1 // SPDX-License-Identifier: GPL-2.0-only
5 * (C) Copyright Al Viro 2000, 2001
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/cred.h>
19 #include <linux/idr.h>
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/file.h>
24 #include <linux/uaccess.h>
25 #include <linux/proc_ns.h>
26 #include <linux/magic.h>
27 #include <linux/memblock.h>
28 #include <linux/task_work.h>
29 #include <linux/sched/task.h>
30 #include <uapi/linux/mount.h>
31 #include <linux/fs_context.h>
32 #include <linux/shmem_fs.h>
37 /* Maximum number of mounts in a mount namespace */
38 unsigned int sysctl_mount_max __read_mostly
= 100000;
40 static unsigned int m_hash_mask __read_mostly
;
41 static unsigned int m_hash_shift __read_mostly
;
42 static unsigned int mp_hash_mask __read_mostly
;
43 static unsigned int mp_hash_shift __read_mostly
;
45 static __initdata
unsigned long mhash_entries
;
46 static int __init
set_mhash_entries(char *str
)
50 mhash_entries
= simple_strtoul(str
, &str
, 0);
53 __setup("mhash_entries=", set_mhash_entries
);
55 static __initdata
unsigned long mphash_entries
;
56 static int __init
set_mphash_entries(char *str
)
60 mphash_entries
= simple_strtoul(str
, &str
, 0);
63 __setup("mphash_entries=", set_mphash_entries
);
66 static DEFINE_IDA(mnt_id_ida
);
67 static DEFINE_IDA(mnt_group_ida
);
69 static struct hlist_head
*mount_hashtable __read_mostly
;
70 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
71 static struct kmem_cache
*mnt_cache __read_mostly
;
72 static DECLARE_RWSEM(namespace_sem
);
73 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
74 static LIST_HEAD(ex_mountpoints
); /* protected by namespace_sem */
77 struct kobject
*fs_kobj
;
78 EXPORT_SYMBOL_GPL(fs_kobj
);
81 * vfsmount lock may be taken for read to prevent changes to the
82 * vfsmount hash, ie. during mountpoint lookups or walking back
85 * It should be taken for write in all cases where the vfsmount
86 * tree or hash is modified or when a vfsmount structure is modified.
88 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
90 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
92 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
93 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
94 tmp
= tmp
+ (tmp
>> m_hash_shift
);
95 return &mount_hashtable
[tmp
& m_hash_mask
];
98 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
100 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
101 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
102 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
105 static int mnt_alloc_id(struct mount
*mnt
)
107 int res
= ida_alloc(&mnt_id_ida
, GFP_KERNEL
);
115 static void mnt_free_id(struct mount
*mnt
)
117 ida_free(&mnt_id_ida
, mnt
->mnt_id
);
121 * Allocate a new peer group ID
123 static int mnt_alloc_group_id(struct mount
*mnt
)
125 int res
= ida_alloc_min(&mnt_group_ida
, 1, GFP_KERNEL
);
129 mnt
->mnt_group_id
= res
;
134 * Release a peer group ID
136 void mnt_release_group_id(struct mount
*mnt
)
138 ida_free(&mnt_group_ida
, mnt
->mnt_group_id
);
139 mnt
->mnt_group_id
= 0;
143 * vfsmount lock must be held for read
145 static inline void mnt_add_count(struct mount
*mnt
, int n
)
148 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
157 * vfsmount lock must be held for write
159 int mnt_get_count(struct mount
*mnt
)
165 for_each_possible_cpu(cpu
) {
166 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
171 return mnt
->mnt_count
;
175 static struct mount
*alloc_vfsmnt(const char *name
)
177 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
181 err
= mnt_alloc_id(mnt
);
186 mnt
->mnt_devname
= kstrdup_const(name
, GFP_KERNEL
);
187 if (!mnt
->mnt_devname
)
192 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
194 goto out_free_devname
;
196 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
199 mnt
->mnt_writers
= 0;
202 INIT_HLIST_NODE(&mnt
->mnt_hash
);
203 INIT_LIST_HEAD(&mnt
->mnt_child
);
204 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
205 INIT_LIST_HEAD(&mnt
->mnt_list
);
206 INIT_LIST_HEAD(&mnt
->mnt_expire
);
207 INIT_LIST_HEAD(&mnt
->mnt_share
);
208 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
209 INIT_LIST_HEAD(&mnt
->mnt_slave
);
210 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
211 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
212 INIT_HLIST_HEAD(&mnt
->mnt_stuck_children
);
218 kfree_const(mnt
->mnt_devname
);
223 kmem_cache_free(mnt_cache
, mnt
);
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
246 bool __mnt_is_readonly(struct vfsmount
*mnt
)
248 return (mnt
->mnt_flags
& MNT_READONLY
) || sb_rdonly(mnt
->mnt_sb
);
250 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
252 static inline void mnt_inc_writers(struct mount
*mnt
)
255 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
261 static inline void mnt_dec_writers(struct mount
*mnt
)
264 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
270 static unsigned int mnt_get_writers(struct mount
*mnt
)
273 unsigned int count
= 0;
276 for_each_possible_cpu(cpu
) {
277 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
282 return mnt
->mnt_writers
;
286 static int mnt_is_readonly(struct vfsmount
*mnt
)
288 if (mnt
->mnt_sb
->s_readonly_remount
)
290 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
292 return __mnt_is_readonly(mnt
);
296 * Most r/o & frozen checks on a fs are for operations that take discrete
297 * amounts of time, like a write() or unlink(). We must keep track of when
298 * those operations start (for permission checks) and when they end, so that we
299 * can determine when writes are able to occur to a filesystem.
302 * __mnt_want_write - get write access to a mount without freeze protection
303 * @m: the mount on which to take a write
305 * This tells the low-level filesystem that a write is about to be performed to
306 * it, and makes sure that writes are allowed (mnt it read-write) before
307 * returning success. This operation does not protect against filesystem being
308 * frozen. When the write operation is finished, __mnt_drop_write() must be
309 * called. This is effectively a refcount.
311 int __mnt_want_write(struct vfsmount
*m
)
313 struct mount
*mnt
= real_mount(m
);
317 mnt_inc_writers(mnt
);
319 * The store to mnt_inc_writers must be visible before we pass
320 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
321 * incremented count after it has set MNT_WRITE_HOLD.
324 while (READ_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
327 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
328 * be set to match its requirements. So we must not load that until
329 * MNT_WRITE_HOLD is cleared.
332 if (mnt_is_readonly(m
)) {
333 mnt_dec_writers(mnt
);
342 * mnt_want_write - get write access to a mount
343 * @m: the mount on which to take a write
345 * This tells the low-level filesystem that a write is about to be performed to
346 * it, and makes sure that writes are allowed (mount is read-write, filesystem
347 * is not frozen) before returning success. When the write operation is
348 * finished, mnt_drop_write() must be called. This is effectively a refcount.
350 int mnt_want_write(struct vfsmount
*m
)
354 sb_start_write(m
->mnt_sb
);
355 ret
= __mnt_want_write(m
);
357 sb_end_write(m
->mnt_sb
);
360 EXPORT_SYMBOL_GPL(mnt_want_write
);
363 * mnt_clone_write - get write access to a mount
364 * @mnt: the mount on which to take a write
366 * This is effectively like mnt_want_write, except
367 * it must only be used to take an extra write reference
368 * on a mountpoint that we already know has a write reference
369 * on it. This allows some optimisation.
371 * After finished, mnt_drop_write must be called as usual to
372 * drop the reference.
374 int mnt_clone_write(struct vfsmount
*mnt
)
376 /* superblock may be r/o */
377 if (__mnt_is_readonly(mnt
))
380 mnt_inc_writers(real_mount(mnt
));
384 EXPORT_SYMBOL_GPL(mnt_clone_write
);
387 * __mnt_want_write_file - get write access to a file's mount
388 * @file: the file who's mount on which to take a write
390 * This is like __mnt_want_write, but it takes a file and can
391 * do some optimisations if the file is open for write already
393 int __mnt_want_write_file(struct file
*file
)
395 if (!(file
->f_mode
& FMODE_WRITER
))
396 return __mnt_want_write(file
->f_path
.mnt
);
398 return mnt_clone_write(file
->f_path
.mnt
);
402 * mnt_want_write_file - get write access to a file's mount
403 * @file: the file who's mount on which to take a write
405 * This is like mnt_want_write, but it takes a file and can
406 * do some optimisations if the file is open for write already
408 int mnt_want_write_file(struct file
*file
)
412 sb_start_write(file_inode(file
)->i_sb
);
413 ret
= __mnt_want_write_file(file
);
415 sb_end_write(file_inode(file
)->i_sb
);
418 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
421 * __mnt_drop_write - give up write access to a mount
422 * @mnt: the mount on which to give up write access
424 * Tells the low-level filesystem that we are done
425 * performing writes to it. Must be matched with
426 * __mnt_want_write() call above.
428 void __mnt_drop_write(struct vfsmount
*mnt
)
431 mnt_dec_writers(real_mount(mnt
));
436 * mnt_drop_write - give up write access to a mount
437 * @mnt: the mount on which to give up write access
439 * Tells the low-level filesystem that we are done performing writes to it and
440 * also allows filesystem to be frozen again. Must be matched with
441 * mnt_want_write() call above.
443 void mnt_drop_write(struct vfsmount
*mnt
)
445 __mnt_drop_write(mnt
);
446 sb_end_write(mnt
->mnt_sb
);
448 EXPORT_SYMBOL_GPL(mnt_drop_write
);
450 void __mnt_drop_write_file(struct file
*file
)
452 __mnt_drop_write(file
->f_path
.mnt
);
455 void mnt_drop_write_file(struct file
*file
)
457 __mnt_drop_write_file(file
);
458 sb_end_write(file_inode(file
)->i_sb
);
460 EXPORT_SYMBOL(mnt_drop_write_file
);
462 static int mnt_make_readonly(struct mount
*mnt
)
467 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
469 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
470 * should be visible before we do.
475 * With writers on hold, if this value is zero, then there are
476 * definitely no active writers (although held writers may subsequently
477 * increment the count, they'll have to wait, and decrement it after
478 * seeing MNT_READONLY).
480 * It is OK to have counter incremented on one CPU and decremented on
481 * another: the sum will add up correctly. The danger would be when we
482 * sum up each counter, if we read a counter before it is incremented,
483 * but then read another CPU's count which it has been subsequently
484 * decremented from -- we would see more decrements than we should.
485 * MNT_WRITE_HOLD protects against this scenario, because
486 * mnt_want_write first increments count, then smp_mb, then spins on
487 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
488 * we're counting up here.
490 if (mnt_get_writers(mnt
) > 0)
493 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
495 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
496 * that become unheld will see MNT_READONLY.
499 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
504 static int __mnt_unmake_readonly(struct mount
*mnt
)
507 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
512 int sb_prepare_remount_readonly(struct super_block
*sb
)
517 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
518 if (atomic_long_read(&sb
->s_remove_count
))
522 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
523 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
524 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
526 if (mnt_get_writers(mnt
) > 0) {
532 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
536 sb
->s_readonly_remount
= 1;
539 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
540 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
541 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
548 static void free_vfsmnt(struct mount
*mnt
)
550 kfree_const(mnt
->mnt_devname
);
552 free_percpu(mnt
->mnt_pcp
);
554 kmem_cache_free(mnt_cache
, mnt
);
557 static void delayed_free_vfsmnt(struct rcu_head
*head
)
559 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
562 /* call under rcu_read_lock */
563 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
566 if (read_seqretry(&mount_lock
, seq
))
570 mnt
= real_mount(bastard
);
571 mnt_add_count(mnt
, 1);
572 smp_mb(); // see mntput_no_expire()
573 if (likely(!read_seqretry(&mount_lock
, seq
)))
575 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
576 mnt_add_count(mnt
, -1);
580 if (unlikely(bastard
->mnt_flags
& MNT_DOOMED
)) {
581 mnt_add_count(mnt
, -1);
586 /* caller will mntput() */
590 /* call under rcu_read_lock */
591 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
593 int res
= __legitimize_mnt(bastard
, seq
);
596 if (unlikely(res
< 0)) {
605 * find the first mount at @dentry on vfsmount @mnt.
