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/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 /* Maximum number of mounts in a mount namespace */
31 unsigned int sysctl_mount_max __read_mostly
= 100000;
33 static unsigned int m_hash_mask __read_mostly
;
34 static unsigned int m_hash_shift __read_mostly
;
35 static unsigned int mp_hash_mask __read_mostly
;
36 static unsigned int mp_hash_shift __read_mostly
;
38 static __initdata
unsigned long mhash_entries
;
39 static int __init
set_mhash_entries(char *str
)
43 mhash_entries
= simple_strtoul(str
, &str
, 0);
46 __setup("mhash_entries=", set_mhash_entries
);
48 static __initdata
unsigned long mphash_entries
;
49 static int __init
set_mphash_entries(char *str
)
53 mphash_entries
= simple_strtoul(str
, &str
, 0);
56 __setup("mphash_entries=", set_mphash_entries
);
59 static DEFINE_IDA(mnt_id_ida
);
60 static DEFINE_IDA(mnt_group_ida
);
61 static DEFINE_SPINLOCK(mnt_id_lock
);
62 static int mnt_id_start
= 0;
63 static int mnt_group_start
= 1;
65 static struct hlist_head
*mount_hashtable __read_mostly
;
66 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
67 static struct kmem_cache
*mnt_cache __read_mostly
;
68 static DECLARE_RWSEM(namespace_sem
);
71 struct kobject
*fs_kobj
;
72 EXPORT_SYMBOL_GPL(fs_kobj
);
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
82 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
84 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
86 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
87 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
88 tmp
= tmp
+ (tmp
>> m_hash_shift
);
89 return &mount_hashtable
[tmp
& m_hash_mask
];
92 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
94 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
95 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
96 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
103 static int mnt_alloc_id(struct mount
*mnt
)
108 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
109 spin_lock(&mnt_id_lock
);
110 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
112 mnt_id_start
= mnt
->mnt_id
+ 1;
113 spin_unlock(&mnt_id_lock
);
120 static void mnt_free_id(struct mount
*mnt
)
122 int id
= mnt
->mnt_id
;
123 spin_lock(&mnt_id_lock
);
124 ida_remove(&mnt_id_ida
, id
);
125 if (mnt_id_start
> id
)
127 spin_unlock(&mnt_id_lock
);
131 * Allocate a new peer group ID
133 * mnt_group_ida is protected by namespace_sem
135 static int mnt_alloc_group_id(struct mount
*mnt
)
139 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
142 res
= ida_get_new_above(&mnt_group_ida
,
146 mnt_group_start
= mnt
->mnt_group_id
+ 1;
152 * Release a peer group ID
154 void mnt_release_group_id(struct mount
*mnt
)
156 int id
= mnt
->mnt_group_id
;
157 ida_remove(&mnt_group_ida
, id
);
158 if (mnt_group_start
> id
)
159 mnt_group_start
= id
;
160 mnt
->mnt_group_id
= 0;
164 * vfsmount lock must be held for read
166 static inline void mnt_add_count(struct mount
*mnt
, int n
)
169 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
178 * vfsmount lock must be held for write
180 unsigned int mnt_get_count(struct mount
*mnt
)
183 unsigned int count
= 0;
186 for_each_possible_cpu(cpu
) {
187 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
192 return mnt
->mnt_count
;
196 static void drop_mountpoint(struct fs_pin
*p
)
198 struct mount
*m
= container_of(p
, struct mount
, mnt_umount
);
199 dput(m
->mnt_ex_mountpoint
);
204 static struct mount
*alloc_vfsmnt(const char *name
)
206 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
210 err
= mnt_alloc_id(mnt
);
215 mnt
->mnt_devname
= kstrdup_const(name
, GFP_KERNEL
);
216 if (!mnt
->mnt_devname
)
221 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
223 goto out_free_devname
;
225 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
228 mnt
->mnt_writers
= 0;
231 INIT_HLIST_NODE(&mnt
->mnt_hash
);
232 INIT_LIST_HEAD(&mnt
->mnt_child
);
233 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
234 INIT_LIST_HEAD(&mnt
->mnt_list
);
235 INIT_LIST_HEAD(&mnt
->mnt_expire
);
236 INIT_LIST_HEAD(&mnt
->mnt_share
);
237 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
238 INIT_LIST_HEAD(&mnt
->mnt_slave
);
239 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
240 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
241 #ifdef CONFIG_FSNOTIFY
242 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
244 init_fs_pin(&mnt
->mnt_umount
, drop_mountpoint
);
250 kfree_const(mnt
->mnt_devname
);
255 kmem_cache_free(mnt_cache
, mnt
);
260 * Most r/o checks on a fs are for operations that take
261 * discrete amounts of time, like a write() or unlink().
262 * We must keep track of when those operations start
263 * (for permission checks) and when they end, so that
264 * we can determine when writes are able to occur to
268 * __mnt_is_readonly: check whether a mount is read-only
269 * @mnt: the mount to check for its write status
271 * This shouldn't be used directly ouside of the VFS.
272 * It does not guarantee that the filesystem will stay
273 * r/w, just that it is right *now*. This can not and
274 * should not be used in place of IS_RDONLY(inode).
275 * mnt_want/drop_write() will _keep_ the filesystem
278 int __mnt_is_readonly(struct vfsmount
*mnt
)
280 if (mnt
->mnt_flags
& MNT_READONLY
)
282 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
286 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
288 static inline void mnt_inc_writers(struct mount
*mnt
)
291 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
297 static inline void mnt_dec_writers(struct mount
*mnt
)
300 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
306 static unsigned int mnt_get_writers(struct mount
*mnt
)
309 unsigned int count
= 0;
312 for_each_possible_cpu(cpu
) {
313 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
318 return mnt
->mnt_writers
;
322 static int mnt_is_readonly(struct vfsmount
*mnt
)
324 if (mnt
->mnt_sb
->s_readonly_remount
)
326 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
328 return __mnt_is_readonly(mnt
);
332 * Most r/o & frozen checks on a fs are for operations that take discrete
333 * amounts of time, like a write() or unlink(). We must keep track of when
334 * those operations start (for permission checks) and when they end, so that we
335 * can determine when writes are able to occur to a filesystem.
338 * __mnt_want_write - get write access to a mount without freeze protection
339 * @m: the mount on which to take a write
341 * This tells the low-level filesystem that a write is about to be performed to
342 * it, and makes sure that writes are allowed (mnt it read-write) before
343 * returning success. This operation does not protect against filesystem being
344 * frozen. When the write operation is finished, __mnt_drop_write() must be
345 * called. This is effectively a refcount.
347 int __mnt_want_write(struct vfsmount
*m
)
349 struct mount
*mnt
= real_mount(m
);
353 mnt_inc_writers(mnt
);
355 * The store to mnt_inc_writers must be visible before we pass
356 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
357 * incremented count after it has set MNT_WRITE_HOLD.
360 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
363 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
364 * be set to match its requirements. So we must not load that until
365 * MNT_WRITE_HOLD is cleared.
368 if (mnt_is_readonly(m
)) {
369 mnt_dec_writers(mnt
);
378 * mnt_want_write - get write access to a mount
379 * @m: the mount on which to take a write
381 * This tells the low-level filesystem that a write is about to be performed to
382 * it, and makes sure that writes are allowed (mount is read-write, filesystem
383 * is not frozen) before returning success. When the write operation is
384 * finished, mnt_drop_write() must be called. This is effectively a refcount.
386 int mnt_want_write(struct vfsmount
*m
)
390 sb_start_write(m
->mnt_sb
);
391 ret
= __mnt_want_write(m
);
393 sb_end_write(m
->mnt_sb
);
396 EXPORT_SYMBOL_GPL(mnt_want_write
);
399 * mnt_clone_write - get write access to a mount
400 * @mnt: the mount on which to take a write
402 * This is effectively like mnt_want_write, except
403 * it must only be used to take an extra write reference
404 * on a mountpoint that we already know has a write reference
405 * on it. This allows some optimisation.
407 * After finished, mnt_drop_write must be called as usual to
408 * drop the reference.
410 int mnt_clone_write(struct vfsmount
*mnt
)
412 /* superblock may be r/o */
413 if (__mnt_is_readonly(mnt
))
416 mnt_inc_writers(real_mount(mnt
));
420 EXPORT_SYMBOL_GPL(mnt_clone_write
);
423 * __mnt_want_write_file - get write access to a file's mount
424 * @file: the file who's mount on which to take a write
426 * This is like __mnt_want_write, but it takes a file and can
427 * do some optimisations if the file is open for write already
429 int __mnt_want_write_file(struct file
*file
)
431 if (!(file
->f_mode
& FMODE_WRITER
))
432 return __mnt_want_write(file
->f_path
.mnt
);
434 return mnt_clone_write(file
->f_path
.mnt
);
438 * mnt_want_write_file - get write access to a file's mount
439 * @file: the file who's mount on which to take a write
441 * This is like mnt_want_write, but it takes a file and can
442 * do some optimisations if the file is open for write already
444 int mnt_want_write_file(struct file
*file
)
448 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
449 ret
= __mnt_want_write_file(file
);
451 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
454 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
457 * __mnt_drop_write - give up write access to a mount
458 * @mnt: the mount on which to give up write access
460 * Tells the low-level filesystem that we are done
461 * performing writes to it. Must be matched with
462 * __mnt_want_write() call above.
