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/proc_fs.h>
29 #include <linux/task_work.h>
30 #include <linux/sched/task.h>
31 #include <uapi/linux/mount.h>
32 #include <linux/fs_context.h>
33 #include <linux/shmem_fs.h>
34 #include <linux/mnt_idmapping.h>
39 /* Maximum number of mounts in a mount namespace */
40 static unsigned int sysctl_mount_max __read_mostly
= 100000;
42 static unsigned int m_hash_mask __read_mostly
;
43 static unsigned int m_hash_shift __read_mostly
;
44 static unsigned int mp_hash_mask __read_mostly
;
45 static unsigned int mp_hash_shift __read_mostly
;
47 static __initdata
unsigned long mhash_entries
;
48 static int __init
set_mhash_entries(char *str
)
52 mhash_entries
= simple_strtoul(str
, &str
, 0);
55 __setup("mhash_entries=", set_mhash_entries
);
57 static __initdata
unsigned long mphash_entries
;
58 static int __init
set_mphash_entries(char *str
)
62 mphash_entries
= simple_strtoul(str
, &str
, 0);
65 __setup("mphash_entries=", set_mphash_entries
);
68 static DEFINE_IDA(mnt_id_ida
);
69 static DEFINE_IDA(mnt_group_ida
);
71 static struct hlist_head
*mount_hashtable __read_mostly
;
72 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
73 static struct kmem_cache
*mnt_cache __read_mostly
;
74 static DECLARE_RWSEM(namespace_sem
);
75 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
76 static LIST_HEAD(ex_mountpoints
); /* protected by namespace_sem */
79 unsigned int attr_set
;
80 unsigned int attr_clr
;
81 unsigned int propagation
;
82 unsigned int lookup_flags
;
84 struct user_namespace
*mnt_userns
;
85 struct mnt_idmap
*mnt_idmap
;
89 struct kobject
*fs_kobj
;
90 EXPORT_SYMBOL_GPL(fs_kobj
);
93 * vfsmount lock may be taken for read to prevent changes to the
94 * vfsmount hash, ie. during mountpoint lookups or walking back
97 * It should be taken for write in all cases where the vfsmount
98 * tree or hash is modified or when a vfsmount structure is modified.
100 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
102 static inline void lock_mount_hash(void)
104 write_seqlock(&mount_lock
);
107 static inline void unlock_mount_hash(void)
109 write_sequnlock(&mount_lock
);
112 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
114 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
115 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
116 tmp
= tmp
+ (tmp
>> m_hash_shift
);
117 return &mount_hashtable
[tmp
& m_hash_mask
];
120 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
122 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
123 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
124 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
127 static int mnt_alloc_id(struct mount
*mnt
)
129 int res
= ida_alloc(&mnt_id_ida
, GFP_KERNEL
);
137 static void mnt_free_id(struct mount
*mnt
)
139 ida_free(&mnt_id_ida
, mnt
->mnt_id
);
143 * Allocate a new peer group ID
145 static int mnt_alloc_group_id(struct mount
*mnt
)
147 int res
= ida_alloc_min(&mnt_group_ida
, 1, GFP_KERNEL
);
151 mnt
->mnt_group_id
= res
;
156 * Release a peer group ID
158 void mnt_release_group_id(struct mount
*mnt
)
160 ida_free(&mnt_group_ida
, mnt
->mnt_group_id
);
161 mnt
->mnt_group_id
= 0;
165 * vfsmount lock must be held for read
167 static inline void mnt_add_count(struct mount
*mnt
, int n
)
170 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
179 * vfsmount lock must be held for write
181 int mnt_get_count(struct mount
*mnt
)
187 for_each_possible_cpu(cpu
) {
188 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
193 return mnt
->mnt_count
;
197 static struct mount
*alloc_vfsmnt(const char *name
)
199 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
203 err
= mnt_alloc_id(mnt
);
208 mnt
->mnt_devname
= kstrdup_const(name
,
210 if (!mnt
->mnt_devname
)
215 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
217 goto out_free_devname
;
219 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
222 mnt
->mnt_writers
= 0;
225 INIT_HLIST_NODE(&mnt
->mnt_hash
);
226 INIT_LIST_HEAD(&mnt
->mnt_child
);
227 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
228 INIT_LIST_HEAD(&mnt
->mnt_list
);
229 INIT_LIST_HEAD(&mnt
->mnt_expire
);
230 INIT_LIST_HEAD(&mnt
->mnt_share
);
231 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
232 INIT_LIST_HEAD(&mnt
->mnt_slave
);
233 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
234 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
235 INIT_HLIST_HEAD(&mnt
->mnt_stuck_children
);
236 mnt
->mnt
.mnt_idmap
= &nop_mnt_idmap
;
242 kfree_const(mnt
->mnt_devname
);
247 kmem_cache_free(mnt_cache
, mnt
);
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
270 bool __mnt_is_readonly(struct vfsmount
*mnt
)
272 return (mnt
->mnt_flags
& MNT_READONLY
) || sb_rdonly(mnt
->mnt_sb
);
274 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
276 static inline void mnt_inc_writers(struct mount
*mnt
)
279 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
285 static inline void mnt_dec_writers(struct mount
*mnt
)
288 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
294 static unsigned int mnt_get_writers(struct mount
*mnt
)
297 unsigned int count
= 0;
300 for_each_possible_cpu(cpu
) {
301 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
306 return mnt
->mnt_writers
;
310 static int mnt_is_readonly(struct vfsmount
*mnt
)
312 if (READ_ONCE(mnt
->mnt_sb
->s_readonly_remount
))
315 * The barrier pairs with the barrier in sb_start_ro_state_change()
316 * making sure if we don't see s_readonly_remount set yet, we also will
317 * not see any superblock / mount flag changes done by remount.
318 * It also pairs with the barrier in sb_end_ro_state_change()
319 * assuring that if we see s_readonly_remount already cleared, we will
320 * see the values of superblock / mount flags updated by remount.
323 return __mnt_is_readonly(mnt
);
327 * Most r/o & frozen checks on a fs are for operations that take discrete
328 * amounts of time, like a write() or unlink(). We must keep track of when
329 * those operations start (for permission checks) and when they end, so that we
330 * can determine when writes are able to occur to a filesystem.
333 * __mnt_want_write - get write access to a mount without freeze protection
334 * @m: the mount on which to take a write
336 * This tells the low-level filesystem that a write is about to be performed to
337 * it, and makes sure that writes are allowed (mnt it read-write) before
338 * returning success. This operation does not protect against filesystem being
339 * frozen. When the write operation is finished, __mnt_drop_write() must be
340 * called. This is effectively a refcount.
342 int __mnt_want_write(struct vfsmount
*m
)
344 struct mount
*mnt
= real_mount(m
);
348 mnt_inc_writers(mnt
);
350 * The store to mnt_inc_writers must be visible before we pass
351 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 * incremented count after it has set MNT_WRITE_HOLD.
355 might_lock(&mount_lock
.lock
);
356 while (READ_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
) {
357 if (!IS_ENABLED(CONFIG_PREEMPT_RT
)) {
361 * This prevents priority inversion, if the task
362 * setting MNT_WRITE_HOLD got preempted on a remote
363 * CPU, and it prevents life lock if the task setting
364 * MNT_WRITE_HOLD has a lower priority and is bound to
365 * the same CPU as the task that is spinning here.
374 * The barrier pairs with the barrier sb_start_ro_state_change() making
375 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
376 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
377 * mnt_is_readonly() and bail in case we are racing with remount
381 if (mnt_is_readonly(m
)) {
382 mnt_dec_writers(mnt
);
391 * mnt_want_write - get write access to a mount
392 * @m: the mount on which to take a write
394 * This tells the low-level filesystem that a write is about to be performed to
395 * it, and makes sure that writes are allowed (mount is read-write, filesystem
396 * is not frozen) before returning success. When the write operation is
397 * finished, mnt_drop_write() must be called. This is effectively a refcount.
399 int mnt_want_write(struct vfsmount
*m
)
403 sb_start_write(m
->mnt_sb
);
404 ret
= __mnt_want_write(m
);
406 sb_end_write(m
->mnt_sb
);
409 EXPORT_SYMBOL_GPL(mnt_want_write
);
412 * __mnt_want_write_file - get write access to a file's mount
413 * @file: the file who's mount on which to take a write
415 * This is like __mnt_want_write, but if the file is already open for writing it
416 * skips incrementing mnt_writers (since the open file already has a reference)
417 * and instead only does the check for emergency r/o remounts. This must be
418 * paired with __mnt_drop_write_file.
420 int __mnt_want_write_file(struct file
*file
)
422 if (file
->f_mode
& FMODE_WRITER
) {
424 * Superblock may have become readonly while there are still
425 * writable fd's, e.g. due to a fs error with errors=remount-ro
427 if (__mnt_is_readonly(file
->f_path
.mnt
))
431 return __mnt_want_write(file
->f_path
.mnt
);
435 * mnt_want_write_file - get write access to a file's mount
436 * @file: the file who's mount on which to take a write
438 * This is like mnt_want_write, but if the file is already open for writing it
439 * skips incrementing mnt_writers (since the open file already has a reference)
440 * and instead only does the freeze protection and the check for emergency r/o
441 * remounts. This must be paired with mnt_drop_write_file.
443 int mnt_want_write_file(struct file
*file
)
447 sb_start_write(file_inode(file
)->i_sb
);
448 ret
= __mnt_want_write_file(file
);
450 sb_end_write(file_inode(file
)->i_sb
);
453 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
456 * __mnt_drop_write - give up write access to a mount
457 * @mnt: the mount on which to give up write access
459 * Tells the low-level filesystem that we are done
460 * performing writes to it. Must be matched with
461 * __mnt_want_write() call above.
463 void __mnt_drop_write(struct vfsmount
*mnt
)
466 mnt_dec_writers(real_mount(mnt
));
471 * mnt_drop_write - give up write access to a mount
472 * @mnt: the mount on which to give up write access
474 * Tells the low-level filesystem that we are done performing writes to it and
475 * also allows filesystem to be frozen again. Must be matched with
476 * mnt_want_write() call above.
478 void mnt_drop_write(struct vfsmount
*mnt
)
480 __mnt_drop_write(mnt
);
481 sb_end_write(mnt
->mnt_sb
);
483 EXPORT_SYMBOL_GPL(mnt_drop_write
);
485 void __mnt_drop_write_file(struct file
*file
)
487 if (!(file
->f_mode
& FMODE_WRITER
))
488 __mnt_drop_write(file
->f_path
.mnt
);
491 void mnt_drop_write_file(struct file
*file
)
493 __mnt_drop_write_file(file
);
494 sb_end_write(file_inode(file
)->i_sb
);
496 EXPORT_SYMBOL(mnt_drop_write_file
);
499 * mnt_hold_writers - prevent write access to the given mount
500 * @mnt: mnt to prevent write access to
502 * Prevents write access to @mnt if there are no active writers for @mnt.
503 * This function needs to be called and return successfully before changing
504 * properties of @mnt that need to remain stable for callers with write access
507 * After this functions has been called successfully callers must pair it with
508 * a call to mnt_unhold_writers() in order to stop preventing write access to
511 * Context: This function expects lock_mount_hash() to be held serializing
512 * setting MNT_WRITE_HOLD.
513 * Return: On success 0 is returned.
514 * On error, -EBUSY is returned.
516 static inline int mnt_hold_writers(struct mount
*mnt
)
518 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
520 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
521 * should be visible before we do.
526 * With writers on hold, if this value is zero, then there are
527 * definitely no active writers (although held writers may subsequently
528 * increment the count, they'll have to wait, and decrement it after
529 * seeing MNT_READONLY).
531 * It is OK to have counter incremented on one CPU and decremented on
532 * another: the sum will add up correctly. The danger would be when we
533 * sum up each counter, if we read a counter before it is incremented,
534 * but then read another CPU's count which it has been subsequently
535 * decremented from -- we would see more decrements than we should.
536 * MNT_WRITE_HOLD protects against this scenario, because
537 * mnt_want_write first increments count, then smp_mb, then spins on
538 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
539 * we're counting up here.
541 if (mnt_get_writers(mnt
) > 0)
548 * mnt_unhold_writers - stop preventing write access to the given mount
549 * @mnt: mnt to stop preventing write access to
551 * Stop preventing write access to @mnt allowing callers to gain write access
554 * This function can only be called after a successful call to
555 * mnt_hold_writers().
557 * Context: This function expects lock_mount_hash() to be held.
559 static inline void mnt_unhold_writers(struct mount
*mnt
)
562 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
563 * that become unheld will see MNT_READONLY.
566 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
569 static int mnt_make_readonly(struct mount
*mnt
)
573 ret
= mnt_hold_writers(mnt
);
575 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
576 mnt_unhold_writers(mnt
);
580 int sb_prepare_remount_readonly(struct super_block
*sb
)
585 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
586 if (atomic_long_read(&sb
->s_remove_count
))
590 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
591 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
592 err
= mnt_hold_writers(mnt
);
597 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
601 sb_start_ro_state_change(sb
);
602 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
603 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
604 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
611 static void free_vfsmnt(struct mount
*mnt
)
613 mnt_idmap_put(mnt_idmap(&mnt
->mnt
));
614 kfree_const(mnt
->mnt_devname
);
616 free_percpu(mnt
->mnt_pcp
);
618 kmem_cache_free(mnt_cache
, mnt
);
621 static void delayed_free_vfsmnt(struct rcu_head
*head
)
623 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
626 /* call under rcu_read_lock */
627 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
630 if (read_seqretry(&mount_lock
, seq
))
634 mnt
= real_mount(bastard
);
635 mnt_add_count(mnt
, 1);
636 smp_mb(); // see mntput_no_expire()
637 if (likely(!read_seqretry(&mount_lock
, seq
)))
639 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
640 mnt_add_count(mnt
, -1);
644 if (unlikely(bastard
->mnt_flags
& MNT_DOOMED
)) {
645 mnt_add_count(mnt
, -1);
650 /* caller will mntput() */
654 /* call under rcu_read_lock */
655 static bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
657 int res
= __legitimize_mnt(bastard
, seq
);
660 if (unlikely(res
< 0)) {
669 * __lookup_mnt - find first child mount
671 * @dentry: mountpoint
673 * If @mnt has a child mount @c mounted @dentry find and return it.
675 * Note that the child mount @c need not be unique. There are cases
676 * where shadow mounts are created. For example, during mount
677 * propagation when a source mount @mnt whose root got overmounted by a
678 * mount @o after path lookup but before @namespace_sem could be
679 * acquired gets copied and propagated. So @mnt gets copied including
680 * @o. When @mnt is propagated to a destination mount @d that already
681 * has another mount @n mounted at the same mountpoint then the source
682 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
683 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
686 * Return: The first child of @mnt mounted @dentry or NULL.
