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 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 void drop_mountpoint(struct fs_pin
*p
)
195 struct mount
*m
= container_of(p
, struct mount
, mnt_umount
);
196 dput(m
->mnt_ex_mountpoint
);
201 static struct mount
*alloc_vfsmnt(const char *name
)
203 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
207 err
= mnt_alloc_id(mnt
);
212 mnt
->mnt_devname
= kstrdup_const(name
, GFP_KERNEL
);
213 if (!mnt
->mnt_devname
)
218 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
220 goto out_free_devname
;
222 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
225 mnt
->mnt_writers
= 0;
228 INIT_HLIST_NODE(&mnt
->mnt_hash
);
229 INIT_LIST_HEAD(&mnt
->mnt_child
);
230 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
231 INIT_LIST_HEAD(&mnt
->mnt_list
);
232 INIT_LIST_HEAD(&mnt
->mnt_expire
);
233 INIT_LIST_HEAD(&mnt
->mnt_share
);
234 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
235 INIT_LIST_HEAD(&mnt
->mnt_slave
);
236 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
240 init_fs_pin(&mnt
->mnt_umount
, drop_mountpoint
);
246 kfree_const(mnt
->mnt_devname
);
251 kmem_cache_free(mnt_cache
, mnt
);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount
*mnt
)
276 if (mnt
->mnt_flags
& MNT_READONLY
)
278 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
284 static inline void mnt_inc_writers(struct mount
*mnt
)
287 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
293 static inline void mnt_dec_writers(struct mount
*mnt
)
296 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
302 static unsigned int mnt_get_writers(struct mount
*mnt
)
305 unsigned int count
= 0;
308 for_each_possible_cpu(cpu
) {
309 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
314 return mnt
->mnt_writers
;
318 static int mnt_is_readonly(struct vfsmount
*mnt
)
320 if (mnt
->mnt_sb
->s_readonly_remount
)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt
);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount
*m
)
345 struct mount
*mnt
= real_mount(m
);
349 mnt_inc_writers(mnt
);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m
)) {
365 mnt_dec_writers(mnt
);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount
*m
)
386 sb_start_write(m
->mnt_sb
);
387 ret
= __mnt_want_write(m
);
389 sb_end_write(m
->mnt_sb
);
392 EXPORT_SYMBOL_GPL(mnt_want_write
);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount
*mnt
)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt
))
412 mnt_inc_writers(real_mount(mnt
));
416 EXPORT_SYMBOL_GPL(mnt_clone_write
);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file
*file
)
427 if (!(file
->f_mode
& FMODE_WRITER
))
428 return __mnt_want_write(file
->f_path
.mnt
);
430 return mnt_clone_write(file
->f_path
.mnt
);
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 int mnt_want_write_file(struct file
*file
)
444 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
445 ret
= __mnt_want_write_file(file
);
447 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
450 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
460 void __mnt_drop_write(struct vfsmount
*mnt
)
463 mnt_dec_writers(real_mount(mnt
));
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
475 void mnt_drop_write(struct vfsmount
*mnt
)
477 __mnt_drop_write(mnt
);
478 sb_end_write(mnt
->mnt_sb
);
480 EXPORT_SYMBOL_GPL(mnt_drop_write
);
482 void __mnt_drop_write_file(struct file
*file
)
484 __mnt_drop_write(file
->f_path
.mnt
);
487 void mnt_drop_write_file(struct file
*file
)
489 mnt_drop_write(file
->f_path
.mnt
);
491 EXPORT_SYMBOL(mnt_drop_write_file
);
493 static int mnt_make_readonly(struct mount
*mnt
)
498 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
521 if (mnt_get_writers(mnt
) > 0)
524 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
530 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
535 static void __mnt_unmake_readonly(struct mount
*mnt
)
538 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
542 int sb_prepare_remount_readonly(struct super_block
*sb
)
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb
->s_remove_count
))
552 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
553 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
554 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
556 if (mnt_get_writers(mnt
) > 0) {
562 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
566 sb
->s_readonly_remount
= 1;
569 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
570 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
571 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
578 static void free_vfsmnt(struct mount
*mnt
)
580 kfree(mnt
->mnt
.data
);
581 kfree_const(mnt
->mnt_devname
);
583 free_percpu(mnt
->mnt_pcp
);
585 kmem_cache_free(mnt_cache
, mnt
);
588 static void delayed_free_vfsmnt(struct rcu_head
*head
)
590 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
593 /* call under rcu_read_lock */
594 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
597 if (read_seqretry(&mount_lock
, seq
))
601 mnt
= real_mount(bastard
);
602 mnt_add_count(mnt
, 1);
603 if (likely(!read_seqretry(&mount_lock
, seq
)))
605 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
606 mnt_add_count(mnt
, -1);
612 /* call under rcu_read_lock */
613 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
615 int res
= __legitimize_mnt(bastard
, seq
);
618 if (unlikely(res
< 0)) {
627 * find the first mount at @dentry on vfsmount @mnt.
628 * call under rcu_read_lock()
630 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
632 struct hlist_head
*head
= m_hash(mnt
, dentry
);
635 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
636 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
642 * find the last mount at @dentry on vfsmount @mnt.
643 * mount_lock must be held.
645 struct mount
*__lookup_mnt_last(struct vfsmount
*mnt
, struct dentry
*dentry
)
647 struct mount
*p
, *res
= NULL
;
648 p
= __lookup_mnt(mnt
, dentry
);
651 if (!(p
->mnt
.mnt_flags
& MNT_UMOUNT
))
653 hlist_for_each_entry_continue(p
, mnt_hash
) {
654 if (&p
->mnt_parent
->mnt
!= mnt
|| p
->mnt_mountpoint
!= dentry
)
656 if (!(p
->mnt
.mnt_flags
& MNT_UMOUNT
))
664 * lookup_mnt - Return the first child mount mounted at path
666 * "First" means first mounted chronologically. If you create the
669 * mount /dev/sda1 /mnt
670 * mount /dev/sda2 /mnt
671 * mount /dev/sda3 /mnt
673 * Then lookup_mnt() on the base /mnt dentry in the root mount will
674 * return successively the root dentry and vfsmount of /dev/sda1, then
675 * /dev/sda2, then /dev/sda3, then NULL.
677 * lookup_mnt takes a reference to the found vfsmount.
679 struct vfsmount
*lookup_mnt(struct path
*path
)
681 struct mount
*child_mnt
;
687 seq
= read_seqbegin(&mount_lock
);
688 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
689 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
690 } while (!legitimize_mnt(m
, seq
));
696 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
697 * current mount namespace.
