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 vfsmount
*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 vfsmount
*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 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
194 goto out_free_devname
;
196 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
199 mnt
->mnt_writers
= 0;
202 INIT_LIST_HEAD(&p
->mnt_hash
);
203 INIT_LIST_HEAD(&mnt
->mnt_child
);
204 INIT_LIST_HEAD(&mnt
->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 vfsmount
*mnt
)
259 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
265 static inline void mnt_dec_writers(struct vfsmount
*mnt
)
268 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
274 static unsigned int mnt_get_writers(struct vfsmount
*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 * @mnt: 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
*mnt
)
313 mnt_inc_writers(mnt
);
315 * The store to mnt_inc_writers must be visible before we pass
316 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
317 * incremented count after it has set MNT_WRITE_HOLD.
320 while (mnt
->mnt_flags
& MNT_WRITE_HOLD
)
323 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
324 * be set to match its requirements. So we must not load that until
325 * MNT_WRITE_HOLD is cleared.
328 if (__mnt_is_readonly(mnt
)) {
329 mnt_dec_writers(mnt
);
337 EXPORT_SYMBOL_GPL(mnt_want_write
);
340 * mnt_clone_write - get write access to a mount
341 * @mnt: the mount on which to take a write
343 * This is effectively like mnt_want_write, except
344 * it must only be used to take an extra write reference
345 * on a mountpoint that we already know has a write reference
346 * on it. This allows some optimisation.
348 * After finished, mnt_drop_write must be called as usual to
349 * drop the reference.
351 int mnt_clone_write(struct vfsmount
*mnt
)
353 /* superblock may be r/o */
354 if (__mnt_is_readonly(mnt
))
357 mnt_inc_writers(mnt
);
361 EXPORT_SYMBOL_GPL(mnt_clone_write
);
364 * mnt_want_write_file - get write access to a file's mount
365 * @file: the file who's mount on which to take a write
367 * This is like mnt_want_write, but it takes a file and can
368 * do some optimisations if the file is open for write already
370 int mnt_want_write_file(struct file
*file
)
372 struct inode
*inode
= file
->f_dentry
->d_inode
;
373 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
374 return mnt_want_write(file
->f_path
.mnt
);
376 return mnt_clone_write(file
->f_path
.mnt
);
378 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
381 * mnt_drop_write - give up write access to a mount
382 * @mnt: the mount on which to give up write access
384 * Tells the low-level filesystem that we are done
385 * performing writes to it. Must be matched with
386 * mnt_want_write() call above.
388 void mnt_drop_write(struct vfsmount
*mnt
)
391 mnt_dec_writers(mnt
);
394 EXPORT_SYMBOL_GPL(mnt_drop_write
);
396 void mnt_drop_write_file(struct file
*file
)
398 mnt_drop_write(file
->f_path
.mnt
);
400 EXPORT_SYMBOL(mnt_drop_write_file
);
402 static int mnt_make_readonly(struct vfsmount
*mnt
)
406 br_write_lock(vfsmount_lock
);
407 mnt
->mnt_flags
|= MNT_WRITE_HOLD
;
409 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
410 * should be visible before we do.
415 * With writers on hold, if this value is zero, then there are
416 * definitely no active writers (although held writers may subsequently
417 * increment the count, they'll have to wait, and decrement it after
418 * seeing MNT_READONLY).
420 * It is OK to have counter incremented on one CPU and decremented on
421 * another: the sum will add up correctly. The danger would be when we
422 * sum up each counter, if we read a counter before it is incremented,
423 * but then read another CPU's count which it has been subsequently
424 * decremented from -- we would see more decrements than we should.
425 * MNT_WRITE_HOLD protects against this scenario, because
426 * mnt_want_write first increments count, then smp_mb, then spins on
427 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
428 * we're counting up here.
430 if (mnt_get_writers(mnt
) > 0)
433 mnt
->mnt_flags
|= MNT_READONLY
;
435 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
436 * that become unheld will see MNT_READONLY.
439 mnt
->mnt_flags
&= ~MNT_WRITE_HOLD
;
440 br_write_unlock(vfsmount_lock
);
444 static void __mnt_unmake_readonly(struct vfsmount
*mnt
)
446 br_write_lock(vfsmount_lock
);
447 mnt
->mnt_flags
&= ~MNT_READONLY
;
448 br_write_unlock(vfsmount_lock
);
451 static void free_vfsmnt(struct mount
*mnt
)
453 kfree(mnt
->mnt
.mnt_devname
);
456 free_percpu(mnt
->mnt
.mnt_pcp
);
458 kmem_cache_free(mnt_cache
, mnt
);
462 * find the first or last mount at @dentry on vfsmount @mnt depending on
463 * @dir. If @dir is set return the first mount else return the last mount.
464 * vfsmount_lock must be held for read or write.
466 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
469 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
470 struct list_head
*tmp
= head
;
471 struct mount
*p
, *found
= NULL
;
474 tmp
= dir
? tmp
->next
: tmp
->prev
;
478 p
= list_entry(tmp
, struct mount
, mnt_hash
);
479 if (p
->mnt
.mnt_parent
== mnt
&& p
->mnt
.mnt_mountpoint
== dentry
) {
488 * lookup_mnt increments the ref count before returning
489 * the vfsmount struct.
491 struct vfsmount
*lookup_mnt(struct path
*path
)
493 struct mount
*child_mnt
;
495 br_read_lock(vfsmount_lock
);
496 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
498 mnt_add_count(child_mnt
, 1);
499 br_read_unlock(vfsmount_lock
);
500 return &child_mnt
->mnt
;
502 br_read_unlock(vfsmount_lock
);
507 static inline int check_mnt(struct vfsmount
*mnt
)
509 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
513 * vfsmount lock must be held for write
515 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
519 wake_up_interruptible(&ns
->poll
);
524 * vfsmount lock must be held for write
526 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
528 if (ns
&& ns
->event
!= event
) {
530 wake_up_interruptible(&ns
->poll
);
535 * Clear dentry's mounted state if it has no remaining mounts.
536 * vfsmount_lock must be held for write.
