97 versions of mount() didn't have a flags word.
2248 * When the flags word was introduced its top half was required
2249 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2250 * Therefore, if this magic number is present, it carries no information
2251 * and must be discarded.
2253 long do_mount(const char *dev_name
, const char *dir_name
,
2254 const char *type_page
, unsigned long flags
, void *data_page
)
2261 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2262 flags
&= ~MS_MGC_MSK
;
2264 /* Basic sanity checks */
2266 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2270 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2272 /* ... and get the mountpoint */
2273 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2277 retval
= security_sb_mount(dev_name
, &path
,
2278 type_page
, flags
, data_page
);
2285 /* Default to relatime unless overriden */
2286 if (!(flags
& MS_NOATIME
))
2287 mnt_flags
|= MNT_RELATIME
;
2289 /* Separate the per-mountpoint flags */
2290 if (flags
& MS_NOSUID
)
2291 mnt_flags
|= MNT_NOSUID
;
2292 if (flags
& MS_NODEV
)
2293 mnt_flags
|= MNT_NODEV
;
2294 if (flags
& MS_NOEXEC
)
2295 mnt_flags
|= MNT_NOEXEC
;
2296 if (flags
& MS_NOATIME
)
2297 mnt_flags
|= MNT_NOATIME
;
2298 if (flags
& MS_NODIRATIME
)
2299 mnt_flags
|= MNT_NODIRATIME
;
2300 if (flags
& MS_STRICTATIME
)
2301 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2302 if (flags
& MS_RDONLY
)
2303 mnt_flags
|= MNT_READONLY
;
2305 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2306 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2309 if (flags
& MS_REMOUNT
)
2310 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2312 else if (flags
& MS_BIND
)
2313 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2314 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2315 retval
= do_change_type(&path
, flags
);
2316 else if (flags
& MS_MOVE
)
2317 retval
= do_move_mount(&path
, dev_name
);
2319 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2320 dev_name
, data_page
);
2326 static void free_mnt_ns(struct mnt_namespace
*ns
)
2328 proc_free_inum(ns
->proc_inum
);
2329 put_user_ns(ns
->user_ns
);
2334 * Assign a sequence number so we can detect when we attempt to bind
2335 * mount a reference to an older mount namespace into the current
2336 * mount namespace, preventing reference counting loops. A 64bit
2337 * number incrementing at 10Ghz will take 12,427 years to wrap which
2338 * is effectively never, so we can ignore the possibility.
2340 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2342 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2344 struct mnt_namespace
*new_ns
;
2347 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2349 return ERR_PTR(-ENOMEM
);
2350 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2353 return ERR_PTR(ret
);
2355 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2356 atomic_set(&new_ns
->count
, 1);
2357 new_ns
->root
= NULL
;
2358 INIT_LIST_HEAD(&new_ns
->list
);
2359 init_waitqueue_head(&new_ns
->poll
);
2361 new_ns
->user_ns
= get_user_ns(user_ns
);
2366 * Allocate a new namespace structure and populate it with contents
2367 * copied from the namespace of the passed in task structure.
2369 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2370 struct user_namespace
*user_ns
, struct fs_struct
*fs
)
2372 struct mnt_namespace
*new_ns
;
2373 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2374 struct mount
*p
, *q
;
2375 struct mount
*old
= mnt_ns
->root
;
2379 new_ns
= alloc_mnt_ns(user_ns
);
2384 /* First pass: copy the tree topology */
2385 copy_flags
= CL_COPY_ALL
| CL_EXPIRE
;
2386 if (user_ns
!= mnt_ns
->user_ns
)
2387 copy_flags
|= CL_SHARED_TO_SLAVE
;
2388 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2391 free_mnt_ns(new_ns
);
2392 return ERR_CAST(new);
2395 br_write_lock(&vfsmount_lock
);
2396 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2397 br_write_unlock(&vfsmount_lock
);
2400 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2401 * as belonging to new namespace. We have already acquired a private
2402 * fs_struct, so tsk->fs->lock is not needed.
