4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
30 static unsigned int m_hash_mask __read_mostly
;
31 static unsigned int m_hash_shift __read_mostly
;
32 static unsigned int mp_hash_mask __read_mostly
;
33 static unsigned int mp_hash_shift __read_mostly
;
35 static __initdata
unsigned long mhash_entries
;
36 static int __init
set_mhash_entries(char *str
)
40 mhash_entries
= simple_strtoul(str
, &str
, 0);
43 __setup("mhash_entries=", set_mhash_entries
);
45 static __initdata
unsigned long mphash_entries
;
46 static int __init
set_mphash_entries(char *str
)
50 mphash_entries
= simple_strtoul(str
, &str
, 0);
53 __setup("mphash_entries=", set_mphash_entries
);
56 static DEFINE_IDA(mnt_id_ida
);
57 static DEFINE_IDA(mnt_group_ida
);
58 static DEFINE_SPINLOCK(mnt_id_lock
);
59 static int mnt_id_start
= 0;
60 static int mnt_group_start
= 1;
62 static struct list_head
*mount_hashtable __read_mostly
;
63 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
64 static struct kmem_cache
*mnt_cache __read_mostly
;
65 static DECLARE_RWSEM(namespace_sem
);
68 struct kobject
*fs_kobj
;
69 EXPORT_SYMBOL_GPL(fs_kobj
);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
81 static inline struct list_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
83 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
84 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
85 tmp
= tmp
+ (tmp
>> m_hash_shift
);
86 return &mount_hashtable
[tmp
& m_hash_mask
];
89 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
91 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
92 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
93 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount
*mnt
)
105 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
106 spin_lock(&mnt_id_lock
);
107 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
109 mnt_id_start
= mnt
->mnt_id
+ 1;
110 spin_unlock(&mnt_id_lock
);
117 static void mnt_free_id(struct mount
*mnt
)
119 int id
= mnt
->mnt_id
;
120 spin_lock(&mnt_id_lock
);
121 ida_remove(&mnt_id_ida
, id
);
122 if (mnt_id_start
> id
)
124 spin_unlock(&mnt_id_lock
);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount
*mnt
)
136 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
139 res
= ida_get_new_above(&mnt_group_ida
,
143 mnt_group_start
= mnt
->mnt_group_id
+ 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount
*mnt
)
153 int id
= mnt
->mnt_group_id
;
154 ida_remove(&mnt_group_ida
, id
);
155 if (mnt_group_start
> id
)
156 mnt_group_start
= id
;
157 mnt
->mnt_group_id
= 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount
*mnt
, int n
)
166 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount
*mnt
)
180 unsigned int count
= 0;
183 for_each_possible_cpu(cpu
) {
184 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
189 return mnt
->mnt_count
;
193 static struct mount
*alloc_vfsmnt(const char *name
)
195 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
199 err
= mnt_alloc_id(mnt
);
204 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
205 if (!mnt
->mnt_devname
)
210 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
212 goto out_free_devname
;
214 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
217 mnt
->mnt_writers
= 0;
220 INIT_LIST_HEAD(&mnt
->mnt_hash
);
221 INIT_LIST_HEAD(&mnt
->mnt_child
);
222 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
223 INIT_LIST_HEAD(&mnt
->mnt_list
);
224 INIT_LIST_HEAD(&mnt
->mnt_expire
);
225 INIT_LIST_HEAD(&mnt
->mnt_share
);
226 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
227 INIT_LIST_HEAD(&mnt
->mnt_slave
);
228 #ifdef CONFIG_FSNOTIFY
229 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
236 kfree(mnt
->mnt_devname
);
241 kmem_cache_free(mnt_cache
, mnt
);
246 * Most r/o checks on a fs are for operations that take
247 * discrete amounts of time, like a write() or unlink().
248 * We must keep track of when those operations start
249 * (for permission checks) and when they end, so that
250 * we can determine when writes are able to occur to
254 * __mnt_is_readonly: check whether a mount is read-only
255 * @mnt: the mount to check for its write status
257 * This shouldn't be used directly ouside of the VFS.
258 * It does not guarantee that the filesystem will stay
259 * r/w, just that it is right *now*. This can not and
260 * should not be used in place of IS_RDONLY(inode).
261 * mnt_want/drop_write() will _keep_ the filesystem
264 int __mnt_is_readonly(struct vfsmount
*mnt
)
266 if (mnt
->mnt_flags
& MNT_READONLY
)
268 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
272 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
274 static inline void mnt_inc_writers(struct mount
*mnt
)
277 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
283 static inline void mnt_dec_writers(struct mount
*mnt
)
286 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
292 static unsigned int mnt_get_writers(struct mount
*mnt
)
295 unsigned int count
= 0;
298 for_each_possible_cpu(cpu
) {
299 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
304 return mnt
->mnt_writers
;
308 static int mnt_is_readonly(struct vfsmount
*mnt
)
310 if (mnt
->mnt_sb
->s_readonly_remount
)
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
314 return __mnt_is_readonly(mnt
);
318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
324 * __mnt_want_write - get write access to a mount without freeze protection
325 * @m: the mount on which to take a write
327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
333 int __mnt_want_write(struct vfsmount
*m
)
335 struct mount
*mnt
= real_mount(m
);
339 mnt_inc_writers(mnt
);
341 * The store to mnt_inc_writers must be visible before we pass
342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
346 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
354 if (mnt_is_readonly(m
)) {
355 mnt_dec_writers(mnt
);
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
372 int mnt_want_write(struct vfsmount
*m
)
376 sb_start_write(m
->mnt_sb
);
377 ret
= __mnt_want_write(m
);
379 sb_end_write(m
->mnt_sb
);
382 EXPORT_SYMBOL_GPL(mnt_want_write
);
385 * mnt_clone_write - get write access to a mount
386 * @mnt: the mount on which to take a write
388 * This is effectively like mnt_want_write, except
389 * it must only be used to take an extra write reference
390 * on a mountpoint that we already know has a write reference
391 * on it. This allows some optimisation.
393 * After finished, mnt_drop_write must be called as usual to
394 * drop the reference.
396 int mnt_clone_write(struct vfsmount
*mnt
)
398 /* superblock may be r/o */
399 if (__mnt_is_readonly(mnt
))
402 mnt_inc_writers(real_mount(mnt
));
406 EXPORT_SYMBOL_GPL(mnt_clone_write
);
409 * __mnt_want_write_file - get write access to a file's mount
410 * @file: the file who's mount on which to take a write
412 * This is like __mnt_want_write, but it takes a file and can
413 * do some optimisations if the file is open for write already
415 int __mnt_want_write_file(struct file
*file
)
417 struct inode
*inode
= file_inode(file
);
419 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
420 return __mnt_want_write(file
->f_path
.mnt
);
422 return mnt_clone_write(file
->f_path
.mnt
);
426 * mnt_want_write_file - get write access to a file's mount
427 * @file: the file who's mount on which to take a write
429 * This is like mnt_want_write, but it takes a file and can
430 * do some optimisations if the file is open for write already
432 int mnt_want_write_file(struct file
*file
)
436 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
437 ret
= __mnt_want_write_file(file
);
439 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
442 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
445 * __mnt_drop_write - give up write access to a mount
446 * @mnt: the mount on which to give up write access
448 * Tells the low-level filesystem that we are done
449 * performing writes to it. Must be matched with
450 * __mnt_want_write() call above.