538 static void dentry_reset_mounted(struct dentry
*dentry
)
542 for (u
= 0; u
< HASH_SIZE
; u
++) {
545 list_for_each_entry(p
, &mount_hashtable
[u
], mnt_hash
) {
546 if (p
->mnt
.mnt_mountpoint
== dentry
)
550 spin_lock(&dentry
->d_lock
);
551 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
552 spin_unlock(&dentry
->d_lock
);
556 * vfsmount lock must be held for write
558 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
560 old_path
->dentry
= mnt
->mnt
.mnt_mountpoint
;
561 old_path
->mnt
= mnt
->mnt
.mnt_parent
;
562 mnt
->mnt
.mnt_parent
= &mnt
->mnt
;
563 mnt
->mnt
.mnt_mountpoint
= mnt
->mnt
.mnt_root
;
564 list_del_init(&mnt
->mnt
.mnt_child
);
565 list_del_init(&mnt
->mnt_hash
);
566 dentry_reset_mounted(old_path
->dentry
);
570 * vfsmount lock must be held for write
572 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
573 struct vfsmount
*child_mnt
)
575 child_mnt
->mnt_parent
= mntget(mnt
);
576 child_mnt
->mnt_mountpoint
= dget(dentry
);
577 spin_lock(&dentry
->d_lock
);
578 dentry
->d_flags
|= DCACHE_MOUNTED
;
579 spin_unlock(&dentry
->d_lock
);
583 * vfsmount lock must be held for write
585 static void attach_mnt(struct mount
*mnt
, struct path
*path
)
587 mnt_set_mountpoint(path
->mnt
, path
->dentry
, &mnt
->mnt
);
588 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
589 hash(path
->mnt
, path
->dentry
));
590 list_add_tail(&mnt
->mnt
.mnt_child
, &path
->mnt
->mnt_mounts
);
593 static inline void __mnt_make_longterm(struct vfsmount
*mnt
)
596 atomic_inc(&mnt
->mnt_longterm
);
600 /* needs vfsmount lock for write */
601 static inline void __mnt_make_shortterm(struct vfsmount
*mnt
)
604 atomic_dec(&mnt
->mnt_longterm
);
609 * vfsmount lock must be held for write
611 static void commit_tree(struct mount
*mnt
)
613 struct vfsmount
*parent
= mnt
->mnt
.mnt_parent
;
616 struct mnt_namespace
*n
= parent
->mnt_ns
;
618 BUG_ON(parent
== &mnt
->mnt
);
620 list_add_tail(&head
, &mnt
->mnt
.mnt_list
);
621 list_for_each_entry(m
, &head
, mnt_list
) {
623 __mnt_make_longterm(m
);
626 list_splice(&head
, n
->list
.prev
);
628 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
629 hash(parent
, mnt
->mnt
.mnt_mountpoint
));
630 list_add_tail(&mnt
->mnt
.mnt_child
, &parent
->mnt_mounts
);
631 touch_mnt_namespace(n
);
634 static struct mount
*next_mnt(struct mount
*p
, struct vfsmount
*root
)
636 struct list_head
*next
= p
->mnt
.mnt_mounts
.next
;
637 if (next
== &p
->mnt
.mnt_mounts
) {
641 next
= p
->mnt
.mnt_child
.next
;
642 if (next
!= &p
->mnt
.mnt_parent
->mnt_mounts
)
644 p
= real_mount(p
->mnt
.mnt_parent
);
647 return list_entry(next
, struct mount
, mnt
.mnt_child
);
650 static struct mount
*skip_mnt_tree(struct mount
*p
)
652 struct list_head
*prev
= p
->mnt
.mnt_mounts
.prev
;
653 while (prev
!= &p
->mnt
.mnt_mounts
) {
654 p
= list_entry(prev
, struct mount
, mnt
.mnt_child
);
655 prev
= p
->mnt
.mnt_mounts
.prev
;
661 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
667 return ERR_PTR(-ENODEV
);
669 mnt
= alloc_vfsmnt(name
);
671 return ERR_PTR(-ENOMEM
);
673 if (flags
& MS_KERNMOUNT
)
674 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
676 root
= mount_fs(type
, flags
, name
, data
);
679 return ERR_CAST(root
);
682 mnt
->mnt
.mnt_root
= root
;
683 mnt
->mnt
.mnt_sb
= root
->d_sb
;
684 mnt
->mnt
.mnt_mountpoint
= mnt
->mnt
.mnt_root
;
685 mnt
->mnt
.mnt_parent
= &mnt
->mnt
;
688 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
690 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
693 struct super_block
*sb
= old
->mnt
.mnt_sb
;
694 struct mount
*mnt
= alloc_vfsmnt(old
->mnt
.mnt_devname
);
697 if (flag
& (CL_SLAVE
| CL_PRIVATE
))
698 mnt
->mnt
.mnt_group_id
= 0; /* not a peer of original */
700 mnt
->mnt
.mnt_group_id
= old
->mnt
.mnt_group_id
;
702 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt
.mnt_group_id
) {
703 int err
= mnt_alloc_group_id(mnt
);
708 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
709 atomic_inc(&sb
->s_active
);
710 mnt
->mnt
.mnt_sb
= sb
;
711 mnt
->mnt
.mnt_root
= dget(root
);
712 mnt
->mnt
.mnt_mountpoint
= mnt
->mnt
.mnt_root
;
713 mnt
->mnt
.mnt_parent
= &mnt
->mnt
;
715 if (flag
& CL_SLAVE
) {
716 list_add(&mnt
->mnt
.mnt_slave
, &old
->mnt
.mnt_slave_list
);
717 mnt
->mnt
.mnt_master
= &old
->mnt
;
718 CLEAR_MNT_SHARED(&mnt
->mnt
);
719 } else if (!(flag
& CL_PRIVATE
)) {
720 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(&old
->mnt
))
721 list_add(&mnt
->mnt
.mnt_share
, &old
->mnt
.mnt_share
);
722 if (IS_MNT_SLAVE(&old
->mnt
))
723 list_add(&mnt
->mnt
.mnt_slave
, &old
->mnt
.mnt_slave
);
724 mnt
->mnt
.mnt_master
= old
->mnt
.mnt_master
;
726 if (flag
& CL_MAKE_SHARED
)
729 /* stick the duplicate mount on the same expiry list
730 * as the original if that was on one */
731 if (flag
& CL_EXPIRE
) {
732 if (!list_empty(&old
->mnt
.mnt_expire
))
733 list_add(&mnt
->mnt
.mnt_expire
, &old
->mnt
.mnt_expire
);
743 static inline void mntfree(struct vfsmount
*mnt
)
745 struct super_block
*sb
= mnt
->mnt_sb
;
748 * This probably indicates that somebody messed
749 * up a mnt_want/drop_write() pair. If this
750 * happens, the filesystem was probably unable
751 * to make r/w->r/o transitions.
754 * The locking used to deal with mnt_count decrement provides barriers,
755 * so mnt_get_writers() below is safe.
757 WARN_ON(mnt_get_writers(mnt
));
758 fsnotify_vfsmount_delete(mnt
);
760 free_vfsmnt(real_mount(mnt
));
761 deactivate_super(sb
);
764 static void mntput_no_expire(struct vfsmount
*mnt
)
768 br_read_lock(vfsmount_lock
);
769 if (likely(atomic_read(&mnt
->mnt_longterm
))) {
770 mnt_add_count(mnt
, -1);
771 br_read_unlock(vfsmount_lock
);
774 br_read_unlock(vfsmount_lock
);
776 br_write_lock(vfsmount_lock
);
777 mnt_add_count(mnt
, -1);
778 if (mnt_get_count(mnt
)) {
779 br_write_unlock(vfsmount_lock
);
783 mnt_add_count(mnt
, -1);
784 if (likely(mnt_get_count(mnt
)))
786 br_write_lock(vfsmount_lock
);
788 if (unlikely(mnt
->mnt_pinned
)) {
789 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
791 br_write_unlock(vfsmount_lock
);
792 acct_auto_close_mnt(mnt
);
795 br_write_unlock(vfsmount_lock
);
799 void mntput(struct vfsmount
*mnt
)
802 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
803 if (unlikely(mnt
->mnt_expiry_mark
))
804 mnt
->mnt_expiry_mark
= 0;
805 mntput_no_expire(mnt
);
808 EXPORT_SYMBOL(mntput
);
810 struct vfsmount
*mntget(struct vfsmount
*mnt
)
813 mnt_add_count(mnt
, 1);
816 EXPORT_SYMBOL(mntget
);
818 void mnt_pin(struct vfsmount
*mnt
)
820 br_write_lock(vfsmount_lock
);
822 br_write_unlock(vfsmount_lock
);
824 EXPORT_SYMBOL(mnt_pin
);
826 void mnt_unpin(struct vfsmount
*mnt
)
828 br_write_lock(vfsmount_lock
);
829 if (mnt
->mnt_pinned
) {
830 mnt_add_count(mnt
, 1);
833 br_write_unlock(vfsmount_lock
);
835 EXPORT_SYMBOL(mnt_unpin
);
837 static inline void mangle(struct seq_file
*m
, const char *s
)
839 seq_escape(m
, s
, " \t\n\\");
843 * Simple .show_options callback for filesystems which don't want to
844 * implement more complex mount option showing.
846 * See also save_mount_options().
848 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
853 options
= rcu_dereference(mnt
->mnt_sb
->s_options
);
855 if (options
!= NULL
&& options
[0]) {
863 EXPORT_SYMBOL(generic_show_options
);
866 * If filesystem uses generic_show_options(), this function should be
867 * called from the fill_super() callback.