2409 if (&p
->mnt
== fs
->root
.mnt
) {
2410 fs
->root
.mnt
= mntget(&q
->mnt
);
2413 if (&p
->mnt
== fs
->pwd
.mnt
) {
2414 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2418 p
= next_mnt(p
, old
);
2419 q
= next_mnt(q
, new);
2431 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2432 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2434 struct mnt_namespace
*new_ns
;
2439 if (!(flags
& CLONE_NEWNS
))
2442 new_ns
= dup_mnt_ns(ns
, user_ns
, new_fs
);
2449 * create_mnt_ns - creates a private namespace and adds a root filesystem
2450 * @mnt: pointer to the new root filesystem mountpoint
2452 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2454 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2455 if (!IS_ERR(new_ns
)) {
2456 struct mount
*mnt
= real_mount(m
);
2457 mnt
->mnt_ns
= new_ns
;
2459 list_add(&new_ns
->list
, &mnt
->mnt_list
);
2466 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2468 struct mnt_namespace
*ns
;
2469 struct super_block
*s
;
2473 ns
= create_mnt_ns(mnt
);
2475 return ERR_CAST(ns
);
2477 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2478 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2483 return ERR_PTR(err
);
2485 /* trade a vfsmount reference for active sb one */
2486 s
= path
.mnt
->mnt_sb
;
2487 atomic_inc(&s
->s_active
);
2489 /* lock the sucker */
2490 down_write(&s
->s_umount
);
2491 /* ... and return the root of (sub)tree on it */
2494 EXPORT_SYMBOL(mount_subtree
);
2496 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2497 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2501 struct filename
*kernel_dir
;
2503 unsigned long data_page
;
2505 ret
= copy_mount_string(type
, &kernel_type
);
2509 kernel_dir
= getname(dir_name
);
2510 if (IS_ERR(kernel_dir
)) {
2511 ret
= PTR_ERR(kernel_dir
);
2515 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2519 ret
= copy_mount_options(data
, &data_page
);
2523 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2524 (void *) data_page
);
2526 free_page(data_page
);
2530 putname(kernel_dir
);
2538 * Return true if path is reachable from root
2540 * namespace_sem or vfsmount_lock is held
2542 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2543 const struct path
*root
)
2545 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2546 dentry
= mnt
->mnt_mountpoint
;
2547 mnt
= mnt
->mnt_parent
;
2549 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2552 int path_is_under(struct path
*path1
, struct path
*path2
)
2555 br_read_lock(&vfsmount_lock
);
2556 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2557 br_read_unlock(&vfsmount_lock
);
2560 EXPORT_SYMBOL(path_is_under
);
2563 * pivot_root Semantics:
2564 * Moves the root file system of the current process to the directory put_old,
2565 * makes new_root as the new root file system of the current process, and sets
2566 * root/cwd of all processes which had them on the current root to new_root.
2569 * The new_root and put_old must be directories, and must not be on the
2570 * same file system as the current process root. The put_old must be
2571 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2572 * pointed to by put_old must yield the same directory as new_root. No other
2573 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2575 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2576 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2577 * in this situation.
2580 * - we don't move root/cwd if they are not at the root (reason: if something
2581 * cared enough to change them, it's probably wrong to force them elsewhere)
2582 * - it's okay to pick a root that isn't the root of a file system, e.g.
2583 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2584 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2587 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2588 const char __user
*, put_old
)
2590 struct path
new, old
, parent_path
, root_parent
, root
;
2591 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2592 struct mountpoint
*old_mp
, *root_mp
;
2598 error
= user_path_dir(new_root
, &new);
2602 error
= user_path_dir(put_old
, &old
);
2606 error
= security_sb_pivotroot(&old
, &new);
2610 get_fs_root(current
->fs
, &root
);
2611 old_mp
= lock_mount(&old
);
2612 error
= PTR_ERR(old_mp
);
2617 new_mnt
= real_mount(new.mnt
);
2618 root_mnt
= real_mount(root
.mnt
);
2619 old_mnt
= real_mount(old
.mnt
);
2620 if (IS_MNT_SHARED(old_mnt
) ||
2621 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2622 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2624 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2627 if (d_unlinked(new.dentry
