4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/percpu.h>
16 #include <linux/init.h>
17 #include <linux/kernel.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/cpumask.h>
21 #include <linux/module.h>
22 #include <linux/sysfs.h>
23 #include <linux/seq_file.h>
24 #include <linux/mnt_namespace.h>
25 #include <linux/namei.h>
26 #include <linux/nsproxy.h>
27 #include <linux/security.h>
28 #include <linux/mount.h>
29 #include <linux/ramfs.h>
30 #include <linux/log2.h>
31 #include <linux/idr.h>
32 #include <linux/fs_struct.h>
33 #include <linux/fsnotify.h>
34 #include <asm/uaccess.h>
35 #include <asm/unistd.h>
39 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40 #define HASH_SIZE (1UL << HASH_SHIFT)
43 static DEFINE_IDA(mnt_id_ida
);
44 static DEFINE_IDA(mnt_group_ida
);
45 static DEFINE_SPINLOCK(mnt_id_lock
);
46 static int mnt_id_start
= 0;
47 static int mnt_group_start
= 1;
49 static struct list_head
*mount_hashtable __read_mostly
;
50 static struct kmem_cache
*mnt_cache __read_mostly
;
51 static struct rw_semaphore namespace_sem
;
54 struct kobject
*fs_kobj
;
55 EXPORT_SYMBOL_GPL(fs_kobj
);
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
65 DEFINE_BRLOCK(vfsmount_lock
);
67 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
69 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
70 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
71 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
72 return tmp
& (HASH_SIZE
- 1);
75 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
81 static int mnt_alloc_id(struct mount
*mnt
)
86 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
87 spin_lock(&mnt_id_lock
);
88 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt
.mnt_id
);
90 mnt_id_start
= mnt
->mnt
.mnt_id
+ 1;
91 spin_unlock(&mnt_id_lock
);
98 static void mnt_free_id(struct mount
*mnt
)
100 int id
= mnt
->mnt
.mnt_id
;
101 spin_lock(&mnt_id_lock
);
102 ida_remove(&mnt_id_ida
, id
);
103 if (mnt_id_start
> id
)
105 spin_unlock(&mnt_id_lock
);
109 * Allocate a new peer group ID
111 * mnt_group_ida is protected by namespace_sem
113 static int mnt_alloc_group_id(struct mount
*mnt
)
117 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
120 res
= ida_get_new_above(&mnt_group_ida
,
122 &mnt
->mnt
.mnt_group_id
);
124 mnt_group_start
= mnt
->mnt
.mnt_group_id
+ 1;
130 * Release a peer group ID
132 void mnt_release_group_id(struct mount
*mnt
)
134 int id
= mnt
->mnt
.mnt_group_id
;
135 ida_remove(&mnt_group_ida
, id
);
136 if (mnt_group_start
> id
)
137 mnt_group_start
= id
;
138 mnt
->mnt
.mnt_group_id
= 0;
142 * vfsmount lock must be held for read
144 static inline void mnt_add_count(struct mount
*mnt
, int n
)
147 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
156 * vfsmount lock must be held for write
158 unsigned int mnt_get_count(struct mount
*mnt
)
161 unsigned int count
= 0;
164 for_each_possible_cpu(cpu
) {
165 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
170 return mnt
->mnt_count
;
174 static struct mount
*alloc_vfsmnt(const char *name
)
176 struct mount
*p
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
178 struct vfsmount
*mnt
= &p
->mnt
;
181 err
= mnt_alloc_id(p
);
186 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
187 if (!mnt
->mnt_devname
)
192 p
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
194 goto out_free_devname
;
196 this_cpu_add(p
->mnt_pcp
->mnt_count
, 1);
202 INIT_LIST_HEAD(&p
->mnt_hash
);
203 INIT_LIST_HEAD(&p
->mnt_child
);
204 INIT_LIST_HEAD(&p
->mnt_mounts
);
205 INIT_LIST_HEAD(&mnt
->mnt_list
);
206 INIT_LIST_HEAD(&mnt
->mnt_expire
);
207 INIT_LIST_HEAD(&mnt
->mnt_share
);
208 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
209 INIT_LIST_HEAD(&mnt
->mnt_slave
);
210 #ifdef CONFIG_FSNOTIFY
211 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
218 kfree(p
->mnt
.mnt_devname
);
223 kmem_cache_free(mnt_cache
, p
);
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
246 int __mnt_is_readonly(struct vfsmount
*mnt
)
248 if (mnt
->mnt_flags
& MNT_READONLY
)
250 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
254 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
256 static inline void mnt_inc_writers(struct mount
*mnt
)
259 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
265 static inline void mnt_dec_writers(struct mount
*mnt
)
268 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
274 static unsigned int mnt_get_writers(struct mount
*mnt
)
277 unsigned int count
= 0;
280 for_each_possible_cpu(cpu
) {
281 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
286 return mnt
->mnt_writers
;
291 * Most r/o checks on a fs are for operations that take
292 * discrete amounts of time, like a write() or unlink().
293 * We must keep track of when those operations start
294 * (for permission checks) and when they end, so that
295 * we can determine when writes are able to occur to
299 * mnt_want_write - get write access to a mount
300 * @m: the mount on which to take a write
302 * This tells the low-level filesystem that a write is
303 * about to be performed to it, and makes sure that
304 * writes are allowed before returning success. When
305 * the write operation is finished, mnt_drop_write()
306 * must be called. This is effectively a refcount.
308 int mnt_want_write(struct vfsmount
*m
)
310 struct mount
*mnt
= real_mount(m
);
314 mnt_inc_writers(mnt
);
316 * The store to mnt_inc_writers must be visible before we pass
317 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
318 * incremented count after it has set MNT_WRITE_HOLD.
321 while (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
324 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
325 * be set to match its requirements. So we must not load that until
326 * MNT_WRITE_HOLD is cleared.
329 if (__mnt_is_readonly(m
)) {
330 mnt_dec_writers(mnt
);
338 EXPORT_SYMBOL_GPL(mnt_want_write
);
341 * mnt_clone_write - get write access to a mount
342 * @mnt: the mount on which to take a write
344 * This is effectively like mnt_want_write, except
345 * it must only be used to take an extra write reference
346 * on a mountpoint that we already know has a write reference
347 * on it. This allows some optimisation.
349 * After finished, mnt_drop_write must be called as usual to
350 * drop the reference.
352 int mnt_clone_write(struct vfsmount
*mnt
)
354 /* superblock may be r/o */
355 if (__mnt_is_readonly(mnt
))
358 mnt_inc_writers(real_mount(mnt
));
362 EXPORT_SYMBOL_GPL(mnt_clone_write
);
365 * mnt_want_write_file - get write access to a file's mount
366 * @file: the file who's mount on which to take a write
368 * This is like mnt_want_write, but it takes a file and can
369 * do some optimisations if the file is open for write already
371 int mnt_want_write_file(struct file
*file
)
373 struct inode
*inode
= file
->f_dentry
->d_inode
;
374 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
375 return mnt_want_write(file
->f_path
.mnt
);
377 return mnt_clone_write(file
->f_path
.mnt
);
379 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
382 * mnt_drop_write - give up write access to a mount
383 * @mnt: the mount on which to give up write access
385 * Tells the low-level filesystem that we are done
386 * performing writes to it. Must be matched with
387 * mnt_want_write() call above.
389 void mnt_drop_write(struct vfsmount
*mnt
)
392 mnt_dec_writers(real_mount(mnt
));
395 EXPORT_SYMBOL_GPL(mnt_drop_write
);
397 void mnt_drop_write_file(struct file
*file
)
399 mnt_drop_write(file
->f_path
.mnt
);
401 EXPORT_SYMBOL(mnt_drop_write_file
);
403 static int mnt_make_readonly(struct mount
*mnt
)
407 br_write_lock(vfsmount_lock
);
408 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
410 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
411 * should be visible before we do.
416 * With writers on hold, if this value is zero, then there are
417 * definitely no active writers (although held writers may subsequently
418 * increment the count, they'll have to wait, and decrement it after
419 * seeing MNT_READONLY).
