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_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 struct rw_semaphore 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);
66 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
69 * allocation is serialized by namespace_sem, but we need the spinlock to
70 * serialize with freeing.
72 static int mnt_alloc_id(struct mount
*mnt
)
77 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
78 spin_lock(&mnt_id_lock
);
79 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
81 mnt_id_start
= mnt
->mnt_id
+ 1;
82 spin_unlock(&mnt_id_lock
);
89 static void mnt_free_id(struct mount
*mnt
)
92 spin_lock(&mnt_id_lock
);
93 ida_remove(&mnt_id_ida
, id
);
94 if (mnt_id_start
> id
)
96 spin_unlock(&mnt_id_lock
);
100 * Allocate a new peer group ID
102 * mnt_group_ida is protected by namespace_sem
104 static int mnt_alloc_group_id(struct mount
*mnt
)
108 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
111 res
= ida_get_new_above(&mnt_group_ida
,
115 mnt_group_start
= mnt
->mnt_group_id
+ 1;
121 * Release a peer group ID
123 void mnt_release_group_id(struct mount
*mnt
)
125 int id
= mnt
->mnt_group_id
;
126 ida_remove(&mnt_group_ida
, id
);
127 if (mnt_group_start
> id
)
128 mnt_group_start
= id
;
129 mnt
->mnt_group_id
= 0;
133 * vfsmount lock must be held for read
135 static inline void mnt_add_count(struct mount
*mnt
, int n
)
138 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
147 * vfsmount lock must be held for write
149 unsigned int mnt_get_count(struct mount
*mnt
)
152 unsigned int count
= 0;
155 for_each_possible_cpu(cpu
) {
156 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
161 return mnt
->mnt_count
;
165 static struct mount
*alloc_vfsmnt(const char *name
)
167 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
171 err
= mnt_alloc_id(mnt
);
176 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
177 if (!mnt
->mnt_devname
)
182 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
184 goto out_free_devname
;
186 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
189 mnt
->mnt_writers
= 0;
192 INIT_LIST_HEAD(&mnt
->mnt_hash
);
193 INIT_LIST_HEAD(&mnt
->mnt_child
);
194 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
195 INIT_LIST_HEAD(&mnt
->mnt_list
);
196 INIT_LIST_HEAD(&mnt
->mnt_expire
);
197 INIT_LIST_HEAD(&mnt
->mnt_share
);
198 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
199 INIT_LIST_HEAD(&mnt
->mnt_slave
);
200 #ifdef CONFIG_FSNOTIFY
201 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
208 kfree(mnt
->mnt_devname
);
213 kmem_cache_free(mnt_cache
, mnt
);
218 * Most r/o checks on a fs are for operations that take
219 * discrete amounts of time, like a write() or unlink().
220 * We must keep track of when those operations start
221 * (for permission checks) and when they end, so that
222 * we can determine when writes are able to occur to
226 * __mnt_is_readonly: check whether a mount is read-only
227 * @mnt: the mount to check for its write status
229 * This shouldn't be used directly ouside of the VFS.
230 * It does not guarantee that the filesystem will stay
231 * r/w, just that it is right *now*. This can not and
232 * should not be used in place of IS_RDONLY(inode).
233 * mnt_want/drop_write() will _keep_ the filesystem
236 int __mnt_is_readonly(struct vfsmount
*mnt
)
238 if (mnt
->mnt_flags
& MNT_READONLY
)
240 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
244 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
246 static inline void mnt_inc_writers(struct mount
*mnt
)
249 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
255 static inline void mnt_dec_writers(struct mount
*mnt
)
258 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
264 static unsigned int mnt_get_writers(struct mount
*mnt
)
267 unsigned int count
= 0;
270 for_each_possible_cpu(cpu
) {
271 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
276 return mnt
->mnt_writers
;
280 static int mnt_is_readonly(struct vfsmount
*mnt
)
282 if (mnt
->mnt_sb
->s_readonly_remount
)
284 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
286 return __mnt_is_readonly(mnt
);
290 * Most r/o & frozen checks on a fs are for operations that take discrete
291 * amounts of time, like a write() or unlink(). We must keep track of when
292 * those operations start (for permission checks) and when they end, so that we
293 * can determine when writes are able to occur to a filesystem.
296 * __mnt_want_write - get write access to a mount without freeze protection
297 * @m: the mount on which to take a write
299 * This tells the low-level filesystem that a write is about to be performed to
300 * it, and makes sure that writes are allowed (mnt it read-write) before
301 * returning success. This operation does not protect against filesystem being
302 * frozen. When the write operation is finished, __mnt_drop_write() must be
303 * called. This is effectively a refcount.
305 int __mnt_want_write(struct vfsmount
*m
)
307 struct mount
*mnt
= real_mount(m
);
311 mnt_inc_writers(mnt
);
313 * The store to mnt_inc_writers must be visible before we pass
314 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
315 * incremented count after it has set MNT_WRITE_HOLD.
318 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
321 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
322 * be set to match its requirements. So we must not load that until
323 * MNT_WRITE_HOLD is cleared.
326 if (mnt_is_readonly(m
)) {
327 mnt_dec_writers(mnt
);
336 * mnt_want_write - get write access to a mount
337 * @m: the mount on which to take a write
339 * This tells the low-level filesystem that a write is about to be performed to
340 * it, and makes sure that writes are allowed (mount is read-write, filesystem
341 * is not frozen) before returning success. When the write operation is
342 * finished, mnt_drop_write() must be called. This is effectively a refcount.
344 int mnt_want_write(struct vfsmount
*m
)
348 sb_start_write(m
->mnt_sb
);
349 ret
= __mnt_want_write(m
);
351 sb_end_write(m
->mnt_sb
);
354 EXPORT_SYMBOL_GPL(mnt_want_write
);
357 * mnt_clone_write - get write access to a mount
358 * @mnt: the mount on which to take a write
360 * This is effectively like mnt_want_write, except
361 * it must only be used to take an extra write reference
362 * on a mountpoint that we already know has a write reference
363 * on it. This allows some optimisation.
365 * After finished, mnt_drop_write must be called as usual to
366 * drop the reference.
368 int mnt_clone_write(struct vfsmount
*mnt
)
370 /* superblock may be r/o */
371 if (__mnt_is_readonly(mnt
))
374 mnt_inc_writers(real_mount(mnt
));
378 EXPORT_SYMBOL_GPL(mnt_clone_write
);
381 * __mnt_want_write_file - get write access to a file's mount
382 * @file: the file who's mount on which to take a write
384 * This is like __mnt_want_write, but it takes a file and can
385 * do some optimisations if the file is open for write already
387 int __mnt_want_write_file(struct file
*file
)
389 struct inode
*inode
= file_inode(file
);
391 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
392 return __mnt_want_write(file
->f_path
.mnt
);
394 return mnt_clone_write(file
->f_path
.mnt
);
398 * mnt_want_write_file - get write access to a file's mount
399 * @file: the file who's mount on which to take a write
401 * This is like mnt_want_write, but it takes a file and can
402 * do some optimisations if the file is open for write already
404 int mnt_want_write_file(struct file
*file
)
408 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
409 ret
= __mnt_want_write_file(file
);
411 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
414 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
417 * __mnt_drop_write - give up write access to a mount
418 * @mnt: the mount on which to give up write access
420 * Tells the low-level filesystem that we are done
421 * performing writes to it. Must be matched with
422 * __mnt_want_write() call above.