869 * The .remount_fs callback usually needs to be handled in a special
870 * way, to make sure, that previous options are not overwritten if the
873 * Also note, that if the filesystem's .remount_fs function doesn't
874 * reset all options to their default value, but changes only newly
875 * given options, then the displayed options will not reflect reality
878 void save_mount_options(struct super_block
*sb
, char *options
)
880 BUG_ON(sb
->s_options
);
881 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
883 EXPORT_SYMBOL(save_mount_options
);
885 void replace_mount_options(struct super_block
*sb
, char *options
)
887 char *old
= sb
->s_options
;
888 rcu_assign_pointer(sb
->s_options
, options
);
894 EXPORT_SYMBOL(replace_mount_options
);
896 #ifdef CONFIG_PROC_FS
898 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
900 struct proc_mounts
*p
= m
->private;
902 down_read(&namespace_sem
);
903 return seq_list_start(&p
->ns
->list
, *pos
);
906 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
908 struct proc_mounts
*p
= m
->private;
910 return seq_list_next(v
, &p
->ns
->list
, pos
);
913 static void m_stop(struct seq_file
*m
, void *v
)
915 up_read(&namespace_sem
);
918 int mnt_had_events(struct proc_mounts
*p
)
920 struct mnt_namespace
*ns
= p
->ns
;
923 br_read_lock(vfsmount_lock
);
924 if (p
->m
.poll_event
!= ns
->event
) {
925 p
->m
.poll_event
= ns
->event
;
928 br_read_unlock(vfsmount_lock
);
933 struct proc_fs_info
{
938 static int show_sb_opts(struct seq_file
*m
, struct super_block
*sb
)
940 static const struct proc_fs_info fs_info
[] = {
941 { MS_SYNCHRONOUS
, ",sync" },
942 { MS_DIRSYNC
, ",dirsync" },
943 { MS_MANDLOCK
, ",mand" },
946 const struct proc_fs_info
*fs_infop
;
948 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
949 if (sb
->s_flags
& fs_infop
->flag
)
950 seq_puts(m
, fs_infop
->str
);
953 return security_sb_show_options(m
, sb
);
956 static void show_mnt_opts(struct seq_file
*m
, struct vfsmount
*mnt
)
958 static const struct proc_fs_info mnt_info
[] = {
959 { MNT_NOSUID
, ",nosuid" },
960 { MNT_NODEV
, ",nodev" },
961 { MNT_NOEXEC
, ",noexec" },
962 { MNT_NOATIME
, ",noatime" },
963 { MNT_NODIRATIME
, ",nodiratime" },
964 { MNT_RELATIME
, ",relatime" },
967 const struct proc_fs_info
*fs_infop
;
969 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
970 if (mnt
->mnt_flags
& fs_infop
->flag
)
971 seq_puts(m
, fs_infop
->str
);
975 static void show_type(struct seq_file
*m
, struct super_block
*sb
)
977 mangle(m
, sb
->s_type
->name
);
978 if (sb
->s_subtype
&& sb
->s_subtype
[0]) {
980 mangle(m
, sb
->s_subtype
);
984 static int show_vfsmnt(struct seq_file
*m
, void *v
)
986 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
988 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
990 if (mnt
->mnt_sb
->s_op
->show_devname
) {
991 err
= mnt
->mnt_sb
->s_op
->show_devname(m
, mnt
);
995 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
998 seq_path(m
, &mnt_path
, " \t\n\\");
1000 show_type(m
, mnt
->mnt_sb
);
1001 seq_puts(m
, __mnt_is_readonly(mnt
) ? " ro" : " rw");
1002 err
= show_sb_opts(m
, mnt
->mnt_sb
);
1005 show_mnt_opts(m
, mnt
);
1006 if (mnt
->mnt_sb
->s_op
->show_options
)
1007 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
1008 seq_puts(m
, " 0 0\n");
1013 const struct seq_operations mounts_op
= {
1020 static int show_mountinfo(struct seq_file
*m
, void *v
)
1022 struct proc_mounts
*p
= m
->private;
1023 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
1024 struct super_block
*sb
= mnt
->mnt_sb
;
1025 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
1026 struct path root
= p
->root
;
1029 seq_printf(m
, "%i %i %u:%u ", mnt
->mnt_id
, mnt
->mnt_parent
->mnt_id
,
1030 MAJOR(sb
->s_dev
), MINOR(sb
->s_dev
));
1031 if (sb
->s_op
->show_path
)
1032 err
= sb
->s_op
->show_path(m
, mnt
);
1034 seq_dentry(m
, mnt
->mnt_root
, " \t\n\\");
1039 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1040 err
= seq_path_root(m
, &mnt_path
, &root
, " \t\n\\");
1044 seq_puts(m
, mnt
->mnt_flags
& MNT_READONLY
? " ro" : " rw");
1045 show_mnt_opts(m
, mnt
);
1047 /* Tagged fields ("foo:X" or "bar") */
1048 if (IS_MNT_SHARED(mnt
))
1049 seq_printf(m
, " shared:%i", mnt
->mnt_group_id
);
1050 if (IS_MNT_SLAVE(mnt
)) {
1051 int master
= mnt
->mnt_master
->mnt_group_id
;
1052 int dom
= get_dominating_id(mnt
, &p
->root
);
1053 seq_printf(m
, " master:%i", master
);
1054 if (dom
&& dom
!= master
)
1055 seq_printf(m
, " propagate_from:%i", dom
);
1057 if (IS_MNT_UNBINDABLE(mnt
))
1058 seq_puts(m
, " unbindable");
1060 /* Filesystem specific data */
1064 if (sb
->s_op
->show_devname
)
1065 err
= sb
->s_op
->show_devname(m
, mnt
);
1067 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
1070 seq_puts(m
, sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
1071 err
= show_sb_opts(m
, sb
);
1074 if (sb
->s_op
->show_options
)
1075 err
= sb
->s_op
->show_options(m
, mnt
);
1081 const struct seq_operations mountinfo_op
= {
1085 .show
= show_mountinfo
,
1088 static int show_vfsstat(struct seq_file
*m
, void *v
)
1090 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
1091 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
1095 if (mnt
->mnt_sb
->s_op
->show_devname
) {
1096 seq_puts(m
, "device ");
1097 err
= mnt
->mnt_sb
->s_op
->show_devname(m
, mnt
);
1099 if (mnt
->mnt_devname
) {
1100 seq_puts(m
, "device ");
1101 mangle(m
, mnt
->mnt_devname
);
1103 seq_puts(m
, "no device");
1107 seq_puts(m
, " mounted on ");
1108 seq_path(m
, &mnt_path
, " \t\n\\");
1111 /* file system type */
1112 seq_puts(m
, "with fstype ");
1113 show_type(m
, mnt
->mnt_sb
);
1115 /* optional statistics */
1116 if (mnt
->mnt_sb
->s_op
->show_stats
) {
1119 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
1126 const struct seq_operations mountstats_op
= {
1130 .show
= show_vfsstat
,
1132 #endif /* CONFIG_PROC_FS */
1135 * may_umount_tree - check if a mount tree is busy
1136 * @mnt: root of mount tree
1138 * This is called to check if a tree of mounts has any
1139 * open files, pwds, chroots or sub mounts that are
1142 int may_umount_tree(struct vfsmount
*mnt
)
1144 int actual_refs
= 0;
1145 int minimum_refs
= 0;
1149 /* write lock needed for mnt_get_count */
1150 br_write_lock(vfsmount_lock
);
1151 for (p
= real_mount(mnt
); p
; p
= next_mnt(p
, mnt
)) {
1152 actual_refs
+= mnt_get_count(&p
->mnt
);
1155 br_write_unlock(vfsmount_lock
);
1157 if (actual_refs
> minimum_refs
)
1163 EXPORT_SYMBOL(may_umount_tree
);
1166 * may_umount - check if a mount point is busy
1167 * @mnt: root of mount
1169 * This is called to check if a mount point has any
1170 * open files, pwds, chroots or sub mounts. If the
1171 * mount has sub mounts this will return busy
1172 * regardless of whether the sub mounts are busy.
