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 hlist_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 hlist_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_HLIST_NODE(&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 hlist_head
*head
= m_hash(mnt
, dentry
);
611 hlist_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 mount
*p
, *res
;
624 res
= p
= __lookup_mnt(mnt
, dentry
);
627 hlist_for_each_entry_continue(p
, mnt_hash
) {
628 if (&p
->mnt_parent
->mnt
!= mnt
|| p
->mnt_mountpoint
!= dentry
)
637 * lookup_mnt - Return the first child mount mounted at path
639 * "First" means first mounted chronologically. If you create the
642 * mount /dev/sda1 /mnt
643 * mount /dev/sda2 /mnt
644 * mount /dev/sda3 /mnt
646 * Then lookup_mnt() on the base /mnt dentry in the root mount will
647 * return successively the root dentry and vfsmount of /dev/sda1, then
648 * /dev/sda2, then /dev/sda3, then NULL.
650 * lookup_mnt takes a reference to the found vfsmount.
652 struct vfsmount
*lookup_mnt(struct path
*path
)
654 struct mount
*child_mnt
;
660 seq
= read_seqbegin(&mount_lock
);
661 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
662 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
663 } while (!legitimize_mnt(m
, seq
));
668 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
670 struct hlist_head
*chain
= mp_hash(dentry
);
671 struct mountpoint
*mp
;
674 hlist_for_each_entry(mp
, chain
, m_hash
) {
675 if (mp
->m_dentry
== dentry
) {
676 /* might be worth a WARN_ON() */
677 if (d_unlinked(dentry
))
678 return ERR_PTR(-ENOENT
);
684 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
686 return ERR_PTR(-ENOMEM
);
688 ret
= d_set_mounted(dentry
);
694 mp
->m_dentry
= dentry
;
696 hlist_add_head(&mp
->m_hash
, chain
);
700 static void put_mountpoint(struct mountpoint
*mp
)
702 if (!--mp
->m_count
) {
703 struct dentry
*dentry
= mp
->m_dentry
;
704 spin_lock(&dentry
->d_lock
);
705 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
706 spin_unlock(&dentry
->d_lock
);
707 hlist_del(&mp
->m_hash
);
712 static inline int check_mnt(struct mount
*mnt
)
714 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
718 * vfsmount lock must be held for write
720 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
724 wake_up_interruptible(&ns
->poll
);
729 * vfsmount lock must be held for write
731 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
733 if (ns
&& ns
->event
!= event
) {
735 wake_up_interruptible(&ns
->poll
);
740 * vfsmount lock must be held for write
742 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
744 old_path
->dentry
= mnt
->mnt_mountpoint
;
745 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
746 mnt
->mnt_parent
= mnt
;
747 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
748 list_del_init(&mnt
->mnt_child
);
749 hlist_del_init_rcu(&mnt
->mnt_hash
);
750 put_mountpoint(mnt
->mnt_mp
);
755 * vfsmount lock must be held for write
757 void mnt_set_mountpoint(struct mount
*mnt
,
758 struct mountpoint
*mp
,
759 struct mount
*child_mnt
)
762 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
763 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
764 child_mnt
->mnt_parent
= mnt
;
765 child_mnt
->mnt_mp
= mp
;
769 * vfsmount lock must be held for write
771 static void attach_mnt(struct mount
*mnt
,
772 struct mount
*parent
,
773 struct mountpoint
*mp
)
775 mnt_set_mountpoint(parent
, mp
, mnt
);
776 hlist_add_head_rcu(&mnt
->mnt_hash
, m_hash(&parent
->mnt
, mp
->m_dentry
));
777 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
781 * vfsmount lock must be held for write
783 static void commit_tree(struct mount
*mnt
, struct mount
*shadows
)
785 struct mount
*parent
= mnt
->mnt_parent
;
788 struct mnt_namespace
*n
= parent
->mnt_ns
;
790 BUG_ON(parent
== mnt
);
792 list_add_tail(&head
, &mnt
->mnt_list
);
793 list_for_each_entry(m
, &head
, mnt_list
)
796 list_splice(&head
, n
->list
.prev
);
799 hlist_add_after_rcu(&shadows
->mnt_hash
, &mnt
->mnt_hash
);
801 hlist_add_head_rcu(&mnt
->mnt_hash
,
802 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
803 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
804 touch_mnt_namespace(n
);
807 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
809 struct list_head
*next
= p
->mnt_mounts
.next
;
810 if (next
== &p
->mnt_mounts
) {
814 next
= p
->mnt_child
.next
;
815 if (next
!= &p
->mnt_parent
->mnt_mounts
)
820 return list_entry(next
, struct mount
, mnt_child
);
823 static struct mount
*skip_mnt_tree(struct mount
*p
)
825 struct list_head
*prev
= p
->mnt_mounts
.prev
;
826 while (prev
!= &p
->mnt_mounts
) {
827 p
= list_entry(prev
, struct mount
, mnt_child
);
828 prev
= p
->mnt_mounts
.prev
;
834 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
840 return ERR_PTR(-ENODEV
);
842 mnt
= alloc_vfsmnt(name
);
844 return ERR_PTR(-ENOMEM
);
846 if (flags
& MS_KERNMOUNT
)
847 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
849 root
= mount_fs(type
, flags
, name
, data
);
852 return ERR_CAST(root
);
855 mnt
->mnt
.mnt_root
= root
;
856 mnt
->mnt
.mnt_sb
= root
->d_sb
;
857 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
858 mnt
->mnt_parent
= mnt
;
860 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
864 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
866 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
869 struct super_block
*sb
= old
->mnt
.mnt_sb
;
873 mnt
= alloc_vfsmnt(old
->mnt_devname
);
875 return ERR_PTR(-ENOMEM
);
877 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
878 mnt
->mnt_group_id
= 0; /* not a peer of original */
880 mnt
->mnt_group_id
= old
->mnt_group_id
;
882 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
883 err
= mnt_alloc_group_id(mnt
);
888 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~(MNT_WRITE_HOLD
|MNT_MARKED
);
889 /* Don't allow unprivileged users to change mount flags */
890 if ((flag
& CL_UNPRIVILEGED
) && (mnt
->mnt
.mnt_flags
& MNT_READONLY
))
891 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
893 /* Don't allow unprivileged users to reveal what is under a mount */
894 if ((flag
& CL_UNPRIVILEGED
) && list_empty(&old
->mnt_expire
))
895 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
897 atomic_inc(&sb
->s_active
);
898 mnt
->mnt
.mnt_sb
= sb
;
899 mnt
->mnt
.mnt_root
= dget(root
);