606 * call under rcu_read_lock()
608 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
610 struct hlist_head
*head
= m_hash(mnt
, dentry
);
613 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
614 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
620 * lookup_mnt - Return the first child mount mounted at path
622 * "First" means first mounted chronologically. If you create the
625 * mount /dev/sda1 /mnt
626 * mount /dev/sda2 /mnt
627 * mount /dev/sda3 /mnt
629 * Then lookup_mnt() on the base /mnt dentry in the root mount will
630 * return successively the root dentry and vfsmount of /dev/sda1, then
631 * /dev/sda2, then /dev/sda3, then NULL.
633 * lookup_mnt takes a reference to the found vfsmount.
635 struct vfsmount
*lookup_mnt(const struct path
*path
)
637 struct mount
*child_mnt
;
643 seq
= read_seqbegin(&mount_lock
);
644 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
645 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
646 } while (!legitimize_mnt(m
, seq
));
651 static inline void lock_ns_list(struct mnt_namespace
*ns
)
653 spin_lock(&ns
->ns_lock
);
656 static inline void unlock_ns_list(struct mnt_namespace
*ns
)
658 spin_unlock(&ns
->ns_lock
);
661 static inline bool mnt_is_cursor(struct mount
*mnt
)
663 return mnt
->mnt
.mnt_flags
& MNT_CURSOR
;
667 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
668 * current mount namespace.
670 * The common case is dentries are not mountpoints at all and that
671 * test is handled inline. For the slow case when we are actually
672 * dealing with a mountpoint of some kind, walk through all of the
673 * mounts in the current mount namespace and test to see if the dentry
676 * The mount_hashtable is not usable in the context because we
677 * need to identify all mounts that may be in the current mount
678 * namespace not just a mount that happens to have some specified
681 bool __is_local_mountpoint(struct dentry
*dentry
)
683 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
685 bool is_covered
= false;
687 down_read(&namespace_sem
);
689 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
690 if (mnt_is_cursor(mnt
))
692 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
697 up_read(&namespace_sem
);
702 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
704 struct hlist_head
*chain
= mp_hash(dentry
);
705 struct mountpoint
*mp
;
707 hlist_for_each_entry(mp
, chain
, m_hash
) {
708 if (mp
->m_dentry
== dentry
) {
716 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
718 struct mountpoint
*mp
, *new = NULL
;
721 if (d_mountpoint(dentry
)) {
722 /* might be worth a WARN_ON() */
723 if (d_unlinked(dentry
))
724 return ERR_PTR(-ENOENT
);
726 read_seqlock_excl(&mount_lock
);
727 mp
= lookup_mountpoint(dentry
);
728 read_sequnlock_excl(&mount_lock
);
734 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
736 return ERR_PTR(-ENOMEM
);
739 /* Exactly one processes may set d_mounted */
740 ret
= d_set_mounted(dentry
);
742 /* Someone else set d_mounted? */
746 /* The dentry is not available as a mountpoint? */
751 /* Add the new mountpoint to the hash table */
752 read_seqlock_excl(&mount_lock
);
753 new->m_dentry
= dget(dentry
);
755 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
756 INIT_HLIST_HEAD(&new->m_list
);
757 read_sequnlock_excl(&mount_lock
);
767 * vfsmount lock must be held. Additionally, the caller is responsible
768 * for serializing calls for given disposal list.
770 static void __put_mountpoint(struct mountpoint
*mp
, struct list_head
*list
)
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 dput_to_list(dentry
, list
);
779 hlist_del(&mp
->m_hash
);
784 /* called with namespace_lock and vfsmount lock */
785 static void put_mountpoint(struct mountpoint
*mp
)
787 __put_mountpoint(mp
, &ex_mountpoints
);
790 static inline int check_mnt(struct mount
*mnt
)
792 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
796 * vfsmount lock must be held for write
798 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
802 wake_up_interruptible(&ns
->poll
);
807 * vfsmount lock must be held for write
809 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
811 if (ns
&& ns
->event
!= event
) {
813 wake_up_interruptible(&ns
->poll
);
818 * vfsmount lock must be held for write
820 static struct mountpoint
*unhash_mnt(struct mount
*mnt
)
822 struct mountpoint
*mp
;
823 mnt
->mnt_parent
= mnt
;
824 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
825 list_del_init(&mnt
->mnt_child
);
826 hlist_del_init_rcu(&mnt
->mnt_hash
);
827 hlist_del_init(&mnt
->mnt_mp_list
);
834 * vfsmount lock must be held for write
836 static void umount_mnt(struct mount
*mnt
)
838 put_mountpoint(unhash_mnt(mnt
));
842 * vfsmount lock must be held for write
844 void mnt_set_mountpoint(struct mount
*mnt
,
845 struct mountpoint
*mp
,
846 struct mount
*child_mnt
)
849 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
850 child_mnt
->mnt_mountpoint
= mp
->m_dentry
;
851 child_mnt
->mnt_parent
= mnt
;
852 child_mnt
->mnt_mp
= mp
;
853 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
856 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
858 hlist_add_head_rcu(&mnt
->mnt_hash
,
859 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
860 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
864 * vfsmount lock must be held for write
866 static void attach_mnt(struct mount
*mnt
,
867 struct mount
*parent
,
868 struct mountpoint
*mp
)
870 mnt_set_mountpoint(parent
, mp
, mnt
);
871 __attach_mnt(mnt
, parent
);
874 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
876 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
877 struct mount
*old_parent
= mnt
->mnt_parent
;
879 list_del_init(&mnt
->mnt_child
);
880 hlist_del_init(&mnt
->mnt_mp_list
);
881 hlist_del_init_rcu(&mnt
->mnt_hash
);
883 attach_mnt(mnt
, parent
, mp
);
885 put_mountpoint(old_mp
);
886 mnt_add_count(old_parent
, -1);
890 * vfsmount lock must be held for write
892 static void commit_tree(struct mount
*mnt
)
894 struct mount
*parent
= mnt
->mnt_parent
;
897 struct mnt_namespace
*n
= parent
->mnt_ns
;
899 BUG_ON(parent
== mnt
);
901 list_add_tail(&head
, &mnt
->mnt_list
);
902 list_for_each_entry(m
, &head
, mnt_list
)
905 list_splice(&head
, n
->list
.prev
);
907 n
->mounts
+= n
->pending_mounts
;
908 n
->pending_mounts
= 0;
910 __attach_mnt(mnt
, parent
);
911 touch_mnt_namespace(n
);
914 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
916 struct list_head
*next
= p
->mnt_mounts
.next
;
917 if (next
== &p
->mnt_mounts
) {
921 next
= p
->mnt_child
.next
;
922 if (next
!= &p
->mnt_parent
->mnt_mounts
)
927 return list_entry(next
, struct mount
, mnt_child
);
930 static struct mount
*skip_mnt_tree(struct mount
*p
)
932 struct list_head
*prev
= p
->mnt_mounts
.prev
;
933 while (prev
!= &p
->mnt_mounts
) {
934 p
= list_entry(prev
, struct mount
, mnt_child
);
935 prev
= p
->mnt_mounts
.prev
;
941 * vfs_create_mount - Create a mount for a configured superblock
942 * @fc: The configuration context with the superblock attached
944 * Create a mount to an already configured superblock. If necessary, the
945 * caller should invoke vfs_get_tree() before calling this.
947 * Note that this does not attach the mount to anything.
949 struct vfsmount
*vfs_create_mount(struct fs_context
*fc
)
954 return ERR_PTR(-EINVAL
);
956 mnt
= alloc_vfsmnt(fc
->source
?: "none");
958 return ERR_PTR(-ENOMEM
);
960 if (fc
->sb_flags
& SB_KERNMOUNT
)
961 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
963 atomic_inc(&fc
->root
->d_sb
->s_active
);
964 mnt
->mnt
.mnt_sb
= fc
->root
->d_sb
;
965 mnt
->mnt
.mnt_root
= dget(fc
->root
);
966 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
967 mnt
->mnt_parent
= mnt
;
970 list_add_tail(&mnt
->mnt_instance
, &mnt
->mnt
.mnt_sb
->s_mounts
);
974 EXPORT_SYMBOL(vfs_create_mount
);
976 struct vfsmount
*fc_mount(struct fs_context
*fc
)
978 int err
= vfs_get_tree(fc
);
980 up_write(&fc
->root
->d_sb
->s_umount
);
981 return vfs_create_mount(fc
);
985 EXPORT_SYMBOL(fc_mount
);
987 struct vfsmount
*vfs_kern_mount(struct file_system_type
*type
,
988 int flags
, const char *name
,
991 struct fs_context
*fc
;
992 struct vfsmount
*mnt
;
996 return ERR_PTR(-EINVAL
);
998 fc
= fs_context_for_mount(type
, flags
);
1000 return ERR_CAST(fc
);
1003 ret
= vfs_parse_fs_string(fc
, "source",
1004 name
, strlen(name
));
1006 ret
= parse_monolithic_mount_data(fc
, data
);
1015 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1018 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1019 const char *name
, void *data
)
1021 /* Until it is worked out how to pass the user namespace
1022 * through from the parent mount to the submount don't support
1023 * unprivileged mounts with submounts.
1025 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1026 return ERR_PTR(-EPERM
);
1028 return vfs_kern_mount(type
, SB_SUBMOUNT
, name
, data
);
1030 EXPORT_SYMBOL_GPL(vfs_submount
);
1032 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1035 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1039 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1041 return ERR_PTR(-ENOMEM
);
1043 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1044 mnt
->mnt_group_id
= 0; /* not a peer of original */
1046 mnt
->mnt_group_id
= old
->mnt_group_id
;
1048 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1049 err
= mnt_alloc_group_id(mnt
);
1054 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
;
1055 mnt
->mnt
.mnt_flags
&= ~(MNT_WRITE_HOLD
|MNT_MARKED
|MNT_INTERNAL
);
1057 atomic_inc(&sb
->s_active
);
1058 mnt
->mnt
.mnt_sb
= sb
;
1059 mnt
->mnt
.mnt_root
= dget(root
);
1060 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1061 mnt
->mnt_parent
= mnt
;
1063 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1064 unlock_mount_hash();
1066 if ((flag
& CL_SLAVE
) ||
1067 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1068 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1069 mnt
->mnt_master
= old
;
1070 CLEAR_MNT_SHARED(mnt
);
1071 } else if (!(flag
& CL_PRIVATE
)) {
1072 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1073 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1074 if (IS_MNT_SLAVE(old
))
1075 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1076 mnt
->mnt_master
= old
->mnt_master
;
1078 CLEAR_MNT_SHARED(mnt
);
1080 if (flag
& CL_MAKE_SHARED
)
1081 set_mnt_shared(mnt
);
1083 /* stick the duplicate mount on the same expiry list
1084 * as the original if that was on one */
1085 if (flag
& CL_EXPIRE
) {
1086 if (!list_empty(&old
->mnt_expire
))
1087 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1095 return ERR_PTR(err
);
1098 static void cleanup_mnt(struct mount
*mnt
)
1100 struct hlist_node
*p
;
1103 * The warning here probably indicates that somebody messed
1104 * up a mnt_want/drop_write() pair. If this happens, the
1105 * filesystem was probably unable to make r/w->r/o transitions.
