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
,
208 if (!mnt
->mnt_devname
)
213 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
215 goto out_free_devname
;
217 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
220 mnt
->mnt_writers
= 0;
223 INIT_HLIST_NODE(&mnt
->mnt_hash
);
224 INIT_LIST_HEAD(&mnt
->mnt_child
);
225 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
226 INIT_LIST_HEAD(&mnt
->mnt_list
);
227 INIT_LIST_HEAD(&mnt
->mnt_expire
);
228 INIT_LIST_HEAD(&mnt
->mnt_share
);
229 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
230 INIT_LIST_HEAD(&mnt
->mnt_slave
);
231 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
232 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
233 INIT_HLIST_HEAD(&mnt
->mnt_stuck_children
);
234 mnt
->mnt
.mnt_userns
= &init_user_ns
;
240 kfree_const(mnt
->mnt_devname
);
245 kmem_cache_free(mnt_cache
, mnt
);
250 * Most r/o checks on a fs are for operations that take
251 * discrete amounts of time, like a write() or unlink().
252 * We must keep track of when those operations start
253 * (for permission checks) and when they end, so that
254 * we can determine when writes are able to occur to
258 * __mnt_is_readonly: check whether a mount is read-only
259 * @mnt: the mount to check for its write status
261 * This shouldn't be used directly ouside of the VFS.
262 * It does not guarantee that the filesystem will stay
263 * r/w, just that it is right *now*. This can not and
264 * should not be used in place of IS_RDONLY(inode).
265 * mnt_want/drop_write() will _keep_ the filesystem
268 bool __mnt_is_readonly(struct vfsmount
*mnt
)
270 return (mnt
->mnt_flags
& MNT_READONLY
) || sb_rdonly(mnt
->mnt_sb
);
272 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
274 static inline void mnt_inc_writers(struct mount
*mnt
)
277 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
283 static inline void mnt_dec_writers(struct mount
*mnt
)
286 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
292 static unsigned int mnt_get_writers(struct mount
*mnt
)
295 unsigned int count
= 0;
298 for_each_possible_cpu(cpu
) {
299 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
304 return mnt
->mnt_writers
;
308 static int mnt_is_readonly(struct vfsmount
*mnt
)
310 if (mnt
->mnt_sb
->s_readonly_remount
)
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
314 return __mnt_is_readonly(mnt
);
318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
324 * __mnt_want_write - get write access to a mount without freeze protection
325 * @m: the mount on which to take a write
327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
333 int __mnt_want_write(struct vfsmount
*m
)
335 struct mount
*mnt
= real_mount(m
);
339 mnt_inc_writers(mnt
);
341 * The store to mnt_inc_writers must be visible before we pass
342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
346 while (READ_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
354 if (mnt_is_readonly(m
)) {
355 mnt_dec_writers(mnt
);
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
372 int mnt_want_write(struct vfsmount
*m
)
376 sb_start_write(m
->mnt_sb
);
377 ret
= __mnt_want_write(m
);
379 sb_end_write(m
->mnt_sb
);
382 EXPORT_SYMBOL_GPL(mnt_want_write
);
385 * __mnt_want_write_file - get write access to a file's mount
386 * @file: the file who's mount on which to take a write
388 * This is like __mnt_want_write, but if the file is already open for writing it
389 * skips incrementing mnt_writers (since the open file already has a reference)
390 * and instead only does the check for emergency r/o remounts. This must be
391 * paired with __mnt_drop_write_file.
393 int __mnt_want_write_file(struct file
*file
)
395 if (file
->f_mode
& FMODE_WRITER
) {
397 * Superblock may have become readonly while there are still
398 * writable fd's, e.g. due to a fs error with errors=remount-ro
400 if (__mnt_is_readonly(file
->f_path
.mnt
))
404 return __mnt_want_write(file
->f_path
.mnt
);
408 * mnt_want_write_file - get write access to a file's mount
409 * @file: the file who's mount on which to take a write
411 * This is like mnt_want_write, but if the file is already open for writing it
412 * skips incrementing mnt_writers (since the open file already has a reference)
413 * and instead only does the freeze protection and the check for emergency r/o
414 * remounts. This must be paired with mnt_drop_write_file.
416 int mnt_want_write_file(struct file
*file
)
420 sb_start_write(file_inode(file
)->i_sb
);
421 ret
= __mnt_want_write_file(file
);
423 sb_end_write(file_inode(file
)->i_sb
);
426 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
429 * __mnt_drop_write - give up write access to a mount
430 * @mnt: the mount on which to give up write access
432 * Tells the low-level filesystem that we are done
433 * performing writes to it. Must be matched with
434 * __mnt_want_write() call above.
436 void __mnt_drop_write(struct vfsmount
*mnt
)
439 mnt_dec_writers(real_mount(mnt
));
442 EXPORT_SYMBOL_GPL(__mnt_drop_write
);
445 * mnt_drop_write - give up write access to a mount
446 * @mnt: the mount on which to give up write access
448 * Tells the low-level filesystem that we are done performing writes to it and
449 * also allows filesystem to be frozen again. Must be matched with
450 * mnt_want_write() call above.