452 void __mnt_drop_write(struct vfsmount
*mnt
)
455 mnt_dec_writers(real_mount(mnt
));
460 * mnt_drop_write - give up write access to a mount
461 * @mnt: the mount on which to give up write access
463 * Tells the low-level filesystem that we are done performing writes to it and
464 * also allows filesystem to be frozen again. Must be matched with
465 * mnt_want_write() call above.
467 void mnt_drop_write(struct vfsmount
*mnt
)
469 __mnt_drop_write(mnt
);
470 sb_end_write(mnt
->mnt_sb
);
472 EXPORT_SYMBOL_GPL(mnt_drop_write
);
474 void __mnt_drop_write_file(struct file
*file
)
476 __mnt_drop_write(file
->f_path
.mnt
);
479 void mnt_drop_write_file(struct file
*file
)
481 mnt_drop_write(file
->f_path
.mnt
);
483 EXPORT_SYMBOL(mnt_drop_write_file
);
485 static int mnt_make_readonly(struct mount
*mnt
)
490 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
492 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
493 * should be visible before we do.
498 * With writers on hold, if this value is zero, then there are
499 * definitely no active writers (although held writers may subsequently
500 * increment the count, they'll have to wait, and decrement it after
501 * seeing MNT_READONLY).
503 * It is OK to have counter incremented on one CPU and decremented on
504 * another: the sum will add up correctly. The danger would be when we
505 * sum up each counter, if we read a counter before it is incremented,
506 * but then read another CPU's count which it has been subsequently
507 * decremented from -- we would see more decrements than we should.
508 * MNT_WRITE_HOLD protects against this scenario, because
509 * mnt_want_write first increments count, then smp_mb, then spins on
510 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
511 * we're counting up here.
513 if (mnt_get_writers(mnt
) > 0)
516 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
518 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
519 * that become unheld will see MNT_READONLY.
522 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
527 static void __mnt_unmake_readonly(struct mount
*mnt
)
530 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
534 int sb_prepare_remount_readonly(struct super_block
*sb
)
539 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
540 if (atomic_long_read(&sb
->s_remove_count
))
544 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
545 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
546 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
548 if (mnt_get_writers(mnt
) > 0) {
554 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
558 sb
->s_readonly_remount
= 1;
561 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
562 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
563 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
570 static void free_vfsmnt(struct mount
*mnt
)
572 kfree(mnt
->mnt_devname
);
575 free_percpu(mnt
->mnt_pcp
);
577 kmem_cache_free(mnt_cache
, mnt
);
580 /* call under rcu_read_lock */
581 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
584 if (read_seqretry(&mount_lock
, seq
))
588 mnt
= real_mount(bastard
);
589 mnt_add_count(mnt
, 1);
590 if (likely(!read_seqretry(&mount_lock
, seq
)))
592 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
593 mnt_add_count(mnt
, -1);
603 * find the first mount at @dentry on vfsmount @mnt.
604 * call under rcu_read_lock()
606 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
608 struct list_head
*head
= m_hash(mnt
, dentry
);
611 list_for_each_entry_rcu(p
, head
, mnt_hash
)
612 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
618 * find the last mount at @dentry on vfsmount @mnt.
619 * mount_lock must be held.
621 struct mount
*__lookup_mnt_last(struct vfsmount
*mnt
, struct dentry
*dentry
)
623 struct list_head
*head
= m_hash(mnt
, dentry
);
624 struct mount
*p
, *res
= NULL
;
626 list_for_each_entry(p
, head
, mnt_hash
)
627 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
632 list_for_each_entry_continue(p
, head
, mnt_hash
) {
633 if (&p
->mnt_parent
->mnt
!= mnt
|| p
->mnt_mountpoint
!= dentry
)
641 * lookup_mnt - Return the first child mount mounted at path
643 * "First" means first mounted chronologically. If you create the
646 * mount /dev/sda1 /mnt
647 * mount /dev/sda2 /mnt
648 * mount /dev/sda3 /mnt
650 * Then lookup_mnt() on the base /mnt dentry in the root mount will
651 * return successively the root dentry and vfsmount of /dev/sda1, then
652 * /dev/sda2, then /dev/sda3, then NULL.
654 * lookup_mnt takes a reference to the found vfsmount.
656 struct vfsmount
*lookup_mnt(struct path
*path
)
658 struct mount
*child_mnt
;
664 seq
= read_seqbegin(&mount_lock
);
665 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
666 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
667 } while (!legitimize_mnt(m
, seq
));
672 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
674 struct hlist_head
*chain
= mp_hash(dentry
);
675 struct mountpoint
*mp
;
678 hlist_for_each_entry(mp
, chain
, m_hash
) {
679 if (mp
->m_dentry
== dentry
) {
680 /* might be worth a WARN_ON() */
681 if (d_unlinked(dentry
))
682 return ERR_PTR(-ENOENT
);
688 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
690 return ERR_PTR(-ENOMEM
);
692 ret
= d_set_mounted(dentry
);
698 mp
->m_dentry
= dentry
;
700 hlist_add_head(&mp
->m_hash
, chain
);
704 static void put_mountpoint(struct mountpoint
*mp
)
706 if (!--mp
->m_count
) {
707 struct dentry
*dentry
= mp
->m_dentry
;
708 spin_lock(&dentry
->d_lock
);
709 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
710 spin_unlock(&dentry
->d_lock
);
711 hlist_del(&mp
->m_hash
);
716 static inline int check_mnt(struct mount
*mnt
)
718 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
722 * vfsmount lock must be held for write
724 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
728 wake_up_interruptible(&ns
->poll
);
733 * vfsmount lock must be held for write
735 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
737 if (ns
&& ns
->event
!= event
) {
739 wake_up_interruptible(&ns
->poll
);
744 * vfsmount lock must be held for write
746 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
748 old_path
->dentry
= mnt
->mnt_mountpoint
;
749 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
750 mnt
->mnt_parent
= mnt
;
751 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
752 list_del_init(&mnt
->mnt_child
);
753 list_del_init(&mnt
->mnt_hash
);
754 put_mountpoint(mnt
->mnt_mp
);
759 * vfsmount lock must be held for write
761 void mnt_set_mountpoint(struct mount
*mnt
,
762 struct mountpoint
*mp
,
763 struct mount
*child_mnt
)