424 void __mnt_drop_write(struct vfsmount
*mnt
)
427 mnt_dec_writers(real_mount(mnt
));
432 * mnt_drop_write - give up write access to a mount
433 * @mnt: the mount on which to give up write access
435 * Tells the low-level filesystem that we are done performing writes to it and
436 * also allows filesystem to be frozen again. Must be matched with
437 * mnt_want_write() call above.
439 void mnt_drop_write(struct vfsmount
*mnt
)
441 __mnt_drop_write(mnt
);
442 sb_end_write(mnt
->mnt_sb
);
444 EXPORT_SYMBOL_GPL(mnt_drop_write
);
446 void __mnt_drop_write_file(struct file
*file
)
448 __mnt_drop_write(file
->f_path
.mnt
);
451 void mnt_drop_write_file(struct file
*file
)
453 mnt_drop_write(file
->f_path
.mnt
);
455 EXPORT_SYMBOL(mnt_drop_write_file
);
457 static int mnt_make_readonly(struct mount
*mnt
)
461 br_write_lock(&vfsmount_lock
);
462 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
464 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
465 * should be visible before we do.
470 * With writers on hold, if this value is zero, then there are
471 * definitely no active writers (although held writers may subsequently
472 * increment the count, they'll have to wait, and decrement it after
473 * seeing MNT_READONLY).
475 * It is OK to have counter incremented on one CPU and decremented on
476 * another: the sum will add up correctly. The danger would be when we
477 * sum up each counter, if we read a counter before it is incremented,
478 * but then read another CPU's count which it has been subsequently
479 * decremented from -- we would see more decrements than we should.
480 * MNT_WRITE_HOLD protects against this scenario, because
481 * mnt_want_write first increments count, then smp_mb, then spins on
482 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
483 * we're counting up here.
485 if (mnt_get_writers(mnt
) > 0)
488 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
490 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
491 * that become unheld will see MNT_READONLY.
494 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
495 br_write_unlock(&vfsmount_lock
);
499 static void __mnt_unmake_readonly(struct mount
*mnt
)
501 br_write_lock(&vfsmount_lock
);
502 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
503 br_write_unlock(&vfsmount_lock
);
506 int sb_prepare_remount_readonly(struct super_block
*sb
)
511 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
512 if (atomic_long_read(&sb
->s_remove_count
))
515 br_write_lock(&vfsmount_lock
);
516 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
517 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
518 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
520 if (mnt_get_writers(mnt
) > 0) {
526 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
530 sb
->s_readonly_remount
= 1;
533 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
534 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
535 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
537 br_write_unlock(&vfsmount_lock
);
542 static void free_vfsmnt(struct mount
*mnt
)
544 kfree(mnt
->mnt_devname
);
547 free_percpu(mnt
->mnt_pcp
);
549 kmem_cache_free(mnt_cache
, mnt
);
553 * find the first or last mount at @dentry on vfsmount @mnt depending on
554 * @dir. If @dir is set return the first mount else return the last mount.
555 * vfsmount_lock must be held for read or write.
557 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
560 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
561 struct list_head
*tmp
= head
;
562 struct mount
*p
, *found
= NULL
;
565 tmp
= dir
? tmp
->next
: tmp
->prev
;
569 p
= list_entry(tmp
, struct mount
, mnt_hash
);
570 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
) {
579 * lookup_mnt - Return the first child mount mounted at path
581 * "First" means first mounted chronologically. If you create the
584 * mount /dev/sda1 /mnt
585 * mount /dev/sda2 /mnt
586 * mount /dev/sda3 /mnt
588 * Then lookup_mnt() on the base /mnt dentry in the root mount will
589 * return successively the root dentry and vfsmount of /dev/sda1, then
590 * /dev/sda2, then /dev/sda3, then NULL.
592 * lookup_mnt takes a reference to the found vfsmount.
594 struct vfsmount
*lookup_mnt(struct path
*path
)
596 struct mount
*child_mnt
;
598 br_read_lock(&vfsmount_lock
);
599 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
601 mnt_add_count(child_mnt
, 1);
602 br_read_unlock(&vfsmount_lock
);
603 return &child_mnt
->mnt
;
605 br_read_unlock(&vfsmount_lock
);
610 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
612 struct list_head
*chain
= mountpoint_hashtable
+ hash(NULL
, dentry
);
613 struct mountpoint
*mp
;
615 list_for_each_entry(mp
, chain
, m_hash
) {
616 if (mp
->m_dentry
== dentry
) {
617 /* might be worth a WARN_ON() */
618 if (d_unlinked(dentry
))
619 return ERR_PTR(-ENOENT
);
625 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
627 return ERR_PTR(-ENOMEM
);
629 spin_lock(&dentry
->d_lock
);
630 if (d_unlinked(dentry
)) {
631 spin_unlock(&dentry
->d_lock
);
633 return ERR_PTR(-ENOENT
);
635 dentry
->d_flags
|= DCACHE_MOUNTED
;
636 spin_unlock(&dentry
->d_lock
);
637 mp
->m_dentry
= dentry
;
639 list_add(&mp
->m_hash
, chain
);
643 static void put_mountpoint(struct mountpoint
*mp
)
645 if (!--mp
->m_count
) {
646 struct dentry
*dentry
= mp
->m_dentry
;
647 spin_lock(&dentry
->d_lock
);
648 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
649 spin_unlock(&dentry
->d_lock
);
650 list_del(&mp
->m_hash
);
655 static inline int check_mnt(struct mount
*mnt
)
657 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
661 * vfsmount lock must be held for write
663 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
667 wake_up_interruptible(&ns
->poll
);
672 * vfsmount lock must be held for write
674 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
676 if (ns
&& ns
->event
!= event
) {
678 wake_up_interruptible(&ns
->poll
);
683 * vfsmount lock must be held for write
685 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
687 old_path
->dentry
= mnt
->mnt_mountpoint
;
688 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
689 mnt
->mnt_parent
= mnt
;
690 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
691 list_del_init(&mnt
->mnt_child
);
692 list_del_init(&mnt
->mnt_hash
);
693 put_mountpoint(mnt
->mnt_mp
);
698 * vfsmount lock must be held for write
700 void mnt_set_mountpoint(struct mount
*mnt
,
701 struct mountpoint
*mp
,
702 struct mount
*child_mnt
)
705 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
706 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