464 void __mnt_drop_write(struct vfsmount
*mnt
)
467 mnt_dec_writers(real_mount(mnt
));
470 EXPORT_SYMBOL_GPL(__mnt_drop_write
);
473 * mnt_drop_write - give up write access to a mount
474 * @mnt: the mount on which to give up write access
476 * Tells the low-level filesystem that we are done performing writes to it and
477 * also allows filesystem to be frozen again. Must be matched with
478 * mnt_want_write() call above.
480 void mnt_drop_write(struct vfsmount
*mnt
)
482 __mnt_drop_write(mnt
);
483 sb_end_write(mnt
->mnt_sb
);
485 EXPORT_SYMBOL_GPL(mnt_drop_write
);
487 void __mnt_drop_write_file(struct file
*file
)
489 __mnt_drop_write(file
->f_path
.mnt
);
492 void mnt_drop_write_file(struct file
*file
)
494 mnt_drop_write(file
->f_path
.mnt
);
496 EXPORT_SYMBOL(mnt_drop_write_file
);
498 static int mnt_make_readonly(struct mount
*mnt
)
503 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
505 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
506 * should be visible before we do.
511 * With writers on hold, if this value is zero, then there are
512 * definitely no active writers (although held writers may subsequently
513 * increment the count, they'll have to wait, and decrement it after
514 * seeing MNT_READONLY).
516 * It is OK to have counter incremented on one CPU and decremented on
517 * another: the sum will add up correctly. The danger would be when we
518 * sum up each counter, if we read a counter before it is incremented,
519 * but then read another CPU's count which it has been subsequently
520 * decremented from -- we would see more decrements than we should.
521 * MNT_WRITE_HOLD protects against this scenario, because
522 * mnt_want_write first increments count, then smp_mb, then spins on
523 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
524 * we're counting up here.
526 if (mnt_get_writers(mnt
) > 0)
529 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
531 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
532 * that become unheld will see MNT_READONLY.
535 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
540 static void __mnt_unmake_readonly(struct mount
*mnt
)
543 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
547 int sb_prepare_remount_readonly(struct super_block
*sb
)
552 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
553 if (atomic_long_read(&sb
->s_remove_count
))
557 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
558 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
559 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
561 if (mnt_get_writers(mnt
) > 0) {
567 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
571 sb
->s_readonly_remount
= 1;
574 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
575 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
576 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
583 static void free_vfsmnt(struct mount
*mnt
)
585 kfree_const(mnt
->mnt_devname
);
587 free_percpu(mnt
->mnt_pcp
);
589 kmem_cache_free(mnt_cache
, mnt
);
592 static void delayed_free_vfsmnt(struct rcu_head
*head
)
594 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
597 /* call under rcu_read_lock */
598 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
601 if (read_seqretry(&mount_lock
, seq
))
605 mnt
= real_mount(bastard
);
606 mnt_add_count(mnt
, 1);
607 if (likely(!read_seqretry(&mount_lock
, seq
)))
609 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
610 mnt_add_count(mnt
, -1);
616 /* call under rcu_read_lock */
617 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
619 int res
= __legitimize_mnt(bastard
, seq
);
622 if (unlikely(res
< 0)) {
631 * find the first mount at @dentry on vfsmount @mnt.
632 * call under rcu_read_lock()
634 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
636 struct hlist_head
*head
= m_hash(mnt
, dentry
);
639 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
640 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
646 * lookup_mnt - Return the first child mount mounted at path
648 * "First" means first mounted chronologically. If you create the
651 * mount /dev/sda1 /mnt
652 * mount /dev/sda2 /mnt
653 * mount /dev/sda3 /mnt
655 * Then lookup_mnt() on the base /mnt dentry in the root mount will
656 * return successively the root dentry and vfsmount of /dev/sda1, then
657 * /dev/sda2, then /dev/sda3, then NULL.
659 * lookup_mnt takes a reference to the found vfsmount.
661 struct vfsmount
*lookup_mnt(struct path
*path
)
663 struct mount
*child_mnt
;
669 seq
= read_seqbegin(&mount_lock
);
670 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
671 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
672 } while (!legitimize_mnt(m
, seq
));
678 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
679 * current mount namespace.
681 * The common case is dentries are not mountpoints at all and that
682 * test is handled inline. For the slow case when we are actually
683 * dealing with a mountpoint of some kind, walk through all of the
684 * mounts in the current mount namespace and test to see if the dentry
687 * The mount_hashtable is not usable in the context because we
688 * need to identify all mounts that may be in the current mount
689 * namespace not just a mount that happens to have some specified
692 bool __is_local_mountpoint(struct dentry
*dentry
)
694 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
696 bool is_covered
= false;
698 if (!d_mountpoint(dentry
))
701 down_read(&namespace_sem
);
702 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
703 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
707 up_read(&namespace_sem
);
712 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
714 struct hlist_head
*chain
= mp_hash(dentry
);
715 struct mountpoint
*mp
;
717 hlist_for_each_entry(mp
, chain
, m_hash
) {
718 if (mp
->m_dentry
== dentry
) {
719 /* might be worth a WARN_ON() */
720 if (d_unlinked(dentry
))
721 return ERR_PTR(-ENOENT
);
729 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
731 struct mountpoint
*mp
, *new = NULL
;
734 if (d_mountpoint(dentry
)) {
736 read_seqlock_excl(&mount_lock
);
737 mp
= lookup_mountpoint(dentry
);
738 read_sequnlock_excl(&mount_lock
);
744 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
746 return ERR_PTR(-ENOMEM
);
749 /* Exactly one processes may set d_mounted */
750 ret
= d_set_mounted(dentry
);
752 /* Someone else set d_mounted? */
756 /* The dentry is not available as a mountpoint? */
761 /* Add the new mountpoint to the hash table */
762 read_seqlock_excl(&mount_lock
);
763 new->m_dentry
= dentry
;
765 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
766 INIT_HLIST_HEAD(&new->m_list
);
767 read_sequnlock_excl(&mount_lock
);
776 static void put_mountpoint(struct mountpoint
*mp
)
778 if (!--mp
->m_count
) {
779 struct dentry
*dentry
= mp
->m_dentry
;
780 BUG_ON(!hlist_empty(&mp
->m_list
));
781 spin_lock(&dentry
->d_lock
);
782 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
783 spin_unlock(&dentry
->d_lock
);
784 hlist_del(&mp
->m_hash
);
789 static inline int check_mnt(struct mount
*mnt
)
791 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
795 * vfsmount lock must be held for write
797 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
801 wake_up_interruptible(&ns
->poll
);
806 * vfsmount lock must be held for write
808 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
810 if (ns
&& ns
->event
!= event
) {
812 wake_up_interruptible(&ns
->poll
);
817 * vfsmount lock must be held for write
819 static void unhash_mnt(struct mount
*mnt
)
821 mnt
->mnt_parent
= mnt
;
822 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
823 list_del_init(&mnt
->mnt_child
);
824 hlist_del_init_rcu(&mnt
->mnt_hash
);
825 hlist_del_init(&mnt
->mnt_mp_list
);
826 put_mountpoint(mnt
->mnt_mp
);
831 * vfsmount lock must be held for write
833 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
835 old_path
->dentry
= mnt
->mnt_mountpoint
;
836 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
841 * vfsmount lock must be held for write
843 static void umount_mnt(struct mount
*mnt
)
845 /* old mountpoint will be dropped when we can do that */
846 mnt
->mnt_ex_mountpoint
= mnt
->mnt_mountpoint
;
851 * vfsmount lock must be held for write
853 void mnt_set_mountpoint(struct mount
*mnt
,
854 struct mountpoint
*mp
,
855 struct mount
*child_mnt
)
858 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
859 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
860 child_mnt
->mnt_parent
= mnt
;
861 child_mnt
->mnt_mp
= mp
;
862 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
865 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
867 hlist_add_head_rcu(&mnt
->mnt_hash
,
868 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
869 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
873 * vfsmount lock must be held for write
875 static void attach_mnt(struct mount
*mnt
,
876 struct mount
*parent
,
877 struct mountpoint
*mp
)
879 mnt_set_mountpoint(parent
, mp
, mnt
);
880 __attach_mnt(mnt
, parent
);
883 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
885 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
886 struct dentry
*old_mountpoint
= mnt
->mnt_mountpoint
;
887 struct mount
*old_parent
= mnt
->mnt_parent
;
889 list_del_init(&mnt
->mnt_child
);
890 hlist_del_init(&mnt
->mnt_mp_list
);
891 hlist_del_init_rcu(&mnt
->mnt_hash
);
893 attach_mnt(mnt
, parent
, mp
);
895 put_mountpoint(old_mp
);
898 * Safely avoid even the suggestion this code might sleep or
899 * lock the mount hash by taking advantage of the knowledge that
900 * mnt_change_mountpoint will not release the final reference
903 * During mounting, the mount passed in as the parent mount will
904 * continue to use the old mountpoint and during unmounting, the
905 * old mountpoint will continue to exist until namespace_unlock,
906 * which happens well after mnt_change_mountpoint.