))
2630 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2631 goto out4
; /* loop, on the same file system */
2633 if (root
.mnt
->mnt_root
!= root
.dentry
)
2634 goto out4
; /* not a mountpoint */
2635 if (!mnt_has_parent(root_mnt
))
2636 goto out4
; /* not attached */
2637 root_mp
= root_mnt
->mnt_mp
;
2638 if (new.mnt
->mnt_root
!= new.dentry
)
2639 goto out4
; /* not a mountpoint */
2640 if (!mnt_has_parent(new_mnt
))
2641 goto out4
; /* not attached */
2642 /* make sure we can reach put_old from new_root */
2643 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2645 root_mp
->m_count
++; /* pin it so it won't go away */
2646 br_write_lock(&vfsmount_lock
);
2647 detach_mnt(new_mnt
, &parent_path
);
2648 detach_mnt(root_mnt
, &root_parent
);
2649 /* mount old root on put_old */
2650 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2651 /* mount new_root on / */
2652 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2653 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2654 br_write_unlock(&vfsmount_lock
);
2655 chroot_fs_refs(&root
, &new);
2656 put_mountpoint(root_mp
);
2659 unlock_mount(old_mp
);
2661 path_put(&root_parent
);
2662 path_put(&parent_path
);
2674 static void __init
init_mount_tree(void)
2676 struct vfsmount
*mnt
;
2677 struct mnt_namespace
*ns
;
2679 struct file_system_type
*type
;
2681 type
= get_fs_type("rootfs");
2683 panic("Can't find rootfs type");
2684 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2685 put_filesystem(type
);
2687 panic("Can't create rootfs");
2689 ns
= create_mnt_ns(mnt
);
2691 panic("Can't allocate initial namespace");
2693 init_task
.nsproxy
->mnt_ns
= ns
;
2697 root
.dentry
= mnt
->mnt_root
;
2699 set_fs_pwd(current
->fs
, &root
);
2700 set_fs_root(current
->fs
, &root
);
2703 void __init
mnt_init(void)
2708 init_rwsem(&namespace_sem
);
2710 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2711 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2713 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2714 mountpoint_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2716 if (!mount_hashtable
|| !mountpoint_hashtable
)
2717 panic("Failed to allocate mount hash table\n");
2719 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2721 for (u
= 0; u
< HASH_SIZE
; u
++)
2722 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2723 for (u
= 0; u
< HASH_SIZE
; u
++)
2724 INIT_LIST_HEAD(&mountpoint_hashtable
[u
]);
2726 br_lock_init(&vfsmount_lock
);
2730 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2732 fs_kobj
= kobject_create_and_add("fs", NULL
);
2734 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2739 void put_mnt_ns(struct mnt_namespace
*ns
)
2741 if (!atomic_dec_and_test(&ns
->count
))
2744 br_write_lock(&vfsmount_lock
);
2745 umount_tree(ns
->root
, 0);
2746 br_write_unlock(&vfsmount_lock
);
2751 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2753 struct vfsmount
*mnt
;
2754 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2757 * it is a longterm mount, don't release mnt until
2758 * we unmount before file sys is unregistered
2760 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2764 EXPORT_SYMBOL_GPL(kern_mount_data
);
2766 void kern_unmount(struct vfsmount
*mnt
)
2768 /* release long term mount so mount point can be released */
2769 if (!IS_ERR_OR_NULL(mnt
)) {
2770 br_write_lock(&vfsmount_lock
);
2771 real_mount(mnt
)->mnt_ns
= NULL
;
2772 br_write_unlock(&vfsmount_lock
);
2776 EXPORT_SYMBOL(kern_unmount
);
2778 bool our_mnt(struct vfsmount
*mnt
)
2780 return check_mnt(real_mount(mnt
));
2783 static void *mntns_get(struct task_struct
*task
)
2785 struct mnt_namespace
*ns
= NULL
;
2786 struct nsproxy
*nsproxy
;
2789 nsproxy
= task_nsproxy(task
);
2791 ns
= nsproxy
->mnt_ns
;
2799 static void mntns_put(void *ns
)
2804 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
2806 struct fs_struct
*fs
= current
->fs
;
2807 struct mnt_namespace
*mnt_ns
= ns
;
2810 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
2811 !nsown_capable(CAP_SYS_CHROOT
) ||
2812 !nsown_capable(CAP_SYS_ADMIN
))
2819 put_mnt_ns(nsproxy
->mnt_ns
);
2820 nsproxy
->mnt_ns
= mnt_ns
;
2823 root
.mnt
= &mnt_ns
->root
->mnt
;
2824 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
2826 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
2829 /* Update the pwd and root */
2830 set_fs_pwd(fs
, &root
);
2831 set_fs_root(fs
, &root
);
2837 static unsigned int mntns_inum(void *ns
)
2839 struct mnt_namespace
*mnt_ns
= ns
;
2840 return mnt_ns
->proc_inum
;
2843 const struct proc_ns_operations mntns_operations
= {
2845 .type
= CLONE_NEWNS
,
2848 .install
= mntns_install
,