1174 * Doesn't take quota and stuff into account. IOW, in some cases it will
1175 * give false negatives. The main reason why it's here is that we need
1176 * a non-destructive way to look for easily umountable filesystems.
1178 int may_umount(struct vfsmount
*mnt
)
1181 down_read(&namespace_sem
);
1182 br_write_lock(vfsmount_lock
);
1183 if (propagate_mount_busy(mnt
, 2))
1185 br_write_unlock(vfsmount_lock
);
1186 up_read(&namespace_sem
);
1190 EXPORT_SYMBOL(may_umount
);
1192 void release_mounts(struct list_head
*head
)
1195 while (!list_empty(head
)) {
1196 mnt
= list_first_entry(head
, struct mount
, mnt_hash
);
1197 list_del_init(&mnt
->mnt_hash
);
1198 if (mnt_has_parent(&mnt
->mnt
)) {
1199 struct dentry
*dentry
;
1202 br_write_lock(vfsmount_lock
);
1203 dentry
= mnt
->mnt
.mnt_mountpoint
;
1204 m
= mnt
->mnt
.mnt_parent
;
1205 mnt
->mnt
.mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1206 mnt
->mnt
.mnt_parent
= &mnt
->mnt
;
1208 br_write_unlock(vfsmount_lock
);
1217 * vfsmount lock must be held for write
1218 * namespace_sem must be held for write
1220 void umount_tree(struct mount
*mnt
, int propagate
, struct list_head
*kill
)
1222 LIST_HEAD(tmp_list
);
1225 for (p
= mnt
; p
; p
= next_mnt(p
, &mnt
->mnt
))
1226 list_move(&p
->mnt_hash
, &tmp_list
);
1229 propagate_umount(&tmp_list
);
1231 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1232 list_del_init(&p
->mnt
.mnt_expire
);
1233 list_del_init(&p
->mnt
.mnt_list
);
1234 __touch_mnt_namespace(p
->mnt
.mnt_ns
);
1235 p
->mnt
.mnt_ns
= NULL
;
1236 __mnt_make_shortterm(&p
->mnt
);
1237 list_del_init(&p
->mnt
.mnt_child
);
1238 if (mnt_has_parent(&p
->mnt
)) {
1239 p
->mnt
.mnt_parent
->mnt_ghosts
++;
1240 dentry_reset_mounted(p
->mnt
.mnt_mountpoint
);
1242 change_mnt_propagation(p
, MS_PRIVATE
);
1244 list_splice(&tmp_list
, kill
);
1247 static void shrink_submounts(struct mount
*mnt
, struct list_head
*umounts
);
1249 static int do_umount(struct vfsmount
*mnt
, int flags
)
1251 struct super_block
*sb
= mnt
->mnt_sb
;
1253 LIST_HEAD(umount_list
);
1255 retval
= security_sb_umount(mnt
, flags
);
1260 * Allow userspace to request a mountpoint be expired rather than
1261 * unmounting unconditionally. Unmount only happens if:
1262 * (1) the mark is already set (the mark is cleared by mntput())
1263 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1265 if (flags
& MNT_EXPIRE
) {
1266 if (mnt
== current
->fs
->root
.mnt
||
1267 flags
& (MNT_FORCE
| MNT_DETACH
))
1271 * probably don't strictly need the lock here if we examined
1272 * all race cases, but it's a slowpath.
1274 br_write_lock(vfsmount_lock
);
1275 if (mnt_get_count(mnt
) != 2) {
1276 br_write_unlock(vfsmount_lock
);
1279 br_write_unlock(vfsmount_lock
);
1281 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1286 * If we may have to abort operations to get out of this
1287 * mount, and they will themselves hold resources we must
1288 * allow the fs to do things. In the Unix tradition of
1289 * 'Gee thats tricky lets do it in userspace' the umount_begin
1290 * might fail to complete on the first run through as other tasks
1291 * must return, and the like. Thats for the mount program to worry
1292 * about for the moment.
1295 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1296 sb
->s_op
->umount_begin(sb
);
1300 * No sense to grab the lock for this test, but test itself looks
1301 * somewhat bogus. Suggestions for better replacement?
1302 * Ho-hum... In principle, we might treat that as umount + switch
1303 * to rootfs. GC would eventually take care of the old vfsmount.
1304 * Actually it makes sense, especially if rootfs would contain a
1305 * /reboot - static binary that would close all descriptors and
1306 * call reboot(9). Then init(8) could umount root and exec /reboot.
1308 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1310 * Special case for "unmounting" root ...
1311 * we just try to remount it readonly.
1313 down_write(&sb
->s_umount
);
1314 if (!(sb
->s_flags
& MS_RDONLY
))
1315 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1316 up_write(&sb
->s_umount
);
1320 down_write(&namespace_sem
);
1321 br_write_lock(vfsmount_lock
);
1324 if (!(flags
& MNT_DETACH
))
1325 shrink_submounts(real_mount(mnt
), &umount_list
);
1328 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1329 if (!list_empty(&mnt
->mnt_list
))
1330 umount_tree(real_mount(mnt
), 1, &umount_list
);
1333 br_write_unlock(vfsmount_lock
);
1334 up_write(&namespace_sem
);
1335 release_mounts(&umount_list
);
1340 * Now umount can handle mount points as well as block devices.
1341 * This is important for filesystems which use unnamed block devices.
1343 * We now support a flag for forced unmount like the other 'big iron'
1344 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1347 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1351 int lookup_flags
= 0;
1353 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1356 if (!(flags
& UMOUNT_NOFOLLOW
))
1357 lookup_flags
|= LOOKUP_FOLLOW
;
1359 retval
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1363 if (path
.dentry
!= path
.mnt
->mnt_root
)
1365 if (!check_mnt(path
.mnt
))
1369 if (!capable(CAP_SYS_ADMIN
))
1372 retval
= do_umount(path
.mnt
, flags
);
1374 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1376 mntput_no_expire(path
.mnt
);
1381 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1384 * The 2.0 compatible umount. No flags.
