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>
38 /* Maximum number of mounts in a mount namespace */
39 unsigned int sysctl_mount_max __read_mostly
= 100000;
41 static unsigned int m_hash_mask __read_mostly
;
42 static unsigned int m_hash_shift __read_mostly
;
43 static unsigned int mp_hash_mask __read_mostly
;
44 static unsigned int mp_hash_shift __read_mostly
;
46 static __initdata
unsigned long mhash_entries
;
47 static int __init
set_mhash_entries(char *str
)
51 mhash_entries
= simple_strtoul(str
, &str
, 0);
54 __setup("mhash_entries=", set_mhash_entries
);
56 static __initdata
unsigned long mphash_entries
;
57 static int __init
set_mphash_entries(char *str
)
61 mphash_entries
= simple_strtoul(str
, &str
, 0);
64 __setup("mphash_entries=", set_mphash_entries
);
67 static DEFINE_IDA(mnt_id_ida
);
68 static DEFINE_IDA(mnt_group_ida
);
70 static struct hlist_head
*mount_hashtable __read_mostly
;
71 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
72 static struct kmem_cache
*mnt_cache __read_mostly
;
73 static DECLARE_RWSEM(namespace_sem
);
74 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
75 static LIST_HEAD(ex_mountpoints
); /* protected by namespace_sem */
78 unsigned int attr_set
;
79 unsigned int attr_clr
;
80 unsigned int propagation
;
81 unsigned int lookup_flags
;
83 struct user_namespace
*mnt_userns
;
87 struct kobject
*fs_kobj
;
88 EXPORT_SYMBOL_GPL(fs_kobj
);
91 * vfsmount lock may be taken for read to prevent changes to the
92 * vfsmount hash, ie. during mountpoint lookups or walking back
95 * It should be taken for write in all cases where the vfsmount
96 * tree or hash is modified or when a vfsmount structure is modified.
98 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
100 static inline void lock_mount_hash(void)
102 write_seqlock(&mount_lock
);
105 static inline void unlock_mount_hash(void)
107 write_sequnlock(&mount_lock
);
110 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
112 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
113 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
114 tmp
= tmp
+ (tmp
>> m_hash_shift
);
115 return &mount_hashtable
[tmp
& m_hash_mask
];
118 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
120 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
121 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
122 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
125 static int mnt_alloc_id(struct mount
*mnt
)
127 int res
= ida_alloc(&mnt_id_ida
, GFP_KERNEL
);
135 static void mnt_free_id(struct mount
*mnt
)
137 ida_free(&mnt_id_ida
, mnt
->mnt_id
);
141 * Allocate a new peer group ID
143 static int mnt_alloc_group_id(struct mount
*mnt
)
145 int res
= ida_alloc_min(&mnt_group_ida
, 1, GFP_KERNEL
);
149 mnt
->mnt_group_id
= res
;
154 * Release a peer group ID
156 void mnt_release_group_id(struct mount
*mnt
)
158 ida_free(&mnt_group_ida
, mnt
->mnt_group_id
);
159 mnt
->mnt_group_id
= 0;
163 * vfsmount lock must be held for read
165 static inline void mnt_add_count(struct mount
*mnt
, int n
)
168 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
177 * vfsmount lock must be held for write
179 int mnt_get_count(struct mount
*mnt
)
185 for_each_possible_cpu(cpu
) {
186 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
191 return mnt
->mnt_count
;
195 static struct mount
*alloc_vfsmnt(const char *name
)
197 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
201 err
= mnt_alloc_id(mnt
);
206 mnt
->mnt_devname
= kstrdup_const(name
, GFP_KERNEL
);
207 if (!mnt
->mnt_devname
)
212 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
214 goto out_free_devname
;
216 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
219 mnt
->mnt_writers
= 0;
222 INIT_HLIST_NODE(&mnt
->mnt_hash
);
223 INIT_LIST_HEAD(&mnt
->mnt_child
);
224 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
225 INIT_LIST_HEAD(&mnt
->mnt_list
);
226 INIT_LIST_HEAD(&mnt
->mnt_expire
);
227 INIT_LIST_HEAD(&mnt
->mnt_share
);
228 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
229 INIT_LIST_HEAD(&mnt
->mnt_slave
);
230 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
231 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
232 INIT_HLIST_HEAD(&mnt
->mnt_stuck_children
);
233 mnt
->mnt
.mnt_userns
= &init_user_ns
;
239 kfree_const(mnt
->mnt_devname
);
244 kmem_cache_free(mnt_cache
, mnt
);
249 * Most r/o checks on a fs are for operations that take
250 * discrete amounts of time, like a write() or unlink().
251 * We must keep track of when those operations start
252 * (for permission checks) and when they end, so that
253 * we can determine when writes are able to occur to
257 * __mnt_is_readonly: check whether a mount is read-only
258 * @mnt: the mount to check for its write status
260 * This shouldn't be used directly ouside of the VFS.
261 * It does not guarantee that the filesystem will stay
262 * r/w, just that it is right *now*. This can not and
263 * should not be used in place of IS_RDONLY(inode).
264 * mnt_want/drop_write() will _keep_ the filesystem
267 bool __mnt_is_readonly(struct vfsmount
*mnt
)
269 return (mnt
->mnt_flags
& MNT_READONLY
) || sb_rdonly(mnt
->mnt_sb
);
271 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
273 static inline void mnt_inc_writers(struct mount
*mnt
)
276 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
282 static inline void mnt_dec_writers(struct mount
*mnt
)
285 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
291 static unsigned int mnt_get_writers(struct mount
*mnt
)
294 unsigned int count
= 0;
297 for_each_possible_cpu(cpu
) {
298 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
303 return mnt
->mnt_writers
;
307 static int mnt_is_readonly(struct vfsmount
*mnt
)
309 if (mnt
->mnt_sb
->s_readonly_remount
)
311 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
313 return __mnt_is_readonly(mnt
);
317 * Most r/o & frozen checks on a fs are for operations that take discrete
318 * amounts of time, like a write() or unlink(). We must keep track of when
319 * those operations start (for permission checks) and when they end, so that we
320 * can determine when writes are able to occur to a filesystem.
323 * __mnt_want_write - get write access to a mount without freeze protection
324 * @m: the mount on which to take a write
326 * This tells the low-level filesystem that a write is about to be performed to
327 * it, and makes sure that writes are allowed (mnt it read-write) before
328 * returning success. This operation does not protect against filesystem being
329 * frozen. When the write operation is finished, __mnt_drop_write() must be
330 * called. This is effectively a refcount.
332 int __mnt_want_write(struct vfsmount
*m
)
334 struct mount
*mnt
= real_mount(m
);
338 mnt_inc_writers(mnt
);
340 * The store to mnt_inc_writers must be visible before we pass
341 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
342 * incremented count after it has set MNT_WRITE_HOLD.
345 while (READ_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
348 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
349 * be set to match its requirements. So we must not load that until
350 * MNT_WRITE_HOLD is cleared.
353 if (mnt_is_readonly(m
)) {
354 mnt_dec_writers(mnt
);
363 * mnt_want_write - get write access to a mount
364 * @m: the mount on which to take a write
366 * This tells the low-level filesystem that a write is about to be performed to
367 * it, and makes sure that writes are allowed (mount is read-write, filesystem
368 * is not frozen) before returning success. When the write operation is
369 * finished, mnt_drop_write() must be called. This is effectively a refcount.
371 int mnt_want_write(struct vfsmount
*m
)
375 sb_start_write(m
->mnt_sb
);
376 ret
= __mnt_want_write(m
);
378 sb_end_write(m
->mnt_sb
);
381 EXPORT_SYMBOL_GPL(mnt_want_write
);
384 * __mnt_want_write_file - get write access to a file's mount
385 * @file: the file who's mount on which to take a write
387 * This is like __mnt_want_write, but if the file is already open for writing it
388 * skips incrementing mnt_writers (since the open file already has a reference)
389 * and instead only does the check for emergency r/o remounts. This must be
390 * paired with __mnt_drop_write_file.
392 int __mnt_want_write_file(struct file
*file
)
394 if (file
->f_mode
& FMODE_WRITER
) {
396 * Superblock may have become readonly while there are still
397 * writable fd's, e.g. due to a fs error with errors=remount-ro
399 if (__mnt_is_readonly(file
->f_path
.mnt
))
403 return __mnt_want_write(file
->f_path
.mnt
);
407 * mnt_want_write_file - get write access to a file's mount
408 * @file: the file who's mount on which to take a write
410 * This is like mnt_want_write, but if the file is already open for writing it
411 * skips incrementing mnt_writers (since the open file already has a reference)
412 * and instead only does the freeze protection and the check for emergency r/o
413 * remounts. This must be paired with mnt_drop_write_file.
415 int mnt_want_write_file(struct file
*file
)
419 sb_start_write(file_inode(file
)->i_sb
);
420 ret
= __mnt_want_write_file(file
);
422 sb_end_write(file_inode(file
)->i_sb
);
425 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
428 * __mnt_drop_write - give up write access to a mount
429 * @mnt: the mount on which to give up write access
431 * Tells the low-level filesystem that we are done
432 * performing writes to it. Must be matched with
433 * __mnt_want_write() call above.
435 void __mnt_drop_write(struct vfsmount
*mnt
)
438 mnt_dec_writers(real_mount(mnt
));
443 * mnt_drop_write - give up write access to a mount
444 * @mnt: the mount on which to give up write access
446 * Tells the low-level filesystem that we are done performing writes to it and
447 * also allows filesystem to be frozen again. Must be matched with
448 * mnt_want_write() call above.
450 void mnt_drop_write(struct vfsmount
*mnt
)
452 __mnt_drop_write(mnt
);
453 sb_end_write(mnt
->mnt_sb
);
455 EXPORT_SYMBOL_GPL(mnt_drop_write
);
457 void __mnt_drop_write_file(struct file
*file
)
459 if (!(file
->f_mode
& FMODE_WRITER
))
460 __mnt_drop_write(file
->f_path
.mnt
);
463 void mnt_drop_write_file(struct file
*file
)
465 __mnt_drop_write_file(file
);
466 sb_end_write(file_inode(file
)->i_sb
);
468 EXPORT_SYMBOL(mnt_drop_write_file
);
470 static inline int mnt_hold_writers(struct mount
*mnt
)
472 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
474 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
475 * should be visible before we do.
480 * With writers on hold, if this value is zero, then there are
481 * definitely no active writers (although held writers may subsequently
482 * increment the count, they'll have to wait, and decrement it after
483 * seeing MNT_READONLY).
485 * It is OK to have counter incremented on one CPU and decremented on
486 * another: the sum will add up correctly. The danger would be when we
487 * sum up each counter, if we read a counter before it is incremented,
488 * but then read another CPU's count which it has been subsequently
489 * decremented from -- we would see more decrements than we should.
490 * MNT_WRITE_HOLD protects against this scenario, because
491 * mnt_want_write first increments count, then smp_mb, then spins on
492 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
493 * we're counting up here.
495 if (mnt_get_writers(mnt
) > 0)
501 static inline void mnt_unhold_writers(struct mount
*mnt
)
504 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
505 * that become unheld will see MNT_READONLY.
508 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
511 static int mnt_make_readonly(struct mount
*mnt
)
515 ret
= mnt_hold_writers(mnt
);
517 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
518 mnt_unhold_writers(mnt
);
522 int sb_prepare_remount_readonly(struct super_block
*sb
)
527 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
528 if (atomic_long_read(&sb
->s_remove_count
))
532 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
533 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
534 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
536 if (mnt_get_writers(mnt
) > 0) {
542 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
546 sb
->s_readonly_remount
= 1;
549 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
550 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
551 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
558 static void free_vfsmnt(struct mount
*mnt
)
560 struct user_namespace
*mnt_userns
;
562 mnt_userns
= mnt_user_ns(&mnt
->mnt
);
563 if (mnt_userns
!= &init_user_ns
)
564 put_user_ns(mnt_userns
);
565 kfree_const(mnt
->mnt_devname
);
567 free_percpu(mnt
->mnt_pcp
);
569 kmem_cache_free(mnt_cache
, mnt
);
572 static void delayed_free_vfsmnt(struct rcu_head
*head
)
574 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
577 /* call under rcu_read_lock */
578 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
581 if (read_seqretry(&mount_lock
, seq
))
585 mnt
= real_mount(bastard
);
586 mnt_add_count(mnt
, 1);
587 smp_mb(); // see mntput_no_expire()
588 if (likely(!read_seqretry(&mount_lock
, seq
)))
590 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
591 mnt_add_count(mnt
, -1);
595 if (unlikely(bastard
->mnt_flags
& MNT_DOOMED
)) {
596 mnt_add_count(mnt
, -1);
601 /* caller will mntput() */
605 /* call under rcu_read_lock */
606 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
608 int res
= __legitimize_mnt(bastard
, seq
);
611 if (unlikely(res
< 0)) {
620 * find the first mount at @dentry on vfsmount @mnt.