688 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
690 struct hlist_head
*head
= m_hash(mnt
, dentry
);
693 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
694 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
700 * lookup_mnt - Return the first child mount mounted at path
702 * "First" means first mounted chronologically. If you create the
705 * mount /dev/sda1 /mnt
706 * mount /dev/sda2 /mnt
707 * mount /dev/sda3 /mnt
709 * Then lookup_mnt() on the base /mnt dentry in the root mount will
710 * return successively the root dentry and vfsmount of /dev/sda1, then
711 * /dev/sda2, then /dev/sda3, then NULL.
713 * lookup_mnt takes a reference to the found vfsmount.
715 struct vfsmount
*lookup_mnt(const struct path
*path
)
717 struct mount
*child_mnt
;
723 seq
= read_seqbegin(&mount_lock
);
724 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
725 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
726 } while (!legitimize_mnt(m
, seq
));
731 static inline void lock_ns_list(struct mnt_namespace
*ns
)
733 spin_lock(&ns
->ns_lock
);
736 static inline void unlock_ns_list(struct mnt_namespace
*ns
)
738 spin_unlock(&ns
->ns_lock
);
741 static inline bool mnt_is_cursor(struct mount
*mnt
)
743 return mnt
->mnt
.mnt_flags
& MNT_CURSOR
;
747 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
748 * current mount namespace.
750 * The common case is dentries are not mountpoints at all and that
751 * test is handled inline. For the slow case when we are actually
752 * dealing with a mountpoint of some kind, walk through all of the
753 * mounts in the current mount namespace and test to see if the dentry
756 * The mount_hashtable is not usable in the context because we
757 * need to identify all mounts that may be in the current mount
758 * namespace not just a mount that happens to have some specified
761 bool __is_local_mountpoint(struct dentry
*dentry
)
763 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
765 bool is_covered
= false;
767 down_read(&namespace_sem
);
769 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
770 if (mnt_is_cursor(mnt
))
772 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
777 up_read(&namespace_sem
);
782 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
784 struct hlist_head
*chain
= mp_hash(dentry
);
785 struct mountpoint
*mp
;
787 hlist_for_each_entry(mp
, chain
, m_hash
) {
788 if (mp
->m_dentry
== dentry
) {
796 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
798 struct mountpoint
*mp
, *new = NULL
;
801 if (d_mountpoint(dentry
)) {
802 /* might be worth a WARN_ON() */
803 if (d_unlinked(dentry
))
804 return ERR_PTR(-ENOENT
);
806 read_seqlock_excl(&mount_lock
);
807 mp
= lookup_mountpoint(dentry
);
808 read_sequnlock_excl(&mount_lock
);
814 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
816 return ERR_PTR(-ENOMEM
);
819 /* Exactly one processes may set d_mounted */
820 ret
= d_set_mounted(dentry
);
822 /* Someone else set d_mounted? */
826 /* The dentry is not available as a mountpoint? */
831 /* Add the new mountpoint to the hash table */
832 read_seqlock_excl(&mount_lock
);
833 new->m_dentry
= dget(dentry
);
835 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
836 INIT_HLIST_HEAD(&new->m_list
);
837 read_sequnlock_excl(&mount_lock
);
847 * vfsmount lock must be held. Additionally, the caller is responsible
848 * for serializing calls for given disposal list.
850 static void __put_mountpoint(struct mountpoint
*mp
, struct list_head
*list
)
852 if (!--mp
->m_count
) {
853 struct dentry
*dentry
= mp
->m_dentry
;
854 BUG_ON(!hlist_empty(&mp
->m_list
));
855 spin_lock(&dentry
->d_lock
);
856 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
857 spin_unlock(&dentry
->d_lock
);
858 dput_to_list(dentry
, list
);
859 hlist_del(&mp
->m_hash
);
864 /* called with namespace_lock and vfsmount lock */
865 static void put_mountpoint(struct mountpoint
*mp
)
867 __put_mountpoint(mp
, &ex_mountpoints
);
870 static inline int check_mnt(struct mount
*mnt
)
872 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
876 * vfsmount lock must be held for write
878 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
882 wake_up_interruptible(&ns
->poll
);
887 * vfsmount lock must be held for write
889 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
891 if (ns
&& ns
->event
!= event
) {
893 wake_up_interruptible(&ns
->poll
);
898 * vfsmount lock must be held for write
900 static struct mountpoint
*unhash_mnt(struct mount
*mnt
)
902 struct mountpoint
*mp
;
903 mnt
->mnt_parent
= mnt
;
904 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
905 list_del_init(&mnt
->mnt_child
);
906 hlist_del_init_rcu(&mnt
->mnt_hash
);
907 hlist_del_init(&mnt
->mnt_mp_list
);
914 * vfsmount lock must be held for write
916 static void umount_mnt(struct mount
*mnt
)
918 put_mountpoint(unhash_mnt(mnt
));
922 * vfsmount lock must be held for write
924 void mnt_set_mountpoint(struct mount
*mnt
,
925 struct mountpoint
*mp
,
926 struct mount
*child_mnt
)
929 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
930 child_mnt
->mnt_mountpoint
= mp
->m_dentry
;
931 child_mnt
->mnt_parent
= mnt
;
932 child_mnt
->mnt_mp
= mp
;
933 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
937 * mnt_set_mountpoint_beneath - mount a mount beneath another one
939 * @new_parent: the source mount
940 * @top_mnt: the mount beneath which @new_parent is mounted
941 * @new_mp: the new mountpoint of @top_mnt on @new_parent
943 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
944 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
945 * @new_mp. And mount @new_parent on the old parent and old
946 * mountpoint of @top_mnt.
948 * Context: This function expects namespace_lock() and lock_mount_hash()
949 * to have been acquired in that order.
951 static void mnt_set_mountpoint_beneath(struct mount
*new_parent
,
952 struct mount
*top_mnt
,
953 struct mountpoint
*new_mp
)
955 struct mount
*old_top_parent
= top_mnt
->mnt_parent
;
956 struct mountpoint
*old_top_mp
= top_mnt
->mnt_mp
;
958 mnt_set_mountpoint(old_top_parent
, old_top_mp
, new_parent
);
959 mnt_change_mountpoint(new_parent
, new_mp
, top_mnt
);
963 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
965 hlist_add_head_rcu(&mnt
->mnt_hash
,
966 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
967 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
971 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
972 * list of child mounts
973 * @parent: the parent
974 * @mnt: the new mount
975 * @mp: the new mountpoint
976 * @beneath: whether to mount @mnt beneath or on top of @parent
978 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
979 * to @parent's child mount list and to @mount_hashtable.
981 * If @beneath is true, remove @mnt from its current parent and
982 * mountpoint and mount it on @mp on @parent, and mount @parent on the
983 * old parent and old mountpoint of @mnt. Finally, attach @parent to
984 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
986 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
987 * to the correct parent.
989 * Context: This function expects namespace_lock() and lock_mount_hash()
990 * to have been acquired in that order.
992 static void attach_mnt(struct mount
*mnt
, struct mount
*parent
,
993 struct mountpoint
*mp
, bool beneath
)
996 mnt_set_mountpoint_beneath(mnt
, parent
, mp
);
998 mnt_set_mountpoint(parent
, mp
, mnt
);
1000 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
1001 * beneath @parent then @mnt will need to be attached to
1002 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
1003 * isn't the same mount as @parent.
1005 __attach_mnt(mnt
, mnt
->mnt_parent
);
1008 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
1010 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
1011 struct mount
*old_parent
= mnt
->mnt_parent
;
1013 list_del_init(&mnt
->mnt_child
);
1014 hlist_del_init(&mnt
->mnt_mp_list
);
1015 hlist_del_init_rcu(&mnt
->mnt_hash
);
1017 attach_mnt(mnt
, parent
, mp
, false);
1019 put_mountpoint(old_mp
);
1020 mnt_add_count(old_parent
, -1);
1024 * vfsmount lock must be held for write
1026 static void commit_tree(struct mount
*mnt
)
1028 struct mount
*parent
= mnt
->mnt_parent
;
1031 struct mnt_namespace
*n
= parent
->mnt_ns
;
1033 BUG_ON(parent
== mnt
);
1035 list_add_tail(&head
, &mnt
->mnt_list
);
1036 list_for_each_entry(m
, &head
, mnt_list
)
1039 list_splice(&head
, n
->list
.prev
);
1041 n
->mounts
+= n
->pending_mounts
;
1042 n
->pending_mounts
= 0;
1044 __attach_mnt(mnt
, parent
);
1045 touch_mnt_namespace(n
);
1048 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
1050 struct list_head
*next
= p
->mnt_mounts
.next
;
1051 if (next
== &p
->mnt_mounts
) {
1055 next
= p
->mnt_child
.next
;
1056 if (next
!= &p
->mnt_parent
->mnt_mounts
)
1061 return list_entry(next
, struct mount
, mnt_child
);
1064 static struct mount
*skip_mnt_tree(struct mount
*p
)
1066 struct list_head
*prev
= p
->mnt_mounts
.prev
;
1067 while (prev
!= &p
->mnt_mounts
) {
1068 p
= list_entry(prev
, struct mount
, mnt_child
);
1069 prev
= p
->mnt_mounts
.prev
;
1075 * vfs_create_mount - Create a mount for a configured superblock
1076 * @fc: The configuration context with the superblock attached
1078 * Create a mount to an already configured superblock. If necessary, the
1079 * caller should invoke vfs_get_tree() before calling this.
1081 * Note that this does not attach the mount to anything.
1083 struct vfsmount
*vfs_create_mount(struct fs_context
*fc
)
1088 return ERR_PTR(-EINVAL
);
1090 mnt
= alloc_vfsmnt(fc
->source
?: "none");
1092 return ERR_PTR(-ENOMEM
);
1094 if (fc
->sb_flags
& SB_KERNMOUNT
)
1095 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
1097 atomic_inc(&fc
->root
->d_sb
->s_active
);
1098 mnt
->mnt
.mnt_sb
= fc
->root
->d_sb
;
1099 mnt
->mnt
.mnt_root
= dget(fc
->root
);
1100 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1101 mnt
->mnt_parent
= mnt
;
1104 list_add_tail(&mnt
->mnt_instance
, &mnt
->mnt
.mnt_sb
->s_mounts
);
1105 unlock_mount_hash();
1108 EXPORT_SYMBOL(vfs_create_mount
);
1110 struct vfsmount
*fc_mount(struct fs_context
*fc
)
1112 int err
= vfs_get_tree(fc
);
1114 up_write(&fc
->root
->d_sb
->s_umount
);
1115 return vfs_create_mount(fc
);
1117 return ERR_PTR(err
);
1119 EXPORT_SYMBOL(fc_mount
);
1121 struct vfsmount
*vfs_kern_mount(struct file_system_type
*type
,
1122 int flags
, const char *name
,
1125 struct fs_context
*fc
;
1126 struct vfsmount
*mnt
;
1130 return ERR_PTR(-EINVAL
);
1132 fc
= fs_context_for_mount(type
, flags
);
1134 return ERR_CAST(fc
);
1137 ret
= vfs_parse_fs_string(fc
, "source",
1138 name
, strlen(name
));
1140 ret
= parse_monolithic_mount_data(fc
, data
);
1149 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1152 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1153 const char *name
, void *data
)
1155 /* Until it is worked out how to pass the user namespace
1156 * through from the parent mount to the submount don't support
1157 * unprivileged mounts with submounts.
1159 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1160 return ERR_PTR(-EPERM
);
1162 return vfs_kern_mount(type
, SB_SUBMOUNT
, name
, data
);
1164 EXPORT_SYMBOL_GPL(vfs_submount
);
1166 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1169 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1173 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1175 return ERR_PTR(-ENOMEM
);
1177 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1178 mnt
->mnt_group_id
= 0; /* not a peer of original */
1180 mnt
->mnt_group_id
= old
->mnt_group_id
;
1182 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1183 err
= mnt_alloc_group_id(mnt
);
1188 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
;
1189 mnt
->mnt
.mnt_flags
&= ~(MNT_WRITE_HOLD
|MNT_MARKED
|MNT_INTERNAL
);
1191 atomic_inc(&sb
->s_active
);
1192 mnt
->mnt
.mnt_idmap
= mnt_idmap_get(mnt_idmap(&old
->mnt
));
1194 mnt
->mnt
.mnt_sb
= sb
;
1195 mnt
->mnt
.mnt_root
= dget(root
);
1196 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1197 mnt
->mnt_parent
= mnt
;
1199 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1200 unlock_mount_hash();
1202 if ((flag
& CL_SLAVE
) ||
1203 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1204 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1205 mnt
->mnt_master
= old
;
1206 CLEAR_MNT_SHARED(mnt
);
1207 } else if (!(flag
& CL_PRIVATE
)) {
1208 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1209 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1210 if (IS_MNT_SLAVE(old
))
1211 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1212 mnt
->mnt_master
= old
->mnt_master
;
1214 CLEAR_MNT_SHARED(mnt
);
1216 if (flag
& CL_MAKE_SHARED
)
1217 set_mnt_shared(mnt
);
1219 /* stick the duplicate mount on the same expiry list
1220 * as the original if that was on one */
1221 if (flag
& CL_EXPIRE
) {
1222 if (!list_empty(&old
->mnt_expire
))
1223 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1231 return ERR_PTR(err
);
1234 static void cleanup_mnt(struct mount
*mnt
)
1236 struct hlist_node
*p
;
1239 * The warning here probably indicates that somebody messed
1240 * up a mnt_want/drop_write() pair. If this happens, the
1241 * filesystem was probably unable to make r/w->r/o transitions.
1242 * The locking used to deal with mnt_count decrement provides barriers,
1243 * so mnt_get_writers() below is safe.
1245 WARN_ON(mnt_get_writers(mnt
));
1246 if (unlikely(mnt
->mnt_pins
.first
))
1248 hlist_for_each_entry_safe(m
, p
, &mnt
->mnt_stuck_children
, mnt_umount
) {
1249 hlist_del(&m
->mnt_umount
);
1252 fsnotify_vfsmount_delete(&mnt
->mnt
);
1253 dput(mnt
->mnt
.mnt_root
);
1254 deactivate_super(mnt
->mnt
.mnt_sb
);
1256 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1259 static void __cleanup_mnt(struct rcu_head
*head
)
1261 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1264 static LLIST_HEAD(delayed_mntput_list
);
1265 static void delayed_mntput(struct work_struct
*unused
)
1267 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1268 struct mount
*m
, *t
;
1270 llist_for_each_entry_safe(m
, t
, node
, mnt_llist
)
1273 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1275 static void mntput_no_expire(struct mount
*mnt
)
1281 if (likely(READ_ONCE(mnt
->mnt_ns
))) {
1283 * Since we don't do lock_mount_hash() here,
1284 * ->mnt_ns can change under us. However, if it's
1285 * non-NULL, then there's a reference that won't
1286 * be dropped until after an RCU delay done after
1287 * turning ->mnt_ns NULL. So if we observe it
1288 * non-NULL under rcu_read_lock(), the reference
1289 * we are dropping is not the final one.