1106 * The locking used to deal with mnt_count decrement provides barriers,
1107 * so mnt_get_writers() below is safe.
1109 WARN_ON(mnt_get_writers(mnt
));
1110 if (unlikely(mnt
->mnt_pins
.first
))
1112 hlist_for_each_entry_safe(m
, p
, &mnt
->mnt_stuck_children
, mnt_umount
) {
1113 hlist_del(&m
->mnt_umount
);
1116 fsnotify_vfsmount_delete(&mnt
->mnt
);
1117 dput(mnt
->mnt
.mnt_root
);
1118 deactivate_super(mnt
->mnt
.mnt_sb
);
1120 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1123 static void __cleanup_mnt(struct rcu_head
*head
)
1125 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1128 static LLIST_HEAD(delayed_mntput_list
);
1129 static void delayed_mntput(struct work_struct
*unused
)
1131 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1132 struct mount
*m
, *t
;
1134 llist_for_each_entry_safe(m
, t
, node
, mnt_llist
)
1137 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1139 static void mntput_no_expire(struct mount
*mnt
)
1145 if (likely(READ_ONCE(mnt
->mnt_ns
))) {
1147 * Since we don't do lock_mount_hash() here,
1148 * ->mnt_ns can change under us. However, if it's
1149 * non-NULL, then there's a reference that won't
1150 * be dropped until after an RCU delay done after
1151 * turning ->mnt_ns NULL. So if we observe it
1152 * non-NULL under rcu_read_lock(), the reference
1153 * we are dropping is not the final one.
1155 mnt_add_count(mnt
, -1);
1161 * make sure that if __legitimize_mnt() has not seen us grab
1162 * mount_lock, we'll see their refcount increment here.
1165 mnt_add_count(mnt
, -1);
1166 count
= mnt_get_count(mnt
);
1170 unlock_mount_hash();
1173 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1175 unlock_mount_hash();
1178 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1181 list_del(&mnt
->mnt_instance
);
1183 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1184 struct mount
*p
, *tmp
;
1185 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1186 __put_mountpoint(unhash_mnt(p
), &list
);
1187 hlist_add_head(&p
->mnt_umount
, &mnt
->mnt_stuck_children
);
1190 unlock_mount_hash();
1191 shrink_dentry_list(&list
);
1193 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1194 struct task_struct
*task
= current
;
1195 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1196 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1197 if (!task_work_add(task
, &mnt
->mnt_rcu
, TWA_RESUME
))
1200 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1201 schedule_delayed_work(&delayed_mntput_work
, 1);
1207 void mntput(struct vfsmount
*mnt
)
1210 struct mount
*m
= real_mount(mnt
);
1211 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1212 if (unlikely(m
->mnt_expiry_mark
))
1213 m
->mnt_expiry_mark
= 0;
1214 mntput_no_expire(m
);
1217 EXPORT_SYMBOL(mntput
);
1219 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1222 mnt_add_count(real_mount(mnt
), 1);
1225 EXPORT_SYMBOL(mntget
);
1227 /* path_is_mountpoint() - Check if path is a mount in the current
1230 * d_mountpoint() can only be used reliably to establish if a dentry is
1231 * not mounted in any namespace and that common case is handled inline.
1232 * d_mountpoint() isn't aware of the possibility there may be multiple
1233 * mounts using a given dentry in a different namespace. This function
1234 * checks if the passed in path is a mountpoint rather than the dentry
1237 bool path_is_mountpoint(const struct path
*path
)
1242 if (!d_mountpoint(path
->dentry
))
1247 seq
= read_seqbegin(&mount_lock
);
1248 res
= __path_is_mountpoint(path
);
1249 } while (read_seqretry(&mount_lock
, seq
));
1254 EXPORT_SYMBOL(path_is_mountpoint
);
1256 struct vfsmount
*mnt_clone_internal(const struct path
*path
)
1259 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1262 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1266 #ifdef CONFIG_PROC_FS
1267 static struct mount
*mnt_list_next(struct mnt_namespace
*ns
,
1268 struct list_head
*p
)
1270 struct mount
*mnt
, *ret
= NULL
;
1273 list_for_each_continue(p
, &ns
->list
) {
1274 mnt
= list_entry(p
, typeof(*mnt
), mnt_list
);
1275 if (!mnt_is_cursor(mnt
)) {
1285 /* iterator; we want it to have access to namespace_sem, thus here... */
1286 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1288 struct proc_mounts
*p
= m
->private;
1289 struct list_head
*prev
;
1291 down_read(&namespace_sem
);
1293 prev
= &p
->ns
->list
;
1295 prev
= &p
->cursor
.mnt_list
;
1297 /* Read after we'd reached the end? */
1298 if (list_empty(prev
))
1302 return mnt_list_next(p
->ns
, prev
);
1305 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1307 struct proc_mounts
*p
= m
->private;
1308 struct mount
*mnt
= v
;
1311 return mnt_list_next(p
->ns
, &mnt
->mnt_list
);
1314 static void m_stop(struct seq_file
*m
, void *v
)
1316 struct proc_mounts
*p
= m
->private;
1317 struct mount
*mnt
= v
;
1319 lock_ns_list(p
->ns
);
1321 list_move_tail(&p
->cursor
.mnt_list
, &mnt
->mnt_list
);
1323 list_del_init(&p
->cursor
.mnt_list
);
1324 unlock_ns_list(p
->ns
);
1325 up_read(&namespace_sem
);
1328 static int m_show(struct seq_file
*m
, void *v
)
1330 struct proc_mounts
*p
= m
->private;
1331 struct mount
*r
= v
;
1332 return p
->show(m
, &r
->mnt
);
1335 const struct seq_operations mounts_op
= {
1342 void mnt_cursor_del(struct mnt_namespace
*ns
, struct mount
*cursor
)
1344 down_read(&namespace_sem
);
1346 list_del(&cursor
->mnt_list
);
1348 up_read(&namespace_sem
);
1350 #endif /* CONFIG_PROC_FS */
1353 * may_umount_tree - check if a mount tree is busy
1354 * @mnt: root of mount tree
1356 * This is called to check if a tree of mounts has any
1357 * open files, pwds, chroots or sub mounts that are
1360 int may_umount_tree(struct vfsmount
*m
)
1362 struct mount
*mnt
= real_mount(m
);
1363 int actual_refs
= 0;
1364 int minimum_refs
= 0;
1368 /* write lock needed for mnt_get_count */
1370 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1371 actual_refs
+= mnt_get_count(p
);
1374 unlock_mount_hash();
1376 if (actual_refs
> minimum_refs
)
1382 EXPORT_SYMBOL(may_umount_tree
);
1385 * may_umount - check if a mount point is busy
1386 * @mnt: root of mount
1388 * This is called to check if a mount point has any
1389 * open files, pwds, chroots or sub mounts. If the
1390 * mount has sub mounts this will return busy
1391 * regardless of whether the sub mounts are busy.
1393 * Doesn't take quota and stuff into account. IOW, in some cases it will
1394 * give false negatives. The main reason why it's here is that we need
1395 * a non-destructive way to look for easily umountable filesystems.
1397 int may_umount(struct vfsmount
*mnt
)
1400 down_read(&namespace_sem
);
1402 if (propagate_mount_busy(real_mount(mnt
), 2))
1404 unlock_mount_hash();
1405 up_read(&namespace_sem
);
1409 EXPORT_SYMBOL(may_umount
);
1411 static void namespace_unlock(void)
1413 struct hlist_head head
;
1414 struct hlist_node
*p
;
1418 hlist_move_list(&unmounted
, &head
);
1419 list_splice_init(&ex_mountpoints
, &list
);
1421 up_write(&namespace_sem
);
1423 shrink_dentry_list(&list
);
1425 if (likely(hlist_empty(&head
)))
1428 synchronize_rcu_expedited();
1430 hlist_for_each_entry_safe(m
, p
, &head
, mnt_umount
) {
1431 hlist_del(&m
->mnt_umount
);
1436 static inline void namespace_lock(void)
1438 down_write(&namespace_sem
);
1441 enum umount_tree_flags
{
1443 UMOUNT_PROPAGATE
= 2,
1444 UMOUNT_CONNECTED
= 4,
1447 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1449 /* Leaving mounts connected is only valid for lazy umounts */
1450 if (how
& UMOUNT_SYNC
)
1453 /* A mount without a parent has nothing to be connected to */
1454 if (!mnt_has_parent(mnt
))
1457 /* Because the reference counting rules change when mounts are
1458 * unmounted and connected, umounted mounts may not be
1459 * connected to mounted mounts.
1461 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1464 /* Has it been requested that the mount remain connected? */
1465 if (how
& UMOUNT_CONNECTED
)
1468 /* Is the mount locked such that it needs to remain connected? */
1469 if (IS_MNT_LOCKED(mnt
))
1472 /* By default disconnect the mount */
1477 * mount_lock must be held
1478 * namespace_sem must be held for write
1480 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1482 LIST_HEAD(tmp_list
);
1485 if (how
& UMOUNT_PROPAGATE
)
1486 propagate_mount_unlock(mnt
);
1488 /* Gather the mounts to umount */
1489 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1490 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1491 list_move(&p
->mnt_list
, &tmp_list
);
1494 /* Hide the mounts from mnt_mounts */
1495 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1496 list_del_init(&p
->mnt_child
);
1499 /* Add propogated mounts to the tmp_list */
1500 if (how
& UMOUNT_PROPAGATE
)
1501 propagate_umount(&tmp_list
);
1503 while (!list_empty(&tmp_list
)) {
1504 struct mnt_namespace
*ns
;
1506 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1507 list_del_init(&p
->mnt_expire
);
1508 list_del_init(&p
->mnt_list
);
1512 __touch_mnt_namespace(ns
);
1515 if (how
& UMOUNT_SYNC
)
1516 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1518 disconnect
= disconnect_mount(p
, how
);
1519 if (mnt_has_parent(p
)) {
1520 mnt_add_count(p
->mnt_parent
, -1);
1522 /* Don't forget about p */
1523 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1528 change_mnt_propagation(p
, MS_PRIVATE
);
1530 hlist_add_head(&p
->mnt_umount
, &unmounted
);
1534 static void shrink_submounts(struct mount
*mnt
);
1536 static int do_umount_root(struct super_block
*sb
)
1540 down_write(&sb
->s_umount
);
1541 if (!sb_rdonly(sb
)) {
1542 struct fs_context
*fc
;
1544 fc
= fs_context_for_reconfigure(sb
->s_root
, SB_RDONLY
,
1549 ret
= parse_monolithic_mount_data(fc
, NULL
);
1551 ret
= reconfigure_super(fc
);
1555 up_write(&sb
->s_umount
);
1559 static int do_umount(struct mount
*mnt
, int flags
)
1561 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1564 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1569 * Allow userspace to request a mountpoint be expired rather than
1570 * unmounting unconditionally. Unmount only happens if:
1571 * (1) the mark is already set (the mark is cleared by mntput())
1572 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1574 if (flags
& MNT_EXPIRE
) {
1575 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1576 flags
& (MNT_FORCE
| MNT_DETACH
))
1580 * probably don't strictly need the lock here if we examined
1581 * all race cases, but it's a slowpath.