621 * call under rcu_read_lock()
623 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
625 struct hlist_head
*head
= m_hash(mnt
, dentry
);
628 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
629 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
635 * lookup_mnt - Return the first child mount mounted at path
637 * "First" means first mounted chronologically. If you create the
640 * mount /dev/sda1 /mnt
641 * mount /dev/sda2 /mnt
642 * mount /dev/sda3 /mnt
644 * Then lookup_mnt() on the base /mnt dentry in the root mount will
645 * return successively the root dentry and vfsmount of /dev/sda1, then
646 * /dev/sda2, then /dev/sda3, then NULL.
648 * lookup_mnt takes a reference to the found vfsmount.
650 struct vfsmount
*lookup_mnt(const struct path
*path
)
652 struct mount
*child_mnt
;
658 seq
= read_seqbegin(&mount_lock
);
659 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
660 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
661 } while (!legitimize_mnt(m
, seq
));
666 static inline void lock_ns_list(struct mnt_namespace
*ns
)
668 spin_lock(&ns
->ns_lock
);
671 static inline void unlock_ns_list(struct mnt_namespace
*ns
)
673 spin_unlock(&ns
->ns_lock
);
676 static inline bool mnt_is_cursor(struct mount
*mnt
)
678 return mnt
->mnt
.mnt_flags
& MNT_CURSOR
;
682 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
683 * current mount namespace.
685 * The common case is dentries are not mountpoints at all and that
686 * test is handled inline. For the slow case when we are actually
687 * dealing with a mountpoint of some kind, walk through all of the
688 * mounts in the current mount namespace and test to see if the dentry
691 * The mount_hashtable is not usable in the context because we
692 * need to identify all mounts that may be in the current mount
693 * namespace not just a mount that happens to have some specified
696 bool __is_local_mountpoint(struct dentry
*dentry
)
698 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
700 bool is_covered
= false;
702 down_read(&namespace_sem
);
704 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
705 if (mnt_is_cursor(mnt
))
707 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
712 up_read(&namespace_sem
);
717 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
719 struct hlist_head
*chain
= mp_hash(dentry
);
720 struct mountpoint
*mp
;
722 hlist_for_each_entry(mp
, chain
, m_hash
) {
723 if (mp
->m_dentry
== dentry
) {
731 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
733 struct mountpoint
*mp
, *new = NULL
;
736 if (d_mountpoint(dentry
)) {
737 /* might be worth a WARN_ON() */
738 if (d_unlinked(dentry
))
739 return ERR_PTR(-ENOENT
);
741 read_seqlock_excl(&mount_lock
);
742 mp
= lookup_mountpoint(dentry
);
743 read_sequnlock_excl(&mount_lock
);
749 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
751 return ERR_PTR(-ENOMEM
);
754 /* Exactly one processes may set d_mounted */
755 ret
= d_set_mounted(dentry
);
757 /* Someone else set d_mounted? */
761 /* The dentry is not available as a mountpoint? */
766 /* Add the new mountpoint to the hash table */
767 read_seqlock_excl(&mount_lock
);
768 new->m_dentry
= dget(dentry
);
770 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
771 INIT_HLIST_HEAD(&new->m_list
);
772 read_sequnlock_excl(&mount_lock
);
782 * vfsmount lock must be held. Additionally, the caller is responsible
783 * for serializing calls for given disposal list.
785 static void __put_mountpoint(struct mountpoint
*mp
, struct list_head
*list
)
787 if (!--mp
->m_count
) {
788 struct dentry
*dentry
= mp
->m_dentry
;
789 BUG_ON(!hlist_empty(&mp
->m_list
));
790 spin_lock(&dentry
->d_lock
);
791 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
792 spin_unlock(&dentry
->d_lock
);
793 dput_to_list(dentry
, list
);
794 hlist_del(&mp
->m_hash
);
799 /* called with namespace_lock and vfsmount lock */
800 static void put_mountpoint(struct mountpoint
*mp
)
802 __put_mountpoint(mp
, &ex_mountpoints
);
805 static inline int check_mnt(struct mount
*mnt
)
807 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
811 * vfsmount lock must be held for write
813 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
817 wake_up_interruptible(&ns
->poll
);
822 * vfsmount lock must be held for write
824 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
826 if (ns
&& ns
->event
!= event
) {
828 wake_up_interruptible(&ns
->poll
);
833 * vfsmount lock must be held for write
835 static struct mountpoint
*unhash_mnt(struct mount
*mnt
)
837 struct mountpoint
*mp
;
838 mnt
->mnt_parent
= mnt
;
839 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
840 list_del_init(&mnt
->mnt_child
);
841 hlist_del_init_rcu(&mnt
->mnt_hash
);
842 hlist_del_init(&mnt
->mnt_mp_list
);
849 * vfsmount lock must be held for write
851 static void umount_mnt(struct mount
*mnt
)
853 put_mountpoint(unhash_mnt(mnt
));
857 * vfsmount lock must be held for write
859 void mnt_set_mountpoint(struct mount
*mnt
,
860 struct mountpoint
*mp
,
861 struct mount
*child_mnt
)
864 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
865 child_mnt
->mnt_mountpoint
= mp
->m_dentry
;
866 child_mnt
->mnt_parent
= mnt
;
867 child_mnt
->mnt_mp
= mp
;
868 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
871 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
873 hlist_add_head_rcu(&mnt
->mnt_hash
,
874 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
875 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
879 * vfsmount lock must be held for write
881 static void attach_mnt(struct mount
*mnt
,
882 struct mount
*parent
,
883 struct mountpoint
*mp
)
885 mnt_set_mountpoint(parent
, mp
, mnt
);
886 __attach_mnt(mnt
, parent
);
889 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
891 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
892 struct mount
*old_parent
= mnt
->mnt_parent
;
894 list_del_init(&mnt
->mnt_child
);
895 hlist_del_init(&mnt
->mnt_mp_list
);
896 hlist_del_init_rcu(&mnt
->mnt_hash
);
898 attach_mnt(mnt
, parent
, mp
);
900 put_mountpoint(old_mp
);
901 mnt_add_count(old_parent
, -1);
905 * vfsmount lock must be held for write
907 static void commit_tree(struct mount
*mnt
)
909 struct mount
*parent
= mnt
->mnt_parent
;
912 struct mnt_namespace
*n
= parent
->mnt_ns
;
914 BUG_ON(parent
== mnt
);
916 list_add_tail(&head
, &mnt
->mnt_list
);
917 list_for_each_entry(m
, &head
, mnt_list
)
920 list_splice(&head
, n
->list
.prev
);
922 n
->mounts
+= n
->pending_mounts
;
923 n
->pending_mounts
= 0;
925 __attach_mnt(mnt
, parent
);
926 touch_mnt_namespace(n
);
929 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
931 struct list_head
*next
= p
->mnt_mounts
.next
;
932 if (next
== &p
->mnt_mounts
) {
936 next
= p
->mnt_child
.next
;
937 if (next
!= &p
->mnt_parent
->mnt_mounts
)
942 return list_entry(next
, struct mount
, mnt_child
);
945 static struct mount
*skip_mnt_tree(struct mount
*p
)
947 struct list_head
*prev
= p
->mnt_mounts
.prev
;
948 while (prev
!= &p
->mnt_mounts
) {
949 p
= list_entry(prev
, struct mount
, mnt_child
);
950 prev
= p
->mnt_mounts
.prev
;
956 * vfs_create_mount - Create a mount for a configured superblock
957 * @fc: The configuration context with the superblock attached
959 * Create a mount to an already configured superblock. If necessary, the
960 * caller should invoke vfs_get_tree() before calling this.
962 * Note that this does not attach the mount to anything.
964 struct vfsmount
*vfs_create_mount(struct fs_context
*fc
)
969 return ERR_PTR(-EINVAL
);
971 mnt
= alloc_vfsmnt(fc
->source
?: "none");
973 return ERR_PTR(-ENOMEM
);
975 if (fc
->sb_flags
& SB_KERNMOUNT
)
976 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
978 atomic_inc(&fc
->root
->d_sb
->s_active
);
979 mnt
->mnt
.mnt_sb
= fc
->root
->d_sb
;
980 mnt
->mnt
.mnt_root
= dget(fc
->root
);
981 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
982 mnt
->mnt_parent
= mnt
;
985 list_add_tail(&mnt
->mnt_instance
, &mnt
->mnt
.mnt_sb
->s_mounts
);
989 EXPORT_SYMBOL(vfs_create_mount
);
991 struct vfsmount
*fc_mount(struct fs_context
*fc
)
993 int err
= vfs_get_tree(fc
);
995 up_write(&fc
->root
->d_sb
->s_umount
);
996 return vfs_create_mount(fc
);
1000 EXPORT_SYMBOL(fc_mount
);
1002 struct vfsmount
*vfs_kern_mount(struct file_system_type
*type
,
1003 int flags
, const char *name
,
1006 struct fs_context
*fc
;
1007 struct vfsmount
*mnt
;
1011 return ERR_PTR(-EINVAL
);
1013 fc
= fs_context_for_mount(type
, flags
);
1015 return ERR_CAST(fc
);
1018 ret
= vfs_parse_fs_string(fc
, "source",
1019 name
, strlen(name
));
1021 ret
= parse_monolithic_mount_data(fc
, data
);
1030 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1033 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1034 const char *name
, void *data
)
1036 /* Until it is worked out how to pass the user namespace
1037 * through from the parent mount to the submount don't support
1038 * unprivileged mounts with submounts.
1040 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1041 return ERR_PTR(-EPERM
);
1043 return vfs_kern_mount(type
, SB_SUBMOUNT
, name
, data
);
1045 EXPORT_SYMBOL_GPL(vfs_submount
);
1047 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1050 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1054 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1056 return ERR_PTR(-ENOMEM
);
1058 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1059 mnt
->mnt_group_id
= 0; /* not a peer of original */
1061 mnt
->mnt_group_id
= old
->mnt_group_id
;
1063 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1064 err
= mnt_alloc_group_id(mnt
);
1069 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
;
1070 mnt
->mnt
.mnt_flags
&= ~(MNT_WRITE_HOLD
|MNT_MARKED
|MNT_INTERNAL
);
1072 atomic_inc(&sb
->s_active
);
1073 mnt
->mnt
.mnt_userns
= mnt_user_ns(&old
->mnt
);
1074 if (mnt
->mnt
.mnt_userns
!= &init_user_ns
)
1075 mnt
->mnt
.mnt_userns
= get_user_ns(mnt
->mnt
.mnt_userns
);
1076 mnt
->mnt
.mnt_sb
= sb
;
1077 mnt
->mnt
.mnt_root
= dget(root
);
1078 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1079 mnt
->mnt_parent
= mnt
;
1081 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1082 unlock_mount_hash();
1084 if ((flag
& CL_SLAVE
) ||
1085 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1086 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1087 mnt
->mnt_master
= old
;
1088 CLEAR_MNT_SHARED(mnt
);
1089 } else if (!(flag
& CL_PRIVATE
)) {
1090 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1091 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1092 if (IS_MNT_SLAVE(old
))
1093 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1094 mnt
->mnt_master
= old
->mnt_master
;
1096 CLEAR_MNT_SHARED(mnt
);
1098 if (flag
& CL_MAKE_SHARED
)
1099 set_mnt_shared(mnt
);
1101 /* stick the duplicate mount on the same expiry list
1102 * as the original if that was on one */
1103 if (flag
& CL_EXPIRE
) {
1104 if (!list_empty(&old
->mnt_expire
))
1105 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1113 return ERR_PTR(err
);
1116 static void cleanup_mnt(struct mount
*mnt
)
1118 struct hlist_node
*p
;
1121 * The warning here probably indicates that somebody messed
1122 * up a mnt_want/drop_write() pair. If this happens, the
1123 * filesystem was probably unable to make r/w->r/o transitions.
1124 * The locking used to deal with mnt_count decrement provides barriers,
1125 * so mnt_get_writers() below is safe.
1127 WARN_ON(mnt_get_writers(mnt
));
1128 if (unlikely(mnt
->mnt_pins
.first
))
1130 hlist_for_each_entry_safe(m
, p
, &mnt
->mnt_stuck_children
, mnt_umount
) {
1131 hlist_del(&m
->mnt_umount
);
1134 fsnotify_vfsmount_delete(&mnt
->mnt
);
1135 dput(mnt
->mnt
.mnt_root
);
1136 deactivate_super(mnt
->mnt
.mnt_sb
);
1138 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1141 static void __cleanup_mnt(struct rcu_head
*head
)
1143 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1146 static LLIST_HEAD(delayed_mntput_list
);
1147 static void delayed_mntput(struct work_struct
*unused
)
1149 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1150 struct mount
*m
, *t
;
1152 llist_for_each_entry_safe(m
, t
, node
, mnt_llist
)
1155 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1157 static void mntput_no_expire(struct mount
*mnt
)
1163 if (likely(READ_ONCE(mnt
->mnt_ns
))) {
1165 * Since we don't do lock_mount_hash() here,
1166 * ->mnt_ns can change under us. However, if it's
1167 * non-NULL, then there's a reference that won't
1168 * be dropped until after an RCU delay done after
1169 * turning ->mnt_ns NULL. So if we observe it
1170 * non-NULL under rcu_read_lock(), the reference
1171 * we are dropping is not the final one.