421 * It is OK to have counter incremented on one CPU and decremented on
422 * another: the sum will add up correctly. The danger would be when we
423 * sum up each counter, if we read a counter before it is incremented,
424 * but then read another CPU's count which it has been subsequently
425 * decremented from -- we would see more decrements than we should.
426 * MNT_WRITE_HOLD protects against this scenario, because
427 * mnt_want_write first increments count, then smp_mb, then spins on
428 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
429 * we're counting up here.
431 if (mnt_get_writers(mnt
) > 0)
434 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
436 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
437 * that become unheld will see MNT_READONLY.
440 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
441 br_write_unlock(vfsmount_lock
);
445 static void __mnt_unmake_readonly(struct mount
*mnt
)
447 br_write_lock(vfsmount_lock
);
448 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
449 br_write_unlock(vfsmount_lock
);
452 static void free_vfsmnt(struct mount
*mnt
)
454 kfree(mnt
->mnt
.mnt_devname
);
457 free_percpu(mnt
->mnt_pcp
);
459 kmem_cache_free(mnt_cache
, mnt
);
463 * find the first or last mount at @dentry on vfsmount @mnt depending on
464 * @dir. If @dir is set return the first mount else return the last mount.
465 * vfsmount_lock must be held for read or write.
467 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
470 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
471 struct list_head
*tmp
= head
;
472 struct mount
*p
, *found
= NULL
;
475 tmp
= dir
? tmp
->next
: tmp
->prev
;
479 p
= list_entry(tmp
, struct mount
, mnt_hash
);
480 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
) {
489 * lookup_mnt increments the ref count before returning
490 * the vfsmount struct.
492 struct vfsmount
*lookup_mnt(struct path
*path
)
494 struct mount
*child_mnt
;
496 br_read_lock(vfsmount_lock
);
497 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
499 mnt_add_count(child_mnt
, 1);
500 br_read_unlock(vfsmount_lock
);
501 return &child_mnt
->mnt
;
503 br_read_unlock(vfsmount_lock
);
508 static inline int check_mnt(struct vfsmount
*mnt
)
510 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
514 * vfsmount lock must be held for write
516 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
520 wake_up_interruptible(&ns
->poll
);
525 * vfsmount lock must be held for write
527 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
529 if (ns
&& ns
->event
!= event
) {
531 wake_up_interruptible(&ns
->poll
);
536 * Clear dentry's mounted state if it has no remaining mounts.
537 * vfsmount_lock must be held for write.
539 static void dentry_reset_mounted(struct dentry
*dentry
)
543 for (u
= 0; u
< HASH_SIZE
; u
++) {
546 list_for_each_entry(p
, &mount_hashtable
[u
], mnt_hash
) {
547 if (p
->mnt_mountpoint
== dentry
)
551 spin_lock(&dentry
->d_lock
);
552 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
553 spin_unlock(&dentry
->d_lock
);
557 * vfsmount lock must be held for write
559 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
561 old_path
->dentry
= mnt
->mnt_mountpoint
;
562 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
563 mnt
->mnt_parent
= mnt
;
564 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
565 list_del_init(&mnt
->mnt_child
);
566 list_del_init(&mnt
->mnt_hash
);
567 dentry_reset_mounted(old_path
->dentry
);
571 * vfsmount lock must be held for write
573 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
574 struct mount
*child_mnt
)
576 child_mnt
->mnt_parent
= real_mount(mntget(mnt
));
577 child_mnt
->mnt_mountpoint
= dget(dentry
);
578 spin_lock(&dentry
->d_lock
);
579 dentry
->d_flags
|= DCACHE_MOUNTED
;
580 spin_unlock(&dentry
->d_lock
);
584 * vfsmount lock must be held for write
586 static void attach_mnt(struct mount
*mnt
, struct path
*path
)
588 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
589 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
590 hash(path
->mnt
, path
->dentry
));
591 list_add_tail(&mnt
->mnt_child
, &real_mount(path
->mnt
)->mnt_mounts
);
594 static inline void __mnt_make_longterm(struct mount
*mnt
)
597 atomic_inc(&mnt
->mnt_longterm
);
601 /* needs vfsmount lock for write */
602 static inline void __mnt_make_shortterm(struct mount
*mnt
)
605 atomic_dec(&mnt
->mnt_longterm
);
610 * vfsmount lock must be held for write
612 static void commit_tree(struct mount
*mnt
)
614 struct mount
*parent
= mnt
->mnt_parent
;
617 struct mnt_namespace
*n
= parent
->mnt
.mnt_ns
;
619 BUG_ON(parent
== mnt
);
621 list_add_tail(&head
, &mnt
->mnt
.mnt_list
);
622 list_for_each_entry(m
, &head
, mnt
.mnt_list
) {
624 __mnt_make_longterm(m
);
627 list_splice(&head
, n
->list
.prev
);
629 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
630 hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
631 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
632 touch_mnt_namespace(n
);
635 static struct mount
*next_mnt(struct mount
*p
, struct vfsmount
*root
)
637 struct list_head
*next
= p
->mnt_mounts
.next
;
638 if (next
== &p
->mnt_mounts
) {
642 next
= p
->mnt_child
.next
;
643 if (next
!= &p
->mnt_parent
->mnt_mounts
)
648 return list_entry(next
, struct mount
, mnt_child
);
651 static struct mount
*skip_mnt_tree(struct mount
*p
)
653 struct list_head
*prev
= p
->mnt_mounts
.prev
;
654 while (prev
!= &p
->mnt_mounts
) {
655 p
= list_entry(prev
, struct mount
, mnt_child
);
656 prev
= p
->mnt_mounts
.prev
;
662 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
668 return ERR_PTR(-ENODEV
);
670 mnt
= alloc_vfsmnt(name
);
672 return ERR_PTR(-ENOMEM
);
674 if (flags
& MS_KERNMOUNT
)
675 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
677 root
= mount_fs(type
, flags
, name
, data
);
680 return ERR_CAST(root
);
683 mnt
->mnt
.mnt_root
= root
;
684 mnt
->mnt
.mnt_sb
= root
->d_sb
;
685 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
686 mnt
->mnt_parent
= mnt
;
689 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
691 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
694 struct super_block
*sb
= old
->mnt
.mnt_sb
;
695 struct mount
*mnt
= alloc_vfsmnt(old
->mnt
.mnt_devname
);
698 if (flag
& (CL_SLAVE
| CL_PRIVATE
))
699 mnt
->mnt
.mnt_group_id
= 0; /* not a peer of original */
701 mnt
->mnt
.mnt_group_id
= old
->mnt
.mnt_group_id
;
703 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt
.mnt_group_id
) {
704 int err
= mnt_alloc_group_id(mnt
);
709 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
710 atomic_inc(&sb
->s_active
);
711 mnt
->mnt
.mnt_sb
= sb
;
712 mnt
->mnt
.mnt_root
= dget(root
);
713 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
714 mnt
->mnt_parent
= mnt
;
716 if (flag
& CL_SLAVE
) {
717 list_add(&mnt
->mnt
.mnt_slave
, &old
->mnt
.mnt_slave_list
);
718 mnt
->mnt
.mnt_master
= &old
->mnt
;
719 CLEAR_MNT_SHARED(&mnt
->mnt
);
720 } else if (!(flag
& CL_PRIVATE
)) {
721 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(&old
->mnt
))
722 list_add(&mnt
->mnt
.mnt_share
, &old
->mnt
.mnt_share
);
723 if (IS_MNT_SLAVE(&old
->mnt
))
724 list_add(&mnt
->mnt
.mnt_slave
, &old
->mnt
.mnt_slave
);
725 mnt
->mnt
.mnt_master
= old
->mnt
.mnt_master
;
727 if (flag
& CL_MAKE_SHARED
)
730 /* stick the duplicate mount on the same expiry list
731 * as the original if that was on one */
732 if (flag
& CL_EXPIRE
) {
733 if (!list_empty(&old
->mnt
.mnt_expire
))
734 list_add(&mnt
->mnt
.mnt_expire
, &old
->mnt
.mnt_expire
);
744 static inline void mntfree(struct mount
*mnt
)
746 struct vfsmount
*m
= &mnt
->mnt
;
747 struct super_block
*sb
= m
->mnt_sb
;
750 * This probably indicates that somebody messed
751 * up a mnt_want/drop_write() pair. If this
752 * happens, the filesystem was probably unable
753 * to make r/w->r/o transitions.