452 void mnt_drop_write(struct vfsmount
*mnt
)
454 __mnt_drop_write(mnt
);
455 sb_end_write(mnt
->mnt_sb
);
457 EXPORT_SYMBOL_GPL(mnt_drop_write
);
459 void __mnt_drop_write_file(struct file
*file
)
461 if (!(file
->f_mode
& FMODE_WRITER
))
462 __mnt_drop_write(file
->f_path
.mnt
);
465 void mnt_drop_write_file(struct file
*file
)
467 __mnt_drop_write_file(file
);
468 sb_end_write(file_inode(file
)->i_sb
);
470 EXPORT_SYMBOL(mnt_drop_write_file
);
472 static inline int mnt_hold_writers(struct mount
*mnt
)
474 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
476 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
477 * should be visible before we do.
482 * With writers on hold, if this value is zero, then there are
483 * definitely no active writers (although held writers may subsequently
484 * increment the count, they'll have to wait, and decrement it after
485 * seeing MNT_READONLY).
487 * It is OK to have counter incremented on one CPU and decremented on
488 * another: the sum will add up correctly. The danger would be when we
489 * sum up each counter, if we read a counter before it is incremented,
490 * but then read another CPU's count which it has been subsequently
491 * decremented from -- we would see more decrements than we should.
492 * MNT_WRITE_HOLD protects against this scenario, because
493 * mnt_want_write first increments count, then smp_mb, then spins on
494 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
495 * we're counting up here.
497 if (mnt_get_writers(mnt
) > 0)
503 static inline void mnt_unhold_writers(struct mount
*mnt
)
506 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
507 * that become unheld will see MNT_READONLY.
510 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
513 static int mnt_make_readonly(struct mount
*mnt
)
517 ret
= mnt_hold_writers(mnt
);
519 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
520 mnt_unhold_writers(mnt
);
524 int sb_prepare_remount_readonly(struct super_block
*sb
)
529 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
530 if (atomic_long_read(&sb
->s_remove_count
))
534 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
535 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
536 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
538 if (mnt_get_writers(mnt
) > 0) {
544 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
548 sb
->s_readonly_remount
= 1;
551 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
552 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
553 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
560 static void free_vfsmnt(struct mount
*mnt
)
562 struct user_namespace
*mnt_userns
;
564 mnt_userns
= mnt_user_ns(&mnt
->mnt
);
565 if (mnt_userns
!= &init_user_ns
)
566 put_user_ns(mnt_userns
);
567 kfree_const(mnt
->mnt_devname
);
569 free_percpu(mnt
->mnt_pcp
);
571 kmem_cache_free(mnt_cache
, mnt
);
574 static void delayed_free_vfsmnt(struct rcu_head
*head
)
576 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
579 /* call under rcu_read_lock */
580 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
583 if (read_seqretry(&mount_lock
, seq
))
587 mnt
= real_mount(bastard
);
588 mnt_add_count(mnt
, 1);
589 smp_mb(); // see mntput_no_expire()
590 if (likely(!read_seqretry(&mount_lock
, seq
)))
592 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
593 mnt_add_count(mnt
, -1);
597 if (unlikely(bastard
->mnt_flags
& MNT_DOOMED
)) {
598 mnt_add_count(mnt
, -1);
603 /* caller will mntput() */
607 /* call under rcu_read_lock */
608 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
610 int res
= __legitimize_mnt(bastard
, seq
);
613 if (unlikely(res
< 0)) {
622 * find the first mount at @dentry on vfsmount @mnt.
623 * call under rcu_read_lock()
625 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
627 struct hlist_head
*head
= m_hash(mnt
, dentry
);
630 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
631 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
637 * lookup_mnt - Return the first child mount mounted at path
639 * "First" means first mounted chronologically. If you create the
642 * mount /dev/sda1 /mnt
643 * mount /dev/sda2 /mnt
644 * mount /dev/sda3 /mnt
646 * Then lookup_mnt() on the base /mnt dentry in the root mount will
647 * return successively the root dentry and vfsmount of /dev/sda1, then
648 * /dev/sda2, then /dev/sda3, then NULL.
650 * lookup_mnt takes a reference to the found vfsmount.
652 struct vfsmount
*lookup_mnt(const struct path
*path
)
654 struct mount
*child_mnt
;
660 seq
= read_seqbegin(&mount_lock
);
661 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
662 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
663 } while (!legitimize_mnt(m
, seq
));
668 static inline void lock_ns_list(struct mnt_namespace
*ns
)
670 spin_lock(&ns
->ns_lock
);
673 static inline void unlock_ns_list(struct mnt_namespace
*ns
)
675 spin_unlock(&ns
->ns_lock
);
678 static inline bool mnt_is_cursor(struct mount
*mnt
)
680 return mnt
->mnt
.mnt_flags
& MNT_CURSOR
;
684 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685 * current mount namespace.