699 * The common case is dentries are not mountpoints at all and that
700 * test is handled inline. For the slow case when we are actually
701 * dealing with a mountpoint of some kind, walk through all of the
702 * mounts in the current mount namespace and test to see if the dentry
705 * The mount_hashtable is not usable in the context because we
706 * need to identify all mounts that may be in the current mount
707 * namespace not just a mount that happens to have some specified
710 bool __is_local_mountpoint(struct dentry
*dentry
)
712 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
714 bool is_covered
= false;
716 if (!d_mountpoint(dentry
))
719 down_read(&namespace_sem
);
720 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
721 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
725 up_read(&namespace_sem
);
730 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
732 struct hlist_head
*chain
= mp_hash(dentry
);
733 struct mountpoint
*mp
;
735 hlist_for_each_entry(mp
, chain
, m_hash
) {
736 if (mp
->m_dentry
== dentry
) {
737 /* might be worth a WARN_ON() */
738 if (d_unlinked(dentry
))
739 return ERR_PTR(-ENOENT
);
747 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
749 struct hlist_head
*chain
= mp_hash(dentry
);
750 struct mountpoint
*mp
;
753 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
755 return ERR_PTR(-ENOMEM
);
757 ret
= d_set_mounted(dentry
);
763 mp
->m_dentry
= dentry
;
765 hlist_add_head(&mp
->m_hash
, chain
);
766 INIT_HLIST_HEAD(&mp
->m_list
);
770 static void put_mountpoint(struct mountpoint
*mp
)
772 if (!--mp
->m_count
) {
773 struct dentry
*dentry
= mp
->m_dentry
;
774 BUG_ON(!hlist_empty(&mp
->m_list
));
775 spin_lock(&dentry
->d_lock
);
776 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
777 spin_unlock(&dentry
->d_lock
);
778 hlist_del(&mp
->m_hash
);
783 static inline int check_mnt(struct mount
*mnt
)
785 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
789 * vfsmount lock must be held for write
791 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
795 wake_up_interruptible(&ns
->poll
);
800 * vfsmount lock must be held for write
802 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
804 if (ns
&& ns
->event
!= event
) {
806 wake_up_interruptible(&ns
->poll
);
811 * vfsmount lock must be held for write
813 static void unhash_mnt(struct mount
*mnt
)
815 mnt
->mnt_parent
= mnt
;
816 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
817 list_del_init(&mnt
->mnt_child
);
818 hlist_del_init_rcu(&mnt
->mnt_hash
);
819 hlist_del_init(&mnt
->mnt_mp_list
);
820 put_mountpoint(mnt
->mnt_mp
);
825 * vfsmount lock must be held for write
827 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
829 old_path
->dentry
= mnt
->mnt_mountpoint
;
830 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
835 * vfsmount lock must be held for write
837 static void umount_mnt(struct mount
*mnt
)
839 /* old mountpoint will be dropped when we can do that */
840 mnt
->mnt_ex_mountpoint
= mnt
->mnt_mountpoint
;
845 * vfsmount lock must be held for write
847 void mnt_set_mountpoint(struct mount
*mnt
,
848 struct mountpoint
*mp
,
849 struct mount
*child_mnt
)
852 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
853 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
854 child_mnt
->mnt_parent
= mnt
;
855 child_mnt
->mnt_mp
= mp
;
856 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
860 * vfsmount lock must be held for write
862 static void attach_mnt(struct mount
*mnt
,
863 struct mount
*parent
,
864 struct mountpoint
*mp
)
866 mnt_set_mountpoint(parent
, mp
, mnt
);
867 hlist_add_head_rcu(&mnt
->mnt_hash
, m_hash(&parent
->mnt
, mp
->m_dentry
));
868 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
871 static void attach_shadowed(struct mount
*mnt
,
872 struct mount
*parent
,
873 struct mount
*shadows
)
876 hlist_add_behind_rcu(&mnt
->mnt_hash
, &shadows
->mnt_hash
);
877 list_add(&mnt
->mnt_child
, &shadows
->mnt_child
);
879 hlist_add_head_rcu(&mnt
->mnt_hash
,
880 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
881 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
886 * vfsmount lock must be held for write
888 static void commit_tree(struct mount
*mnt
, struct mount
*shadows
)
890 struct mount
*parent
= mnt
->mnt_parent
;
893 struct mnt_namespace
*n
= parent
->mnt_ns
;
895 BUG_ON(parent
== mnt
);
897 list_add_tail(&head
, &mnt
->mnt_list
);
898 list_for_each_entry(m
, &head
, mnt_list
)
901 list_splice(&head
, n
->list
.prev
);
903 attach_shadowed(mnt
, parent
, shadows
);
904 touch_mnt_namespace(n
);
907 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
909 struct list_head
*next
= p
->mnt_mounts
.next
;
910 if (next
== &p
->mnt_mounts
) {
914 next
= p
->mnt_child
.next
;
915 if (next
!= &p
->mnt_parent
->mnt_mounts
)
920 return list_entry(next
, struct mount
, mnt_child
);
923 static struct mount
*skip_mnt_tree(struct mount
*p
)
925 struct list_head
*prev
= p
->mnt_mounts
.prev
;
926 while (prev
!= &p
->mnt_mounts
) {
927 p
= list_entry(prev
, struct mount
, mnt_child
);
928 prev
= p
->mnt_mounts
.prev
;
934 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
940 return ERR_PTR(-ENODEV
);
942 mnt
= alloc_vfsmnt(name
);
944 return ERR_PTR(-ENOMEM
);
946 mnt
->mnt
.data
= NULL
;
947 if (type
->alloc_mnt_data
) {
948 mnt
->mnt
.data
= type
->alloc_mnt_data();
949 if (!mnt
->mnt
.data
) {
952 return ERR_PTR(-ENOMEM
);
955 if (flags
& MS_KERNMOUNT
)
956 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
958 root
= mount_fs(type
, flags
, name
, data
);
960 kfree(mnt
->mnt
.data
);
963 return ERR_CAST(root
);
966 mnt
->mnt
.mnt_root
= root
;
967 mnt
->mnt
.mnt_sb
= root
->d_sb
;
968 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
969 mnt
->mnt_parent
= mnt
;
971 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
975 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
977 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
980 struct super_block
*sb
= old
->mnt
.mnt_sb
;
984 mnt
= alloc_vfsmnt(old
->mnt_devname
);
986 return ERR_PTR(-ENOMEM
);
988 if (sb
->s_op
->clone_mnt_data
) {
989 mnt
->mnt
.data
= sb
->s_op
->clone_mnt_data(old
->mnt
.data
);
990 if (!mnt
->mnt
.data
) {
996 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
997 mnt
->mnt_group_id
= 0; /* not a peer of original */
999 mnt
->mnt_group_id
= old
->mnt_group_id
;
1001 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1002 err
= mnt_alloc_group_id(mnt
);
1007 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~(MNT_WRITE_HOLD
|MNT_MARKED
);
1008 /* Don't allow unprivileged users to change mount flags */
1009 if (flag
& CL_UNPRIVILEGED
) {
1010 mnt
->mnt
.mnt_flags
|= MNT_LOCK_ATIME
;
1012 if (mnt
->mnt
.mnt_flags
& MNT_READONLY
)
1013 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
1015 if (mnt
->mnt
.mnt_flags
& MNT_NODEV
)
1016 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NODEV
;
1018 if (mnt
->mnt
.mnt_flags
& MNT_NOSUID
)
1019 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOSUID
;
1021 if (mnt
->mnt
.mnt_flags
& MNT_NOEXEC
)
1022 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOEXEC
;
1025 /* Don't allow unprivileged users to reveal what is under a mount */
1026 if ((flag
& CL_UNPRIVILEGED
) &&
1027 (!(flag
& CL_EXPIRE
) || list_empty(&old
->mnt_expire
)))
1028 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
1030 atomic_inc(&sb
->s_active
);
1031 mnt
->mnt
.mnt_sb
= sb
;
1032 mnt
->mnt
.mnt_root
= dget(root
);
1033 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1034 mnt
->mnt_parent
= mnt
;
1036 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1037 unlock_mount_hash();
1039 if ((flag
& CL_SLAVE
) ||
1040 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1041 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1042 mnt
->mnt_master
= old
;
1043 CLEAR_MNT_SHARED(mnt
);
1044 } else if (!(flag
& CL_PRIVATE
)) {
1045 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1046 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1047 if (IS_MNT_SLAVE(old
))
1048 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1049 mnt
->mnt_master
= old
->mnt_master
;
1051 if (flag
& CL_MAKE_SHARED
)
1052 set_mnt_shared(mnt
);
1054 /* stick the duplicate mount on the same expiry list
1055 * as the original if that was on one */
1056 if (flag
& CL_EXPIRE
) {
1057 if (!list_empty(&old
->mnt_expire
))
1058 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1064 kfree(mnt
->mnt
.data
);
1067 return ERR_PTR(err
);
1070 static void cleanup_mnt(struct mount
*mnt
)
1073 * This probably indicates that somebody messed
1074 * up a mnt_want/drop_write() pair. If this
1075 * happens, the filesystem was probably unable
1076 * to make r/w->r/o transitions.
1079 * The locking used to deal with mnt_count decrement provides barriers,
1080 * so mnt_get_writers() below is safe.