1584 if (mnt_get_count(mnt
) != 2) {
1585 unlock_mount_hash();
1588 unlock_mount_hash();
1590 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1595 * If we may have to abort operations to get out of this
1596 * mount, and they will themselves hold resources we must
1597 * allow the fs to do things. In the Unix tradition of
1598 * 'Gee thats tricky lets do it in userspace' the umount_begin
1599 * might fail to complete on the first run through as other tasks
1600 * must return, and the like. Thats for the mount program to worry
1601 * about for the moment.
1604 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1605 sb
->s_op
->umount_begin(sb
);
1609 * No sense to grab the lock for this test, but test itself looks
1610 * somewhat bogus. Suggestions for better replacement?
1611 * Ho-hum... In principle, we might treat that as umount + switch
1612 * to rootfs. GC would eventually take care of the old vfsmount.
1613 * Actually it makes sense, especially if rootfs would contain a
1614 * /reboot - static binary that would close all descriptors and
1615 * call reboot(9). Then init(8) could umount root and exec /reboot.
1617 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1619 * Special case for "unmounting" root ...
1620 * we just try to remount it readonly.
1622 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1624 return do_umount_root(sb
);
1630 /* Recheck MNT_LOCKED with the locks held */
1632 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1636 if (flags
& MNT_DETACH
) {
1637 if (!list_empty(&mnt
->mnt_list
))
1638 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1641 shrink_submounts(mnt
);
1643 if (!propagate_mount_busy(mnt
, 2)) {
1644 if (!list_empty(&mnt
->mnt_list
))
1645 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1650 unlock_mount_hash();
1656 * __detach_mounts - lazily unmount all mounts on the specified dentry
1658 * During unlink, rmdir, and d_drop it is possible to loose the path
1659 * to an existing mountpoint, and wind up leaking the mount.
1660 * detach_mounts allows lazily unmounting those mounts instead of
1663 * The caller may hold dentry->d_inode->i_mutex.
1665 void __detach_mounts(struct dentry
*dentry
)
1667 struct mountpoint
*mp
;
1672 mp
= lookup_mountpoint(dentry
);
1677 while (!hlist_empty(&mp
->m_list
)) {
1678 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1679 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1681 hlist_add_head(&mnt
->mnt_umount
, &unmounted
);
1683 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1687 unlock_mount_hash();
1692 * Is the caller allowed to modify his namespace?
1694 static inline bool may_mount(void)
1696 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1699 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1700 static inline bool may_mandlock(void)
1702 return capable(CAP_SYS_ADMIN
);
1705 static inline bool may_mandlock(void)
1707 pr_warn("VFS: \"mand\" mount option not supported");
1712 static int can_umount(const struct path
*path
, int flags
)
1714 struct mount
*mnt
= real_mount(path
->mnt
);
1718 if (path
->dentry
!= path
->mnt
->mnt_root
)
1720 if (!check_mnt(mnt
))
1722 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
) /* Check optimistically */
1724 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1729 // caller is responsible for flags being sane
1730 int path_umount(struct path
*path
, int flags
)
1732 struct mount
*mnt
= real_mount(path
->mnt
);
1735 ret
= can_umount(path
, flags
);
1737 ret
= do_umount(mnt
, flags
);
1739 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1741 mntput_no_expire(mnt
);
1745 static int ksys_umount(char __user
*name
, int flags
)
1747 int lookup_flags
= LOOKUP_MOUNTPOINT
;
1751 // basic validity checks done first
1752 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1755 if (!(flags
& UMOUNT_NOFOLLOW
))
1756 lookup_flags
|= LOOKUP_FOLLOW
;
1757 ret
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1760 return path_umount(&path
, flags
);
1763 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1765 return ksys_umount(name
, flags
);
1768 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1771 * The 2.0 compatible umount. No flags.
1773 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1775 return ksys_umount(name
, 0);
1780 static bool is_mnt_ns_file(struct dentry
*dentry
)
1782 /* Is this a proxy for a mount namespace? */
1783 return dentry
->d_op
== &ns_dentry_operations
&&
1784 dentry
->d_fsdata
== &mntns_operations
;
1787 static struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1789 return container_of(ns
, struct mnt_namespace
, ns
);
1792 struct ns_common
*from_mnt_ns(struct mnt_namespace
*mnt
)
1797 static bool mnt_ns_loop(struct dentry
*dentry
)
1799 /* Could bind mounting the mount namespace inode cause a
1800 * mount namespace loop?
1802 struct mnt_namespace
*mnt_ns
;
1803 if (!is_mnt_ns_file(dentry
))
1806 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1807 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1810 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1813 struct mount
*res
, *p
, *q
, *r
, *parent
;
1815 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1816 return ERR_PTR(-EINVAL
);
1818 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1819 return ERR_PTR(-EINVAL
);
1821 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1825 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1828 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1830 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1833 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1834 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1835 IS_MNT_UNBINDABLE(s
)) {
1836 if (s
->mnt
.mnt_flags
& MNT_LOCKED
) {
1837 /* Both unbindable and locked. */
1838 q
= ERR_PTR(-EPERM
);
1841 s
= skip_mnt_tree(s
);
1845 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1846 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1847 s
= skip_mnt_tree(s
);
1850 while (p
!= s
->mnt_parent
) {
1856 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1860 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1861 attach_mnt(q
, parent
, p
->mnt_mp
);
1862 unlock_mount_hash();
1869 umount_tree(res
, UMOUNT_SYNC
);
1870 unlock_mount_hash();
1875 /* Caller should check returned pointer for errors */
1877 struct vfsmount
*collect_mounts(const struct path
*path
)
1881 if (!check_mnt(real_mount(path
->mnt
)))
1882 tree
= ERR_PTR(-EINVAL
);
1884 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1885 CL_COPY_ALL
| CL_PRIVATE
);
1888 return ERR_CAST(tree
);
1892 static void free_mnt_ns(struct mnt_namespace
*);
1893 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*, bool);
1895 void dissolve_on_fput(struct vfsmount
*mnt
)
1897 struct mnt_namespace
*ns
;
1900 ns
= real_mount(mnt
)->mnt_ns
;
1903 umount_tree(real_mount(mnt
), UMOUNT_CONNECTED
);
1907 unlock_mount_hash();
1913 void drop_collected_mounts(struct vfsmount
*mnt
)
1917 umount_tree(real_mount(mnt
), 0);
1918 unlock_mount_hash();
1923 * clone_private_mount - create a private clone of a path
1925 * This creates a new vfsmount, which will be the clone of @path. The new will
1926 * not be attached anywhere in the namespace and will be private (i.e. changes
1927 * to the originating mount won't be propagated into this).
1929 * Release with mntput().
1931 struct vfsmount
*clone_private_mount(const struct path
*path
)
1933 struct mount
*old_mnt
= real_mount(path
->mnt
);
1934 struct mount
*new_mnt
;
1936 if (IS_MNT_UNBINDABLE(old_mnt
))
1937 return ERR_PTR(-EINVAL
);
1939 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1940 if (IS_ERR(new_mnt
))
1941 return ERR_CAST(new_mnt
);
1943 /* Longterm mount to be removed by kern_unmount*() */
1944 new_mnt
->mnt_ns
= MNT_NS_INTERNAL
;
1946 return &new_mnt
->mnt
;
1948 EXPORT_SYMBOL_GPL(clone_private_mount
);
1950 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1951 struct vfsmount
*root
)
1954 int res
= f(root
, arg
);
1957 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1958 res
= f(&mnt
->mnt
, arg
);
1965 static void lock_mnt_tree(struct mount
*mnt
)
1969 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1970 int flags
= p
->mnt
.mnt_flags
;
1971 /* Don't allow unprivileged users to change mount flags */
1972 flags
|= MNT_LOCK_ATIME
;
1974 if (flags
& MNT_READONLY
)
1975 flags
|= MNT_LOCK_READONLY
;
1977 if (flags
& MNT_NODEV
)
1978 flags
|= MNT_LOCK_NODEV
;
1980 if (flags
& MNT_NOSUID
)
1981 flags
|= MNT_LOCK_NOSUID
;
1983 if (flags
& MNT_NOEXEC
)
1984 flags
|= MNT_LOCK_NOEXEC
;
1985 /* Don't allow unprivileged users to reveal what is under a mount */
1986 if (list_empty(&p
->mnt_expire
))
1987 flags
|= MNT_LOCKED
;
1988 p
->mnt
.mnt_flags
= flags
;
1992 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1996 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1997 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1998 mnt_release_group_id(p
);
2002 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
2006 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
2007 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
2008 int err
= mnt_alloc_group_id(p
);
2010 cleanup_group_ids(mnt
, p
);
2019 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
2021 unsigned int max
= READ_ONCE(sysctl_mount_max
);
2022 unsigned int mounts
= 0, old
, pending
, sum
;
2025 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
2029 pending
= ns
->pending_mounts
;
2030 sum
= old
+ pending
;
2034 (mounts
> (max
- sum
)))
2037 ns
->pending_mounts
= pending
+ mounts
;
2042 * @source_mnt : mount tree to be attached
2043 * @nd : place the mount tree @source_mnt is attached
2044 * @parent_nd : if non-null, detach the source_mnt from its parent and
2045 * store the parent mount and mountpoint dentry.
2046 * (done when source_mnt is moved)
2048 * NOTE: in the table below explains the semantics when a source mount
2049 * of a given type is attached to a destination mount of a given type.
2050 * ---------------------------------------------------------------------------
2051 * | BIND MOUNT OPERATION |
2052 * |**************************************************************************
2053 * | source-->| shared | private | slave | unbindable |
2057 * |**************************************************************************
2058 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2060 * |non-shared| shared (+) | private | slave (*) | invalid |
2061 * ***************************************************************************
2062 * A bind operation clones the source mount and mounts the clone on the
2063 * destination mount.
2065 * (++) the cloned mount is propagated to all the mounts in the propagation
2066 * tree of the destination mount and the cloned mount is added to
2067 * the peer group of the source mount.
2068 * (+) the cloned mount is created under the destination mount and is marked
2069 * as shared. The cloned mount is added to the peer group of the source
2071 * (+++) the mount is propagated to all the mounts in the propagation tree
2072 * of the destination mount and the cloned mount is made slave
2073 * of the same master as that of the source mount. The cloned mount
2074 * is marked as 'shared and slave'.
2075 * (*) the cloned mount is made a slave of the same master as that of the
2078 * ---------------------------------------------------------------------------
2079 * | MOVE MOUNT OPERATION |
2080 * |**************************************************************************
2081 * | source-->| shared | private | slave | unbindable |
2085 * |**************************************************************************
2086 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2088 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2089 * ***************************************************************************
2091 * (+) the mount is moved to the destination. And is then propagated to
2092 * all the mounts in the propagation tree of the destination mount.
2093 * (+*) the mount is moved to the destination.
2094 * (+++) the mount is moved to the destination and is then propagated to
2095 * all the mounts belonging to the destination mount's propagation tree.
2096 * the mount is marked as 'shared and slave'.
2097 * (*) the mount continues to be a slave at the new location.
2099 * if the source mount is a tree, the operations explained above is
2100 * applied to each mount in the tree.
2101 * Must be called without spinlocks held, since this function can sleep
2104 static int attach_recursive_mnt(struct mount
*source_mnt
,
2105 struct mount
*dest_mnt
,
2106 struct mountpoint
*dest_mp
,
2109 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2110 HLIST_HEAD(tree_list
);
2111 struct mnt_namespace
*ns
= dest_mnt
->mnt_ns
;
2112 struct mountpoint
*smp
;
2113 struct mount
*child
, *p
;
2114 struct hlist_node
*n
;
2117 /* Preallocate a mountpoint in case the new mounts need
2118 * to be tucked under other mounts.