756 * The locking used to deal with mnt_count decrement provides barriers,
757 * so mnt_get_writers() below is safe.
759 WARN_ON(mnt_get_writers(mnt
));
760 fsnotify_vfsmount_delete(m
);
763 deactivate_super(sb
);
766 static void mntput_no_expire(struct vfsmount
*m
)
768 struct mount
*mnt
= real_mount(m
);
771 br_read_lock(vfsmount_lock
);
772 if (likely(atomic_read(&mnt
->mnt_longterm
))) {
773 mnt_add_count(mnt
, -1);
774 br_read_unlock(vfsmount_lock
);
777 br_read_unlock(vfsmount_lock
);
779 br_write_lock(vfsmount_lock
);
780 mnt_add_count(mnt
, -1);
781 if (mnt_get_count(mnt
)) {
782 br_write_unlock(vfsmount_lock
);
786 mnt_add_count(mnt
, -1);
787 if (likely(mnt_get_count(mnt
)))
789 br_write_lock(vfsmount_lock
);
791 if (unlikely(mnt
->mnt
.mnt_pinned
)) {
792 mnt_add_count(mnt
, mnt
->mnt
.mnt_pinned
+ 1);
793 mnt
->mnt
.mnt_pinned
= 0;
794 br_write_unlock(vfsmount_lock
);
795 acct_auto_close_mnt(m
);
798 br_write_unlock(vfsmount_lock
);
802 void mntput(struct vfsmount
*mnt
)
805 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
806 if (unlikely(mnt
->mnt_expiry_mark
))
807 mnt
->mnt_expiry_mark
= 0;
808 mntput_no_expire(mnt
);
811 EXPORT_SYMBOL(mntput
);
813 struct vfsmount
*mntget(struct vfsmount
*mnt
)
816 mnt_add_count(real_mount(mnt
), 1);
819 EXPORT_SYMBOL(mntget
);
821 void mnt_pin(struct vfsmount
*mnt
)
823 br_write_lock(vfsmount_lock
);
825 br_write_unlock(vfsmount_lock
);
827 EXPORT_SYMBOL(mnt_pin
);
829 void mnt_unpin(struct vfsmount
*mnt
)
831 br_write_lock(vfsmount_lock
);
832 if (mnt
->mnt_pinned
) {
833 mnt_add_count(real_mount(mnt
), 1);
836 br_write_unlock(vfsmount_lock
);
838 EXPORT_SYMBOL(mnt_unpin
);
840 static inline void mangle(struct seq_file
*m
, const char *s
)
842 seq_escape(m
, s
, " \t\n\\");
846 * Simple .show_options callback for filesystems which don't want to
847 * implement more complex mount option showing.
849 * See also save_mount_options().
851 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
856 options
= rcu_dereference(mnt
->mnt_sb
->s_options
);
858 if (options
!= NULL
&& options
[0]) {
866 EXPORT_SYMBOL(generic_show_options
);
869 * If filesystem uses generic_show_options(), this function should be
870 * called from the fill_super() callback.
872 * The .remount_fs callback usually needs to be handled in a special
873 * way, to make sure, that previous options are not overwritten if the
876 * Also note, that if the filesystem's .remount_fs function doesn't
877 * reset all options to their default value, but changes only newly
878 * given options, then the displayed options will not reflect reality
881 void save_mount_options(struct super_block
*sb
, char *options
)
883 BUG_ON(sb
->s_options
);
884 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
886 EXPORT_SYMBOL(save_mount_options
);
888 void replace_mount_options(struct super_block
*sb
, char *options
)
890 char *old
= sb
->s_options
;
891 rcu_assign_pointer(sb
->s_options
, options
);
897 EXPORT_SYMBOL(replace_mount_options
);
899 #ifdef CONFIG_PROC_FS
901 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
903 struct proc_mounts
*p
= m
->private;
905 down_read(&namespace_sem
);
906 return seq_list_start(&p
->ns
->list
, *pos
);
909 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
911 struct proc_mounts
*p
= m
->private;
913 return seq_list_next(v
, &p
->ns
->list
, pos
);
916 static void m_stop(struct seq_file
*m
, void *v
)
918 up_read(&namespace_sem
);
921 int mnt_had_events(struct proc_mounts
*p
)
923 struct mnt_namespace
*ns
= p
->ns
;
926 br_read_lock(vfsmount_lock
);
927 if (p
->m
.poll_event
!= ns
->event
) {
928 p
->m
.poll_event
= ns
->event
;
931 br_read_unlock(vfsmount_lock
);
936 struct proc_fs_info
{
941 static int show_sb_opts(struct seq_file
*m
, struct super_block
*sb
)
943 static const struct proc_fs_info fs_info
[] = {
944 { MS_SYNCHRONOUS
, ",sync" },
945 { MS_DIRSYNC
, ",dirsync" },
946 { MS_MANDLOCK
, ",mand" },
949 const struct proc_fs_info
*fs_infop
;
951 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
952 if (sb
->s_flags
& fs_infop
->flag
)
953 seq_puts(m
, fs_infop
->str
);
956 return security_sb_show_options(m
, sb
);
959 static void show_mnt_opts(struct seq_file
*m
, struct vfsmount
*mnt
)
961 static const struct proc_fs_info mnt_info
[] = {
962 { MNT_NOSUID
, ",nosuid" },
963 { MNT_NODEV
, ",nodev" },
964 { MNT_NOEXEC
, ",noexec" },
965 { MNT_NOATIME
, ",noatime" },
966 { MNT_NODIRATIME
, ",nodiratime" },
967 { MNT_RELATIME
, ",relatime" },
970 const struct proc_fs_info
*fs_infop
;
972 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
973 if (mnt
->mnt_flags
& fs_infop
->flag
)
974 seq_puts(m
, fs_infop
->str
);
978 static void show_type(struct seq_file
*m
, struct super_block
*sb
)
980 mangle(m
, sb
->s_type
->name
);
981 if (sb
->s_subtype
&& sb
->s_subtype
[0]) {
983 mangle(m
, sb
->s_subtype
);
987 static int show_vfsmnt(struct seq_file
*m
, void *v
)
989 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
991 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
993 if (mnt
->mnt_sb
->s_op
->show_devname
) {
994 err
= mnt
->mnt_sb
->s_op
->show_devname(m
, mnt
);
998 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
1001 seq_path(m
, &mnt_path
, " \t\n\\");
1003 show_type(m
, mnt
->mnt_sb
);
1004 seq_puts(m
, __mnt_is_readonly(mnt
) ? " ro" : " rw");
1005 err
= show_sb_opts(m
, mnt
->mnt_sb
);
1008 show_mnt_opts(m
, mnt
);
1009 if (mnt
->mnt_sb
->s_op
->show_options
)
1010 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
1011 seq_puts(m
, " 0 0\n");
1016 const struct seq_operations mounts_op
= {
1023 static int show_mountinfo(struct seq_file
*m
, void *v
)
1025 struct proc_mounts
*p
= m
->private;
1026 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
1027 struct mount
*r
= real_mount(mnt
);
1028 struct super_block
*sb
= mnt
->mnt_sb
;
1029 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
1030 struct path root
= p
->root
;
1033 seq_printf(m
, "%i %i %u:%u ", mnt
->mnt_id
, r
->mnt_parent
->mnt
.mnt_id
,
1034 MAJOR(sb
->s_dev
), MINOR(sb
->s_dev
));
1035 if (sb
->s_op
->show_path
)
1036 err
= sb
->s_op
->show_path(m
, mnt
);
1038 seq_dentry(m
, mnt
->mnt_root
, " \t\n\\");
1043 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1044 err
= seq_path_root(m
, &mnt_path
, &root
, " \t\n\\");
1048 seq_puts(m
, mnt
->mnt_flags
& MNT_READONLY
? " ro" : " rw");
1049 show_mnt_opts(m
, mnt
);
1051 /* Tagged fields ("foo:X" or "bar") */
1052 if (IS_MNT_SHARED(mnt
))
1053 seq_printf(m
, " shared:%i", mnt
->mnt_group_id
);
1054 if (IS_MNT_SLAVE(mnt
)) {
1055 int master
= mnt
->mnt_master
->mnt_group_id
;
1056 int dom
= get_dominating_id(mnt
, &p
->root
);
1057 seq_printf(m
, " master:%i", master
);
1058 if (dom
&& dom
!= master
)
1059 seq_printf(m
, " propagate_from:%i", dom
);
1061 if (IS_MNT_UNBINDABLE(mnt