1291 mnt_add_count(mnt
, -1);
1297 * make sure that if __legitimize_mnt() has not seen us grab
1298 * mount_lock, we'll see their refcount increment here.
1301 mnt_add_count(mnt
, -1);
1302 count
= mnt_get_count(mnt
);
1306 unlock_mount_hash();
1309 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1311 unlock_mount_hash();
1314 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1317 list_del(&mnt
->mnt_instance
);
1319 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1320 struct mount
*p
, *tmp
;
1321 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1322 __put_mountpoint(unhash_mnt(p
), &list
);
1323 hlist_add_head(&p
->mnt_umount
, &mnt
->mnt_stuck_children
);
1326 unlock_mount_hash();
1327 shrink_dentry_list(&list
);
1329 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1330 struct task_struct
*task
= current
;
1331 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1332 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1333 if (!task_work_add(task
, &mnt
->mnt_rcu
, TWA_RESUME
))
1336 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1337 schedule_delayed_work(&delayed_mntput_work
, 1);
1343 void mntput(struct vfsmount
*mnt
)
1346 struct mount
*m
= real_mount(mnt
);
1347 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1348 if (unlikely(m
->mnt_expiry_mark
))
1349 m
->mnt_expiry_mark
= 0;
1350 mntput_no_expire(m
);
1353 EXPORT_SYMBOL(mntput
);
1355 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1358 mnt_add_count(real_mount(mnt
), 1);
1361 EXPORT_SYMBOL(mntget
);
1364 * Make a mount point inaccessible to new lookups.
1365 * Because there may still be current users, the caller MUST WAIT
1366 * for an RCU grace period before destroying the mount point.
1368 void mnt_make_shortterm(struct vfsmount
*mnt
)
1371 real_mount(mnt
)->mnt_ns
= NULL
;
1375 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1376 * @path: path to check
1378 * d_mountpoint() can only be used reliably to establish if a dentry is
1379 * not mounted in any namespace and that common case is handled inline.
1380 * d_mountpoint() isn't aware of the possibility there may be multiple
1381 * mounts using a given dentry in a different namespace. This function
1382 * checks if the passed in path is a mountpoint rather than the dentry
1385 bool path_is_mountpoint(const struct path
*path
)
1390 if (!d_mountpoint(path
->dentry
))
1395 seq
= read_seqbegin(&mount_lock
);
1396 res
= __path_is_mountpoint(path
);
1397 } while (read_seqretry(&mount_lock
, seq
));
1402 EXPORT_SYMBOL(path_is_mountpoint
);
1404 struct vfsmount
*mnt_clone_internal(const struct path
*path
)
1407 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1410 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1414 #ifdef CONFIG_PROC_FS
1415 static struct mount
*mnt_list_next(struct mnt_namespace
*ns
,
1416 struct list_head
*p
)
1418 struct mount
*mnt
, *ret
= NULL
;
1421 list_for_each_continue(p
, &ns
->list
) {
1422 mnt
= list_entry(p
, typeof(*mnt
), mnt_list
);
1423 if (!mnt_is_cursor(mnt
)) {
1433 /* iterator; we want it to have access to namespace_sem, thus here... */
1434 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1436 struct proc_mounts
*p
= m
->private;
1437 struct list_head
*prev
;
1439 down_read(&namespace_sem
);
1441 prev
= &p
->ns
->list
;
1443 prev
= &p
->cursor
.mnt_list
;
1445 /* Read after we'd reached the end? */
1446 if (list_empty(prev
))
1450 return mnt_list_next(p
->ns
, prev
);
1453 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1455 struct proc_mounts
*p
= m
->private;
1456 struct mount
*mnt
= v
;
1459 return mnt_list_next(p
->ns
, &mnt
->mnt_list
);
1462 static void m_stop(struct seq_file
*m
, void *v
)
1464 struct proc_mounts
*p
= m
->private;
1465 struct mount
*mnt
= v
;
1467 lock_ns_list(p
->ns
);
1469 list_move_tail(&p
->cursor
.mnt_list
, &mnt
->mnt_list
);
1471 list_del_init(&p
->cursor
.mnt_list
);
1472 unlock_ns_list(p
->ns
);
1473 up_read(&namespace_sem
);
1476 static int m_show(struct seq_file
*m
, void *v
)
1478 struct proc_mounts
*p
= m
->private;
1479 struct mount
*r
= v
;
1480 return p
->show(m
, &r
->mnt
);
1483 const struct seq_operations mounts_op
= {
1490 void mnt_cursor_del(struct mnt_namespace
*ns
, struct mount
*cursor
)
1492 down_read(&namespace_sem
);
1494 list_del(&cursor
->mnt_list
);
1496 up_read(&namespace_sem
);
1498 #endif /* CONFIG_PROC_FS */
1501 * may_umount_tree - check if a mount tree is busy
1502 * @m: root of mount tree
1504 * This is called to check if a tree of mounts has any
1505 * open files, pwds, chroots or sub mounts that are
1508 int may_umount_tree(struct vfsmount
*m
)
1510 struct mount
*mnt
= real_mount(m
);
1511 int actual_refs
= 0;
1512 int minimum_refs
= 0;
1516 /* write lock needed for mnt_get_count */
1518 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1519 actual_refs
+= mnt_get_count(p
);
1522 unlock_mount_hash();
1524 if (actual_refs
> minimum_refs
)
1530 EXPORT_SYMBOL(may_umount_tree
);
1533 * may_umount - check if a mount point is busy
1534 * @mnt: root of mount
1536 * This is called to check if a mount point has any
1537 * open files, pwds, chroots or sub mounts. If the
1538 * mount has sub mounts this will return busy
1539 * regardless of whether the sub mounts are busy.
1541 * Doesn't take quota and stuff into account. IOW, in some cases it will
1542 * give false negatives. The main reason why it's here is that we need
1543 * a non-destructive way to look for easily umountable filesystems.
1545 int may_umount(struct vfsmount
*mnt
)
1548 down_read(&namespace_sem
);
1550 if (propagate_mount_busy(real_mount(mnt
), 2))
1552 unlock_mount_hash();
1553 up_read(&namespace_sem
);
1557 EXPORT_SYMBOL(may_umount
);
1559 static void namespace_unlock(void)
1561 struct hlist_head head
;
1562 struct hlist_node
*p
;
1566 hlist_move_list(&unmounted
, &head
);
1567 list_splice_init(&ex_mountpoints
, &list
);
1569 up_write(&namespace_sem
);
1571 shrink_dentry_list(&list
);
1573 if (likely(hlist_empty(&head
)))
1576 synchronize_rcu_expedited();
1578 hlist_for_each_entry_safe(m
, p
, &head
, mnt_umount
) {
1579 hlist_del(&m
->mnt_umount
);
1584 static inline void namespace_lock(void)
1586 down_write(&namespace_sem
);
1589 enum umount_tree_flags
{
1591 UMOUNT_PROPAGATE
= 2,
1592 UMOUNT_CONNECTED
= 4,
1595 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1597 /* Leaving mounts connected is only valid for lazy umounts */
1598 if (how
& UMOUNT_SYNC
)
1601 /* A mount without a parent has nothing to be connected to */
1602 if (!mnt_has_parent(mnt
))
1605 /* Because the reference counting rules change when mounts are
1606 * unmounted and connected, umounted mounts may not be
1607 * connected to mounted mounts.
1609 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1612 /* Has it been requested that the mount remain connected? */
1613 if (how
& UMOUNT_CONNECTED
)
1616 /* Is the mount locked such that it needs to remain connected? */
1617 if (IS_MNT_LOCKED(mnt
))
1620 /* By default disconnect the mount */
1625 * mount_lock must be held
1626 * namespace_sem must be held for write
1628 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1630 LIST_HEAD(tmp_list
);
1633 if (how
& UMOUNT_PROPAGATE
)
1634 propagate_mount_unlock(mnt
);
1636 /* Gather the mounts to umount */
1637 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1638 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1639 list_move(&p
->mnt_list
, &tmp_list
);
1642 /* Hide the mounts from mnt_mounts */
1643 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1644 list_del_init(&p
->mnt_child
);
1647 /* Add propogated mounts to the tmp_list */
1648 if (how
& UMOUNT_PROPAGATE
)
1649 propagate_umount(&tmp_list
);
1651 while (!list_empty(&tmp_list
)) {
1652 struct mnt_namespace
*ns
;
1654 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1655 list_del_init(&p
->mnt_expire
);
1656 list_del_init(&p
->mnt_list
);
1660 __touch_mnt_namespace(ns
);
1663 if (how
& UMOUNT_SYNC
)
1664 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1666 disconnect
= disconnect_mount(p
, how
);
1667 if (mnt_has_parent(p
)) {
1668 mnt_add_count(p
->mnt_parent
, -1);
1670 /* Don't forget about p */
1671 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1676 change_mnt_propagation(p
, MS_PRIVATE
);
1678 hlist_add_head(&p
->mnt_umount
, &unmounted
);
1682 static void shrink_submounts(struct mount
*mnt
);
1684 static int do_umount_root(struct super_block
*sb
)
1688 down_write(&sb
->s_umount
);
1689 if (!sb_rdonly(sb
)) {
1690 struct fs_context
*fc
;
1692 fc
= fs_context_for_reconfigure(sb
->s_root
, SB_RDONLY
,
1697 ret
= parse_monolithic_mount_data(fc
, NULL
);
1699 ret
= reconfigure_super(fc
);
1703 up_write(&sb
->s_umount
);
1707 static int do_umount(struct mount
*mnt
, int flags
)
1709 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1712 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1717 * Allow userspace to request a mountpoint be expired rather than
1718 * unmounting unconditionally. Unmount only happens if:
1719 * (1) the mark is already set (the mark is cleared by mntput())
1720 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1722 if (flags
& MNT_EXPIRE
) {
1723 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1724 flags
& (MNT_FORCE
| MNT_DETACH
))
1728 * probably don't strictly need the lock here if we examined
1729 * all race cases, but it's a slowpath.
1732 if (mnt_get_count(mnt
) != 2) {
1733 unlock_mount_hash();
1736 unlock_mount_hash();
1738 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1743 * If we may have to abort operations to get out of this
1744 * mount, and they will themselves hold resources we must
1745 * allow the fs to do things. In the Unix tradition of
1746 * 'Gee thats tricky lets do it in userspace' the umount_begin
1747 * might fail to complete on the first run through as other tasks
1748 * must return, and the like. Thats for the mount program to worry
1749 * about for the moment.
1752 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1753 sb
->s_op
->umount_begin(sb
);
1757 * No sense to grab the lock for this test, but test itself looks
1758 * somewhat bogus. Suggestions for better replacement?
1759 * Ho-hum... In principle, we might treat that as umount + switch
1760 * to rootfs. GC would eventually take care of the old vfsmount.
1761 * Actually it makes sense, especially if rootfs would contain a
1762 * /reboot - static binary that would close all descriptors and
1763 * call reboot(9). Then init(8) could umount root and exec /reboot.
1765 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1767 * Special case for "unmounting" root ...
1768 * we just try to remount it readonly.
1770 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1772 return do_umount_root(sb
);
1778 /* Recheck MNT_LOCKED with the locks held */
1780 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1784 if (flags
& MNT_DETACH
) {
1785 if (!list_empty(&mnt
->mnt_list
))
1786 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1789 shrink_submounts(mnt
);
1791 if (!propagate_mount_busy(mnt
, 2)) {
1792 if (!list_empty(&mnt
->mnt_list
))
1793 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1798 unlock_mount_hash();
1804 * __detach_mounts - lazily unmount all mounts on the specified dentry
1806 * During unlink, rmdir, and d_drop it is possible to loose the path
1807 * to an existing mountpoint, and wind up leaking the mount.
1808 * detach_mounts allows lazily unmounting those mounts instead of
1811 * The caller may hold dentry->d_inode->i_mutex.
1813 void __detach_mounts(struct dentry
*dentry
)
1815 struct mountpoint
*mp
;
1820 mp
= lookup_mountpoint(dentry
);
1825 while (!hlist_empty(&mp
->m_list
)) {
1826 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1827 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1829 hlist_add_head(&mnt
->mnt_umount
, &unmounted
);
1831 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1835 unlock_mount_hash();
1840 * Is the caller allowed to modify his namespace?
1842 bool may_mount(void)
1844 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1848 * path_mounted - check whether path is mounted
1849 * @path: path to check
1851 * Determine whether @path refers to the root of a mount.
1853 * Return: true if @path is the root of a mount, false if not.
1855 static inline bool path_mounted(const struct path
*path
)
1857 return path
->mnt
->mnt_root
== path
->dentry
;
1860 static void warn_mandlock(void)
1862 pr_warn_once("=======================================================\n"
1863 "WARNING: The mand mount option has been deprecated and\n"
1864 " and is ignored by this kernel. Remove the mand\n"
1865 " option from the mount to silence this warning.\n"
1866 "=======================================================\n");
1869 static int can_umount(const struct path
*path
, int flags
)
1871 struct mount
*mnt
= real_mount(path
->mnt
);
1875 if (!path_mounted(path
))
1877 if (!check_mnt(mnt
))
1879 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
) /* Check optimistically */
1881 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1886 // caller is responsible for flags being sane
1887 int path_umount(struct path
*path
, int flags
)
1889 struct mount
*mnt
= real_mount(path
->mnt
);
1892 ret
= can_umount(path
, flags
);
1894 ret
= do_umount(mnt
, flags
);
1896 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1898 mntput_no_expire(mnt
);
1902 static int ksys_umount(char __user
*name
, int flags
)
1904 int lookup_flags
= LOOKUP_MOUNTPOINT
;
1908 // basic validity checks done first
1909 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1912 if (!(flags
& UMOUNT_NOFOLLOW
))
1913 lookup_flags
|= LOOKUP_FOLLOW
;
1914 ret
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1917 return path_umount(&path
, flags
);
1920 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1922 return ksys_umount(name
, flags
);
1925 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1928 * The 2.0 compatible umount. No flags.
1930 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1932 return ksys_umount(name
, 0);
1937 static bool is_mnt_ns_file(struct dentry
*dentry
)
1939 /* Is this a proxy for a mount namespace? */
1940 return dentry
->d_op
== &ns_dentry_operations
&&
1941 dentry
->d_fsdata
== &mntns_operations
;
1944 static struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1946 return container_of(ns
, struct mnt_namespace
, ns
);
1949 struct ns_common
*from_mnt_ns(struct mnt_namespace
*mnt
)
1954 static bool mnt_ns_loop(struct dentry
*dentry
)
1956 /* Could bind mounting the mount namespace inode cause a
1957 * mount namespace loop?