1386 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1388 return sys_umount(name
, 0);
1393 static int mount_is_safe(struct path
*path
)
1395 if (capable(CAP_SYS_ADMIN
))
1399 if (S_ISLNK(path
->dentry
->d_inode
->i_mode
))
1401 if (path
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
1402 if (current_uid() != path
->dentry
->d_inode
->i_uid
)
1405 if (inode_permission(path
->dentry
->d_inode
, MAY_WRITE
))
1411 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1414 struct mount
*res
, *p
, *q
;
1418 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(&mnt
->mnt
))
1421 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1424 q
->mnt
.mnt_mountpoint
= mnt
->mnt
.mnt_mountpoint
;
1427 list_for_each_entry(r
, &mnt
->mnt
.mnt_mounts
, mnt_child
) {
1429 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1432 for (s
= real_mount(r
); s
; s
= next_mnt(s
, r
)) {
1433 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(&s
->mnt
)) {
1434 s
= skip_mnt_tree(s
);
1437 while (p
!= real_mount(s
->mnt
.mnt_parent
)) {
1438 p
= real_mount(p
->mnt
.mnt_parent
);
1439 q
= real_mount(q
->mnt
.mnt_parent
);
1443 path
.dentry
= p
->mnt
.mnt_mountpoint
;
1444 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1447 br_write_lock(vfsmount_lock
);
1448 list_add_tail(&q
->mnt
.mnt_list
, &res
->mnt
.mnt_list
);
1449 attach_mnt(q
, &path
);
1450 br_write_unlock(vfsmount_lock
);
1456 LIST_HEAD(umount_list
);
1457 br_write_lock(vfsmount_lock
);
1458 umount_tree(res
, 0, &umount_list
);
1459 br_write_unlock(vfsmount_lock
);
1460 release_mounts(&umount_list
);
1465 struct vfsmount
*collect_mounts(struct path
*path
)
1468 down_write(&namespace_sem
);
1469 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1470 CL_COPY_ALL
| CL_PRIVATE
);
1471 up_write(&namespace_sem
);
1472 return tree
? &tree
->mnt
: NULL
;
1475 void drop_collected_mounts(struct vfsmount
*mnt
)
1477 LIST_HEAD(umount_list
);
1478 down_write(&namespace_sem
);
1479 br_write_lock(vfsmount_lock
);
1480 umount_tree(real_mount(mnt
), 0, &umount_list
);
1481 br_write_unlock(vfsmount_lock
);
1482 up_write(&namespace_sem
);
1483 release_mounts(&umount_list
);
1486 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1487 struct vfsmount
*root
)
1489 struct vfsmount
*mnt
;
1490 int res
= f(root
, arg
);
1493 list_for_each_entry(mnt
, &root
->mnt_list
, mnt_list
) {
1501 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1505 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, &mnt
->mnt
)) {
1506 if (p
->mnt
.mnt_group_id
&& !IS_MNT_SHARED(&p
->mnt
))
1507 mnt_release_group_id(p
);
1511 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1515 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, &mnt
->mnt
) : NULL
) {
1516 if (!p
->mnt
.mnt_group_id
&& !IS_MNT_SHARED(&p
->mnt
)) {
1517 int err
= mnt_alloc_group_id(p
);
1519 cleanup_group_ids(mnt
, p
);
1529 * @source_mnt : mount tree to be attached
1530 * @nd : place the mount tree @source_mnt is attached
1531 * @parent_nd : if non-null, detach the source_mnt from its parent and
1532 * store the parent mount and mountpoint dentry.
1533 * (done when source_mnt is moved)
1535 * NOTE: in the table below explains the semantics when a source mount
1536 * of a given type is attached to a destination mount of a given type.
1537 * ---------------------------------------------------------------------------
1538 * | BIND MOUNT OPERATION |
1539 * |**************************************************************************
1540 * | source-->| shared | private | slave | unbindable |
1544 * |**************************************************************************
1545 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1547 * |non-shared| shared (+) | private | slave (*) | invalid |
1548 * ***************************************************************************
1549 * A bind operation clones the source mount and mounts the clone on the
1550 * destination mount.
1552 * (++) the cloned mount is propagated to all the mounts in the propagation
1553 * tree of the destination mount and the cloned mount is added to
1554 * the peer group of the source mount.
1555 * (+) the cloned mount is created under the destination mount and is marked
1556 * as shared. The cloned mount is added to the peer group of the source
1558 * (+++) the mount is propagated to all the mounts in the propagation tree
1559 * of the destination mount and the cloned mount is made slave
1560 * of the same master as that of the source mount. The cloned mount
1561 * is marked as 'shared and slave'.
1562 * (*) the cloned mount is made a slave of the same master as that of the
1565 * ---------------------------------------------------------------------------
1566 * | MOVE MOUNT OPERATION |
1567 * |**************************************************************************
1568 * | source-->| shared | private | slave | unbindable |
1572 * |**************************************************************************
1573 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1575 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1576 * ***************************************************************************
1578 * (+) the mount is moved to the destination. And is then propagated to
1579 * all the mounts in the propagation tree of the destination mount.
1580 * (+*) the mount is moved to the destination.
1581 * (+++) the mount is moved to the destination and is then propagated to
1582 * all the mounts belonging to the destination mount's propagation tree.
1583 * the mount is marked as 'shared and slave'.
1584 * (*) the mount continues to be a slave at the new location.
1586 * if the source mount is a tree, the operations explained above is
1587 * applied to each mount in the tree.
1588 * Must be called without spinlocks held, since this function can sleep
1591 static int attach_recursive_mnt(struct mount
*source_mnt
,
1592 struct path
*path
, struct path
*parent_path
)
1594 LIST_HEAD(tree_list
);
1595 struct vfsmount
*dest_mnt
= path
->mnt
;
1596 struct dentry
*dest_dentry
= path
->dentry
;
1597 struct mount
*child
, *p
;
1600 if (IS_MNT_SHARED(dest_mnt
)) {
1601 err
= invent_group_ids(source_mnt
, true);
1605 err
= propagate_mnt(dest_mnt
, dest_dentry
, &source_mnt
->mnt
, &tree_list
);
1607 goto out_cleanup_ids
;
1609 br_write_lock(vfsmount_lock
);
1611 if (IS_MNT_SHARED(dest_mnt
)) {
1612 for (p
= source_mnt
; p
; p
= next_mnt(p
, &source_mnt
->mnt
))
1616 detach_mnt(source_mnt
, parent_path
);
1617 attach_mnt(source_mnt
, path
);
1618 touch_mnt_namespace(parent_path
->mnt
->mnt_ns
);
1620 mnt_set_mountpoint(dest_mnt
, dest_dentry
, &source_mnt
->mnt
);
1621 commit_tree(source_mnt
);
1624 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1625 list_del_init(&child
->mnt_hash
);
1628 br_write_unlock(vfsmount_lock
);
1633 if (IS_MNT_SHARED(dest_mnt
))
1634 cleanup_group_ids(source_mnt
, NULL
);
1639 static int lock_mount(struct path
*path
)
1641 struct vfsmount
*mnt
;
1643 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1644 if (unlikely(cant_mount(path
->dentry
))) {
1645 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1648 down_write(&namespace_sem
);
1649 mnt
= lookup_mnt(path
);
1652 up_write(&namespace_sem
);
1653 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1656 path
->dentry
= dget(mnt
->mnt_root
);
1660 static void unlock_mount(struct path
*path
)
1662 up_write(&namespace_sem
);
1663 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1666 static int graft_tree(struct vfsmount
*mnt
, struct path
*path
)
1668 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
1671 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1672 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
1675 if (d_unlinked(path
->dentry
))
1678 return attach_recursive_mnt(real_mount(mnt
), path
, NULL
);
1682 * Sanity check the flags to change_mnt_propagation.
1685 static int flags_to_propagation_type(int flags
)
1687 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1689 /* Fail if any non-propagation flags are set */
1690 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1692 /* Only one propagation flag should be set */
1693 if (!is_power_of_2(type
))
1699 * recursively change the type of the mountpoint.
1701 static int do_change_type(struct path
*path
, int flag
)
1704 struct mount
*mnt
= real_mount(path
->mnt
);
1705 int recurse
= flag
& MS_REC
;
1709 if (!capable(CAP_SYS_ADMIN
))
1712 if (path
->dentry
!= path
->mnt
->mnt_root
)
1715 type
= flags_to_propagation_type(flag
);
1719 down_write(&namespace_sem
);
1720 if (type
== MS_SHARED
) {
1721 err
= invent_group_ids(mnt
, recurse
);
1726 br_write_lock(vfsmount_lock
);
1727 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, &mnt
->mnt
) : NULL
))
1728 change_mnt_propagation(m
, type
);
1729 br_write_unlock(vfsmount_lock
);
1732 up_write(&namespace_sem
);
1737 * do loopback mount.