))
1062 seq_puts(m
, " unbindable");
1064 /* Filesystem specific data */
1068 if (sb
->s_op
->show_devname
)
1069 err
= sb
->s_op
->show_devname(m
, mnt
);
1071 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
1074 seq_puts(m
, sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
1075 err
= show_sb_opts(m
, sb
);
1078 if (sb
->s_op
->show_options
)
1079 err
= sb
->s_op
->show_options(m
, mnt
);
1085 const struct seq_operations mountinfo_op
= {
1089 .show
= show_mountinfo
,
1092 static int show_vfsstat(struct seq_file
*m
, void *v
)
1094 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
1095 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
1099 if (mnt
->mnt_sb
->s_op
->show_devname
) {
1100 seq_puts(m
, "device ");
1101 err
= mnt
->mnt_sb
->s_op
->show_devname(m
, mnt
);
1103 if (mnt
->mnt_devname
) {
1104 seq_puts(m
, "device ");
1105 mangle(m
, mnt
->mnt_devname
);
1107 seq_puts(m
, "no device");
1111 seq_puts(m
, " mounted on ");
1112 seq_path(m
, &mnt_path
, " \t\n\\");
1115 /* file system type */
1116 seq_puts(m
, "with fstype ");
1117 show_type(m
, mnt
->mnt_sb
);
1119 /* optional statistics */
1120 if (mnt
->mnt_sb
->s_op
->show_stats
) {
1123 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
1130 const struct seq_operations mountstats_op
= {
1134 .show
= show_vfsstat
,
1136 #endif /* CONFIG_PROC_FS */
1139 * may_umount_tree - check if a mount tree is busy
1140 * @mnt: root of mount tree
1142 * This is called to check if a tree of mounts has any
1143 * open files, pwds, chroots or sub mounts that are
1146 int may_umount_tree(struct vfsmount
*mnt
)
1148 int actual_refs
= 0;
1149 int minimum_refs
= 0;
1153 /* write lock needed for mnt_get_count */
1154 br_write_lock(vfsmount_lock
);
1155 for (p
= real_mount(mnt
); p
; p
= next_mnt(p
, mnt
)) {
1156 actual_refs
+= mnt_get_count(p
);
1159 br_write_unlock(vfsmount_lock
);
1161 if (actual_refs
> minimum_refs
)
1167 EXPORT_SYMBOL(may_umount_tree
);
1170 * may_umount - check if a mount point is busy
1171 * @mnt: root of mount
1173 * This is called to check if a mount point has any
1174 * open files, pwds, chroots or sub mounts. If the
1175 * mount has sub mounts this will return busy
1176 * regardless of whether the sub mounts are busy.
1178 * Doesn't take quota and stuff into account. IOW, in some cases it will
1179 * give false negatives. The main reason why it's here is that we need
1180 * a non-destructive way to look for easily umountable filesystems.
1182 int may_umount(struct vfsmount
*mnt
)
1185 down_read(&namespace_sem
);
1186 br_write_lock(vfsmount_lock
);
1187 if (propagate_mount_busy(real_mount(mnt
), 2))
1189 br_write_unlock(vfsmount_lock
);
1190 up_read(&namespace_sem
);
1194 EXPORT_SYMBOL(may_umount
);
1196 void release_mounts(struct list_head
*head
)
1199 while (!list_empty(head
)) {
1200 mnt
= list_first_entry(head
, struct mount
, mnt_hash
);
1201 list_del_init(&mnt
->mnt_hash
);
1202 if (mnt_has_parent(mnt
)) {
1203 struct dentry
*dentry
;
1206 br_write_lock(vfsmount_lock
);
1207 dentry
= mnt
->mnt_mountpoint
;
1208 m
= &mnt
->mnt_parent
->mnt
;
1209 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1210 mnt
->mnt_parent
= mnt
;
1212 br_write_unlock(vfsmount_lock
);
1221 * vfsmount lock must be held for write
1222 * namespace_sem must be held for write
1224 void umount_tree(struct mount
*mnt
, int propagate
, struct list_head
*kill
)
1226 LIST_HEAD(tmp_list
);
1229 for (p
= mnt
; p
; p
= next_mnt(p
, &mnt
->mnt
))
1230 list_move(&p
->mnt_hash
, &tmp_list
);
1233 propagate_umount(&tmp_list
);
1235 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1236 list_del_init(&p
->mnt
.mnt_expire
);
1237 list_del_init(&p
->mnt
.mnt_list
);
1238 __touch_mnt_namespace(p
->mnt
.mnt_ns
);
1239 p
->mnt
.mnt_ns
= NULL
;
1240 __mnt_make_shortterm(p
);
1241 list_del_init(&p
->mnt_child
);
1242 if (mnt_has_parent(p
)) {
1243 p
->mnt_parent
->mnt
.mnt_ghosts
++;
1244 dentry_reset_mounted(p
->mnt_mountpoint
);
1246 change_mnt_propagation(p
, MS_PRIVATE
);
1248 list_splice(&tmp_list
, kill
);
1251 static void shrink_submounts(struct mount
*mnt
, struct list_head
*umounts
);
1253 static int do_umount(struct mount
*mnt
, int flags
)
1255 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1257 LIST_HEAD(umount_list
);
1259 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1264 * Allow userspace to request a mountpoint be expired rather than
1265 * unmounting unconditionally. Unmount only happens if:
1266 * (1) the mark is already set (the mark is cleared by mntput())
1267 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1269 if (flags
& MNT_EXPIRE
) {
1270 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1271 flags
& (MNT_FORCE
| MNT_DETACH
))
1275 * probably don't strictly need the lock here if we examined
1276 * all race cases, but it's a slowpath.
1278 br_write_lock(vfsmount_lock
);
1279 if (mnt_get_count(mnt
) != 2) {
1280 br_write_unlock(vfsmount_lock
);
1283 br_write_unlock(vfsmount_lock
);
1285 if (!xchg(&mnt
->mnt
.mnt_expiry_mark
, 1))
1290 * If we may have to abort operations to get out of this
1291 * mount, and they will themselves hold resources we must
1292 * allow the fs to do things. In the Unix tradition of
1293 * 'Gee thats tricky lets do it in userspace' the umount_begin
1294 * might fail to complete on the first run through as other tasks
1295 * must return, and the like. Thats for the mount program to worry
1296 * about for the moment.
1299 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1300 sb
->s_op
->umount_begin(sb
);
1304 * No sense to grab the lock for this test, but test itself looks
1305 * somewhat bogus. Suggestions for better replacement?
1306 * Ho-hum... In principle, we might treat that as umount + switch
1307 * to rootfs. GC would eventually take care of the old vfsmount.
1308 * Actually it makes sense, especially if rootfs would contain a
1309 * /reboot - static binary that would close all descriptors and
1310 * call reboot(9). Then init(8) could umount root and exec /reboot.
1312 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1314 * Special case for "unmounting" root ...
1315 * we just try to remount it readonly.
1317 down_write(&sb
->s_umount
);
1318 if (!(sb
->s_flags
& MS_RDONLY
))
1319 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1320 up_write(&sb
->s_umount
);
1324 down_write(&namespace_sem
);
1325 br_write_lock(vfsmount_lock
);
1328 if (!(flags
& MNT_DETACH
))
1329 shrink_submounts(mnt
, &umount_list
);
1332 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1333 if (!list_empty(&mnt
->mnt
.mnt_list
))
1334 umount_tree(mnt
, 1, &umount_list
);
1337 br_write_unlock(vfsmount_lock
);
1338 up_write(&namespace_sem
);
1339 release_mounts(&umount_list
);
1344 * Now umount can handle mount points as well as block devices.