1959 struct mnt_namespace
*mnt_ns
;
1960 if (!is_mnt_ns_file(dentry
))
1963 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1964 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1967 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1970 struct mount
*res
, *p
, *q
, *r
, *parent
;
1972 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1973 return ERR_PTR(-EINVAL
);
1975 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1976 return ERR_PTR(-EINVAL
);
1978 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1982 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1985 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1987 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1990 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1991 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1992 IS_MNT_UNBINDABLE(s
)) {
1993 if (s
->mnt
.mnt_flags
& MNT_LOCKED
) {
1994 /* Both unbindable and locked. */
1995 q
= ERR_PTR(-EPERM
);
1998 s
= skip_mnt_tree(s
);
2002 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
2003 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
2004 s
= skip_mnt_tree(s
);
2007 while (p
!= s
->mnt_parent
) {
2013 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
2017 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
2018 attach_mnt(q
, parent
, p
->mnt_mp
, false);
2019 unlock_mount_hash();
2026 umount_tree(res
, UMOUNT_SYNC
);
2027 unlock_mount_hash();
2032 /* Caller should check returned pointer for errors */
2034 struct vfsmount
*collect_mounts(const struct path
*path
)
2038 if (!check_mnt(real_mount(path
->mnt
)))
2039 tree
= ERR_PTR(-EINVAL
);
2041 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
2042 CL_COPY_ALL
| CL_PRIVATE
);
2045 return ERR_CAST(tree
);
2049 static void free_mnt_ns(struct mnt_namespace
*);
2050 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*, bool);
2052 void dissolve_on_fput(struct vfsmount
*mnt
)
2054 struct mnt_namespace
*ns
;
2057 ns
= real_mount(mnt
)->mnt_ns
;
2060 umount_tree(real_mount(mnt
), UMOUNT_CONNECTED
);
2064 unlock_mount_hash();
2070 void drop_collected_mounts(struct vfsmount
*mnt
)
2074 umount_tree(real_mount(mnt
), 0);
2075 unlock_mount_hash();
2079 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2081 struct mount
*child
;
2083 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2084 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2087 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2094 * clone_private_mount - create a private clone of a path
2095 * @path: path to clone
2097 * This creates a new vfsmount, which will be the clone of @path. The new mount
2098 * will not be attached anywhere in the namespace and will be private (i.e.
2099 * changes to the originating mount won't be propagated into this).
2101 * Release with mntput().
2103 struct vfsmount
*clone_private_mount(const struct path
*path
)
2105 struct mount
*old_mnt
= real_mount(path
->mnt
);
2106 struct mount
*new_mnt
;
2108 down_read(&namespace_sem
);
2109 if (IS_MNT_UNBINDABLE(old_mnt
))
2112 if (!check_mnt(old_mnt
))
2115 if (has_locked_children(old_mnt
, path
->dentry
))
2118 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
2119 up_read(&namespace_sem
);
2121 if (IS_ERR(new_mnt
))
2122 return ERR_CAST(new_mnt
);
2124 /* Longterm mount to be removed by kern_unmount*() */
2125 new_mnt
->mnt_ns
= MNT_NS_INTERNAL
;
2127 return &new_mnt
->mnt
;
2130 up_read(&namespace_sem
);
2131 return ERR_PTR(-EINVAL
);
2133 EXPORT_SYMBOL_GPL(clone_private_mount
);
2135 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
2136 struct vfsmount
*root
)
2139 int res
= f(root
, arg
);
2142 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
2143 res
= f(&mnt
->mnt
, arg
);
2150 static void lock_mnt_tree(struct mount
*mnt
)
2154 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2155 int flags
= p
->mnt
.mnt_flags
;
2156 /* Don't allow unprivileged users to change mount flags */
2157 flags
|= MNT_LOCK_ATIME
;
2159 if (flags
& MNT_READONLY
)
2160 flags
|= MNT_LOCK_READONLY
;
2162 if (flags
& MNT_NODEV
)
2163 flags
|= MNT_LOCK_NODEV
;
2165 if (flags
& MNT_NOSUID
)
2166 flags
|= MNT_LOCK_NOSUID
;
2168 if (flags
& MNT_NOEXEC
)
2169 flags
|= MNT_LOCK_NOEXEC
;
2170 /* Don't allow unprivileged users to reveal what is under a mount */
2171 if (list_empty(&p
->mnt_expire
))
2172 flags
|= MNT_LOCKED
;
2173 p
->mnt
.mnt_flags
= flags
;
2177 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
2181 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
2182 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
2183 mnt_release_group_id(p
);
2187 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
2191 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
2192 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
2193 int err
= mnt_alloc_group_id(p
);
2195 cleanup_group_ids(mnt
, p
);
2204 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
2206 unsigned int max
= READ_ONCE(sysctl_mount_max
);
2207 unsigned int mounts
= 0;
2210 if (ns
->mounts
>= max
)
2213 if (ns
->pending_mounts
>= max
)
2215 max
-= ns
->pending_mounts
;
2217 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
2223 ns
->pending_mounts
+= mounts
;
2227 enum mnt_tree_flags_t
{
2228 MNT_TREE_MOVE
= BIT(0),
2229 MNT_TREE_BENEATH
= BIT(1),
2233 * attach_recursive_mnt - attach a source mount tree
2234 * @source_mnt: mount tree to be attached
2235 * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath
2236 * @dest_mp: the mountpoint @source_mnt will be mounted at
2237 * @flags: modify how @source_mnt is supposed to be attached
2239 * NOTE: in the table below explains the semantics when a source mount
2240 * of a given type is attached to a destination mount of a given type.
2241 * ---------------------------------------------------------------------------
2242 * | BIND MOUNT OPERATION |
2243 * |**************************************************************************
2244 * | source-->| shared | private | slave | unbindable |
2248 * |**************************************************************************
2249 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2251 * |non-shared| shared (+) | private | slave (*) | invalid |
2252 * ***************************************************************************
2253 * A bind operation clones the source mount and mounts the clone on the
2254 * destination mount.
2256 * (++) the cloned mount is propagated to all the mounts in the propagation
2257 * tree of the destination mount and the cloned mount is added to
2258 * the peer group of the source mount.
2259 * (+) the cloned mount is created under the destination mount and is marked
2260 * as shared. The cloned mount is added to the peer group of the source
2262 * (+++) the mount is propagated to all the mounts in the propagation tree
2263 * of the destination mount and the cloned mount is made slave
2264 * of the same master as that of the source mount. The cloned mount
2265 * is marked as 'shared and slave'.
2266 * (*) the cloned mount is made a slave of the same master as that of the
2269 * ---------------------------------------------------------------------------
2270 * | MOVE MOUNT OPERATION |
2271 * |**************************************************************************
2272 * | source-->| shared | private | slave | unbindable |
2276 * |**************************************************************************
2277 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2279 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2280 * ***************************************************************************
2282 * (+) the mount is moved to the destination. And is then propagated to
2283 * all the mounts in the propagation tree of the destination mount.
2284 * (+*) the mount is moved to the destination.
2285 * (+++) the mount is moved to the destination and is then propagated to
2286 * all the mounts belonging to the destination mount's propagation tree.
2287 * the mount is marked as 'shared and slave'.
2288 * (*) the mount continues to be a slave at the new location.
2290 * if the source mount is a tree, the operations explained above is
2291 * applied to each mount in the tree.
2292 * Must be called without spinlocks held, since this function can sleep
2295 * Context: The function expects namespace_lock() to be held.
2296 * Return: If @source_mnt was successfully attached 0 is returned.
2297 * Otherwise a negative error code is returned.
2299 static int attach_recursive_mnt(struct mount
*source_mnt
,
2300 struct mount
*top_mnt
,
2301 struct mountpoint
*dest_mp
,
2302 enum mnt_tree_flags_t flags
)
2304 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2305 HLIST_HEAD(tree_list
);
2306 struct mnt_namespace
*ns
= top_mnt
->mnt_ns
;
2307 struct mountpoint
*smp
;
2308 struct mount
*child
, *dest_mnt
, *p
;
2309 struct hlist_node
*n
;
2311 bool moving
= flags
& MNT_TREE_MOVE
, beneath
= flags
& MNT_TREE_BENEATH
;
2314 * Preallocate a mountpoint in case the new mounts need to be
2315 * mounted beneath mounts on the same mountpoint.
2317 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
2319 return PTR_ERR(smp
);
2321 /* Is there space to add these mounts to the mount namespace? */
2323 err
= count_mounts(ns
, source_mnt
);
2329 dest_mnt
= top_mnt
->mnt_parent
;
2333 if (IS_MNT_SHARED(dest_mnt
)) {
2334 err
= invent_group_ids(source_mnt
, true);
2337 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2341 goto out_cleanup_ids
;
2343 if (IS_MNT_SHARED(dest_mnt
)) {
2344 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2351 unhash_mnt(source_mnt
);
2352 attach_mnt(source_mnt
, top_mnt
, dest_mp
, beneath
);
2353 touch_mnt_namespace(source_mnt
->mnt_ns
);
2355 if (source_mnt
->mnt_ns
) {
2356 /* move from anon - the caller will destroy */
2357 list_del_init(&source_mnt
->mnt_ns
->list
);
2360 mnt_set_mountpoint_beneath(source_mnt
, top_mnt
, smp
);
2362 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2363 commit_tree(source_mnt
);
2366 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2368 hlist_del_init(&child
->mnt_hash
);
2369 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2370 child
->mnt_mountpoint
);
2372 mnt_change_mountpoint(child
, smp
, q
);
2373 /* Notice when we are propagating across user namespaces */
2374 if (child
->mnt_parent
->mnt_ns
->user_ns
!= user_ns
)
2375 lock_mnt_tree(child
);
2376 child
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2379 put_mountpoint(smp
);
2380 unlock_mount_hash();
2385 while (!hlist_empty(&tree_list
)) {
2386 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2387 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2388 umount_tree(child
, UMOUNT_SYNC
);
2390 unlock_mount_hash();
2391 cleanup_group_ids(source_mnt
, NULL
);
2393 ns
->pending_mounts
= 0;
2395 read_seqlock_excl(&mount_lock
);
2396 put_mountpoint(smp
);
2397 read_sequnlock_excl(&mount_lock
);
2403 * do_lock_mount - lock mount and mountpoint
2404 * @path: target path
2405 * @beneath: whether the intention is to mount beneath @path
2407 * Follow the mount stack on @path until the top mount @mnt is found. If
2408 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2409 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2410 * until nothing is stacked on top of it anymore.
2412 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2413 * against concurrent removal of the new mountpoint from another mount
2416 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2417 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2418 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2419 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2420 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2421 * on top of it for @beneath.
2423 * In addition, @beneath needs to make sure that @mnt hasn't been
2424 * unmounted or moved from its current mountpoint in between dropping
2425 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2426 * being unmounted would be detected later by e.g., calling
2427 * check_mnt(mnt) in the function it's called from. For the @beneath
2428 * case however, it's useful to detect it directly in do_lock_mount().
2429 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2430 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2431 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2433 * Return: Either the target mountpoint on the top mount or the top
2434 * mount's mountpoint.
2436 static struct mountpoint
*do_lock_mount(struct path
*path
, bool beneath
)
2438 struct vfsmount
*mnt
= path
->mnt
;
2439 struct dentry
*dentry
;
2440 struct mountpoint
*mp
= ERR_PTR(-ENOENT
);
2446 m
= real_mount(mnt
);
2447 read_seqlock_excl(&mount_lock
);
2448 dentry
= dget(m
->mnt_mountpoint
);
2449 read_sequnlock_excl(&mount_lock
);
2451 dentry
= path
->dentry
;
2454 inode_lock(dentry
->d_inode
);
2455 if (unlikely(cant_mount(dentry
))) {
2456 inode_unlock(dentry
->d_inode
);
2462 if (beneath
&& (!is_mounted(mnt
) || m
->mnt_mountpoint
!= dentry
)) {
2464 inode_unlock(dentry
->d_inode
);
2468 mnt
= lookup_mnt(path
);
2473 inode_unlock(dentry
->d_inode
);
2478 path
->dentry
= dget(mnt
->mnt_root
);
2481 mp
= get_mountpoint(dentry
);
2484 inode_unlock(dentry
->d_inode
);
2494 static inline struct mountpoint
*lock_mount(struct path
*path
)
2496 return do_lock_mount(path
, false);
2499 static void unlock_mount(struct mountpoint
*where
)
2501 struct dentry
*dentry
= where
->m_dentry
;
2503 read_seqlock_excl(&mount_lock
);
2504 put_mountpoint(where
);
2505 read_sequnlock_excl(&mount_lock
);
2508 inode_unlock(dentry
->d_inode
);
2511 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2513 if (mnt
->mnt
.mnt_sb
->s_flags
& SB_NOUSER
)
2516 if (d_is_dir(mp
->m_dentry
) !=
2517 d_is_dir(mnt
->mnt
.mnt_root
))
2520 return attach_recursive_mnt(mnt
, p
, mp
, 0);
2524 * Sanity check the flags to change_mnt_propagation.
2527 static int flags_to_propagation_type(int ms_flags
)
2529 int type
= ms_flags
& ~(MS_REC
| MS_SILENT
);
2531 /* Fail if any non-propagation flags are set */
2532 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2534 /* Only one propagation flag should be set */
2535 if (!is_power_of_2(type
))
2541 * recursively change the type of the mountpoint.
2543 static int do_change_type(struct path
*path
, int ms_flags
)
2546 struct mount
*mnt
= real_mount(path
->mnt
);
2547 int recurse
= ms_flags
& MS_REC
;
2551 if (!path_mounted(path
))
2554 type
= flags_to_propagation_type(ms_flags
);
2559 if (type
== MS_SHARED
) {
2560 err
= invent_group_ids(mnt
, recurse
);
2566 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2567 change_mnt_propagation(m
, type
);
2568 unlock_mount_hash();
2575 static struct mount
*__do_loopback(struct path
*old_path
, int recurse
)
2577 struct mount
*mnt
= ERR_PTR(-EINVAL
), *old
= real_mount(old_path
->mnt
);
2579 if (IS_MNT_UNBINDABLE(old
))
2582 if (!check_mnt(old
) && old_path
->dentry
->d_op
!= &ns_dentry_operations
)
2585 if (!recurse
&& has_locked_children(old
, old_path
->dentry
))
2589 mnt
= copy_tree(old
, old_path
->dentry
, CL_COPY_MNT_NS_FILE
);
2591 mnt
= clone_mnt(old
, old_path
->dentry
, 0);
2594 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2600 * do loopback mount.