2120 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
2122 return PTR_ERR(smp
);
2124 /* Is there space to add these mounts to the mount namespace? */
2126 err
= count_mounts(ns
, source_mnt
);
2131 if (IS_MNT_SHARED(dest_mnt
)) {
2132 err
= invent_group_ids(source_mnt
, true);
2135 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2138 goto out_cleanup_ids
;
2139 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2145 unhash_mnt(source_mnt
);
2146 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
2147 touch_mnt_namespace(source_mnt
->mnt_ns
);
2149 if (source_mnt
->mnt_ns
) {
2150 /* move from anon - the caller will destroy */
2151 list_del_init(&source_mnt
->mnt_ns
->list
);
2153 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2154 commit_tree(source_mnt
);
2157 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2159 hlist_del_init(&child
->mnt_hash
);
2160 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2161 child
->mnt_mountpoint
);
2163 mnt_change_mountpoint(child
, smp
, q
);
2164 /* Notice when we are propagating across user namespaces */
2165 if (child
->mnt_parent
->mnt_ns
->user_ns
!= user_ns
)
2166 lock_mnt_tree(child
);
2167 child
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2170 put_mountpoint(smp
);
2171 unlock_mount_hash();
2176 while (!hlist_empty(&tree_list
)) {
2177 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2178 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2179 umount_tree(child
, UMOUNT_SYNC
);
2181 unlock_mount_hash();
2182 cleanup_group_ids(source_mnt
, NULL
);
2184 ns
->pending_mounts
= 0;
2186 read_seqlock_excl(&mount_lock
);
2187 put_mountpoint(smp
);
2188 read_sequnlock_excl(&mount_lock
);
2193 static struct mountpoint
*lock_mount(struct path
*path
)
2195 struct vfsmount
*mnt
;
2196 struct dentry
*dentry
= path
->dentry
;
2198 inode_lock(dentry
->d_inode
);
2199 if (unlikely(cant_mount(dentry
))) {
2200 inode_unlock(dentry
->d_inode
);
2201 return ERR_PTR(-ENOENT
);
2204 mnt
= lookup_mnt(path
);
2206 struct mountpoint
*mp
= get_mountpoint(dentry
);
2209 inode_unlock(dentry
->d_inode
);
2215 inode_unlock(path
->dentry
->d_inode
);
2218 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2222 static void unlock_mount(struct mountpoint
*where
)
2224 struct dentry
*dentry
= where
->m_dentry
;
2226 read_seqlock_excl(&mount_lock
);
2227 put_mountpoint(where
);
2228 read_sequnlock_excl(&mount_lock
);
2231 inode_unlock(dentry
->d_inode
);
2234 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2236 if (mnt
->mnt
.mnt_sb
->s_flags
& SB_NOUSER
)
2239 if (d_is_dir(mp
->m_dentry
) !=
2240 d_is_dir(mnt
->mnt
.mnt_root
))
2243 return attach_recursive_mnt(mnt
, p
, mp
, false);
2247 * Sanity check the flags to change_mnt_propagation.
2250 static int flags_to_propagation_type(int ms_flags
)
2252 int type
= ms_flags
& ~(MS_REC
| MS_SILENT
);
2254 /* Fail if any non-propagation flags are set */
2255 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2257 /* Only one propagation flag should be set */
2258 if (!is_power_of_2(type
))
2264 * recursively change the type of the mountpoint.
2266 static int do_change_type(struct path
*path
, int ms_flags
)
2269 struct mount
*mnt
= real_mount(path
->mnt
);
2270 int recurse
= ms_flags
& MS_REC
;
2274 if (path
->dentry
!= path
->mnt
->mnt_root
)
2277 type
= flags_to_propagation_type(ms_flags
);
2282 if (type
== MS_SHARED
) {
2283 err
= invent_group_ids(mnt
, recurse
);
2289 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2290 change_mnt_propagation(m
, type
);
2291 unlock_mount_hash();
2298 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2300 struct mount
*child
;
2301 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2302 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2305 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2311 static struct mount
*__do_loopback(struct path
*old_path
, int recurse
)
2313 struct mount
*mnt
= ERR_PTR(-EINVAL
), *old
= real_mount(old_path
->mnt
);
2315 if (IS_MNT_UNBINDABLE(old
))
2318 if (!check_mnt(old
) && old_path
->dentry
->d_op
!= &ns_dentry_operations
)
2321 if (!recurse
&& has_locked_children(old
, old_path
->dentry
))
2325 mnt
= copy_tree(old
, old_path
->dentry
, CL_COPY_MNT_NS_FILE
);
2327 mnt
= clone_mnt(old
, old_path
->dentry
, 0);
2330 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2336 * do loopback mount.
2338 static int do_loopback(struct path
*path
, const char *old_name
,
2341 struct path old_path
;
2342 struct mount
*mnt
= NULL
, *parent
;
2343 struct mountpoint
*mp
;
2345 if (!old_name
|| !*old_name
)
2347 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2352 if (mnt_ns_loop(old_path
.dentry
))
2355 mp
= lock_mount(path
);
2361 parent
= real_mount(path
->mnt
);
2362 if (!check_mnt(parent
))
2365 mnt
= __do_loopback(&old_path
, recurse
);
2371 err
= graft_tree(mnt
, parent
, mp
);
2374 umount_tree(mnt
, UMOUNT_SYNC
);
2375 unlock_mount_hash();
2380 path_put(&old_path
);
2384 static struct file
*open_detached_copy(struct path
*path
, bool recursive
)
2386 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2387 struct mnt_namespace
*ns
= alloc_mnt_ns(user_ns
, true);
2388 struct mount
*mnt
, *p
;
2392 return ERR_CAST(ns
);
2395 mnt
= __do_loopback(path
, recursive
);
2399 return ERR_CAST(mnt
);
2403 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2408 list_add_tail(&ns
->list
, &mnt
->mnt_list
);
2410 unlock_mount_hash();
2414 path
->mnt
= &mnt
->mnt
;
2415 file
= dentry_open(path
, O_PATH
, current_cred());
2417 dissolve_on_fput(path
->mnt
);
2419 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
2423 SYSCALL_DEFINE3(open_tree
, int, dfd
, const char __user
*, filename
, unsigned, flags
)
2427 int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
2428 bool detached
= flags
& OPEN_TREE_CLONE
;
2432 BUILD_BUG_ON(OPEN_TREE_CLOEXEC
!= O_CLOEXEC
);
2434 if (flags
& ~(AT_EMPTY_PATH
| AT_NO_AUTOMOUNT
| AT_RECURSIVE
|
2435 AT_SYMLINK_NOFOLLOW
| OPEN_TREE_CLONE
|
2439 if ((flags
& (AT_RECURSIVE
| OPEN_TREE_CLONE
)) == AT_RECURSIVE
)
2442 if (flags
& AT_NO_AUTOMOUNT
)
2443 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
2444 if (flags
& AT_SYMLINK_NOFOLLOW
)
2445 lookup_flags
&= ~LOOKUP_FOLLOW
;
2446 if (flags
& AT_EMPTY_PATH
)
2447 lookup_flags
|= LOOKUP_EMPTY
;
2449 if (detached
&& !may_mount())
2452 fd
= get_unused_fd_flags(flags
& O_CLOEXEC
);
2456 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
2457 if (unlikely(error
)) {
2458 file
= ERR_PTR(error
);
2461 file
= open_detached_copy(&path
, flags
& AT_RECURSIVE
);
2463 file
= dentry_open(&path
, O_PATH
, current_cred());
2468 return PTR_ERR(file
);
2470 fd_install(fd
, file
);
2475 * Don't allow locked mount flags to be cleared.
2477 * No locks need to be held here while testing the various MNT_LOCK
2478 * flags because those flags can never be cleared once they are set.
2480 static bool can_change_locked_flags(struct mount
*mnt
, unsigned int mnt_flags
)
2482 unsigned int fl
= mnt
->mnt
.mnt_flags
;
2484 if ((fl
& MNT_LOCK_READONLY
) &&
2485 !(mnt_flags
& MNT_READONLY
))
2488 if ((fl
& MNT_LOCK_NODEV
) &&
2489 !(mnt_flags
& MNT_NODEV
))
2492 if ((fl
& MNT_LOCK_NOSUID
) &&
2493 !(mnt_flags
& MNT_NOSUID
))
2496 if ((fl
& MNT_LOCK_NOEXEC
) &&
2497 !(mnt_flags
& MNT_NOEXEC
))
2500 if ((fl
& MNT_LOCK_ATIME
) &&
2501 ((fl
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
)))
2507 static int change_mount_ro_state(struct mount
*mnt
, unsigned int mnt_flags
)
2509 bool readonly_request
= (mnt_flags
& MNT_READONLY
);
2511 if (readonly_request
== __mnt_is_readonly(&mnt
->mnt
))
2514 if (readonly_request
)
2515 return mnt_make_readonly(mnt
);
2517 return __mnt_unmake_readonly(mnt
);
2521 * Update the user-settable attributes on a mount. The caller must hold
2522 * sb->s_umount for writing.
2524 static void set_mount_attributes(struct mount
*mnt
, unsigned int mnt_flags
)
2527 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2528 mnt
->mnt
.mnt_flags
= mnt_flags
;
2529 touch_mnt_namespace(mnt
->mnt_ns
);
2530 unlock_mount_hash();
2533 static void mnt_warn_timestamp_expiry(struct path
*mountpoint
, struct vfsmount
*mnt
)
2535 struct super_block
*sb
= mnt
->mnt_sb
;
2537 if (!__mnt_is_readonly(mnt
) &&
2538 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX
> sb
->s_time_max
)) {
2539 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
2540 char *mntpath
= buf
? d_path(mountpoint
, buf
, PAGE_SIZE
) : ERR_PTR(-ENOMEM
);
2543 time64_to_tm(sb
->s_time_max
, 0, &tm
);
2545 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2547 is_mounted(mnt
) ? "remounted" : "mounted",
2549 tm
.tm_year
+1900, (unsigned long long)sb
->s_time_max
);
2551 free_page((unsigned long)buf
);
2556 * Handle reconfiguration of the mountpoint only without alteration of the
2557 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2560 static int do_reconfigure_mnt(struct path
*path
, unsigned int mnt_flags
)
2562 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2563 struct mount
*mnt
= real_mount(path
->mnt
);
2566 if (!check_mnt(mnt
))
2569 if (path
->dentry
!= mnt
->mnt
.mnt_root
)
2572 if (!can_change_locked_flags(mnt
, mnt_flags
))
2575 down_write(&sb
->s_umount
);
2576 ret
= change_mount_ro_state(mnt
, mnt_flags
);
2578 set_mount_attributes(mnt
, mnt_flags
);
2579 up_write(&sb
->s_umount
);
2581 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2587 * change filesystem flags. dir should be a physical root of filesystem.
2588 * If you've mounted a non-root directory somewhere and want to do remount
2589 * on it - tough luck.