1345 * This is important for filesystems which use unnamed block devices.
1347 * We now support a flag for forced unmount like the other 'big iron'
1348 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1351 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1355 int lookup_flags
= 0;
1357 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1360 if (!(flags
& UMOUNT_NOFOLLOW
))
1361 lookup_flags
|= LOOKUP_FOLLOW
;
1363 retval
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1367 if (path
.dentry
!= path
.mnt
->mnt_root
)
1369 if (!check_mnt(path
.mnt
))
1373 if (!capable(CAP_SYS_ADMIN
))
1376 retval
= do_umount(real_mount(path
.mnt
), flags
);
1378 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1380 mntput_no_expire(path
.mnt
);
1385 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1388 * The 2.0 compatible umount. No flags.
1390 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1392 return sys_umount(name
, 0);
1397 static int mount_is_safe(struct path
*path
)
1399 if (capable(CAP_SYS_ADMIN
))
1403 if (S_ISLNK(path
->dentry
->d_inode
->i_mode
))
1405 if (path
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
1406 if (current_uid() != path
->dentry
->d_inode
->i_uid
)
1409 if (inode_permission(path
->dentry
->d_inode
, MAY_WRITE
))
1415 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1418 struct mount
*res
, *p
, *q
, *r
;
1421 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(&mnt
->mnt
))
1424 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1427 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1430 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1432 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1435 for (s
= r
; s
; s
= next_mnt(s
, &r
->mnt
)) {
1436 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(&s
->mnt
)) {
1437 s
= skip_mnt_tree(s
);
1440 while (p
!= s
->mnt_parent
) {
1446 path
.dentry
= p
->mnt_mountpoint
;
1447 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1450 br_write_lock(vfsmount_lock
);
1451 list_add_tail(&q
->mnt
.mnt_list
, &res
->mnt
.mnt_list
);
1452 attach_mnt(q
, &path
);
1453 br_write_unlock(vfsmount_lock
);
1459 LIST_HEAD(umount_list
);
1460 br_write_lock(vfsmount_lock
);
1461 umount_tree(res
, 0, &umount_list
);
1462 br_write_unlock(vfsmount_lock
);
1463 release_mounts(&umount_list
);
1468 struct vfsmount
*collect_mounts(struct path
*path
)
1471 down_write(&namespace_sem
);
1472 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1473 CL_COPY_ALL
| CL_PRIVATE
);
1474 up_write(&namespace_sem
);
1475 return tree
? &tree
->mnt
: NULL
;
1478 void drop_collected_mounts(struct vfsmount
*mnt
)
1480 LIST_HEAD(umount_list
);
1481 down_write(&namespace_sem
);
1482 br_write_lock(vfsmount_lock
);
1483 umount_tree(real_mount(mnt
), 0, &umount_list
);
1484 br_write_unlock(vfsmount_lock
);
1485 up_write(&namespace_sem
);
1486 release_mounts(&umount_list
);
1489 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1490 struct vfsmount
*root
)
1492 struct vfsmount
*mnt
;
1493 int res
= f(root
, arg
);
1496 list_for_each_entry(mnt
, &root
->mnt_list
, mnt_list
) {
1504 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1508 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, &mnt
->mnt
)) {
1509 if (p
->mnt
.mnt_group_id
&& !IS_MNT_SHARED(&p
->mnt
))
1510 mnt_release_group_id(p
);
1514 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1518 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, &mnt
->mnt
) : NULL
) {
1519 if (!p
->mnt
.mnt_group_id
&& !IS_MNT_SHARED(&p
->mnt
)) {
1520 int err
= mnt_alloc_group_id(p
);
1522 cleanup_group_ids(mnt
, p
);
1532 * @source_mnt : mount tree to be attached
1533 * @nd : place the mount tree @source_mnt is attached
1534 * @parent_nd : if non-null, detach the source_mnt from its parent and
1535 * store the parent mount and mountpoint dentry.
1536 * (done when source_mnt is moved)
1538 * NOTE: in the table below explains the semantics when a source mount
1539 * of a given type is attached to a destination mount of a given type.
1540 * ---------------------------------------------------------------------------
1541 * | BIND MOUNT OPERATION |
1542 * |**************************************************************************
1543 * | source-->| shared | private | slave | unbindable |
1547 * |**************************************************************************
1548 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1550 * |non-shared| shared (+) | private | slave (*) | invalid |
1551 * ***************************************************************************
1552 * A bind operation clones the source mount and mounts the clone on the
1553 * destination mount.
1555 * (++) the cloned mount is propagated to all the mounts in the propagation
1556 * tree of the destination mount and the cloned mount is added to
1557 * the peer group of the source mount.
1558 * (+) the cloned mount is created under the destination mount and is marked
1559 * as shared. The cloned mount is added to the peer group of the source
1561 * (+++) the mount is propagated to all the mounts in the propagation tree
1562 * of the destination mount and the cloned mount is made slave
1563 * of the same master as that of the source mount. The cloned mount
1564 * is marked as 'shared and slave'.
1565 * (*) the cloned mount is made a slave of the same master as that of the
1568 * ---------------------------------------------------------------------------
1569 * | MOVE MOUNT OPERATION |
1570 * |**************************************************************************
1571 * | source-->| shared | private | slave | unbindable |
1575 * |**************************************************************************
1576 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1578 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1579 * ***************************************************************************
1581 * (+) the mount is moved to the destination. And is then propagated to
1582 * all the mounts in the propagation tree of the destination mount.
1583 * (+*) the mount is moved to the destination.
1584 * (+++) the mount is moved to the destination and is then propagated to
1585 * all the mounts belonging to the destination mount's propagation tree.
1586 * the mount is marked as 'shared and slave'.
1587 * (*) the mount continues to be a slave at the new location.
1589 * if the source mount is a tree, the operations explained above is
1590 * applied to each mount in the tree.
1591 * Must be called without spinlocks held, since this function can sleep
1594 static int attach_recursive_mnt(struct mount
*source_mnt
,
1595 struct path
*path
, struct path
*parent_path
)
1597 LIST_HEAD(tree_list
);
1598 struct vfsmount
*dest_mnt
= path
->mnt
;
1599 struct dentry
*dest_dentry
= path
->dentry
;
1600 struct mount
*child
, *p
;
1603 if (IS_MNT_SHARED(dest_mnt
)) {
1604 err
= invent_group_ids(source_mnt
, true);
1608 err
= propagate_mnt(dest_mnt
, dest_dentry
, &source_mnt
->mnt
, &tree_list
);
1610 goto out_cleanup_ids
;
1612 br_write_lock(vfsmount_lock
);
1614 if (IS_MNT_SHARED(dest_mnt
)) {
1615 for (p
= source_mnt
; p
; p
= next_mnt(p
, &source_mnt
->mnt
))
1619 detach_mnt(source_mnt
, parent_path
);
1620 attach_mnt(source_mnt
, path
);
1621 touch_mnt_namespace(parent_path
->mnt
->mnt_ns
);
1623 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
1624 commit_tree(source_mnt
);
1627 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1628 list_del_init(&child
->mnt_hash
);
1631 br_write_unlock(vfsmount_lock
);
1636 if (IS_MNT_SHARED(dest_mnt
))
1637 cleanup_group_ids(source_mnt
, NULL
);
1642 static int lock_mount(struct path
*path
)
1644 struct vfsmount
*mnt
;
1646 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1647 if (unlikely(cant_mount(path
->dentry
))) {
1648 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1651 down_write(&namespace_sem
);
1652 mnt
= lookup_mnt(path
);
1655 up_write(&namespace_sem
);
1656 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1659 path
->dentry
= dget(mnt
->mnt_root
);
1663 static void unlock_mount(struct path
*path
)
1665 up_write(&namespace_sem
);
1666 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1669 static int graft_tree(struct vfsmount
*mnt
, struct path
*path
)
1671 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
1674 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1675 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
1678 if (d_unlinked(path
->dentry
))
1681 return attach_recursive_mnt(real_mount(mnt
), path
, NULL
);
1685 * Sanity check the flags to change_mnt_propagation.