766 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
767 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
768 child_mnt
->mnt_parent
= mnt
;
769 child_mnt
->mnt_mp
= mp
;
773 * vfsmount lock must be held for write
775 static void attach_mnt(struct mount
*mnt
,
776 struct mount
*parent
,
777 struct mountpoint
*mp
)
779 mnt_set_mountpoint(parent
, mp
, mnt
);
780 list_add(&mnt
->mnt_hash
, m_hash(&parent
->mnt
, mp
->m_dentry
));
781 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
785 * vfsmount lock must be held for write
787 static void commit_tree(struct mount
*mnt
, struct mount
*shadows
)
789 struct mount
*parent
= mnt
->mnt_parent
;
792 struct mnt_namespace
*n
= parent
->mnt_ns
;
794 BUG_ON(parent
== mnt
);
796 list_add_tail(&head
, &mnt
->mnt_list
);
797 list_for_each_entry(m
, &head
, mnt_list
)
800 list_splice(&head
, n
->list
.prev
);
803 list_add(&mnt
->mnt_hash
, &shadows
->mnt_hash
);
805 list_add(&mnt
->mnt_hash
,
806 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
807 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
808 touch_mnt_namespace(n
);
811 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
813 struct list_head
*next
= p
->mnt_mounts
.next
;
814 if (next
== &p
->mnt_mounts
) {
818 next
= p
->mnt_child
.next
;
819 if (next
!= &p
->mnt_parent
->mnt_mounts
)
824 return list_entry(next
, struct mount
, mnt_child
);
827 static struct mount
*skip_mnt_tree(struct mount
*p
)
829 struct list_head
*prev
= p
->mnt_mounts
.prev
;
830 while (prev
!= &p
->mnt_mounts
) {
831 p
= list_entry(prev
, struct mount
, mnt_child
);
832 prev
= p
->mnt_mounts
.prev
;
838 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
844 return ERR_PTR(-ENODEV
);
846 mnt
= alloc_vfsmnt(name
);
848 return ERR_PTR(-ENOMEM
);
850 if (flags
& MS_KERNMOUNT
)
851 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
853 root
= mount_fs(type
, flags
, name
, data
);
856 return ERR_CAST(root
);
859 mnt
->mnt
.mnt_root
= root
;
860 mnt
->mnt
.mnt_sb
= root
->d_sb
;
861 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
862 mnt
->mnt_parent
= mnt
;
864 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
868 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
870 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
873 struct super_block
*sb
= old
->mnt
.mnt_sb
;
877 mnt
= alloc_vfsmnt(old
->mnt_devname
);
879 return ERR_PTR(-ENOMEM
);
881 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
882 mnt
->mnt_group_id
= 0; /* not a peer of original */
884 mnt
->mnt_group_id
= old
->mnt_group_id
;
886 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
887 err
= mnt_alloc_group_id(mnt
);
892 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~MNT_WRITE_HOLD
;
893 /* Don't allow unprivileged users to change mount flags */
894 if ((flag
& CL_UNPRIVILEGED
) && (mnt
->mnt
.mnt_flags
& MNT_READONLY
))
895 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
897 /* Don't allow unprivileged users to reveal what is under a mount */
898 if ((flag
& CL_UNPRIVILEGED
) && list_empty(&old
->mnt_expire
))
899 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
901 atomic_inc(&sb
->s_active
);
902 mnt
->mnt
.mnt_sb
= sb
;
903 mnt
->mnt
.mnt_root
= dget(root
);
904 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
905 mnt
->mnt_parent
= mnt
;
907 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
910 if ((flag
& CL_SLAVE
) ||
911 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
912 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
913 mnt
->mnt_master
= old
;
914 CLEAR_MNT_SHARED(mnt
);
915 } else if (!(flag
& CL_PRIVATE
)) {
916 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
917 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
918 if (IS_MNT_SLAVE(old
))
919 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
920 mnt
->mnt_master
= old
->mnt_master
;
922 if (flag
& CL_MAKE_SHARED
)
925 /* stick the duplicate mount on the same expiry list
926 * as the original if that was on one */
927 if (flag
& CL_EXPIRE
) {
928 if (!list_empty(&old
->mnt_expire
))
929 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
939 static void delayed_free(struct rcu_head
*head
)
941 struct mount
*mnt
= container_of(head
, struct mount
, mnt_rcu
);
942 kfree(mnt
->mnt_devname
);
944 free_percpu(mnt
->mnt_pcp
);
946 kmem_cache_free(mnt_cache
, mnt
);
949 static void mntput_no_expire(struct mount
*mnt
)
953 mnt_add_count(mnt
, -1);
954 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
959 if (mnt_get_count(mnt
)) {
964 if (unlikely(mnt
->mnt_pinned
)) {
965 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
969 acct_auto_close_mnt(&mnt
->mnt
);
972 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
977 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
980 list_del(&mnt
->mnt_instance
);
984 * This probably indicates that somebody messed
985 * up a mnt_want/drop_write() pair. If this
986 * happens, the filesystem was probably unable
987 * to make r/w->r/o transitions.
990 * The locking used to deal with mnt_count decrement provides barriers,
991 * so mnt_get_writers() below is safe.
993 WARN_ON(mnt_get_writers(mnt
));
994 fsnotify_vfsmount_delete(&mnt
->mnt
);
995 dput(mnt
->mnt
.mnt_root
);
996 deactivate_super(mnt
->mnt
.mnt_sb
);
998 call_rcu(&mnt
->mnt_rcu
, delayed_free
);
1001 void mntput(struct vfsmount
*mnt
)
1004 struct mount
*m
= real_mount(mnt
);
1005 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1006 if (unlikely(m
->mnt_expiry_mark
))
1007 m
->mnt_expiry_mark
= 0;
1008 mntput_no_expire(m
);
1011 EXPORT_SYMBOL(mntput
);
1013 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1016 mnt_add_count(real_mount(mnt
), 1);
1019 EXPORT_SYMBOL(mntget
);
1021 void mnt_pin(struct vfsmount
*mnt
)
1024 real_mount(mnt
)->mnt_pinned
++;
1025 unlock_mount_hash();
1027 EXPORT_SYMBOL(mnt_pin
);
1029 void mnt_unpin(struct vfsmount
*m
)
1031 struct mount
*mnt
= real_mount(m
);
1033 if (mnt
->mnt_pinned
) {
1034 mnt_add_count(mnt
, 1);
1037 unlock_mount_hash();
1039 EXPORT_SYMBOL(mnt_unpin
);
1041 static inline void mangle(struct seq_file
*m
, const char *s
)
1043 seq_escape(m
, s
, " \t\n\\");
1047 * Simple .show_options callback for filesystems which don't want to
1048 * implement more complex mount option showing.
1050 * See also save_mount_options().
1052 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1054 const char *options
;
1057 options
= rcu_dereference(root
->d_sb
->s_options
);
1059 if (options
!= NULL
&& options
[0]) {
1067 EXPORT_SYMBOL(generic_show_options
);
1070 * If filesystem uses generic_show_options(), this function should be
1071 * called from the fill_super() callback.