2602 static int do_loopback(struct path
*path
, const char *old_name
,
2605 struct path old_path
;
2606 struct mount
*mnt
= NULL
, *parent
;
2607 struct mountpoint
*mp
;
2609 if (!old_name
|| !*old_name
)
2611 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2616 if (mnt_ns_loop(old_path
.dentry
))
2619 mp
= lock_mount(path
);
2625 parent
= real_mount(path
->mnt
);
2626 if (!check_mnt(parent
))
2629 mnt
= __do_loopback(&old_path
, recurse
);
2635 err
= graft_tree(mnt
, parent
, mp
);
2638 umount_tree(mnt
, UMOUNT_SYNC
);
2639 unlock_mount_hash();
2644 path_put(&old_path
);
2648 static struct file
*open_detached_copy(struct path
*path
, bool recursive
)
2650 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2651 struct mnt_namespace
*ns
= alloc_mnt_ns(user_ns
, true);
2652 struct mount
*mnt
, *p
;
2656 return ERR_CAST(ns
);
2659 mnt
= __do_loopback(path
, recursive
);
2663 return ERR_CAST(mnt
);
2667 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2672 list_add_tail(&ns
->list
, &mnt
->mnt_list
);
2674 unlock_mount_hash();
2678 path
->mnt
= &mnt
->mnt
;
2679 file
= dentry_open(path
, O_PATH
, current_cred());
2681 dissolve_on_fput(path
->mnt
);
2683 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
2687 SYSCALL_DEFINE3(open_tree
, int, dfd
, const char __user
*, filename
, unsigned, flags
)
2691 int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
2692 bool detached
= flags
& OPEN_TREE_CLONE
;
2696 BUILD_BUG_ON(OPEN_TREE_CLOEXEC
!= O_CLOEXEC
);
2698 if (flags
& ~(AT_EMPTY_PATH
| AT_NO_AUTOMOUNT
| AT_RECURSIVE
|
2699 AT_SYMLINK_NOFOLLOW
| OPEN_TREE_CLONE
|
2703 if ((flags
& (AT_RECURSIVE
| OPEN_TREE_CLONE
)) == AT_RECURSIVE
)
2706 if (flags
& AT_NO_AUTOMOUNT
)
2707 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
2708 if (flags
& AT_SYMLINK_NOFOLLOW
)
2709 lookup_flags
&= ~LOOKUP_FOLLOW
;
2710 if (flags
& AT_EMPTY_PATH
)
2711 lookup_flags
|= LOOKUP_EMPTY
;
2713 if (detached
&& !may_mount())
2716 fd
= get_unused_fd_flags(flags
& O_CLOEXEC
);
2720 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
2721 if (unlikely(error
)) {
2722 file
= ERR_PTR(error
);
2725 file
= open_detached_copy(&path
, flags
& AT_RECURSIVE
);
2727 file
= dentry_open(&path
, O_PATH
, current_cred());
2732 return PTR_ERR(file
);
2734 fd_install(fd
, file
);
2739 * Don't allow locked mount flags to be cleared.
2741 * No locks need to be held here while testing the various MNT_LOCK
2742 * flags because those flags can never be cleared once they are set.
2744 static bool can_change_locked_flags(struct mount
*mnt
, unsigned int mnt_flags
)
2746 unsigned int fl
= mnt
->mnt
.mnt_flags
;
2748 if ((fl
& MNT_LOCK_READONLY
) &&
2749 !(mnt_flags
& MNT_READONLY
))
2752 if ((fl
& MNT_LOCK_NODEV
) &&
2753 !(mnt_flags
& MNT_NODEV
))
2756 if ((fl
& MNT_LOCK_NOSUID
) &&
2757 !(mnt_flags
& MNT_NOSUID
))
2760 if ((fl
& MNT_LOCK_NOEXEC
) &&
2761 !(mnt_flags
& MNT_NOEXEC
))
2764 if ((fl
& MNT_LOCK_ATIME
) &&
2765 ((fl
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
)))
2771 static int change_mount_ro_state(struct mount
*mnt
, unsigned int mnt_flags
)
2773 bool readonly_request
= (mnt_flags
& MNT_READONLY
);
2775 if (readonly_request
== __mnt_is_readonly(&mnt
->mnt
))
2778 if (readonly_request
)
2779 return mnt_make_readonly(mnt
);
2781 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
2785 static void set_mount_attributes(struct mount
*mnt
, unsigned int mnt_flags
)
2787 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2788 mnt
->mnt
.mnt_flags
= mnt_flags
;
2789 touch_mnt_namespace(mnt
->mnt_ns
);
2792 static void mnt_warn_timestamp_expiry(struct path
*mountpoint
, struct vfsmount
*mnt
)
2794 struct super_block
*sb
= mnt
->mnt_sb
;
2796 if (!__mnt_is_readonly(mnt
) &&
2797 (!(sb
->s_iflags
& SB_I_TS_EXPIRY_WARNED
)) &&
2798 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX
> sb
->s_time_max
)) {
2799 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
2800 char *mntpath
= buf
? d_path(mountpoint
, buf
, PAGE_SIZE
) : ERR_PTR(-ENOMEM
);
2802 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2804 is_mounted(mnt
) ? "remounted" : "mounted",
2805 mntpath
, &sb
->s_time_max
,
2806 (unsigned long long)sb
->s_time_max
);
2808 free_page((unsigned long)buf
);
2809 sb
->s_iflags
|= SB_I_TS_EXPIRY_WARNED
;
2814 * Handle reconfiguration of the mountpoint only without alteration of the
2815 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2818 static int do_reconfigure_mnt(struct path
*path
, unsigned int mnt_flags
)
2820 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2821 struct mount
*mnt
= real_mount(path
->mnt
);
2824 if (!check_mnt(mnt
))
2827 if (!path_mounted(path
))
2830 if (!can_change_locked_flags(mnt
, mnt_flags
))
2834 * We're only checking whether the superblock is read-only not
2835 * changing it, so only take down_read(&sb->s_umount).
2837 down_read(&sb
->s_umount
);
2839 ret
= change_mount_ro_state(mnt
, mnt_flags
);
2841 set_mount_attributes(mnt
, mnt_flags
);
2842 unlock_mount_hash();
2843 up_read(&sb
->s_umount
);
2845 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2851 * change filesystem flags. dir should be a physical root of filesystem.
2852 * If you've mounted a non-root directory somewhere and want to do remount
2853 * on it - tough luck.
2855 static int do_remount(struct path
*path
, int ms_flags
, int sb_flags
,
2856 int mnt_flags
, void *data
)
2859 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2860 struct mount
*mnt
= real_mount(path
->mnt
);
2861 struct fs_context
*fc
;
2863 if (!check_mnt(mnt
))
2866 if (!path_mounted(path
))
2869 if (!can_change_locked_flags(mnt
, mnt_flags
))
2872 fc
= fs_context_for_reconfigure(path
->dentry
, sb_flags
, MS_RMT_MASK
);
2877 err
= parse_monolithic_mount_data(fc
, data
);
2879 down_write(&sb
->s_umount
);
2881 if (ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
)) {
2882 err
= reconfigure_super(fc
);
2885 set_mount_attributes(mnt
, mnt_flags
);
2886 unlock_mount_hash();
2889 up_write(&sb
->s_umount
);
2892 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2898 static inline int tree_contains_unbindable(struct mount
*mnt
)
2901 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2902 if (IS_MNT_UNBINDABLE(p
))
2909 * Check that there aren't references to earlier/same mount namespaces in the
2910 * specified subtree. Such references can act as pins for mount namespaces
2911 * that aren't checked by the mount-cycle checking code, thereby allowing
2912 * cycles to be made.
2914 static bool check_for_nsfs_mounts(struct mount
*subtree
)
2920 for (p
= subtree
; p
; p
= next_mnt(p
, subtree
))
2921 if (mnt_ns_loop(p
->mnt
.mnt_root
))
2926 unlock_mount_hash();
2930 static int do_set_group(struct path
*from_path
, struct path
*to_path
)
2932 struct mount
*from
, *to
;
2935 from
= real_mount(from_path
->mnt
);
2936 to
= real_mount(to_path
->mnt
);
2941 /* To and From must be mounted */
2942 if (!is_mounted(&from
->mnt
))
2944 if (!is_mounted(&to
->mnt
))
2948 /* We should be allowed to modify mount namespaces of both mounts */
2949 if (!ns_capable(from
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
))
2951 if (!ns_capable(to
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
))
2955 /* To and From paths should be mount roots */
2956 if (!path_mounted(from_path
))
2958 if (!path_mounted(to_path
))
2961 /* Setting sharing groups is only allowed across same superblock */
2962 if (from
->mnt
.mnt_sb
!= to
->mnt
.mnt_sb
)
2965 /* From mount root should be wider than To mount root */
2966 if (!is_subdir(to
->mnt
.mnt_root
, from
->mnt
.mnt_root
))
2969 /* From mount should not have locked children in place of To's root */
2970 if (has_locked_children(from
, to
->mnt
.mnt_root
))
2973 /* Setting sharing groups is only allowed on private mounts */
2974 if (IS_MNT_SHARED(to
) || IS_MNT_SLAVE(to
))
2977 /* From should not be private */
2978 if (!IS_MNT_SHARED(from
) && !IS_MNT_SLAVE(from
))
2981 if (IS_MNT_SLAVE(from
)) {
2982 struct mount
*m
= from
->mnt_master
;
2984 list_add(&to
->mnt_slave
, &m
->mnt_slave_list
);
2988 if (IS_MNT_SHARED(from
)) {
2989 to
->mnt_group_id
= from
->mnt_group_id
;
2990 list_add(&to
->mnt_share
, &from
->mnt_share
);
2993 unlock_mount_hash();
3003 * path_overmounted - check if path is overmounted
3004 * @path: path to check
3006 * Check if path is overmounted, i.e., if there's a mount on top of
3007 * @path->mnt with @path->dentry as mountpoint.
3009 * Context: This function expects namespace_lock() to be held.
3010 * Return: If path is overmounted true is returned, false if not.
3012 static inline bool path_overmounted(const struct path
*path
)
3015 if (unlikely(__lookup_mnt(path
->mnt
, path
->dentry
))) {
3024 * can_move_mount_beneath - check that we can mount beneath the top mount
3025 * @from: mount to mount beneath
3026 * @to: mount under which to mount
3028 * - Make sure that @to->dentry is actually the root of a mount under
3029 * which we can mount another mount.
3030 * - Make sure that nothing can be mounted beneath the caller's current
3031 * root or the rootfs of the namespace.
3032 * - Make sure that the caller can unmount the topmost mount ensuring
3033 * that the caller could reveal the underlying mountpoint.
3034 * - Ensure that nothing has been mounted on top of @from before we
3035 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3036 * - Prevent mounting beneath a mount if the propagation relationship
3037 * between the source mount, parent mount, and top mount would lead to
3038 * nonsensical mount trees.
3040 * Context: This function expects namespace_lock() to be held.
3041 * Return: On success 0, and on error a negative error code is returned.
3043 static int can_move_mount_beneath(const struct path
*from
,
3044 const struct path
*to
,
3045 const struct mountpoint
*mp
)
3047 struct mount
*mnt_from
= real_mount(from
->mnt
),
3048 *mnt_to
= real_mount(to
->mnt
),
3049 *parent_mnt_to
= mnt_to
->mnt_parent
;
3051 if (!mnt_has_parent(mnt_to
))
3054 if (!path_mounted(to
))
3057 if (IS_MNT_LOCKED(mnt_to
))
3060 /* Avoid creating shadow mounts during mount propagation. */
3061 if (path_overmounted(from
))
3065 * Mounting beneath the rootfs only makes sense when the
3066 * semantics of pivot_root(".", ".") are used.
3068 if (&mnt_to
->mnt
== current
->fs
->root
.mnt
)
3070 if (parent_mnt_to
== current
->nsproxy
->mnt_ns
->root
)
3073 for (struct mount
*p
= mnt_from
; mnt_has_parent(p
); p
= p
->mnt_parent
)
3078 * If the parent mount propagates to the child mount this would
3079 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3080 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3081 * defeats the whole purpose of mounting beneath another mount.
3083 if (propagation_would_overmount(parent_mnt_to
, mnt_to
, mp
))
3087 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3088 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3089 * Afterwards @mnt_from would be mounted on top of
3090 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3091 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3092 * already mounted on @mnt_from, @mnt_to would ultimately be
3093 * remounted on top of @c. Afterwards, @mnt_from would be
3094 * covered by a copy @c of @mnt_from and @c would be covered by
3095 * @mnt_from itself. This defeats the whole purpose of mounting
3096 * @mnt_from beneath @mnt_to.
3098 if (propagation_would_overmount(parent_mnt_to
, mnt_from
, mp
))
3104 static int do_move_mount(struct path
*old_path
, struct path
*new_path
,
3107 struct mnt_namespace
*ns
;
3110 struct mount
*parent
;
3111 struct mountpoint
*mp
, *old_mp
;
3114 enum mnt_tree_flags_t flags
= 0;
3116 mp
= do_lock_mount(new_path
, beneath
);
3120 old
= real_mount(old_path
->mnt
);
3121 p
= real_mount(new_path
->mnt
);
3122 parent
= old
->mnt_parent
;
3123 attached
= mnt_has_parent(old
);
3125 flags
|= MNT_TREE_MOVE
;
3126 old_mp
= old
->mnt_mp
;
3130 /* The mountpoint must be in our namespace. */
3134 /* The thing moved must be mounted... */
3135 if (!is_mounted(&old
->mnt
))
3138 /* ... and either ours or the root of anon namespace */
3139 if (!(attached
? check_mnt(old
) : is_anon_ns(ns
)))
3142 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
3145 if (!path_mounted(old_path
))
3148 if (d_is_dir(new_path
->dentry
) !=
3149 d_is_dir(old_path
->dentry
))
3152 * Don't move a mount residing in a shared parent.
3154 if (attached
&& IS_MNT_SHARED(parent
))
3158 err
= can_move_mount_beneath(old_path
, new_path
, mp
);
3164 flags
|= MNT_TREE_BENEATH
;
3168 * Don't move a mount tree containing unbindable mounts to a destination
3169 * mount which is shared.