900 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
901 mnt
->mnt_parent
= mnt
;
903 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
906 if ((flag
& CL_SLAVE
) ||
907 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
908 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
909 mnt
->mnt_master
= old
;
910 CLEAR_MNT_SHARED(mnt
);
911 } else if (!(flag
& CL_PRIVATE
)) {
912 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
913 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
914 if (IS_MNT_SLAVE(old
))
915 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
916 mnt
->mnt_master
= old
->mnt_master
;
918 if (flag
& CL_MAKE_SHARED
)
921 /* stick the duplicate mount on the same expiry list
922 * as the original if that was on one */
923 if (flag
& CL_EXPIRE
) {
924 if (!list_empty(&old
->mnt_expire
))
925 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
935 static void delayed_free(struct rcu_head
*head
)
937 struct mount
*mnt
= container_of(head
, struct mount
, mnt_rcu
);
938 kfree(mnt
->mnt_devname
);
940 free_percpu(mnt
->mnt_pcp
);
942 kmem_cache_free(mnt_cache
, mnt
);
945 static void mntput_no_expire(struct mount
*mnt
)
949 mnt_add_count(mnt
, -1);
950 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
955 if (mnt_get_count(mnt
)) {
960 if (unlikely(mnt
->mnt_pinned
)) {
961 mnt_add_count(mnt
, mnt
->mnt_pinned
+ 1);
965 acct_auto_close_mnt(&mnt
->mnt
);
968 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
973 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
976 list_del(&mnt
->mnt_instance
);
980 * This probably indicates that somebody messed
981 * up a mnt_want/drop_write() pair. If this
982 * happens, the filesystem was probably unable
983 * to make r/w->r/o transitions.
986 * The locking used to deal with mnt_count decrement provides barriers,
987 * so mnt_get_writers() below is safe.
989 WARN_ON(mnt_get_writers(mnt
));
990 fsnotify_vfsmount_delete(&mnt
->mnt
);
991 dput(mnt
->mnt
.mnt_root
);
992 deactivate_super(mnt
->mnt
.mnt_sb
);
994 call_rcu(&mnt
->mnt_rcu
, delayed_free
);
997 void mntput(struct vfsmount
*mnt
)
1000 struct mount
*m
= real_mount(mnt
);
1001 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1002 if (unlikely(m
->mnt_expiry_mark
))
1003 m
->mnt_expiry_mark
= 0;
1004 mntput_no_expire(m
);
1007 EXPORT_SYMBOL(mntput
);
1009 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1012 mnt_add_count(real_mount(mnt
), 1);
1015 EXPORT_SYMBOL(mntget
);
1017 void mnt_pin(struct vfsmount
*mnt
)
1020 real_mount(mnt
)->mnt_pinned
++;
1021 unlock_mount_hash();
1023 EXPORT_SYMBOL(mnt_pin
);
1025 void mnt_unpin(struct vfsmount
*m
)
1027 struct mount
*mnt
= real_mount(m
);
1029 if (mnt
->mnt_pinned
) {
1030 mnt_add_count(mnt
, 1);
1033 unlock_mount_hash();
1035 EXPORT_SYMBOL(mnt_unpin
);
1037 static inline void mangle(struct seq_file
*m
, const char *s
)
1039 seq_escape(m
, s
, " \t\n\\");
1043 * Simple .show_options callback for filesystems which don't want to
1044 * implement more complex mount option showing.
1046 * See also save_mount_options().
1048 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1050 const char *options
;
1053 options
= rcu_dereference(root
->d_sb
->s_options
);
1055 if (options
!= NULL
&& options
[0]) {
1063 EXPORT_SYMBOL(generic_show_options
);
1066 * If filesystem uses generic_show_options(), this function should be
1067 * called from the fill_super() callback.
1069 * The .remount_fs callback usually needs to be handled in a special
1070 * way, to make sure, that previous options are not overwritten if the
1073 * Also note, that if the filesystem's .remount_fs function doesn't
1074 * reset all options to their default value, but changes only newly
1075 * given options, then the displayed options will not reflect reality
1078 void save_mount_options(struct super_block
*sb
, char *options
)
1080 BUG_ON(sb
->s_options
);
1081 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1083 EXPORT_SYMBOL(save_mount_options
);
1085 void replace_mount_options(struct super_block
*sb
, char *options
)
1087 char *old
= sb
->s_options
;
1088 rcu_assign_pointer(sb
->s_options
, options
);
1094 EXPORT_SYMBOL(replace_mount_options
);
1096 #ifdef CONFIG_PROC_FS
1097 /* iterator; we want it to have access to namespace_sem, thus here... */
1098 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1100 struct proc_mounts
*p
= proc_mounts(m
);
1102 down_read(&namespace_sem
);
1103 if (p
->cached_event
== p
->ns
->event
) {
1104 void *v
= p
->cached_mount
;
1105 if (*pos
== p
->cached_index
)
1107 if (*pos
== p
->cached_index
+ 1) {
1108 v
= seq_list_next(v
, &p
->ns
->list
, &p
->cached_index
);
1109 return p
->cached_mount
= v
;
1113 p
->cached_event
= p
->ns
->event
;
1114 p
->cached_mount
= seq_list_start(&p
->ns
->list
, *pos
);
1115 p
->cached_index
= *pos
;
1116 return p
->cached_mount
;
1119 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1121 struct proc_mounts
*p
= proc_mounts(m
);
1123 p
->cached_mount
= seq_list_next(v
, &p
->ns
->list
, pos
);
1124 p
->cached_index
= *pos
;
1125 return p
->cached_mount
;
1128 static void m_stop(struct seq_file
*m
, void *v
)
1130 up_read(&namespace_sem
);
1133 static int m_show(struct seq_file
*m
, void *v
)
1135 struct proc_mounts
*p
= proc_mounts(m
);
1136 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1137 return p
->show(m
, &r
->mnt
);
1140 const struct seq_operations mounts_op
= {
1146 #endif /* CONFIG_PROC_FS */
1149 * may_umount_tree - check if a mount tree is busy
1150 * @mnt: root of mount tree
1152 * This is called to check if a tree of mounts has any
1153 * open files, pwds, chroots or sub mounts that are
1156 int may_umount_tree(struct vfsmount
*m
)
1158 struct mount
*mnt
= real_mount(m
);
1159 int actual_refs
= 0;
1160 int minimum_refs
= 0;
1164 /* write lock needed for mnt_get_count */
1166 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1167 actual_refs
+= mnt_get_count(p
);
1170 unlock_mount_hash();
1172 if (actual_refs
> minimum_refs
)
1178 EXPORT_SYMBOL(may_umount_tree
);
1181 * may_umount - check if a mount point is busy
1182 * @mnt: root of mount
1184 * This is called to check if a mount point has any
1185 * open files, pwds, chroots or sub mounts. If the
1186 * mount has sub mounts this will return busy
1187 * regardless of whether the sub mounts are busy.