707 child_mnt
->mnt_parent
= mnt
;
708 child_mnt
->mnt_mp
= mp
;
712 * vfsmount lock must be held for write
714 static void attach_mnt(struct mount
*mnt
,
715 struct mount
*parent
,
716 struct mountpoint
*mp
)
718 mnt_set_mountpoint(parent
, mp
, mnt
);
719 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
720 hash(&parent
->mnt
, mp
->m_dentry
));
721 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
725 * vfsmount lock must be held for write
727 static void commit_tree(struct mount
*mnt
)
729 struct mount
*parent
= mnt
->mnt_parent
;
732 struct mnt_namespace
*n
= parent
->mnt_ns
;
734 BUG_ON(parent
== mnt
);
736 list_add_tail(&head
, &mnt
->mnt_list
);
737 list_for_each_entry(m
, &head
, mnt_list
)
740 list_splice(&head
, n
->list
.prev
);
742 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
743 hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
744 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
745 touch_mnt_namespace(n
);
748 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
750 struct list_head
*next
= p
->mnt_mounts
.next
;
751 if (next
== &p
->mnt_mounts
) {
755 next
= p
->mnt_child
.next
;
756 if (next
!= &p
->mnt_parent
->mnt_mounts
)
761 return list_entry(next
, struct mount
, mnt_child
);
764 static struct mount
*skip_mnt_tree(struct mount
*p
)
766 struct list_head
*prev
= p
->mnt_mounts
.prev
;
767 while (prev
!= &p
->mnt_mounts
) {
768 p
= list_entry(prev
, struct mount
, mnt_child
);
769 prev
= p
->mnt_mounts
.prev
;
775 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
781 return ERR_PTR(-ENODEV
);
783 mnt
= alloc_vfsmnt(name
);
785 return ERR_PTR(-ENOMEM
);
787 if (flags
& MS_KERNMOUNT
)
788 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
790 root
= mount_fs(type
, flags
, name
, data
);
793 return ERR_CAST(root
);
796 mnt
->mnt
.mnt_root
= root
;
797 mnt
->mnt
.mnt_sb
= root
->d_sb
;
798 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
799 mnt
->mnt_parent
= mnt
;
800 br_write_lock(&vfsmount_lock
);
801 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
802 br_write_unlock(&vfsmount_lock
);
805 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
807 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
810 struct super_block
*sb
= old
->mnt
.mnt_sb
;
814 mnt
= alloc_vfsmnt(old
->mnt_devname
);
816 return ERR_PTR(-ENOMEM
);
818 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
819 mnt
->mnt_group_id
= 0; /* not a peer of original */
821 mnt
->mnt_group_id
= old
->mnt_group_id
;
823 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
824 err
= mnt_alloc_group_id(mnt
);
829 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
830 /* Don't allow unprivileged users to change mount flags */
831 if (flag
& CL_UNPRIVILEGED
) {
832 mnt
->mnt
.mnt_flags
|= MNT_LOCK_ATIME
;
834 if (mnt
->mnt
.mnt_flags
& MNT_READONLY
)
835 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
837 if (mnt
->mnt
.mnt_flags
& MNT_NODEV
)
838 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NODEV
;
840 if (mnt
->mnt
.mnt_flags
& MNT_NOSUID
)
841 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOSUID
;
843 if (mnt
->mnt
.mnt_flags
& MNT_NOEXEC
)
844 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOEXEC
;
847 atomic_inc(&sb
->s_active
);
848 mnt
->mnt
.mnt_sb
= sb
;
849 mnt
->mnt
.mnt_root
= dget(root
);
850 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
851 mnt
->mnt_parent
= mnt
;
852 br_write_lock(&vfsmount_lock
);
853 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
854 br_write_unlock(&vfsmount_lock
);
856 if ((flag
& CL_SLAVE
) ||
857 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
858 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
859 mnt
->mnt_master
= old
;
860 CLEAR_MNT_SHARED(mnt
);
861 } else if (!(flag
& CL_PRIVATE
)) {
862 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
863 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
864 if (IS_MNT_SLAVE(old
))
865 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
866 mnt
->mnt_master
= old
->mnt_master
;
868 if (flag
& CL_MAKE_SHARED
)
871 /* stick the duplicate mount on the same expiry list
872 * as the original if that was on one */
873 if (flag
& CL_EXPIRE
) {
874 if (!list_empty(&old
->mnt_expire
))
875 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
885 static inline void mntfree(struct mount
*mnt
)
887 struct vfsmount
*m
= &mnt
->mnt
;
888 struct super_block
*sb
= m
->mnt_sb
;
891 * This probably indicates that somebody messed
892 * up a mnt_want/drop_write() pair. If this
893 * happens, the filesystem was probably unable
894 * to make r/w->r/o transitions.
897 * The locking used to deal with mnt_count decrement provides barriers,
898 * so mnt_get_writers() below is safe.
900 WARN_ON(mnt_get_writers(mnt
));
901 fsnotify_vfsmount_delete(m
);
904 deactivate_super(sb
);
907 static void mntput_no_expire(struct mount
*mnt
)
911 br_read_lock(&vfsmount_lock
);
912 if (likely(mnt
->mnt_ns
)) {
913 /* shouldn't be the last one */
914 mnt_add_count(mnt
, -1);
915 br_read_unlock(&vfsmount_lock
);
918 br_read_unlock(&vfsmount_lock
);
920 br_write_lock(&vfsmount_lock
);
921 mnt_add_count(mnt
, -1);
922 if (mnt_get_count(mnt
)) {
923 br_write_unlock(&vfsmount_lock
);
927 mnt_add_count(mnt
, -1);
928 if (likely(mnt_get_count(mnt
)))
930 br_write_lock(&vfsmount_lock
);
932 if (unlikely(mnt
->mnt_pinned
)) {
933 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
935 br_write_unlock(&vfsmount_lock
);
936 acct_auto_close_mnt(&mnt
->mnt
);
940 list_del(&mnt
->mnt_instance
);
941 br_write_unlock(&vfsmount_lock
);
945 void mntput(struct vfsmount
*mnt
)
948 struct mount
*m
= real_mount(mnt
);
949 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
950 if (unlikely(m
->mnt_expiry_mark
))
951 m
->mnt_expiry_mark
= 0;
955 EXPORT_SYMBOL(mntput
);
957 struct vfsmount
*mntget(struct vfsmount
*mnt
)
960 mnt_add_count(real_mount(mnt
), 1);
963 EXPORT_SYMBOL(mntget
);
965 void mnt_pin(struct vfsmount
*mnt
)
967 br_write_lock(&vfsmount_lock
);
968 real_mount(mnt
)->mnt_pinned
++;
969 br_write_unlock(&vfsmount_lock
);
971 EXPORT_SYMBOL(mnt_pin
);
973 void mnt_unpin(struct vfsmount
*m
)
975 struct mount
*mnt
= real_mount(m
);
976 br_write_lock(&vfsmount_lock
);
977 if (mnt
->mnt_pinned
) {
978 mnt_add_count(mnt
, 1);
981 br_write_unlock(&vfsmount_lock
);
983 EXPORT_SYMBOL(mnt_unpin
);
985 static inline void mangle(struct seq_file
*m
, const char *s
)
987 seq_escape(m
, s
, " \t\n\\");
991 * Simple .show_options callback for filesystems which don't want to
992 * implement more complex mount option showing.