3171 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
3174 if (!check_for_nsfs_mounts(old
))
3176 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
3180 err
= attach_recursive_mnt(old
, real_mount(new_path
->mnt
), mp
, flags
);
3184 /* if the mount is moved, it should no longer be expire
3186 list_del_init(&old
->mnt_expire
);
3188 put_mountpoint(old_mp
);
3193 mntput_no_expire(parent
);
3200 static int do_move_mount_old(struct path
*path
, const char *old_name
)
3202 struct path old_path
;
3205 if (!old_name
|| !*old_name
)
3208 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
3212 err
= do_move_mount(&old_path
, path
, false);
3213 path_put(&old_path
);
3218 * add a mount into a namespace's mount tree
3220 static int do_add_mount(struct mount
*newmnt
, struct mountpoint
*mp
,
3221 const struct path
*path
, int mnt_flags
)
3223 struct mount
*parent
= real_mount(path
->mnt
);
3225 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
3227 if (unlikely(!check_mnt(parent
))) {
3228 /* that's acceptable only for automounts done in private ns */
3229 if (!(mnt_flags
& MNT_SHRINKABLE
))
3231 /* ... and for those we'd better have mountpoint still alive */
3232 if (!parent
->mnt_ns
)
3236 /* Refuse the same filesystem on the same mount point */
3237 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&& path_mounted(path
))
3240 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
3243 newmnt
->mnt
.mnt_flags
= mnt_flags
;
3244 return graft_tree(newmnt
, parent
, mp
);
3247 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
);
3250 * Create a new mount using a superblock configuration and request it
3251 * be added to the namespace tree.
3253 static int do_new_mount_fc(struct fs_context
*fc
, struct path
*mountpoint
,
3254 unsigned int mnt_flags
)
3256 struct vfsmount
*mnt
;
3257 struct mountpoint
*mp
;
3258 struct super_block
*sb
= fc
->root
->d_sb
;
3261 error
= security_sb_kern_mount(sb
);
3262 if (!error
&& mount_too_revealing(sb
, &mnt_flags
))
3265 if (unlikely(error
)) {
3270 up_write(&sb
->s_umount
);
3272 mnt
= vfs_create_mount(fc
);
3274 return PTR_ERR(mnt
);
3276 mnt_warn_timestamp_expiry(mountpoint
, mnt
);
3278 mp
= lock_mount(mountpoint
);
3283 error
= do_add_mount(real_mount(mnt
), mp
, mountpoint
, mnt_flags
);
3291 * create a new mount for userspace and request it to be added into the
3294 static int do_new_mount(struct path
*path
, const char *fstype
, int sb_flags
,
3295 int mnt_flags
, const char *name
, void *data
)
3297 struct file_system_type
*type
;
3298 struct fs_context
*fc
;
3299 const char *subtype
= NULL
;
3305 type
= get_fs_type(fstype
);
3309 if (type
->fs_flags
& FS_HAS_SUBTYPE
) {
3310 subtype
= strchr(fstype
, '.');
3314 put_filesystem(type
);
3320 fc
= fs_context_for_mount(type
, sb_flags
);
3321 put_filesystem(type
);
3326 err
= vfs_parse_fs_string(fc
, "subtype",
3327 subtype
, strlen(subtype
));
3329 err
= vfs_parse_fs_string(fc
, "source", name
, strlen(name
));
3331 err
= parse_monolithic_mount_data(fc
, data
);
3332 if (!err
&& !mount_capable(fc
))
3335 err
= vfs_get_tree(fc
);
3337 err
= do_new_mount_fc(fc
, path
, mnt_flags
);
3343 int finish_automount(struct vfsmount
*m
, const struct path
*path
)
3345 struct dentry
*dentry
= path
->dentry
;
3346 struct mountpoint
*mp
;
3355 mnt
= real_mount(m
);
3356 /* The new mount record should have at least 2 refs to prevent it being
3357 * expired before we get a chance to add it
3359 BUG_ON(mnt_get_count(mnt
) < 2);
3361 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
3362 m
->mnt_root
== dentry
) {
3368 * we don't want to use lock_mount() - in this case finding something
3369 * that overmounts our mountpoint to be means "quitely drop what we've
3370 * got", not "try to mount it on top".
3372 inode_lock(dentry
->d_inode
);
3374 if (unlikely(cant_mount(dentry
))) {
3376 goto discard_locked
;
3378 if (path_overmounted(path
)) {
3380 goto discard_locked
;
3382 mp
= get_mountpoint(dentry
);
3385 goto discard_locked
;
3388 err
= do_add_mount(mnt
, mp
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
3397 inode_unlock(dentry
->d_inode
);
3399 /* remove m from any expiration list it may be on */
3400 if (!list_empty(&mnt
->mnt_expire
)) {
3402 list_del_init(&mnt
->mnt_expire
);
3411 * mnt_set_expiry - Put a mount on an expiration list
3412 * @mnt: The mount to list.
3413 * @expiry_list: The list to add the mount to.
3415 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
3419 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
3423 EXPORT_SYMBOL(mnt_set_expiry
);
3426 * process a list of expirable mountpoints with the intent of discarding any
3427 * mountpoints that aren't in use and haven't been touched since last we came
3430 void mark_mounts_for_expiry(struct list_head
*mounts
)
3432 struct mount
*mnt
, *next
;
3433 LIST_HEAD(graveyard
);
3435 if (list_empty(mounts
))
3441 /* extract from the expiration list every vfsmount that matches the
3442 * following criteria:
3443 * - only referenced by its parent vfsmount
3444 * - still marked for expiry (marked on the last call here; marks are
3445 * cleared by mntput())
3447 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
3448 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
3449 propagate_mount_busy(mnt
, 1))
3451 list_move(&mnt
->mnt_expire
, &graveyard
);
3453 while (!list_empty(&graveyard
)) {
3454 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
3455 touch_mnt_namespace(mnt
->mnt_ns
);
3456 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3458 unlock_mount_hash();
3462 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
3465 * Ripoff of 'select_parent()'
3467 * search the list of submounts for a given mountpoint, and move any
3468 * shrinkable submounts to the 'graveyard' list.
3470 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
3472 struct mount
*this_parent
= parent
;
3473 struct list_head
*next
;
3477 next
= this_parent
->mnt_mounts
.next
;
3479 while (next
!= &this_parent
->mnt_mounts
) {
3480 struct list_head
*tmp
= next
;
3481 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
3484 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
3487 * Descend a level if the d_mounts list is non-empty.
3489 if (!list_empty(&mnt
->mnt_mounts
)) {
3494 if (!propagate_mount_busy(mnt
, 1)) {
3495 list_move_tail(&mnt
->mnt_expire
, graveyard
);
3500 * All done at this level ... ascend and resume the search
3502 if (this_parent
!= parent
) {
3503 next
= this_parent
->mnt_child
.next
;
3504 this_parent
= this_parent
->mnt_parent
;
3511 * process a list of expirable mountpoints with the intent of discarding any
3512 * submounts of a specific parent mountpoint
3514 * mount_lock must be held for write
3516 static void shrink_submounts(struct mount
*mnt
)
3518 LIST_HEAD(graveyard
);
3521 /* extract submounts of 'mountpoint' from the expiration list */
3522 while (select_submounts(mnt
, &graveyard
)) {
3523 while (!list_empty(&graveyard
)) {
3524 m
= list_first_entry(&graveyard
, struct mount
,
3526 touch_mnt_namespace(m
->mnt_ns
);
3527 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3532 static void *copy_mount_options(const void __user
* data
)
3535 unsigned left
, offset
;
3540 copy
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3542 return ERR_PTR(-ENOMEM
);
3544 left
= copy_from_user(copy
, data
, PAGE_SIZE
);
3547 * Not all architectures have an exact copy_from_user(). Resort to
3550 offset
= PAGE_SIZE
- left
;
3553 if (get_user(c
, (const char __user
*)data
+ offset
))
3560 if (left
== PAGE_SIZE
) {
3562 return ERR_PTR(-EFAULT
);
3568 static char *copy_mount_string(const void __user
*data
)
3570 return data
? strndup_user(data
, PATH_MAX
) : NULL
;
3574 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3575 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3577 * data is a (void *) that can point to any structure up to
3578 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3579 * information (or be NULL).
3581 * Pre-0.97 versions of mount() didn't have a flags word.
3582 * When the flags word was introduced its top half was required
3583 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3584 * Therefore, if this magic number is present, it carries no information
3585 * and must be discarded.
3587 int path_mount(const char *dev_name
, struct path
*path
,
3588 const char *type_page
, unsigned long flags
, void *data_page
)
3590 unsigned int mnt_flags
= 0, sb_flags
;
3594 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
3595 flags
&= ~MS_MGC_MSK
;
3597 /* Basic sanity checks */
3599 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
3601 if (flags
& MS_NOUSER
)
3604 ret
= security_sb_mount(dev_name
, path
, type_page
, flags
, data_page
);
3609 if (flags
& SB_MANDLOCK
)
3612 /* Default to relatime unless overriden */
3613 if (!(flags
& MS_NOATIME
))
3614 mnt_flags
|= MNT_RELATIME
;
3616 /* Separate the per-mountpoint flags */
3617 if (flags
& MS_NOSUID
)
3618 mnt_flags
|= MNT_NOSUID
;
3619 if (flags
& MS_NODEV
)
3620 mnt_flags
|= MNT_NODEV
;
3621 if (flags
& MS_NOEXEC
)
3622 mnt_flags
|= MNT_NOEXEC
;
3623 if (flags
& MS_NOATIME
)
3624 mnt_flags
|= MNT_NOATIME
;
3625 if (flags
& MS_NODIRATIME
)
3626 mnt_flags
|= MNT_NODIRATIME
;
3627 if (flags
& MS_STRICTATIME
)
3628 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
3629 if (flags
& MS_RDONLY
)
3630 mnt_flags
|= MNT_READONLY
;
3631 if (flags
& MS_NOSYMFOLLOW
)
3632 mnt_flags
|= MNT_NOSYMFOLLOW
;
3634 /* The default atime for remount is preservation */
3635 if ((flags
& MS_REMOUNT
) &&
3636 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
3637 MS_STRICTATIME
)) == 0)) {
3638 mnt_flags
&= ~MNT_ATIME_MASK
;
3639 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_ATIME_MASK
;
3642 sb_flags
= flags
& (SB_RDONLY
|
3651 if ((flags
& (MS_REMOUNT
| MS_BIND
)) == (MS_REMOUNT
| MS_BIND
))
3652 return do_reconfigure_mnt(path
, mnt_flags
);
3653 if (flags
& MS_REMOUNT
)
3654 return do_remount(path
, flags
, sb_flags
, mnt_flags
, data_page
);
3655 if (flags
& MS_BIND
)
3656 return do_loopback(path
, dev_name
, flags
& MS_REC
);
3657 if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
3658 return do_change_type(path
, flags
);
3659 if (flags
& MS_MOVE
)
3660 return do_move_mount_old(path
, dev_name
);
3662 return do_new_mount(path
, type_page
, sb_flags
, mnt_flags
, dev_name
,
3666 long do_mount(const char *dev_name
, const char __user
*dir_name
,
3667 const char *type_page
, unsigned long flags
, void *data_page
)
3672 ret
= user_path_at(AT_FDCWD
, dir_name
, LOOKUP_FOLLOW
, &path
);
3675 ret
= path_mount(dev_name
, &path
, type_page
, flags
, data_page
);
3680 static struct ucounts
*inc_mnt_namespaces(struct user_namespace
*ns
)
3682 return inc_ucount(ns
, current_euid(), UCOUNT_MNT_NAMESPACES
);
3685 static void dec_mnt_namespaces(struct ucounts
*ucounts
)
3687 dec_ucount(ucounts
, UCOUNT_MNT_NAMESPACES
);
3690 static void free_mnt_ns(struct mnt_namespace
*ns
)
3692 if (!is_anon_ns(ns
))
3693 ns_free_inum(&ns
->ns
);
3694 dec_mnt_namespaces(ns
->ucounts
);
3695 put_user_ns(ns
->user_ns
);
3700 * Assign a sequence number so we can detect when we attempt to bind
3701 * mount a reference to an older mount namespace into the current
3702 * mount namespace, preventing reference counting loops. A 64bit
3703 * number incrementing at 10Ghz will take 12,427 years to wrap which
3704 * is effectively never, so we can ignore the possibility.
3706 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
3708 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
, bool anon
)
3710 struct mnt_namespace
*new_ns
;
3711 struct ucounts
*ucounts
;
3714 ucounts
= inc_mnt_namespaces(user_ns
);
3716 return ERR_PTR(-ENOSPC
);
3718 new_ns
= kzalloc(sizeof(struct mnt_namespace
), GFP_KERNEL_ACCOUNT
);
3720 dec_mnt_namespaces(ucounts
);
3721 return ERR_PTR(-ENOMEM
);
3724 ret
= ns_alloc_inum(&new_ns
->ns
);
3727 dec_mnt_namespaces(ucounts
);
3728 return ERR_PTR(ret
);
3731 new_ns
->ns
.ops
= &mntns_operations
;
3733 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
3734 refcount_set(&new_ns
->ns
.count
, 1);
3735 INIT_LIST_HEAD(&new_ns
->list
);
3736 init_waitqueue_head(&new_ns
->poll
);
3737 spin_lock_init(&new_ns
->ns_lock
);
3738 new_ns
->user_ns
= get_user_ns(user_ns
);
3739 new_ns
->ucounts
= ucounts
;
3744 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
3745 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
3747 struct mnt_namespace
*new_ns
;
3748 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
3749 struct mount
*p
, *q
;
3756 if (likely(!(flags
& CLONE_NEWNS
))) {
3763 new_ns
= alloc_mnt_ns(user_ns
, false);
3768 /* First pass: copy the tree topology */
3769 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
3770 if (user_ns
!= ns
->user_ns
)
3771 copy_flags
|= CL_SHARED_TO_SLAVE
;
3772 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
3775 free_mnt_ns(new_ns
);
3776 return ERR_CAST(new);
3778 if (user_ns
!= ns
->user_ns
) {
3781 unlock_mount_hash();
3784 list_add_tail(&new_ns
->list
, &new->mnt_list
);
3787 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3788 * as belonging to new namespace. We have already acquired a private
3789 * fs_struct, so tsk->fs->lock is not needed.
3797 if (&p
->mnt
== new_fs
->root
.mnt
) {
3798 new_fs
->root
.mnt
= mntget(&q
->mnt
);
3801 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
3802 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
3806 p
= next_mnt(p
, old
);
3807 q
= next_mnt(q
, new);
3810 // an mntns binding we'd skipped?