687 * The common case is dentries are not mountpoints at all and that
688 * test is handled inline. For the slow case when we are actually
689 * dealing with a mountpoint of some kind, walk through all of the
690 * mounts in the current mount namespace and test to see if the dentry
693 * The mount_hashtable is not usable in the context because we
694 * need to identify all mounts that may be in the current mount
695 * namespace not just a mount that happens to have some specified
698 bool __is_local_mountpoint(struct dentry
*dentry
)
700 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
702 bool is_covered
= false;
704 down_read(&namespace_sem
);
706 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
707 if (mnt_is_cursor(mnt
))
709 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
714 up_read(&namespace_sem
);
719 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
721 struct hlist_head
*chain
= mp_hash(dentry
);
722 struct mountpoint
*mp
;
724 hlist_for_each_entry(mp
, chain
, m_hash
) {
725 if (mp
->m_dentry
== dentry
) {
733 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
735 struct mountpoint
*mp
, *new = NULL
;
738 if (d_mountpoint(dentry
)) {
739 /* might be worth a WARN_ON() */
740 if (d_unlinked(dentry
))
741 return ERR_PTR(-ENOENT
);
743 read_seqlock_excl(&mount_lock
);
744 mp
= lookup_mountpoint(dentry
);
745 read_sequnlock_excl(&mount_lock
);
751 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
753 return ERR_PTR(-ENOMEM
);
756 /* Exactly one processes may set d_mounted */
757 ret
= d_set_mounted(dentry
);
759 /* Someone else set d_mounted? */
763 /* The dentry is not available as a mountpoint? */
768 /* Add the new mountpoint to the hash table */
769 read_seqlock_excl(&mount_lock
);
770 new->m_dentry
= dget(dentry
);
772 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
773 INIT_HLIST_HEAD(&new->m_list
);
774 read_sequnlock_excl(&mount_lock
);
784 * vfsmount lock must be held. Additionally, the caller is responsible
785 * for serializing calls for given disposal list.
787 static void __put_mountpoint(struct mountpoint
*mp
, struct list_head
*list
)
789 if (!--mp
->m_count
) {
790 struct dentry
*dentry
= mp
->m_dentry
;
791 BUG_ON(!hlist_empty(&mp
->m_list
));
792 spin_lock(&dentry
->d_lock
);
793 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
794 spin_unlock(&dentry
->d_lock
);
795 dput_to_list(dentry
, list
);
796 hlist_del(&mp
->m_hash
);
801 /* called with namespace_lock and vfsmount lock */
802 static void put_mountpoint(struct mountpoint
*mp
)
804 __put_mountpoint(mp
, &ex_mountpoints
);
807 static inline int check_mnt(struct mount
*mnt
)
809 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
812 /* for aufs, CONFIG_AUFS_BR_FUSE */
813 int is_current_mnt_ns(struct vfsmount
*mnt
)
815 return check_mnt(real_mount(mnt
));
817 EXPORT_SYMBOL_GPL(is_current_mnt_ns
);
820 * vfsmount lock must be held for write
822 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
826 wake_up_interruptible(&ns
->poll
);
831 * vfsmount lock must be held for write
833 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
835 if (ns
&& ns
->event
!= event
) {
837 wake_up_interruptible(&ns
->poll
);
842 * vfsmount lock must be held for write
844 static struct mountpoint
*unhash_mnt(struct mount
*mnt
)
846 struct mountpoint
*mp
;
847 mnt
->mnt_parent
= mnt
;
848 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
849 list_del_init(&mnt
->mnt_child
);
850 hlist_del_init_rcu(&mnt
->mnt_hash
);
851 hlist_del_init(&mnt
->mnt_mp_list
);
858 * vfsmount lock must be held for write
860 static void umount_mnt(struct mount
*mnt
)
862 put_mountpoint(unhash_mnt(mnt
));
866 * vfsmount lock must be held for write
868 void mnt_set_mountpoint(struct mount
*mnt
,
869 struct mountpoint
*mp
,
870 struct mount
*child_mnt
)
873 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
874 child_mnt
->mnt_mountpoint
= mp
->m_dentry
;
875 child_mnt
->mnt_parent
= mnt
;
876 child_mnt
->mnt_mp
= mp
;
877 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
880 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
882 hlist_add_head_rcu(&mnt
->mnt_hash
,
883 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
884 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
888 * vfsmount lock must be held for write
890 static void attach_mnt(struct mount
*mnt
,
891 struct mount
*parent
,
892 struct mountpoint
*mp
)
894 mnt_set_mountpoint(parent
, mp
, mnt
);
895 __attach_mnt(mnt
, parent
);
898 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
900 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
901 struct mount
*old_parent
= mnt
->mnt_parent
;
903 list_del_init(&mnt
->mnt_child
);
904 hlist_del_init(&mnt
->mnt_mp_list
);
905 hlist_del_init_rcu(&mnt
->mnt_hash
);
907 attach_mnt(mnt
, parent
, mp
);
909 put_mountpoint(old_mp
);
910 mnt_add_count(old_parent
, -1);
914 * vfsmount lock must be held for write
916 static void commit_tree(struct mount
*mnt
)
918 struct mount
*parent
= mnt
->mnt_parent
;
921 struct mnt_namespace
*n
= parent
->mnt_ns
;
923 BUG_ON(parent
== mnt
);
925 list_add_tail(&head
, &mnt
->mnt_list
);
926 list_for_each_entry(m
, &head
, mnt_list
)
929 list_splice(&head
, n
->list
.prev
);
931 n
->mounts
+= n
->pending_mounts
;
932 n
->pending_mounts
= 0;
934 __attach_mnt(mnt
, parent
);
935 touch_mnt_namespace(n
);
938 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
940 struct list_head
*next
= p
->mnt_mounts
.next
;
941 if (next
== &p
->mnt_mounts
) {
945 next
= p
->mnt_child
.next
;
946 if (next
!= &p
->mnt_parent
->mnt_mounts
)
951 return list_entry(next
, struct mount
, mnt_child
);
954 static struct mount
*skip_mnt_tree(struct mount
*p
)
956 struct list_head
*prev
= p
->mnt_mounts
.prev
;
957 while (prev
!= &p
->mnt_mounts
) {
958 p
= list_entry(prev
, struct mount
, mnt_child
);
959 prev
= p
->mnt_mounts
.prev
;
965 * vfs_create_mount - Create a mount for a configured superblock
966 * @fc: The configuration context with the superblock attached
968 * Create a mount to an already configured superblock. If necessary, the
969 * caller should invoke vfs_get_tree() before calling this.