2591 static int do_remount(struct path
*path
, int ms_flags
, int sb_flags
,
2592 int mnt_flags
, void *data
)
2595 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2596 struct mount
*mnt
= real_mount(path
->mnt
);
2597 struct fs_context
*fc
;
2599 if (!check_mnt(mnt
))
2602 if (path
->dentry
!= path
->mnt
->mnt_root
)
2605 if (!can_change_locked_flags(mnt
, mnt_flags
))
2608 fc
= fs_context_for_reconfigure(path
->dentry
, sb_flags
, MS_RMT_MASK
);
2613 err
= parse_monolithic_mount_data(fc
, data
);
2615 down_write(&sb
->s_umount
);
2617 if (ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
)) {
2618 err
= reconfigure_super(fc
);
2620 set_mount_attributes(mnt
, mnt_flags
);
2622 up_write(&sb
->s_umount
);
2625 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2631 static inline int tree_contains_unbindable(struct mount
*mnt
)
2634 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2635 if (IS_MNT_UNBINDABLE(p
))
2642 * Check that there aren't references to earlier/same mount namespaces in the
2643 * specified subtree. Such references can act as pins for mount namespaces
2644 * that aren't checked by the mount-cycle checking code, thereby allowing
2645 * cycles to be made.
2647 static bool check_for_nsfs_mounts(struct mount
*subtree
)
2653 for (p
= subtree
; p
; p
= next_mnt(p
, subtree
))
2654 if (mnt_ns_loop(p
->mnt
.mnt_root
))
2659 unlock_mount_hash();
2663 static int do_move_mount(struct path
*old_path
, struct path
*new_path
)
2665 struct mnt_namespace
*ns
;
2668 struct mount
*parent
;
2669 struct mountpoint
*mp
, *old_mp
;
2673 mp
= lock_mount(new_path
);
2677 old
= real_mount(old_path
->mnt
);
2678 p
= real_mount(new_path
->mnt
);
2679 parent
= old
->mnt_parent
;
2680 attached
= mnt_has_parent(old
);
2681 old_mp
= old
->mnt_mp
;
2685 /* The mountpoint must be in our namespace. */
2689 /* The thing moved must be mounted... */
2690 if (!is_mounted(&old
->mnt
))
2693 /* ... and either ours or the root of anon namespace */
2694 if (!(attached
? check_mnt(old
) : is_anon_ns(ns
)))
2697 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2700 if (old_path
->dentry
!= old_path
->mnt
->mnt_root
)
2703 if (d_is_dir(new_path
->dentry
) !=
2704 d_is_dir(old_path
->dentry
))
2707 * Don't move a mount residing in a shared parent.
2709 if (attached
&& IS_MNT_SHARED(parent
))
2712 * Don't move a mount tree containing unbindable mounts to a destination
2713 * mount which is shared.
2715 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2718 if (!check_for_nsfs_mounts(old
))
2720 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2724 err
= attach_recursive_mnt(old
, real_mount(new_path
->mnt
), mp
,
2729 /* if the mount is moved, it should no longer be expire
2731 list_del_init(&old
->mnt_expire
);
2733 put_mountpoint(old_mp
);
2738 mntput_no_expire(parent
);
2745 static int do_move_mount_old(struct path
*path
, const char *old_name
)
2747 struct path old_path
;
2750 if (!old_name
|| !*old_name
)
2753 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2757 err
= do_move_mount(&old_path
, path
);
2758 path_put(&old_path
);
2763 * add a mount into a namespace's mount tree
2765 static int do_add_mount(struct mount
*newmnt
, struct mountpoint
*mp
,
2766 struct path
*path
, int mnt_flags
)
2768 struct mount
*parent
= real_mount(path
->mnt
);
2770 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2772 if (unlikely(!check_mnt(parent
))) {
2773 /* that's acceptable only for automounts done in private ns */
2774 if (!(mnt_flags
& MNT_SHRINKABLE
))
2776 /* ... and for those we'd better have mountpoint still alive */
2777 if (!parent
->mnt_ns
)
2781 /* Refuse the same filesystem on the same mount point */
2782 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2783 path
->mnt
->mnt_root
== path
->dentry
)
2786 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2789 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2790 return graft_tree(newmnt
, parent
, mp
);
2793 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
);
2796 * Create a new mount using a superblock configuration and request it
2797 * be added to the namespace tree.
2799 static int do_new_mount_fc(struct fs_context
*fc
, struct path
*mountpoint
,
2800 unsigned int mnt_flags
)
2802 struct vfsmount
*mnt
;
2803 struct mountpoint
*mp
;
2804 struct super_block
*sb
= fc
->root
->d_sb
;
2807 error
= security_sb_kern_mount(sb
);
2808 if (!error
&& mount_too_revealing(sb
, &mnt_flags
))
2811 if (unlikely(error
)) {
2816 up_write(&sb
->s_umount
);
2818 mnt
= vfs_create_mount(fc
);
2820 return PTR_ERR(mnt
);
2822 mnt_warn_timestamp_expiry(mountpoint
, mnt
);
2824 mp
= lock_mount(mountpoint
);
2829 error
= do_add_mount(real_mount(mnt
), mp
, mountpoint
, mnt_flags
);
2837 * create a new mount for userspace and request it to be added into the
2840 static int do_new_mount(struct path
*path
, const char *fstype
, int sb_flags
,
2841 int mnt_flags
, const char *name
, void *data
)
2843 struct file_system_type
*type
;
2844 struct fs_context
*fc
;
2845 const char *subtype
= NULL
;
2851 type
= get_fs_type(fstype
);
2855 if (type
->fs_flags
& FS_HAS_SUBTYPE
) {
2856 subtype
= strchr(fstype
, '.');
2860 put_filesystem(type
);
2866 fc
= fs_context_for_mount(type
, sb_flags
);
2867 put_filesystem(type
);
2872 err
= vfs_parse_fs_string(fc
, "subtype",
2873 subtype
, strlen(subtype
));
2875 err
= vfs_parse_fs_string(fc
, "source", name
, strlen(name
));
2877 err
= parse_monolithic_mount_data(fc
, data
);
2878 if (!err
&& !mount_capable(fc
))
2881 err
= vfs_get_tree(fc
);
2883 err
= do_new_mount_fc(fc
, path
, mnt_flags
);
2889 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2891 struct dentry
*dentry
= path
->dentry
;
2892 struct mountpoint
*mp
;
2901 mnt
= real_mount(m
);
2902 /* The new mount record should have at least 2 refs to prevent it being
2903 * expired before we get a chance to add it
2905 BUG_ON(mnt_get_count(mnt
) < 2);
2907 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2908 m
->mnt_root
== dentry
) {
2914 * we don't want to use lock_mount() - in this case finding something
2915 * that overmounts our mountpoint to be means "quitely drop what we've
2916 * got", not "try to mount it on top".
2918 inode_lock(dentry
->d_inode
);
2920 if (unlikely(cant_mount(dentry
))) {
2922 goto discard_locked
;
2925 if (unlikely(__lookup_mnt(path
->mnt
, dentry
))) {
2928 goto discard_locked
;
2931 mp
= get_mountpoint(dentry
);
2934 goto discard_locked
;
2937 err
= do_add_mount(mnt
, mp
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2946 inode_unlock(dentry
->d_inode
);
2948 /* remove m from any expiration list it may be on */
2949 if (!list_empty(&mnt
->mnt_expire
)) {
2951 list_del_init(&mnt
->mnt_expire
);
2960 * mnt_set_expiry - Put a mount on an expiration list
2961 * @mnt: The mount to list.
2962 * @expiry_list: The list to add the mount to.
2964 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2968 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2972 EXPORT_SYMBOL(mnt_set_expiry
);
2975 * process a list of expirable mountpoints with the intent of discarding any
2976 * mountpoints that aren't in use and haven't been touched since last we came
2979 void mark_mounts_for_expiry(struct list_head
*mounts
)
2981 struct mount
*mnt
, *next
;
2982 LIST_HEAD(graveyard
);
2984 if (list_empty(mounts
))
2990 /* extract from the expiration list every vfsmount that matches the
2991 * following criteria:
2992 * - only referenced by its parent vfsmount
2993 * - still marked for expiry (marked on the last call here; marks are
2994 * cleared by mntput())
2996 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2997 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2998 propagate_mount_busy(mnt
, 1))
3000 list_move(&mnt
->mnt_expire
, &graveyard
);
3002 while (!list_empty(&graveyard
)) {
3003 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
3004 touch_mnt_namespace(mnt
->mnt_ns
);
3005 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3007 unlock_mount_hash();
3011 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
3014 * Ripoff of 'select_parent()'
3016 * search the list of submounts for a given mountpoint, and move any
3017 * shrinkable submounts to the 'graveyard' list.
3019 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
3021 struct mount
*this_parent
= parent
;
3022 struct list_head
*next
;
3026 next
= this_parent
->mnt_mounts
.next
;
3028 while (next
!= &this_parent
->mnt_mounts
) {
3029 struct list_head
*tmp
= next
;
3030 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
3033 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
3036 * Descend a level if the d_mounts list is non-empty.
3038 if (!list_empty(&mnt
->mnt_mounts
)) {
3043 if (!propagate_mount_busy(mnt
, 1)) {
3044 list_move_tail(&mnt
->mnt_expire
, graveyard
);
3049 * All done at this level ... ascend and resume the search
3051 if (this_parent
!= parent
) {
3052 next
= this_parent
->mnt_child
.next
;
3053 this_parent
= this_parent
->mnt_parent
;
3060 * process a list of expirable mountpoints with the intent of discarding any
3061 * submounts of a specific parent mountpoint
3063 * mount_lock must be held for write
3065 static void shrink_submounts(struct mount
*mnt
)
3067 LIST_HEAD(graveyard
);
3070 /* extract submounts of 'mountpoint' from the expiration list */
3071 while (select_submounts(mnt
, &graveyard
)) {
3072 while (!list_empty(&graveyard
)) {
3073 m
= list_first_entry(&graveyard
, struct mount
,
3075 touch_mnt_namespace(m
->mnt_ns
);
3076 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3081 static void *copy_mount_options(const void __user
* data
)
3084 unsigned left
, offset
;
3089 copy
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3091 return ERR_PTR(-ENOMEM
);
3093 left
= copy_from_user(copy
, data
, PAGE_SIZE
);
3096 * Not all architectures have an exact copy_from_user(). Resort to
3099 offset
= PAGE_SIZE
- left
;
3102 if (get_user(c
, (const char __user
*)data
+ offset
))
3109 if (left
== PAGE_SIZE
) {
3111 return ERR_PTR(-EFAULT
);
3117 static char *copy_mount_string(const void __user
*data
)
3119 return data
? strndup_user(data
, PATH_MAX
) : NULL
;
3123 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3124 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3126 * data is a (void *) that can point to any structure up to
3127 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3128 * information (or be NULL).
3130 * Pre-0.97 versions of mount() didn't have a flags word.
3131 * When the flags word was introduced its top half was required
3132 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3133 * Therefore, if this magic number is present, it carries no information
3134 * and must be discarded.