1173 mnt_add_count(mnt
, -1);
1179 * make sure that if __legitimize_mnt() has not seen us grab
1180 * mount_lock, we'll see their refcount increment here.
1183 mnt_add_count(mnt
, -1);
1184 count
= mnt_get_count(mnt
);
1188 unlock_mount_hash();
1191 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1193 unlock_mount_hash();
1196 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1199 list_del(&mnt
->mnt_instance
);
1201 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1202 struct mount
*p
, *tmp
;
1203 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1204 __put_mountpoint(unhash_mnt(p
), &list
);
1205 hlist_add_head(&p
->mnt_umount
, &mnt
->mnt_stuck_children
);
1208 unlock_mount_hash();
1209 shrink_dentry_list(&list
);
1211 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1212 struct task_struct
*task
= current
;
1213 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1214 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1215 if (!task_work_add(task
, &mnt
->mnt_rcu
, TWA_RESUME
))
1218 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1219 schedule_delayed_work(&delayed_mntput_work
, 1);
1225 void mntput(struct vfsmount
*mnt
)
1228 struct mount
*m
= real_mount(mnt
);
1229 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1230 if (unlikely(m
->mnt_expiry_mark
))
1231 m
->mnt_expiry_mark
= 0;
1232 mntput_no_expire(m
);
1235 EXPORT_SYMBOL(mntput
);
1237 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1240 mnt_add_count(real_mount(mnt
), 1);
1243 EXPORT_SYMBOL(mntget
);
1246 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1247 * @path: path to check
1249 * d_mountpoint() can only be used reliably to establish if a dentry is
1250 * not mounted in any namespace and that common case is handled inline.
1251 * d_mountpoint() isn't aware of the possibility there may be multiple
1252 * mounts using a given dentry in a different namespace. This function
1253 * checks if the passed in path is a mountpoint rather than the dentry
1256 bool path_is_mountpoint(const struct path
*path
)
1261 if (!d_mountpoint(path
->dentry
))
1266 seq
= read_seqbegin(&mount_lock
);
1267 res
= __path_is_mountpoint(path
);
1268 } while (read_seqretry(&mount_lock
, seq
));
1273 EXPORT_SYMBOL(path_is_mountpoint
);
1275 struct vfsmount
*mnt_clone_internal(const struct path
*path
)
1278 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1281 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1285 #ifdef CONFIG_PROC_FS
1286 static struct mount
*mnt_list_next(struct mnt_namespace
*ns
,
1287 struct list_head
*p
)
1289 struct mount
*mnt
, *ret
= NULL
;
1292 list_for_each_continue(p
, &ns
->list
) {
1293 mnt
= list_entry(p
, typeof(*mnt
), mnt_list
);
1294 if (!mnt_is_cursor(mnt
)) {
1304 /* iterator; we want it to have access to namespace_sem, thus here... */
1305 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1307 struct proc_mounts
*p
= m
->private;
1308 struct list_head
*prev
;
1310 down_read(&namespace_sem
);
1312 prev
= &p
->ns
->list
;
1314 prev
= &p
->cursor
.mnt_list
;
1316 /* Read after we'd reached the end? */
1317 if (list_empty(prev
))
1321 return mnt_list_next(p
->ns
, prev
);
1324 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1326 struct proc_mounts
*p
= m
->private;
1327 struct mount
*mnt
= v
;
1330 return mnt_list_next(p
->ns
, &mnt
->mnt_list
);
1333 static void m_stop(struct seq_file
*m
, void *v
)
1335 struct proc_mounts
*p
= m
->private;
1336 struct mount
*mnt
= v
;
1338 lock_ns_list(p
->ns
);
1340 list_move_tail(&p
->cursor
.mnt_list
, &mnt
->mnt_list
);
1342 list_del_init(&p
->cursor
.mnt_list
);
1343 unlock_ns_list(p
->ns
);
1344 up_read(&namespace_sem
);
1347 static int m_show(struct seq_file
*m
, void *v
)
1349 struct proc_mounts
*p
= m
->private;
1350 struct mount
*r
= v
;
1351 return p
->show(m
, &r
->mnt
);
1354 const struct seq_operations mounts_op
= {
1361 void mnt_cursor_del(struct mnt_namespace
*ns
, struct mount
*cursor
)
1363 down_read(&namespace_sem
);
1365 list_del(&cursor
->mnt_list
);
1367 up_read(&namespace_sem
);
1369 #endif /* CONFIG_PROC_FS */
1372 * may_umount_tree - check if a mount tree is busy
1373 * @m: root of mount tree
1375 * This is called to check if a tree of mounts has any
1376 * open files, pwds, chroots or sub mounts that are
1379 int may_umount_tree(struct vfsmount
*m
)
1381 struct mount
*mnt
= real_mount(m
);
1382 int actual_refs
= 0;
1383 int minimum_refs
= 0;
1387 /* write lock needed for mnt_get_count */
1389 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1390 actual_refs
+= mnt_get_count(p
);
1393 unlock_mount_hash();
1395 if (actual_refs
> minimum_refs
)
1401 EXPORT_SYMBOL(may_umount_tree
);
1404 * may_umount - check if a mount point is busy
1405 * @mnt: root of mount
1407 * This is called to check if a mount point has any
1408 * open files, pwds, chroots or sub mounts. If the
1409 * mount has sub mounts this will return busy
1410 * regardless of whether the sub mounts are busy.
1412 * Doesn't take quota and stuff into account. IOW, in some cases it will
1413 * give false negatives. The main reason why it's here is that we need
1414 * a non-destructive way to look for easily umountable filesystems.
1416 int may_umount(struct vfsmount
*mnt
)
1419 down_read(&namespace_sem
);
1421 if (propagate_mount_busy(real_mount(mnt
), 2))
1423 unlock_mount_hash();
1424 up_read(&namespace_sem
);
1428 EXPORT_SYMBOL(may_umount
);
1430 static void namespace_unlock(void)
1432 struct hlist_head head
;
1433 struct hlist_node
*p
;
1437 hlist_move_list(&unmounted
, &head
);
1438 list_splice_init(&ex_mountpoints
, &list
);
1440 up_write(&namespace_sem
);
1442 shrink_dentry_list(&list
);
1444 if (likely(hlist_empty(&head
)))
1447 synchronize_rcu_expedited();
1449 hlist_for_each_entry_safe(m
, p
, &head
, mnt_umount
) {
1450 hlist_del(&m
->mnt_umount
);
1455 static inline void namespace_lock(void)
1457 down_write(&namespace_sem
);
1460 enum umount_tree_flags
{
1462 UMOUNT_PROPAGATE
= 2,
1463 UMOUNT_CONNECTED
= 4,
1466 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1468 /* Leaving mounts connected is only valid for lazy umounts */
1469 if (how
& UMOUNT_SYNC
)
1472 /* A mount without a parent has nothing to be connected to */
1473 if (!mnt_has_parent(mnt
))
1476 /* Because the reference counting rules change when mounts are
1477 * unmounted and connected, umounted mounts may not be
1478 * connected to mounted mounts.
1480 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1483 /* Has it been requested that the mount remain connected? */
1484 if (how
& UMOUNT_CONNECTED
)
1487 /* Is the mount locked such that it needs to remain connected? */
1488 if (IS_MNT_LOCKED(mnt
))
1491 /* By default disconnect the mount */
1496 * mount_lock must be held
1497 * namespace_sem must be held for write
1499 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1501 LIST_HEAD(tmp_list
);
1504 if (how
& UMOUNT_PROPAGATE
)
1505 propagate_mount_unlock(mnt
);
1507 /* Gather the mounts to umount */
1508 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1509 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1510 list_move(&p
->mnt_list
, &tmp_list
);
1513 /* Hide the mounts from mnt_mounts */
1514 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1515 list_del_init(&p
->mnt_child
);
1518 /* Add propogated mounts to the tmp_list */
1519 if (how
& UMOUNT_PROPAGATE
)
1520 propagate_umount(&tmp_list
);
1522 while (!list_empty(&tmp_list
)) {
1523 struct mnt_namespace
*ns
;
1525 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1526 list_del_init(&p
->mnt_expire
);
1527 list_del_init(&p
->mnt_list
);
1531 __touch_mnt_namespace(ns
);
1534 if (how
& UMOUNT_SYNC
)
1535 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1537 disconnect
= disconnect_mount(p
, how
);
1538 if (mnt_has_parent(p
)) {
1539 mnt_add_count(p
->mnt_parent
, -1);
1541 /* Don't forget about p */
1542 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1547 change_mnt_propagation(p
, MS_PRIVATE
);
1549 hlist_add_head(&p
->mnt_umount
, &unmounted
);
1553 static void shrink_submounts(struct mount
*mnt
);
1555 static int do_umount_root(struct super_block
*sb
)
1559 down_write(&sb
->s_umount
);
1560 if (!sb_rdonly(sb
)) {
1561 struct fs_context
*fc
;
1563 fc
= fs_context_for_reconfigure(sb
->s_root
, SB_RDONLY
,
1568 ret
= parse_monolithic_mount_data(fc
, NULL
);
1570 ret
= reconfigure_super(fc
);
1574 up_write(&sb
->s_umount
);
1578 static int do_umount(struct mount
*mnt
, int flags
)
1580 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1583 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1588 * Allow userspace to request a mountpoint be expired rather than
1589 * unmounting unconditionally. Unmount only happens if:
1590 * (1) the mark is already set (the mark is cleared by mntput())
1591 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1593 if (flags
& MNT_EXPIRE
) {
1594 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1595 flags
& (MNT_FORCE
| MNT_DETACH
))
1599 * probably don't strictly need the lock here if we examined
1600 * all race cases, but it's a slowpath.
1603 if (mnt_get_count(mnt
) != 2) {
1604 unlock_mount_hash();
1607 unlock_mount_hash();
1609 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1614 * If we may have to abort operations to get out of this
1615 * mount, and they will themselves hold resources we must
1616 * allow the fs to do things. In the Unix tradition of
1617 * 'Gee thats tricky lets do it in userspace' the umount_begin
1618 * might fail to complete on the first run through as other tasks
1619 * must return, and the like. Thats for the mount program to worry
1620 * about for the moment.
1623 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1624 sb
->s_op
->umount_begin(sb
);
1628 * No sense to grab the lock for this test, but test itself looks
1629 * somewhat bogus. Suggestions for better replacement?
1630 * Ho-hum... In principle, we might treat that as umount + switch
1631 * to rootfs. GC would eventually take care of the old vfsmount.
1632 * Actually it makes sense, especially if rootfs would contain a
1633 * /reboot - static binary that would close all descriptors and
1634 * call reboot(9). Then init(8) could umount root and exec /reboot.
1636 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1638 * Special case for "unmounting" root ...
1639 * we just try to remount it readonly.
1641 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1643 return do_umount_root(sb
);
1649 /* Recheck MNT_LOCKED with the locks held */
1651 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1655 if (flags
& MNT_DETACH
) {
1656 if (!list_empty(&mnt
->mnt_list
))
1657 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1660 shrink_submounts(mnt
);
1662 if (!propagate_mount_busy(mnt
, 2)) {
1663 if (!list_empty(&mnt
->mnt_list
))
1664 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1669 unlock_mount_hash();
1675 * __detach_mounts - lazily unmount all mounts on the specified dentry
1677 * During unlink, rmdir, and d_drop it is possible to loose the path
1678 * to an existing mountpoint, and wind up leaking the mount.
1679 * detach_mounts allows lazily unmounting those mounts instead of
1682 * The caller may hold dentry->d_inode->i_mutex.
1684 void __detach_mounts(struct dentry
*dentry
)
1686 struct mountpoint
*mp
;
1691 mp
= lookup_mountpoint(dentry
);
1696 while (!hlist_empty(&mp
->m_list
)) {
1697 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1698 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1700 hlist_add_head(&mnt
->mnt_umount
, &unmounted
);
1702 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1706 unlock_mount_hash();
1711 * Is the caller allowed to modify his namespace?
1713 static inline bool may_mount(void)
1715 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1718 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1719 static inline bool may_mandlock(void)
1721 return capable(CAP_SYS_ADMIN
);
1724 static inline bool may_mandlock(void)
1726 pr_warn("VFS: \"mand\" mount option not supported");
1731 static int can_umount(const struct path
*path
, int flags
)
1733 struct mount
*mnt
= real_mount(path
->mnt
);
1737 if (path
->dentry
!= path
->mnt
->mnt_root
)
1739 if (!check_mnt(mnt
))
1741 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
) /* Check optimistically */
1743 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1748 // caller is responsible for flags being sane
1749 int path_umount(struct path
*path
, int flags
)
1751 struct mount
*mnt
= real_mount(path
->mnt
);
1754 ret
= can_umount(path
, flags
);
1756 ret
= do_umount(mnt
, flags
);
1758 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1760 mntput_no_expire(mnt
);
1764 static int ksys_umount(char __user
*name
, int flags
)
1766 int lookup_flags
= LOOKUP_MOUNTPOINT
;
1770 // basic validity checks done first
1771 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1774 if (!(flags
& UMOUNT_NOFOLLOW
))
1775 lookup_flags
|= LOOKUP_FOLLOW
;
1776 ret
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1779 return path_umount(&path
, flags
);
1782 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1784 return ksys_umount(name
, flags
);
1787 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1790 * The 2.0 compatible umount. No flags.