1189 * Doesn't take quota and stuff into account. IOW, in some cases it will
1190 * give false negatives. The main reason why it's here is that we need
1191 * a non-destructive way to look for easily umountable filesystems.
1193 int may_umount(struct vfsmount
*mnt
)
1196 down_read(&namespace_sem
);
1198 if (propagate_mount_busy(real_mount(mnt
), 2))
1200 unlock_mount_hash();
1201 up_read(&namespace_sem
);
1205 EXPORT_SYMBOL(may_umount
);
1207 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
1209 static void namespace_unlock(void)
1212 struct hlist_head head
= unmounted
;
1214 if (likely(hlist_empty(&head
))) {
1215 up_write(&namespace_sem
);
1219 head
.first
->pprev
= &head
.first
;
1220 INIT_HLIST_HEAD(&unmounted
);
1222 up_write(&namespace_sem
);
1226 while (!hlist_empty(&head
)) {
1227 mnt
= hlist_entry(head
.first
, struct mount
, mnt_hash
);
1228 hlist_del_init(&mnt
->mnt_hash
);
1229 if (mnt
->mnt_ex_mountpoint
.mnt
)
1230 path_put(&mnt
->mnt_ex_mountpoint
);
1235 static inline void namespace_lock(void)
1237 down_write(&namespace_sem
);
1241 * mount_lock must be held
1242 * namespace_sem must be held for write
1243 * how = 0 => just this tree, don't propagate
1244 * how = 1 => propagate; we know that nobody else has reference to any victims
1245 * how = 2 => lazy umount
1247 void umount_tree(struct mount
*mnt
, int how
)
1249 HLIST_HEAD(tmp_list
);
1251 struct mount
*last
= NULL
;
1253 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1254 hlist_del_init_rcu(&p
->mnt_hash
);
1255 hlist_add_head(&p
->mnt_hash
, &tmp_list
);
1259 propagate_umount(&tmp_list
);
1261 hlist_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1262 list_del_init(&p
->mnt_expire
);
1263 list_del_init(&p
->mnt_list
);
1264 __touch_mnt_namespace(p
->mnt_ns
);
1267 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1268 list_del_init(&p
->mnt_child
);
1269 if (mnt_has_parent(p
)) {
1270 put_mountpoint(p
->mnt_mp
);
1271 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1272 p
->mnt_ex_mountpoint
.dentry
= p
->mnt_mountpoint
;
1273 p
->mnt_ex_mountpoint
.mnt
= &p
->mnt_parent
->mnt
;
1274 p
->mnt_mountpoint
= p
->mnt
.mnt_root
;
1278 change_mnt_propagation(p
, MS_PRIVATE
);
1282 last
->mnt_hash
.next
= unmounted
.first
;
1283 unmounted
.first
= tmp_list
.first
;
1284 unmounted
.first
->pprev
= &unmounted
.first
;
1288 static void shrink_submounts(struct mount
*mnt
);
1290 static int do_umount(struct mount
*mnt
, int flags
)
1292 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1295 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1300 * Allow userspace to request a mountpoint be expired rather than
1301 * unmounting unconditionally. Unmount only happens if:
1302 * (1) the mark is already set (the mark is cleared by mntput())
1303 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1305 if (flags
& MNT_EXPIRE
) {
1306 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1307 flags
& (MNT_FORCE
| MNT_DETACH
))
1311 * probably don't strictly need the lock here if we examined
1312 * all race cases, but it's a slowpath.
1315 if (mnt_get_count(mnt
) != 2) {
1316 unlock_mount_hash();
1319 unlock_mount_hash();
1321 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1326 * If we may have to abort operations to get out of this
1327 * mount, and they will themselves hold resources we must
1328 * allow the fs to do things. In the Unix tradition of
1329 * 'Gee thats tricky lets do it in userspace' the umount_begin
1330 * might fail to complete on the first run through as other tasks
1331 * must return, and the like. Thats for the mount program to worry
1332 * about for the moment.
1335 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1336 sb
->s_op
->umount_begin(sb
);
1340 * No sense to grab the lock for this test, but test itself looks
1341 * somewhat bogus. Suggestions for better replacement?
1342 * Ho-hum... In principle, we might treat that as umount + switch
1343 * to rootfs. GC would eventually take care of the old vfsmount.
1344 * Actually it makes sense, especially if rootfs would contain a
1345 * /reboot - static binary that would close all descriptors and
1346 * call reboot(9). Then init(8) could umount root and exec /reboot.
1348 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1350 * Special case for "unmounting" root ...
1351 * we just try to remount it readonly.
1353 down_write(&sb
->s_umount
);
1354 if (!(sb
->s_flags
& MS_RDONLY
))
1355 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1356 up_write(&sb
->s_umount
);
1364 if (flags
& MNT_DETACH
) {
1365 if (!list_empty(&mnt
->mnt_list
))
1366 umount_tree(mnt
, 2);
1369 shrink_submounts(mnt
);
1371 if (!propagate_mount_busy(mnt
, 2)) {
1372 if (!list_empty(&mnt
->mnt_list
))
1373 umount_tree(mnt
, 1);
1377 unlock_mount_hash();
1383 * Is the caller allowed to modify his namespace?