908 spin_lock(&old_mountpoint
->d_lock
);
909 old_mountpoint
->d_lockref
.count
--;
910 spin_unlock(&old_mountpoint
->d_lock
);
912 mnt_add_count(old_parent
, -1);
916 * vfsmount lock must be held for write
918 static void commit_tree(struct mount
*mnt
)
920 struct mount
*parent
= mnt
->mnt_parent
;
923 struct mnt_namespace
*n
= parent
->mnt_ns
;
925 BUG_ON(parent
== mnt
);
927 list_add_tail(&head
, &mnt
->mnt_list
);
928 list_for_each_entry(m
, &head
, mnt_list
)
931 list_splice(&head
, n
->list
.prev
);
933 n
->mounts
+= n
->pending_mounts
;
934 n
->pending_mounts
= 0;
936 __attach_mnt(mnt
, parent
);
937 touch_mnt_namespace(n
);
940 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
942 struct list_head
*next
= p
->mnt_mounts
.next
;
943 if (next
== &p
->mnt_mounts
) {
947 next
= p
->mnt_child
.next
;
948 if (next
!= &p
->mnt_parent
->mnt_mounts
)
953 return list_entry(next
, struct mount
, mnt_child
);
956 static struct mount
*skip_mnt_tree(struct mount
*p
)
958 struct list_head
*prev
= p
->mnt_mounts
.prev
;
959 while (prev
!= &p
->mnt_mounts
) {
960 p
= list_entry(prev
, struct mount
, mnt_child
);
961 prev
= p
->mnt_mounts
.prev
;
967 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
973 return ERR_PTR(-ENODEV
);
975 mnt
= alloc_vfsmnt(name
);
977 return ERR_PTR(-ENOMEM
);
979 if (flags
& MS_KERNMOUNT
)
980 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
982 root
= mount_fs(type
, flags
, name
, data
);
986 return ERR_CAST(root
);
989 mnt
->mnt
.mnt_root
= root
;
990 mnt
->mnt
.mnt_sb
= root
->d_sb
;
991 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
992 mnt
->mnt_parent
= mnt
;
994 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
998 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1001 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1002 const char *name
, void *data
)
1004 /* Until it is worked out how to pass the user namespace
1005 * through from the parent mount to the submount don't support
1006 * unprivileged mounts with submounts.
1008 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1009 return ERR_PTR(-EPERM
);
1011 return vfs_kern_mount(type
, MS_SUBMOUNT
, name
, data
);
1013 EXPORT_SYMBOL_GPL(vfs_submount
);
1015 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1018 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1022 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1024 return ERR_PTR(-ENOMEM
);
1026 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1027 mnt
->mnt_group_id
= 0; /* not a peer of original */
1029 mnt
->mnt_group_id
= old
->mnt_group_id
;
1031 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1032 err
= mnt_alloc_group_id(mnt
);
1037 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~(MNT_WRITE_HOLD
|MNT_MARKED
);
1038 /* Don't allow unprivileged users to change mount flags */
1039 if (flag
& CL_UNPRIVILEGED
) {
1040 mnt
->mnt
.mnt_flags
|= MNT_LOCK_ATIME
;
1042 if (mnt
->mnt
.mnt_flags
& MNT_READONLY
)
1043 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
1045 if (mnt
->mnt
.mnt_flags
& MNT_NODEV
)
1046 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NODEV
;
1048 if (mnt
->mnt
.mnt_flags
& MNT_NOSUID
)
1049 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOSUID
;
1051 if (mnt
->mnt
.mnt_flags
& MNT_NOEXEC
)
1052 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOEXEC
;
1055 /* Don't allow unprivileged users to reveal what is under a mount */
1056 if ((flag
& CL_UNPRIVILEGED
) &&
1057 (!(flag
& CL_EXPIRE
) || list_empty(&old
->mnt_expire
)))
1058 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
1060 atomic_inc(&sb
->s_active
);
1061 mnt
->mnt
.mnt_sb
= sb
;
1062 mnt
->mnt
.mnt_root
= dget(root
);
1063 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1064 mnt
->mnt_parent
= mnt
;
1066 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1067 unlock_mount_hash();
1069 if ((flag
& CL_SLAVE
) ||
1070 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1071 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1072 mnt
->mnt_master
= old
;
1073 CLEAR_MNT_SHARED(mnt
);
1074 } else if (!(flag
& CL_PRIVATE
)) {
1075 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1076 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1077 if (IS_MNT_SLAVE(old
))
1078 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1079 mnt
->mnt_master
= old
->mnt_master
;
1081 if (flag
& CL_MAKE_SHARED
)
1082 set_mnt_shared(mnt
);
1084 /* stick the duplicate mount on the same expiry list
1085 * as the original if that was on one */
1086 if (flag
& CL_EXPIRE
) {
1087 if (!list_empty(&old
->mnt_expire
))
1088 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1096 return ERR_PTR(err
);
1099 static void cleanup_mnt(struct mount
*mnt
)
1102 * This probably indicates that somebody messed
1103 * up a mnt_want/drop_write() pair. If this
1104 * happens, the filesystem was probably unable
1105 * to make r/w->r/o transitions.
1108 * The locking used to deal with mnt_count decrement provides barriers,
1109 * so mnt_get_writers() below is safe.
1111 WARN_ON(mnt_get_writers(mnt
));
1112 if (unlikely(mnt
->mnt_pins
.first
))
1114 fsnotify_vfsmount_delete(&mnt
->mnt
);
1115 dput(mnt
->mnt
.mnt_root
);
1116 deactivate_super(mnt
->mnt
.mnt_sb
);
1118 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1121 static void __cleanup_mnt(struct rcu_head
*head
)
1123 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1126 static LLIST_HEAD(delayed_mntput_list
);
1127 static void delayed_mntput(struct work_struct
*unused
)
1129 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1130 struct llist_node
*next
;
1132 for (; node
; node
= next
) {
1133 next
= llist_next(node
);
1134 cleanup_mnt(llist_entry(node
, struct mount
, mnt_llist
));
1137 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1139 static void mntput_no_expire(struct mount
*mnt
)
1142 mnt_add_count(mnt
, -1);
1143 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
1148 if (mnt_get_count(mnt
)) {
1150 unlock_mount_hash();
1153 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1155 unlock_mount_hash();
1158 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1161 list_del(&mnt
->mnt_instance
);
1163 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1164 struct mount
*p
, *tmp
;
1165 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1169 unlock_mount_hash();
1171 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1172 struct task_struct
*task
= current
;
1173 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1174 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1175 if (!task_work_add(task
, &mnt
->mnt_rcu
, true))
1178 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1179 schedule_delayed_work(&delayed_mntput_work
, 1);
1185 void mntput(struct vfsmount
*mnt
)
1188 struct mount
*m
= real_mount(mnt
);
1189 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1190 if (unlikely(m
->mnt_expiry_mark
))
1191 m
->mnt_expiry_mark
= 0;
1192 mntput_no_expire(m
);
1195 EXPORT_SYMBOL(mntput
);
1197 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1200 mnt_add_count(real_mount(mnt
), 1);
1203 EXPORT_SYMBOL(mntget
);
1205 struct vfsmount
*mnt_clone_internal(struct path
*path
)
1208 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1211 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1215 static inline void mangle(struct seq_file
*m
, const char *s
)
1217 seq_escape(m
, s
, " \t\n\\");
1221 * Simple .show_options callback for filesystems which don't want to
1222 * implement more complex mount option showing.
1224 * See also save_mount_options().
1226 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1228 const char *options
;
1231 options
= rcu_dereference(root
->d_sb
->s_options
);
1233 if (options
!= NULL
&& options
[0]) {
1241 EXPORT_SYMBOL(generic_show_options
);
1244 * If filesystem uses generic_show_options(), this function should be
1245 * called from the fill_super() callback.