1792 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1794 return ksys_umount(name
, 0);
1799 static bool is_mnt_ns_file(struct dentry
*dentry
)
1801 /* Is this a proxy for a mount namespace? */
1802 return dentry
->d_op
== &ns_dentry_operations
&&
1803 dentry
->d_fsdata
== &mntns_operations
;
1806 static struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1808 return container_of(ns
, struct mnt_namespace
, ns
);
1811 struct ns_common
*from_mnt_ns(struct mnt_namespace
*mnt
)
1816 static bool mnt_ns_loop(struct dentry
*dentry
)
1818 /* Could bind mounting the mount namespace inode cause a
1819 * mount namespace loop?
1821 struct mnt_namespace
*mnt_ns
;
1822 if (!is_mnt_ns_file(dentry
))
1825 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1826 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1829 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1832 struct mount
*res
, *p
, *q
, *r
, *parent
;
1834 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1835 return ERR_PTR(-EINVAL
);
1837 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1838 return ERR_PTR(-EINVAL
);
1840 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1844 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1847 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1849 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1852 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1853 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1854 IS_MNT_UNBINDABLE(s
)) {
1855 if (s
->mnt
.mnt_flags
& MNT_LOCKED
) {
1856 /* Both unbindable and locked. */
1857 q
= ERR_PTR(-EPERM
);
1860 s
= skip_mnt_tree(s
);
1864 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1865 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1866 s
= skip_mnt_tree(s
);
1869 while (p
!= s
->mnt_parent
) {
1875 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1879 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1880 attach_mnt(q
, parent
, p
->mnt_mp
);
1881 unlock_mount_hash();
1888 umount_tree(res
, UMOUNT_SYNC
);
1889 unlock_mount_hash();
1894 /* Caller should check returned pointer for errors */
1896 struct vfsmount
*collect_mounts(const struct path
*path
)
1900 if (!check_mnt(real_mount(path
->mnt
)))
1901 tree
= ERR_PTR(-EINVAL
);
1903 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1904 CL_COPY_ALL
| CL_PRIVATE
);
1907 return ERR_CAST(tree
);
1911 static void free_mnt_ns(struct mnt_namespace
*);
1912 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*, bool);
1914 void dissolve_on_fput(struct vfsmount
*mnt
)
1916 struct mnt_namespace
*ns
;
1919 ns
= real_mount(mnt
)->mnt_ns
;
1922 umount_tree(real_mount(mnt
), UMOUNT_CONNECTED
);
1926 unlock_mount_hash();
1932 void drop_collected_mounts(struct vfsmount
*mnt
)
1936 umount_tree(real_mount(mnt
), 0);
1937 unlock_mount_hash();
1942 * clone_private_mount - create a private clone of a path
1943 * @path: path to clone
1945 * This creates a new vfsmount, which will be the clone of @path. The new mount
1946 * will not be attached anywhere in the namespace and will be private (i.e.
1947 * changes to the originating mount won't be propagated into this).
1949 * Release with mntput().
1951 struct vfsmount
*clone_private_mount(const struct path
*path
)
1953 struct mount
*old_mnt
= real_mount(path
->mnt
);
1954 struct mount
*new_mnt
;
1956 if (IS_MNT_UNBINDABLE(old_mnt
))
1957 return ERR_PTR(-EINVAL
);
1959 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1960 if (IS_ERR(new_mnt
))
1961 return ERR_CAST(new_mnt
);
1963 /* Longterm mount to be removed by kern_unmount*() */
1964 new_mnt
->mnt_ns
= MNT_NS_INTERNAL
;
1966 return &new_mnt
->mnt
;
1968 EXPORT_SYMBOL_GPL(clone_private_mount
);
1970 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1971 struct vfsmount
*root
)
1974 int res
= f(root
, arg
);
1977 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1978 res
= f(&mnt
->mnt
, arg
);
1985 static void lock_mnt_tree(struct mount
*mnt
)
1989 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1990 int flags
= p
->mnt
.mnt_flags
;
1991 /* Don't allow unprivileged users to change mount flags */
1992 flags
|= MNT_LOCK_ATIME
;
1994 if (flags
& MNT_READONLY
)
1995 flags
|= MNT_LOCK_READONLY
;
1997 if (flags
& MNT_NODEV
)
1998 flags
|= MNT_LOCK_NODEV
;
2000 if (flags
& MNT_NOSUID
)
2001 flags
|= MNT_LOCK_NOSUID
;
2003 if (flags
& MNT_NOEXEC
)
2004 flags
|= MNT_LOCK_NOEXEC
;
2005 /* Don't allow unprivileged users to reveal what is under a mount */
2006 if (list_empty(&p
->mnt_expire
))
2007 flags
|= MNT_LOCKED
;
2008 p
->mnt
.mnt_flags
= flags
;
2012 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
2016 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
2017 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
2018 mnt_release_group_id(p
);
2022 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
2026 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
2027 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
2028 int err
= mnt_alloc_group_id(p
);
2030 cleanup_group_ids(mnt
, p
);
2039 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
2041 unsigned int max
= READ_ONCE(sysctl_mount_max
);
2042 unsigned int mounts
= 0, old
, pending
, sum
;
2045 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
2049 pending
= ns
->pending_mounts
;
2050 sum
= old
+ pending
;
2054 (mounts
> (max
- sum
)))
2057 ns
->pending_mounts
= pending
+ mounts
;
2062 * @source_mnt : mount tree to be attached
2063 * @nd : place the mount tree @source_mnt is attached
2064 * @parent_nd : if non-null, detach the source_mnt from its parent and
2065 * store the parent mount and mountpoint dentry.
2066 * (done when source_mnt is moved)
2068 * NOTE: in the table below explains the semantics when a source mount
2069 * of a given type is attached to a destination mount of a given type.
2070 * ---------------------------------------------------------------------------
2071 * | BIND MOUNT OPERATION |
2072 * |**************************************************************************
2073 * | source-->| shared | private | slave | unbindable |
2077 * |**************************************************************************
2078 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2080 * |non-shared| shared (+) | private | slave (*) | invalid |
2081 * ***************************************************************************
2082 * A bind operation clones the source mount and mounts the clone on the
2083 * destination mount.
2085 * (++) the cloned mount is propagated to all the mounts in the propagation
2086 * tree of the destination mount and the cloned mount is added to
2087 * the peer group of the source mount.
2088 * (+) the cloned mount is created under the destination mount and is marked
2089 * as shared. The cloned mount is added to the peer group of the source
2091 * (+++) the mount is propagated to all the mounts in the propagation tree
2092 * of the destination mount and the cloned mount is made slave
2093 * of the same master as that of the source mount. The cloned mount
2094 * is marked as 'shared and slave'.
2095 * (*) the cloned mount is made a slave of the same master as that of the
2098 * ---------------------------------------------------------------------------
2099 * | MOVE MOUNT OPERATION |
2100 * |**************************************************************************
2101 * | source-->| shared | private | slave | unbindable |
2105 * |**************************************************************************
2106 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2108 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2109 * ***************************************************************************
2111 * (+) the mount is moved to the destination. And is then propagated to
2112 * all the mounts in the propagation tree of the destination mount.
2113 * (+*) the mount is moved to the destination.
2114 * (+++) the mount is moved to the destination and is then propagated to
2115 * all the mounts belonging to the destination mount's propagation tree.
2116 * the mount is marked as 'shared and slave'.
2117 * (*) the mount continues to be a slave at the new location.
2119 * if the source mount is a tree, the operations explained above is
2120 * applied to each mount in the tree.
2121 * Must be called without spinlocks held, since this function can sleep
2124 static int attach_recursive_mnt(struct mount
*source_mnt
,
2125 struct mount
*dest_mnt
,
2126 struct mountpoint
*dest_mp
,
2129 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2130 HLIST_HEAD(tree_list
);
2131 struct mnt_namespace
*ns
= dest_mnt
->mnt_ns
;
2132 struct mountpoint
*smp
;
2133 struct mount
*child
, *p
;
2134 struct hlist_node
*n
;
2137 /* Preallocate a mountpoint in case the new mounts need
2138 * to be tucked under other mounts.
2140 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
2142 return PTR_ERR(smp
);
2144 /* Is there space to add these mounts to the mount namespace? */
2146 err
= count_mounts(ns
, source_mnt
);
2151 if (IS_MNT_SHARED(dest_mnt
)) {
2152 err
= invent_group_ids(source_mnt
, true);
2155 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2158 goto out_cleanup_ids
;
2159 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2165 unhash_mnt(source_mnt
);
2166 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
2167 touch_mnt_namespace(source_mnt
->mnt_ns
);
2169 if (source_mnt
->mnt_ns
) {
2170 /* move from anon - the caller will destroy */
2171 list_del_init(&source_mnt
->mnt_ns
->list
);
2173 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2174 commit_tree(source_mnt
);
2177 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2179 hlist_del_init(&child
->mnt_hash
);
2180 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2181 child
->mnt_mountpoint
);
2183 mnt_change_mountpoint(child
, smp
, q
);
2184 /* Notice when we are propagating across user namespaces */
2185 if (child
->mnt_parent
->mnt_ns
->user_ns
!= user_ns
)
2186 lock_mnt_tree(child
);
2187 child
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2190 put_mountpoint(smp
);
2191 unlock_mount_hash();
2196 while (!hlist_empty(&tree_list
)) {
2197 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2198 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2199 umount_tree(child
, UMOUNT_SYNC
);
2201 unlock_mount_hash();
2202 cleanup_group_ids(source_mnt
, NULL
);
2204 ns
->pending_mounts
= 0;
2206 read_seqlock_excl(&mount_lock
);
2207 put_mountpoint(smp
);
2208 read_sequnlock_excl(&mount_lock
);
2213 static struct mountpoint
*lock_mount(struct path
*path
)
2215 struct vfsmount
*mnt
;
2216 struct dentry
*dentry
= path
->dentry
;
2218 inode_lock(dentry
->d_inode
);
2219 if (unlikely(cant_mount(dentry
))) {
2220 inode_unlock(dentry
->d_inode
);
2221 return ERR_PTR(-ENOENT
);
2224 mnt
= lookup_mnt(path
);
2226 struct mountpoint
*mp
= get_mountpoint(dentry
);
2229 inode_unlock(dentry
->d_inode
);
2235 inode_unlock(path
->dentry
->d_inode
);
2238 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2242 static void unlock_mount(struct mountpoint
*where
)
2244 struct dentry
*dentry
= where
->m_dentry
;
2246 read_seqlock_excl(&mount_lock
);
2247 put_mountpoint(where
);
2248 read_sequnlock_excl(&mount_lock
);
2251 inode_unlock(dentry
->d_inode
);
2254 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2256 if (mnt
->mnt
.mnt_sb
->s_flags
& SB_NOUSER
)
2259 if (d_is_dir(mp
->m_dentry
) !=
2260 d_is_dir(mnt
->mnt
.mnt_root
))
2263 return attach_recursive_mnt(mnt
, p
, mp
, false);
2267 * Sanity check the flags to change_mnt_propagation.
2270 static int flags_to_propagation_type(int ms_flags
)
2272 int type
= ms_flags
& ~(MS_REC
| MS_SILENT
);
2274 /* Fail if any non-propagation flags are set */
2275 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2277 /* Only one propagation flag should be set */
2278 if (!is_power_of_2(type
))
2284 * recursively change the type of the mountpoint.
2286 static int do_change_type(struct path
*path
, int ms_flags
)
2289 struct mount
*mnt
= real_mount(path
->mnt
);
2290 int recurse
= ms_flags
& MS_REC
;
2294 if (path
->dentry
!= path
->mnt
->mnt_root
)
2297 type
= flags_to_propagation_type(ms_flags
);
2302 if (type
== MS_SHARED
) {
2303 err
= invent_group_ids(mnt
, recurse
);
2309 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2310 change_mnt_propagation(m
, type
);
2311 unlock_mount_hash();
2318 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2320 struct mount
*child
;
2321 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2322 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2325 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2331 static struct mount
*__do_loopback(struct path
*old_path
, int recurse
)
2333 struct mount
*mnt
= ERR_PTR(-EINVAL
), *old
= real_mount(old_path
->mnt
);
2335 if (IS_MNT_UNBINDABLE(old
))
2338 if (!check_mnt(old
) && old_path
->dentry
->d_op
!= &ns_dentry_operations
)
2341 if (!recurse
&& has_locked_children(old
, old_path
->dentry
))
2345 mnt
= copy_tree(old
, old_path
->dentry
, CL_COPY_MNT_NS_FILE
);
2347 mnt
= clone_mnt(old
, old_path
->dentry
, 0);
2350 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2356 * do loopback mount.