1385 static inline bool may_mount(void)
1387 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1391 * Now umount can handle mount points as well as block devices.
1392 * This is important for filesystems which use unnamed block devices.
1394 * We now support a flag for forced unmount like the other 'big iron'
1395 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1398 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1403 int lookup_flags
= 0;
1405 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1411 if (!(flags
& UMOUNT_NOFOLLOW
))
1412 lookup_flags
|= LOOKUP_FOLLOW
;
1414 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1417 mnt
= real_mount(path
.mnt
);
1419 if (path
.dentry
!= path
.mnt
->mnt_root
)
1421 if (!check_mnt(mnt
))
1423 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1426 retval
= do_umount(mnt
, flags
);
1428 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1430 mntput_no_expire(mnt
);
1435 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1438 * The 2.0 compatible umount. No flags.
1440 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1442 return sys_umount(name
, 0);
1447 static bool is_mnt_ns_file(struct dentry
*dentry
)
1449 /* Is this a proxy for a mount namespace? */
1450 struct inode
*inode
= dentry
->d_inode
;
1453 if (!proc_ns_inode(inode
))
1456 ei
= get_proc_ns(inode
);
1457 if (ei
->ns_ops
!= &mntns_operations
)
1463 static bool mnt_ns_loop(struct dentry
*dentry
)
1465 /* Could bind mounting the mount namespace inode cause a
1466 * mount namespace loop?
1468 struct mnt_namespace
*mnt_ns
;
1469 if (!is_mnt_ns_file(dentry
))
1472 mnt_ns
= get_proc_ns(dentry
->d_inode
)->ns
;
1473 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1476 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1479 struct mount
*res
, *p
, *q
, *r
, *parent
;
1481 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1482 return ERR_PTR(-EINVAL
);
1484 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1485 return ERR_PTR(-EINVAL
);
1487 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1491 q
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1492 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1495 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1497 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1500 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1501 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1502 IS_MNT_UNBINDABLE(s
)) {
1503 s
= skip_mnt_tree(s
);
1506 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1507 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1508 s
= skip_mnt_tree(s
);
1511 while (p
!= s
->mnt_parent
) {
1517 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1521 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1522 attach_mnt(q
, parent
, p
->mnt_mp
);
1523 unlock_mount_hash();
1530 umount_tree(res
, 0);
1531 unlock_mount_hash();
1536 /* Caller should check returned pointer for errors */
1538 struct vfsmount
*collect_mounts(struct path
*path
)
1542 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1543 CL_COPY_ALL
| CL_PRIVATE
);
1546 return ERR_CAST(tree
);
1550 void drop_collected_mounts(struct vfsmount
*mnt
)
1554 umount_tree(real_mount(mnt
), 0);
1555 unlock_mount_hash();
1559 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1560 struct vfsmount
*root
)
1563 int res
= f(root
, arg
);
1566 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1567 res
= f(&mnt
->mnt
, arg
);
1574 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1578 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1579 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1580 mnt_release_group_id(p
);
1584 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1588 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1589 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1590 int err
= mnt_alloc_group_id(p
);
1592 cleanup_group_ids(mnt
, p
);
1602 * @source_mnt : mount tree to be attached
1603 * @nd : place the mount tree @source_mnt is attached
1604 * @parent_nd : if non-null, detach the source_mnt from its parent and
1605 * store the parent mount and mountpoint dentry.
1606 * (done when source_mnt is moved)
1608 * NOTE: in the table below explains the semantics when a source mount
1609 * of a given type is attached to a destination mount of a given type.
1610 * ---------------------------------------------------------------------------
1611 * | BIND MOUNT OPERATION |
1612 * |**************************************************************************
1613 * | source-->| shared | private | slave | unbindable |
1617 * |**************************************************************************
1618 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1620 * |non-shared| shared (+) | private | slave (*) | invalid |
1621 * ***************************************************************************
1622 * A bind operation clones the source mount and mounts the clone on the
1623 * destination mount.
1625 * (++) the cloned mount is propagated to all the mounts in the propagation
1626 * tree of the destination mount and the cloned mount is added to
1627 * the peer group of the source mount.
1628 * (+) the cloned mount is created under the destination mount and is marked
1629 * as shared. The cloned mount is added to the peer group of the source
1631 * (+++) the mount is propagated to all the mounts in the propagation tree
1632 * of the destination mount and the cloned mount is made slave
1633 * of the same master as that of the source mount. The cloned mount
1634 * is marked as 'shared and slave'.
1635 * (*) the cloned mount is made a slave of the same master as that of the
1638 * ---------------------------------------------------------------------------
1639 * | MOVE MOUNT OPERATION |
1640 * |**************************************************************************
1641 * | source-->| shared | private | slave | unbindable |
1645 * |**************************************************************************
1646 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1648 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1649 * ***************************************************************************
1651 * (+) the mount is moved to the destination. And is then propagated to
1652 * all the mounts in the propagation tree of the destination mount.
1653 * (+*) the mount is moved to the destination.
1654 * (+++) the mount is moved to the destination and is then propagated to
1655 * all the mounts belonging to the destination mount's propagation tree.
1656 * the mount is marked as 'shared and slave'.
1657 * (*) the mount continues to be a slave at the new location.
1659 * if the source mount is a tree, the operations explained above is
1660 * applied to each mount in the tree.