1739 static int do_loopback(struct path
*path
, char *old_name
,
1742 LIST_HEAD(umount_list
);
1743 struct path old_path
;
1744 struct mount
*mnt
= NULL
, *old
;
1745 int err
= mount_is_safe(path
);
1748 if (!old_name
|| !*old_name
)
1750 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1754 err
= lock_mount(path
);
1758 old
= real_mount(old_path
.mnt
);
1761 if (IS_MNT_UNBINDABLE(old_path
.mnt
))
1764 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1769 mnt
= copy_tree(old
, old_path
.dentry
, 0);
1771 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1776 err
= graft_tree(&mnt
->mnt
, path
);
1778 br_write_lock(vfsmount_lock
);
1779 umount_tree(mnt
, 0, &umount_list
);
1780 br_write_unlock(vfsmount_lock
);
1784 release_mounts(&umount_list
);
1786 path_put(&old_path
);
1790 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1793 int readonly_request
= 0;
1795 if (ms_flags
& MS_RDONLY
)
1796 readonly_request
= 1;
1797 if (readonly_request
== __mnt_is_readonly(mnt
))
1800 if (readonly_request
)
1801 error
= mnt_make_readonly(mnt
);
1803 __mnt_unmake_readonly(mnt
);
1808 * change filesystem flags. dir should be a physical root of filesystem.
1809 * If you've mounted a non-root directory somewhere and want to do remount
1810 * on it - tough luck.
1812 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1816 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1818 if (!capable(CAP_SYS_ADMIN
))
1821 if (!check_mnt(path
->mnt
))
1824 if (path
->dentry
!= path
->mnt
->mnt_root
)
1827 err
= security_sb_remount(sb
, data
);
1831 down_write(&sb
->s_umount
);
1832 if (flags
& MS_BIND
)
1833 err
= change_mount_flags(path
->mnt
, flags
);
1835 err
= do_remount_sb(sb
, flags
, data
, 0);
1837 br_write_lock(vfsmount_lock
);
1838 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_PROPAGATION_MASK
;
1839 path
->mnt
->mnt_flags
= mnt_flags
;
1840 br_write_unlock(vfsmount_lock
);
1842 up_write(&sb
->s_umount
);
1844 br_write_lock(vfsmount_lock
);
1845 touch_mnt_namespace(path
->mnt
->mnt_ns
);
1846 br_write_unlock(vfsmount_lock
);
1851 static inline int tree_contains_unbindable(struct mount
*mnt
)
1854 for (p
= mnt
; p
; p
= next_mnt(p
, &mnt
->mnt
)) {
1855 if (IS_MNT_UNBINDABLE(&p
->mnt
))
1861 static int do_move_mount(struct path
*path
, char *old_name
)
1863 struct path old_path
, parent_path
;
1867 if (!capable(CAP_SYS_ADMIN
))
1869 if (!old_name
|| !*old_name
)
1871 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1875 err
= lock_mount(path
);
1880 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1883 if (d_unlinked(path
->dentry
))
1887 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1890 old
= real_mount(old_path
.mnt
);
1892 if (!mnt_has_parent(old_path
.mnt
))
1895 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1896 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1899 * Don't move a mount residing in a shared parent.
1901 if (IS_MNT_SHARED(old_path
.mnt
->mnt_parent
))
1904 * Don't move a mount tree containing unbindable mounts to a destination
1905 * mount which is shared.
1907 if (IS_MNT_SHARED(path
->mnt
) &&
1908 tree_contains_unbindable(old
))
1911 for (p
= path
->mnt
; mnt_has_parent(p
); p
= p
->mnt_parent
)
1912 if (p
== old_path
.mnt
)
1915 err
= attach_recursive_mnt(old
, path
, &parent_path
);
1919 /* if the mount is moved, it should no longer be expire
1921 list_del_init(&old_path
.mnt
->mnt_expire
);
1926 path_put(&parent_path
);
1927 path_put(&old_path
);
1931 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
1934 const char *subtype
= strchr(fstype
, '.');
1943 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
1945 if (!mnt
->mnt_sb
->s_subtype
)
1951 return ERR_PTR(err
);
1954 static struct vfsmount
*
1955 do_kern_mount(const char *fstype
, int flags
, const char *name
, void *data
)
1957 struct file_system_type
*type
= get_fs_type(fstype
);
1958 struct vfsmount
*mnt
;
1960 return ERR_PTR(-ENODEV
);
1961 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
1962 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
1963 !mnt
->mnt_sb
->s_subtype
)
1964 mnt
= fs_set_subtype(mnt
, fstype
);
1965 put_filesystem(type
);
1970 * add a mount into a namespace's mount tree
1972 static int do_add_mount(struct vfsmount
*newmnt
, struct path
*path
, int mnt_flags
)
1976 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
);
1978 err
= lock_mount(path
);
1983 if (!(mnt_flags
& MNT_SHRINKABLE
) && !check_mnt(path
->mnt
))
1986 /* Refuse the same filesystem on the same mount point */
1988 if (path
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1989 path
->mnt
->mnt_root
== path
->dentry
)
1993 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1996 newmnt
->mnt_flags
= mnt_flags
;
1997 err
= graft_tree(newmnt
, path
);
2005 * create a new mount for userspace and request it to be added into the
2008 static int do_new_mount(struct path
*path
, char *type
, int flags
,
2009 int mnt_flags
, char *name
, void *data
)
2011 struct vfsmount
*mnt
;
2017 /* we need capabilities... */
2018 if (!capable(CAP_SYS_ADMIN
))
2021 mnt
= do_kern_mount(type
, flags
, name
, data
);
2023 return PTR_ERR(mnt
);
2025 err
= do_add_mount(mnt
, path
, mnt_flags
);
2031 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2034 /* The new mount record should have at least 2 refs to prevent it being
2035 * expired before we get a chance to add it
2037 BUG_ON(mnt_get_count(m
) < 2);
2039 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2040 m
->mnt_root
== path
->dentry
) {
2045 err
= do_add_mount(m
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2049 /* remove m from any expiration list it may be on */
2050 if (!list_empty(&m
->mnt_expire
)) {
2051 down_write(&namespace_sem
);
2052 br_write_lock(vfsmount_lock
);
2053 list_del_init(&m
->mnt_expire
);
2054 br_write_unlock(vfsmount_lock
);
2055 up_write(&namespace_sem
);
2063 * mnt_set_expiry - Put a mount on an expiration list
2064 * @mnt: The mount to list.
2065 * @expiry_list: The list to add the mount to.
2067 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2069 down_write(&namespace_sem
);
2070 br_write_lock(vfsmount_lock
);
2072 list_add_tail(&mnt
->mnt_expire
, expiry_list
);
2074 br_write_unlock(vfsmount_lock
);
2075 up_write(&namespace_sem
);
2077 EXPORT_SYMBOL(mnt_set_expiry
);
2080 * process a list of expirable mountpoints with the intent of discarding any
2081 * mountpoints that aren't in use and haven't been touched since last we came
2084 void mark_mounts_for_expiry(struct list_head
*mounts
)
2086 struct mount
*mnt
, *next
;
2087 LIST_HEAD(graveyard
);
2090 if (list_empty(mounts
))
2093 down_write(&namespace_sem
);
2094 br_write_lock(vfsmount_lock
);
2096 /* extract from the expiration list every vfsmount that matches the
2097 * following criteria:
2098 * - only referenced by its parent vfsmount
2099 * - still marked for expiry (marked on the last call here; marks are
2100 * cleared by mntput())
2102 list_for_each_entry_safe(mnt
, next
, mounts
, mnt
.mnt_expire
) {
2103 if (!xchg(&mnt
->mnt
.mnt_expiry_mark
, 1) ||
2104 propagate_mount_busy(&mnt
->mnt
, 1))
2106 list_move(&mnt
->mnt
.mnt_expire
, &graveyard
);
2108 while (!list_empty(&graveyard
)) {
2109 mnt
= list_first_entry(&graveyard
, struct mount
, mnt
.mnt_expire
);
2110 touch_mnt_namespace(mnt
->mnt
.mnt_ns
);
2111 umount_tree(mnt
, 1, &umounts
);
2113 br_write_unlock(vfsmount_lock
);
2114 up_write(&namespace_sem
);
2116 release_mounts(&umounts
);
2119 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2122 * Ripoff of 'select_parent()'
2124 * search the list of submounts for a given mountpoint, and move any
2125 * shrinkable submounts to the 'graveyard' list.