1247 * The .remount_fs callback usually needs to be handled in a special
1248 * way, to make sure, that previous options are not overwritten if the
1251 * Also note, that if the filesystem's .remount_fs function doesn't
1252 * reset all options to their default value, but changes only newly
1253 * given options, then the displayed options will not reflect reality
1256 void save_mount_options(struct super_block
*sb
, char *options
)
1258 BUG_ON(sb
->s_options
);
1259 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1261 EXPORT_SYMBOL(save_mount_options
);
1263 void replace_mount_options(struct super_block
*sb
, char *options
)
1265 char *old
= sb
->s_options
;
1266 rcu_assign_pointer(sb
->s_options
, options
);
1272 EXPORT_SYMBOL(replace_mount_options
);
1274 #ifdef CONFIG_PROC_FS
1275 /* iterator; we want it to have access to namespace_sem, thus here... */
1276 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1278 struct proc_mounts
*p
= m
->private;
1280 down_read(&namespace_sem
);
1281 if (p
->cached_event
== p
->ns
->event
) {
1282 void *v
= p
->cached_mount
;
1283 if (*pos
== p
->cached_index
)
1285 if (*pos
== p
->cached_index
+ 1) {
1286 v
= seq_list_next(v
, &p
->ns
->list
, &p
->cached_index
);
1287 return p
->cached_mount
= v
;
1291 p
->cached_event
= p
->ns
->event
;
1292 p
->cached_mount
= seq_list_start(&p
->ns
->list
, *pos
);
1293 p
->cached_index
= *pos
;
1294 return p
->cached_mount
;
1297 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1299 struct proc_mounts
*p
= m
->private;
1301 p
->cached_mount
= seq_list_next(v
, &p
->ns
->list
, pos
);
1302 p
->cached_index
= *pos
;
1303 return p
->cached_mount
;
1306 static void m_stop(struct seq_file
*m
, void *v
)
1308 up_read(&namespace_sem
);
1311 static int m_show(struct seq_file
*m
, void *v
)
1313 struct proc_mounts
*p
= m
->private;
1314 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1315 return p
->show(m
, &r
->mnt
);
1318 const struct seq_operations mounts_op
= {
1324 #endif /* CONFIG_PROC_FS */
1327 * may_umount_tree - check if a mount tree is busy
1328 * @mnt: root of mount tree
1330 * This is called to check if a tree of mounts has any
1331 * open files, pwds, chroots or sub mounts that are
1334 int may_umount_tree(struct vfsmount
*m
)
1336 struct mount
*mnt
= real_mount(m
);
1337 int actual_refs
= 0;
1338 int minimum_refs
= 0;
1342 /* write lock needed for mnt_get_count */
1344 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1345 actual_refs
+= mnt_get_count(p
);
1348 unlock_mount_hash();
1350 if (actual_refs
> minimum_refs
)
1356 EXPORT_SYMBOL(may_umount_tree
);
1359 * may_umount - check if a mount point is busy
1360 * @mnt: root of mount
1362 * This is called to check if a mount point has any
1363 * open files, pwds, chroots or sub mounts. If the
1364 * mount has sub mounts this will return busy
1365 * regardless of whether the sub mounts are busy.
1367 * Doesn't take quota and stuff into account. IOW, in some cases it will
1368 * give false negatives. The main reason why it's here is that we need
1369 * a non-destructive way to look for easily umountable filesystems.
1371 int may_umount(struct vfsmount
*mnt
)
1374 down_read(&namespace_sem
);
1376 if (propagate_mount_busy(real_mount(mnt
), 2))
1378 unlock_mount_hash();
1379 up_read(&namespace_sem
);
1383 EXPORT_SYMBOL(may_umount
);
1385 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
1387 static void namespace_unlock(void)
1389 struct hlist_head head
;
1391 hlist_move_list(&unmounted
, &head
);
1393 up_write(&namespace_sem
);
1395 if (likely(hlist_empty(&head
)))
1400 group_pin_kill(&head
);
1403 static inline void namespace_lock(void)
1405 down_write(&namespace_sem
);
1408 enum umount_tree_flags
{
1410 UMOUNT_PROPAGATE
= 2,
1411 UMOUNT_CONNECTED
= 4,
1414 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1416 /* Leaving mounts connected is only valid for lazy umounts */
1417 if (how
& UMOUNT_SYNC
)
1420 /* A mount without a parent has nothing to be connected to */
1421 if (!mnt_has_parent(mnt
))
1424 /* Because the reference counting rules change when mounts are
1425 * unmounted and connected, umounted mounts may not be
1426 * connected to mounted mounts.
1428 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1431 /* Has it been requested that the mount remain connected? */
1432 if (how
& UMOUNT_CONNECTED
)
1435 /* Is the mount locked such that it needs to remain connected? */
1436 if (IS_MNT_LOCKED(mnt
))
1439 /* By default disconnect the mount */
1444 * mount_lock must be held
1445 * namespace_sem must be held for write
1447 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1449 LIST_HEAD(tmp_list
);
1452 if (how
& UMOUNT_PROPAGATE
)
1453 propagate_mount_unlock(mnt
);
1455 /* Gather the mounts to umount */
1456 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1457 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1458 list_move(&p
->mnt_list
, &tmp_list
);
1461 /* Hide the mounts from mnt_mounts */
1462 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1463 list_del_init(&p
->mnt_child
);
1466 /* Add propogated mounts to the tmp_list */
1467 if (how
& UMOUNT_PROPAGATE
)
1468 propagate_umount(&tmp_list
);
1470 while (!list_empty(&tmp_list
)) {
1471 struct mnt_namespace
*ns
;
1473 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1474 list_del_init(&p
->mnt_expire
);
1475 list_del_init(&p
->mnt_list
);
1479 __touch_mnt_namespace(ns
);
1482 if (how
& UMOUNT_SYNC
)
1483 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1485 disconnect
= disconnect_mount(p
, how
);
1487 pin_insert_group(&p
->mnt_umount
, &p
->mnt_parent
->mnt
,
1488 disconnect
? &unmounted
: NULL
);
1489 if (mnt_has_parent(p
)) {
1490 mnt_add_count(p
->mnt_parent
, -1);
1492 /* Don't forget about p */
1493 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1498 change_mnt_propagation(p
, MS_PRIVATE
);
1502 static void shrink_submounts(struct mount
*mnt
);
1504 static int do_umount(struct mount
*mnt
, int flags
)
1506 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1509 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1514 * Allow userspace to request a mountpoint be expired rather than
1515 * unmounting unconditionally. Unmount only happens if:
1516 * (1) the mark is already set (the mark is cleared by mntput())
1517 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1519 if (flags
& MNT_EXPIRE
) {
1520 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1521 flags
& (MNT_FORCE
| MNT_DETACH
))
1525 * probably don't strictly need the lock here if we examined
1526 * all race cases, but it's a slowpath.
1529 if (mnt_get_count(mnt
) != 2) {
1530 unlock_mount_hash();
1533 unlock_mount_hash();
1535 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1540 * If we may have to abort operations to get out of this
1541 * mount, and they will themselves hold resources we must
1542 * allow the fs to do things. In the Unix tradition of
1543 * 'Gee thats tricky lets do it in userspace' the umount_begin
1544 * might fail to complete on the first run through as other tasks
1545 * must return, and the like. Thats for the mount program to worry
1546 * about for the moment.
1549 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1550 sb
->s_op
->umount_begin(sb
);
1554 * No sense to grab the lock for this test, but test itself looks
1555 * somewhat bogus. Suggestions for better replacement?
1556 * Ho-hum... In principle, we might treat that as umount + switch
1557 * to rootfs. GC would eventually take care of the old vfsmount.
1558 * Actually it makes sense, especially if rootfs would contain a
1559 * /reboot - static binary that would close all descriptors and
1560 * call reboot(9). Then init(8) could umount root and exec /reboot.
1562 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1564 * Special case for "unmounting" root ...
1565 * we just try to remount it readonly.
1567 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1569 down_write(&sb
->s_umount
);
1570 if (!(sb
->s_flags
& MS_RDONLY
))
1571 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1572 up_write(&sb
->s_umount
);
1580 if (flags
& MNT_DETACH
) {
1581 if (!list_empty(&mnt
->mnt_list
))
1582 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1585 shrink_submounts(mnt
);
1587 if (!propagate_mount_busy(mnt
, 2)) {
1588 if (!list_empty(&mnt
->mnt_list
))
1589 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1593 unlock_mount_hash();
1599 * __detach_mounts - lazily unmount all mounts on the specified dentry
1601 * During unlink, rmdir, and d_drop it is possible to loose the path
1602 * to an existing mountpoint, and wind up leaking the mount.
1603 * detach_mounts allows lazily unmounting those mounts instead of
1606 * The caller may hold dentry->d_inode->i_mutex.
1608 void __detach_mounts(struct dentry
*dentry
)
1610 struct mountpoint
*mp
;
1615 mp
= lookup_mountpoint(dentry
);
1616 if (IS_ERR_OR_NULL(mp
))
1620 while (!hlist_empty(&mp
->m_list
)) {
1621 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1622 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1623 hlist_add_head(&mnt
->mnt_umount
.s_list
, &unmounted
);
1626 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1630 unlock_mount_hash();
1635 * Is the caller allowed to modify his namespace?
1637 static inline bool may_mount(void)
1639 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1643 * Now umount can handle mount points as well as block devices.
1644 * This is important for filesystems which use unnamed block devices.
1646 * We now support a flag for forced unmount like the other 'big iron'
1647 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1650 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1655 int lookup_flags
= 0;
1657 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1663 if (!(flags
& UMOUNT_NOFOLLOW
))
1664 lookup_flags
|= LOOKUP_FOLLOW
;
1666 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1669 mnt
= real_mount(path
.mnt
);
1671 if (path
.dentry
!= path
.mnt
->mnt_root
)
1673 if (!check_mnt(mnt
))
1675 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1678 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1681 retval
= do_umount(mnt
, flags
);
1683 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1685 mntput_no_expire(mnt
);
1690 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1693 * The 2.0 compatible umount. No flags.