1661 * Must be called without spinlocks held, since this function can sleep
1664 static int attach_recursive_mnt(struct mount
*source_mnt
,
1665 struct mount
*dest_mnt
,
1666 struct mountpoint
*dest_mp
,
1667 struct path
*parent_path
)
1669 HLIST_HEAD(tree_list
);
1670 struct mount
*child
, *p
;
1671 struct hlist_node
*n
;
1674 if (IS_MNT_SHARED(dest_mnt
)) {
1675 err
= invent_group_ids(source_mnt
, true);
1678 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1681 goto out_cleanup_ids
;
1682 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1688 detach_mnt(source_mnt
, parent_path
);
1689 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1690 touch_mnt_namespace(source_mnt
->mnt_ns
);
1692 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1693 commit_tree(source_mnt
, NULL
);
1696 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
1698 hlist_del_init(&child
->mnt_hash
);
1699 q
= __lookup_mnt_last(&child
->mnt_parent
->mnt
,
1700 child
->mnt_mountpoint
);
1701 commit_tree(child
, q
);
1703 unlock_mount_hash();
1708 while (!hlist_empty(&tree_list
)) {
1709 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
1710 umount_tree(child
, 0);
1712 unlock_mount_hash();
1713 cleanup_group_ids(source_mnt
, NULL
);
1718 static struct mountpoint
*lock_mount(struct path
*path
)
1720 struct vfsmount
*mnt
;
1721 struct dentry
*dentry
= path
->dentry
;
1723 mutex_lock(&dentry
->d_inode
->i_mutex
);
1724 if (unlikely(cant_mount(dentry
))) {
1725 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1726 return ERR_PTR(-ENOENT
);
1729 mnt
= lookup_mnt(path
);
1731 struct mountpoint
*mp
= new_mountpoint(dentry
);
1734 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1740 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1743 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1747 static void unlock_mount(struct mountpoint
*where
)
1749 struct dentry
*dentry
= where
->m_dentry
;
1750 put_mountpoint(where
);
1752 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1755 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1757 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1760 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1761 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1764 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1768 * Sanity check the flags to change_mnt_propagation.
1771 static int flags_to_propagation_type(int flags
)
1773 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1775 /* Fail if any non-propagation flags are set */
1776 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1778 /* Only one propagation flag should be set */
1779 if (!is_power_of_2(type
))
1785 * recursively change the type of the mountpoint.
1787 static int do_change_type(struct path
*path
, int flag
)
1790 struct mount
*mnt
= real_mount(path
->mnt
);
1791 int recurse
= flag
& MS_REC
;
1795 if (path
->dentry
!= path
->mnt
->mnt_root
)
1798 type
= flags_to_propagation_type(flag
);
1803 if (type
== MS_SHARED
) {
1804 err
= invent_group_ids(mnt
, recurse
);
1810 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1811 change_mnt_propagation(m
, type
);
1812 unlock_mount_hash();
1819 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
1821 struct mount
*child
;
1822 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
1823 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
1826 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
1833 * do loopback mount.
1835 static int do_loopback(struct path
*path
, const char *old_name
,
1838 struct path old_path
;
1839 struct mount
*mnt
= NULL
, *old
, *parent
;
1840 struct mountpoint
*mp
;
1842 if (!old_name
|| !*old_name
)
1844 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1849 if (mnt_ns_loop(old_path
.dentry
))
1852 mp
= lock_mount(path
);
1857 old
= real_mount(old_path
.mnt
);
1858 parent
= real_mount(path
->mnt
);
1861 if (IS_MNT_UNBINDABLE(old
))
1864 if (!check_mnt(parent
) || !check_mnt(old
))
1867 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
1871 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
1873 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
1880 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1882 err
= graft_tree(mnt
, parent
, mp
);
1885 umount_tree(mnt
, 0);
1886 unlock_mount_hash();
1891 path_put(&old_path
);
1895 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1898 int readonly_request
= 0;
1900 if (ms_flags
& MS_RDONLY
)
1901 readonly_request
= 1;
1902 if (readonly_request
== __mnt_is_readonly(mnt
))
1905 if (mnt
->mnt_flags
& MNT_LOCK_READONLY
)
1908 if (readonly_request
)
1909 error
= mnt_make_readonly(real_mount(mnt
));
1911 __mnt_unmake_readonly(real_mount(mnt
));
1916 * change filesystem flags. dir should be a physical root of filesystem.
1917 * If you've mounted a non-root directory somewhere and want to do remount
1918 * on it - tough luck.
1920 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1924 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1925 struct mount
*mnt
= real_mount(path
->mnt
);
1927 if (!check_mnt(mnt
))
1930 if (path
->dentry
!= path
->mnt
->mnt_root
)
1933 err
= security_sb_remount(sb
, data
);
1937 down_write(&sb
->s_umount
);
1938 if (flags
& MS_BIND
)
1939 err
= change_mount_flags(path
->mnt
, flags
);
1940 else if (!capable(CAP_SYS_ADMIN
))
1943 err
= do_remount_sb(sb
, flags
, data
, 0);
1946 mnt_flags
|= mnt
->mnt
.mnt_flags
& MNT_PROPAGATION_MASK
;
1947 mnt
->mnt
.mnt_flags
= mnt_flags
;
1948 touch_mnt_namespace(mnt
->mnt_ns
);
1949 unlock_mount_hash();
1951 up_write(&sb
->s_umount
);
1955 static inline int tree_contains_unbindable(struct mount
*mnt
)
1958 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1959 if (IS_MNT_UNBINDABLE(p
))
1965 static int do_move_mount(struct path
*path
, const char *old_name
)
1967 struct path old_path
, parent_path
;
1970 struct mountpoint
*mp
;
1972 if (!old_name
|| !*old_name
)
1974 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1978 mp
= lock_mount(path
);
1983 old
= real_mount(old_path
.mnt
);
1984 p
= real_mount(path
->mnt
);
1987 if (!check_mnt(p
) || !check_mnt(old
))
1990 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
1994 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1997 if (!mnt_has_parent(old
))
2000 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
2001 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
2004 * Don't move a mount residing in a shared parent.
2006 if (IS_MNT_SHARED(old
->mnt_parent
))
2009 * Don't move a mount tree containing unbindable mounts to a destination
2010 * mount which is shared.
2012 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2015 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2019 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2023 /* if the mount is moved, it should no longer be expire
2025 list_del_init(&old
->mnt_expire
);
2030 path_put(&parent_path
);
2031 path_put(&old_path
);
2035 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2038 const char *subtype
= strchr(fstype
, '.');
2047 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2049 if (!mnt
->mnt_sb
->s_subtype
)
2055 return ERR_PTR(err
);
2059 * add a mount into a namespace's mount tree
2061 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2063 struct mountpoint
*mp
;
2064 struct mount
*parent
;
2067 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2069 mp
= lock_mount(path
);
2073 parent
= real_mount(path
->mnt
);
2075 if (unlikely(!check_mnt(parent
))) {
2076 /* that's acceptable only for automounts done in private ns */
2077 if (!(mnt_flags
& MNT_SHRINKABLE
))
2079 /* ... and for those we'd better have mountpoint still alive */
2080 if (!parent
->mnt_ns
)
2084 /* Refuse the same filesystem on the same mount point */
2086 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2087 path
->mnt
->mnt_root
== path
->dentry
)
2091 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
2094 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2095 err
= graft_tree(newmnt
, parent
, mp
);
2103 * create a new mount for userspace and request it to be added into the
2106 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2107 int mnt_flags
, const char *name
, void *data
)
2109 struct file_system_type
*type
;
2110 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2111 struct vfsmount
*mnt
;
2117 type
= get_fs_type(fstype
);
2121 if (user_ns
!= &init_user_ns
) {
2122 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2123 put_filesystem(type
);
2126 /* Only in special cases allow devices from mounts
2127 * created outside the initial user namespace.