1082 WARN_ON(mnt_get_writers(mnt
));
1083 if (unlikely(mnt
->mnt_pins
.first
))
1085 fsnotify_vfsmount_delete(&mnt
->mnt
);
1086 dput(mnt
->mnt
.mnt_root
);
1087 deactivate_super(mnt
->mnt
.mnt_sb
);
1089 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1092 static void __cleanup_mnt(struct rcu_head
*head
)
1094 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1097 static LLIST_HEAD(delayed_mntput_list
);
1098 static void delayed_mntput(struct work_struct
*unused
)
1100 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1101 struct llist_node
*next
;
1103 for (; node
; node
= next
) {
1104 next
= llist_next(node
);
1105 cleanup_mnt(llist_entry(node
, struct mount
, mnt_llist
));
1108 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1110 static void mntput_no_expire(struct mount
*mnt
)
1113 mnt_add_count(mnt
, -1);
1114 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
1119 if (mnt_get_count(mnt
)) {
1121 unlock_mount_hash();
1124 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1126 unlock_mount_hash();
1129 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1132 list_del(&mnt
->mnt_instance
);
1134 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1135 struct mount
*p
, *tmp
;
1136 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1140 unlock_mount_hash();
1142 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1143 struct task_struct
*task
= current
;
1144 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1145 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1146 if (!task_work_add(task
, &mnt
->mnt_rcu
, true))
1149 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1150 schedule_delayed_work(&delayed_mntput_work
, 1);
1156 void mntput(struct vfsmount
*mnt
)
1159 struct mount
*m
= real_mount(mnt
);
1160 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1161 if (unlikely(m
->mnt_expiry_mark
))
1162 m
->mnt_expiry_mark
= 0;
1163 mntput_no_expire(m
);
1166 EXPORT_SYMBOL(mntput
);
1168 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1171 mnt_add_count(real_mount(mnt
), 1);
1174 EXPORT_SYMBOL(mntget
);
1176 struct vfsmount
*mnt_clone_internal(struct path
*path
)
1179 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1182 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1186 static inline void mangle(struct seq_file
*m
, const char *s
)
1188 seq_escape(m
, s
, " \t\n\\");
1192 * Simple .show_options callback for filesystems which don't want to
1193 * implement more complex mount option showing.
1195 * See also save_mount_options().
1197 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1199 const char *options
;
1202 options
= rcu_dereference(root
->d_sb
->s_options
);
1204 if (options
!= NULL
&& options
[0]) {
1212 EXPORT_SYMBOL(generic_show_options
);
1215 * If filesystem uses generic_show_options(), this function should be
1216 * called from the fill_super() callback.
1218 * The .remount_fs callback usually needs to be handled in a special
1219 * way, to make sure, that previous options are not overwritten if the
1222 * Also note, that if the filesystem's .remount_fs function doesn't
1223 * reset all options to their default value, but changes only newly
1224 * given options, then the displayed options will not reflect reality
1227 void save_mount_options(struct super_block
*sb
, char *options
)
1229 BUG_ON(sb
->s_options
);
1230 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1232 EXPORT_SYMBOL(save_mount_options
);
1234 void replace_mount_options(struct super_block
*sb
, char *options
)
1236 char *old
= sb
->s_options
;
1237 rcu_assign_pointer(sb
->s_options
, options
);
1243 EXPORT_SYMBOL(replace_mount_options
);
1245 #ifdef CONFIG_PROC_FS
1246 /* iterator; we want it to have access to namespace_sem, thus here... */
1247 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1249 struct proc_mounts
*p
= m
->private;
1251 down_read(&namespace_sem
);
1252 if (p
->cached_event
== p
->ns
->event
) {
1253 void *v
= p
->cached_mount
;
1254 if (*pos
== p
->cached_index
)
1256 if (*pos
== p
->cached_index
+ 1) {
1257 v
= seq_list_next(v
, &p
->ns
->list
, &p
->cached_index
);
1258 return p
->cached_mount
= v
;
1262 p
->cached_event
= p
->ns
->event
;
1263 p
->cached_mount
= seq_list_start(&p
->ns
->list
, *pos
);
1264 p
->cached_index
= *pos
;
1265 return p
->cached_mount
;
1268 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1270 struct proc_mounts
*p
= m
->private;
1272 p
->cached_mount
= seq_list_next(v
, &p
->ns
->list
, pos
);
1273 p
->cached_index
= *pos
;
1274 return p
->cached_mount
;
1277 static void m_stop(struct seq_file
*m
, void *v
)
1279 up_read(&namespace_sem
);
1282 static int m_show(struct seq_file
*m
, void *v
)
1284 struct proc_mounts
*p
= m
->private;
1285 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1286 return p
->show(m
, &r
->mnt
);
1289 const struct seq_operations mounts_op
= {
1295 #endif /* CONFIG_PROC_FS */
1298 * may_umount_tree - check if a mount tree is busy
1299 * @mnt: root of mount tree
1301 * This is called to check if a tree of mounts has any
1302 * open files, pwds, chroots or sub mounts that are
1305 int may_umount_tree(struct vfsmount
*m
)
1307 struct mount
*mnt
= real_mount(m
);
1308 int actual_refs
= 0;
1309 int minimum_refs
= 0;
1313 /* write lock needed for mnt_get_count */
1315 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1316 actual_refs
+= mnt_get_count(p
);
1319 unlock_mount_hash();
1321 if (actual_refs
> minimum_refs
)
1327 EXPORT_SYMBOL(may_umount_tree
);
1330 * may_umount - check if a mount point is busy
1331 * @mnt: root of mount
1333 * This is called to check if a mount point has any
1334 * open files, pwds, chroots or sub mounts. If the
1335 * mount has sub mounts this will return busy
1336 * regardless of whether the sub mounts are busy.
1338 * Doesn't take quota and stuff into account. IOW, in some cases it will
1339 * give false negatives. The main reason why it's here is that we need
1340 * a non-destructive way to look for easily umountable filesystems.
1342 int may_umount(struct vfsmount
*mnt
)
1345 down_read(&namespace_sem
);
1347 if (propagate_mount_busy(real_mount(mnt
), 2))
1349 unlock_mount_hash();
1350 up_read(&namespace_sem
);
1354 EXPORT_SYMBOL(may_umount
);
1356 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
1358 static void namespace_unlock(void)
1360 struct hlist_head head
;
1362 hlist_move_list(&unmounted
, &head
);
1364 up_write(&namespace_sem
);
1366 if (likely(hlist_empty(&head
)))
1371 group_pin_kill(&head
);
1374 static inline void namespace_lock(void)
1376 down_write(&namespace_sem
);
1379 enum umount_tree_flags
{
1381 UMOUNT_PROPAGATE
= 2,
1382 UMOUNT_CONNECTED
= 4,
1385 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1387 /* Leaving mounts connected is only valid for lazy umounts */
1388 if (how
& UMOUNT_SYNC
)
1391 /* A mount without a parent has nothing to be connected to */
1392 if (!mnt_has_parent(mnt
))
1395 /* Because the reference counting rules change when mounts are
1396 * unmounted and connected, umounted mounts may not be
1397 * connected to mounted mounts.