3136 int path_mount(const char *dev_name
, struct path
*path
,
3137 const char *type_page
, unsigned long flags
, void *data_page
)
3139 unsigned int mnt_flags
= 0, sb_flags
;
3143 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
3144 flags
&= ~MS_MGC_MSK
;
3146 /* Basic sanity checks */
3148 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
3150 if (flags
& MS_NOUSER
)
3153 ret
= security_sb_mount(dev_name
, path
, type_page
, flags
, data_page
);
3158 if ((flags
& SB_MANDLOCK
) && !may_mandlock())
3161 /* Default to relatime unless overriden */
3162 if (!(flags
& MS_NOATIME
))
3163 mnt_flags
|= MNT_RELATIME
;
3165 /* Separate the per-mountpoint flags */
3166 if (flags
& MS_NOSUID
)
3167 mnt_flags
|= MNT_NOSUID
;
3168 if (flags
& MS_NODEV
)
3169 mnt_flags
|= MNT_NODEV
;
3170 if (flags
& MS_NOEXEC
)
3171 mnt_flags
|= MNT_NOEXEC
;
3172 if (flags
& MS_NOATIME
)
3173 mnt_flags
|= MNT_NOATIME
;
3174 if (flags
& MS_NODIRATIME
)
3175 mnt_flags
|= MNT_NODIRATIME
;
3176 if (flags
& MS_STRICTATIME
)
3177 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
3178 if (flags
& MS_RDONLY
)
3179 mnt_flags
|= MNT_READONLY
;
3180 if (flags
& MS_NOSYMFOLLOW
)
3181 mnt_flags
|= MNT_NOSYMFOLLOW
;
3183 /* The default atime for remount is preservation */
3184 if ((flags
& MS_REMOUNT
) &&
3185 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
3186 MS_STRICTATIME
)) == 0)) {
3187 mnt_flags
&= ~MNT_ATIME_MASK
;
3188 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_ATIME_MASK
;
3191 sb_flags
= flags
& (SB_RDONLY
|
3200 if ((flags
& (MS_REMOUNT
| MS_BIND
)) == (MS_REMOUNT
| MS_BIND
))
3201 return do_reconfigure_mnt(path
, mnt_flags
);
3202 if (flags
& MS_REMOUNT
)
3203 return do_remount(path
, flags
, sb_flags
, mnt_flags
, data_page
);
3204 if (flags
& MS_BIND
)
3205 return do_loopback(path
, dev_name
, flags
& MS_REC
);
3206 if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
3207 return do_change_type(path
, flags
);
3208 if (flags
& MS_MOVE
)
3209 return do_move_mount_old(path
, dev_name
);
3211 return do_new_mount(path
, type_page
, sb_flags
, mnt_flags
, dev_name
,
3215 long do_mount(const char *dev_name
, const char __user
*dir_name
,
3216 const char *type_page
, unsigned long flags
, void *data_page
)
3221 ret
= user_path_at(AT_FDCWD
, dir_name
, LOOKUP_FOLLOW
, &path
);
3224 ret
= path_mount(dev_name
, &path
, type_page
, flags
, data_page
);
3229 static struct ucounts
*inc_mnt_namespaces(struct user_namespace
*ns
)
3231 return inc_ucount(ns
, current_euid(), UCOUNT_MNT_NAMESPACES
);
3234 static void dec_mnt_namespaces(struct ucounts
*ucounts
)
3236 dec_ucount(ucounts
, UCOUNT_MNT_NAMESPACES
);
3239 static void free_mnt_ns(struct mnt_namespace
*ns
)
3241 if (!is_anon_ns(ns
))
3242 ns_free_inum(&ns
->ns
);
3243 dec_mnt_namespaces(ns
->ucounts
);
3244 put_user_ns(ns
->user_ns
);
3249 * Assign a sequence number so we can detect when we attempt to bind
3250 * mount a reference to an older mount namespace into the current
3251 * mount namespace, preventing reference counting loops. A 64bit
3252 * number incrementing at 10Ghz will take 12,427 years to wrap which
3253 * is effectively never, so we can ignore the possibility.
3255 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
3257 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
, bool anon
)
3259 struct mnt_namespace
*new_ns
;
3260 struct ucounts
*ucounts
;
3263 ucounts
= inc_mnt_namespaces(user_ns
);
3265 return ERR_PTR(-ENOSPC
);
3267 new_ns
= kzalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
3269 dec_mnt_namespaces(ucounts
);
3270 return ERR_PTR(-ENOMEM
);
3273 ret
= ns_alloc_inum(&new_ns
->ns
);
3276 dec_mnt_namespaces(ucounts
);
3277 return ERR_PTR(ret
);
3280 new_ns
->ns
.ops
= &mntns_operations
;
3282 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
3283 refcount_set(&new_ns
->ns
.count
, 1);
3284 INIT_LIST_HEAD(&new_ns
->list
);
3285 init_waitqueue_head(&new_ns
->poll
);
3286 spin_lock_init(&new_ns
->ns_lock
);
3287 new_ns
->user_ns
= get_user_ns(user_ns
);
3288 new_ns
->ucounts
= ucounts
;
3293 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
3294 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
3296 struct mnt_namespace
*new_ns
;
3297 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
3298 struct mount
*p
, *q
;
3305 if (likely(!(flags
& CLONE_NEWNS
))) {
3312 new_ns
= alloc_mnt_ns(user_ns
, false);
3317 /* First pass: copy the tree topology */
3318 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
3319 if (user_ns
!= ns
->user_ns
)
3320 copy_flags
|= CL_SHARED_TO_SLAVE
;
3321 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
3324 free_mnt_ns(new_ns
);
3325 return ERR_CAST(new);
3327 if (user_ns
!= ns
->user_ns
) {
3330 unlock_mount_hash();
3333 list_add_tail(&new_ns
->list
, &new->mnt_list
);
3336 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3337 * as belonging to new namespace. We have already acquired a private
3338 * fs_struct, so tsk->fs->lock is not needed.
3346 if (&p
->mnt
== new_fs
->root
.mnt
) {
3347 new_fs
->root
.mnt
= mntget(&q
->mnt
);
3350 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
3351 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
3355 p
= next_mnt(p
, old
);
3356 q
= next_mnt(q
, new);
3359 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
3360 p
= next_mnt(p
, old
);
3372 struct dentry
*mount_subtree(struct vfsmount
*m
, const char *name
)
3374 struct mount
*mnt
= real_mount(m
);
3375 struct mnt_namespace
*ns
;
3376 struct super_block
*s
;
3380 ns
= alloc_mnt_ns(&init_user_ns
, true);
3383 return ERR_CAST(ns
);
3388 list_add(&mnt
->mnt_list
, &ns
->list
);
3390 err
= vfs_path_lookup(m
->mnt_root
, m
,
3391 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
3396 return ERR_PTR(err
);
3398 /* trade a vfsmount reference for active sb one */
3399 s
= path
.mnt
->mnt_sb
;
3400 atomic_inc(&s
->s_active
);
3402 /* lock the sucker */
3403 down_write(&s
->s_umount
);
3404 /* ... and return the root of (sub)tree on it */
3407 EXPORT_SYMBOL(mount_subtree
);
3409 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
3410 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3417 kernel_type
= copy_mount_string(type
);
3418 ret
= PTR_ERR(kernel_type
);
3419 if (IS_ERR(kernel_type
))
3422 kernel_dev
= copy_mount_string(dev_name
);
3423 ret
= PTR_ERR(kernel_dev
);
3424 if (IS_ERR(kernel_dev
))
3427 options
= copy_mount_options(data
);
3428 ret
= PTR_ERR(options
);
3429 if (IS_ERR(options
))
3432 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
, options
);
3444 * Create a kernel mount representation for a new, prepared superblock
3445 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3447 SYSCALL_DEFINE3(fsmount
, int, fs_fd
, unsigned int, flags
,
3448 unsigned int, attr_flags
)
3450 struct mnt_namespace
*ns
;
3451 struct fs_context
*fc
;
3453 struct path newmount
;
3456 unsigned int mnt_flags
= 0;
3462 if ((flags
& ~(FSMOUNT_CLOEXEC
)) != 0)
3465 if (attr_flags
& ~(MOUNT_ATTR_RDONLY
|
3470 MOUNT_ATTR_NODIRATIME
))
3473 if (attr_flags
& MOUNT_ATTR_RDONLY
)
3474 mnt_flags
|= MNT_READONLY
;
3475 if (attr_flags
& MOUNT_ATTR_NOSUID
)
3476 mnt_flags
|= MNT_NOSUID
;
3477 if (attr_flags
& MOUNT_ATTR_NODEV
)
3478 mnt_flags
|= MNT_NODEV
;
3479 if (attr_flags
& MOUNT_ATTR_NOEXEC
)
3480 mnt_flags
|= MNT_NOEXEC
;
3481 if (attr_flags
& MOUNT_ATTR_NODIRATIME
)
3482 mnt_flags
|= MNT_NODIRATIME
;
3484 switch (attr_flags
& MOUNT_ATTR__ATIME
) {
3485 case MOUNT_ATTR_STRICTATIME
:
3487 case MOUNT_ATTR_NOATIME
:
3488 mnt_flags
|= MNT_NOATIME
;
3490 case MOUNT_ATTR_RELATIME
:
3491 mnt_flags
|= MNT_RELATIME
;
3502 if (f
.file
->f_op
!= &fscontext_fops
)
3505 fc
= f
.file
->private_data
;
3507 ret
= mutex_lock_interruptible(&fc
->uapi_mutex
);
3511 /* There must be a valid superblock or we can't mount it */
3517 if (mount_too_revealing(fc
->root
->d_sb
, &mnt_flags
)) {
3518 pr_warn("VFS: Mount too revealing\n");
3523 if (fc
->phase
!= FS_CONTEXT_AWAITING_MOUNT
)
3527 if ((fc
->sb_flags
& SB_MANDLOCK
) && !may_mandlock())
3530 newmount
.mnt
= vfs_create_mount(fc
);
3531 if (IS_ERR(newmount
.mnt
)) {
3532 ret
= PTR_ERR(newmount
.mnt
);
3535 newmount
.dentry
= dget(fc
->root
);
3536 newmount
.mnt
->mnt_flags
= mnt_flags
;
3538 /* We've done the mount bit - now move the file context into more or
3539 * less the same state as if we'd done an fspick(). We don't want to
3540 * do any memory allocation or anything like that at this point as we
3541 * don't want to have to handle any errors incurred.
3543 vfs_clean_context(fc
);
3545 ns
= alloc_mnt_ns(current
->nsproxy
->mnt_ns
->user_ns
, true);
3550 mnt
= real_mount(newmount
.mnt
);
3554 list_add(&mnt
->mnt_list
, &ns
->list
);
3555 mntget(newmount
.mnt
);
3557 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3558 * it, not just simply put it.
3560 file
= dentry_open(&newmount
, O_PATH
, fc
->cred
);
3562 dissolve_on_fput(newmount
.mnt
);
3563 ret
= PTR_ERR(file
);
3566 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
3568 ret
= get_unused_fd_flags((flags
& FSMOUNT_CLOEXEC
) ? O_CLOEXEC
: 0);
3570 fd_install(ret
, file
);
3575 path_put(&newmount
);
3577 mutex_unlock(&fc
->uapi_mutex
);
3584 * Move a mount from one place to another. In combination with
3585 * fsopen()/fsmount() this is used to install a new mount and in combination
3586 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3589 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3591 SYSCALL_DEFINE5(move_mount
,
3592 int, from_dfd
, const char __user
*, from_pathname
,
3593 int, to_dfd
, const char __user
*, to_pathname
,
3594 unsigned int, flags
)
3596 struct path from_path
, to_path
;
3597 unsigned int lflags
;
3603 if (flags
& ~MOVE_MOUNT__MASK
)
3606 /* If someone gives a pathname, they aren't permitted to move
3607 * from an fd that requires unmount as we can't get at the flag
3608 * to clear it afterwards.