1073 * The .remount_fs callback usually needs to be handled in a special
1074 * way, to make sure, that previous options are not overwritten if the
1077 * Also note, that if the filesystem's .remount_fs function doesn't
1078 * reset all options to their default value, but changes only newly
1079 * given options, then the displayed options will not reflect reality
1082 void save_mount_options(struct super_block
*sb
, char *options
)
1084 BUG_ON(sb
->s_options
);
1085 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1087 EXPORT_SYMBOL(save_mount_options
);
1089 void replace_mount_options(struct super_block
*sb
, char *options
)
1091 char *old
= sb
->s_options
;
1092 rcu_assign_pointer(sb
->s_options
, options
);
1098 EXPORT_SYMBOL(replace_mount_options
);
1100 #ifdef CONFIG_PROC_FS
1101 /* iterator; we want it to have access to namespace_sem, thus here... */
1102 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1104 struct proc_mounts
*p
= proc_mounts(m
);
1106 down_read(&namespace_sem
);
1107 return seq_list_start(&p
->ns
->list
, *pos
);
1110 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1112 struct proc_mounts
*p
= proc_mounts(m
);
1114 return seq_list_next(v
, &p
->ns
->list
, pos
);
1117 static void m_stop(struct seq_file
*m
, void *v
)
1119 up_read(&namespace_sem
);
1122 static int m_show(struct seq_file
*m
, void *v
)
1124 struct proc_mounts
*p
= proc_mounts(m
);
1125 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1126 return p
->show(m
, &r
->mnt
);
1129 const struct seq_operations mounts_op
= {
1135 #endif /* CONFIG_PROC_FS */
1138 * may_umount_tree - check if a mount tree is busy
1139 * @mnt: root of mount tree
1141 * This is called to check if a tree of mounts has any
1142 * open files, pwds, chroots or sub mounts that are
1145 int may_umount_tree(struct vfsmount
*m
)
1147 struct mount
*mnt
= real_mount(m
);
1148 int actual_refs
= 0;
1149 int minimum_refs
= 0;
1153 /* write lock needed for mnt_get_count */
1155 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1156 actual_refs
+= mnt_get_count(p
);
1159 unlock_mount_hash();
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
);
1187 if (propagate_mount_busy(real_mount(mnt
), 2))
1189 unlock_mount_hash();
1190 up_read(&namespace_sem
);
1194 EXPORT_SYMBOL(may_umount
);
1196 static LIST_HEAD(unmounted
); /* protected by namespace_sem */
1198 static void namespace_unlock(void)
1203 if (likely(list_empty(&unmounted
))) {
1204 up_write(&namespace_sem
);
1208 list_splice_init(&unmounted
, &head
);
1209 up_write(&namespace_sem
);
1213 while (!list_empty(&head
)) {
1214 mnt
= list_first_entry(&head
, struct mount
, mnt_hash
);
1215 list_del_init(&mnt
->mnt_hash
);
1216 if (mnt
->mnt_ex_mountpoint
.mnt
)
1217 path_put(&mnt
->mnt_ex_mountpoint
);
1222 static inline void namespace_lock(void)
1224 down_write(&namespace_sem
);
1228 * mount_lock must be held
1229 * namespace_sem must be held for write
1230 * how = 0 => just this tree, don't propagate
1231 * how = 1 => propagate; we know that nobody else has reference to any victims
1232 * how = 2 => lazy umount
1234 void umount_tree(struct mount
*mnt
, int how
)
1236 LIST_HEAD(tmp_list
);
1239 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1240 list_move(&p
->mnt_hash
, &tmp_list
);
1243 propagate_umount(&tmp_list
);
1245 list_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1246 list_del_init(&p
->mnt_expire
);
1247 list_del_init(&p
->mnt_list
);
1248 __touch_mnt_namespace(p
->mnt_ns
);
1251 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1252 list_del_init(&p
->mnt_child
);
1253 if (mnt_has_parent(p
)) {
1254 put_mountpoint(p
->mnt_mp
);
1255 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1256 p
->mnt_ex_mountpoint
.dentry
= p
->mnt_mountpoint
;
1257 p
->mnt_ex_mountpoint
.mnt
= &p
->mnt_parent
->mnt
;
1258 p
->mnt_mountpoint
= p
->mnt
.mnt_root
;
1262 change_mnt_propagation(p
, MS_PRIVATE
);
1264 list_splice(&tmp_list
, &unmounted
);
1267 static void shrink_submounts(struct mount
*mnt
);
1269 static int do_umount(struct mount
*mnt
, int flags
)
1271 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1274 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1279 * Allow userspace to request a mountpoint be expired rather than
1280 * unmounting unconditionally. Unmount only happens if:
1281 * (1) the mark is already set (the mark is cleared by mntput())
1282 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1284 if (flags
& MNT_EXPIRE
) {
1285 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1286 flags
& (MNT_FORCE
| MNT_DETACH
))
1290 * probably don't strictly need the lock here if we examined
1291 * all race cases, but it's a slowpath.
1294 if (mnt_get_count(mnt
) != 2) {
1295 unlock_mount_hash();
1298 unlock_mount_hash();
1300 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1305 * If we may have to abort operations to get out of this
1306 * mount, and they will themselves hold resources we must
1307 * allow the fs to do things. In the Unix tradition of
1308 * 'Gee thats tricky lets do it in userspace' the umount_begin
1309 * might fail to complete on the first run through as other tasks
1310 * must return, and the like. Thats for the mount program to worry
1311 * about for the moment.
1314 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1315 sb
->s_op
->umount_begin(sb
);
1319 * No sense to grab the lock for this test, but test itself looks
1320 * somewhat bogus. Suggestions for better replacement?
1321 * Ho-hum... In principle, we might treat that as umount + switch
1322 * to rootfs. GC would eventually take care of the old vfsmount.
1323 * Actually it makes sense, especially if rootfs would contain a
1324 * /reboot - static binary that would close all descriptors and
1325 * call reboot(9). Then init(8) could umount root and exec /reboot.
1327 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1329 * Special case for "unmounting" root ...
1330 * we just try to remount it readonly.
1332 down_write(&sb
->s_umount
);
1333 if (!(sb
->s_flags
& MS_RDONLY
))
1334 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1335 up_write(&sb
->s_umount
);
1343 if (flags
& MNT_DETACH
) {
1344 if (!list_empty(&mnt
->mnt_list
))
1345 umount_tree(mnt
, 2);
1348 shrink_submounts(mnt
);
1350 if (!propagate_mount_busy(mnt
, 2)) {
1351 if (!list_empty(&mnt
->mnt_list
))
1352 umount_tree(mnt
, 1);
1356 unlock_mount_hash();
1362 * Is the caller allowed to modify his namespace?
1364 static inline bool may_mount(void)
1366 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1370 * Now umount can handle mount points as well as block devices.
1371 * This is important for filesystems which use unnamed block devices.
1373 * We now support a flag for forced unmount like the other 'big iron'
1374 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1377 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1382 int lookup_flags
= 0;
1384 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1390 if (!(flags
& UMOUNT_NOFOLLOW
))
1391 lookup_flags
|= LOOKUP_FOLLOW
;
1393 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1396 mnt
= real_mount(path
.mnt
);
1398 if (path
.dentry
!= path
.mnt
->mnt_root
)
1400 if (!check_mnt(mnt
))
1402 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1405 retval
= do_umount(mnt
, flags
);
1407 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1409 mntput_no_expire(mnt
);
1414 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1417 * The 2.0 compatible umount. No flags.
1419 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1421 return sys_umount(name
, 0);
1426 static bool is_mnt_ns_file(struct dentry
*dentry
)
1428 /* Is this a proxy for a mount namespace? */
1429 struct inode
*inode
= dentry
->d_inode
;
1432 if (!proc_ns_inode(inode
))
1435 ei
= get_proc_ns(inode
);
1436 if (ei
->ns_ops
!= &mntns_operations
)
1442 static bool mnt_ns_loop(struct dentry
*dentry
)
1444 /* Could bind mounting the mount namespace inode cause a
1445 * mount namespace loop?