971 * Note that this does not attach the mount to anything.
973 struct vfsmount
*vfs_create_mount(struct fs_context
*fc
)
978 return ERR_PTR(-EINVAL
);
980 mnt
= alloc_vfsmnt(fc
->source
?: "none");
982 return ERR_PTR(-ENOMEM
);
984 if (fc
->sb_flags
& SB_KERNMOUNT
)
985 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
987 atomic_inc(&fc
->root
->d_sb
->s_active
);
988 mnt
->mnt
.mnt_sb
= fc
->root
->d_sb
;
989 mnt
->mnt
.mnt_root
= dget(fc
->root
);
990 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
991 mnt
->mnt_parent
= mnt
;
994 list_add_tail(&mnt
->mnt_instance
, &mnt
->mnt
.mnt_sb
->s_mounts
);
998 EXPORT_SYMBOL(vfs_create_mount
);
1000 struct vfsmount
*fc_mount(struct fs_context
*fc
)
1002 int err
= vfs_get_tree(fc
);
1004 up_write(&fc
->root
->d_sb
->s_umount
);
1005 return vfs_create_mount(fc
);
1007 return ERR_PTR(err
);
1009 EXPORT_SYMBOL(fc_mount
);
1011 struct vfsmount
*vfs_kern_mount(struct file_system_type
*type
,
1012 int flags
, const char *name
,
1015 struct fs_context
*fc
;
1016 struct vfsmount
*mnt
;
1020 return ERR_PTR(-EINVAL
);
1022 fc
= fs_context_for_mount(type
, flags
);
1024 return ERR_CAST(fc
);
1027 ret
= vfs_parse_fs_string(fc
, "source",
1028 name
, strlen(name
));
1030 ret
= parse_monolithic_mount_data(fc
, data
);
1039 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1042 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1043 const char *name
, void *data
)
1045 /* Until it is worked out how to pass the user namespace
1046 * through from the parent mount to the submount don't support
1047 * unprivileged mounts with submounts.
1049 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1050 return ERR_PTR(-EPERM
);
1052 return vfs_kern_mount(type
, SB_SUBMOUNT
, name
, data
);
1054 EXPORT_SYMBOL_GPL(vfs_submount
);
1056 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1059 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1063 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1065 return ERR_PTR(-ENOMEM
);
1067 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1068 mnt
->mnt_group_id
= 0; /* not a peer of original */
1070 mnt
->mnt_group_id
= old
->mnt_group_id
;
1072 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1073 err
= mnt_alloc_group_id(mnt
);
1078 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
;
1079 mnt
->mnt
.mnt_flags
&= ~(MNT_WRITE_HOLD
|MNT_MARKED
|MNT_INTERNAL
);
1081 atomic_inc(&sb
->s_active
);
1082 mnt
->mnt
.mnt_userns
= mnt_user_ns(&old
->mnt
);
1083 if (mnt
->mnt
.mnt_userns
!= &init_user_ns
)
1084 mnt
->mnt
.mnt_userns
= get_user_ns(mnt
->mnt
.mnt_userns
);
1085 mnt
->mnt
.mnt_sb
= sb
;
1086 mnt
->mnt
.mnt_root
= dget(root
);
1087 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1088 mnt
->mnt_parent
= mnt
;
1090 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1091 unlock_mount_hash();
1093 if ((flag
& CL_SLAVE
) ||
1094 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1095 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1096 mnt
->mnt_master
= old
;
1097 CLEAR_MNT_SHARED(mnt
);
1098 } else if (!(flag
& CL_PRIVATE
)) {
1099 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1100 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1101 if (IS_MNT_SLAVE(old
))
1102 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1103 mnt
->mnt_master
= old
->mnt_master
;
1105 CLEAR_MNT_SHARED(mnt
);
1107 if (flag
& CL_MAKE_SHARED
)
1108 set_mnt_shared(mnt
);
1110 /* stick the duplicate mount on the same expiry list
1111 * as the original if that was on one */
1112 if (flag
& CL_EXPIRE
) {
1113 if (!list_empty(&old
->mnt_expire
))
1114 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1122 return ERR_PTR(err
);
1125 static void cleanup_mnt(struct mount
*mnt
)
1127 struct hlist_node
*p
;
1130 * The warning here probably indicates that somebody messed
1131 * up a mnt_want/drop_write() pair. If this happens, the
1132 * filesystem was probably unable to make r/w->r/o transitions.
1133 * The locking used to deal with mnt_count decrement provides barriers,
1134 * so mnt_get_writers() below is safe.