1399 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1402 /* Has it been requested that the mount remain connected? */
1403 if (how
& UMOUNT_CONNECTED
)
1406 /* Is the mount locked such that it needs to remain connected? */
1407 if (IS_MNT_LOCKED(mnt
))
1410 /* By default disconnect the mount */
1415 * mount_lock must be held
1416 * namespace_sem must be held for write
1418 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1420 LIST_HEAD(tmp_list
);
1423 if (how
& UMOUNT_PROPAGATE
)
1424 propagate_mount_unlock(mnt
);
1426 /* Gather the mounts to umount */
1427 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1428 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1429 list_move(&p
->mnt_list
, &tmp_list
);
1432 /* Hide the mounts from mnt_mounts */
1433 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1434 list_del_init(&p
->mnt_child
);
1437 /* Add propogated mounts to the tmp_list */
1438 if (how
& UMOUNT_PROPAGATE
)
1439 propagate_umount(&tmp_list
);
1441 while (!list_empty(&tmp_list
)) {
1443 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1444 list_del_init(&p
->mnt_expire
);
1445 list_del_init(&p
->mnt_list
);
1446 __touch_mnt_namespace(p
->mnt_ns
);
1448 if (how
& UMOUNT_SYNC
)
1449 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1451 disconnect
= disconnect_mount(p
, how
);
1453 pin_insert_group(&p
->mnt_umount
, &p
->mnt_parent
->mnt
,
1454 disconnect
? &unmounted
: NULL
);
1455 if (mnt_has_parent(p
)) {
1456 mnt_add_count(p
->mnt_parent
, -1);
1458 /* Don't forget about p */
1459 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1464 change_mnt_propagation(p
, MS_PRIVATE
);
1468 static void shrink_submounts(struct mount
*mnt
);
1470 static int do_umount(struct mount
*mnt
, int flags
)
1472 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1475 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1480 * Allow userspace to request a mountpoint be expired rather than
1481 * unmounting unconditionally. Unmount only happens if:
1482 * (1) the mark is already set (the mark is cleared by mntput())
1483 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1485 if (flags
& MNT_EXPIRE
) {
1486 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1487 flags
& (MNT_FORCE
| MNT_DETACH
))
1491 * probably don't strictly need the lock here if we examined
1492 * all race cases, but it's a slowpath.
1495 if (mnt_get_count(mnt
) != 2) {
1496 unlock_mount_hash();
1499 unlock_mount_hash();
1501 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1506 * If we may have to abort operations to get out of this
1507 * mount, and they will themselves hold resources we must
1508 * allow the fs to do things. In the Unix tradition of
1509 * 'Gee thats tricky lets do it in userspace' the umount_begin
1510 * might fail to complete on the first run through as other tasks
1511 * must return, and the like. Thats for the mount program to worry
1512 * about for the moment.
1515 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1516 sb
->s_op
->umount_begin(sb
);
1520 * No sense to grab the lock for this test, but test itself looks
1521 * somewhat bogus. Suggestions for better replacement?
1522 * Ho-hum... In principle, we might treat that as umount + switch
1523 * to rootfs. GC would eventually take care of the old vfsmount.
1524 * Actually it makes sense, especially if rootfs would contain a
1525 * /reboot - static binary that would close all descriptors and
1526 * call reboot(9). Then init(8) could umount root and exec /reboot.
1528 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1530 * Special case for "unmounting" root ...
1531 * we just try to remount it readonly.
1533 if (!capable(CAP_SYS_ADMIN
))
1535 down_write(&sb
->s_umount
);
1536 if (!(sb
->s_flags
& MS_RDONLY
))
1537 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1538 up_write(&sb
->s_umount
);
1546 if (flags
& MNT_DETACH
) {
1547 if (!list_empty(&mnt
->mnt_list
))
1548 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1551 shrink_submounts(mnt
);
1553 if (!propagate_mount_busy(mnt
, 2)) {
1554 if (!list_empty(&mnt
->mnt_list
))
1555 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1559 unlock_mount_hash();
1565 * __detach_mounts - lazily unmount all mounts on the specified dentry
1567 * During unlink, rmdir, and d_drop it is possible to loose the path
1568 * to an existing mountpoint, and wind up leaking the mount.
1569 * detach_mounts allows lazily unmounting those mounts instead of
1572 * The caller may hold dentry->d_inode->i_mutex.
1574 void __detach_mounts(struct dentry
*dentry
)
1576 struct mountpoint
*mp
;
1580 mp
= lookup_mountpoint(dentry
);
1581 if (IS_ERR_OR_NULL(mp
))
1585 while (!hlist_empty(&mp
->m_list
)) {
1586 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1587 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1588 hlist_add_head(&mnt
->mnt_umount
.s_list
, &unmounted
);
1591 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1593 unlock_mount_hash();
1600 * Is the caller allowed to modify his namespace?
1602 static inline bool may_mount(void)
1604 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1608 * Now umount can handle mount points as well as block devices.
1609 * This is important for filesystems which use unnamed block devices.
1611 * We now support a flag for forced unmount like the other 'big iron'
1612 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1615 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1620 int lookup_flags
= 0;
1622 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1628 if (!(flags
& UMOUNT_NOFOLLOW
))
1629 lookup_flags
|= LOOKUP_FOLLOW
;
1631 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1634 mnt
= real_mount(path
.mnt
);
1636 if (path
.dentry
!= path
.mnt
->mnt_root
)
1638 if (!check_mnt(mnt
))
1640 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1643 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1646 retval
= do_umount(mnt
, flags
);
1648 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1650 mntput_no_expire(mnt
);
1655 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1658 * The 2.0 compatible umount. No flags.
1660 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1662 return sys_umount(name
, 0);
1667 static bool is_mnt_ns_file(struct dentry
*dentry
)
1669 /* Is this a proxy for a mount namespace? */
1670 return dentry
->d_op
== &ns_dentry_operations
&&
1671 dentry
->d_fsdata
== &mntns_operations
;
1674 struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1676 return container_of(ns
, struct mnt_namespace
, ns
);
1679 static bool mnt_ns_loop(struct dentry
*dentry
)
1681 /* Could bind mounting the mount namespace inode cause a
1682 * mount namespace loop?
1684 struct mnt_namespace
*mnt_ns
;
1685 if (!is_mnt_ns_file(dentry
))
1688 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1689 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1692 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1695 struct mount
*res
, *p
, *q
, *r
, *parent
;
1697 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1698 return ERR_PTR(-EINVAL
);
1700 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1701 return ERR_PTR(-EINVAL
);
1703 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1707 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1710 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1712 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1715 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1716 struct mount
*t
= NULL
;
1717 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1718 IS_MNT_UNBINDABLE(s
)) {
1719 s
= skip_mnt_tree(s
);
1722 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1723 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1724 s
= skip_mnt_tree(s
);
1727 while (p
!= s
->mnt_parent
) {
1733 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1737 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1738 mnt_set_mountpoint(parent
, p
->mnt_mp
, q
);
1739 if (!list_empty(&parent
->mnt_mounts
)) {
1740 t
= list_last_entry(&parent
->mnt_mounts
,
1741 struct mount
, mnt_child
);
1742 if (t
->mnt_mp
!= p
->mnt_mp
)
1745 attach_shadowed(q
, parent
, t
);
1746 unlock_mount_hash();
1753 umount_tree(res
, UMOUNT_SYNC
);
1754 unlock_mount_hash();
1759 /* Caller should check returned pointer for errors */
1761 struct vfsmount
*collect_mounts(struct path
*path
)
1765 if (!check_mnt(real_mount(path
->mnt
)))
1766 tree
= ERR_PTR(-EINVAL
);
1768 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1769 CL_COPY_ALL
| CL_PRIVATE
);
1772 return ERR_CAST(tree
);
1776 void drop_collected_mounts(struct vfsmount
*mnt
)
1780 umount_tree(real_mount(mnt
), UMOUNT_SYNC
);
1781 unlock_mount_hash();
1786 * clone_private_mount - create a private clone of a path
1788 * This creates a new vfsmount, which will be the clone of @path. The new will
1789 * not be attached anywhere in the namespace and will be private (i.e. changes
1790 * to the originating mount won't be propagated into this).
1792 * Release with mntput().