1447 struct mnt_namespace
*mnt_ns
;
1448 if (!is_mnt_ns_file(dentry
))
1451 mnt_ns
= get_proc_ns(dentry
->d_inode
)->ns
;
1452 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1455 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1458 struct mount
*res
, *p
, *q
, *r
, *parent
;
1460 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1461 return ERR_PTR(-EINVAL
);
1463 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1464 return ERR_PTR(-EINVAL
);
1466 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1470 q
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1471 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1474 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1476 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1479 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1480 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1481 IS_MNT_UNBINDABLE(s
)) {
1482 s
= skip_mnt_tree(s
);
1485 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1486 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1487 s
= skip_mnt_tree(s
);
1490 while (p
!= s
->mnt_parent
) {
1496 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1500 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1501 attach_mnt(q
, parent
, p
->mnt_mp
);
1502 unlock_mount_hash();
1509 umount_tree(res
, 0);
1510 unlock_mount_hash();
1515 /* Caller should check returned pointer for errors */
1517 struct vfsmount
*collect_mounts(struct path
*path
)
1521 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1522 CL_COPY_ALL
| CL_PRIVATE
);
1525 return ERR_CAST(tree
);
1529 void drop_collected_mounts(struct vfsmount
*mnt
)
1533 umount_tree(real_mount(mnt
), 0);
1534 unlock_mount_hash();
1538 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1539 struct vfsmount
*root
)
1542 int res
= f(root
, arg
);
1545 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1546 res
= f(&mnt
->mnt
, arg
);
1553 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1557 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1558 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1559 mnt_release_group_id(p
);
1563 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1567 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1568 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1569 int err
= mnt_alloc_group_id(p
);
1571 cleanup_group_ids(mnt
, p
);
1581 * @source_mnt : mount tree to be attached
1582 * @nd : place the mount tree @source_mnt is attached
1583 * @parent_nd : if non-null, detach the source_mnt from its parent and
1584 * store the parent mount and mountpoint dentry.
1585 * (done when source_mnt is moved)
1587 * NOTE: in the table below explains the semantics when a source mount
1588 * of a given type is attached to a destination mount of a given type.
1589 * ---------------------------------------------------------------------------
1590 * | BIND MOUNT OPERATION |
1591 * |**************************************************************************
1592 * | source-->| shared | private | slave | unbindable |
1596 * |**************************************************************************
1597 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1599 * |non-shared| shared (+) | private | slave (*) | invalid |
1600 * ***************************************************************************
1601 * A bind operation clones the source mount and mounts the clone on the
1602 * destination mount.
1604 * (++) the cloned mount is propagated to all the mounts in the propagation
1605 * tree of the destination mount and the cloned mount is added to
1606 * the peer group of the source mount.
1607 * (+) the cloned mount is created under the destination mount and is marked
1608 * as shared. The cloned mount is added to the peer group of the source
1610 * (+++) the mount is propagated to all the mounts in the propagation tree
1611 * of the destination mount and the cloned mount is made slave
1612 * of the same master as that of the source mount. The cloned mount
1613 * is marked as 'shared and slave'.
1614 * (*) the cloned mount is made a slave of the same master as that of the
1617 * ---------------------------------------------------------------------------
1618 * | MOVE MOUNT OPERATION |
1619 * |**************************************************************************
1620 * | source-->| shared | private | slave | unbindable |
1624 * |**************************************************************************
1625 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1627 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1628 * ***************************************************************************
1630 * (+) the mount is moved to the destination. And is then propagated to
1631 * all the mounts in the propagation tree of the destination mount.
1632 * (+*) the mount is moved to the destination.
1633 * (+++) the mount is moved to the destination and is then propagated to
1634 * all the mounts belonging to the destination mount's propagation tree.
1635 * the mount is marked as 'shared and slave'.
1636 * (*) the mount continues to be a slave at the new location.
1638 * if the source mount is a tree, the operations explained above is
1639 * applied to each mount in the tree.
1640 * Must be called without spinlocks held, since this function can sleep
1643 static int attach_recursive_mnt(struct mount
*source_mnt
,
1644 struct mount
*dest_mnt
,
1645 struct mountpoint
*dest_mp
,
1646 struct path
*parent_path
)
1648 LIST_HEAD(tree_list
);
1649 struct mount
*child
, *p
;
1652 if (IS_MNT_SHARED(dest_mnt
)) {
1653 err
= invent_group_ids(source_mnt
, true);
1657 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1659 goto out_cleanup_ids
;
1663 if (IS_MNT_SHARED(dest_mnt
)) {
1664 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1668 detach_mnt(source_mnt
, parent_path
);
1669 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1670 touch_mnt_namespace(source_mnt
->mnt_ns
);
1672 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1673 commit_tree(source_mnt
, NULL
);
1676 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1678 list_del_init(&child
->mnt_hash
);
1679 q
= __lookup_mnt_last(&child
->mnt_parent
->mnt
,
1680 child
->mnt_mountpoint
);
1681 commit_tree(child
, q
);
1683 unlock_mount_hash();
1688 if (IS_MNT_SHARED(dest_mnt
))
1689 cleanup_group_ids(source_mnt
, NULL
);
1694 static struct mountpoint
*lock_mount(struct path
*path
)
1696 struct vfsmount
*mnt
;
1697 struct dentry
*dentry
= path
->dentry
;
1699 mutex_lock(&dentry
->d_inode
->i_mutex
);
1700 if (unlikely(cant_mount(dentry
))) {
1701 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1702 return ERR_PTR(-ENOENT
);
1705 mnt
= lookup_mnt(path
);
1707 struct mountpoint
*mp
= new_mountpoint(dentry
);
1710 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1716 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1719 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1723 static void unlock_mount(struct mountpoint
*where
)
1725 struct dentry
*dentry
= where
->m_dentry
;
1726 put_mountpoint(where
);
1728 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1731 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1733 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1736 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1737 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1740 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1744 * Sanity check the flags to change_mnt_propagation.
1747 static int flags_to_propagation_type(int flags
)
1749 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1751 /* Fail if any non-propagation flags are set */
1752 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1754 /* Only one propagation flag should be set */
1755 if (!is_power_of_2(type
))
1761 * recursively change the type of the mountpoint.
1763 static int do_change_type(struct path
*path
, int flag
)
1766 struct mount
*mnt
= real_mount(path
->mnt
);
1767 int recurse
= flag
& MS_REC
;
1771 if (path
->dentry
!= path
->mnt
->mnt_root
)
1774 type
= flags_to_propagation_type(flag
);
1779 if (type
== MS_SHARED
) {
1780 err
= invent_group_ids(mnt
, recurse
);
1786 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1787 change_mnt_propagation(m
, type
);
1788 unlock_mount_hash();
1795 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
1797 struct mount
*child
;
1798 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
1799 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
1802 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
1809 * do loopback mount.
1811 static int do_loopback(struct path
*path
, const char *old_name
,
1814 struct path old_path
;
1815 struct mount
*mnt
= NULL
, *old
, *parent
;
1816 struct mountpoint
*mp
;
1818 if (!old_name
|| !*old_name
)
1820 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1825 if (mnt_ns_loop(old_path
.dentry
))
1828 mp
= lock_mount(path
);
1833 old
= real_mount(old_path
.mnt
);
1834 parent
= real_mount(path
->mnt
);
1837 if (IS_MNT_UNBINDABLE(old
))
1840 if (!check_mnt(parent
) || !check_mnt(old
))
1843 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
1847 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
1849 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1856 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1858 err
= graft_tree(mnt
, parent
, mp
);
1861 umount_tree(mnt
, 0);
1862 unlock_mount_hash();
1867 path_put(&old_path
);
1871 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1874 int readonly_request
= 0;
1876 if (ms_flags
& MS_RDONLY
)
1877 readonly_request
= 1;
1878 if (readonly_request
== __mnt_is_readonly(mnt
))
1881 if (mnt
->mnt_flags
& MNT_LOCK_READONLY
)
1884 if (readonly_request
)
1885 error
= mnt_make_readonly(real_mount(mnt
));
1887 __mnt_unmake_readonly(real_mount(mnt
));
1892 * change filesystem flags. dir should be a physical root of filesystem.