1136 WARN_ON(mnt_get_writers(mnt
));
1137 if (unlikely(mnt
->mnt_pins
.first
))
1139 hlist_for_each_entry_safe(m
, p
, &mnt
->mnt_stuck_children
, mnt_umount
) {
1140 hlist_del(&m
->mnt_umount
);
1143 fsnotify_vfsmount_delete(&mnt
->mnt
);
1144 dput(mnt
->mnt
.mnt_root
);
1145 deactivate_super(mnt
->mnt
.mnt_sb
);
1147 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1150 static void __cleanup_mnt(struct rcu_head
*head
)
1152 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1155 static LLIST_HEAD(delayed_mntput_list
);
1156 static void delayed_mntput(struct work_struct
*unused
)
1158 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1159 struct mount
*m
, *t
;
1161 llist_for_each_entry_safe(m
, t
, node
, mnt_llist
)
1164 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1166 static void mntput_no_expire(struct mount
*mnt
)
1172 if (likely(READ_ONCE(mnt
->mnt_ns
))) {
1174 * Since we don't do lock_mount_hash() here,
1175 * ->mnt_ns can change under us. However, if it's
1176 * non-NULL, then there's a reference that won't
1177 * be dropped until after an RCU delay done after
1178 * turning ->mnt_ns NULL. So if we observe it
1179 * non-NULL under rcu_read_lock(), the reference
1180 * we are dropping is not the final one.
1182 mnt_add_count(mnt
, -1);
1188 * make sure that if __legitimize_mnt() has not seen us grab
1189 * mount_lock, we'll see their refcount increment here.
1192 mnt_add_count(mnt
, -1);
1193 count
= mnt_get_count(mnt
);
1197 unlock_mount_hash();
1200 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1202 unlock_mount_hash();
1205 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1208 list_del(&mnt
->mnt_instance
);
1210 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1211 struct mount
*p
, *tmp
;
1212 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1213 __put_mountpoint(unhash_mnt(p
), &list
);
1214 hlist_add_head(&p
->mnt_umount
, &mnt
->mnt_stuck_children
);
1217 unlock_mount_hash();
1218 shrink_dentry_list(&list
);
1220 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1221 struct task_struct
*task
= current
;
1222 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1223 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1224 if (!task_work_add(task
, &mnt
->mnt_rcu
, TWA_RESUME
))
1227 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1228 schedule_delayed_work(&delayed_mntput_work
, 1);
1234 void mntput(struct vfsmount
*mnt
)
1237 struct mount
*m
= real_mount(mnt
);
1238 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1239 if (unlikely(m
->mnt_expiry_mark
))
1240 m
->mnt_expiry_mark
= 0;
1241 mntput_no_expire(m
);
1244 EXPORT_SYMBOL(mntput
);
1246 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1249 mnt_add_count(real_mount(mnt
), 1);
1252 EXPORT_SYMBOL(mntget
);
1255 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1256 * @path: path to check
1258 * d_mountpoint() can only be used reliably to establish if a dentry is
1259 * not mounted in any namespace and that common case is handled inline.
1260 * d_mountpoint() isn't aware of the possibility there may be multiple
1261 * mounts using a given dentry in a different namespace. This function
1262 * checks if the passed in path is a mountpoint rather than the dentry
1265 bool path_is_mountpoint(const struct path
*path
)
1270 if (!d_mountpoint(path
->dentry
))
1275 seq
= read_seqbegin(&mount_lock
);
1276 res
= __path_is_mountpoint(path
);
1277 } while (read_seqretry(&mount_lock
, seq
));
1282 EXPORT_SYMBOL(path_is_mountpoint
);
1284 struct vfsmount
*mnt_clone_internal(const struct path
*path
)
1287 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1290 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1294 #ifdef CONFIG_PROC_FS
1295 static struct mount
*mnt_list_next(struct mnt_namespace
*ns
,
1296 struct list_head
*p
)
1298 struct mount
*mnt
, *ret
= NULL
;
1301 list_for_each_continue(p
, &ns
->list
) {
1302 mnt
= list_entry(p
, typeof(*mnt
), mnt_list
);
1303 if (!mnt_is_cursor(mnt
)) {
1313 /* iterator; we want it to have access to namespace_sem, thus here... */
1314 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1316 struct proc_mounts
*p
= m
->private;
1317 struct list_head
*prev
;
1319 down_read(&namespace_sem
);
1321 prev
= &p
->ns
->list
;
1323 prev
= &p
->cursor
.mnt_list
;
1325 /* Read after we'd reached the end? */
1326 if (list_empty(prev
))
1330 return mnt_list_next(p
->ns
, prev
);
1333 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1335 struct proc_mounts
*p
= m
->private;
1336 struct mount
*mnt
= v
;
1339 return mnt_list_next(p
->ns
, &mnt
->mnt_list
);
1342 static void m_stop(struct seq_file
*m
, void *v
)
1344 struct proc_mounts
*p
= m
->private;
1345 struct mount
*mnt
= v
;
1347 lock_ns_list(p
->ns
);
1349 list_move_tail(&p
->cursor
.mnt_list
, &mnt
->mnt_list
);
1351 list_del_init(&p
->cursor
.mnt_list
);
1352 unlock_ns_list(p
->ns
);
1353 up_read(&namespace_sem
);
1356 static int m_show(struct seq_file
*m
, void *v
)
1358 struct proc_mounts
*p
= m
->private;
1359 struct mount
*r
= v
;
1360 return p
->show(m
, &r
->mnt
);
1363 const struct seq_operations mounts_op
= {
1370 void mnt_cursor_del(struct mnt_namespace
*ns
, struct mount
*cursor
)
1372 down_read(&namespace_sem
);
1374 list_del(&cursor
->mnt_list
);
1376 up_read(&namespace_sem
);
1378 #endif /* CONFIG_PROC_FS */
1381 * may_umount_tree - check if a mount tree is busy
1382 * @m: root of mount tree
1384 * This is called to check if a tree of mounts has any
1385 * open files, pwds, chroots or sub mounts that are
1388 int may_umount_tree(struct vfsmount
*m
)
1390 struct mount
*mnt
= real_mount(m
);
1391 int actual_refs
= 0;
1392 int minimum_refs
= 0;
1396 /* write lock needed for mnt_get_count */
1398 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1399 actual_refs
+= mnt_get_count(p
);
1402 unlock_mount_hash();
1404 if (actual_refs
> minimum_refs
)
1410 EXPORT_SYMBOL(may_umount_tree
);
1413 * may_umount - check if a mount point is busy
1414 * @mnt: root of mount
1416 * This is called to check if a mount point has any
1417 * open files, pwds, chroots or sub mounts. If the
1418 * mount has sub mounts this will return busy
1419 * regardless of whether the sub mounts are busy.