2358 static int do_loopback(struct path
*path
, const char *old_name
,
2361 struct path old_path
;
2362 struct mount
*mnt
= NULL
, *parent
;
2363 struct mountpoint
*mp
;
2365 if (!old_name
|| !*old_name
)
2367 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2372 if (mnt_ns_loop(old_path
.dentry
))
2375 mp
= lock_mount(path
);
2381 parent
= real_mount(path
->mnt
);
2382 if (!check_mnt(parent
))
2385 mnt
= __do_loopback(&old_path
, recurse
);
2391 err
= graft_tree(mnt
, parent
, mp
);
2394 umount_tree(mnt
, UMOUNT_SYNC
);
2395 unlock_mount_hash();
2400 path_put(&old_path
);
2404 static struct file
*open_detached_copy(struct path
*path
, bool recursive
)
2406 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2407 struct mnt_namespace
*ns
= alloc_mnt_ns(user_ns
, true);
2408 struct mount
*mnt
, *p
;
2412 return ERR_CAST(ns
);
2415 mnt
= __do_loopback(path
, recursive
);
2419 return ERR_CAST(mnt
);
2423 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2428 list_add_tail(&ns
->list
, &mnt
->mnt_list
);
2430 unlock_mount_hash();
2434 path
->mnt
= &mnt
->mnt
;
2435 file
= dentry_open(path
, O_PATH
, current_cred());
2437 dissolve_on_fput(path
->mnt
);
2439 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
2443 SYSCALL_DEFINE3(open_tree
, int, dfd
, const char __user
*, filename
, unsigned, flags
)
2447 int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
2448 bool detached
= flags
& OPEN_TREE_CLONE
;
2452 BUILD_BUG_ON(OPEN_TREE_CLOEXEC
!= O_CLOEXEC
);
2454 if (flags
& ~(AT_EMPTY_PATH
| AT_NO_AUTOMOUNT
| AT_RECURSIVE
|
2455 AT_SYMLINK_NOFOLLOW
| OPEN_TREE_CLONE
|
2459 if ((flags
& (AT_RECURSIVE
| OPEN_TREE_CLONE
)) == AT_RECURSIVE
)
2462 if (flags
& AT_NO_AUTOMOUNT
)
2463 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
2464 if (flags
& AT_SYMLINK_NOFOLLOW
)
2465 lookup_flags
&= ~LOOKUP_FOLLOW
;
2466 if (flags
& AT_EMPTY_PATH
)
2467 lookup_flags
|= LOOKUP_EMPTY
;
2469 if (detached
&& !may_mount())
2472 fd
= get_unused_fd_flags(flags
& O_CLOEXEC
);
2476 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
2477 if (unlikely(error
)) {
2478 file
= ERR_PTR(error
);
2481 file
= open_detached_copy(&path
, flags
& AT_RECURSIVE
);
2483 file
= dentry_open(&path
, O_PATH
, current_cred());
2488 return PTR_ERR(file
);
2490 fd_install(fd
, file
);
2495 * Don't allow locked mount flags to be cleared.
2497 * No locks need to be held here while testing the various MNT_LOCK
2498 * flags because those flags can never be cleared once they are set.
2500 static bool can_change_locked_flags(struct mount
*mnt
, unsigned int mnt_flags
)
2502 unsigned int fl
= mnt
->mnt
.mnt_flags
;
2504 if ((fl
& MNT_LOCK_READONLY
) &&
2505 !(mnt_flags
& MNT_READONLY
))
2508 if ((fl
& MNT_LOCK_NODEV
) &&
2509 !(mnt_flags
& MNT_NODEV
))
2512 if ((fl
& MNT_LOCK_NOSUID
) &&
2513 !(mnt_flags
& MNT_NOSUID
))
2516 if ((fl
& MNT_LOCK_NOEXEC
) &&
2517 !(mnt_flags
& MNT_NOEXEC
))
2520 if ((fl
& MNT_LOCK_ATIME
) &&
2521 ((fl
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
)))
2527 static int change_mount_ro_state(struct mount
*mnt
, unsigned int mnt_flags
)
2529 bool readonly_request
= (mnt_flags
& MNT_READONLY
);
2531 if (readonly_request
== __mnt_is_readonly(&mnt
->mnt
))
2534 if (readonly_request
)
2535 return mnt_make_readonly(mnt
);
2537 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
2541 static void set_mount_attributes(struct mount
*mnt
, unsigned int mnt_flags
)
2543 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2544 mnt
->mnt
.mnt_flags
= mnt_flags
;
2545 touch_mnt_namespace(mnt
->mnt_ns
);
2548 static void mnt_warn_timestamp_expiry(struct path
*mountpoint
, struct vfsmount
*mnt
)
2550 struct super_block
*sb
= mnt
->mnt_sb
;
2552 if (!__mnt_is_readonly(mnt
) &&
2553 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX
> sb
->s_time_max
)) {
2554 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
2555 char *mntpath
= buf
? d_path(mountpoint
, buf
, PAGE_SIZE
) : ERR_PTR(-ENOMEM
);
2558 time64_to_tm(sb
->s_time_max
, 0, &tm
);
2560 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2562 is_mounted(mnt
) ? "remounted" : "mounted",
2564 tm
.tm_year
+1900, (unsigned long long)sb
->s_time_max
);
2566 free_page((unsigned long)buf
);
2571 * Handle reconfiguration of the mountpoint only without alteration of the
2572 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2575 static int do_reconfigure_mnt(struct path
*path
, unsigned int mnt_flags
)
2577 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2578 struct mount
*mnt
= real_mount(path
->mnt
);
2581 if (!check_mnt(mnt
))
2584 if (path
->dentry
!= mnt
->mnt
.mnt_root
)
2587 if (!can_change_locked_flags(mnt
, mnt_flags
))
2591 * We're only checking whether the superblock is read-only not
2592 * changing it, so only take down_read(&sb->s_umount).
2594 down_read(&sb
->s_umount
);
2596 ret
= change_mount_ro_state(mnt
, mnt_flags
);
2598 set_mount_attributes(mnt
, mnt_flags
);
2599 unlock_mount_hash();
2600 up_read(&sb
->s_umount
);
2602 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2608 * change filesystem flags. dir should be a physical root of filesystem.
2609 * If you've mounted a non-root directory somewhere and want to do remount
2610 * on it - tough luck.
2612 static int do_remount(struct path
*path
, int ms_flags
, int sb_flags
,
2613 int mnt_flags
, void *data
)
2616 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2617 struct mount
*mnt
= real_mount(path
->mnt
);
2618 struct fs_context
*fc
;
2620 if (!check_mnt(mnt
))
2623 if (path
->dentry
!= path
->mnt
->mnt_root
)
2626 if (!can_change_locked_flags(mnt
, mnt_flags
))
2629 fc
= fs_context_for_reconfigure(path
->dentry
, sb_flags
, MS_RMT_MASK
);
2634 err
= parse_monolithic_mount_data(fc
, data
);
2636 down_write(&sb
->s_umount
);
2638 if (ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
)) {
2639 err
= reconfigure_super(fc
);
2642 set_mount_attributes(mnt
, mnt_flags
);
2643 unlock_mount_hash();
2646 up_write(&sb
->s_umount
);
2649 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2655 static inline int tree_contains_unbindable(struct mount
*mnt
)
2658 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2659 if (IS_MNT_UNBINDABLE(p
))
2666 * Check that there aren't references to earlier/same mount namespaces in the
2667 * specified subtree. Such references can act as pins for mount namespaces
2668 * that aren't checked by the mount-cycle checking code, thereby allowing
2669 * cycles to be made.
2671 static bool check_for_nsfs_mounts(struct mount
*subtree
)
2677 for (p
= subtree
; p
; p
= next_mnt(p
, subtree
))
2678 if (mnt_ns_loop(p
->mnt
.mnt_root
))
2683 unlock_mount_hash();
2687 static int do_move_mount(struct path
*old_path
, struct path
*new_path
)
2689 struct mnt_namespace
*ns
;
2692 struct mount
*parent
;
2693 struct mountpoint
*mp
, *old_mp
;
2697 mp
= lock_mount(new_path
);
2701 old
= real_mount(old_path
->mnt
);
2702 p
= real_mount(new_path
->mnt
);
2703 parent
= old
->mnt_parent
;
2704 attached
= mnt_has_parent(old
);
2705 old_mp
= old
->mnt_mp
;
2709 /* The mountpoint must be in our namespace. */
2713 /* The thing moved must be mounted... */
2714 if (!is_mounted(&old
->mnt
))
2717 /* ... and either ours or the root of anon namespace */
2718 if (!(attached
? check_mnt(old
) : is_anon_ns(ns
)))
2721 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2724 if (old_path
->dentry
!= old_path
->mnt
->mnt_root
)
2727 if (d_is_dir(new_path
->dentry
) !=
2728 d_is_dir(old_path
->dentry
))
2731 * Don't move a mount residing in a shared parent.
2733 if (attached
&& IS_MNT_SHARED(parent
))
2736 * Don't move a mount tree containing unbindable mounts to a destination
2737 * mount which is shared.
2739 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2742 if (!check_for_nsfs_mounts(old
))
2744 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2748 err
= attach_recursive_mnt(old
, real_mount(new_path
->mnt
), mp
,
2753 /* if the mount is moved, it should no longer be expire
2755 list_del_init(&old
->mnt_expire
);
2757 put_mountpoint(old_mp
);
2762 mntput_no_expire(parent
);
2769 static int do_move_mount_old(struct path
*path
, const char *old_name
)
2771 struct path old_path
;
2774 if (!old_name
|| !*old_name
)
2777 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2781 err
= do_move_mount(&old_path
, path
);
2782 path_put(&old_path
);
2787 * add a mount into a namespace's mount tree
2789 static int do_add_mount(struct mount
*newmnt
, struct mountpoint
*mp
,
2790 struct path
*path
, int mnt_flags
)
2792 struct mount
*parent
= real_mount(path
->mnt
);
2794 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2796 if (unlikely(!check_mnt(parent
))) {
2797 /* that's acceptable only for automounts done in private ns */
2798 if (!(mnt_flags
& MNT_SHRINKABLE
))
2800 /* ... and for those we'd better have mountpoint still alive */
2801 if (!parent
->mnt_ns
)
2805 /* Refuse the same filesystem on the same mount point */
2806 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2807 path
->mnt
->mnt_root
== path
->dentry
)
2810 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2813 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2814 return graft_tree(newmnt
, parent
, mp
);
2817 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
);
2820 * Create a new mount using a superblock configuration and request it
2821 * be added to the namespace tree.
2823 static int do_new_mount_fc(struct fs_context
*fc
, struct path
*mountpoint
,
2824 unsigned int mnt_flags
)
2826 struct vfsmount
*mnt
;
2827 struct mountpoint
*mp
;
2828 struct super_block
*sb
= fc
->root
->d_sb
;
2831 error
= security_sb_kern_mount(sb
);
2832 if (!error
&& mount_too_revealing(sb
, &mnt_flags
))
2835 if (unlikely(error
)) {
2840 up_write(&sb
->s_umount
);
2842 mnt
= vfs_create_mount(fc
);
2844 return PTR_ERR(mnt
);
2846 mnt_warn_timestamp_expiry(mountpoint
, mnt
);
2848 mp
= lock_mount(mountpoint
);
2853 error
= do_add_mount(real_mount(mnt
), mp
, mountpoint
, mnt_flags
);
2861 * create a new mount for userspace and request it to be added into the
2864 static int do_new_mount(struct path
*path
, const char *fstype
, int sb_flags
,
2865 int mnt_flags
, const char *name
, void *data
)
2867 struct file_system_type
*type
;
2868 struct fs_context
*fc
;
2869 const char *subtype
= NULL
;
2875 type
= get_fs_type(fstype
);
2879 if (type
->fs_flags
& FS_HAS_SUBTYPE
) {
2880 subtype
= strchr(fstype
, '.');
2884 put_filesystem(type
);
2890 fc
= fs_context_for_mount(type
, sb_flags
);
2891 put_filesystem(type
);
2896 err
= vfs_parse_fs_string(fc
, "subtype",
2897 subtype
, strlen(subtype
));
2899 err
= vfs_parse_fs_string(fc
, "source", name
, strlen(name
));
2901 err
= parse_monolithic_mount_data(fc
, data
);
2902 if (!err
&& !mount_capable(fc
))
2905 err
= vfs_get_tree(fc
);
2907 err
= do_new_mount_fc(fc
, path
, mnt_flags
);
2913 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2915 struct dentry
*dentry
= path
->dentry
;
2916 struct mountpoint
*mp
;
2925 mnt
= real_mount(m
);
2926 /* The new mount record should have at least 2 refs to prevent it being
2927 * expired before we get a chance to add it
2929 BUG_ON(mnt_get_count(mnt
) < 2);
2931 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2932 m
->mnt_root
== dentry
) {
2938 * we don't want to use lock_mount() - in this case finding something
2939 * that overmounts our mountpoint to be means "quitely drop what we've
2940 * got", not "try to mount it on top".