1893 * If you've mounted a non-root directory somewhere and want to do remount
1894 * on it - tough luck.
1896 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1900 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1901 struct mount
*mnt
= real_mount(path
->mnt
);
1903 if (!check_mnt(mnt
))
1906 if (path
->dentry
!= path
->mnt
->mnt_root
)
1909 err
= security_sb_remount(sb
, data
);
1913 down_write(&sb
->s_umount
);
1914 if (flags
& MS_BIND
)
1915 err
= change_mount_flags(path
->mnt
, flags
);
1916 else if (!capable(CAP_SYS_ADMIN
))
1919 err
= do_remount_sb(sb
, flags
, data
, 0);
1922 mnt_flags
|= mnt
->mnt
.mnt_flags
& MNT_PROPAGATION_MASK
;
1923 mnt
->mnt
.mnt_flags
= mnt_flags
;
1924 touch_mnt_namespace(mnt
->mnt_ns
);
1925 unlock_mount_hash();
1927 up_write(&sb
->s_umount
);
1931 static inline int tree_contains_unbindable(struct mount
*mnt
)
1934 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1935 if (IS_MNT_UNBINDABLE(p
))
1941 static int do_move_mount(struct path
*path
, const char *old_name
)
1943 struct path old_path
, parent_path
;
1946 struct mountpoint
*mp
;
1948 if (!old_name
|| !*old_name
)
1950 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1954 mp
= lock_mount(path
);
1959 old
= real_mount(old_path
.mnt
);
1960 p
= real_mount(path
->mnt
);
1963 if (!check_mnt(p
) || !check_mnt(old
))
1966 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
1970 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1973 if (!mnt_has_parent(old
))
1976 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1977 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1980 * Don't move a mount residing in a shared parent.
1982 if (IS_MNT_SHARED(old
->mnt_parent
))
1985 * Don't move a mount tree containing unbindable mounts to a destination
1986 * mount which is shared.
1988 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
1991 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
1995 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
1999 /* if the mount is moved, it should no longer be expire
2001 list_del_init(&old
->mnt_expire
);
2006 path_put(&parent_path
);
2007 path_put(&old_path
);
2011 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2014 const char *subtype
= strchr(fstype
, '.');
2023 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2025 if (!mnt
->mnt_sb
->s_subtype
)
2031 return ERR_PTR(err
);
2035 * add a mount into a namespace's mount tree
2037 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2039 struct mountpoint
*mp
;
2040 struct mount
*parent
;
2043 mnt_flags
&= ~(MNT_SHARED
| MNT_WRITE_HOLD
| MNT_INTERNAL
| MNT_DOOMED
| MNT_SYNC_UMOUNT
);
2045 mp
= lock_mount(path
);
2049 parent
= real_mount(path
->mnt
);
2051 if (unlikely(!check_mnt(parent
))) {
2052 /* that's acceptable only for automounts done in private ns */
2053 if (!(mnt_flags
& MNT_SHRINKABLE
))
2055 /* ... and for those we'd better have mountpoint still alive */
2056 if (!parent
->mnt_ns
)
2060 /* Refuse the same filesystem on the same mount point */
2062 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2063 path
->mnt
->mnt_root
== path
->dentry
)
2067 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
2070 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2071 err
= graft_tree(newmnt
, parent
, mp
);
2079 * create a new mount for userspace and request it to be added into the
2082 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2083 int mnt_flags
, const char *name
, void *data
)
2085 struct file_system_type
*type
;
2086 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2087 struct vfsmount
*mnt
;
2093 type
= get_fs_type(fstype
);
2097 if (user_ns
!= &init_user_ns
) {
2098 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2099 put_filesystem(type
);
2102 /* Only in special cases allow devices from mounts
2103 * created outside the initial user namespace.
2105 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2107 mnt_flags
|= MNT_NODEV
;
2111 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2112 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2113 !mnt
->mnt_sb
->s_subtype
)
2114 mnt
= fs_set_subtype(mnt
, fstype
);
2116 put_filesystem(type
);
2118 return PTR_ERR(mnt
);
2120 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2126 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2128 struct mount
*mnt
= real_mount(m
);
2130 /* The new mount record should have at least 2 refs to prevent it being
2131 * expired before we get a chance to add it
2133 BUG_ON(mnt_get_count(mnt
) < 2);
2135 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2136 m
->mnt_root
== path
->dentry
) {
2141 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2145 /* remove m from any expiration list it may be on */
2146 if (!list_empty(&mnt
->mnt_expire
)) {
2148 list_del_init(&mnt
->mnt_expire
);
2157 * mnt_set_expiry - Put a mount on an expiration list
2158 * @mnt: The mount to list.
2159 * @expiry_list: The list to add the mount to.
2161 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2165 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2169 EXPORT_SYMBOL(mnt_set_expiry
);
2172 * process a list of expirable mountpoints with the intent of discarding any
2173 * mountpoints that aren't in use and haven't been touched since last we came
2176 void mark_mounts_for_expiry(struct list_head
*mounts
)
2178 struct mount
*mnt
, *next
;
2179 LIST_HEAD(graveyard
);
2181 if (list_empty(mounts
))
2187 /* extract from the expiration list every vfsmount that matches the
2188 * following criteria:
2189 * - only referenced by its parent vfsmount
2190 * - still marked for expiry (marked on the last call here; marks are
2191 * cleared by mntput())
2193 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2194 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2195 propagate_mount_busy(mnt
, 1))
2197 list_move(&mnt
->mnt_expire
, &graveyard
);
2199 while (!list_empty(&graveyard
)) {
2200 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2201 touch_mnt_namespace(mnt
->mnt_ns
);
2202 umount_tree(mnt
, 1);
2204 unlock_mount_hash();
2208 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2211 * Ripoff of 'select_parent()'
2213 * search the list of submounts for a given mountpoint, and move any
2214 * shrinkable submounts to the 'graveyard' list.
2216 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2218 struct mount
*this_parent
= parent
;
2219 struct list_head
*next
;
2223 next
= this_parent
->mnt_mounts
.next
;
2225 while (next
!= &this_parent
->mnt_mounts
) {
2226 struct list_head
*tmp
= next
;
2227 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2230 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2233 * Descend a level if the d_mounts list is non-empty.