1421 * Doesn't take quota and stuff into account. IOW, in some cases it will
1422 * give false negatives. The main reason why it's here is that we need
1423 * a non-destructive way to look for easily umountable filesystems.
1425 int may_umount(struct vfsmount
*mnt
)
1428 down_read(&namespace_sem
);
1430 if (propagate_mount_busy(real_mount(mnt
), 2))
1432 unlock_mount_hash();
1433 up_read(&namespace_sem
);
1437 EXPORT_SYMBOL(may_umount
);
1439 static void namespace_unlock(void)
1441 struct hlist_head head
;
1442 struct hlist_node
*p
;
1446 hlist_move_list(&unmounted
, &head
);
1447 list_splice_init(&ex_mountpoints
, &list
);
1449 up_write(&namespace_sem
);
1451 shrink_dentry_list(&list
);
1453 if (likely(hlist_empty(&head
)))
1456 synchronize_rcu_expedited();
1458 hlist_for_each_entry_safe(m
, p
, &head
, mnt_umount
) {
1459 hlist_del(&m
->mnt_umount
);
1464 static inline void namespace_lock(void)
1466 down_write(&namespace_sem
);
1469 enum umount_tree_flags
{
1471 UMOUNT_PROPAGATE
= 2,
1472 UMOUNT_CONNECTED
= 4,
1475 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1477 /* Leaving mounts connected is only valid for lazy umounts */
1478 if (how
& UMOUNT_SYNC
)
1481 /* A mount without a parent has nothing to be connected to */
1482 if (!mnt_has_parent(mnt
))
1485 /* Because the reference counting rules change when mounts are
1486 * unmounted and connected, umounted mounts may not be
1487 * connected to mounted mounts.
1489 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1492 /* Has it been requested that the mount remain connected? */
1493 if (how
& UMOUNT_CONNECTED
)
1496 /* Is the mount locked such that it needs to remain connected? */
1497 if (IS_MNT_LOCKED(mnt
))
1500 /* By default disconnect the mount */
1505 * mount_lock must be held
1506 * namespace_sem must be held for write
1508 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1510 LIST_HEAD(tmp_list
);
1513 if (how
& UMOUNT_PROPAGATE
)
1514 propagate_mount_unlock(mnt
);
1516 /* Gather the mounts to umount */
1517 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1518 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1519 list_move(&p
->mnt_list
, &tmp_list
);
1522 /* Hide the mounts from mnt_mounts */
1523 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1524 list_del_init(&p
->mnt_child
);
1527 /* Add propogated mounts to the tmp_list */
1528 if (how
& UMOUNT_PROPAGATE
)
1529 propagate_umount(&tmp_list
);
1531 while (!list_empty(&tmp_list
)) {
1532 struct mnt_namespace
*ns
;
1534 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1535 list_del_init(&p
->mnt_expire
);
1536 list_del_init(&p
->mnt_list
);
1540 __touch_mnt_namespace(ns
);
1543 if (how
& UMOUNT_SYNC
)
1544 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1546 disconnect
= disconnect_mount(p
, how
);
1547 if (mnt_has_parent(p
)) {
1548 mnt_add_count(p
->mnt_parent
, -1);
1550 /* Don't forget about p */
1551 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1556 change_mnt_propagation(p
, MS_PRIVATE
);
1558 hlist_add_head(&p
->mnt_umount
, &unmounted
);
1562 static void shrink_submounts(struct mount
*mnt
);
1564 static int do_umount_root(struct super_block
*sb
)
1568 down_write(&sb
->s_umount
);
1569 if (!sb_rdonly(sb
)) {
1570 struct fs_context
*fc
;
1572 fc
= fs_context_for_reconfigure(sb
->s_root
, SB_RDONLY
,
1577 ret
= parse_monolithic_mount_data(fc
, NULL
);
1579 ret
= reconfigure_super(fc
);
1583 up_write(&sb
->s_umount
);
1587 static int do_umount(struct mount
*mnt
, int flags
)
1589 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1592 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1597 * Allow userspace to request a mountpoint be expired rather than
1598 * unmounting unconditionally. Unmount only happens if:
1599 * (1) the mark is already set (the mark is cleared by mntput())
1600 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1602 if (flags
& MNT_EXPIRE
) {
1603 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1604 flags
& (MNT_FORCE
| MNT_DETACH
))
1608 * probably don't strictly need the lock here if we examined
1609 * all race cases, but it's a slowpath.