1695 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1697 return sys_umount(name
, 0);
1702 static bool is_mnt_ns_file(struct dentry
*dentry
)
1704 /* Is this a proxy for a mount namespace? */
1705 return dentry
->d_op
== &ns_dentry_operations
&&
1706 dentry
->d_fsdata
== &mntns_operations
;
1709 struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1711 return container_of(ns
, struct mnt_namespace
, ns
);
1714 static bool mnt_ns_loop(struct dentry
*dentry
)
1716 /* Could bind mounting the mount namespace inode cause a
1717 * mount namespace loop?
1719 struct mnt_namespace
*mnt_ns
;
1720 if (!is_mnt_ns_file(dentry
))
1723 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1724 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1727 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1730 struct mount
*res
, *p
, *q
, *r
, *parent
;
1732 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1733 return ERR_PTR(-EINVAL
);
1735 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1736 return ERR_PTR(-EINVAL
);
1738 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1742 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1745 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1747 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1750 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1751 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1752 IS_MNT_UNBINDABLE(s
)) {
1753 s
= skip_mnt_tree(s
);
1756 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1757 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1758 s
= skip_mnt_tree(s
);
1761 while (p
!= s
->mnt_parent
) {
1767 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1771 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1772 attach_mnt(q
, parent
, p
->mnt_mp
);
1773 unlock_mount_hash();
1780 umount_tree(res
, UMOUNT_SYNC
);
1781 unlock_mount_hash();
1786 /* Caller should check returned pointer for errors */
1788 struct vfsmount
*collect_mounts(struct path
*path
)
1792 if (!check_mnt(real_mount(path
->mnt
)))
1793 tree
= ERR_PTR(-EINVAL
);
1795 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1796 CL_COPY_ALL
| CL_PRIVATE
);
1799 return ERR_CAST(tree
);
1803 void drop_collected_mounts(struct vfsmount
*mnt
)
1807 umount_tree(real_mount(mnt
), UMOUNT_SYNC
);
1808 unlock_mount_hash();
1813 * clone_private_mount - create a private clone of a path
1815 * This creates a new vfsmount, which will be the clone of @path. The new will
1816 * not be attached anywhere in the namespace and will be private (i.e. changes
1817 * to the originating mount won't be propagated into this).
1819 * Release with mntput().
1821 struct vfsmount
*clone_private_mount(struct path
*path
)
1823 struct mount
*old_mnt
= real_mount(path
->mnt
);
1824 struct mount
*new_mnt
;
1826 if (IS_MNT_UNBINDABLE(old_mnt
))
1827 return ERR_PTR(-EINVAL
);
1829 down_read(&namespace_sem
);
1830 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1831 up_read(&namespace_sem
);
1832 if (IS_ERR(new_mnt
))
1833 return ERR_CAST(new_mnt
);
1835 return &new_mnt
->mnt
;
1837 EXPORT_SYMBOL_GPL(clone_private_mount
);
1839 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1840 struct vfsmount
*root
)
1843 int res
= f(root
, arg
);
1846 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1847 res
= f(&mnt
->mnt
, arg
);
1853 EXPORT_SYMBOL(iterate_mounts
);
1855 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1859 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1860 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1861 mnt_release_group_id(p
);
1865 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1869 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1870 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1871 int err
= mnt_alloc_group_id(p
);
1873 cleanup_group_ids(mnt
, p
);
1882 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
1884 unsigned int max
= READ_ONCE(sysctl_mount_max
);
1885 unsigned int mounts
= 0, old
, pending
, sum
;
1888 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1892 pending
= ns
->pending_mounts
;
1893 sum
= old
+ pending
;
1897 (mounts
> (max
- sum
)))
1900 ns
->pending_mounts
= pending
+ mounts
;
1905 * @source_mnt : mount tree to be attached
1906 * @nd : place the mount tree @source_mnt is attached
1907 * @parent_nd : if non-null, detach the source_mnt from its parent and
1908 * store the parent mount and mountpoint dentry.
1909 * (done when source_mnt is moved)
1911 * NOTE: in the table below explains the semantics when a source mount
1912 * of a given type is attached to a destination mount of a given type.
1913 * ---------------------------------------------------------------------------
1914 * | BIND MOUNT OPERATION |
1915 * |**************************************************************************
1916 * | source-->| shared | private | slave | unbindable |
1920 * |**************************************************************************
1921 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1923 * |non-shared| shared (+) | private | slave (*) | invalid |
1924 * ***************************************************************************
1925 * A bind operation clones the source mount and mounts the clone on the
1926 * destination mount.
1928 * (++) the cloned mount is propagated to all the mounts in the propagation
1929 * tree of the destination mount and the cloned mount is added to
1930 * the peer group of the source mount.
1931 * (+) the cloned mount is created under the destination mount and is marked
1932 * as shared. The cloned mount is added to the peer group of the source
1934 * (+++) the mount is propagated to all the mounts in the propagation tree
1935 * of the destination mount and the cloned mount is made slave
1936 * of the same master as that of the source mount. The cloned mount
1937 * is marked as 'shared and slave'.
1938 * (*) the cloned mount is made a slave of the same master as that of the
1941 * ---------------------------------------------------------------------------
1942 * | MOVE MOUNT OPERATION |
1943 * |**************************************************************************
1944 * | source-->| shared | private | slave | unbindable |
1948 * |**************************************************************************
1949 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1951 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1952 * ***************************************************************************
1954 * (+) the mount is moved to the destination. And is then propagated to
1955 * all the mounts in the propagation tree of the destination mount.
1956 * (+*) the mount is moved to the destination.
1957 * (+++) the mount is moved to the destination and is then propagated to
1958 * all the mounts belonging to the destination mount's propagation tree.
1959 * the mount is marked as 'shared and slave'.
1960 * (*) the mount continues to be a slave at the new location.
1962 * if the source mount is a tree, the operations explained above is
1963 * applied to each mount in the tree.
1964 * Must be called without spinlocks held, since this function can sleep
1967 static int attach_recursive_mnt(struct mount
*source_mnt
,
1968 struct mount
*dest_mnt
,
1969 struct mountpoint
*dest_mp
,
1970 struct path
*parent_path
)
1972 HLIST_HEAD(tree_list
);
1973 struct mnt_namespace
*ns
= dest_mnt
->mnt_ns
;
1974 struct mountpoint
*smp
;
1975 struct mount
*child
, *p
;
1976 struct hlist_node
*n
;
1979 /* Is there space to add these mounts to the mount namespace? */
1981 err
= count_mounts(ns
, source_mnt
);
1986 /* Preallocate a mountpoint in case the new mounts need
1987 * to be tucked under other mounts.
1989 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
1991 return PTR_ERR(smp
);
1993 if (IS_MNT_SHARED(dest_mnt
)) {
1994 err
= invent_group_ids(source_mnt
, true);
1997 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2000 goto out_cleanup_ids
;
2001 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2007 detach_mnt(source_mnt
, parent_path
);
2008 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
2009 touch_mnt_namespace(source_mnt
->mnt_ns
);
2011 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2012 commit_tree(source_mnt
);
2015 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2017 hlist_del_init(&child
->mnt_hash
);
2018 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2019 child
->mnt_mountpoint
);
2021 mnt_change_mountpoint(child
, smp
, q
);
2024 put_mountpoint(smp
);
2025 unlock_mount_hash();
2030 while (!hlist_empty(&tree_list
)) {
2031 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2032 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2033 umount_tree(child
, UMOUNT_SYNC
);
2035 unlock_mount_hash();
2036 cleanup_group_ids(source_mnt
, NULL
);
2038 ns
->pending_mounts
= 0;
2039 read_seqlock_excl(&mount_lock
);
2040 put_mountpoint(smp
);
2041 read_sequnlock_excl(&mount_lock
);
2046 static struct mountpoint
*lock_mount(struct path
*path
)
2048 struct vfsmount
*mnt
;
2049 struct dentry
*dentry
= path
->dentry
;
2051 mutex_lock(&dentry
->d_inode
->i_mutex
);
2052 if (unlikely(cant_mount(dentry
))) {
2053 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2054 return ERR_PTR(-ENOENT
);
2057 mnt
= lookup_mnt(path
);
2059 struct mountpoint
*mp
= get_mountpoint(dentry
);
2062 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2068 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
2071 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2075 static void unlock_mount(struct mountpoint
*where
)
2077 struct dentry
*dentry
= where
->m_dentry
;
2079 read_seqlock_excl(&mount_lock
);
2080 put_mountpoint(where
);
2081 read_sequnlock_excl(&mount_lock
);
2084 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2087 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2089 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
2092 if (d_is_dir(mp
->m_dentry
) !=
2093 d_is_dir(mnt
->mnt
.mnt_root
))
2096 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
2100 * Sanity check the flags to change_mnt_propagation.
2103 static int flags_to_propagation_type(int flags
)
2105 int type
= flags
& ~(MS_REC
| MS_SILENT
);
2107 /* Fail if any non-propagation flags are set */
2108 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2110 /* Only one propagation flag should be set */
2111 if (!is_power_of_2(type
))
2117 * recursively change the type of the mountpoint.