3611 if (flags
& MOVE_MOUNT_F_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3612 if (flags
& MOVE_MOUNT_F_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3613 if (flags
& MOVE_MOUNT_F_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3615 ret
= user_path_at(from_dfd
, from_pathname
, lflags
, &from_path
);
3620 if (flags
& MOVE_MOUNT_T_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3621 if (flags
& MOVE_MOUNT_T_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3622 if (flags
& MOVE_MOUNT_T_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3624 ret
= user_path_at(to_dfd
, to_pathname
, lflags
, &to_path
);
3628 ret
= security_move_mount(&from_path
, &to_path
);
3632 ret
= do_move_mount(&from_path
, &to_path
);
3637 path_put(&from_path
);
3642 * Return true if path is reachable from root
3644 * namespace_sem or mount_lock is held
3646 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
3647 const struct path
*root
)
3649 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
3650 dentry
= mnt
->mnt_mountpoint
;
3651 mnt
= mnt
->mnt_parent
;
3653 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
3656 bool path_is_under(const struct path
*path1
, const struct path
*path2
)
3659 read_seqlock_excl(&mount_lock
);
3660 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
3661 read_sequnlock_excl(&mount_lock
);
3664 EXPORT_SYMBOL(path_is_under
);
3667 * pivot_root Semantics:
3668 * Moves the root file system of the current process to the directory put_old,
3669 * makes new_root as the new root file system of the current process, and sets
3670 * root/cwd of all processes which had them on the current root to new_root.
3673 * The new_root and put_old must be directories, and must not be on the
3674 * same file system as the current process root. The put_old must be
3675 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3676 * pointed to by put_old must yield the same directory as new_root. No other
3677 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3679 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3680 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3681 * in this situation.
3684 * - we don't move root/cwd if they are not at the root (reason: if something
3685 * cared enough to change them, it's probably wrong to force them elsewhere)
3686 * - it's okay to pick a root that isn't the root of a file system, e.g.
3687 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3688 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3691 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3692 const char __user
*, put_old
)
3694 struct path
new, old
, root
;
3695 struct mount
*new_mnt
, *root_mnt
, *old_mnt
, *root_parent
, *ex_parent
;
3696 struct mountpoint
*old_mp
, *root_mp
;
3702 error
= user_path_at(AT_FDCWD
, new_root
,
3703 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &new);
3707 error
= user_path_at(AT_FDCWD
, put_old
,
3708 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old
);
3712 error
= security_sb_pivotroot(&old
, &new);
3716 get_fs_root(current
->fs
, &root
);
3717 old_mp
= lock_mount(&old
);
3718 error
= PTR_ERR(old_mp
);
3723 new_mnt
= real_mount(new.mnt
);
3724 root_mnt
= real_mount(root
.mnt
);
3725 old_mnt
= real_mount(old
.mnt
);
3726 ex_parent
= new_mnt
->mnt_parent
;
3727 root_parent
= root_mnt
->mnt_parent
;
3728 if (IS_MNT_SHARED(old_mnt
) ||
3729 IS_MNT_SHARED(ex_parent
) ||
3730 IS_MNT_SHARED(root_parent
))
3732 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3734 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3737 if (d_unlinked(new.dentry
))
3740 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3741 goto out4
; /* loop, on the same file system */
3743 if (root
.mnt
->mnt_root
!= root
.dentry
)
3744 goto out4
; /* not a mountpoint */
3745 if (!mnt_has_parent(root_mnt
))
3746 goto out4
; /* not attached */
3747 if (new.mnt
->mnt_root
!= new.dentry
)
3748 goto out4
; /* not a mountpoint */
3749 if (!mnt_has_parent(new_mnt
))
3750 goto out4
; /* not attached */
3751 /* make sure we can reach put_old from new_root */
3752 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3754 /* make certain new is below the root */
3755 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3758 umount_mnt(new_mnt
);
3759 root_mp
= unhash_mnt(root_mnt
); /* we'll need its mountpoint */
3760 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3761 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3762 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3764 /* mount old root on put_old */
3765 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3766 /* mount new_root on / */
3767 attach_mnt(new_mnt
, root_parent
, root_mp
);
3768 mnt_add_count(root_parent
, -1);
3769 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3770 /* A moved mount should not expire automatically */
3771 list_del_init(&new_mnt
->mnt_expire
);
3772 put_mountpoint(root_mp
);
3773 unlock_mount_hash();
3774 chroot_fs_refs(&root
, &new);
3777 unlock_mount(old_mp
);
3779 mntput_no_expire(ex_parent
);
3790 static void __init
init_mount_tree(void)
3792 struct vfsmount
*mnt
;
3794 struct mnt_namespace
*ns
;
3797 mnt
= vfs_kern_mount(&rootfs_fs_type
, 0, "rootfs", NULL
);
3799 panic("Can't create rootfs");
3801 ns
= alloc_mnt_ns(&init_user_ns
, false);
3803 panic("Can't allocate initial namespace");
3804 m
= real_mount(mnt
);
3808 list_add(&m
->mnt_list
, &ns
->list
);
3809 init_task
.nsproxy
->mnt_ns
= ns
;
3813 root
.dentry
= mnt
->mnt_root
;
3814 mnt
->mnt_flags
|= MNT_LOCKED
;
3816 set_fs_pwd(current
->fs
, &root
);
3817 set_fs_root(current
->fs
, &root
);
3820 void __init
mnt_init(void)
3824 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
3825 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3827 mount_hashtable
= alloc_large_system_hash("Mount-cache",
3828 sizeof(struct hlist_head
),
3831 &m_hash_shift
, &m_hash_mask
, 0, 0);
3832 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
3833 sizeof(struct hlist_head
),
3836 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
3838 if (!mount_hashtable
|| !mountpoint_hashtable
)
3839 panic("Failed to allocate mount hash table\n");
3845 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
3847 fs_kobj
= kobject_create_and_add("fs", NULL
);
3849 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
3855 void put_mnt_ns(struct mnt_namespace
*ns
)
3857 if (!refcount_dec_and_test(&ns
->ns
.count
))
3859 drop_collected_mounts(&ns
->root
->mnt
);
3863 struct vfsmount
*kern_mount(struct file_system_type
*type
)
3865 struct vfsmount
*mnt
;
3866 mnt
= vfs_kern_mount(type
, SB_KERNMOUNT
, type
->name
, NULL
);
3869 * it is a longterm mount, don't release mnt until
3870 * we unmount before file sys is unregistered
3872 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
3876 EXPORT_SYMBOL_GPL(kern_mount
);
3878 void kern_unmount(struct vfsmount
*mnt
)
3880 /* release long term mount so mount point can be released */
3881 if (!IS_ERR_OR_NULL(mnt
)) {
3882 real_mount(mnt
)->mnt_ns
= NULL
;
3883 synchronize_rcu(); /* yecchhh... */
3887 EXPORT_SYMBOL(kern_unmount
);
3889 void kern_unmount_array(struct vfsmount
*mnt
[], unsigned int num
)
3893 for (i
= 0; i
< num
; i
++)
3895 real_mount(mnt
[i
])->mnt_ns
= NULL
;
3896 synchronize_rcu_expedited();
3897 for (i
= 0; i
< num
; i
++)
3900 EXPORT_SYMBOL(kern_unmount_array
);
3902 bool our_mnt(struct vfsmount
*mnt
)
3904 return check_mnt(real_mount(mnt
));
3907 bool current_chrooted(void)
3909 /* Does the current process have a non-standard root */
3910 struct path ns_root
;
3911 struct path fs_root
;
3914 /* Find the namespace root */
3915 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
3916 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
3918 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
3921 get_fs_root(current
->fs
, &fs_root
);
3923 chrooted
= !path_equal(&fs_root
, &ns_root
);
3931 static bool mnt_already_visible(struct mnt_namespace
*ns
,
3932 const struct super_block
*sb
,
3935 int new_flags
= *new_mnt_flags
;
3937 bool visible
= false;
3939 down_read(&namespace_sem
);
3941 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
3942 struct mount
*child
;
3945 if (mnt_is_cursor(mnt
))
3948 if (mnt
->mnt
.mnt_sb
->s_type
!= sb
->s_type
)
3951 /* This mount is not fully visible if it's root directory
3952 * is not the root directory of the filesystem.
3954 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
3957 /* A local view of the mount flags */
3958 mnt_flags
= mnt
->mnt
.mnt_flags
;
3960 /* Don't miss readonly hidden in the superblock flags */
3961 if (sb_rdonly(mnt
->mnt
.mnt_sb
))
3962 mnt_flags
|= MNT_LOCK_READONLY
;
3964 /* Verify the mount flags are equal to or more permissive
3965 * than the proposed new mount.
3967 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
3968 !(new_flags
& MNT_READONLY
))
3970 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
3971 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
3974 /* This mount is not fully visible if there are any
3975 * locked child mounts that cover anything except for
3976 * empty directories.
3978 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
3979 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
3980 /* Only worry about locked mounts */
3981 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
3983 /* Is the directory permanetly empty? */
3984 if (!is_empty_dir_inode(inode
))
3987 /* Preserve the locked attributes */
3988 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
3996 up_read(&namespace_sem
);
4000 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
)
4002 const unsigned long required_iflags
= SB_I_NOEXEC
| SB_I_NODEV
;
4003 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
4004 unsigned long s_iflags
;
4006 if (ns
->user_ns
== &init_user_ns
)
4009 /* Can this filesystem be too revealing? */
4010 s_iflags
= sb
->s_iflags
;
4011 if (!(s_iflags
& SB_I_USERNS_VISIBLE
))
4014 if ((s_iflags
& required_iflags
) != required_iflags
) {
4015 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4020 return !mnt_already_visible(ns
, sb
, new_mnt_flags
);
4023 bool mnt_may_suid(struct vfsmount
*mnt
)
4026 * Foreign mounts (accessed via fchdir or through /proc
4027 * symlinks) are always treated as if they are nosuid. This
4028 * prevents namespaces from trusting potentially unsafe
4029 * suid/sgid bits, file caps, or security labels that originate
4030 * in other namespaces.
4032 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
4033 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
4036 static struct ns_common
*mntns_get(struct task_struct
*task
)
4038 struct ns_common
*ns
= NULL
;
4039 struct nsproxy
*nsproxy
;
4042 nsproxy
= task
->nsproxy
;
4044 ns
= &nsproxy
->mnt_ns
->ns
;
4045 get_mnt_ns(to_mnt_ns(ns
));
4052 static void mntns_put(struct ns_common
*ns
)
4054 put_mnt_ns(to_mnt_ns(ns
));
4057 static int mntns_install(struct nsset
*nsset
, struct ns_common
*ns
)
4059 struct nsproxy
*nsproxy
= nsset
->nsproxy
;
4060 struct fs_struct
*fs
= nsset
->fs
;
4061 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
), *old_mnt_ns
;
4062 struct user_namespace
*user_ns
= nsset
->cred
->user_ns
;
4066 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
4067 !ns_capable(user_ns
, CAP_SYS_CHROOT
) ||
4068 !ns_capable(user_ns
, CAP_SYS_ADMIN
))
4071 if (is_anon_ns(mnt_ns
))
4078 old_mnt_ns
= nsproxy
->mnt_ns
;
4079 nsproxy
->mnt_ns
= mnt_ns
;
4082 err
= vfs_path_lookup(mnt_ns
->root
->mnt
.mnt_root
, &mnt_ns
->root
->mnt
,
4083 "/", LOOKUP_DOWN
, &root
);
4085 /* revert to old namespace */
4086 nsproxy
->mnt_ns
= old_mnt_ns
;
4091 put_mnt_ns(old_mnt_ns
);
4093 /* Update the pwd and root */
4094 set_fs_pwd(fs
, &root
);
4095 set_fs_root(fs
, &root
);
4101 static struct user_namespace
*mntns_owner(struct ns_common
*ns
)
4103 return to_mnt_ns(ns
)->user_ns
;
4106 const struct proc_ns_operations mntns_operations
= {
4108 .type
= CLONE_NEWNS
,
4111 .install
= mntns_install
,
4112 .owner
= mntns_owner
,