994 * See also save_mount_options().
996 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1001 options
= rcu_dereference(root
->d_sb
->s_options
);
1003 if (options
!= NULL
&& options
[0]) {
1011 EXPORT_SYMBOL(generic_show_options
);
1014 * If filesystem uses generic_show_options(), this function should be
1015 * called from the fill_super() callback.
1017 * The .remount_fs callback usually needs to be handled in a special
1018 * way, to make sure, that previous options are not overwritten if the
1021 * Also note, that if the filesystem's .remount_fs function doesn't
1022 * reset all options to their default value, but changes only newly
1023 * given options, then the displayed options will not reflect reality
1026 void save_mount_options(struct super_block
*sb
, char *options
)
1028 BUG_ON(sb
->s_options
);
1029 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1031 EXPORT_SYMBOL(save_mount_options
);
1033 void replace_mount_options(struct super_block
*sb
, char *options
)
1035 char *old
= sb
->s_options
;
1036 rcu_assign_pointer(sb
->s_options
, options
);
1042 EXPORT_SYMBOL(replace_mount_options
);
1044 #ifdef CONFIG_PROC_FS
1045 /* iterator; we want it to have access to namespace_sem, thus here... */
1046 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1048 struct proc_mounts
*p
= proc_mounts(m
);
1050 down_read(&namespace_sem
);
1051 return seq_list_start(&p
->ns
->list
, *pos
);
1054 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1056 struct proc_mounts
*p
= proc_mounts(m
);
1058 return seq_list_next(v
, &p
->ns
->list
, pos
);
1061 static void m_stop(struct seq_file
*m
, void *v
)
1063 up_read(&namespace_sem
);
1066 static int m_show(struct seq_file
*m
, void *v
)
1068 struct proc_mounts
*p
= proc_mounts(m
);
1069 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1070 return p
->show(m
, &r
->mnt
);
1073 const struct seq_operations mounts_op
= {
1079 #endif /* CONFIG_PROC_FS */
1082 * may_umount_tree - check if a mount tree is busy
1083 * @mnt: root of mount tree
1085 * This is called to check if a tree of mounts has any
1086 * open files, pwds, chroots or sub mounts that are
1089 int may_umount_tree(struct vfsmount
*m
)
1091 struct mount
*mnt
= real_mount(m
);
1092 int actual_refs
= 0;
1093 int minimum_refs
= 0;
1097 /* write lock needed for mnt_get_count */
1098 br_write_lock(&vfsmount_lock
);
1099 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1100 actual_refs
+= mnt_get_count(p
);
1103 br_write_unlock(&vfsmount_lock
);
1105 if (actual_refs
> minimum_refs
)
1111 EXPORT_SYMBOL(may_umount_tree
);
1114 * may_umount - check if a mount point is busy
1115 * @mnt: root of mount
1117 * This is called to check if a mount point has any
1118 * open files, pwds, chroots or sub mounts. If the
1119 * mount has sub mounts this will return busy
1120 * regardless of whether the sub mounts are busy.
1122 * Doesn't take quota and stuff into account. IOW, in some cases it will
1123 * give false negatives. The main reason why it's here is that we need
1124 * a non-destructive way to look for easily umountable filesystems.
1126 int may_umount(struct vfsmount
*mnt
)
1129 down_read(&namespace_sem
);
1130 br_write_lock(&vfsmount_lock
);
1131 if (propagate_mount_busy(real_mount(mnt
), 2))
1133 br_write_unlock(&vfsmount_lock
);
1134 up_read(&namespace_sem
);
1138 EXPORT_SYMBOL(may_umount
);
1140 static LIST_HEAD(unmounted
); /* protected by namespace_sem */
1142 static void namespace_unlock(void)
1147 if (likely(list_empty(&unmounted
))) {
1148 up_write(&namespace_sem
);
1152 list_splice_init(&unmounted
, &head
);
1153 up_write(&namespace_sem
);
1155 while (!list_empty(&head
)) {
1156 mnt
= list_first_entry(&head
, struct mount
, mnt_hash
);
1157 list_del_init(&mnt
->mnt_hash
);
1158 if (mnt_has_parent(mnt
)) {
1159 struct dentry
*dentry
;
1162 br_write_lock(&vfsmount_lock
);
1163 dentry
= mnt
->mnt_mountpoint
;
1164 m
= mnt
->mnt_parent
;
1165 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1166 mnt
->mnt_parent
= mnt
;
1168 br_write_unlock(&vfsmount_lock
);
1176 static inline void namespace_lock(void)
1178 down_write(&namespace_sem
);
1182 * vfsmount lock must be held for write
1183 * namespace_sem must be held for write
1185 void umount_tree(struct mount
*mnt
, int propagate
)
1187 LIST_HEAD(tmp_list
);
1190 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1191 list_move(&p
->mnt_hash
, &tmp_list
);
1194 propagate_umount(&tmp_list
);
1196 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1197 list_del_init(&p
->mnt_expire
);
1198 list_del_init(&p
->mnt_list
);
1199 __touch_mnt_namespace(p
->mnt_ns
);
1201 list_del_init(&p
->mnt_child
);
1202 if (mnt_has_parent(p
)) {
1203 p
->mnt_parent
->mnt_ghosts
++;
1204 put_mountpoint(p
->mnt_mp
);
1207 change_mnt_propagation(p
, MS_PRIVATE
);
1209 list_splice(&tmp_list
, &unmounted
);
1212 static void shrink_submounts(struct mount
*mnt
);
1214 static int do_umount(struct mount
*mnt
, int flags
)
1216 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1219 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1224 * Allow userspace to request a mountpoint be expired rather than
1225 * unmounting unconditionally. Unmount only happens if:
1226 * (1) the mark is already set (the mark is cleared by mntput())
1227 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1229 if (flags
& MNT_EXPIRE
) {
1230 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1231 flags
& (MNT_FORCE
| MNT_DETACH
))
1235 * probably don't strictly need the lock here if we examined
1236 * all race cases, but it's a slowpath.
1238 br_write_lock(&vfsmount_lock
);
1239 if (mnt_get_count(mnt
) != 2) {
1240 br_write_unlock(&vfsmount_lock
);
1243 br_write_unlock(&vfsmount_lock
);
1245 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1250 * If we may have to abort operations to get out of this
1251 * mount, and they will themselves hold resources we must
1252 * allow the fs to do things. In the Unix tradition of
1253 * 'Gee thats tricky lets do it in userspace' the umount_begin
1254 * might fail to complete on the first run through as other tasks
1255 * must return, and the like. Thats for the mount program to worry
1256 * about for the moment.
1259 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1260 sb
->s_op
->umount_begin(sb
);
1264 * No sense to grab the lock for this test, but test itself looks
1265 * somewhat bogus. Suggestions for better replacement?