3811 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
3812 p
= next_mnt(skip_mnt_tree(p
), old
);
3824 struct dentry
*mount_subtree(struct vfsmount
*m
, const char *name
)
3826 struct mount
*mnt
= real_mount(m
);
3827 struct mnt_namespace
*ns
;
3828 struct super_block
*s
;
3832 ns
= alloc_mnt_ns(&init_user_ns
, true);
3835 return ERR_CAST(ns
);
3840 list_add(&mnt
->mnt_list
, &ns
->list
);
3842 err
= vfs_path_lookup(m
->mnt_root
, m
,
3843 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
3848 return ERR_PTR(err
);
3850 /* trade a vfsmount reference for active sb one */
3851 s
= path
.mnt
->mnt_sb
;
3852 atomic_inc(&s
->s_active
);
3854 /* lock the sucker */
3855 down_write(&s
->s_umount
);
3856 /* ... and return the root of (sub)tree on it */
3859 EXPORT_SYMBOL(mount_subtree
);
3861 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
3862 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3869 kernel_type
= copy_mount_string(type
);
3870 ret
= PTR_ERR(kernel_type
);
3871 if (IS_ERR(kernel_type
))
3874 kernel_dev
= copy_mount_string(dev_name
);
3875 ret
= PTR_ERR(kernel_dev
);
3876 if (IS_ERR(kernel_dev
))
3879 options
= copy_mount_options(data
);
3880 ret
= PTR_ERR(options
);
3881 if (IS_ERR(options
))
3884 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
, options
);
3895 #define FSMOUNT_VALID_FLAGS \
3896 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3897 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3898 MOUNT_ATTR_NOSYMFOLLOW)
3900 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3902 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3903 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3905 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags
)
3907 unsigned int mnt_flags
= 0;
3909 if (attr_flags
& MOUNT_ATTR_RDONLY
)
3910 mnt_flags
|= MNT_READONLY
;
3911 if (attr_flags
& MOUNT_ATTR_NOSUID
)
3912 mnt_flags
|= MNT_NOSUID
;
3913 if (attr_flags
& MOUNT_ATTR_NODEV
)
3914 mnt_flags
|= MNT_NODEV
;
3915 if (attr_flags
& MOUNT_ATTR_NOEXEC
)
3916 mnt_flags
|= MNT_NOEXEC
;
3917 if (attr_flags
& MOUNT_ATTR_NODIRATIME
)
3918 mnt_flags
|= MNT_NODIRATIME
;
3919 if (attr_flags
& MOUNT_ATTR_NOSYMFOLLOW
)
3920 mnt_flags
|= MNT_NOSYMFOLLOW
;
3926 * Create a kernel mount representation for a new, prepared superblock
3927 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3929 SYSCALL_DEFINE3(fsmount
, int, fs_fd
, unsigned int, flags
,
3930 unsigned int, attr_flags
)
3932 struct mnt_namespace
*ns
;
3933 struct fs_context
*fc
;
3935 struct path newmount
;
3938 unsigned int mnt_flags
= 0;
3944 if ((flags
& ~(FSMOUNT_CLOEXEC
)) != 0)
3947 if (attr_flags
& ~FSMOUNT_VALID_FLAGS
)
3950 mnt_flags
= attr_flags_to_mnt_flags(attr_flags
);
3952 switch (attr_flags
& MOUNT_ATTR__ATIME
) {
3953 case MOUNT_ATTR_STRICTATIME
:
3955 case MOUNT_ATTR_NOATIME
:
3956 mnt_flags
|= MNT_NOATIME
;
3958 case MOUNT_ATTR_RELATIME
:
3959 mnt_flags
|= MNT_RELATIME
;
3970 if (f
.file
->f_op
!= &fscontext_fops
)
3973 fc
= f
.file
->private_data
;
3975 ret
= mutex_lock_interruptible(&fc
->uapi_mutex
);
3979 /* There must be a valid superblock or we can't mount it */
3985 if (mount_too_revealing(fc
->root
->d_sb
, &mnt_flags
)) {
3986 pr_warn("VFS: Mount too revealing\n");
3991 if (fc
->phase
!= FS_CONTEXT_AWAITING_MOUNT
)
3994 if (fc
->sb_flags
& SB_MANDLOCK
)
3997 newmount
.mnt
= vfs_create_mount(fc
);
3998 if (IS_ERR(newmount
.mnt
)) {
3999 ret
= PTR_ERR(newmount
.mnt
);
4002 newmount
.dentry
= dget(fc
->root
);
4003 newmount
.mnt
->mnt_flags
= mnt_flags
;
4005 /* We've done the mount bit - now move the file context into more or
4006 * less the same state as if we'd done an fspick(). We don't want to
4007 * do any memory allocation or anything like that at this point as we
4008 * don't want to have to handle any errors incurred.
4010 vfs_clean_context(fc
);
4012 ns
= alloc_mnt_ns(current
->nsproxy
->mnt_ns
->user_ns
, true);
4017 mnt
= real_mount(newmount
.mnt
);
4021 list_add(&mnt
->mnt_list
, &ns
->list
);
4022 mntget(newmount
.mnt
);
4024 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4025 * it, not just simply put it.
4027 file
= dentry_open(&newmount
, O_PATH
, fc
->cred
);
4029 dissolve_on_fput(newmount
.mnt
);
4030 ret
= PTR_ERR(file
);
4033 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
4035 ret
= get_unused_fd_flags((flags
& FSMOUNT_CLOEXEC
) ? O_CLOEXEC
: 0);
4037 fd_install(ret
, file
);
4042 path_put(&newmount
);
4044 mutex_unlock(&fc
->uapi_mutex
);
4051 * Move a mount from one place to another. In combination with
4052 * fsopen()/fsmount() this is used to install a new mount and in combination
4053 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4056 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4058 SYSCALL_DEFINE5(move_mount
,
4059 int, from_dfd
, const char __user
*, from_pathname
,
4060 int, to_dfd
, const char __user
*, to_pathname
,
4061 unsigned int, flags
)
4063 struct path from_path
, to_path
;
4064 unsigned int lflags
;
4070 if (flags
& ~MOVE_MOUNT__MASK
)
4073 if ((flags
& (MOVE_MOUNT_BENEATH
| MOVE_MOUNT_SET_GROUP
)) ==
4074 (MOVE_MOUNT_BENEATH
| MOVE_MOUNT_SET_GROUP
))
4077 /* If someone gives a pathname, they aren't permitted to move
4078 * from an fd that requires unmount as we can't get at the flag
4079 * to clear it afterwards.
4082 if (flags
& MOVE_MOUNT_F_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
4083 if (flags
& MOVE_MOUNT_F_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
4084 if (flags
& MOVE_MOUNT_F_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
4086 ret
= user_path_at(from_dfd
, from_pathname
, lflags
, &from_path
);
4091 if (flags
& MOVE_MOUNT_T_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
4092 if (flags
& MOVE_MOUNT_T_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
4093 if (flags
& MOVE_MOUNT_T_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
4095 ret
= user_path_at(to_dfd
, to_pathname
, lflags
, &to_path
);
4099 ret
= security_move_mount(&from_path
, &to_path
);
4103 if (flags
& MOVE_MOUNT_SET_GROUP
)
4104 ret
= do_set_group(&from_path
, &to_path
);
4106 ret
= do_move_mount(&from_path
, &to_path
,
4107 (flags
& MOVE_MOUNT_BENEATH
));
4112 path_put(&from_path
);
4117 * Return true if path is reachable from root
4119 * namespace_sem or mount_lock is held
4121 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
4122 const struct path
*root
)
4124 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
4125 dentry
= mnt
->mnt_mountpoint
;
4126 mnt
= mnt
->mnt_parent
;
4128 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
4131 bool path_is_under(const struct path
*path1
, const struct path
*path2
)
4134 read_seqlock_excl(&mount_lock
);
4135 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
4136 read_sequnlock_excl(&mount_lock
);
4139 EXPORT_SYMBOL(path_is_under
);
4142 * pivot_root Semantics:
4143 * Moves the root file system of the current process to the directory put_old,
4144 * makes new_root as the new root file system of the current process, and sets
4145 * root/cwd of all processes which had them on the current root to new_root.
4148 * The new_root and put_old must be directories, and must not be on the
4149 * same file system as the current process root. The put_old must be
4150 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4151 * pointed to by put_old must yield the same directory as new_root. No other
4152 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4154 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
4155 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4156 * in this situation.
4159 * - we don't move root/cwd if they are not at the root (reason: if something
4160 * cared enough to change them, it's probably wrong to force them elsewhere)
4161 * - it's okay to pick a root that isn't the root of a file system, e.g.
4162 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4163 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4166 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
4167 const char __user
*, put_old
)
4169 struct path
new, old
, root
;
4170 struct mount
*new_mnt
, *root_mnt
, *old_mnt
, *root_parent
, *ex_parent
;
4171 struct mountpoint
*old_mp
, *root_mp
;
4177 error
= user_path_at(AT_FDCWD
, new_root
,
4178 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &new);
4182 error
= user_path_at(AT_FDCWD
, put_old
,
4183 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old
);
4187 error
= security_sb_pivotroot(&old
, &new);
4191 get_fs_root(current
->fs
, &root
);
4192 old_mp
= lock_mount(&old
);
4193 error
= PTR_ERR(old_mp
);
4198 new_mnt
= real_mount(new.mnt
);
4199 root_mnt
= real_mount(root
.mnt
);
4200 old_mnt
= real_mount(old
.mnt
);
4201 ex_parent
= new_mnt
->mnt_parent
;
4202 root_parent
= root_mnt
->mnt_parent
;
4203 if (IS_MNT_SHARED(old_mnt
) ||
4204 IS_MNT_SHARED(ex_parent
) ||
4205 IS_MNT_SHARED(root_parent
))
4207 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
4209 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
4212 if (d_unlinked(new.dentry
))
4215 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
4216 goto out4
; /* loop, on the same file system */
4218 if (!path_mounted(&root
))
4219 goto out4
; /* not a mountpoint */
4220 if (!mnt_has_parent(root_mnt
))
4221 goto out4
; /* not attached */
4222 if (!path_mounted(&new))
4223 goto out4
; /* not a mountpoint */
4224 if (!mnt_has_parent(new_mnt
))
4225 goto out4
; /* not attached */
4226 /* make sure we can reach put_old from new_root */
4227 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
4229 /* make certain new is below the root */
4230 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
4233 umount_mnt(new_mnt
);
4234 root_mp
= unhash_mnt(root_mnt
); /* we'll need its mountpoint */
4235 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
4236 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
4237 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
4239 /* mount old root on put_old */
4240 attach_mnt(root_mnt
, old_mnt
, old_mp
, false);
4241 /* mount new_root on / */
4242 attach_mnt(new_mnt
, root_parent
, root_mp
, false);
4243 mnt_add_count(root_parent
, -1);
4244 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
4245 /* A moved mount should not expire automatically */
4246 list_del_init(&new_mnt
->mnt_expire
);
4247 put_mountpoint(root_mp
);
4248 unlock_mount_hash();
4249 chroot_fs_refs(&root
, &new);
4252 unlock_mount(old_mp
);
4254 mntput_no_expire(ex_parent
);
4265 static unsigned int recalc_flags(struct mount_kattr
*kattr
, struct mount
*mnt
)
4267 unsigned int flags
= mnt
->mnt
.mnt_flags
;
4269 /* flags to clear */
4270 flags
&= ~kattr
->attr_clr
;
4271 /* flags to raise */
4272 flags
|= kattr
->attr_set
;
4277 static int can_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
4279 struct vfsmount
*m
= &mnt
->mnt
;
4280 struct user_namespace
*fs_userns
= m
->mnt_sb
->s_user_ns
;
4282 if (!kattr
->mnt_idmap
)
4286 * Creating an idmapped mount with the filesystem wide idmapping
4287 * doesn't make sense so block that. We don't allow mushy semantics.
4289 if (!check_fsmapping(kattr
->mnt_idmap
, m
->mnt_sb
))
4293 * Once a mount has been idmapped we don't allow it to change its
4294 * mapping. It makes things simpler and callers can just create
4295 * another bind-mount they can idmap if they want to.
4297 if (is_idmapped_mnt(m
))
4300 /* The underlying filesystem doesn't support idmapped mounts yet. */
4301 if (!(m
->mnt_sb
->s_type
->fs_flags
& FS_ALLOW_IDMAP
))
4304 /* We're not controlling the superblock. */
4305 if (!ns_capable(fs_userns
, CAP_SYS_ADMIN
))
4308 /* Mount has already been visible in the filesystem hierarchy. */
4309 if (!is_anon_ns(mnt
->mnt_ns
))
4316 * mnt_allow_writers() - check whether the attribute change allows writers
4317 * @kattr: the new mount attributes
4318 * @mnt: the mount to which @kattr will be applied
4320 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4322 * Return: true if writers need to be held, false if not
4324 static inline bool mnt_allow_writers(const struct mount_kattr
*kattr
,
4325 const struct mount
*mnt
)
4327 return (!(kattr
->attr_set
& MNT_READONLY
) ||
4328 (mnt
->mnt
.mnt_flags
& MNT_READONLY
)) &&
4332 static int mount_setattr_prepare(struct mount_kattr
*kattr
, struct mount
*mnt
)
4337 for (m
= mnt
; m
; m
= next_mnt(m
, mnt
)) {
4338 if (!can_change_locked_flags(m
, recalc_flags(kattr
, m
))) {
4343 err
= can_idmap_mount(kattr
, m
);
4347 if (!mnt_allow_writers(kattr
, m
)) {
4348 err
= mnt_hold_writers(m
);
4353 if (!kattr
->recurse
)
4361 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4362 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4363 * mounts and needs to take care to include the first mount.
4365 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
4366 /* If we had to hold writers unblock them. */
4367 if (p
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
4368 mnt_unhold_writers(p
);
4371 * We're done once the first mount we changed got
4372 * MNT_WRITE_HOLD unset.
4381 static void do_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
4383 if (!kattr
->mnt_idmap
)
4387 * Pairs with smp_load_acquire() in mnt_idmap().
4389 * Since we only allow a mount to change the idmapping once and
4390 * verified this in can_idmap_mount() we know that the mount has
4391 * @nop_mnt_idmap attached to it. So there's no need to drop any
4394 smp_store_release(&mnt
->mnt
.mnt_idmap
, mnt_idmap_get(kattr
->mnt_idmap
));
4397 static void mount_setattr_commit(struct mount_kattr
*kattr
, struct mount
*mnt
)
4401 for (m
= mnt
; m
; m
= next_mnt(m
, mnt
)) {
4404 do_idmap_mount(kattr
, m
);
4405 flags
= recalc_flags(kattr
, m
);
4406 WRITE_ONCE(m
->mnt
.mnt_flags
, flags
);
4408 /* If we had to hold writers unblock them. */
4409 if (m
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
4410 mnt_unhold_writers(m
);
4412 if (kattr
->propagation
)
4413 change_mnt_propagation(m
, kattr
->propagation
);
4414 if (!kattr
->recurse
)
4417 touch_mnt_namespace(mnt
->mnt_ns
);
4420 static int do_mount_setattr(struct path
*path
, struct mount_kattr
*kattr
)
4422 struct mount
*mnt
= real_mount(path
->mnt
);
4425 if (!path_mounted(path
))
4428 if (kattr
->mnt_userns
) {
4429 struct mnt_idmap
*mnt_idmap
;
4431 mnt_idmap
= alloc_mnt_idmap(kattr
->mnt_userns
);
4432 if (IS_ERR(mnt_idmap
))
4433 return PTR_ERR(mnt_idmap
);
4434 kattr
->mnt_idmap
= mnt_idmap
;
4437 if (kattr
->propagation
) {
4439 * Only take namespace_lock() if we're actually changing
4443 if (kattr
->propagation
== MS_SHARED
) {
4444 err
= invent_group_ids(mnt
, kattr
->recurse
);
4455 /* Ensure that this isn't anything purely vfs internal. */
4456 if (!is_mounted(&mnt
->mnt
))
4460 * If this is an attached mount make sure it's located in the callers
4461 * mount namespace. If it's not don't let the caller interact with it.