1688 static int flags_to_propagation_type(int flags
)
1690 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1692 /* Fail if any non-propagation flags are set */
1693 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1695 /* Only one propagation flag should be set */
1696 if (!is_power_of_2(type
))
1702 * recursively change the type of the mountpoint.
1704 static int do_change_type(struct path
*path
, int flag
)
1707 struct mount
*mnt
= real_mount(path
->mnt
);
1708 int recurse
= flag
& MS_REC
;
1712 if (!capable(CAP_SYS_ADMIN
))
1715 if (path
->dentry
!= path
->mnt
->mnt_root
)
1718 type
= flags_to_propagation_type(flag
);
1722 down_write(&namespace_sem
);
1723 if (type
== MS_SHARED
) {
1724 err
= invent_group_ids(mnt
, recurse
);
1729 br_write_lock(vfsmount_lock
);
1730 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, &mnt
->mnt
) : NULL
))
1731 change_mnt_propagation(m
, type
);
1732 br_write_unlock(vfsmount_lock
);
1735 up_write(&namespace_sem
);
1740 * do loopback mount.
1742 static int do_loopback(struct path
*path
, char *old_name
,
1745 LIST_HEAD(umount_list
);
1746 struct path old_path
;
1747 struct mount
*mnt
= NULL
, *old
;
1748 int err
= mount_is_safe(path
);
1751 if (!old_name
|| !*old_name
)
1753 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1757 err
= lock_mount(path
);
1761 old
= real_mount(old_path
.mnt
);
1764 if (IS_MNT_UNBINDABLE(old_path
.mnt
))
1767 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1772 mnt
= copy_tree(old
, old_path
.dentry
, 0);
1774 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1779 err
= graft_tree(&mnt
->mnt
, path
);
1781 br_write_lock(vfsmount_lock
);
1782 umount_tree(mnt
, 0, &umount_list
);
1783 br_write_unlock(vfsmount_lock
);
1787 release_mounts(&umount_list
);
1789 path_put(&old_path
);
1793 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1796 int readonly_request
= 0;
1798 if (ms_flags
& MS_RDONLY
)
1799 readonly_request
= 1;
1800 if (readonly_request
== __mnt_is_readonly(mnt
))
1803 if (readonly_request
)
1804 error
= mnt_make_readonly(real_mount(mnt
));
1806 __mnt_unmake_readonly(real_mount(mnt
));
1811 * change filesystem flags. dir should be a physical root of filesystem.
1812 * If you've mounted a non-root directory somewhere and want to do remount
1813 * on it - tough luck.
1815 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1819 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1821 if (!capable(CAP_SYS_ADMIN
))
1824 if (!check_mnt(path
->mnt
))
1827 if (path
->dentry
!= path
->mnt
->mnt_root
)
1830 err
= security_sb_remount(sb
, data
);
1834 down_write(&sb
->s_umount
);
1835 if (flags
& MS_BIND
)
1836 err
= change_mount_flags(path
->mnt
, flags
);
1838 err
= do_remount_sb(sb
, flags
, data
, 0);
1840 br_write_lock(vfsmount_lock
);
1841 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_PROPAGATION_MASK
;
1842 path
->mnt
->mnt_flags
= mnt_flags
;
1843 br_write_unlock(vfsmount_lock
);
1845 up_write(&sb
->s_umount
);
1847 br_write_lock(vfsmount_lock
);
1848 touch_mnt_namespace(path
->mnt
->mnt_ns
);
1849 br_write_unlock(vfsmount_lock
);
1854 static inline int tree_contains_unbindable(struct mount
*mnt
)
1857 for (p
= mnt
; p
; p
= next_mnt(p
, &mnt
->mnt
)) {
1858 if (IS_MNT_UNBINDABLE(&p
->mnt
))
1864 static int do_move_mount(struct path
*path
, char *old_name
)
1866 struct path old_path
, parent_path
;
1870 if (!capable(CAP_SYS_ADMIN
))
1872 if (!old_name
|| !*old_name
)
1874 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1878 err
= lock_mount(path
);
1883 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1886 if (d_unlinked(path
->dentry
))
1890 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1893 old
= real_mount(old_path
.mnt
);
1895 if (!mnt_has_parent(old
))
1898 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1899 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1902 * Don't move a mount residing in a shared parent.
1904 if (IS_MNT_SHARED(&old
->mnt_parent
->mnt
))
1907 * Don't move a mount tree containing unbindable mounts to a destination
1908 * mount which is shared.
1910 if (IS_MNT_SHARED(path
->mnt
) &&
1911 tree_contains_unbindable(old
))
1914 for (p
= real_mount(path
->mnt
); mnt_has_parent(p
); p
= p
->mnt_parent
)
1918 err
= attach_recursive_mnt(old
, path
, &parent_path
);
1922 /* if the mount is moved, it should no longer be expire
1924 list_del_init(&old_path
.mnt
->mnt_expire
);
1929 path_put(&parent_path
);
1930 path_put(&old_path
);
1934 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
1937 const char *subtype
= strchr(fstype
, '.');
1946 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
1948 if (!mnt
->mnt_sb
->s_subtype
)
1954 return ERR_PTR(err
);
1957 static struct vfsmount
*
1958 do_kern_mount(const char *fstype
, int flags
, const char *name
, void *data
)
1960 struct file_system_type
*type
= get_fs_type(fstype
);
1961 struct vfsmount
*mnt
;
1963 return ERR_PTR(-ENODEV
);
1964 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
1965 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
1966 !mnt
->mnt_sb
->s_subtype
)
1967 mnt
= fs_set_subtype(mnt
, fstype
);
1968 put_filesystem(type
);
1973 * add a mount into a namespace's mount tree
1975 static int do_add_mount(struct vfsmount
*newmnt
, struct path
*path
, int mnt_flags
)
1979 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
1981 err
= lock_mount(path
);
1986 if (!(mnt_flags
& MNT_SHRINKABLE
) && !check_mnt(path
->mnt
))
1989 /* Refuse the same filesystem on the same mount point */
1991 if (path
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1992 path
->mnt
->mnt_root
== path
->dentry
)
1996 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1999 newmnt
->mnt_flags
= mnt_flags
;
2000 err
= graft_tree(newmnt
, path
);
2008 * create a new mount for userspace and request it to be added into the
2011 static int do_new_mount(struct path
*path
, char *type
, int flags
,
2012 int mnt_flags
, char *name
, void *data
)
2014 struct vfsmount
*mnt
;
2020 /* we need capabilities... */
2021 if (!capable(CAP_SYS_ADMIN
))
2024 mnt
= do_kern_mount(type
, flags
, name
, data
);
2026 return PTR_ERR(mnt
);
2028 err
= do_add_mount(mnt
, path
, mnt_flags
);
2034 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2037 /* The new mount record should have at least 2 refs to prevent it being
2038 * expired before we get a chance to add it
2040 BUG_ON(mnt_get_count(real_mount(m
)) < 2);
2042 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2043 m
->mnt_root
== path
->dentry
) {
2048 err
= do_add_mount(m
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2052 /* remove m from any expiration list it may be on */
2053 if (!list_empty(&m
->mnt_expire
)) {
2054 down_write(&namespace_sem
);
2055 br_write_lock(vfsmount_lock
);
2056 list_del_init(&m
->mnt_expire
);
2057 br_write_unlock(vfsmount_lock
);
2058 up_write(&namespace_sem
);
2066 * mnt_set_expiry - Put a mount on an expiration list
2067 * @mnt: The mount to list.
2068 * @expiry_list: The list to add the mount to.