2235 if (!list_empty(&mnt
->mnt_mounts
)) {
2240 if (!propagate_mount_busy(mnt
, 1)) {
2241 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2246 * All done at this level ... ascend and resume the search
2248 if (this_parent
!= parent
) {
2249 next
= this_parent
->mnt_child
.next
;
2250 this_parent
= this_parent
->mnt_parent
;
2257 * process a list of expirable mountpoints with the intent of discarding any
2258 * submounts of a specific parent mountpoint
2260 * mount_lock must be held for write
2262 static void shrink_submounts(struct mount
*mnt
)
2264 LIST_HEAD(graveyard
);
2267 /* extract submounts of 'mountpoint' from the expiration list */
2268 while (select_submounts(mnt
, &graveyard
)) {
2269 while (!list_empty(&graveyard
)) {
2270 m
= list_first_entry(&graveyard
, struct mount
,
2272 touch_mnt_namespace(m
->mnt_ns
);
2279 * Some copy_from_user() implementations do not return the exact number of
2280 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2281 * Note that this function differs from copy_from_user() in that it will oops
2282 * on bad values of `to', rather than returning a short copy.
2284 static long exact_copy_from_user(void *to
, const void __user
* from
,
2288 const char __user
*f
= from
;
2291 if (!access_ok(VERIFY_READ
, from
, n
))
2295 if (__get_user(c
, f
)) {
2306 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2316 if (!(page
= __get_free_page(GFP_KERNEL
)))
2319 /* We only care that *some* data at the address the user
2320 * gave us is valid. Just in case, we'll zero
2321 * the remainder of the page.
2323 /* copy_from_user cannot cross TASK_SIZE ! */
2324 size
= TASK_SIZE
- (unsigned long)data
;
2325 if (size
> PAGE_SIZE
)
2328 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2334 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2339 int copy_mount_string(const void __user
*data
, char **where
)
2348 tmp
= strndup_user(data
, PAGE_SIZE
);
2350 return PTR_ERR(tmp
);
2357 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2358 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2360 * data is a (void *) that can point to any structure up to
2361 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2362 * information (or be NULL).
2364 * Pre-0.97 versions of mount() didn't have a flags word.
2365 * When the flags word was introduced its top half was required
2366 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2367 * Therefore, if this magic number is present, it carries no information
2368 * and must be discarded.
2370 long do_mount(const char *dev_name
, const char *dir_name
,
2371 const char *type_page
, unsigned long flags
, void *data_page
)
2378 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2379 flags
&= ~MS_MGC_MSK
;
2381 /* Basic sanity checks */
2383 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2387 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2389 /* ... and get the mountpoint */
2390 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2394 retval
= security_sb_mount(dev_name
, &path
,
2395 type_page
, flags
, data_page
);
2396 if (!retval
&& !may_mount())
2401 /* Default to relatime unless overriden */
2402 if (!(flags
& MS_NOATIME
))
2403 mnt_flags
|= MNT_RELATIME
;
2405 /* Separate the per-mountpoint flags */
2406 if (flags
& MS_NOSUID
)
2407 mnt_flags
|= MNT_NOSUID
;
2408 if (flags
& MS_NODEV
)
2409 mnt_flags
|= MNT_NODEV
;
2410 if (flags
& MS_NOEXEC
)
2411 mnt_flags
|= MNT_NOEXEC
;
2412 if (flags
& MS_NOATIME
)
2413 mnt_flags
|= MNT_NOATIME
;
2414 if (flags
& MS_NODIRATIME
)
2415 mnt_flags
|= MNT_NODIRATIME
;
2416 if (flags
& MS_STRICTATIME
)
2417 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2418 if (flags
& MS_RDONLY
)
2419 mnt_flags
|= MNT_READONLY
;
2421 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2422 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2425 if (flags
& MS_REMOUNT
)
2426 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2428 else if (flags
& MS_BIND
)
2429 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2430 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2431 retval
= do_change_type(&path
, flags
);
2432 else if (flags
& MS_MOVE
)
2433 retval
= do_move_mount(&path
, dev_name
);
2435 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2436 dev_name
, data_page
);
2442 static void free_mnt_ns(struct mnt_namespace
*ns
)
2444 proc_free_inum(ns
->proc_inum
);
2445 put_user_ns(ns
->user_ns
);
2450 * Assign a sequence number so we can detect when we attempt to bind
2451 * mount a reference to an older mount namespace into the current
2452 * mount namespace, preventing reference counting loops. A 64bit
2453 * number incrementing at 10Ghz will take 12,427 years to wrap which
2454 * is effectively never, so we can ignore the possibility.
2456 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2458 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2460 struct mnt_namespace
*new_ns
;
2463 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2465 return ERR_PTR(-ENOMEM
);
2466 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2469 return ERR_PTR(ret
);
2471 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2472 atomic_set(&new_ns
->count
, 1);
2473 new_ns
->root
= NULL
;
2474 INIT_LIST_HEAD(&new_ns
->list
);
2475 init_waitqueue_head(&new_ns
->poll
);
2477 new_ns
->user_ns
= get_user_ns(user_ns
);
2481 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2482 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2484 struct mnt_namespace
*new_ns
;
2485 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2486 struct mount
*p
, *q
;
2493 if (likely(!(flags
& CLONE_NEWNS
))) {
2500 new_ns
= alloc_mnt_ns(user_ns
);
2505 /* First pass: copy the tree topology */
2506 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2507 if (user_ns
!= ns
->user_ns
)
2508 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2509 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2512 free_mnt_ns(new_ns
);
2513 return ERR_CAST(new);
2516 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2519 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2520 * as belonging to new namespace. We have already acquired a private
2521 * fs_struct, so tsk->fs->lock is not needed.
2528 if (&p
->mnt
== new_fs
->root
.mnt
) {
2529 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2532 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2533 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2537 p
= next_mnt(p
, old
);
2538 q
= next_mnt(q
, new);
2541 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2542 p
= next_mnt(p
, old
);
2555 * create_mnt_ns - creates a private namespace and adds a root filesystem
2556 * @mnt: pointer to the new root filesystem mountpoint
2558 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2560 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2561 if (!IS_ERR(new_ns
)) {
2562 struct mount
*mnt
= real_mount(m
);
2563 mnt
->mnt_ns
= new_ns
;
2565 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2572 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2574 struct mnt_namespace
*ns
;
2575 struct super_block
*s
;
2579 ns
= create_mnt_ns(mnt
);
2581 return ERR_CAST(ns
);
2583 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2584 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2589 return ERR_PTR(err
);
2591 /* trade a vfsmount reference for active sb one */
2592 s
= path
.mnt
->mnt_sb
;
2593 atomic_inc(&s
->s_active
);
2595 /* lock the sucker */
2596 down_write(&s
->s_umount
);
2597 /* ... and return the root of (sub)tree on it */
2600 EXPORT_SYMBOL(mount_subtree
);
2602 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2603 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2607 struct filename
*kernel_dir
;
2609 unsigned long data_page
;
2611 ret
= copy_mount_string(type
, &kernel_type
);
2615 kernel_dir
= getname(dir_name
);
2616 if (IS_ERR(kernel_dir
)) {
2617 ret
= PTR_ERR(kernel_dir
);
2621 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2625 ret
= copy_mount_options(data
, &data_page
);
2629 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2630 (void *) data_page
);
2632 free_page(data_page
);
2636 putname(kernel_dir
);
2644 * Return true if path is reachable from root
2646 * namespace_sem or mount_lock is held
2648 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2649 const struct path
*root
)
2651 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2652 dentry
= mnt
->mnt_mountpoint
;
2653 mnt
= mnt
->mnt_parent
;
2655 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2658 int path_is_under(struct path
*path1
, struct path
*path2
)
2661 read_seqlock_excl(&mount_lock
);
2662 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2663 read_sequnlock_excl(&mount_lock
);
2666 EXPORT_SYMBOL(path_is_under
);
2669 * pivot_root Semantics:
2670 * Moves the root file system of the current process to the directory put_old,
2671 * makes new_root as the new root file system of the current process, and sets
2672 * root/cwd of all processes which had them on the current root to new_root.