1612 if (mnt_get_count(mnt
) != 2) {
1613 unlock_mount_hash();
1616 unlock_mount_hash();
1618 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1623 * If we may have to abort operations to get out of this
1624 * mount, and they will themselves hold resources we must
1625 * allow the fs to do things. In the Unix tradition of
1626 * 'Gee thats tricky lets do it in userspace' the umount_begin
1627 * might fail to complete on the first run through as other tasks
1628 * must return, and the like. Thats for the mount program to worry
1629 * about for the moment.
1632 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1633 sb
->s_op
->umount_begin(sb
);
1637 * No sense to grab the lock for this test, but test itself looks
1638 * somewhat bogus. Suggestions for better replacement?
1639 * Ho-hum... In principle, we might treat that as umount + switch
1640 * to rootfs. GC would eventually take care of the old vfsmount.
1641 * Actually it makes sense, especially if rootfs would contain a
1642 * /reboot - static binary that would close all descriptors and
1643 * call reboot(9). Then init(8) could umount root and exec /reboot.
1645 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1647 * Special case for "unmounting" root ...
1648 * we just try to remount it readonly.
1650 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1652 return do_umount_root(sb
);
1658 /* Recheck MNT_LOCKED with the locks held */
1660 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1664 if (flags
& MNT_DETACH
) {
1665 if (!list_empty(&mnt
->mnt_list
))
1666 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1669 shrink_submounts(mnt
);
1671 if (!propagate_mount_busy(mnt
, 2)) {
1672 if (!list_empty(&mnt
->mnt_list
))
1673 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1678 unlock_mount_hash();
1684 * __detach_mounts - lazily unmount all mounts on the specified dentry
1686 * During unlink, rmdir, and d_drop it is possible to loose the path
1687 * to an existing mountpoint, and wind up leaking the mount.
1688 * detach_mounts allows lazily unmounting those mounts instead of
1691 * The caller may hold dentry->d_inode->i_mutex.
1693 void __detach_mounts(struct dentry
*dentry
)
1695 struct mountpoint
*mp
;
1700 mp
= lookup_mountpoint(dentry
);
1705 while (!hlist_empty(&mp
->m_list
)) {
1706 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1707 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1709 hlist_add_head(&mnt
->mnt_umount
, &unmounted
);
1711 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1715 unlock_mount_hash();
1720 * Is the caller allowed to modify his namespace?
1722 static inline bool may_mount(void)
1724 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1727 static void warn_mandlock(void)
1729 pr_warn_once("=======================================================\n"
1730 "WARNING: The mand mount option has been deprecated and\n"
1731 " and is ignored by this kernel. Remove the mand\n"
1732 " option from the mount to silence this warning.\n"
1733 "=======================================================\n");
1736 static int can_umount(const struct path
*path
, int flags
)
1738 struct mount
*mnt
= real_mount(path
->mnt
);
1742 if (path
->dentry
!= path
->mnt
->mnt_root
)
1744 if (!check_mnt(mnt
))
1746 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
) /* Check optimistically */
1748 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1753 // caller is responsible for flags being sane
1754 int path_umount(struct path
*path
, int flags
)
1756 struct mount
*mnt
= real_mount(path
->mnt
);
1759 ret
= can_umount(path
, flags
);
1761 ret
= do_umount(mnt
, flags
);
1763 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1765 mntput_no_expire(mnt
);
1769 static int ksys_umount(char __user
*name
, int flags
)
1771 int lookup_flags
= LOOKUP_MOUNTPOINT
;
1775 // basic validity checks done first
1776 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1779 if (!(flags
& UMOUNT_NOFOLLOW
))
1780 lookup_flags
|= LOOKUP_FOLLOW
;
1781 ret
= user_path_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1784 return path_umount(&path
, flags
);
1787 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1789 return ksys_umount(name
, flags
);
1792 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1795 * The 2.0 compatible umount. No flags.
1797 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1799 return ksys_umount(name
, 0);
1804 static bool is_mnt_ns_file(struct dentry
*dentry
)
1806 /* Is this a proxy for a mount namespace? */
1807 return dentry
->d_op
== &ns_dentry_operations
&&
1808 dentry
->d_fsdata
== &mntns_operations
;
1811 static struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1813 return container_of(ns
, struct mnt_namespace
, ns
);
1816 struct ns_common
*from_mnt_ns(struct mnt_namespace
*mnt
)
1821 static bool mnt_ns_loop(struct dentry
*dentry
)
1823 /* Could bind mounting the mount namespace inode cause a
1824 * mount namespace loop?