2119 static int do_change_type(struct path
*path
, int flag
)
2122 struct mount
*mnt
= real_mount(path
->mnt
);
2123 int recurse
= flag
& MS_REC
;
2127 if (path
->dentry
!= path
->mnt
->mnt_root
)
2130 type
= flags_to_propagation_type(flag
);
2135 if (type
== MS_SHARED
) {
2136 err
= invent_group_ids(mnt
, recurse
);
2142 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2143 change_mnt_propagation(m
, type
);
2144 unlock_mount_hash();
2151 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2153 struct mount
*child
;
2154 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2155 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2158 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2165 * do loopback mount.
2167 static int do_loopback(struct path
*path
, const char *old_name
,
2170 struct path old_path
;
2171 struct mount
*mnt
= NULL
, *old
, *parent
;
2172 struct mountpoint
*mp
;
2174 if (!old_name
|| !*old_name
)
2176 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2181 if (mnt_ns_loop(old_path
.dentry
))
2184 mp
= lock_mount(path
);
2189 old
= real_mount(old_path
.mnt
);
2190 parent
= real_mount(path
->mnt
);
2193 if (IS_MNT_UNBINDABLE(old
))
2196 if (!check_mnt(parent
))
2199 if (!check_mnt(old
) && old_path
.dentry
->d_op
!= &ns_dentry_operations
)
2202 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
2206 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
2208 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
2215 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2217 err
= graft_tree(mnt
, parent
, mp
);
2220 umount_tree(mnt
, UMOUNT_SYNC
);
2221 unlock_mount_hash();
2226 path_put(&old_path
);
2230 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
2233 int readonly_request
= 0;
2235 if (ms_flags
& MS_RDONLY
)
2236 readonly_request
= 1;
2237 if (readonly_request
== __mnt_is_readonly(mnt
))
2240 if (readonly_request
)
2241 error
= mnt_make_readonly(real_mount(mnt
));
2243 __mnt_unmake_readonly(real_mount(mnt
));
2248 * change filesystem flags. dir should be a physical root of filesystem.
2249 * If you've mounted a non-root directory somewhere and want to do remount
2250 * on it - tough luck.
2252 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
2256 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2257 struct mount
*mnt
= real_mount(path
->mnt
);
2259 if (!check_mnt(mnt
))
2262 if (path
->dentry
!= path
->mnt
->mnt_root
)
2265 /* Don't allow changing of locked mnt flags.
2267 * No locks need to be held here while testing the various
2268 * MNT_LOCK flags because those flags can never be cleared
2269 * once they are set.
2271 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_READONLY
) &&
2272 !(mnt_flags
& MNT_READONLY
)) {
2275 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NODEV
) &&
2276 !(mnt_flags
& MNT_NODEV
)) {
2277 /* Was the nodev implicitly added in mount? */
2278 if ((mnt
->mnt_ns
->user_ns
!= &init_user_ns
) &&
2279 !(sb
->s_type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2280 mnt_flags
|= MNT_NODEV
;
2285 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOSUID
) &&
2286 !(mnt_flags
& MNT_NOSUID
)) {
2289 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOEXEC
) &&
2290 !(mnt_flags
& MNT_NOEXEC
)) {
2293 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_ATIME
) &&
2294 ((mnt
->mnt
.mnt_flags
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
))) {
2298 err
= security_sb_remount(sb
, data
);
2302 down_write(&sb
->s_umount
);
2303 if (flags
& MS_BIND
)
2304 err
= change_mount_flags(path
->mnt
, flags
);
2305 else if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
2308 err
= do_remount_sb(sb
, flags
, data
, 0);
2311 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2312 mnt
->mnt
.mnt_flags
= mnt_flags
;
2313 touch_mnt_namespace(mnt
->mnt_ns
);
2314 unlock_mount_hash();
2316 up_write(&sb
->s_umount
);
2320 static inline int tree_contains_unbindable(struct mount
*mnt
)
2323 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2324 if (IS_MNT_UNBINDABLE(p
))
2330 static int do_move_mount(struct path
*path
, const char *old_name
)
2332 struct path old_path
, parent_path
;
2335 struct mountpoint
*mp
;
2337 if (!old_name
|| !*old_name
)
2339 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2343 mp
= lock_mount(path
);
2348 old
= real_mount(old_path
.mnt
);
2349 p
= real_mount(path
->mnt
);
2352 if (!check_mnt(p
) || !check_mnt(old
))
2355 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2359 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
2362 if (!mnt_has_parent(old
))
2365 if (d_is_dir(path
->dentry
) !=
2366 d_is_dir(old_path
.dentry
))
2369 * Don't move a mount residing in a shared parent.
2371 if (IS_MNT_SHARED(old
->mnt_parent
))
2374 * Don't move a mount tree containing unbindable mounts to a destination
2375 * mount which is shared.
2377 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2380 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2384 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2388 /* if the mount is moved, it should no longer be expire
2390 list_del_init(&old
->mnt_expire
);
2395 path_put(&parent_path
);
2396 path_put(&old_path
);
2400 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2403 const char *subtype
= strchr(fstype
, '.');
2412 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2414 if (!mnt
->mnt_sb
->s_subtype
)
2420 return ERR_PTR(err
);
2424 * add a mount into a namespace's mount tree
2426 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2428 struct mountpoint
*mp
;
2429 struct mount
*parent
;
2432 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2434 mp
= lock_mount(path
);
2438 parent
= real_mount(path
->mnt
);
2440 if (unlikely(!check_mnt(parent
))) {
2441 /* that's acceptable only for automounts done in private ns */
2442 if (!(mnt_flags
& MNT_SHRINKABLE
))
2444 /* ... and for those we'd better have mountpoint still alive */
2445 if (!parent
->mnt_ns
)
2449 /* Refuse the same filesystem on the same mount point */
2451 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2452 path
->mnt
->mnt_root
== path
->dentry
)
2456 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2459 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2460 err
= graft_tree(newmnt
, parent
, mp
);
2467 static bool fs_fully_visible(struct file_system_type
*fs_type
, int *new_mnt_flags
);
2470 * create a new mount for userspace and request it to be added into the
2473 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2474 int mnt_flags
, const char *name
, void *data
)
2476 struct file_system_type
*type
;
2477 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2478 struct vfsmount
*mnt
;
2481 if (!ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
2487 type
= get_fs_type(fstype
);
2491 if (user_ns
!= &init_user_ns
) {
2492 /* Only in special cases allow devices from mounts
2493 * created outside the initial user namespace.
2495 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2497 mnt_flags
|= MNT_NODEV
| MNT_LOCK_NODEV
;
2499 if (type
->fs_flags
& FS_USERNS_VISIBLE
) {
2500 if (!fs_fully_visible(type
, &mnt_flags
)) {
2501 put_filesystem(type
);
2507 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2508 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2509 !mnt
->mnt_sb
->s_subtype
)
2510 mnt
= fs_set_subtype(mnt
, fstype
);
2512 put_filesystem(type
);
2514 return PTR_ERR(mnt
);
2516 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2522 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2524 struct mount
*mnt
= real_mount(m
);
2526 /* The new mount record should have at least 2 refs to prevent it being
2527 * expired before we get a chance to add it
2529 BUG_ON(mnt_get_count(mnt
) < 2);
2531 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2532 m
->mnt_root
== path
->dentry
) {
2537 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2541 /* remove m from any expiration list it may be on */
2542 if (!list_empty(&mnt
->mnt_expire
)) {
2544 list_del_init(&mnt
->mnt_expire
);
2553 * mnt_set_expiry - Put a mount on an expiration list
2554 * @mnt: The mount to list.
2555 * @expiry_list: The list to add the mount to.
2557 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2561 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2565 EXPORT_SYMBOL(mnt_set_expiry
);
2568 * process a list of expirable mountpoints with the intent of discarding any
2569 * mountpoints that aren't in use and haven't been touched since last we came
2572 void mark_mounts_for_expiry(struct list_head
*mounts
)
2574 struct mount
*mnt
, *next
;
2575 LIST_HEAD(graveyard
);
2577 if (list_empty(mounts
))
2583 /* extract from the expiration list every vfsmount that matches the
2584 * following criteria:
2585 * - only referenced by its parent vfsmount
2586 * - still marked for expiry (marked on the last call here; marks are
2587 * cleared by mntput())
2589 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2590 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2591 propagate_mount_busy(mnt
, 1))
2593 list_move(&mnt
->mnt_expire
, &graveyard
);
2595 while (!list_empty(&graveyard
)) {
2596 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2597 touch_mnt_namespace(mnt
->mnt_ns
);
2598 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2600 unlock_mount_hash();
2604 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2607 * Ripoff of 'select_parent()'
2609 * search the list of submounts for a given mountpoint, and move any
2610 * shrinkable submounts to the 'graveyard' list.