2127 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2129 struct mount
*this_parent
= parent
;
2130 struct list_head
*next
;
2134 next
= this_parent
->mnt
.mnt_mounts
.next
;
2136 while (next
!= &this_parent
->mnt
.mnt_mounts
) {
2137 struct list_head
*tmp
= next
;
2138 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt
.mnt_child
);
2141 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2144 * Descend a level if the d_mounts list is non-empty.
2146 if (!list_empty(&mnt
->mnt
.mnt_mounts
)) {
2151 if (!propagate_mount_busy(&mnt
->mnt
, 1)) {
2152 list_move_tail(&mnt
->mnt
.mnt_expire
, graveyard
);
2157 * All done at this level ... ascend and resume the search
2159 if (this_parent
!= parent
) {
2160 next
= this_parent
->mnt
.mnt_child
.next
;
2161 this_parent
= real_mount(this_parent
->mnt
.mnt_parent
);
2168 * process a list of expirable mountpoints with the intent of discarding any
2169 * submounts of a specific parent mountpoint
2171 * vfsmount_lock must be held for write
2173 static void shrink_submounts(struct mount
*mnt
, struct list_head
*umounts
)
2175 LIST_HEAD(graveyard
);
2178 /* extract submounts of 'mountpoint' from the expiration list */
2179 while (select_submounts(mnt
, &graveyard
)) {
2180 while (!list_empty(&graveyard
)) {
2181 m
= list_first_entry(&graveyard
, struct mount
,
2183 touch_mnt_namespace(m
->mnt
.mnt_ns
);
2184 umount_tree(m
, 1, umounts
);
2190 * Some copy_from_user() implementations do not return the exact number of
2191 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2192 * Note that this function differs from copy_from_user() in that it will oops
2193 * on bad values of `to', rather than returning a short copy.
2195 static long exact_copy_from_user(void *to
, const void __user
* from
,
2199 const char __user
*f
= from
;
2202 if (!access_ok(VERIFY_READ
, from
, n
))
2206 if (__get_user(c
, f
)) {
2217 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2227 if (!(page
= __get_free_page(GFP_KERNEL
)))
2230 /* We only care that *some* data at the address the user
2231 * gave us is valid. Just in case, we'll zero
2232 * the remainder of the page.
2234 /* copy_from_user cannot cross TASK_SIZE ! */
2235 size
= TASK_SIZE
- (unsigned long)data
;
2236 if (size
> PAGE_SIZE
)
2239 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2245 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2250 int copy_mount_string(const void __user
*data
, char **where
)
2259 tmp
= strndup_user(data
, PAGE_SIZE
);
2261 return PTR_ERR(tmp
);
2268 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2269 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2271 * data is a (void *) that can point to any structure up to
2272 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2273 * information (or be NULL).
2275 * Pre-0.97 versions of mount() didn't have a flags word.
2276 * When the flags word was introduced its top half was required
2277 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2278 * Therefore, if this magic number is present, it carries no information
2279 * and must be discarded.
2281 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
2282 unsigned long flags
, void *data_page
)
2289 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2290 flags
&= ~MS_MGC_MSK
;
2292 /* Basic sanity checks */
2294 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2298 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2300 /* ... and get the mountpoint */
2301 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2305 retval
= security_sb_mount(dev_name
, &path
,
2306 type_page
, flags
, data_page
);
2310 /* Default to relatime unless overriden */
2311 if (!(flags
& MS_NOATIME
))
2312 mnt_flags
|= MNT_RELATIME
;
2314 /* Separate the per-mountpoint flags */
2315 if (flags
& MS_NOSUID
)
2316 mnt_flags
|= MNT_NOSUID
;
2317 if (flags
& MS_NODEV
)
2318 mnt_flags
|= MNT_NODEV
;
2319 if (flags
& MS_NOEXEC
)
2320 mnt_flags
|= MNT_NOEXEC
;
2321 if (flags
& MS_NOATIME
)
2322 mnt_flags
|= MNT_NOATIME
;
2323 if (flags
& MS_NODIRATIME
)
2324 mnt_flags
|= MNT_NODIRATIME
;
2325 if (flags
& MS_STRICTATIME
)
2326 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2327 if (flags
& MS_RDONLY
)
2328 mnt_flags
|= MNT_READONLY
;
2330 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2331 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2334 if (flags
& MS_REMOUNT
)
2335 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2337 else if (flags
& MS_BIND
)
2338 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2339 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2340 retval
= do_change_type(&path
, flags
);
2341 else if (flags
& MS_MOVE
)
2342 retval
= do_move_mount(&path
, dev_name
);
2344 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2345 dev_name
, data_page
);
2351 static struct mnt_namespace
*alloc_mnt_ns(void)
2353 struct mnt_namespace
*new_ns
;
2355 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2357 return ERR_PTR(-ENOMEM
);
2358 atomic_set(&new_ns
->count
, 1);
2359 new_ns
->root
= NULL
;
2360 INIT_LIST_HEAD(&new_ns
->list
);
2361 init_waitqueue_head(&new_ns
->poll
);
2366 void mnt_make_longterm(struct vfsmount
*mnt
)
2368 __mnt_make_longterm(mnt
);
2371 void mnt_make_shortterm(struct vfsmount
*mnt
)
2374 if (atomic_add_unless(&mnt
->mnt_longterm
, -1, 1))
2376 br_write_lock(vfsmount_lock
);
2377 atomic_dec(&mnt
->mnt_longterm
);
2378 br_write_unlock(vfsmount_lock
);
2383 * Allocate a new namespace structure and populate it with contents
2384 * copied from the namespace of the passed in task structure.
2386 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
2387 struct fs_struct
*fs
)
2389 struct mnt_namespace
*new_ns
;
2390 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2391 struct mount
*p
, *q
;
2394 new_ns
= alloc_mnt_ns();
2398 down_write(&namespace_sem
);
2399 /* First pass: copy the tree topology */
2400 new = copy_tree(real_mount(mnt_ns
->root
), mnt_ns
->root
->mnt_root
,
2401 CL_COPY_ALL
| CL_EXPIRE
);
2403 up_write(&namespace_sem
);
2405 return ERR_PTR(-ENOMEM
);
2407 new_ns
->root
= &new->mnt
;
2408 br_write_lock(vfsmount_lock
);
2409 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2410 br_write_unlock(vfsmount_lock
);
2413 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2414 * as belonging to new namespace. We have already acquired a private
2415 * fs_struct, so tsk->fs->lock is not needed.