4462 * If this is a detached mount make sure it has an anonymous mount
4463 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4465 if (!(mnt_has_parent(mnt
) ? check_mnt(mnt
) : is_anon_ns(mnt
->mnt_ns
)))
4469 * First, we get the mount tree in a shape where we can change mount
4470 * properties without failure. If we succeeded to do so we commit all
4471 * changes and if we failed we clean up.
4473 err
= mount_setattr_prepare(kattr
, mnt
);
4475 mount_setattr_commit(kattr
, mnt
);
4478 unlock_mount_hash();
4480 if (kattr
->propagation
) {
4482 cleanup_group_ids(mnt
, NULL
);
4489 static int build_mount_idmapped(const struct mount_attr
*attr
, size_t usize
,
4490 struct mount_kattr
*kattr
, unsigned int flags
)
4493 struct ns_common
*ns
;
4494 struct user_namespace
*mnt_userns
;
4497 if (!((attr
->attr_set
| attr
->attr_clr
) & MOUNT_ATTR_IDMAP
))
4501 * We currently do not support clearing an idmapped mount. If this ever
4502 * is a use-case we can revisit this but for now let's keep it simple
4505 if (attr
->attr_clr
& MOUNT_ATTR_IDMAP
)
4508 if (attr
->userns_fd
> INT_MAX
)
4511 f
= fdget(attr
->userns_fd
);
4515 if (!proc_ns_file(f
.file
)) {
4520 ns
= get_proc_ns(file_inode(f
.file
));
4521 if (ns
->ops
->type
!= CLONE_NEWUSER
) {
4527 * The initial idmapping cannot be used to create an idmapped
4528 * mount. We use the initial idmapping as an indicator of a mount
4529 * that is not idmapped. It can simply be passed into helpers that
4530 * are aware of idmapped mounts as a convenient shortcut. A user
4531 * can just create a dedicated identity mapping to achieve the same
4534 mnt_userns
= container_of(ns
, struct user_namespace
, ns
);
4535 if (mnt_userns
== &init_user_ns
) {
4540 /* We're not controlling the target namespace. */
4541 if (!ns_capable(mnt_userns
, CAP_SYS_ADMIN
)) {
4546 kattr
->mnt_userns
= get_user_ns(mnt_userns
);
4553 static int build_mount_kattr(const struct mount_attr
*attr
, size_t usize
,
4554 struct mount_kattr
*kattr
, unsigned int flags
)
4556 unsigned int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
4558 if (flags
& AT_NO_AUTOMOUNT
)
4559 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
4560 if (flags
& AT_SYMLINK_NOFOLLOW
)
4561 lookup_flags
&= ~LOOKUP_FOLLOW
;
4562 if (flags
& AT_EMPTY_PATH
)
4563 lookup_flags
|= LOOKUP_EMPTY
;
4565 *kattr
= (struct mount_kattr
) {
4566 .lookup_flags
= lookup_flags
,
4567 .recurse
= !!(flags
& AT_RECURSIVE
),
4570 if (attr
->propagation
& ~MOUNT_SETATTR_PROPAGATION_FLAGS
)
4572 if (hweight32(attr
->propagation
& MOUNT_SETATTR_PROPAGATION_FLAGS
) > 1)
4574 kattr
->propagation
= attr
->propagation
;
4576 if ((attr
->attr_set
| attr
->attr_clr
) & ~MOUNT_SETATTR_VALID_FLAGS
)
4579 kattr
->attr_set
= attr_flags_to_mnt_flags(attr
->attr_set
);
4580 kattr
->attr_clr
= attr_flags_to_mnt_flags(attr
->attr_clr
);
4583 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4584 * users wanting to transition to a different atime setting cannot
4585 * simply specify the atime setting in @attr_set, but must also
4586 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4587 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4588 * @attr_clr and that @attr_set can't have any atime bits set if
4589 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4591 if (attr
->attr_clr
& MOUNT_ATTR__ATIME
) {
4592 if ((attr
->attr_clr
& MOUNT_ATTR__ATIME
) != MOUNT_ATTR__ATIME
)
4596 * Clear all previous time settings as they are mutually
4599 kattr
->attr_clr
|= MNT_RELATIME
| MNT_NOATIME
;
4600 switch (attr
->attr_set
& MOUNT_ATTR__ATIME
) {
4601 case MOUNT_ATTR_RELATIME
:
4602 kattr
->attr_set
|= MNT_RELATIME
;
4604 case MOUNT_ATTR_NOATIME
:
4605 kattr
->attr_set
|= MNT_NOATIME
;
4607 case MOUNT_ATTR_STRICTATIME
:
4613 if (attr
->attr_set
& MOUNT_ATTR__ATIME
)
4617 return build_mount_idmapped(attr
, usize
, kattr
, flags
);
4620 static void finish_mount_kattr(struct mount_kattr
*kattr
)
4622 put_user_ns(kattr
->mnt_userns
);
4623 kattr
->mnt_userns
= NULL
;
4625 if (kattr
->mnt_idmap
)
4626 mnt_idmap_put(kattr
->mnt_idmap
);
4629 SYSCALL_DEFINE5(mount_setattr
, int, dfd
, const char __user
*, path
,
4630 unsigned int, flags
, struct mount_attr __user
*, uattr
,
4635 struct mount_attr attr
;
4636 struct mount_kattr kattr
;
4638 BUILD_BUG_ON(sizeof(struct mount_attr
) != MOUNT_ATTR_SIZE_VER0
);
4640 if (flags
& ~(AT_EMPTY_PATH
|
4642 AT_SYMLINK_NOFOLLOW
|
4646 if (unlikely(usize
> PAGE_SIZE
))
4648 if (unlikely(usize
< MOUNT_ATTR_SIZE_VER0
))
4654 err
= copy_struct_from_user(&attr
, sizeof(attr
), uattr
, usize
);
4658 /* Don't bother walking through the mounts if this is a nop. */
4659 if (attr
.attr_set
== 0 &&
4660 attr
.attr_clr
== 0 &&
4661 attr
.propagation
== 0)
4664 err
= build_mount_kattr(&attr
, usize
, &kattr
, flags
);
4668 err
= user_path_at(dfd
, path
, kattr
.lookup_flags
, &target
);
4670 err
= do_mount_setattr(&target
, &kattr
);
4673 finish_mount_kattr(&kattr
);
4677 static void __init
init_mount_tree(void)
4679 struct vfsmount
*mnt
;
4681 struct mnt_namespace
*ns
;
4684 mnt
= vfs_kern_mount(&rootfs_fs_type
, 0, "rootfs", NULL
);
4686 panic("Can't create rootfs");
4688 ns
= alloc_mnt_ns(&init_user_ns
, false);
4690 panic("Can't allocate initial namespace");
4691 m
= real_mount(mnt
);
4695 list_add(&m
->mnt_list
, &ns
->list
);
4696 init_task
.nsproxy
->mnt_ns
= ns
;
4700 root
.dentry
= mnt
->mnt_root
;
4701 mnt
->mnt_flags
|= MNT_LOCKED
;
4703 set_fs_pwd(current
->fs
, &root
);
4704 set_fs_root(current
->fs
, &root
);
4707 void __init
mnt_init(void)
4711 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
4712 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
, NULL
);
4714 mount_hashtable
= alloc_large_system_hash("Mount-cache",
4715 sizeof(struct hlist_head
),
4718 &m_hash_shift
, &m_hash_mask
, 0, 0);
4719 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
4720 sizeof(struct hlist_head
),
4723 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
4725 if (!mount_hashtable
|| !mountpoint_hashtable
)
4726 panic("Failed to allocate mount hash table\n");
4732 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
4734 fs_kobj
= kobject_create_and_add("fs", NULL
);
4736 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
4742 void put_mnt_ns(struct mnt_namespace
*ns
)
4744 if (!refcount_dec_and_test(&ns
->ns
.count
))
4746 drop_collected_mounts(&ns
->root
->mnt
);
4750 struct vfsmount
*kern_mount(struct file_system_type
*type
)
4752 struct vfsmount
*mnt
;
4753 mnt
= vfs_kern_mount(type
, SB_KERNMOUNT
, type
->name
, NULL
);
4756 * it is a longterm mount, don't release mnt until
4757 * we unmount before file sys is unregistered
4759 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
4763 EXPORT_SYMBOL_GPL(kern_mount
);
4765 void kern_unmount(struct vfsmount
*mnt
)
4767 /* release long term mount so mount point can be released */
4769 mnt_make_shortterm(mnt
);
4770 synchronize_rcu(); /* yecchhh... */
4774 EXPORT_SYMBOL(kern_unmount
);
4776 void kern_unmount_array(struct vfsmount
*mnt
[], unsigned int num
)
4780 for (i
= 0; i
< num
; i
++)
4781 mnt_make_shortterm(mnt
[i
]);
4782 synchronize_rcu_expedited();
4783 for (i
= 0; i
< num
; i
++)
4786 EXPORT_SYMBOL(kern_unmount_array
);
4788 bool our_mnt(struct vfsmount
*mnt
)
4790 return check_mnt(real_mount(mnt
));
4793 bool current_chrooted(void)
4795 /* Does the current process have a non-standard root */
4796 struct path ns_root
;
4797 struct path fs_root
;
4800 /* Find the namespace root */
4801 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
4802 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
4804 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
4807 get_fs_root(current
->fs
, &fs_root
);
4809 chrooted
= !path_equal(&fs_root
, &ns_root
);
4817 static bool mnt_already_visible(struct mnt_namespace
*ns
,
4818 const struct super_block
*sb
,
4821 int new_flags
= *new_mnt_flags
;
4823 bool visible
= false;
4825 down_read(&namespace_sem
);
4827 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
4828 struct mount
*child
;
4831 if (mnt_is_cursor(mnt
))
4834 if (mnt
->mnt
.mnt_sb
->s_type
!= sb
->s_type
)
4837 /* This mount is not fully visible if it's root directory
4838 * is not the root directory of the filesystem.
4840 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
4843 /* A local view of the mount flags */
4844 mnt_flags
= mnt
->mnt
.mnt_flags
;
4846 /* Don't miss readonly hidden in the superblock flags */
4847 if (sb_rdonly(mnt
->mnt
.mnt_sb
))
4848 mnt_flags
|= MNT_LOCK_READONLY
;
4850 /* Verify the mount flags are equal to or more permissive
4851 * than the proposed new mount.
4853 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
4854 !(new_flags
& MNT_READONLY
))
4856 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
4857 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
4860 /* This mount is not fully visible if there are any
4861 * locked child mounts that cover anything except for
4862 * empty directories.
4864 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
4865 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
4866 /* Only worry about locked mounts */
4867 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
4869 /* Is the directory permanetly empty? */
4870 if (!is_empty_dir_inode(inode
))
4873 /* Preserve the locked attributes */
4874 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
4882 up_read(&namespace_sem
);
4886 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
)
4888 const unsigned long required_iflags
= SB_I_NOEXEC
| SB_I_NODEV
;
4889 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
4890 unsigned long s_iflags
;
4892 if (ns
->user_ns
== &init_user_ns
)
4895 /* Can this filesystem be too revealing? */
4896 s_iflags
= sb
->s_iflags
;
4897 if (!(s_iflags
& SB_I_USERNS_VISIBLE
))
4900 if ((s_iflags
& required_iflags
) != required_iflags
) {
4901 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4906 return !mnt_already_visible(ns
, sb
, new_mnt_flags
);
4909 bool mnt_may_suid(struct vfsmount
*mnt
)
4912 * Foreign mounts (accessed via fchdir or through /proc
4913 * symlinks) are always treated as if they are nosuid. This
4914 * prevents namespaces from trusting potentially unsafe
4915 * suid/sgid bits, file caps, or security labels that originate
4916 * in other namespaces.
4918 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
4919 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
4922 static struct ns_common
*mntns_get(struct task_struct
*task
)
4924 struct ns_common
*ns
= NULL
;
4925 struct nsproxy
*nsproxy
;
4928 nsproxy
= task
->nsproxy
;
4930 ns
= &nsproxy
->mnt_ns
->ns
;
4931 get_mnt_ns(to_mnt_ns(ns
));
4938 static void mntns_put(struct ns_common
*ns
)
4940 put_mnt_ns(to_mnt_ns(ns
));
4943 static int mntns_install(struct nsset
*nsset
, struct ns_common
*ns
)
4945 struct nsproxy
*nsproxy
= nsset
->nsproxy
;
4946 struct fs_struct
*fs
= nsset
->fs
;
4947 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
), *old_mnt_ns
;
4948 struct user_namespace
*user_ns
= nsset
->cred
->user_ns
;
4952 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
4953 !ns_capable(user_ns
, CAP_SYS_CHROOT
) ||
4954 !ns_capable(user_ns
, CAP_SYS_ADMIN
))
4957 if (is_anon_ns(mnt_ns
))
4964 old_mnt_ns
= nsproxy
->mnt_ns
;
4965 nsproxy
->mnt_ns
= mnt_ns
;
4968 err
= vfs_path_lookup(mnt_ns
->root
->mnt
.mnt_root
, &mnt_ns
->root
->mnt
,
4969 "/", LOOKUP_DOWN
, &root
);
4971 /* revert to old namespace */
4972 nsproxy
->mnt_ns
= old_mnt_ns
;
4977 put_mnt_ns(old_mnt_ns
);
4979 /* Update the pwd and root */
4980 set_fs_pwd(fs
, &root
);
4981 set_fs_root(fs
, &root
);
4987 static struct user_namespace
*mntns_owner(struct ns_common
*ns
)
4989 return to_mnt_ns(ns
)->user_ns
;
4992 const struct proc_ns_operations mntns_operations
= {
4994 .type
= CLONE_NEWNS
,
4997 .install
= mntns_install
,
4998 .owner
= mntns_owner
,
5001 #ifdef CONFIG_SYSCTL
5002 static struct ctl_table fs_namespace_sysctls
[] = {
5004 .procname
= "mount-max",
5005 .data
= &sysctl_mount_max
,
5006 .maxlen
= sizeof(unsigned int),
5008 .proc_handler
= proc_dointvec_minmax
,
5009 .extra1
= SYSCTL_ONE
,
5014 static int __init
init_fs_namespace_sysctls(void)
5016 register_sysctl_init("fs", fs_namespace_sysctls
);
5019 fs_initcall(init_fs_namespace_sysctls
);
5021 #endif /* CONFIG_SYSCTL */