2129 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2131 mnt_flags
|= MNT_NODEV
;
2135 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2136 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2137 !mnt
->mnt_sb
->s_subtype
)
2138 mnt
= fs_set_subtype(mnt
, fstype
);
2140 put_filesystem(type
);
2142 return PTR_ERR(mnt
);
2144 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2150 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2152 struct mount
*mnt
= real_mount(m
);
2154 /* The new mount record should have at least 2 refs to prevent it being
2155 * expired before we get a chance to add it
2157 BUG_ON(mnt_get_count(mnt
) < 2);
2159 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2160 m
->mnt_root
== path
->dentry
) {
2165 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2169 /* remove m from any expiration list it may be on */
2170 if (!list_empty(&mnt
->mnt_expire
)) {
2172 list_del_init(&mnt
->mnt_expire
);
2181 * mnt_set_expiry - Put a mount on an expiration list
2182 * @mnt: The mount to list.
2183 * @expiry_list: The list to add the mount to.
2185 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2189 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2193 EXPORT_SYMBOL(mnt_set_expiry
);
2196 * process a list of expirable mountpoints with the intent of discarding any
2197 * mountpoints that aren't in use and haven't been touched since last we came
2200 void mark_mounts_for_expiry(struct list_head
*mounts
)
2202 struct mount
*mnt
, *next
;
2203 LIST_HEAD(graveyard
);
2205 if (list_empty(mounts
))
2211 /* extract from the expiration list every vfsmount that matches the
2212 * following criteria:
2213 * - only referenced by its parent vfsmount
2214 * - still marked for expiry (marked on the last call here; marks are
2215 * cleared by mntput())
2217 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2218 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2219 propagate_mount_busy(mnt
, 1))
2221 list_move(&mnt
->mnt_expire
, &graveyard
);
2223 while (!list_empty(&graveyard
)) {
2224 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2225 touch_mnt_namespace(mnt
->mnt_ns
);
2226 umount_tree(mnt
, 1);
2228 unlock_mount_hash();
2232 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2235 * Ripoff of 'select_parent()'
2237 * search the list of submounts for a given mountpoint, and move any
2238 * shrinkable submounts to the 'graveyard' list.
2240 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2242 struct mount
*this_parent
= parent
;
2243 struct list_head
*next
;
2247 next
= this_parent
->mnt_mounts
.next
;
2249 while (next
!= &this_parent
->mnt_mounts
) {
2250 struct list_head
*tmp
= next
;
2251 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2254 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2257 * Descend a level if the d_mounts list is non-empty.
2259 if (!list_empty(&mnt
->mnt_mounts
)) {
2264 if (!propagate_mount_busy(mnt
, 1)) {
2265 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2270 * All done at this level ... ascend and resume the search
2272 if (this_parent
!= parent
) {
2273 next
= this_parent
->mnt_child
.next
;
2274 this_parent
= this_parent
->mnt_parent
;
2281 * process a list of expirable mountpoints with the intent of discarding any
2282 * submounts of a specific parent mountpoint
2284 * mount_lock must be held for write
2286 static void shrink_submounts(struct mount
*mnt
)
2288 LIST_HEAD(graveyard
);
2291 /* extract submounts of 'mountpoint' from the expiration list */
2292 while (select_submounts(mnt
, &graveyard
)) {
2293 while (!list_empty(&graveyard
)) {
2294 m
= list_first_entry(&graveyard
, struct mount
,
2296 touch_mnt_namespace(m
->mnt_ns
);
2303 * Some copy_from_user() implementations do not return the exact number of
2304 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2305 * Note that this function differs from copy_from_user() in that it will oops
2306 * on bad values of `to', rather than returning a short copy.
2308 static long exact_copy_from_user(void *to
, const void __user
* from
,
2312 const char __user
*f
= from
;
2315 if (!access_ok(VERIFY_READ
, from
, n
))
2319 if (__get_user(c
, f
)) {
2330 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2340 if (!(page
= __get_free_page(GFP_KERNEL
)))
2343 /* We only care that *some* data at the address the user
2344 * gave us is valid. Just in case, we'll zero
2345 * the remainder of the page.
2347 /* copy_from_user cannot cross TASK_SIZE ! */
2348 size
= TASK_SIZE
- (unsigned long)data
;
2349 if (size
> PAGE_SIZE
)
2352 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2358 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2363 int copy_mount_string(const void __user
*data
, char **where
)
2372 tmp
= strndup_user(data
, PAGE_SIZE
);
2374 return PTR_ERR(tmp
);
2381 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2382 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2384 * data is a (void *) that can point to any structure up to
2385 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2386 * information (or be NULL).
2388 * Pre-0.97 versions of mount() didn't have a flags word.
2389 * When the flags word was introduced its top half was required
2390 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2391 * Therefore, if this magic number is present, it carries no information
2392 * and must be discarded.