2417 p
= real_mount(mnt_ns
->root
);
2420 q
->mnt
.mnt_ns
= new_ns
;
2421 __mnt_make_longterm(&q
->mnt
);
2423 if (&p
->mnt
== fs
->root
.mnt
) {
2424 fs
->root
.mnt
= mntget(&q
->mnt
);
2425 __mnt_make_longterm(&q
->mnt
);
2426 mnt_make_shortterm(&p
->mnt
);
2429 if (&p
->mnt
== fs
->pwd
.mnt
) {
2430 fs
->pwd
.mnt
= mntget(&q
->mnt
);
2431 __mnt_make_longterm(&q
->mnt
);
2432 mnt_make_shortterm(&p
->mnt
);
2436 p
= next_mnt(p
, mnt_ns
->root
);
2437 q
= next_mnt(q
, new_ns
->root
);
2439 up_write(&namespace_sem
);
2449 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2450 struct fs_struct
*new_fs
)
2452 struct mnt_namespace
*new_ns
;
2457 if (!(flags
& CLONE_NEWNS
))
2460 new_ns
= dup_mnt_ns(ns
, new_fs
);
2467 * create_mnt_ns - creates a private namespace and adds a root filesystem
2468 * @mnt: pointer to the new root filesystem mountpoint
2470 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*mnt
)
2472 struct mnt_namespace
*new_ns
;
2474 new_ns
= alloc_mnt_ns();
2475 if (!IS_ERR(new_ns
)) {
2476 mnt
->mnt_ns
= new_ns
;
2477 __mnt_make_longterm(mnt
);
2479 list_add(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2486 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2488 struct mnt_namespace
*ns
;
2489 struct super_block
*s
;
2493 ns
= create_mnt_ns(mnt
);
2495 return ERR_CAST(ns
);
2497 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2498 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2503 return ERR_PTR(err
);
2505 /* trade a vfsmount reference for active sb one */
2506 s
= path
.mnt
->mnt_sb
;
2507 atomic_inc(&s
->s_active
);
2509 /* lock the sucker */
2510 down_write(&s
->s_umount
);
2511 /* ... and return the root of (sub)tree on it */
2514 EXPORT_SYMBOL(mount_subtree
);
2516 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2517 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2523 unsigned long data_page
;
2525 ret
= copy_mount_string(type
, &kernel_type
);
2529 kernel_dir
= getname(dir_name
);
2530 if (IS_ERR(kernel_dir
)) {
2531 ret
= PTR_ERR(kernel_dir
);
2535 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2539 ret
= copy_mount_options(data
, &data_page
);
2543 ret
= do_mount(kernel_dev
, kernel_dir
, kernel_type
, flags
,
2544 (void *) data_page
);
2546 free_page(data_page
);
2550 putname(kernel_dir
);
2558 * Return true if path is reachable from root
2560 * namespace_sem or vfsmount_lock is held
2562 bool is_path_reachable(struct vfsmount
*mnt
, struct dentry
*dentry
,
2563 const struct path
*root
)
2565 while (mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2566 dentry
= mnt
->mnt_mountpoint
;
2567 mnt
= mnt
->mnt_parent
;
2569 return mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2572 int path_is_under(struct path
*path1
, struct path
*path2
)
2575 br_read_lock(vfsmount_lock
);
2576 res
= is_path_reachable(path1
->mnt
, path1
->dentry
, path2
);
2577 br_read_unlock(vfsmount_lock
);
2580 EXPORT_SYMBOL(path_is_under
);
2583 * pivot_root Semantics:
2584 * Moves the root file system of the current process to the directory put_old,
2585 * makes new_root as the new root file system of the current process, and sets
2586 * root/cwd of all processes which had them on the current root to new_root.
2589 * The new_root and put_old must be directories, and must not be on the
2590 * same file system as the current process root. The put_old must be
2591 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2592 * pointed to by put_old must yield the same directory as new_root. No other
2593 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2595 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2596 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2597 * in this situation.
2600 * - we don't move root/cwd if they are not at the root (reason: if something
2601 * cared enough to change them, it's probably wrong to force them elsewhere)
2602 * - it's okay to pick a root that isn't the root of a file system, e.g.
2603 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2604 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2607 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2608 const char __user
*, put_old
)
2610 struct path
new, old
, parent_path
, root_parent
, root
;
2611 struct mount
*new_mnt
, *root_mnt
;
2614 if (!capable(CAP_SYS_ADMIN
))
2617 error
= user_path_dir(new_root
, &new);
2621 error
= user_path_dir(put_old
, &old
);
2625 error
= security_sb_pivotroot(&old
, &new);
2629 get_fs_root(current
->fs
, &root
);
2630 error
= lock_mount(&old
);
2635 new_mnt
= real_mount(new.mnt
);
2636 root_mnt
= real_mount(root
.mnt
);
2637 if (IS_MNT_SHARED(old
.mnt
) ||
2638 IS_MNT_SHARED(new.mnt
->mnt_parent
) ||
2639 IS_MNT_SHARED(root
.mnt
->mnt_parent
))
2641 if (!check_mnt(root
.mnt
) || !check_mnt(new.mnt
))
2644 if (d_unlinked(new.dentry
))
2646 if (d_unlinked(old
.dentry
))
2649 if (new.mnt
== root
.mnt
||
2650 old
.mnt
== root
.mnt
)
2651 goto out4
; /* loop, on the same file system */
2653 if (root
.mnt
->mnt_root
!= root
.dentry
)
2654 goto out4
; /* not a mountpoint */
2655 if (!mnt_has_parent(root
.mnt
))
2656 goto out4
; /* not attached */
2657 if (new.mnt
->mnt_root
!= new.dentry
)
2658 goto out4
; /* not a mountpoint */
2659 if (!mnt_has_parent(new.mnt
))
2660 goto out4
; /* not attached */
2661 /* make sure we can reach put_old from new_root */
2662 if (!is_path_reachable(old
.mnt
, old
.dentry
, &new))
2664 br_write_lock(vfsmount_lock
);
2665 detach_mnt(new_mnt
, &parent_path
);
2666 detach_mnt(root_mnt
, &root_parent
);
2667 /* mount old root on put_old */
2668 attach_mnt(root_mnt
, &old
);
2669 /* mount new_root on / */
2670 attach_mnt(new_mnt
, &root_parent
);
2671 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2672 br_write_unlock(vfsmount_lock
);
2673 chroot_fs_refs(&root
, &new);
2678 path_put(&root_parent
);
2679 path_put(&parent_path
);
2691 static void __init
init_mount_tree(void)
2693 struct vfsmount
*mnt
;
2694 struct mnt_namespace
*ns
;
2697 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
2699 panic("Can't create rootfs");
2701 ns
= create_mnt_ns(mnt
);
2703 panic("Can't allocate initial namespace");
2705 init_task
.nsproxy
->mnt_ns
= ns
;
2708 root
.mnt
= ns
->root
;
2709 root
.dentry
= ns
->root
->mnt_root
;
2711 set_fs_pwd(current
->fs
, &root
);
2712 set_fs_root(current
->fs
, &root
);
2715 void __init
mnt_init(void)
2720 init_rwsem(&namespace_sem
);
2722 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2723 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2725 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2727 if (!mount_hashtable
)
2728 panic("Failed to allocate mount hash table\n");
2730 printk(KERN_INFO
"Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2732 for (u
= 0; u
< HASH_SIZE
; u
++)
2733 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2735 br_lock_init(vfsmount_lock
);
2739 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2741 fs_kobj
= kobject_create_and_add("fs", NULL
);
2743 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2748 void put_mnt_ns(struct mnt_namespace
*ns
)
2750 LIST_HEAD(umount_list
);
2752 if (!atomic_dec_and_test(&ns
->count
))
2754 down_write(&namespace_sem
);
2755 br_write_lock(vfsmount_lock
);
2756 umount_tree(real_mount(ns
->root
), 0, &umount_list
);
2757 br_write_unlock(vfsmount_lock
);
2758 up_write(&namespace_sem
);
2759 release_mounts(&umount_list
);
2763 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2765 struct vfsmount
*mnt
;
2766 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2769 * it is a longterm mount, don't release mnt until
2770 * we unmount before file sys is unregistered
2772 mnt_make_longterm(mnt
);
2776 EXPORT_SYMBOL_GPL(kern_mount_data
);
2778 void kern_unmount(struct vfsmount
*mnt
)
2780 /* release long term mount so mount point can be released */
2781 if (!IS_ERR_OR_NULL(mnt
)) {
2782 mnt_make_shortterm(mnt
);
2786 EXPORT_SYMBOL(kern_unmount
);
2788 bool our_mnt(struct vfsmount
*mnt
)
2790 return check_mnt(mnt
);