2942 inode_lock(dentry
->d_inode
);
2944 if (unlikely(cant_mount(dentry
))) {
2946 goto discard_locked
;
2949 if (unlikely(__lookup_mnt(path
->mnt
, dentry
))) {
2952 goto discard_locked
;
2955 mp
= get_mountpoint(dentry
);
2958 goto discard_locked
;
2961 err
= do_add_mount(mnt
, mp
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2970 inode_unlock(dentry
->d_inode
);
2972 /* remove m from any expiration list it may be on */
2973 if (!list_empty(&mnt
->mnt_expire
)) {
2975 list_del_init(&mnt
->mnt_expire
);
2984 * mnt_set_expiry - Put a mount on an expiration list
2985 * @mnt: The mount to list.
2986 * @expiry_list: The list to add the mount to.
2988 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2992 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2996 EXPORT_SYMBOL(mnt_set_expiry
);
2999 * process a list of expirable mountpoints with the intent of discarding any
3000 * mountpoints that aren't in use and haven't been touched since last we came
3003 void mark_mounts_for_expiry(struct list_head
*mounts
)
3005 struct mount
*mnt
, *next
;
3006 LIST_HEAD(graveyard
);
3008 if (list_empty(mounts
))
3014 /* extract from the expiration list every vfsmount that matches the
3015 * following criteria:
3016 * - only referenced by its parent vfsmount
3017 * - still marked for expiry (marked on the last call here; marks are
3018 * cleared by mntput())
3020 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
3021 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
3022 propagate_mount_busy(mnt
, 1))
3024 list_move(&mnt
->mnt_expire
, &graveyard
);
3026 while (!list_empty(&graveyard
)) {
3027 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
3028 touch_mnt_namespace(mnt
->mnt_ns
);
3029 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3031 unlock_mount_hash();
3035 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
3038 * Ripoff of 'select_parent()'
3040 * search the list of submounts for a given mountpoint, and move any
3041 * shrinkable submounts to the 'graveyard' list.
3043 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
3045 struct mount
*this_parent
= parent
;
3046 struct list_head
*next
;
3050 next
= this_parent
->mnt_mounts
.next
;
3052 while (next
!= &this_parent
->mnt_mounts
) {
3053 struct list_head
*tmp
= next
;
3054 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
3057 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
3060 * Descend a level if the d_mounts list is non-empty.
3062 if (!list_empty(&mnt
->mnt_mounts
)) {
3067 if (!propagate_mount_busy(mnt
, 1)) {
3068 list_move_tail(&mnt
->mnt_expire
, graveyard
);
3073 * All done at this level ... ascend and resume the search
3075 if (this_parent
!= parent
) {
3076 next
= this_parent
->mnt_child
.next
;
3077 this_parent
= this_parent
->mnt_parent
;
3084 * process a list of expirable mountpoints with the intent of discarding any
3085 * submounts of a specific parent mountpoint
3087 * mount_lock must be held for write
3089 static void shrink_submounts(struct mount
*mnt
)
3091 LIST_HEAD(graveyard
);
3094 /* extract submounts of 'mountpoint' from the expiration list */
3095 while (select_submounts(mnt
, &graveyard
)) {
3096 while (!list_empty(&graveyard
)) {
3097 m
= list_first_entry(&graveyard
, struct mount
,
3099 touch_mnt_namespace(m
->mnt_ns
);
3100 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3105 static void *copy_mount_options(const void __user
* data
)
3108 unsigned left
, offset
;
3113 copy
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3115 return ERR_PTR(-ENOMEM
);
3117 left
= copy_from_user(copy
, data
, PAGE_SIZE
);
3120 * Not all architectures have an exact copy_from_user(). Resort to
3123 offset
= PAGE_SIZE
- left
;
3126 if (get_user(c
, (const char __user
*)data
+ offset
))
3133 if (left
== PAGE_SIZE
) {
3135 return ERR_PTR(-EFAULT
);
3141 static char *copy_mount_string(const void __user
*data
)
3143 return data
? strndup_user(data
, PATH_MAX
) : NULL
;
3147 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3148 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3150 * data is a (void *) that can point to any structure up to
3151 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3152 * information (or be NULL).
3154 * Pre-0.97 versions of mount() didn't have a flags word.
3155 * When the flags word was introduced its top half was required
3156 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3157 * Therefore, if this magic number is present, it carries no information
3158 * and must be discarded.
3160 int path_mount(const char *dev_name
, struct path
*path
,
3161 const char *type_page
, unsigned long flags
, void *data_page
)
3163 unsigned int mnt_flags
= 0, sb_flags
;
3167 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
3168 flags
&= ~MS_MGC_MSK
;
3170 /* Basic sanity checks */
3172 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
3174 if (flags
& MS_NOUSER
)
3177 ret
= security_sb_mount(dev_name
, path
, type_page
, flags
, data_page
);
3182 if ((flags
& SB_MANDLOCK
) && !may_mandlock())
3185 /* Default to relatime unless overriden */
3186 if (!(flags
& MS_NOATIME
))
3187 mnt_flags
|= MNT_RELATIME
;
3189 /* Separate the per-mountpoint flags */
3190 if (flags
& MS_NOSUID
)
3191 mnt_flags
|= MNT_NOSUID
;
3192 if (flags
& MS_NODEV
)
3193 mnt_flags
|= MNT_NODEV
;
3194 if (flags
& MS_NOEXEC
)
3195 mnt_flags
|= MNT_NOEXEC
;
3196 if (flags
& MS_NOATIME
)
3197 mnt_flags
|= MNT_NOATIME
;
3198 if (flags
& MS_NODIRATIME
)
3199 mnt_flags
|= MNT_NODIRATIME
;
3200 if (flags
& MS_STRICTATIME
)
3201 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
3202 if (flags
& MS_RDONLY
)
3203 mnt_flags
|= MNT_READONLY
;
3204 if (flags
& MS_NOSYMFOLLOW
)
3205 mnt_flags
|= MNT_NOSYMFOLLOW
;
3207 /* The default atime for remount is preservation */
3208 if ((flags
& MS_REMOUNT
) &&
3209 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
3210 MS_STRICTATIME
)) == 0)) {
3211 mnt_flags
&= ~MNT_ATIME_MASK
;
3212 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_ATIME_MASK
;
3215 sb_flags
= flags
& (SB_RDONLY
|
3224 if ((flags
& (MS_REMOUNT
| MS_BIND
)) == (MS_REMOUNT
| MS_BIND
))
3225 return do_reconfigure_mnt(path
, mnt_flags
);
3226 if (flags
& MS_REMOUNT
)
3227 return do_remount(path
, flags
, sb_flags
, mnt_flags
, data_page
);
3228 if (flags
& MS_BIND
)
3229 return do_loopback(path
, dev_name
, flags
& MS_REC
);
3230 if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
3231 return do_change_type(path
, flags
);
3232 if (flags
& MS_MOVE
)
3233 return do_move_mount_old(path
, dev_name
);
3235 return do_new_mount(path
, type_page
, sb_flags
, mnt_flags
, dev_name
,
3239 long do_mount(const char *dev_name
, const char __user
*dir_name
,
3240 const char *type_page
, unsigned long flags
, void *data_page
)
3245 ret
= user_path_at(AT_FDCWD
, dir_name
, LOOKUP_FOLLOW
, &path
);
3248 ret
= path_mount(dev_name
, &path
, type_page
, flags
, data_page
);
3253 static struct ucounts
*inc_mnt_namespaces(struct user_namespace
*ns
)
3255 return inc_ucount(ns
, current_euid(), UCOUNT_MNT_NAMESPACES
);
3258 static void dec_mnt_namespaces(struct ucounts
*ucounts
)
3260 dec_ucount(ucounts
, UCOUNT_MNT_NAMESPACES
);
3263 static void free_mnt_ns(struct mnt_namespace
*ns
)
3265 if (!is_anon_ns(ns
))
3266 ns_free_inum(&ns
->ns
);
3267 dec_mnt_namespaces(ns
->ucounts
);
3268 put_user_ns(ns
->user_ns
);
3273 * Assign a sequence number so we can detect when we attempt to bind
3274 * mount a reference to an older mount namespace into the current
3275 * mount namespace, preventing reference counting loops. A 64bit
3276 * number incrementing at 10Ghz will take 12,427 years to wrap which
3277 * is effectively never, so we can ignore the possibility.
3279 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
3281 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
, bool anon
)
3283 struct mnt_namespace
*new_ns
;
3284 struct ucounts
*ucounts
;
3287 ucounts
= inc_mnt_namespaces(user_ns
);
3289 return ERR_PTR(-ENOSPC
);
3291 new_ns
= kzalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
3293 dec_mnt_namespaces(ucounts
);
3294 return ERR_PTR(-ENOMEM
);
3297 ret
= ns_alloc_inum(&new_ns
->ns
);
3300 dec_mnt_namespaces(ucounts
);
3301 return ERR_PTR(ret
);
3304 new_ns
->ns
.ops
= &mntns_operations
;
3306 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
3307 refcount_set(&new_ns
->ns
.count
, 1);
3308 INIT_LIST_HEAD(&new_ns
->list
);
3309 init_waitqueue_head(&new_ns
->poll
);
3310 spin_lock_init(&new_ns
->ns_lock
);
3311 new_ns
->user_ns
= get_user_ns(user_ns
);
3312 new_ns
->ucounts
= ucounts
;
3317 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
3318 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
3320 struct mnt_namespace
*new_ns
;
3321 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
3322 struct mount
*p
, *q
;
3329 if (likely(!(flags
& CLONE_NEWNS
))) {
3336 new_ns
= alloc_mnt_ns(user_ns
, false);
3341 /* First pass: copy the tree topology */
3342 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
3343 if (user_ns
!= ns
->user_ns
)
3344 copy_flags
|= CL_SHARED_TO_SLAVE
;
3345 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
3348 free_mnt_ns(new_ns
);
3349 return ERR_CAST(new);
3351 if (user_ns
!= ns
->user_ns
) {
3354 unlock_mount_hash();
3357 list_add_tail(&new_ns
->list
, &new->mnt_list
);
3360 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3361 * as belonging to new namespace. We have already acquired a private
3362 * fs_struct, so tsk->fs->lock is not needed.