1266 * Ho-hum... In principle, we might treat that as umount + switch
1267 * to rootfs. GC would eventually take care of the old vfsmount.
1268 * Actually it makes sense, especially if rootfs would contain a
1269 * /reboot - static binary that would close all descriptors and
1270 * call reboot(9). Then init(8) could umount root and exec /reboot.
1272 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1274 * Special case for "unmounting" root ...
1275 * we just try to remount it readonly.
1277 if (!capable(CAP_SYS_ADMIN
))
1279 down_write(&sb
->s_umount
);
1280 if (!(sb
->s_flags
& MS_RDONLY
))
1281 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1282 up_write(&sb
->s_umount
);
1287 br_write_lock(&vfsmount_lock
);
1290 if (!(flags
& MNT_DETACH
))
1291 shrink_submounts(mnt
);
1294 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1295 if (!list_empty(&mnt
->mnt_list
))
1296 umount_tree(mnt
, 1);
1299 br_write_unlock(&vfsmount_lock
);
1305 * Is the caller allowed to modify his namespace?
1307 static inline bool may_mount(void)
1309 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1313 * Now umount can handle mount points as well as block devices.
1314 * This is important for filesystems which use unnamed block devices.
1316 * We now support a flag for forced unmount like the other 'big iron'
1317 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1320 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1325 int lookup_flags
= 0;
1327 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1333 if (!(flags
& UMOUNT_NOFOLLOW
))
1334 lookup_flags
|= LOOKUP_FOLLOW
;
1336 retval
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1339 mnt
= real_mount(path
.mnt
);
1341 if (path
.dentry
!= path
.mnt
->mnt_root
)
1343 if (!check_mnt(mnt
))
1346 retval
= do_umount(mnt
, flags
);
1348 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1350 mntput_no_expire(mnt
);
1355 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1358 * The 2.0 compatible umount. No flags.
1360 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1362 return sys_umount(name
, 0);
1367 static bool mnt_ns_loop(struct path
*path
)
1369 /* Could bind mounting the mount namespace inode cause a
1370 * mount namespace loop?
1372 struct inode
*inode
= path
->dentry
->d_inode
;
1374 struct mnt_namespace
*mnt_ns
;
1376 if (!proc_ns_inode(inode
))
1379 ei
= get_proc_ns(inode
);
1380 if (ei
->ns_ops
!= &mntns_operations
)
1384 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1387 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1390 struct mount
*res
, *p
, *q
, *r
, *parent
;
1392 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1393 return ERR_PTR(-EINVAL
);
1395 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1399 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1402 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1404 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1407 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1408 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
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
);
1715 * do loopback mount.
1717 static int do_loopback(struct path
*path
, const char *old_name
,
1720 struct path old_path
;
1721 struct mount
*mnt
= NULL
, *old
, *parent
;
1722 struct mountpoint
*mp
;
1724 if (!old_name
|| !*old_name
)
1726 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1731 if (mnt_ns_loop(&old_path
))
1734 mp
= lock_mount(path
);
1739 old
= real_mount(old_path
.mnt
);
1740 parent
= real_mount(path
->mnt
);
1743 if (IS_MNT_UNBINDABLE(old
))
1746 if (!check_mnt(parent
) || !check_mnt(old
))
1750 mnt
= copy_tree(old
, old_path
.dentry
, 0);
1752 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1759 err
= graft_tree(mnt
, parent
, mp
);
1761 br_write_lock(&vfsmount_lock
);
1762 umount_tree(mnt
, 0);
1763 br_write_unlock(&vfsmount_lock
);
1768 path_put(&old_path
);
1772 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1775 int readonly_request
= 0;
1777 if (ms_flags
& MS_RDONLY
)
1778 readonly_request
= 1;
1779 if (readonly_request
== __mnt_is_readonly(mnt
))
1782 if (readonly_request
)
1783 error
= mnt_make_readonly(real_mount(mnt
));
1785 __mnt_unmake_readonly(real_mount(mnt
));
1790 * change filesystem flags. dir should be a physical root of filesystem.
1791 * If you've mounted a non-root directory somewhere and want to do remount
1792 * on it - tough luck.
1794 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1798 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1799 struct mount
*mnt
= real_mount(path
->mnt
);
1801 if (!check_mnt(mnt
))
1804 if (path
->dentry
!= path
->mnt
->mnt_root
)
1807 /* Don't allow changing of locked mnt flags.
1809 * No locks need to be held here while testing the various
1810 * MNT_LOCK flags because those flags can never be cleared
1811 * once they are set.
1813 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_READONLY
) &&
1814 !(mnt_flags
& MNT_READONLY
)) {
1817 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NODEV
) &&
1818 !(mnt_flags
& MNT_NODEV
)) {
1819 /* Was the nodev implicitly added in mount? */
1820 if ((mnt
->mnt_ns
->user_ns
!= &init_user_ns
) &&
1821 !(sb
->s_type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
1822 mnt_flags
|= MNT_NODEV
;
1827 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOSUID
) &&
1828 !(mnt_flags
& MNT_NOSUID
)) {
1831 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOEXEC
) &&
1832 !(mnt_flags
& MNT_NOEXEC
)) {
1835 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_ATIME
) &&
1836 ((mnt
->mnt
.mnt_flags
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
))) {
1840 err
= security_sb_remount(sb
, data
);
1844 down_write(&sb
->s_umount
);
1845 if (flags
& MS_BIND
)
1846 err
= change_mount_flags(path
->mnt
, flags
);
1847 else if (!capable(CAP_SYS_ADMIN
))
1850 err
= do_remount_sb(sb
, flags
, data
, 0);
1852 br_write_lock(&vfsmount_lock
);
1853 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
1854 mnt
->mnt
.mnt_flags
= mnt_flags
;
1855 br_write_unlock(&vfsmount_lock
);
1857 up_write(&sb
->s_umount
);
1859 br_write_lock(&vfsmount_lock
);
1860 touch_mnt_namespace(mnt
->mnt_ns
);
1861 br_write_unlock(&vfsmount_lock
);
1866 static inline int tree_contains_unbindable(struct mount
*mnt
)
1869 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1870 if (IS_MNT_UNBINDABLE(p
))
1876 static int do_move_mount(struct path
*path
, const char *old_name
)
1878 struct path old_path
, parent_path
;
1881 struct mountpoint
*mp
;
1883 if (!old_name
|| !*old_name
)
1885 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1889 mp
= lock_mount(path
);
1894 old
= real_mount(old_path
.mnt
);
1895 p
= real_mount(path
->mnt
);
1898 if (!check_mnt(p
) || !check_mnt(old
))
1902 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1905 if (!mnt_has_parent(old
))
1908 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1909 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1912 * Don't move a mount residing in a shared parent.
1914 if (IS_MNT_SHARED(old
->mnt_parent
))
1917 * Don't move a mount tree containing unbindable mounts to a destination
1918 * mount which is shared.