2612 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2614 struct mount
*this_parent
= parent
;
2615 struct list_head
*next
;
2619 next
= this_parent
->mnt_mounts
.next
;
2621 while (next
!= &this_parent
->mnt_mounts
) {
2622 struct list_head
*tmp
= next
;
2623 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2626 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2629 * Descend a level if the d_mounts list is non-empty.
2631 if (!list_empty(&mnt
->mnt_mounts
)) {
2636 if (!propagate_mount_busy(mnt
, 1)) {
2637 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2642 * All done at this level ... ascend and resume the search
2644 if (this_parent
!= parent
) {
2645 next
= this_parent
->mnt_child
.next
;
2646 this_parent
= this_parent
->mnt_parent
;
2653 * process a list of expirable mountpoints with the intent of discarding any
2654 * submounts of a specific parent mountpoint
2656 * mount_lock must be held for write
2658 static void shrink_submounts(struct mount
*mnt
)
2660 LIST_HEAD(graveyard
);
2663 /* extract submounts of 'mountpoint' from the expiration list */
2664 while (select_submounts(mnt
, &graveyard
)) {
2665 while (!list_empty(&graveyard
)) {
2666 m
= list_first_entry(&graveyard
, struct mount
,
2668 touch_mnt_namespace(m
->mnt_ns
);
2669 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2675 * Some copy_from_user() implementations do not return the exact number of
2676 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2677 * Note that this function differs from copy_from_user() in that it will oops
2678 * on bad values of `to', rather than returning a short copy.
2680 static long exact_copy_from_user(void *to
, const void __user
* from
,
2684 const char __user
*f
= from
;
2687 if (!access_ok(VERIFY_READ
, from
, n
))
2691 if (__get_user(c
, f
)) {
2702 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2712 if (!(page
= __get_free_page(GFP_KERNEL
)))
2715 /* We only care that *some* data at the address the user
2716 * gave us is valid. Just in case, we'll zero
2717 * the remainder of the page.
2719 /* copy_from_user cannot cross TASK_SIZE ! */
2720 size
= TASK_SIZE
- (unsigned long)data
;
2721 if (size
> PAGE_SIZE
)
2724 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2730 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2735 char *copy_mount_string(const void __user
*data
)
2737 return data
? strndup_user(data
, PAGE_SIZE
) : NULL
;
2741 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2742 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2744 * data is a (void *) that can point to any structure up to
2745 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2746 * information (or be NULL).
2748 * Pre-0.97 versions of mount() didn't have a flags word.
2749 * When the flags word was introduced its top half was required
2750 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2751 * Therefore, if this magic number is present, it carries no information
2752 * and must be discarded.
2754 long do_mount(const char *dev_name
, const char __user
*dir_name
,
2755 const char *type_page
, unsigned long flags
, void *data_page
)
2762 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2763 flags
&= ~MS_MGC_MSK
;
2765 /* Basic sanity checks */
2767 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2769 /* ... and get the mountpoint */
2770 retval
= user_path(dir_name
, &path
);
2774 retval
= security_sb_mount(dev_name
, &path
,
2775 type_page
, flags
, data_page
);
2776 if (!retval
&& !may_mount())
2781 /* Default to relatime unless overriden */
2782 if (!(flags
& MS_NOATIME
))
2783 mnt_flags
|= MNT_RELATIME
;
2785 /* Separate the per-mountpoint flags */
2786 if (flags
& MS_NOSUID
)
2787 mnt_flags
|= MNT_NOSUID
;
2788 if (flags
& MS_NODEV
)
2789 mnt_flags
|= MNT_NODEV
;
2790 if (flags
& MS_NOEXEC
)
2791 mnt_flags
|= MNT_NOEXEC
;
2792 if (flags
& MS_NOATIME
)
2793 mnt_flags
|= MNT_NOATIME
;
2794 if (flags
& MS_NODIRATIME
)
2795 mnt_flags
|= MNT_NODIRATIME
;
2796 if (flags
& MS_STRICTATIME
)
2797 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2798 if (flags
& MS_RDONLY
)
2799 mnt_flags
|= MNT_READONLY
;
2801 /* The default atime for remount is preservation */
2802 if ((flags
& MS_REMOUNT
) &&
2803 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
2804 MS_STRICTATIME
)) == 0)) {
2805 mnt_flags
&= ~MNT_ATIME_MASK
;
2806 mnt_flags
|= path
.mnt
->mnt_flags
& MNT_ATIME_MASK
;
2809 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2810 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2811 MS_STRICTATIME
| MS_SUBMOUNT
);
2813 if (flags
& MS_REMOUNT
)
2814 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2816 else if (flags
& MS_BIND
)
2817 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2818 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2819 retval
= do_change_type(&path
, flags
);
2820 else if (flags
& MS_MOVE
)
2821 retval
= do_move_mount(&path
, dev_name
);
2823 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2824 dev_name
, data_page
);
2830 static void free_mnt_ns(struct mnt_namespace
*ns
)
2832 ns_free_inum(&ns
->ns
);
2833 put_user_ns(ns
->user_ns
);
2838 * Assign a sequence number so we can detect when we attempt to bind
2839 * mount a reference to an older mount namespace into the current
2840 * mount namespace, preventing reference counting loops. A 64bit
2841 * number incrementing at 10Ghz will take 12,427 years to wrap which
2842 * is effectively never, so we can ignore the possibility.
2844 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2846 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2848 struct mnt_namespace
*new_ns
;
2851 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2853 return ERR_PTR(-ENOMEM
);
2854 ret
= ns_alloc_inum(&new_ns
->ns
);
2857 return ERR_PTR(ret
);
2859 new_ns
->ns
.ops
= &mntns_operations
;
2860 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2861 atomic_set(&new_ns
->count
, 1);
2862 new_ns
->root
= NULL
;
2863 INIT_LIST_HEAD(&new_ns
->list
);
2864 init_waitqueue_head(&new_ns
->poll
);
2866 new_ns
->user_ns
= get_user_ns(user_ns
);
2868 new_ns
->pending_mounts
= 0;
2872 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2873 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2875 struct mnt_namespace
*new_ns
;
2876 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2877 struct mount
*p
, *q
;
2884 if (likely(!(flags
& CLONE_NEWNS
))) {
2891 new_ns
= alloc_mnt_ns(user_ns
);
2896 /* First pass: copy the tree topology */
2897 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2898 if (user_ns
!= ns
->user_ns
)
2899 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2900 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2903 free_mnt_ns(new_ns
);
2904 return ERR_CAST(new);
2907 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2910 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2911 * as belonging to new namespace. We have already acquired a private
2912 * fs_struct, so tsk->fs->lock is not needed.
2920 if (&p
->mnt
== new_fs
->root
.mnt
) {
2921 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2924 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2925 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2929 p
= next_mnt(p
, old
);
2930 q
= next_mnt(q
, new);
2933 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2934 p
= next_mnt(p
, old
);
2947 * create_mnt_ns - creates a private namespace and adds a root filesystem
2948 * @mnt: pointer to the new root filesystem mountpoint
2950 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2952 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2953 if (!IS_ERR(new_ns
)) {
2954 struct mount
*mnt
= real_mount(m
);
2955 mnt
->mnt_ns
= new_ns
;
2958 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2965 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2967 struct mnt_namespace
*ns
;
2968 struct super_block
*s
;
2972 ns
= create_mnt_ns(mnt
);
2974 return ERR_CAST(ns
);
2976 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2977 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2982 return ERR_PTR(err
);
2984 /* trade a vfsmount reference for active sb one */
2985 s
= path
.mnt
->mnt_sb
;
2986 atomic_inc(&s
->s_active
);
2988 /* lock the sucker */
2989 down_write(&s
->s_umount
);
2990 /* ... and return the root of (sub)tree on it */
2993 EXPORT_SYMBOL(mount_subtree
);
2995 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2996 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3001 unsigned long data_page
;
3003 kernel_type
= copy_mount_string(type
);
3004 ret
= PTR_ERR(kernel_type
);
3005 if (IS_ERR(kernel_type
))
3008 kernel_dev
= copy_mount_string(dev_name
);
3009 ret
= PTR_ERR(kernel_dev
);
3010 if (IS_ERR(kernel_dev
))
3013 ret
= copy_mount_options(data
, &data_page
);
3017 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
,
3018 (void *) data_page
);
3020 free_page(data_page
);
3030 * Return true if path is reachable from root
3032 * namespace_sem or mount_lock is held
3034 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
3035 const struct path
*root
)
3037 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
3038 dentry
= mnt
->mnt_mountpoint
;
3039 mnt
= mnt
->mnt_parent
;
3041 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
3044 int path_is_under(struct path
*path1
, struct path
*path2
)
3047 read_seqlock_excl(&mount_lock
);
3048 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
3049 read_sequnlock_excl(&mount_lock
);
3052 EXPORT_SYMBOL(path_is_under
);
3055 * pivot_root Semantics:
3056 * Moves the root file system of the current process to the directory put_old,
3057 * makes new_root as the new root file system of the current process, and sets
3058 * root/cwd of all processes which had them on the current root to new_root.