2070 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2072 down_write(&namespace_sem
);
2073 br_write_lock(vfsmount_lock
);
2075 list_add_tail(&mnt
->mnt_expire
, expiry_list
);
2077 br_write_unlock(vfsmount_lock
);
2078 up_write(&namespace_sem
);
2080 EXPORT_SYMBOL(mnt_set_expiry
);
2083 * process a list of expirable mountpoints with the intent of discarding any
2084 * mountpoints that aren't in use and haven't been touched since last we came
2087 void mark_mounts_for_expiry(struct list_head
*mounts
)
2089 struct mount
*mnt
, *next
;
2090 LIST_HEAD(graveyard
);
2093 if (list_empty(mounts
))
2096 down_write(&namespace_sem
);
2097 br_write_lock(vfsmount_lock
);
2099 /* extract from the expiration list every vfsmount that matches the
2100 * following criteria:
2101 * - only referenced by its parent vfsmount
2102 * - still marked for expiry (marked on the last call here; marks are
2103 * cleared by mntput())
2105 list_for_each_entry_safe(mnt
, next
, mounts
, mnt
.mnt_expire
) {
2106 if (!xchg(&mnt
->mnt
.mnt_expiry_mark
, 1) ||
2107 propagate_mount_busy(mnt
, 1))
2109 list_move(&mnt
->mnt
.mnt_expire
, &graveyard
);
2111 while (!list_empty(&graveyard
)) {
2112 mnt
= list_first_entry(&graveyard
, struct mount
, mnt
.mnt_expire
);
2113 touch_mnt_namespace(mnt
->mnt
.mnt_ns
);
2114 umount_tree(mnt
, 1, &umounts
);
2116 br_write_unlock(vfsmount_lock
);
2117 up_write(&namespace_sem
);
2119 release_mounts(&umounts
);
2122 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2125 * Ripoff of 'select_parent()'
2127 * search the list of submounts for a given mountpoint, and move any
2128 * shrinkable submounts to the 'graveyard' list.
2130 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2132 struct mount
*this_parent
= parent
;
2133 struct list_head
*next
;
2137 next
= this_parent
->mnt_mounts
.next
;
2139 while (next
!= &this_parent
->mnt_mounts
) {
2140 struct list_head
*tmp
= next
;
2141 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2144 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2147 * Descend a level if the d_mounts list is non-empty.
2149 if (!list_empty(&mnt
->mnt_mounts
)) {
2154 if (!propagate_mount_busy(mnt
, 1)) {
2155 list_move_tail(&mnt
->mnt
.mnt_expire
, graveyard
);
2160 * All done at this level ... ascend and resume the search
2162 if (this_parent
!= parent
) {
2163 next
= this_parent
->mnt_child
.next
;
2164 this_parent
= this_parent
->mnt_parent
;
2171 * process a list of expirable mountpoints with the intent of discarding any
2172 * submounts of a specific parent mountpoint
2174 * vfsmount_lock must be held for write
2176 static void shrink_submounts(struct mount
*mnt
, struct list_head
*umounts
)
2178 LIST_HEAD(graveyard
);
2181 /* extract submounts of 'mountpoint' from the expiration list */
2182 while (select_submounts(mnt
, &graveyard
)) {
2183 while (!list_empty(&graveyard
)) {
2184 m
= list_first_entry(&graveyard
, struct mount
,
2186 touch_mnt_namespace(m
->mnt
.mnt_ns
);
2187 umount_tree(m
, 1, umounts
);
2193 * Some copy_from_user() implementations do not return the exact number of
2194 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2195 * Note that this function differs from copy_from_user() in that it will oops
2196 * on bad values of `to', rather than returning a short copy.
2198 static long exact_copy_from_user(void *to
, const void __user
* from
,
2202 const char __user
*f
= from
;
2205 if (!access_ok(VERIFY_READ
, from
, n
))
2209 if (__get_user(c
, f
)) {
2220 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2230 if (!(page
= __get_free_page(GFP_KERNEL
)))
2233 /* We only care that *some* data at the address the user
2234 * gave us is valid. Just in case, we'll zero
2235 * the remainder of the page.
2237 /* copy_from_user cannot cross TASK_SIZE ! */
2238 size
= TASK_SIZE
- (unsigned long)data
;
2239 if (size
> PAGE_SIZE
)
2242 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2248 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2253 int copy_mount_string(const void __user
*data
, char **where
)
2262 tmp
= strndup_user(data
, PAGE_SIZE
);
2264 return PTR_ERR(tmp
);
2271 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2272 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2274 * data is a (void *) that can point to any structure up to
2275 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2276 * information (or be NULL).
2278 * Pre-0.97 versions of mount() didn't have a flags word.
2279 * When the flags word was introduced its top half was required
2280 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2281 * Therefore, if this magic number is present, it carries no information
2282 * and must be discarded.
2284 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
2285 unsigned long flags
, void *data_page
)
2292 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2293 flags
&= ~MS_MGC_MSK
;
2295 /* Basic sanity checks */
2297 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2301 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2303 /* ... and get the mountpoint */
2304 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2308 retval
= security_sb_mount(dev_name
, &path
,
2309 type_page
, flags
, data_page
);
2313 /* Default to relatime unless overriden */
2314 if (!(flags
& MS_NOATIME
))
2315 mnt_flags
|= MNT_RELATIME
;
2317 /* Separate the per-mountpoint flags */
2318 if (flags
& MS_NOSUID
)
2319 mnt_flags
|= MNT_NOSUID
;
2320 if (flags
& MS_NODEV
)
2321 mnt_flags
|= MNT_NODEV
;
2322 if (flags
& MS_NOEXEC
)
2323 mnt_flags
|= MNT_NOEXEC
;
2324 if (flags
& MS_NOATIME
)
2325 mnt_flags
|= MNT_NOATIME
;
2326 if (flags
& MS_NODIRATIME
)
2327 mnt_flags
|= MNT_NODIRATIME
;
2328 if (flags
& MS_STRICTATIME
)
2329 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2330 if (flags
& MS_RDONLY
)
2331 mnt_flags
|= MNT_READONLY
;
2333 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2334 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2337 if (flags
& MS_REMOUNT
)
2338 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2340 else if (flags
& MS_BIND
)
2341 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2342 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2343 retval
= do_change_type(&path
, flags
);
2344 else if (flags
& MS_MOVE
)
2345 retval
= do_move_mount(&path
, dev_name
);
2347 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2348 dev_name
, data_page
);
2354 static struct mnt_namespace
*alloc_mnt_ns(void)
2356 struct mnt_namespace
*new_ns
;
2358 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2360 return ERR_PTR(-ENOMEM
);
2361 atomic_set(&new_ns
->count
, 1);
2362 new_ns
->root
= NULL
;
2363 INIT_LIST_HEAD(&new_ns
->list
);
2364 init_waitqueue_head(&new_ns
->poll
);
2369 void mnt_make_longterm(struct vfsmount
*mnt
)
2371 __mnt_make_longterm(real_mount(mnt
));
2374 void mnt_make_shortterm(struct vfsmount
*m
)
2377 struct mount
*mnt
= real_mount(m
);
2378 if (atomic_add_unless(&mnt
->mnt_longterm
, -1, 1))
2380 br_write_lock(vfsmount_lock
);
2381 atomic_dec(&mnt
->mnt_longterm
);
2382 br_write_unlock(vfsmount_lock
);
2387 * Allocate a new namespace structure and populate it with contents
2388 * copied from the namespace of the passed in task structure.
2390 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2391 struct fs_struct
*fs
)
2393 struct mnt_namespace
*new_ns
;
2394 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2395 struct mount
*p
, *q
;
2398 new_ns
= alloc_mnt_ns();
2402 down_write(&namespace_sem
);
2403 /* First pass: copy the tree topology */
2404 new = copy_tree(real_mount(mnt_ns
->root
), mnt_ns
->root
->mnt_root
,
2405 CL_COPY_ALL
| CL_EXPIRE
);
2407 up_write(&namespace_sem
);
2409 return ERR_PTR(-ENOMEM
);
2411 new_ns
->root
= &new->mnt
;
2412 br_write_lock(vfsmount_lock
);
2413 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2414 br_write_unlock(vfsmount_lock
);
2417 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2418 * as belonging to new namespace. We have already acquired a private
2419 * fs_struct, so tsk->fs->lock is not needed.