1920 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
1923 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
1927 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
1931 /* if the mount is moved, it should no longer be expire
1933 list_del_init(&old
->mnt_expire
);
1938 path_put(&parent_path
);
1939 path_put(&old_path
);
1943 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
1946 const char *subtype
= strchr(fstype
, '.');
1955 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
1957 if (!mnt
->mnt_sb
->s_subtype
)
1963 return ERR_PTR(err
);
1967 * add a mount into a namespace's mount tree
1969 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
1971 struct mountpoint
*mp
;
1972 struct mount
*parent
;
1975 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
1977 mp
= lock_mount(path
);
1981 parent
= real_mount(path
->mnt
);
1983 if (unlikely(!check_mnt(parent
))) {
1984 /* that's acceptable only for automounts done in private ns */
1985 if (!(mnt_flags
& MNT_SHRINKABLE
))
1987 /* ... and for those we'd better have mountpoint still alive */
1988 if (!parent
->mnt_ns
)
1992 /* Refuse the same filesystem on the same mount point */
1994 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
1995 path
->mnt
->mnt_root
== path
->dentry
)
1999 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
2002 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2003 err
= graft_tree(newmnt
, parent
, mp
);
2011 * create a new mount for userspace and request it to be added into the
2014 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2015 int mnt_flags
, const char *name
, void *data
)
2017 struct file_system_type
*type
;
2018 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2019 struct vfsmount
*mnt
;
2025 type
= get_fs_type(fstype
);
2029 if (user_ns
!= &init_user_ns
) {
2030 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2031 put_filesystem(type
);
2034 /* Only in special cases allow devices from mounts
2035 * created outside the initial user namespace.
2037 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2039 mnt_flags
|= MNT_NODEV
| MNT_LOCK_NODEV
;
2043 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2044 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2045 !mnt
->mnt_sb
->s_subtype
)
2046 mnt
= fs_set_subtype(mnt
, fstype
);
2048 put_filesystem(type
);
2050 return PTR_ERR(mnt
);
2052 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2058 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2060 struct mount
*mnt
= real_mount(m
);
2062 /* The new mount record should have at least 2 refs to prevent it being
2063 * expired before we get a chance to add it
2065 BUG_ON(mnt_get_count(mnt
) < 2);
2067 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2068 m
->mnt_root
== path
->dentry
) {
2073 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2077 /* remove m from any expiration list it may be on */
2078 if (!list_empty(&mnt
->mnt_expire
)) {
2080 br_write_lock(&vfsmount_lock
);
2081 list_del_init(&mnt
->mnt_expire
);
2082 br_write_unlock(&vfsmount_lock
);
2091 * mnt_set_expiry - Put a mount on an expiration list
2092 * @mnt: The mount to list.
2093 * @expiry_list: The list to add the mount to.
2095 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2098 br_write_lock(&vfsmount_lock
);
2100 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2102 br_write_unlock(&vfsmount_lock
);
2105 EXPORT_SYMBOL(mnt_set_expiry
);
2108 * process a list of expirable mountpoints with the intent of discarding any
2109 * mountpoints that aren't in use and haven't been touched since last we came
2112 void mark_mounts_for_expiry(struct list_head
*mounts
)
2114 struct mount
*mnt
, *next
;
2115 LIST_HEAD(graveyard
);
2117 if (list_empty(mounts
))
2121 br_write_lock(&vfsmount_lock
);
2123 /* extract from the expiration list every vfsmount that matches the
2124 * following criteria:
2125 * - only referenced by its parent vfsmount
2126 * - still marked for expiry (marked on the last call here; marks are
2127 * cleared by mntput())
2129 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2130 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2131 propagate_mount_busy(mnt
, 1))
2133 list_move(&mnt
->mnt_expire
, &graveyard
);
2135 while (!list_empty(&graveyard
)) {
2136 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2137 touch_mnt_namespace(mnt
->mnt_ns
);
2138 umount_tree(mnt
, 1);
2140 br_write_unlock(&vfsmount_lock
);
2144 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2147 * Ripoff of 'select_parent()'
2149 * search the list of submounts for a given mountpoint, and move any
2150 * shrinkable submounts to the 'graveyard' list.
2152 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2154 struct mount
*this_parent
= parent
;
2155 struct list_head
*next
;
2159 next
= this_parent
->mnt_mounts
.next
;
2161 while (next
!= &this_parent
->mnt_mounts
) {
2162 struct list_head
*tmp
= next
;
2163 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2166 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2169 * Descend a level if the d_mounts list is non-empty.
2171 if (!list_empty(&mnt
->mnt_mounts
)) {
2176 if (!propagate_mount_busy(mnt
, 1)) {
2177 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2182 * All done at this level ... ascend and resume the search
2184 if (this_parent
!= parent
) {
2185 next
= this_parent
->mnt_child
.next
;
2186 this_parent
= this_parent
->mnt_parent
;
2193 * process a list of expirable mountpoints with the intent of discarding any
2194 * submounts of a specific parent mountpoint
2196 * vfsmount_lock must be held for write
2198 static void shrink_submounts(struct mount
*mnt
)
2200 LIST_HEAD(graveyard
);
2203 /* extract submounts of 'mountpoint' from the expiration list */
2204 while (select_submounts(mnt
, &graveyard
)) {
2205 while (!list_empty(&graveyard
)) {
2206 m
= list_first_entry(&graveyard
, struct mount
,
2208 touch_mnt_namespace(m
->mnt_ns
);
2215 * Some copy_from_user() implementations do not return the exact number of
2216 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2217 * Note that this function differs from copy_from_user() in that it will oops
2218 * on bad values of `to', rather than returning a short copy.
2220 static long exact_copy_from_user(void *to
, const void __user
* from
,
2224 const char __user
*f
= from
;
2227 if (!access_ok(VERIFY_READ
, from
, n
))
2231 if (__get_user(c
, f
)) {
2242 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2252 if (!(page
= __get_free_page(GFP_KERNEL
)))
2255 /* We only care that *some* data at the address the user
2256 * gave us is valid. Just in case, we'll zero
2257 * the remainder of the page.
2259 /* copy_from_user cannot cross TASK_SIZE ! */
2260 size
= TASK_SIZE
- (unsigned long)data
;
2261 if (size
> PAGE_SIZE
)
2264 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2270 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2275 int copy_mount_string(const void __user
*data
, char **where
)
2284 tmp
= strndup_user(data
, PAGE_SIZE
);
2286 return PTR_ERR(tmp
);
2293 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2294 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2296 * data is a (void *) that can point to any structure up to
2297 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2298 * information (or be NULL).
2300 * Pre-0.97 versions of mount() didn't have a flags word.
2301 * When the flags word was introduced its top half was required
2302 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2303 * Therefore, if this magic number is present, it carries no information
2304 * and must be discarded.