1794 struct vfsmount
*clone_private_mount(struct path
*path
)
1796 struct mount
*old_mnt
= real_mount(path
->mnt
);
1797 struct mount
*new_mnt
;
1799 if (IS_MNT_UNBINDABLE(old_mnt
))
1800 return ERR_PTR(-EINVAL
);
1802 down_read(&namespace_sem
);
1803 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1804 up_read(&namespace_sem
);
1805 if (IS_ERR(new_mnt
))
1806 return ERR_CAST(new_mnt
);
1808 return &new_mnt
->mnt
;
1810 EXPORT_SYMBOL_GPL(clone_private_mount
);
1812 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1813 struct vfsmount
*root
)
1816 int res
= f(root
, arg
);
1819 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1820 res
= f(&mnt
->mnt
, arg
);
1827 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1831 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1832 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1833 mnt_release_group_id(p
);
1837 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1841 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1842 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1843 int err
= mnt_alloc_group_id(p
);
1845 cleanup_group_ids(mnt
, p
);
1855 * @source_mnt : mount tree to be attached
1856 * @nd : place the mount tree @source_mnt is attached
1857 * @parent_nd : if non-null, detach the source_mnt from its parent and
1858 * store the parent mount and mountpoint dentry.
1859 * (done when source_mnt is moved)
1861 * NOTE: in the table below explains the semantics when a source mount
1862 * of a given type is attached to a destination mount of a given type.
1863 * ---------------------------------------------------------------------------
1864 * | BIND MOUNT OPERATION |
1865 * |**************************************************************************
1866 * | source-->| shared | private | slave | unbindable |
1870 * |**************************************************************************
1871 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1873 * |non-shared| shared (+) | private | slave (*) | invalid |
1874 * ***************************************************************************
1875 * A bind operation clones the source mount and mounts the clone on the
1876 * destination mount.
1878 * (++) the cloned mount is propagated to all the mounts in the propagation
1879 * tree of the destination mount and the cloned mount is added to
1880 * the peer group of the source mount.
1881 * (+) the cloned mount is created under the destination mount and is marked
1882 * as shared. The cloned mount is added to the peer group of the source
1884 * (+++) the mount is propagated to all the mounts in the propagation tree
1885 * of the destination mount and the cloned mount is made slave
1886 * of the same master as that of the source mount. The cloned mount
1887 * is marked as 'shared and slave'.
1888 * (*) the cloned mount is made a slave of the same master as that of the
1891 * ---------------------------------------------------------------------------
1892 * | MOVE MOUNT OPERATION |
1893 * |**************************************************************************
1894 * | source-->| shared | private | slave | unbindable |
1898 * |**************************************************************************
1899 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1901 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1902 * ***************************************************************************
1904 * (+) the mount is moved to the destination. And is then propagated to
1905 * all the mounts in the propagation tree of the destination mount.
1906 * (+*) the mount is moved to the destination.
1907 * (+++) the mount is moved to the destination and is then propagated to
1908 * all the mounts belonging to the destination mount's propagation tree.
1909 * the mount is marked as 'shared and slave'.
1910 * (*) the mount continues to be a slave at the new location.
1912 * if the source mount is a tree, the operations explained above is
1913 * applied to each mount in the tree.
1914 * Must be called without spinlocks held, since this function can sleep
1917 static int attach_recursive_mnt(struct mount
*source_mnt
,
1918 struct mount
*dest_mnt
,
1919 struct mountpoint
*dest_mp
,
1920 struct path
*parent_path
)
1922 HLIST_HEAD(tree_list
);
1923 struct mount
*child
, *p
;
1924 struct hlist_node
*n
;
1927 if (IS_MNT_SHARED(dest_mnt
)) {
1928 err
= invent_group_ids(source_mnt
, true);
1931 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1934 goto out_cleanup_ids
;
1935 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1941 detach_mnt(source_mnt
, parent_path
);
1942 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1943 touch_mnt_namespace(source_mnt
->mnt_ns
);
1945 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1946 commit_tree(source_mnt
, NULL
);
1949 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
1951 hlist_del_init(&child
->mnt_hash
);
1952 q
= __lookup_mnt_last(&child
->mnt_parent
->mnt
,
1953 child
->mnt_mountpoint
);
1954 commit_tree(child
, q
);
1956 unlock_mount_hash();
1961 while (!hlist_empty(&tree_list
)) {
1962 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
1963 umount_tree(child
, UMOUNT_SYNC
);
1965 unlock_mount_hash();
1966 cleanup_group_ids(source_mnt
, NULL
);
1971 static struct mountpoint
*lock_mount(struct path
*path
)
1973 struct vfsmount
*mnt
;
1974 struct dentry
*dentry
= path
->dentry
;
1976 mutex_lock(&dentry
->d_inode
->i_mutex
);
1977 if (unlikely(cant_mount(dentry
))) {
1978 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1979 return ERR_PTR(-ENOENT
);
1982 mnt
= lookup_mnt(path
);
1984 struct mountpoint
*mp
= lookup_mountpoint(dentry
);
1986 mp
= new_mountpoint(dentry
);
1989 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1995 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1998 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2002 static void unlock_mount(struct mountpoint
*where
)
2004 struct dentry
*dentry
= where
->m_dentry
;
2005 put_mountpoint(where
);
2007 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2010 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2012 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
2015 if (d_is_dir(mp
->m_dentry
) !=
2016 d_is_dir(mnt
->mnt
.mnt_root
))
2019 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
2023 * Sanity check the flags to change_mnt_propagation.
2026 static int flags_to_propagation_type(int flags
)
2028 int type
= flags
& ~(MS_REC
| MS_SILENT
);
2030 /* Fail if any non-propagation flags are set */
2031 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2033 /* Only one propagation flag should be set */
2034 if (!is_power_of_2(type
))
2040 * recursively change the type of the mountpoint.
2042 static int do_change_type(struct path
*path
, int flag
)
2045 struct mount
*mnt
= real_mount(path
->mnt
);
2046 int recurse
= flag
& MS_REC
;
2050 if (path
->dentry
!= path
->mnt
->mnt_root
)
2053 type
= flags_to_propagation_type(flag
);
2058 if (type
== MS_SHARED
) {
2059 err
= invent_group_ids(mnt
, recurse
);
2065 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2066 change_mnt_propagation(m
, type
);
2067 unlock_mount_hash();
2074 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2076 struct mount
*child
;
2077 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2078 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2081 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2088 * do loopback mount.
2090 static int do_loopback(struct path
*path
, const char *old_name
,
2093 struct path old_path
;
2094 struct mount
*mnt
= NULL
, *old
, *parent
;
2095 struct mountpoint
*mp
;
2097 if (!old_name
|| !*old_name
)
2099 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2104 if (mnt_ns_loop(old_path
.dentry
))
2107 mp
= lock_mount(path
);
2112 old
= real_mount(old_path
.mnt
);
2113 parent
= real_mount(path
->mnt
);
2116 if (IS_MNT_UNBINDABLE(old
))
2119 if (!check_mnt(parent
))
2122 if (!check_mnt(old
) && old_path
.dentry
->d_op
!= &ns_dentry_operations
)
2125 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
2129 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
2131 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
2138 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2140 err
= graft_tree(mnt
, parent
, mp
);
2143 umount_tree(mnt
, UMOUNT_SYNC
);
2144 unlock_mount_hash();
2149 path_put(&old_path
);
2153 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
2156 int readonly_request
= 0;
2158 if (ms_flags
& MS_RDONLY
)
2159 readonly_request
= 1;
2160 if (readonly_request
== __mnt_is_readonly(mnt
))
2163 if (readonly_request
)
2164 error
= mnt_make_readonly(real_mount(mnt
));
2166 __mnt_unmake_readonly(real_mount(mnt
));
2171 * change filesystem flags. dir should be a physical root of filesystem.
2172 * If you've mounted a non-root directory somewhere and want to do remount
2173 * on it - tough luck.
2175 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
2179 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2180 struct mount
*mnt
= real_mount(path
->mnt
);
2182 if (!check_mnt(mnt
))
2185 if (path
->dentry
!= path
->mnt
->mnt_root
)
2188 /* Don't allow changing of locked mnt flags.
2190 * No locks need to be held here while testing the various
2191 * MNT_LOCK flags because those flags can never be cleared
2192 * once they are set.