3370 if (&p
->mnt
== new_fs
->root
.mnt
) {
3371 new_fs
->root
.mnt
= mntget(&q
->mnt
);
3374 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
3375 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
3379 p
= next_mnt(p
, old
);
3380 q
= next_mnt(q
, new);
3383 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
3384 p
= next_mnt(p
, old
);
3396 struct dentry
*mount_subtree(struct vfsmount
*m
, const char *name
)
3398 struct mount
*mnt
= real_mount(m
);
3399 struct mnt_namespace
*ns
;
3400 struct super_block
*s
;
3404 ns
= alloc_mnt_ns(&init_user_ns
, true);
3407 return ERR_CAST(ns
);
3412 list_add(&mnt
->mnt_list
, &ns
->list
);
3414 err
= vfs_path_lookup(m
->mnt_root
, m
,
3415 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
3420 return ERR_PTR(err
);
3422 /* trade a vfsmount reference for active sb one */
3423 s
= path
.mnt
->mnt_sb
;
3424 atomic_inc(&s
->s_active
);
3426 /* lock the sucker */
3427 down_write(&s
->s_umount
);
3428 /* ... and return the root of (sub)tree on it */
3431 EXPORT_SYMBOL(mount_subtree
);
3433 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
3434 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3441 kernel_type
= copy_mount_string(type
);
3442 ret
= PTR_ERR(kernel_type
);
3443 if (IS_ERR(kernel_type
))
3446 kernel_dev
= copy_mount_string(dev_name
);
3447 ret
= PTR_ERR(kernel_dev
);
3448 if (IS_ERR(kernel_dev
))
3451 options
= copy_mount_options(data
);
3452 ret
= PTR_ERR(options
);
3453 if (IS_ERR(options
))
3456 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
, options
);
3467 #define FSMOUNT_VALID_FLAGS \
3468 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3469 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3470 MOUNT_ATTR_NOSYMFOLLOW)
3472 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3474 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3475 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3477 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags
)
3479 unsigned int mnt_flags
= 0;
3481 if (attr_flags
& MOUNT_ATTR_RDONLY
)
3482 mnt_flags
|= MNT_READONLY
;
3483 if (attr_flags
& MOUNT_ATTR_NOSUID
)
3484 mnt_flags
|= MNT_NOSUID
;
3485 if (attr_flags
& MOUNT_ATTR_NODEV
)
3486 mnt_flags
|= MNT_NODEV
;
3487 if (attr_flags
& MOUNT_ATTR_NOEXEC
)
3488 mnt_flags
|= MNT_NOEXEC
;
3489 if (attr_flags
& MOUNT_ATTR_NODIRATIME
)
3490 mnt_flags
|= MNT_NODIRATIME
;
3491 if (attr_flags
& MOUNT_ATTR_NOSYMFOLLOW
)
3492 mnt_flags
|= MNT_NOSYMFOLLOW
;
3498 * Create a kernel mount representation for a new, prepared superblock
3499 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3501 SYSCALL_DEFINE3(fsmount
, int, fs_fd
, unsigned int, flags
,
3502 unsigned int, attr_flags
)
3504 struct mnt_namespace
*ns
;
3505 struct fs_context
*fc
;
3507 struct path newmount
;
3510 unsigned int mnt_flags
= 0;
3516 if ((flags
& ~(FSMOUNT_CLOEXEC
)) != 0)
3519 if (attr_flags
& ~FSMOUNT_VALID_FLAGS
)
3522 mnt_flags
= attr_flags_to_mnt_flags(attr_flags
);
3524 switch (attr_flags
& MOUNT_ATTR__ATIME
) {
3525 case MOUNT_ATTR_STRICTATIME
:
3527 case MOUNT_ATTR_NOATIME
:
3528 mnt_flags
|= MNT_NOATIME
;
3530 case MOUNT_ATTR_RELATIME
:
3531 mnt_flags
|= MNT_RELATIME
;
3542 if (f
.file
->f_op
!= &fscontext_fops
)
3545 fc
= f
.file
->private_data
;
3547 ret
= mutex_lock_interruptible(&fc
->uapi_mutex
);
3551 /* There must be a valid superblock or we can't mount it */
3557 if (mount_too_revealing(fc
->root
->d_sb
, &mnt_flags
)) {
3558 pr_warn("VFS: Mount too revealing\n");
3563 if (fc
->phase
!= FS_CONTEXT_AWAITING_MOUNT
)
3567 if ((fc
->sb_flags
& SB_MANDLOCK
) && !may_mandlock())
3570 newmount
.mnt
= vfs_create_mount(fc
);
3571 if (IS_ERR(newmount
.mnt
)) {
3572 ret
= PTR_ERR(newmount
.mnt
);
3575 newmount
.dentry
= dget(fc
->root
);
3576 newmount
.mnt
->mnt_flags
= mnt_flags
;
3578 /* We've done the mount bit - now move the file context into more or
3579 * less the same state as if we'd done an fspick(). We don't want to
3580 * do any memory allocation or anything like that at this point as we
3581 * don't want to have to handle any errors incurred.
3583 vfs_clean_context(fc
);
3585 ns
= alloc_mnt_ns(current
->nsproxy
->mnt_ns
->user_ns
, true);
3590 mnt
= real_mount(newmount
.mnt
);
3594 list_add(&mnt
->mnt_list
, &ns
->list
);
3595 mntget(newmount
.mnt
);
3597 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3598 * it, not just simply put it.
3600 file
= dentry_open(&newmount
, O_PATH
, fc
->cred
);
3602 dissolve_on_fput(newmount
.mnt
);
3603 ret
= PTR_ERR(file
);
3606 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
3608 ret
= get_unused_fd_flags((flags
& FSMOUNT_CLOEXEC
) ? O_CLOEXEC
: 0);
3610 fd_install(ret
, file
);
3615 path_put(&newmount
);
3617 mutex_unlock(&fc
->uapi_mutex
);
3624 * Move a mount from one place to another. In combination with
3625 * fsopen()/fsmount() this is used to install a new mount and in combination
3626 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3629 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3631 SYSCALL_DEFINE5(move_mount
,
3632 int, from_dfd
, const char __user
*, from_pathname
,
3633 int, to_dfd
, const char __user
*, to_pathname
,
3634 unsigned int, flags
)
3636 struct path from_path
, to_path
;
3637 unsigned int lflags
;
3643 if (flags
& ~MOVE_MOUNT__MASK
)
3646 /* If someone gives a pathname, they aren't permitted to move
3647 * from an fd that requires unmount as we can't get at the flag
3648 * to clear it afterwards.
3651 if (flags
& MOVE_MOUNT_F_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3652 if (flags
& MOVE_MOUNT_F_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3653 if (flags
& MOVE_MOUNT_F_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3655 ret
= user_path_at(from_dfd
, from_pathname
, lflags
, &from_path
);
3660 if (flags
& MOVE_MOUNT_T_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3661 if (flags
& MOVE_MOUNT_T_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3662 if (flags
& MOVE_MOUNT_T_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3664 ret
= user_path_at(to_dfd
, to_pathname
, lflags
, &to_path
);
3668 ret
= security_move_mount(&from_path
, &to_path
);
3672 ret
= do_move_mount(&from_path
, &to_path
);
3677 path_put(&from_path
);
3682 * Return true if path is reachable from root
3684 * namespace_sem or mount_lock is held
3686 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
3687 const struct path
*root
)
3689 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
3690 dentry
= mnt
->mnt_mountpoint
;
3691 mnt
= mnt
->mnt_parent
;
3693 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
3696 bool path_is_under(const struct path
*path1
, const struct path
*path2
)
3699 read_seqlock_excl(&mount_lock
);
3700 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
3701 read_sequnlock_excl(&mount_lock
);
3704 EXPORT_SYMBOL(path_is_under
);
3707 * pivot_root Semantics:
3708 * Moves the root file system of the current process to the directory put_old,
3709 * makes new_root as the new root file system of the current process, and sets
3710 * root/cwd of all processes which had them on the current root to new_root.
3713 * The new_root and put_old must be directories, and must not be on the
3714 * same file system as the current process root. The put_old must be
3715 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3716 * pointed to by put_old must yield the same directory as new_root. No other
3717 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3719 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3720 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3721 * in this situation.
3724 * - we don't move root/cwd if they are not at the root (reason: if something
3725 * cared enough to change them, it's probably wrong to force them elsewhere)
3726 * - it's okay to pick a root that isn't the root of a file system, e.g.
3727 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3728 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3731 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3732 const char __user
*, put_old
)
3734 struct path
new, old
, root
;
3735 struct mount
*new_mnt
, *root_mnt
, *old_mnt
, *root_parent
, *ex_parent
;
3736 struct mountpoint
*old_mp
, *root_mp
;
3742 error
= user_path_at(AT_FDCWD
, new_root
,
3743 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &new);
3747 error
= user_path_at(AT_FDCWD
, put_old
,
3748 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old
);
3752 error
= security_sb_pivotroot(&old
, &new);
3756 get_fs_root(current
->fs
, &root
);
3757 old_mp
= lock_mount(&old
);
3758 error
= PTR_ERR(old_mp
);
3763 new_mnt
= real_mount(new.mnt
);
3764 root_mnt
= real_mount(root
.mnt
);
3765 old_mnt
= real_mount(old
.mnt
);
3766 ex_parent
= new_mnt
->mnt_parent
;
3767 root_parent
= root_mnt
->mnt_parent
;
3768 if (IS_MNT_SHARED(old_mnt
) ||
3769 IS_MNT_SHARED(ex_parent
) ||
3770 IS_MNT_SHARED(root_parent
))
3772 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3774 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3777 if (d_unlinked(new.dentry
))
3780 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3781 goto out4
; /* loop, on the same file system */
3783 if (root
.mnt
->mnt_root
!= root
.dentry
)
3784 goto out4
; /* not a mountpoint */
3785 if (!mnt_has_parent(root_mnt
))
3786 goto out4
; /* not attached */
3787 if (new.mnt
->mnt_root
!= new.dentry
)
3788 goto out4
; /* not a mountpoint */
3789 if (!mnt_has_parent(new_mnt
))
3790 goto out4
; /* not attached */
3791 /* make sure we can reach put_old from new_root */
3792 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3794 /* make certain new is below the root */
3795 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3798 umount_mnt(new_mnt
);
3799 root_mp
= unhash_mnt(root_mnt
); /* we'll need its mountpoint */
3800 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3801 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3802 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3804 /* mount old root on put_old */
3805 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3806 /* mount new_root on / */
3807 attach_mnt(new_mnt
, root_parent
, root_mp
);
3808 mnt_add_count(root_parent
, -1);
3809 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3810 /* A moved mount should not expire automatically */
3811 list_del_init(&new_mnt
->mnt_expire
);
3812 put_mountpoint(root_mp
);
3813 unlock_mount_hash();
3814 chroot_fs_refs(&root
, &new);
3817 unlock_mount(old_mp
);
3819 mntput_no_expire(ex_parent
);
3830 static unsigned int recalc_flags(struct mount_kattr
*kattr
, struct mount
*mnt
)
3832 unsigned int flags
= mnt
->mnt
.mnt_flags
;
3834 /* flags to clear */
3835 flags
&= ~kattr
->attr_clr
;
3836 /* flags to raise */
3837 flags
|= kattr
->attr_set
;
3842 static int can_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
3844 struct vfsmount
*m
= &mnt
->mnt
;
3846 if (!kattr
->mnt_userns
)
3850 * Once a mount has been idmapped we don't allow it to change its
3851 * mapping. It makes things simpler and callers can just create
3852 * another bind-mount they can idmap if they want to.
3854 if (mnt_user_ns(m
) != &init_user_ns
)
3857 /* The underlying filesystem doesn't support idmapped mounts yet. */
3858 if (!(m
->mnt_sb
->s_type
->fs_flags
& FS_ALLOW_IDMAP
))
3861 /* Don't yet support filesystem mountable in user namespaces. */
3862 if (m
->mnt_sb
->s_user_ns
!= &init_user_ns
)
3865 /* We're not controlling the superblock. */
3866 if (!capable(CAP_SYS_ADMIN
))
3869 /* Mount has already been visible in the filesystem hierarchy. */
3870 if (!is_anon_ns(mnt
->mnt_ns
))
3876 static struct mount
*mount_setattr_prepare(struct mount_kattr
*kattr
,
3877 struct mount
*mnt
, int *err
)
3879 struct mount
*m
= mnt
, *last
= NULL
;
3881 if (!is_mounted(&m
->mnt
)) {
3886 if (!(mnt_has_parent(m
) ? check_mnt(m
) : is_anon_ns(m
->mnt_ns
))) {
3894 flags
= recalc_flags(kattr
, m
);
3895 if (!can_change_locked_flags(m
, flags
)) {
3900 *err
= can_idmap_mount(kattr
, m
);
3906 if ((kattr
->attr_set
& MNT_READONLY
) &&
3907 !(m
->mnt
.mnt_flags
& MNT_READONLY
)) {
3908 *err
= mnt_hold_writers(m
);
3912 } while (kattr
->recurse
&& (m
= next_mnt(m
, mnt
)));
3918 static void do_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
3920 struct user_namespace
*mnt_userns
;
3922 if (!kattr
->mnt_userns
)
3925 mnt_userns
= get_user_ns(kattr
->mnt_userns
);
3926 /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3927 smp_store_release(&mnt
->mnt
.mnt_userns
, mnt_userns
);
3930 static void mount_setattr_commit(struct mount_kattr
*kattr
,
3931 struct mount
*mnt
, struct mount
*last
,
3934 struct mount
*m
= mnt
;
3940 do_idmap_mount(kattr
, m
);
3941 flags
= recalc_flags(kattr
, m
);
3942 WRITE_ONCE(m
->mnt
.mnt_flags
, flags
);
3946 * We either set MNT_READONLY above so make it visible
3947 * before ~MNT_WRITE_HOLD or we failed to recursively
3948 * apply mount options.
3950 if ((kattr
->attr_set
& MNT_READONLY
) &&
3951 (m
->mnt
.mnt_flags
& MNT_WRITE_HOLD
))
3952 mnt_unhold_writers(m
);
3954 if (!err
&& kattr
->propagation
)
3955 change_mnt_propagation(m
, kattr
->propagation
);
3958 * On failure, only cleanup until we found the first mount
3959 * we failed to handle.
3961 if (err
&& m
== last
)
3963 } while (kattr
->recurse
&& (m
= next_mnt(m
, mnt
)));
3966 touch_mnt_namespace(mnt
->mnt_ns
);
3969 static int do_mount_setattr(struct path
*path
, struct mount_kattr
*kattr
)
3971 struct mount
*mnt
= real_mount(path
->mnt
), *last
= NULL
;
3974 if (path
->dentry
!= mnt
->mnt
.mnt_root
)
3977 if (kattr
->propagation
) {
3979 * Only take namespace_lock() if we're actually changing
3983 if (kattr
->propagation
== MS_SHARED
) {
3984 err
= invent_group_ids(mnt
, kattr
->recurse
);
3995 * Get the mount tree in a shape where we can change mount
3996 * properties without failure.