3061 * The new_root and put_old must be directories, and must not be on the
3062 * same file system as the current process root. The put_old must be
3063 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3064 * pointed to by put_old must yield the same directory as new_root. No other
3065 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3067 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3068 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3069 * in this situation.
3072 * - we don't move root/cwd if they are not at the root (reason: if something
3073 * cared enough to change them, it's probably wrong to force them elsewhere)
3074 * - it's okay to pick a root that isn't the root of a file system, e.g.
3075 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3076 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3079 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3080 const char __user
*, put_old
)
3082 struct path
new, old
, parent_path
, root_parent
, root
;
3083 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
3084 struct mountpoint
*old_mp
, *root_mp
;
3090 error
= user_path_dir(new_root
, &new);
3094 error
= user_path_dir(put_old
, &old
);
3098 error
= security_sb_pivotroot(&old
, &new);
3102 get_fs_root(current
->fs
, &root
);
3103 old_mp
= lock_mount(&old
);
3104 error
= PTR_ERR(old_mp
);
3109 new_mnt
= real_mount(new.mnt
);
3110 root_mnt
= real_mount(root
.mnt
);
3111 old_mnt
= real_mount(old
.mnt
);
3112 if (IS_MNT_SHARED(old_mnt
) ||
3113 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
3114 IS_MNT_SHARED(root_mnt
->mnt_parent
))
3116 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3118 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3121 if (d_unlinked(new.dentry
))
3124 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3125 goto out4
; /* loop, on the same file system */
3127 if (root
.mnt
->mnt_root
!= root
.dentry
)
3128 goto out4
; /* not a mountpoint */
3129 if (!mnt_has_parent(root_mnt
))
3130 goto out4
; /* not attached */
3131 root_mp
= root_mnt
->mnt_mp
;
3132 if (new.mnt
->mnt_root
!= new.dentry
)
3133 goto out4
; /* not a mountpoint */
3134 if (!mnt_has_parent(new_mnt
))
3135 goto out4
; /* not attached */
3136 /* make sure we can reach put_old from new_root */
3137 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3139 /* make certain new is below the root */
3140 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3142 root_mp
->m_count
++; /* pin it so it won't go away */
3144 detach_mnt(new_mnt
, &parent_path
);
3145 detach_mnt(root_mnt
, &root_parent
);
3146 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3147 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3148 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3150 /* mount old root on put_old */
3151 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3152 /* mount new_root on / */
3153 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
3154 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3155 /* A moved mount should not expire automatically */
3156 list_del_init(&new_mnt
->mnt_expire
);
3157 put_mountpoint(root_mp
);
3158 unlock_mount_hash();
3159 chroot_fs_refs(&root
, &new);
3162 unlock_mount(old_mp
);
3164 path_put(&root_parent
);
3165 path_put(&parent_path
);
3177 static void __init
init_mount_tree(void)
3179 struct vfsmount
*mnt
;
3180 struct mnt_namespace
*ns
;
3182 struct file_system_type
*type
;
3184 type
= get_fs_type("rootfs");
3186 panic("Can't find rootfs type");
3187 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
3188 put_filesystem(type
);
3190 panic("Can't create rootfs");
3192 ns
= create_mnt_ns(mnt
);
3194 panic("Can't allocate initial namespace");
3196 init_task
.nsproxy
->mnt_ns
= ns
;
3200 root
.dentry
= mnt
->mnt_root
;
3201 mnt
->mnt_flags
|= MNT_LOCKED
;
3203 set_fs_pwd(current
->fs
, &root
);
3204 set_fs_root(current
->fs
, &root
);
3207 void __init
mnt_init(void)
3212 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
3213 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3215 mount_hashtable
= alloc_large_system_hash("Mount-cache",
3216 sizeof(struct hlist_head
),
3219 &m_hash_shift
, &m_hash_mask
, 0, 0);
3220 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
3221 sizeof(struct hlist_head
),
3224 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
3226 if (!mount_hashtable
|| !mountpoint_hashtable
)
3227 panic("Failed to allocate mount hash table\n");
3229 for (u
= 0; u
<= m_hash_mask
; u
++)
3230 INIT_HLIST_HEAD(&mount_hashtable
[u
]);
3231 for (u
= 0; u
<= mp_hash_mask
; u
++)
3232 INIT_HLIST_HEAD(&mountpoint_hashtable
[u
]);
3238 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
3240 fs_kobj
= kobject_create_and_add("fs", NULL
);
3242 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
3247 void put_mnt_ns(struct mnt_namespace
*ns
)
3249 if (!atomic_dec_and_test(&ns
->count
))
3251 drop_collected_mounts(&ns
->root
->mnt
);
3255 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
3257 struct vfsmount
*mnt
;
3258 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
3261 * it is a longterm mount, don't release mnt until
3262 * we unmount before file sys is unregistered
3264 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
3268 EXPORT_SYMBOL_GPL(kern_mount_data
);
3270 void kern_unmount(struct vfsmount
*mnt
)
3272 /* release long term mount so mount point can be released */
3273 if (!IS_ERR_OR_NULL(mnt
)) {
3274 real_mount(mnt
)->mnt_ns
= NULL
;
3275 synchronize_rcu(); /* yecchhh... */
3279 EXPORT_SYMBOL(kern_unmount
);
3281 bool our_mnt(struct vfsmount
*mnt
)
3283 return check_mnt(real_mount(mnt
));
3286 bool current_chrooted(void)
3288 /* Does the current process have a non-standard root */
3289 struct path ns_root
;
3290 struct path fs_root
;
3293 /* Find the namespace root */
3294 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
3295 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
3297 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
3300 get_fs_root(current
->fs
, &fs_root
);
3302 chrooted
= !path_equal(&fs_root
, &ns_root
);
3310 static bool fs_fully_visible(struct file_system_type
*type
, int *new_mnt_flags
)
3312 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
3313 int new_flags
= *new_mnt_flags
;
3315 bool visible
= false;
3320 down_read(&namespace_sem
);
3321 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
3322 struct mount
*child
;
3325 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
3328 /* This mount is not fully visible if it's root directory
3329 * is not the root directory of the filesystem.
3331 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
3334 /* Read the mount flags and filter out flags that
3335 * may safely be ignored.
3337 mnt_flags
= mnt
->mnt
.mnt_flags
;
3338 if (mnt
->mnt
.mnt_sb
->s_iflags
& SB_I_NOEXEC
)
3339 mnt_flags
&= ~(MNT_LOCK_NOSUID
| MNT_LOCK_NOEXEC
);
3341 /* Don't miss readonly hidden in the superblock flags */
3342 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_RDONLY
)
3343 mnt_flags
|= MNT_LOCK_READONLY
;
3345 /* Verify the mount flags are equal to or more permissive
3346 * than the proposed new mount.
3348 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
3349 !(new_flags
& MNT_READONLY
))
3351 if ((mnt_flags
& MNT_LOCK_NODEV
) &&
3352 !(new_flags
& MNT_NODEV
))
3354 if ((mnt_flags
& MNT_LOCK_NOSUID
) &&
3355 !(new_flags
& MNT_NOSUID
))
3357 if ((mnt_flags
& MNT_LOCK_NOEXEC
) &&
3358 !(new_flags
& MNT_NOEXEC
))
3360 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
3361 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
3364 /* This mount is not fully visible if there are any
3365 * locked child mounts that cover anything except for
3366 * empty directories.
3368 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
3369 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
3370 /* Only worry about locked mounts */
3371 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
3373 /* Is the directory permanetly empty? */
3374 if (!is_empty_dir_inode(inode
))
3377 /* Preserve the locked attributes */
3378 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
3388 up_read(&namespace_sem
);
3392 bool mnt_may_suid(struct vfsmount
*mnt
)
3395 * Foreign mounts (accessed via fchdir or through /proc
3396 * symlinks) are always treated as if they are nosuid. This
3397 * prevents namespaces from trusting potentially unsafe
3398 * suid/sgid bits, file caps, or security labels that originate
3399 * in other namespaces.
3401 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
3402 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
3405 static struct ns_common
*mntns_get(struct task_struct
*task
)
3407 struct ns_common
*ns
= NULL
;
3408 struct nsproxy
*nsproxy
;
3411 nsproxy
= task
->nsproxy
;
3413 ns
= &nsproxy
->mnt_ns
->ns
;
3414 get_mnt_ns(to_mnt_ns(ns
));
3421 static void mntns_put(struct ns_common
*ns
)
3423 put_mnt_ns(to_mnt_ns(ns
));
3426 static int mntns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
3428 struct fs_struct
*fs
= current
->fs
;
3429 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
);
3432 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3433 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
3434 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
3441 put_mnt_ns(nsproxy
->mnt_ns
);
3442 nsproxy
->mnt_ns
= mnt_ns
;
3445 root
.mnt
= &mnt_ns
->root
->mnt
;
3446 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
3448 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
3451 /* Update the pwd and root */
3452 set_fs_pwd(fs
, &root
);
3453 set_fs_root(fs
, &root
);
3459 const struct proc_ns_operations mntns_operations
= {
3461 .type
= CLONE_NEWNS
,
3464 .install
= mntns_install
,