2194 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_READONLY
) &&
2195 !(mnt_flags
& MNT_READONLY
)) {
2198 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NODEV
) &&
2199 !(mnt_flags
& MNT_NODEV
)) {
2200 /* Was the nodev implicitly added in mount? */
2201 if ((mnt
->mnt_ns
->user_ns
!= &init_user_ns
) &&
2202 !(sb
->s_type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2203 mnt_flags
|= MNT_NODEV
;
2208 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOSUID
) &&
2209 !(mnt_flags
& MNT_NOSUID
)) {
2212 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOEXEC
) &&
2213 !(mnt_flags
& MNT_NOEXEC
)) {
2216 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_ATIME
) &&
2217 ((mnt
->mnt
.mnt_flags
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
))) {
2221 err
= security_sb_remount(sb
, data
);
2225 down_write(&sb
->s_umount
);
2226 if (flags
& MS_BIND
)
2227 err
= change_mount_flags(path
->mnt
, flags
);
2228 else if (!capable(CAP_SYS_ADMIN
))
2231 err
= do_remount_sb(sb
, flags
, data
, 0);
2234 propagate_remount(mnt
);
2235 unlock_mount_hash();
2240 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2241 mnt
->mnt
.mnt_flags
= mnt_flags
;
2242 touch_mnt_namespace(mnt
->mnt_ns
);
2243 unlock_mount_hash();
2245 up_write(&sb
->s_umount
);
2249 static inline int tree_contains_unbindable(struct mount
*mnt
)
2252 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2253 if (IS_MNT_UNBINDABLE(p
))
2259 static int do_move_mount(struct path
*path
, const char *old_name
)
2261 struct path old_path
, parent_path
;
2264 struct mountpoint
*mp
;
2266 if (!old_name
|| !*old_name
)
2268 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2272 mp
= lock_mount(path
);
2277 old
= real_mount(old_path
.mnt
);
2278 p
= real_mount(path
->mnt
);
2281 if (!check_mnt(p
) || !check_mnt(old
))
2284 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2288 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
2291 if (!mnt_has_parent(old
))
2294 if (d_is_dir(path
->dentry
) !=
2295 d_is_dir(old_path
.dentry
))
2298 * Don't move a mount residing in a shared parent.
2300 if (IS_MNT_SHARED(old
->mnt_parent
))
2303 * Don't move a mount tree containing unbindable mounts to a destination
2304 * mount which is shared.
2306 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2309 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2313 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2317 /* if the mount is moved, it should no longer be expire
2319 list_del_init(&old
->mnt_expire
);
2324 path_put(&parent_path
);
2325 path_put(&old_path
);
2329 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2332 const char *subtype
= strchr(fstype
, '.');
2341 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2343 if (!mnt
->mnt_sb
->s_subtype
)
2349 return ERR_PTR(err
);
2353 * add a mount into a namespace's mount tree
2355 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2357 struct mountpoint
*mp
;
2358 struct mount
*parent
;
2361 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2363 mp
= lock_mount(path
);
2367 parent
= real_mount(path
->mnt
);
2369 if (unlikely(!check_mnt(parent
))) {
2370 /* that's acceptable only for automounts done in private ns */
2371 if (!(mnt_flags
& MNT_SHRINKABLE
))
2373 /* ... and for those we'd better have mountpoint still alive */
2374 if (!parent
->mnt_ns
)
2378 /* Refuse the same filesystem on the same mount point */
2380 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2381 path
->mnt
->mnt_root
== path
->dentry
)
2385 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2388 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2389 err
= graft_tree(newmnt
, parent
, mp
);
2396 static bool fs_fully_visible(struct file_system_type
*fs_type
, int *new_mnt_flags
);
2399 * create a new mount for userspace and request it to be added into the
2402 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2403 int mnt_flags
, const char *name
, void *data
)
2405 struct file_system_type
*type
;
2406 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2407 struct vfsmount
*mnt
;
2413 type
= get_fs_type(fstype
);
2417 if (user_ns
!= &init_user_ns
) {
2418 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2419 put_filesystem(type
);
2422 /* Only in special cases allow devices from mounts
2423 * created outside the initial user namespace.
2425 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2427 mnt_flags
|= MNT_NODEV
| MNT_LOCK_NODEV
;
2429 if (type
->fs_flags
& FS_USERNS_VISIBLE
) {
2430 if (!fs_fully_visible(type
, &mnt_flags
))
2435 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2436 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2437 !mnt
->mnt_sb
->s_subtype
)
2438 mnt
= fs_set_subtype(mnt
, fstype
);
2440 put_filesystem(type
);
2442 return PTR_ERR(mnt
);
2444 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2450 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2452 struct mount
*mnt
= real_mount(m
);
2454 /* The new mount record should have at least 2 refs to prevent it being
2455 * expired before we get a chance to add it
2457 BUG_ON(mnt_get_count(mnt
) < 2);
2459 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2460 m
->mnt_root
== path
->dentry
) {
2465 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2469 /* remove m from any expiration list it may be on */
2470 if (!list_empty(&mnt
->mnt_expire
)) {
2472 list_del_init(&mnt
->mnt_expire
);
2481 * mnt_set_expiry - Put a mount on an expiration list
2482 * @mnt: The mount to list.
2483 * @expiry_list: The list to add the mount to.
2485 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2489 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2493 EXPORT_SYMBOL(mnt_set_expiry
);
2496 * process a list of expirable mountpoints with the intent of discarding any
2497 * mountpoints that aren't in use and haven't been touched since last we came
2500 void mark_mounts_for_expiry(struct list_head
*mounts
)
2502 struct mount
*mnt
, *next
;
2503 LIST_HEAD(graveyard
);
2505 if (list_empty(mounts
))
2511 /* extract from the expiration list every vfsmount that matches the
2512 * following criteria:
2513 * - only referenced by its parent vfsmount
2514 * - still marked for expiry (marked on the last call here; marks are
2515 * cleared by mntput())
2517 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2518 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2519 propagate_mount_busy(mnt
, 1))
2521 list_move(&mnt
->mnt_expire
, &graveyard
);
2523 while (!list_empty(&graveyard
)) {
2524 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2525 touch_mnt_namespace(mnt
->mnt_ns
);
2526 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2528 unlock_mount_hash();
2532 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2535 * Ripoff of 'select_parent()'
2537 * search the list of submounts for a given mountpoint, and move any
2538 * shrinkable submounts to the 'graveyard' list.
2540 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2542 struct mount
*this_parent
= parent
;
2543 struct list_head
*next
;
2547 next
= this_parent
->mnt_mounts
.next
;
2549 while (next
!= &this_parent
->mnt_mounts
) {
2550 struct list_head
*tmp
= next
;
2551 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2554 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2557 * Descend a level if the d_mounts list is non-empty.
2559 if (!list_empty(&mnt
->mnt_mounts
)) {
2564 if (!propagate_mount_busy(mnt
, 1)) {
2565 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2570 * All done at this level ... ascend and resume the search
2572 if (this_parent
!= parent
) {
2573 next
= this_parent
->mnt_child
.next
;
2574 this_parent
= this_parent
->mnt_parent
;
2581 * process a list of expirable mountpoints with the intent of discarding any
2582 * submounts of a specific parent mountpoint
2584 * mount_lock must be held for write
2586 static void shrink_submounts(struct mount
*mnt
)
2588 LIST_HEAD(graveyard
);
2591 /* extract submounts of 'mountpoint' from the expiration list */
2592 while (select_submounts(mnt
, &graveyard
)) {
2593 while (!list_empty(&graveyard
)) {
2594 m
= list_first_entry(&graveyard
, struct mount
,
2596 touch_mnt_namespace(m
->mnt_ns
);
2597 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2603 * Some copy_from_user() implementations do not return the exact number of
2604 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2605 * Note that this function differs from copy_from_user() in that it will oops
2606 * on bad values of `to', rather than returning a short copy.
2608 static long exact_copy_from_user(void *to
, const void __user
* from
,
2612 const char __user
*f
= from
;
2615 if (!access_ok(VERIFY_READ
, from
, n
))
2619 if (__get_user(c
, f
)) {
2630 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2640 if (!(page
= __get_free_page(GFP_KERNEL
)))
2643 /* We only care that *some* data at the address the user
2644 * gave us is valid. Just in case, we'll zero
2645 * the remainder of the page.