3998 last
= mount_setattr_prepare(kattr
, mnt
, &err
);
3999 if (last
) /* Commit all changes or revert to the old state. */
4000 mount_setattr_commit(kattr
, mnt
, last
, err
);
4002 unlock_mount_hash();
4004 if (kattr
->propagation
) {
4007 cleanup_group_ids(mnt
, NULL
);
4013 static int build_mount_idmapped(const struct mount_attr
*attr
, size_t usize
,
4014 struct mount_kattr
*kattr
, unsigned int flags
)
4017 struct ns_common
*ns
;
4018 struct user_namespace
*mnt_userns
;
4021 if (!((attr
->attr_set
| attr
->attr_clr
) & MOUNT_ATTR_IDMAP
))
4025 * We currently do not support clearing an idmapped mount. If this ever
4026 * is a use-case we can revisit this but for now let's keep it simple
4029 if (attr
->attr_clr
& MOUNT_ATTR_IDMAP
)
4032 if (attr
->userns_fd
> INT_MAX
)
4035 file
= fget(attr
->userns_fd
);
4039 if (!proc_ns_file(file
)) {
4044 ns
= get_proc_ns(file_inode(file
));
4045 if (ns
->ops
->type
!= CLONE_NEWUSER
) {
4051 * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4052 * This is simpler than just having to treat NULL as unmapped. Users
4053 * wanting to idmap a mount to init_user_ns can just use a namespace
4054 * with an identity mapping.
4056 mnt_userns
= container_of(ns
, struct user_namespace
, ns
);
4057 if (mnt_userns
== &init_user_ns
) {
4061 kattr
->mnt_userns
= get_user_ns(mnt_userns
);
4068 static int build_mount_kattr(const struct mount_attr
*attr
, size_t usize
,
4069 struct mount_kattr
*kattr
, unsigned int flags
)
4071 unsigned int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
4073 if (flags
& AT_NO_AUTOMOUNT
)
4074 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
4075 if (flags
& AT_SYMLINK_NOFOLLOW
)
4076 lookup_flags
&= ~LOOKUP_FOLLOW
;
4077 if (flags
& AT_EMPTY_PATH
)
4078 lookup_flags
|= LOOKUP_EMPTY
;
4080 *kattr
= (struct mount_kattr
) {
4081 .lookup_flags
= lookup_flags
,
4082 .recurse
= !!(flags
& AT_RECURSIVE
),
4085 if (attr
->propagation
& ~MOUNT_SETATTR_PROPAGATION_FLAGS
)
4087 if (hweight32(attr
->propagation
& MOUNT_SETATTR_PROPAGATION_FLAGS
) > 1)
4089 kattr
->propagation
= attr
->propagation
;
4091 if ((attr
->attr_set
| attr
->attr_clr
) & ~MOUNT_SETATTR_VALID_FLAGS
)
4094 kattr
->attr_set
= attr_flags_to_mnt_flags(attr
->attr_set
);
4095 kattr
->attr_clr
= attr_flags_to_mnt_flags(attr
->attr_clr
);
4098 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4099 * users wanting to transition to a different atime setting cannot
4100 * simply specify the atime setting in @attr_set, but must also
4101 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4102 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4103 * @attr_clr and that @attr_set can't have any atime bits set if
4104 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4106 if (attr
->attr_clr
& MOUNT_ATTR__ATIME
) {
4107 if ((attr
->attr_clr
& MOUNT_ATTR__ATIME
) != MOUNT_ATTR__ATIME
)
4111 * Clear all previous time settings as they are mutually
4114 kattr
->attr_clr
|= MNT_RELATIME
| MNT_NOATIME
;
4115 switch (attr
->attr_set
& MOUNT_ATTR__ATIME
) {
4116 case MOUNT_ATTR_RELATIME
:
4117 kattr
->attr_set
|= MNT_RELATIME
;
4119 case MOUNT_ATTR_NOATIME
:
4120 kattr
->attr_set
|= MNT_NOATIME
;
4122 case MOUNT_ATTR_STRICTATIME
:
4128 if (attr
->attr_set
& MOUNT_ATTR__ATIME
)
4132 return build_mount_idmapped(attr
, usize
, kattr
, flags
);
4135 static void finish_mount_kattr(struct mount_kattr
*kattr
)
4137 put_user_ns(kattr
->mnt_userns
);
4138 kattr
->mnt_userns
= NULL
;
4141 SYSCALL_DEFINE5(mount_setattr
, int, dfd
, const char __user
*, path
,
4142 unsigned int, flags
, struct mount_attr __user
*, uattr
,
4147 struct mount_attr attr
;
4148 struct mount_kattr kattr
;
4150 BUILD_BUG_ON(sizeof(struct mount_attr
) != MOUNT_ATTR_SIZE_VER0
);
4152 if (flags
& ~(AT_EMPTY_PATH
|
4154 AT_SYMLINK_NOFOLLOW
|
4158 if (unlikely(usize
> PAGE_SIZE
))
4160 if (unlikely(usize
< MOUNT_ATTR_SIZE_VER0
))
4166 err
= copy_struct_from_user(&attr
, sizeof(attr
), uattr
, usize
);
4170 /* Don't bother walking through the mounts if this is a nop. */
4171 if (attr
.attr_set
== 0 &&
4172 attr
.attr_clr
== 0 &&
4173 attr
.propagation
== 0)
4176 err
= build_mount_kattr(&attr
, usize
, &kattr
, flags
);
4180 err
= user_path_at(dfd
, path
, kattr
.lookup_flags
, &target
);
4184 err
= do_mount_setattr(&target
, &kattr
);
4185 finish_mount_kattr(&kattr
);
4190 static void __init
init_mount_tree(void)
4192 struct vfsmount
*mnt
;
4194 struct mnt_namespace
*ns
;
4197 mnt
= vfs_kern_mount(&rootfs_fs_type
, 0, "rootfs", NULL
);
4199 panic("Can't create rootfs");
4201 ns
= alloc_mnt_ns(&init_user_ns
, false);
4203 panic("Can't allocate initial namespace");
4204 m
= real_mount(mnt
);
4208 list_add(&m
->mnt_list
, &ns
->list
);
4209 init_task
.nsproxy
->mnt_ns
= ns
;
4213 root
.dentry
= mnt
->mnt_root
;
4214 mnt
->mnt_flags
|= MNT_LOCKED
;
4216 set_fs_pwd(current
->fs
, &root
);
4217 set_fs_root(current
->fs
, &root
);
4220 void __init
mnt_init(void)
4224 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
4225 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
4227 mount_hashtable
= alloc_large_system_hash("Mount-cache",
4228 sizeof(struct hlist_head
),
4231 &m_hash_shift
, &m_hash_mask
, 0, 0);
4232 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
4233 sizeof(struct hlist_head
),
4236 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
4238 if (!mount_hashtable
|| !mountpoint_hashtable
)
4239 panic("Failed to allocate mount hash table\n");
4245 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
4247 fs_kobj
= kobject_create_and_add("fs", NULL
);
4249 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
4255 void put_mnt_ns(struct mnt_namespace
*ns
)
4257 if (!refcount_dec_and_test(&ns
->ns
.count
))
4259 drop_collected_mounts(&ns
->root
->mnt
);
4263 struct vfsmount
*kern_mount(struct file_system_type
*type
)
4265 struct vfsmount
*mnt
;
4266 mnt
= vfs_kern_mount(type
, SB_KERNMOUNT
, type
->name
, NULL
);
4269 * it is a longterm mount, don't release mnt until
4270 * we unmount before file sys is unregistered
4272 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
4276 EXPORT_SYMBOL_GPL(kern_mount
);
4278 void kern_unmount(struct vfsmount
*mnt
)
4280 /* release long term mount so mount point can be released */
4281 if (!IS_ERR_OR_NULL(mnt
)) {
4282 real_mount(mnt
)->mnt_ns
= NULL
;
4283 synchronize_rcu(); /* yecchhh... */
4287 EXPORT_SYMBOL(kern_unmount
);
4289 void kern_unmount_array(struct vfsmount
*mnt
[], unsigned int num
)
4293 for (i
= 0; i
< num
; i
++)
4295 real_mount(mnt
[i
])->mnt_ns
= NULL
;
4296 synchronize_rcu_expedited();
4297 for (i
= 0; i
< num
; i
++)
4300 EXPORT_SYMBOL(kern_unmount_array
);
4302 bool our_mnt(struct vfsmount
*mnt
)
4304 return check_mnt(real_mount(mnt
));
4307 bool current_chrooted(void)
4309 /* Does the current process have a non-standard root */
4310 struct path ns_root
;
4311 struct path fs_root
;
4314 /* Find the namespace root */
4315 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
4316 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
4318 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
4321 get_fs_root(current
->fs
, &fs_root
);
4323 chrooted
= !path_equal(&fs_root
, &ns_root
);
4331 static bool mnt_already_visible(struct mnt_namespace
*ns
,
4332 const struct super_block
*sb
,
4335 int new_flags
= *new_mnt_flags
;
4337 bool visible
= false;
4339 down_read(&namespace_sem
);
4341 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
4342 struct mount
*child
;
4345 if (mnt_is_cursor(mnt
))
4348 if (mnt
->mnt
.mnt_sb
->s_type
!= sb
->s_type
)
4351 /* This mount is not fully visible if it's root directory
4352 * is not the root directory of the filesystem.
4354 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
4357 /* A local view of the mount flags */
4358 mnt_flags
= mnt
->mnt
.mnt_flags
;
4360 /* Don't miss readonly hidden in the superblock flags */
4361 if (sb_rdonly(mnt
->mnt
.mnt_sb
))
4362 mnt_flags
|= MNT_LOCK_READONLY
;
4364 /* Verify the mount flags are equal to or more permissive
4365 * than the proposed new mount.
4367 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
4368 !(new_flags
& MNT_READONLY
))
4370 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
4371 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
4374 /* This mount is not fully visible if there are any
4375 * locked child mounts that cover anything except for
4376 * empty directories.
4378 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
4379 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
4380 /* Only worry about locked mounts */
4381 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
4383 /* Is the directory permanetly empty? */
4384 if (!is_empty_dir_inode(inode
))
4387 /* Preserve the locked attributes */
4388 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
4396 up_read(&namespace_sem
);
4400 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
)
4402 const unsigned long required_iflags
= SB_I_NOEXEC
| SB_I_NODEV
;
4403 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
4404 unsigned long s_iflags
;
4406 if (ns
->user_ns
== &init_user_ns
)
4409 /* Can this filesystem be too revealing? */
4410 s_iflags
= sb
->s_iflags
;
4411 if (!(s_iflags
& SB_I_USERNS_VISIBLE
))
4414 if ((s_iflags
& required_iflags
) != required_iflags
) {
4415 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4420 return !mnt_already_visible(ns
, sb
, new_mnt_flags
);
4423 bool mnt_may_suid(struct vfsmount
*mnt
)
4426 * Foreign mounts (accessed via fchdir or through /proc
4427 * symlinks) are always treated as if they are nosuid. This
4428 * prevents namespaces from trusting potentially unsafe
4429 * suid/sgid bits, file caps, or security labels that originate
4430 * in other namespaces.
4432 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
4433 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
4436 static struct ns_common
*mntns_get(struct task_struct
*task
)
4438 struct ns_common
*ns
= NULL
;
4439 struct nsproxy
*nsproxy
;
4442 nsproxy
= task
->nsproxy
;
4444 ns
= &nsproxy
->mnt_ns
->ns
;
4445 get_mnt_ns(to_mnt_ns(ns
));
4452 static void mntns_put(struct ns_common
*ns
)
4454 put_mnt_ns(to_mnt_ns(ns
));
4457 static int mntns_install(struct nsset
*nsset
, struct ns_common
*ns
)
4459 struct nsproxy
*nsproxy
= nsset
->nsproxy
;
4460 struct fs_struct
*fs
= nsset
->fs
;
4461 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
), *old_mnt_ns
;
4462 struct user_namespace
*user_ns
= nsset
->cred
->user_ns
;
4466 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
4467 !ns_capable(user_ns
, CAP_SYS_CHROOT
) ||
4468 !ns_capable(user_ns
, CAP_SYS_ADMIN
))
4471 if (is_anon_ns(mnt_ns
))
4478 old_mnt_ns
= nsproxy
->mnt_ns
;
4479 nsproxy
->mnt_ns
= mnt_ns
;
4482 err
= vfs_path_lookup(mnt_ns
->root
->mnt
.mnt_root
, &mnt_ns
->root
->mnt
,
4483 "/", LOOKUP_DOWN
, &root
);
4485 /* revert to old namespace */
4486 nsproxy
->mnt_ns
= old_mnt_ns
;
4491 put_mnt_ns(old_mnt_ns
);
4493 /* Update the pwd and root */
4494 set_fs_pwd(fs
, &root
);
4495 set_fs_root(fs
, &root
);
4501 static struct user_namespace
*mntns_owner(struct ns_common
*ns
)
4503 return to_mnt_ns(ns
)->user_ns
;
4506 const struct proc_ns_operations mntns_operations
= {
4508 .type
= CLONE_NEWNS
,
4511 .install
= mntns_install
,
4512 .owner
= mntns_owner
,