2394 long do_mount(const char *dev_name
, const char *dir_name
,
2395 const char *type_page
, unsigned long flags
, void *data_page
)
2402 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2403 flags
&= ~MS_MGC_MSK
;
2405 /* Basic sanity checks */
2407 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
2411 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2413 /* ... and get the mountpoint */
2414 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
2418 retval
= security_sb_mount(dev_name
, &path
,
2419 type_page
, flags
, data_page
);
2420 if (!retval
&& !may_mount())
2425 /* Default to relatime unless overriden */
2426 if (!(flags
& MS_NOATIME
))
2427 mnt_flags
|= MNT_RELATIME
;
2429 /* Separate the per-mountpoint flags */
2430 if (flags
& MS_NOSUID
)
2431 mnt_flags
|= MNT_NOSUID
;
2432 if (flags
& MS_NODEV
)
2433 mnt_flags
|= MNT_NODEV
;
2434 if (flags
& MS_NOEXEC
)
2435 mnt_flags
|= MNT_NOEXEC
;
2436 if (flags
& MS_NOATIME
)
2437 mnt_flags
|= MNT_NOATIME
;
2438 if (flags
& MS_NODIRATIME
)
2439 mnt_flags
|= MNT_NODIRATIME
;
2440 if (flags
& MS_STRICTATIME
)
2441 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2442 if (flags
& MS_RDONLY
)
2443 mnt_flags
|= MNT_READONLY
;
2445 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2446 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2449 if (flags
& MS_REMOUNT
)
2450 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2452 else if (flags
& MS_BIND
)
2453 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2454 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2455 retval
= do_change_type(&path
, flags
);
2456 else if (flags
& MS_MOVE
)
2457 retval
= do_move_mount(&path
, dev_name
);
2459 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2460 dev_name
, data_page
);
2466 static void free_mnt_ns(struct mnt_namespace
*ns
)
2468 proc_free_inum(ns
->proc_inum
);
2469 put_user_ns(ns
->user_ns
);
2474 * Assign a sequence number so we can detect when we attempt to bind
2475 * mount a reference to an older mount namespace into the current
2476 * mount namespace, preventing reference counting loops. A 64bit
2477 * number incrementing at 10Ghz will take 12,427 years to wrap which
2478 * is effectively never, so we can ignore the possibility.
2480 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2482 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2484 struct mnt_namespace
*new_ns
;
2487 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2489 return ERR_PTR(-ENOMEM
);
2490 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2493 return ERR_PTR(ret
);
2495 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2496 atomic_set(&new_ns
->count
, 1);
2497 new_ns
->root
= NULL
;
2498 INIT_LIST_HEAD(&new_ns
->list
);
2499 init_waitqueue_head(&new_ns
->poll
);
2501 new_ns
->user_ns
= get_user_ns(user_ns
);
2505 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2506 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2508 struct mnt_namespace
*new_ns
;
2509 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2510 struct mount
*p
, *q
;
2517 if (likely(!(flags
& CLONE_NEWNS
))) {
2524 new_ns
= alloc_mnt_ns(user_ns
);
2529 /* First pass: copy the tree topology */
2530 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2531 if (user_ns
!= ns
->user_ns
)
2532 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2533 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2536 free_mnt_ns(new_ns
);
2537 return ERR_CAST(new);
2540 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2543 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2544 * as belonging to new namespace. We have already acquired a private
2545 * fs_struct, so tsk->fs->lock is not needed.
2552 if (&p
->mnt
== new_fs
->root
.mnt
) {
2553 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2556 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2557 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2561 p
= next_mnt(p
, old
);
2562 q
= next_mnt(q
, new);
2565 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2566 p
= next_mnt(p
, old
);
2579 * create_mnt_ns - creates a private namespace and adds a root filesystem
2580 * @mnt: pointer to the new root filesystem mountpoint
2582 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2584 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2585 if (!IS_ERR(new_ns
)) {
2586 struct mount
*mnt
= real_mount(m
);
2587 mnt
->mnt_ns
= new_ns
;
2589 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2596 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2598 struct mnt_namespace
*ns
;
2599 struct super_block
*s
;
2603 ns
= create_mnt_ns(mnt
);
2605 return ERR_CAST(ns
);
2607 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2608 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2613 return ERR_PTR(err
);
2615 /* trade a vfsmount reference for active sb one */
2616 s
= path
.mnt
->mnt_sb
;
2617 atomic_inc(&s
->s_active
);
2619 /* lock the sucker */
2620 down_write(&s
->s_umount
);
2621 /* ... and return the root of (sub)tree on it */
2624 EXPORT_SYMBOL(mount_subtree
);
2626 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2627 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2631 struct filename
*kernel_dir
;
2633 unsigned long data_page
;
2635 ret
= copy_mount_string(type
, &kernel_type
);
2639 kernel_dir
= getname(dir_name
);
2640 if (IS_ERR(kernel_dir
)) {
2641 ret
= PTR_ERR(kernel_dir
);
2645 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2649 ret
= copy_mount_options(data
, &data_page
);
2653 ret
= do_mount(kernel_dev
, kernel_dir
->name
, kernel_type
, flags
,
2654 (void *) data_page
);
2656 free_page(data_page
);
2660 putname(kernel_dir
);
2668 * Return true if path is reachable from root
2670 * namespace_sem or mount_lock is held
2672 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2673 const struct path
*root
)
2675 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2676 dentry
= mnt
->mnt_mountpoint
;
2677 mnt
= mnt
->mnt_parent
;
2679 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2682 int path_is_under(struct path
*path1
, struct path
*path2
)
2685 read_seqlock_excl(&mount_lock
);
2686 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2687 read_sequnlock_excl(&mount_lock
);
2690 EXPORT_SYMBOL(path_is_under
);
2693 * pivot_root Semantics:
2694 * Moves the root file system of the current process to the directory put_old,
2695 * makes new_root as the new root file system of the current process, and sets
2696 * root/cwd of all processes which had them on the current root to new_root.
2699 * The new_root and put_old must be directories, and must not be on the
2700 * same file system as the current process root. The put_old must be
2701 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2702 * pointed to by put_old must yield the same directory as new_root. No other
2703 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2705 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2706 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2707 * in this situation.
2710 * - we don't move root/cwd if they are not at the root (reason: if something
2711 * cared enough to change them, it's probably wrong to force them elsewhere)
2712 * - it's okay to pick a root that isn't the root of a file system, e.g.