2306 long do_mount(const char *dev_name
, const char *dir_name
,
2307 const char *type_page
, unsigned long flags
, void *data_page
)
2314 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2315 flags
&= ~MS_MGC_MSK
;
2317 /* Basic sanity checks */
2319 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2323 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2325 /* ... and get the mountpoint */
2326 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2330 retval
= security_sb_mount(dev_name
, &path
,
2331 type_page
, flags
, data_page
);
2332 if (!retval
&& !may_mount())
2337 /* Default to relatime unless overriden */
2338 if (!(flags
& MS_NOATIME
))
2339 mnt_flags
|= MNT_RELATIME
;
2341 /* Separate the per-mountpoint flags */
2342 if (flags
& MS_NOSUID
)
2343 mnt_flags
|= MNT_NOSUID
;
2344 if (flags
& MS_NODEV
)
2345 mnt_flags
|= MNT_NODEV
;
2346 if (flags
& MS_NOEXEC
)
2347 mnt_flags
|= MNT_NOEXEC
;
2348 if (flags
& MS_NOATIME
)
2349 mnt_flags
|= MNT_NOATIME
;
2350 if (flags
& MS_NODIRATIME
)
2351 mnt_flags
|= MNT_NODIRATIME
;
2352 if (flags
& MS_STRICTATIME
)
2353 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2354 if (flags
& MS_RDONLY
)
2355 mnt_flags
|= MNT_READONLY
;
2357 /* The default atime for remount is preservation */
2358 if ((flags
& MS_REMOUNT
) &&
2359 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
2360 MS_STRICTATIME
)) == 0)) {
2361 mnt_flags
&= ~MNT_ATIME_MASK
;
2362 mnt_flags
|= path
.mnt
->mnt_flags
& MNT_ATIME_MASK
;
2365 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2366 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2369 if (flags
& MS_REMOUNT
)
2370 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2372 else if (flags
& MS_BIND
)
2373 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2374 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2375 retval
= do_change_type(&path
, flags
);
2376 else if (flags
& MS_MOVE
)
2377 retval
= do_move_mount(&path
, dev_name
);
2379 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2380 dev_name
, data_page
);
2386 static void free_mnt_ns(struct mnt_namespace
*ns
)
2388 proc_free_inum(ns
->proc_inum
);
2389 put_user_ns(ns
->user_ns
);
2394 * Assign a sequence number so we can detect when we attempt to bind
2395 * mount a reference to an older mount namespace into the current
2396 * mount namespace, preventing reference counting loops. A 64bit
2397 * number incrementing at 10Ghz will take 12,427 years to wrap which
2398 * is effectively never, so we can ignore the possibility.
2400 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2402 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2404 struct mnt_namespace
*new_ns
;
2407 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2409 return ERR_PTR(-ENOMEM
);
2410 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2413 return ERR_PTR(ret
);
2415 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2416 atomic_set(&new_ns
->count
, 1);
2417 new_ns
->root
= NULL
;
2418 INIT_LIST_HEAD(&new_ns
->list
);
2419 init_waitqueue_head(&new_ns
->poll
);
2421 new_ns
->user_ns
= get_user_ns(user_ns
);
2426 * Allocate a new namespace structure and populate it with contents
2427 * copied from the namespace of the passed in task structure.
2429 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2430 struct user_namespace
*user_ns
, struct fs_struct
*fs
)
2432 struct mnt_namespace
*new_ns
;
2433 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2434 struct mount
*p
, *q
;
2435 struct mount
*old
= mnt_ns
->root
;
2439 new_ns
= alloc_mnt_ns(user_ns
);
2444 /* First pass: copy the tree topology */
2445 copy_flags
= CL_COPY_ALL
| CL_EXPIRE
;
2446 if (user_ns
!= mnt_ns
->user_ns
)
2447 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2448 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2451 free_mnt_ns(new_ns
);
2452 return ERR_CAST(new);
2455 br_write_lock(&vfsmount_lock
);
2456 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2457 br_write_unlock(&vfsmount_lock
);
2460 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2461 * as belonging to new namespace. We have already acquired a private
2462 * fs_struct, so tsk->fs->lock is not needed.
2469 if (&p
->mnt
== fs
->root
.mnt
) {
2470 fs
->root
.mnt
= mntget(&q
->mnt
);
2473 if (&p
->mnt
== fs
->pwd
.mnt
) {
2474 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2478 p
= next_mnt(p
, old
);
2479 q
= next_mnt(q
, new);
2491 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2492 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2494 struct mnt_namespace
*new_ns
;
2499 if (!(flags
& CLONE_NEWNS
))
2502 new_ns
= dup_mnt_ns(ns
, user_ns
, new_fs
);
2509 * create_mnt_ns - creates a private namespace and adds a root filesystem
2510 * @mnt: pointer to the new root filesystem mountpoint
2512 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2514 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2515 if (!IS_ERR(new_ns
)) {
2516 struct mount
*mnt
= real_mount(m
);
2517 mnt
->mnt_ns
= new_ns
;
2519 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2526 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2528 struct mnt_namespace
*ns
;
2529 struct super_block
*s
;
2533 ns
= create_mnt_ns(mnt
);
2535 return ERR_CAST(ns
);
2537 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2538 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2543 return ERR_PTR(err
);
2545 /* trade a vfsmount reference for active sb one */
2546 s
= path
.mnt
->mnt_sb
;
2547 atomic_inc(&s
->s_active
);
2549 /* lock the sucker */
2550 down_write(&s
->s_umount
);
2551 /* ... and return the root of (sub)tree on it */
2554 EXPORT_SYMBOL(mount_subtree
);
2556 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2557 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2561 struct filename
*kernel_dir
;
2563 unsigned long data_page
;
2565 ret
= copy_mount_string(type
, &kernel_type
);
2569 kernel_dir
= getname(dir_name
);
2570 if (IS_ERR(kernel_dir
)) {
2571 ret
= PTR_ERR(kernel_dir
);
2575 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2579 ret
= copy_mount_options(data
, &data_page
);
2583 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2584 (void *) data_page
);
2586 free_page(data_page
);
2590 putname(kernel_dir
);
2598 * Return true if path is reachable from root
2600 * namespace_sem or vfsmount_lock is held
2602 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2603 const struct path
*root
)
2605 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2606 dentry
= mnt
->mnt_mountpoint
;
2607 mnt
= mnt
->mnt_parent
;
2609 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2612 int path_is_under(struct path
*path1
, struct path
*path2
)
2615 br_read_lock(&vfsmount_lock
);
2616 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2617 br_read_unlock(&vfsmount_lock
);
2620 EXPORT_SYMBOL(path_is_under
);
2623 * pivot_root Semantics:
2624 * Moves the root file system of the current process to the directory put_old,
2625 * makes new_root as the new root file system of the current process, and sets
2626 * root/cwd of all processes which had them on the current root to new_root.
2629 * The new_root and put_old must be directories, and must not be on the
2630 * same file system as the current process root. The put_old must be
2631 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2632 * pointed to by put_old must yield the same directory as new_root. No other
2633 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2635 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2636 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2637 * in this situation.