2647 /* copy_from_user cannot cross TASK_SIZE ! */
2648 size
= TASK_SIZE
- (unsigned long)data
;
2649 if (size
> PAGE_SIZE
)
2652 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2658 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2663 char *copy_mount_string(const void __user
*data
)
2665 return data
? strndup_user(data
, PAGE_SIZE
) : NULL
;
2669 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2670 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2672 * data is a (void *) that can point to any structure up to
2673 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2674 * information (or be NULL).
2676 * Pre-0.97 versions of mount() didn't have a flags word.
2677 * When the flags word was introduced its top half was required
2678 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2679 * Therefore, if this magic number is present, it carries no information
2680 * and must be discarded.
2682 long do_mount(const char *dev_name
, const char __user
*dir_name
,
2683 const char *type_page
, unsigned long flags
, void *data_page
)
2690 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2691 flags
&= ~MS_MGC_MSK
;
2693 /* Basic sanity checks */
2695 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2697 /* ... and get the mountpoint */
2698 retval
= user_path(dir_name
, &path
);
2702 retval
= security_sb_mount(dev_name
, &path
,
2703 type_page
, flags
, data_page
);
2704 if (!retval
&& !may_mount())
2709 /* Default to relatime unless overriden */
2710 if (!(flags
& MS_NOATIME
))
2711 mnt_flags
|= MNT_RELATIME
;
2713 /* Separate the per-mountpoint flags */
2714 if (flags
& MS_NOSUID
)
2715 mnt_flags
|= MNT_NOSUID
;
2716 if (flags
& MS_NODEV
)
2717 mnt_flags
|= MNT_NODEV
;
2718 if (flags
& MS_NOEXEC
)
2719 mnt_flags
|= MNT_NOEXEC
;
2720 if (flags
& MS_NOATIME
)
2721 mnt_flags
|= MNT_NOATIME
;
2722 if (flags
& MS_NODIRATIME
)
2723 mnt_flags
|= MNT_NODIRATIME
;
2724 if (flags
& MS_STRICTATIME
)
2725 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2726 if (flags
& MS_RDONLY
)
2727 mnt_flags
|= MNT_READONLY
;
2729 /* The default atime for remount is preservation */
2730 if ((flags
& MS_REMOUNT
) &&
2731 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
2732 MS_STRICTATIME
)) == 0)) {
2733 mnt_flags
&= ~MNT_ATIME_MASK
;
2734 mnt_flags
|= path
.mnt
->mnt_flags
& MNT_ATIME_MASK
;
2737 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2738 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2741 if (flags
& MS_REMOUNT
)
2742 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2744 else if (flags
& MS_BIND
)
2745 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2746 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2747 retval
= do_change_type(&path
, flags
);
2748 else if (flags
& MS_MOVE
)
2749 retval
= do_move_mount(&path
, dev_name
);
2751 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2752 dev_name
, data_page
);
2758 static void free_mnt_ns(struct mnt_namespace
*ns
)
2760 ns_free_inum(&ns
->ns
);
2761 put_user_ns(ns
->user_ns
);
2766 * Assign a sequence number so we can detect when we attempt to bind
2767 * mount a reference to an older mount namespace into the current
2768 * mount namespace, preventing reference counting loops. A 64bit
2769 * number incrementing at 10Ghz will take 12,427 years to wrap which
2770 * is effectively never, so we can ignore the possibility.
2772 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2774 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2776 struct mnt_namespace
*new_ns
;
2779 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2781 return ERR_PTR(-ENOMEM
);
2782 ret
= ns_alloc_inum(&new_ns
->ns
);
2785 return ERR_PTR(ret
);
2787 new_ns
->ns
.ops
= &mntns_operations
;
2788 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2789 atomic_set(&new_ns
->count
, 1);
2790 new_ns
->root
= NULL
;
2791 INIT_LIST_HEAD(&new_ns
->list
);
2792 init_waitqueue_head(&new_ns
->poll
);
2794 new_ns
->user_ns
= get_user_ns(user_ns
);
2798 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2799 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2801 struct mnt_namespace
*new_ns
;
2802 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2803 struct mount
*p
, *q
;
2810 if (likely(!(flags
& CLONE_NEWNS
))) {
2817 new_ns
= alloc_mnt_ns(user_ns
);
2822 /* First pass: copy the tree topology */
2823 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2824 if (user_ns
!= ns
->user_ns
)
2825 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2826 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2829 free_mnt_ns(new_ns
);
2830 return ERR_CAST(new);
2833 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2836 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2837 * as belonging to new namespace. We have already acquired a private
2838 * fs_struct, so tsk->fs->lock is not needed.
2845 if (&p
->mnt
== new_fs
->root
.mnt
) {
2846 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2849 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2850 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2854 p
= next_mnt(p
, old
);
2855 q
= next_mnt(q
, new);
2858 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2859 p
= next_mnt(p
, old
);
2872 * create_mnt_ns - creates a private namespace and adds a root filesystem
2873 * @mnt: pointer to the new root filesystem mountpoint
2875 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2877 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2878 if (!IS_ERR(new_ns
)) {
2879 struct mount
*mnt
= real_mount(m
);
2880 mnt
->mnt_ns
= new_ns
;
2882 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2889 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2891 struct mnt_namespace
*ns
;
2892 struct super_block
*s
;
2896 ns
= create_mnt_ns(mnt
);
2898 return ERR_CAST(ns
);
2900 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2901 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2906 return ERR_PTR(err
);
2908 /* trade a vfsmount reference for active sb one */
2909 s
= path
.mnt
->mnt_sb
;
2910 atomic_inc(&s
->s_active
);
2912 /* lock the sucker */
2913 down_write(&s
->s_umount
);
2914 /* ... and return the root of (sub)tree on it */
2917 EXPORT_SYMBOL(mount_subtree
);
2919 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2920 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2925 unsigned long data_page
;
2927 kernel_type
= copy_mount_string(type
);
2928 ret
= PTR_ERR(kernel_type
);
2929 if (IS_ERR(kernel_type
))
2932 kernel_dev
= copy_mount_string(dev_name
);
2933 ret
= PTR_ERR(kernel_dev
);
2934 if (IS_ERR(kernel_dev
))
2937 ret
= copy_mount_options(data
, &data_page
);
2941 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
,
2942 (void *) data_page
);
2944 free_page(data_page
);
2954 * Return true if path is reachable from root
2956 * namespace_sem or mount_lock is held
2958 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2959 const struct path
*root
)
2961 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2962 dentry
= mnt
->mnt_mountpoint
;
2963 mnt
= mnt
->mnt_parent
;
2965 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2968 int path_is_under(struct path
*path1
, struct path
*path2
)
2971 read_seqlock_excl(&mount_lock
);
2972 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2973 read_sequnlock_excl(&mount_lock
);
2976 EXPORT_SYMBOL(path_is_under
);
2979 * pivot_root Semantics:
2980 * Moves the root file system of the current process to the directory put_old,
2981 * makes new_root as the new root file system of the current process, and sets
2982 * root/cwd of all processes which had them on the current root to new_root.
2985 * The new_root and put_old must be directories, and must not be on the
2986 * same file system as the current process root. The put_old must be
2987 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2988 * pointed to by put_old must yield the same directory as new_root. No other
2989 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2991 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2992 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2993 * in this situation.
2996 * - we don't move root/cwd if they are not at the root (reason: if something
2997 * cared enough to change them, it's probably wrong to force them elsewhere)
2998 * - it's okay to pick a root that isn't the root of a file system, e.g.