2421 p
= real_mount(mnt_ns
->root
);
2424 q
->mnt
.mnt_ns
= new_ns
;
2425 __mnt_make_longterm(q
);
2427 if (&p
->mnt
== fs
->root
.mnt
) {
2428 fs
->root
.mnt
= mntget(&q
->mnt
);
2429 __mnt_make_longterm(q
);
2430 mnt_make_shortterm(&p
->mnt
);
2433 if (&p
->mnt
== fs
->pwd
.mnt
) {
2434 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2435 __mnt_make_longterm(q
);
2436 mnt_make_shortterm(&p
->mnt
);
2440 p
= next_mnt(p
, mnt_ns
->root
);
2441 q
= next_mnt(q
, new_ns
->root
);
2443 up_write(&namespace_sem
);
2453 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2454 struct fs_struct
*new_fs
)
2456 struct mnt_namespace
*new_ns
;
2461 if (!(flags
& CLONE_NEWNS
))
2464 new_ns
= dup_mnt_ns(ns
, new_fs
);
2471 * create_mnt_ns - creates a private namespace and adds a root filesystem
2472 * @mnt: pointer to the new root filesystem mountpoint
2474 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*mnt
)
2476 struct mnt_namespace
*new_ns
;
2478 new_ns
= alloc_mnt_ns();
2479 if (!IS_ERR(new_ns
)) {
2480 mnt
->mnt_ns
= new_ns
;
2481 __mnt_make_longterm(real_mount(mnt
));
2483 list_add(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2490 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2492 struct mnt_namespace
*ns
;
2493 struct super_block
*s
;
2497 ns
= create_mnt_ns(mnt
);
2499 return ERR_CAST(ns
);
2501 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2502 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2507 return ERR_PTR(err
);
2509 /* trade a vfsmount reference for active sb one */
2510 s
= path
.mnt
->mnt_sb
;
2511 atomic_inc(&s
->s_active
);
2513 /* lock the sucker */
2514 down_write(&s
->s_umount
);
2515 /* ... and return the root of (sub)tree on it */
2518 EXPORT_SYMBOL(mount_subtree
);
2520 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2521 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2527 unsigned long data_page
;
2529 ret
= copy_mount_string(type
, &kernel_type
);
2533 kernel_dir
= getname(dir_name
);
2534 if (IS_ERR(kernel_dir
)) {
2535 ret
= PTR_ERR(kernel_dir
);
2539 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2543 ret
= copy_mount_options(data
, &data_page
);
2547 ret
= do_mount(kernel_dev
, kernel_dir
, kernel_type
, flags
,
2548 (void *) data_page
);
2550 free_page(data_page
);
2554 putname(kernel_dir
);
2562 * Return true if path is reachable from root
2564 * namespace_sem or vfsmount_lock is held
2566 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2567 const struct path
*root
)
2569 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2570 dentry
= mnt
->mnt_mountpoint
;
2571 mnt
= mnt
->mnt_parent
;
2573 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2576 int path_is_under(struct path
*path1
, struct path
*path2
)
2579 br_read_lock(vfsmount_lock
);
2580 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2581 br_read_unlock(vfsmount_lock
);
2584 EXPORT_SYMBOL(path_is_under
);
2587 * pivot_root Semantics:
2588 * Moves the root file system of the current process to the directory put_old,
2589 * makes new_root as the new root file system of the current process, and sets
2590 * root/cwd of all processes which had them on the current root to new_root.
2593 * The new_root and put_old must be directories, and must not be on the
2594 * same file system as the current process root. The put_old must be
2595 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2596 * pointed to by put_old must yield the same directory as new_root. No other
2597 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2599 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2600 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2601 * in this situation.
2604 * - we don't move root/cwd if they are not at the root (reason: if something
2605 * cared enough to change them, it's probably wrong to force them elsewhere)
2606 * - it's okay to pick a root that isn't the root of a file system, e.g.
2607 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2608 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2611 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2612 const char __user
*, put_old
)
2614 struct path
new, old
, parent_path
, root_parent
, root
;
2615 struct mount
*new_mnt
, *root_mnt
;
2618 if (!capable(CAP_SYS_ADMIN
))
2621 error
= user_path_dir(new_root
, &new);
2625 error
= user_path_dir(put_old
, &old
);
2629 error
= security_sb_pivotroot(&old
, &new);
2633 get_fs_root(current
->fs
, &root
);
2634 error
= lock_mount(&old
);
2639 new_mnt
= real_mount(new.mnt
);
2640 root_mnt
= real_mount(root
.mnt
);
2641 if (IS_MNT_SHARED(old
.mnt
) ||
2642 IS_MNT_SHARED(&new_mnt
->mnt_parent
->mnt
) ||
2643 IS_MNT_SHARED(&root_mnt
->mnt_parent
->mnt
))
2645 if (!check_mnt(root
.mnt
) || !check_mnt(new.mnt
))
2648 if (d_unlinked(new.dentry
))
2650 if (d_unlinked(old
.dentry
))
2653 if (new.mnt
== root
.mnt
||
2654 old
.mnt
== root
.mnt
)
2655 goto out4
; /* loop, on the same file system */
2657 if (root
.mnt
->mnt_root
!= root
.dentry
)
2658 goto out4
; /* not a mountpoint */
2659 if (!mnt_has_parent(root_mnt
))
2660 goto out4
; /* not attached */
2661 if (new.mnt
->mnt_root
!= new.dentry
)
2662 goto out4
; /* not a mountpoint */
2663 if (!mnt_has_parent(new_mnt
))
2664 goto out4
; /* not attached */
2665 /* make sure we can reach put_old from new_root */
2666 if (!is_path_reachable(real_mount(old
.mnt
), old
.dentry
, &new))
2668 br_write_lock(vfsmount_lock
);
2669 detach_mnt(new_mnt
, &parent_path
);
2670 detach_mnt(root_mnt
, &root_parent
);
2671 /* mount old root on put_old */
2672 attach_mnt(root_mnt
, &old
);
2673 /* mount new_root on / */
2674 attach_mnt(new_mnt
, &root_parent
);
2675 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2676 br_write_unlock(vfsmount_lock
);
2677 chroot_fs_refs(&root
, &new);
2682 path_put(&root_parent
);
2683 path_put(&parent_path
);
2695 static void __init
init_mount_tree(void)
2697 struct vfsmount
*mnt
;
2698 struct mnt_namespace
*ns
;
2701 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
2703 panic("Can't create rootfs");
2705 ns
= create_mnt_ns(mnt
);
2707 panic("Can't allocate initial namespace");
2709 init_task
.nsproxy
->mnt_ns
= ns
;
2712 root
.mnt
= ns
->root
;
2713 root
.dentry
= ns
->root
->mnt_root
;
2715 set_fs_pwd(current
->fs
, &root
);
2716 set_fs_root(current
->fs
, &root
);
2719 void __init
mnt_init(void)
2724 init_rwsem(&namespace_sem
);
2726 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2727 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2729 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2731 if (!mount_hashtable
)
2732 panic("Failed to allocate mount hash table\n");
2734 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2736 for (u
= 0; u
< HASH_SIZE
; u
++)
2737 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2739 br_lock_init(vfsmount_lock
);
2743 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2745 fs_kobj
= kobject_create_and_add("fs", NULL
);
2747 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2752 void put_mnt_ns(struct mnt_namespace
*ns
)
2754 LIST_HEAD(umount_list
);
2756 if (!atomic_dec_and_test(&ns
->count
))
2758 down_write(&namespace_sem
);
2759 br_write_lock(vfsmount_lock
);
2760 umount_tree(real_mount(ns
->root
), 0, &umount_list
);
2761 br_write_unlock(vfsmount_lock
);
2762 up_write(&namespace_sem
);
2763 release_mounts(&umount_list
);
2767 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2769 struct vfsmount
*mnt
;
2770 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2773 * it is a longterm mount, don't release mnt until
2774 * we unmount before file sys is unregistered
2776 mnt_make_longterm(mnt
);
2780 EXPORT_SYMBOL_GPL(kern_mount_data
);
2782 void kern_unmount(struct vfsmount
*mnt
)
2784 /* release long term mount so mount point can be released */
2785 if (!IS_ERR_OR_NULL(mnt
)) {
2786 mnt_make_shortterm(mnt
);
2790 EXPORT_SYMBOL(kern_unmount
);
2792 bool our_mnt(struct vfsmount
*mnt
)
2794 return check_mnt(mnt
);