2640 * - we don't move root/cwd if they are not at the root (reason: if something
2641 * cared enough to change them, it's probably wrong to force them elsewhere)
2642 * - it's okay to pick a root that isn't the root of a file system, e.g.
2643 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2644 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2647 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2648 const char __user
*, put_old
)
2650 struct path
new, old
, parent_path
, root_parent
, root
;
2651 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2652 struct mountpoint
*old_mp
, *root_mp
;
2658 error
= user_path_dir(new_root
, &new);
2662 error
= user_path_dir(put_old
, &old
);
2666 error
= security_sb_pivotroot(&old
, &new);
2670 get_fs_root(current
->fs
, &root
);
2671 old_mp
= lock_mount(&old
);
2672 error
= PTR_ERR(old_mp
);
2677 new_mnt
= real_mount(new.mnt
);
2678 root_mnt
= real_mount(root
.mnt
);
2679 old_mnt
= real_mount(old
.mnt
);
2680 if (IS_MNT_SHARED(old_mnt
) ||
2681 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2682 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2684 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2687 if (d_unlinked(new.dentry
))
2690 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2691 goto out4
; /* loop, on the same file system */
2693 if (root
.mnt
->mnt_root
!= root
.dentry
)
2694 goto out4
; /* not a mountpoint */
2695 if (!mnt_has_parent(root_mnt
))
2696 goto out4
; /* not attached */
2697 root_mp
= root_mnt
->mnt_mp
;
2698 if (new.mnt
->mnt_root
!= new.dentry
)
2699 goto out4
; /* not a mountpoint */
2700 if (!mnt_has_parent(new_mnt
))
2701 goto out4
; /* not attached */
2702 /* make sure we can reach put_old from new_root */
2703 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2705 /* make certain new is below the root */
2706 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
2708 root_mp
->m_count
++; /* pin it so it won't go away */
2709 br_write_lock(&vfsmount_lock
);
2710 detach_mnt(new_mnt
, &parent_path
);
2711 detach_mnt(root_mnt
, &root_parent
);
2712 /* mount old root on put_old */
2713 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2714 /* mount new_root on / */
2715 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2716 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2717 br_write_unlock(&vfsmount_lock
);
2718 chroot_fs_refs(&root
, &new);
2719 put_mountpoint(root_mp
);
2722 unlock_mount(old_mp
);
2724 path_put(&root_parent
);
2725 path_put(&parent_path
);
2737 static void __init
init_mount_tree(void)
2739 struct vfsmount
*mnt
;
2740 struct mnt_namespace
*ns
;
2742 struct file_system_type
*type
;
2744 type
= get_fs_type("rootfs");
2746 panic("Can't find rootfs type");
2747 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2748 put_filesystem(type
);
2750 panic("Can't create rootfs");
2752 ns
= create_mnt_ns(mnt
);
2754 panic("Can't allocate initial namespace");
2756 init_task
.nsproxy
->mnt_ns
= ns
;
2760 root
.dentry
= mnt
->mnt_root
;
2762 set_fs_pwd(current
->fs
, &root
);
2763 set_fs_root(current
->fs
, &root
);
2766 void __init
mnt_init(void)
2771 init_rwsem(&namespace_sem
);
2773 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2774 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2776 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2777 mountpoint_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2779 if (!mount_hashtable
|| !mountpoint_hashtable
)
2780 panic("Failed to allocate mount hash table\n");
2782 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2784 for (u
= 0; u
< HASH_SIZE
; u
++)
2785 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2786 for (u
= 0; u
< HASH_SIZE
; u
++)
2787 INIT_LIST_HEAD(&mountpoint_hashtable
[u
]);
2789 br_lock_init(&vfsmount_lock
);
2793 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2795 fs_kobj
= kobject_create_and_add("fs", NULL
);
2797 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2802 void put_mnt_ns(struct mnt_namespace
*ns
)
2804 if (!atomic_dec_and_test(&ns
->count
))
2807 br_write_lock(&vfsmount_lock
);
2808 umount_tree(ns
->root
, 0);
2809 br_write_unlock(&vfsmount_lock
);
2814 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2816 struct vfsmount
*mnt
;
2817 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2820 * it is a longterm mount, don't release mnt until
2821 * we unmount before file sys is unregistered
2823 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2827 EXPORT_SYMBOL_GPL(kern_mount_data
);
2829 void kern_unmount(struct vfsmount
*mnt
)
2831 /* release long term mount so mount point can be released */
2832 if (!IS_ERR_OR_NULL(mnt
)) {
2833 br_write_lock(&vfsmount_lock
);
2834 real_mount(mnt
)->mnt_ns
= NULL
;
2835 br_write_unlock(&vfsmount_lock
);
2839 EXPORT_SYMBOL(kern_unmount
);
2841 bool our_mnt(struct vfsmount
*mnt
)
2843 return check_mnt(real_mount(mnt
));
2846 bool current_chrooted(void)
2848 /* Does the current process have a non-standard root */
2849 struct path ns_root
;
2850 struct path fs_root
;
2853 /* Find the namespace root */
2854 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
2855 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
2857 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
2860 get_fs_root(current
->fs
, &fs_root
);
2862 chrooted
= !path_equal(&fs_root
, &ns_root
);
2870 void update_mnt_policy(struct user_namespace
*userns
)
2872 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
2875 down_read(&namespace_sem
);
2876 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
2877 switch (mnt
->mnt
.mnt_sb
->s_magic
) {
2879 userns
->may_mount_sysfs
= true;
2881 case PROC_SUPER_MAGIC
:
2882 userns
->may_mount_proc
= true;
2885 if (userns
->may_mount_sysfs
&& userns
->may_mount_proc
)
2888 up_read(&namespace_sem
);
2891 static void *mntns_get(struct task_struct
*task
)
2893 struct mnt_namespace
*ns
= NULL
;
2894 struct nsproxy
*nsproxy
;
2897 nsproxy
= task_nsproxy(task
);
2899 ns
= nsproxy
->mnt_ns
;
2907 static void mntns_put(void *ns
)
2912 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
2914 struct fs_struct
*fs
= current
->fs
;
2915 struct mnt_namespace
*mnt_ns
= ns
;
2918 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
2919 !nsown_capable(CAP_SYS_CHROOT
) ||
2920 !nsown_capable(CAP_SYS_ADMIN
))
2927 put_mnt_ns(nsproxy
->mnt_ns
);
2928 nsproxy
->mnt_ns
= mnt_ns
;
2931 root
.mnt
= &mnt_ns
->root
->mnt
;
2932 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
2934 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
2937 /* Update the pwd and root */
2938 set_fs_pwd(fs
, &root
);
2939 set_fs_root(fs
, &root
);
2945 static unsigned int mntns_inum(void *ns
)
2947 struct mnt_namespace
*mnt_ns
= ns
;
2948 return mnt_ns
->proc_inum
;
2951 const struct proc_ns_operations mntns_operations
= {
2953 .type
= CLONE_NEWNS
,
2956 .install
= mntns_install
,