2999 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3000 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3003 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3004 const char __user
*, put_old
)
3006 struct path
new, old
, parent_path
, root_parent
, root
;
3007 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
3008 struct mountpoint
*old_mp
, *root_mp
;
3014 error
= user_path_dir(new_root
, &new);
3018 error
= user_path_dir(put_old
, &old
);
3022 error
= security_sb_pivotroot(&old
, &new);
3026 get_fs_root(current
->fs
, &root
);
3027 old_mp
= lock_mount(&old
);
3028 error
= PTR_ERR(old_mp
);
3033 new_mnt
= real_mount(new.mnt
);
3034 root_mnt
= real_mount(root
.mnt
);
3035 old_mnt
= real_mount(old
.mnt
);
3036 if (IS_MNT_SHARED(old_mnt
) ||
3037 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
3038 IS_MNT_SHARED(root_mnt
->mnt_parent
))
3040 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3042 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3045 if (d_unlinked(new.dentry
))
3048 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3049 goto out4
; /* loop, on the same file system */
3051 if (root
.mnt
->mnt_root
!= root
.dentry
)
3052 goto out4
; /* not a mountpoint */
3053 if (!mnt_has_parent(root_mnt
))
3054 goto out4
; /* not attached */
3055 root_mp
= root_mnt
->mnt_mp
;
3056 if (new.mnt
->mnt_root
!= new.dentry
)
3057 goto out4
; /* not a mountpoint */
3058 if (!mnt_has_parent(new_mnt
))
3059 goto out4
; /* not attached */
3060 /* make sure we can reach put_old from new_root */
3061 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3063 /* make certain new is below the root */
3064 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3066 root_mp
->m_count
++; /* pin it so it won't go away */
3068 detach_mnt(new_mnt
, &parent_path
);
3069 detach_mnt(root_mnt
, &root_parent
);
3070 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3071 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3072 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3074 /* mount old root on put_old */
3075 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3076 /* mount new_root on / */
3077 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
3078 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3079 /* A moved mount should not expire automatically */
3080 list_del_init(&new_mnt
->mnt_expire
);
3081 unlock_mount_hash();
3082 chroot_fs_refs(&root
, &new);
3083 put_mountpoint(root_mp
);
3086 unlock_mount(old_mp
);
3088 path_put(&root_parent
);
3089 path_put(&parent_path
);
3101 static void __init
init_mount_tree(void)
3103 struct vfsmount
*mnt
;
3104 struct mnt_namespace
*ns
;
3106 struct file_system_type
*type
;
3108 type
= get_fs_type("rootfs");
3110 panic("Can't find rootfs type");
3111 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
3112 put_filesystem(type
);
3114 panic("Can't create rootfs");
3116 ns
= create_mnt_ns(mnt
);
3118 panic("Can't allocate initial namespace");
3120 init_task
.nsproxy
->mnt_ns
= ns
;
3124 root
.dentry
= mnt
->mnt_root
;
3125 mnt
->mnt_flags
|= MNT_LOCKED
;
3127 set_fs_pwd(current
->fs
, &root
);
3128 set_fs_root(current
->fs
, &root
);
3131 void __init
mnt_init(void)
3136 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
3137 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3139 mount_hashtable
= alloc_large_system_hash("Mount-cache",
3140 sizeof(struct hlist_head
),
3143 &m_hash_shift
, &m_hash_mask
, 0, 0);
3144 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
3145 sizeof(struct hlist_head
),
3148 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
3150 if (!mount_hashtable
|| !mountpoint_hashtable
)
3151 panic("Failed to allocate mount hash table\n");
3153 for (u
= 0; u
<= m_hash_mask
; u
++)
3154 INIT_HLIST_HEAD(&mount_hashtable
[u
]);
3155 for (u
= 0; u
<= mp_hash_mask
; u
++)
3156 INIT_HLIST_HEAD(&mountpoint_hashtable
[u
]);
3162 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
3164 fs_kobj
= kobject_create_and_add("fs", NULL
);
3166 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
3171 void put_mnt_ns(struct mnt_namespace
*ns
)
3173 if (!atomic_dec_and_test(&ns
->count
))
3175 drop_collected_mounts(&ns
->root
->mnt
);
3179 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
3181 struct vfsmount
*mnt
;
3182 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
3185 * it is a longterm mount, don't release mnt until
3186 * we unmount before file sys is unregistered
3188 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
3192 EXPORT_SYMBOL_GPL(kern_mount_data
);
3194 void kern_unmount(struct vfsmount
*mnt
)
3196 /* release long term mount so mount point can be released */
3197 if (!IS_ERR_OR_NULL(mnt
)) {
3198 real_mount(mnt
)->mnt_ns
= NULL
;
3199 synchronize_rcu(); /* yecchhh... */
3203 EXPORT_SYMBOL(kern_unmount
);
3205 bool our_mnt(struct vfsmount
*mnt
)
3207 return check_mnt(real_mount(mnt
));
3210 bool current_chrooted(void)
3212 /* Does the current process have a non-standard root */
3213 struct path ns_root
;
3214 struct path fs_root
;
3217 /* Find the namespace root */
3218 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
3219 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
3221 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
3224 get_fs_root(current
->fs
, &fs_root
);
3226 chrooted
= !path_equal(&fs_root
, &ns_root
);
3234 static bool fs_fully_visible(struct file_system_type
*type
, int *new_mnt_flags
)
3236 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
3237 int new_flags
= *new_mnt_flags
;
3239 bool visible
= false;
3244 down_read(&namespace_sem
);
3245 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
3246 struct mount
*child
;
3249 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
3252 /* This mount is not fully visible if it's root directory
3253 * is not the root directory of the filesystem.
3255 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
3258 /* Read the mount flags and filter out flags that
3259 * may safely be ignored.
3261 mnt_flags
= mnt
->mnt
.mnt_flags
;
3262 if (mnt
->mnt
.mnt_sb
->s_iflags
& SB_I_NOEXEC
)
3263 mnt_flags
&= ~(MNT_LOCK_NOSUID
| MNT_LOCK_NOEXEC
);
3265 /* Verify the mount flags are equal to or more permissive
3266 * than the proposed new mount.
3268 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
3269 !(new_flags
& MNT_READONLY
))
3271 if ((mnt_flags
& MNT_LOCK_NODEV
) &&
3272 !(new_flags
& MNT_NODEV
))
3274 if ((mnt_flags
& MNT_LOCK_NOSUID
) &&
3275 !(new_flags
& MNT_NOSUID
))
3277 if ((mnt_flags
& MNT_LOCK_NOEXEC
) &&
3278 !(new_flags
& MNT_NOEXEC
))
3280 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
3281 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
3284 /* This mount is not fully visible if there are any
3285 * locked child mounts that cover anything except for
3286 * empty directories.
3288 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
3289 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
3290 /* Only worry about locked mounts */
3291 if (!(mnt_flags
& MNT_LOCKED
))
3293 /* Is the directory permanetly empty? */
3294 if (!is_empty_dir_inode(inode
))
3297 /* Preserve the locked attributes */
3298 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
3308 up_read(&namespace_sem
);
3312 static struct ns_common
*mntns_get(struct task_struct
*task
)
3314 struct ns_common
*ns
= NULL
;
3315 struct nsproxy
*nsproxy
;
3318 nsproxy
= task
->nsproxy
;
3320 ns
= &nsproxy
->mnt_ns
->ns
;
3321 get_mnt_ns(to_mnt_ns(ns
));
3328 static void mntns_put(struct ns_common
*ns
)
3330 put_mnt_ns(to_mnt_ns(ns
));
3333 static int mntns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
3335 struct fs_struct
*fs
= current
->fs
;
3336 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
);
3339 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3340 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
3341 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
3348 put_mnt_ns(nsproxy
->mnt_ns
);
3349 nsproxy
->mnt_ns
= mnt_ns
;
3352 root
.mnt
= &mnt_ns
->root
->mnt
;
3353 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
3355 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
3358 /* Update the pwd and root */
3359 set_fs_pwd(fs
, &root
);
3360 set_fs_root(fs
, &root
);
3366 const struct proc_ns_operations mntns_operations
= {
3368 .type
= CLONE_NEWNS
,
3371 .install
= mntns_install
,