2713 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2714 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2717 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2718 const char __user
*, put_old
)
2720 struct path
new, old
, parent_path
, root_parent
, root
;
2721 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2722 struct mountpoint
*old_mp
, *root_mp
;
2728 error
= user_path_dir(new_root
, &new);
2732 error
= user_path_dir(put_old
, &old
);
2736 error
= security_sb_pivotroot(&old
, &new);
2740 get_fs_root(current
->fs
, &root
);
2741 old_mp
= lock_mount(&old
);
2742 error
= PTR_ERR(old_mp
);
2747 new_mnt
= real_mount(new.mnt
);
2748 root_mnt
= real_mount(root
.mnt
);
2749 old_mnt
= real_mount(old
.mnt
);
2750 if (IS_MNT_SHARED(old_mnt
) ||
2751 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2752 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2754 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2756 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
2759 if (d_unlinked(new.dentry
))
2762 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2763 goto out4
; /* loop, on the same file system */
2765 if (root
.mnt
->mnt_root
!= root
.dentry
)
2766 goto out4
; /* not a mountpoint */
2767 if (!mnt_has_parent(root_mnt
))
2768 goto out4
; /* not attached */
2769 root_mp
= root_mnt
->mnt_mp
;
2770 if (new.mnt
->mnt_root
!= new.dentry
)
2771 goto out4
; /* not a mountpoint */
2772 if (!mnt_has_parent(new_mnt
))
2773 goto out4
; /* not attached */
2774 /* make sure we can reach put_old from new_root */
2775 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2777 root_mp
->m_count
++; /* pin it so it won't go away */
2779 detach_mnt(new_mnt
, &parent_path
);
2780 detach_mnt(root_mnt
, &root_parent
);
2781 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
2782 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
2783 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2785 /* mount old root on put_old */
2786 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2787 /* mount new_root on / */
2788 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2789 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2790 unlock_mount_hash();
2791 chroot_fs_refs(&root
, &new);
2792 put_mountpoint(root_mp
);
2795 unlock_mount(old_mp
);
2797 path_put(&root_parent
);
2798 path_put(&parent_path
);
2810 static void __init
init_mount_tree(void)
2812 struct vfsmount
*mnt
;
2813 struct mnt_namespace
*ns
;
2815 struct file_system_type
*type
;
2817 type
= get_fs_type("rootfs");
2819 panic("Can't find rootfs type");
2820 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2821 put_filesystem(type
);
2823 panic("Can't create rootfs");
2825 ns
= create_mnt_ns(mnt
);
2827 panic("Can't allocate initial namespace");
2829 init_task
.nsproxy
->mnt_ns
= ns
;
2833 root
.dentry
= mnt
->mnt_root
;
2835 set_fs_pwd(current
->fs
, &root
);
2836 set_fs_root(current
->fs
, &root
);
2839 void __init
mnt_init(void)
2844 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2845 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2847 mount_hashtable
= alloc_large_system_hash("Mount-cache",
2848 sizeof(struct hlist_head
),
2851 &m_hash_shift
, &m_hash_mask
, 0, 0);
2852 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
2853 sizeof(struct hlist_head
),
2856 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
2858 if (!mount_hashtable
|| !mountpoint_hashtable
)
2859 panic("Failed to allocate mount hash table\n");
2861 for (u
= 0; u
<= m_hash_mask
; u
++)
2862 INIT_HLIST_HEAD(&mount_hashtable
[u
]);
2863 for (u
= 0; u
<= mp_hash_mask
; u
++)
2864 INIT_HLIST_HEAD(&mountpoint_hashtable
[u
]);
2870 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2872 fs_kobj
= kobject_create_and_add("fs", NULL
);
2874 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2879 void put_mnt_ns(struct mnt_namespace
*ns
)
2881 if (!atomic_dec_and_test(&ns
->count
))
2883 drop_collected_mounts(&ns
->root
->mnt
);
2887 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
2889 struct vfsmount
*mnt
;
2890 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
2893 * it is a longterm mount, don't release mnt until
2894 * we unmount before file sys is unregistered
2896 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
2900 EXPORT_SYMBOL_GPL(kern_mount_data
);
2902 void kern_unmount(struct vfsmount
*mnt
)
2904 /* release long term mount so mount point can be released */
2905 if (!IS_ERR_OR_NULL(mnt
)) {
2906 real_mount(mnt
)->mnt_ns
= NULL
;
2907 synchronize_rcu(); /* yecchhh... */
2911 EXPORT_SYMBOL(kern_unmount
);
2913 bool our_mnt(struct vfsmount
*mnt
)
2915 return check_mnt(real_mount(mnt
));
2918 bool current_chrooted(void)
2920 /* Does the current process have a non-standard root */
2921 struct path ns_root
;
2922 struct path fs_root
;
2925 /* Find the namespace root */
2926 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
2927 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
2929 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
2932 get_fs_root(current
->fs
, &fs_root
);
2934 chrooted
= !path_equal(&fs_root
, &ns_root
);
2942 bool fs_fully_visible(struct file_system_type
*type
)
2944 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
2946 bool visible
= false;
2951 down_read(&namespace_sem
);
2952 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
2953 struct mount
*child
;
2954 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
2957 /* This mount is not fully visible if there are any child mounts
2958 * that cover anything except for empty directories.
2960 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2961 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
2962 if (!S_ISDIR(inode
->i_mode
))
2964 if (inode
->i_nlink
> 2)
2972 up_read(&namespace_sem
);
2976 static void *mntns_get(struct task_struct
*task
)
2978 struct mnt_namespace
*ns
= NULL
;
2979 struct nsproxy
*nsproxy
;
2982 nsproxy
= task_nsproxy(task
);
2984 ns
= nsproxy
->mnt_ns
;
2992 static void mntns_put(void *ns
)
2997 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
2999 struct fs_struct
*fs
= current
->fs
;
3000 struct mnt_namespace
*mnt_ns
= ns
;
3003 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3004 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
3005 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
3012 put_mnt_ns(nsproxy
->mnt_ns
);
3013 nsproxy
->mnt_ns
= mnt_ns
;
3016 root
.mnt
= &mnt_ns
->root
->mnt
;
3017 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
3019 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
3022 /* Update the pwd and root */
3023 set_fs_pwd(fs
, &root
);
3024 set_fs_root(fs
, &root
);
3030 static unsigned int mntns_inum(void *ns
)
3032 struct mnt_namespace
*mnt_ns
= ns
;
3033 return mnt_ns
->proc_inum
;
3036 const struct proc_ns_operations mntns_operations
= {
3038 .type
= CLONE_NEWNS
,
3041 .install
= mntns_install
,