2675 * The new_root and put_old must be directories, and must not be on the
2676 * same file system as the current process root. The put_old must be
2677 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2678 * pointed to by put_old must yield the same directory as new_root. No other
2679 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2681 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2682 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2683 * in this situation.
2686 * - we don't move root/cwd if they are not at the root (reason: if something
2687 * cared enough to change them, it's probably wrong to force them elsewhere)
2688 * - it's okay to pick a root that isn't the root of a file system, e.g.
2689 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2690 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2693 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2694 const char __user
*, put_old
)
2696 struct path
new, old
, parent_path
, root_parent
, root
;
2697 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2698 struct mountpoint
*old_mp
, *root_mp
;
2704 error
= user_path_dir(new_root
, &new);
2708 error
= user_path_dir(put_old
, &old
);
2712 error
= security_sb_pivotroot(&old
, &new);
2716 get_fs_root(current
->fs
, &root
);
2717 old_mp
= lock_mount(&old
);
2718 error
= PTR_ERR(old_mp
);
2723 new_mnt
= real_mount(new.mnt
);
2724 root_mnt
= real_mount(root
.mnt
);
2725 old_mnt
= real_mount(old
.mnt
);
2726 if (IS_MNT_SHARED(old_mnt
) ||
2727 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2728 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2730 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2732 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
2735 if (d_unlinked(new.dentry
))
2738 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2739 goto out4
; /* loop, on the same file system */
2741 if (root
.mnt
->mnt_root
!= root
.dentry
)
2742 goto out4
; /* not a mountpoint */
2743 if (!mnt_has_parent(root_mnt
))
2744 goto out4
; /* not attached */
2745 root_mp
= root_mnt
->mnt_mp
;
2746 if (new.mnt
->mnt_root
!= new.dentry
)
2747 goto out4
; /* not a mountpoint */
2748 if (!mnt_has_parent(new_mnt
))
2749 goto out4
; /* not attached */
2750 /* make sure we can reach put_old from new_root */
2751 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2753 root_mp
->m_count
++; /* pin it so it won't go away */
2755 detach_mnt(new_mnt
, &parent_path
);
2756 detach_mnt(root_mnt
, &root_parent
);
2757 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
2758 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
2759 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2761 /* mount old root on put_old */
2762 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2763 /* mount new_root on / */
2764 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2765 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2766 unlock_mount_hash();
2767 chroot_fs_refs(&root
, &new);
2768 put_mountpoint(root_mp
);
2771 unlock_mount(old_mp
);
2773 path_put(&root_parent
);
2774 path_put(&parent_path
);
2786 static void __init
init_mount_tree(void)
2788 struct vfsmount
*mnt
;
2789 struct mnt_namespace
*ns
;
2791 struct file_system_type
*type
;
2793 type
= get_fs_type("rootfs");
2795 panic("Can't find rootfs type");
2796 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2797 put_filesystem(type
);
2799 panic("Can't create rootfs");
2801 ns
= create_mnt_ns(mnt
);
2803 panic("Can't allocate initial namespace");
2805 init_task
.nsproxy
->mnt_ns
= ns
;
2809 root
.dentry
= mnt
->mnt_root
;
2811 set_fs_pwd(current
->fs
, &root
);
2812 set_fs_root(current
->fs
, &root
);
2815 void __init
mnt_init(void)
2820 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2821 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2823 mount_hashtable
= alloc_large_system_hash("Mount-cache",
2824 sizeof(struct list_head
),
2827 &m_hash_shift
, &m_hash_mask
, 0, 0);
2828 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
2829 sizeof(struct hlist_head
),
2832 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
2834 if (!mount_hashtable
|| !mountpoint_hashtable
)
2835 panic("Failed to allocate mount hash table\n");
2837 for (u
= 0; u
<= m_hash_mask
; u
++)
2838 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2839 for (u
= 0; u
<= mp_hash_mask
; u
++)
2840 INIT_HLIST_HEAD(&mountpoint_hashtable
[u
]);
2846 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2848 fs_kobj
= kobject_create_and_add("fs", NULL
);
2850 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2855 void put_mnt_ns(struct mnt_namespace
*ns
)
2857 if (!atomic_dec_and_test(&ns
->count
))
2859 drop_collected_mounts(&ns
->root
->mnt
);
2863 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2865 struct vfsmount
*mnt
;
2866 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2869 * it is a longterm mount, don't release mnt until
2870 * we unmount before file sys is unregistered
2872 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2876 EXPORT_SYMBOL_GPL(kern_mount_data
);
2878 void kern_unmount(struct vfsmount
*mnt
)
2880 /* release long term mount so mount point can be released */
2881 if (!IS_ERR_OR_NULL(mnt
)) {
2882 real_mount(mnt
)->mnt_ns
= NULL
;
2883 synchronize_rcu(); /* yecchhh... */
2887 EXPORT_SYMBOL(kern_unmount
);
2889 bool our_mnt(struct vfsmount
*mnt
)
2891 return check_mnt(real_mount(mnt
));
2894 bool current_chrooted(void)
2896 /* Does the current process have a non-standard root */
2897 struct path ns_root
;
2898 struct path fs_root
;
2901 /* Find the namespace root */
2902 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
2903 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
2905 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
2908 get_fs_root(current
->fs
, &fs_root
);
2910 chrooted
= !path_equal(&fs_root
, &ns_root
);
2918 bool fs_fully_visible(struct file_system_type
*type
)
2920 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
2922 bool visible
= false;
2927 down_read(&namespace_sem
);
2928 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
2929 struct mount
*child
;
2930 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
2933 /* This mount is not fully visible if there are any child mounts
2934 * that cover anything except for empty directories.
2936 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2937 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
2938 if (!S_ISDIR(inode
->i_mode
))
2940 if (inode
->i_nlink
> 2)
2948 up_read(&namespace_sem
);
2952 static void *mntns_get(struct task_struct
*task
)
2954 struct mnt_namespace
*ns
= NULL
;
2955 struct nsproxy
*nsproxy
;
2958 nsproxy
= task_nsproxy(task
);
2960 ns
= nsproxy
->mnt_ns
;
2968 static void mntns_put(void *ns
)
2973 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
2975 struct fs_struct
*fs
= current
->fs
;
2976 struct mnt_namespace
*mnt_ns
= ns
;
2979 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
2980 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
2981 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
2988 put_mnt_ns(nsproxy
->mnt_ns
);
2989 nsproxy
->mnt_ns
= mnt_ns
;
2992 root
.mnt
= &mnt_ns
->root
->mnt
;
2993 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
2995 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
2998 /* Update the pwd and root */
2999 set_fs_pwd(fs
, &root
);
3000 set_fs_root(fs
, &root
);
3006 static unsigned int mntns_inum(void *ns
)
3008 struct mnt_namespace
*mnt_ns
= ns
;
3009 return mnt_ns
->proc_inum
;
3012 const struct proc_ns_operations mntns_operations
= {
3014 .type
= CLONE_NEWNS
,
3017 .install
= mntns_install
,