1826 struct mnt_namespace
*mnt_ns
;
1827 if (!is_mnt_ns_file(dentry
))
1830 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1831 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1834 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1837 struct mount
*res
, *p
, *q
, *r
, *parent
;
1839 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1840 return ERR_PTR(-EINVAL
);
1842 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1843 return ERR_PTR(-EINVAL
);
1845 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1849 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1852 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1854 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1857 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1858 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1859 IS_MNT_UNBINDABLE(s
)) {
1860 if (s
->mnt
.mnt_flags
& MNT_LOCKED
) {
1861 /* Both unbindable and locked. */
1862 q
= ERR_PTR(-EPERM
);
1865 s
= skip_mnt_tree(s
);
1869 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1870 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1871 s
= skip_mnt_tree(s
);
1874 while (p
!= s
->mnt_parent
) {
1880 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1884 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1885 attach_mnt(q
, parent
, p
->mnt_mp
);
1886 unlock_mount_hash();
1893 umount_tree(res
, UMOUNT_SYNC
);
1894 unlock_mount_hash();
1899 /* Caller should check returned pointer for errors */
1901 struct vfsmount
*collect_mounts(const struct path
*path
)
1905 if (!check_mnt(real_mount(path
->mnt
)))
1906 tree
= ERR_PTR(-EINVAL
);
1908 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1909 CL_COPY_ALL
| CL_PRIVATE
);
1912 return ERR_CAST(tree
);
1916 static void free_mnt_ns(struct mnt_namespace
*);
1917 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*, bool);
1919 void dissolve_on_fput(struct vfsmount
*mnt
)
1921 struct mnt_namespace
*ns
;
1924 ns
= real_mount(mnt
)->mnt_ns
;
1927 umount_tree(real_mount(mnt
), UMOUNT_CONNECTED
);
1931 unlock_mount_hash();
1937 void drop_collected_mounts(struct vfsmount
*mnt
)
1941 umount_tree(real_mount(mnt
), 0);
1942 unlock_mount_hash();
1946 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
1948 struct mount
*child
;
1950 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
1951 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
1954 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
1961 * clone_private_mount - create a private clone of a path
1962 * @path: path to clone
1964 * This creates a new vfsmount, which will be the clone of @path. The new mount
1965 * will not be attached anywhere in the namespace and will be private (i.e.
1966 * changes to the originating mount won't be propagated into this).
1968 * Release with mntput().
1970 struct vfsmount
*clone_private_mount(const struct path
*path
)
1972 struct mount
*old_mnt
= real_mount(path
->mnt
);
1973 struct mount
*new_mnt
;
1975 down_read(&namespace_sem
);
1976 if (IS_MNT_UNBINDABLE(old_mnt
))
1979 if (!check_mnt(old_mnt
))
1982 if (has_locked_children(old_mnt
, path
->dentry
))
1985 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1986 up_read(&namespace_sem
);
1988 if (IS_ERR(new_mnt
))
1989 return ERR_CAST(new_mnt
);
1991 /* Longterm mount to be removed by kern_unmount*() */
1992 new_mnt
->mnt_ns
= MNT_NS_INTERNAL
;
1994 return &new_mnt
->mnt
;
1997 up_read(&namespace_sem
);
1998 return ERR_PTR(-EINVAL
);
2000 EXPORT_SYMBOL_GPL(clone_private_mount
);
2002 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
2003 struct vfsmount
*root
)
2006 int res
= f(root
, arg
);
2009 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
2010 res
= f(&mnt
->mnt
, arg
);
2016 EXPORT_SYMBOL_GPL(iterate_mounts
);
2018 static void lock_mnt_tree(struct mount
*mnt
)
2022 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2023 int flags
= p
->mnt
.mnt_flags
;
2024 /* Don't allow unprivileged users to change mount flags */
2025 flags
|= MNT_LOCK_ATIME
;
2027 if (flags
& MNT_READONLY
)
2028 flags
|= MNT_LOCK_READONLY
;
2030 if (flags
& MNT_NODEV
)
2031 flags
|= MNT_LOCK_NODEV
;
2033 if (flags
& MNT_NOSUID
)
2034 flags
|= MNT_LOCK_NOSUID
;
2036 if (flags
& MNT_NOEXEC
)
2037 flags
|= MNT_LOCK_NOEXEC
;
2038 /* Don't allow unprivileged users to reveal what is under a mount */
2039 if (list_empty(&p
->mnt_expire
))
2040 flags
|= MNT_LOCKED
;
2041 p
->mnt
.mnt_flags
= flags
;
2045 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
2049 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
2050 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
2051 mnt_release_group_id(p
);
2055 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
2059 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
2060 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
2061 int err
= mnt_alloc_group_id(p
);
2063 cleanup_group_ids(mnt
, p
);
2072 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
2074 unsigned int max
= READ_ONCE(sysctl_mount_max
);
2075 unsigned int mounts
= 0, old
, pending
, sum
;
2078 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
2082 pending
= ns
->pending_mounts
;
2083 sum
= old
+ pending
;
2087 (mounts
> (max
- sum
)))
2090 ns
->pending_mounts
= pending
+ mounts
;
2095 * @source_mnt : mount tree to be attached
2096 * @nd : place the mount tree @source_mnt is attached
2097 * @parent_nd : if non-null, detach the source_mnt from its parent and
2098 * store the parent mount and mountpoint dentry.
2099 * (done when source_mnt is moved)
2101 * NOTE: in the table below explains the semantics when a source mount
2102 * of a given type is attached to a destination mount of a given type.
2103 * ---------------------------------------------------------------------------
2104 * | BIND MOUNT OPERATION |
2105 * |**************************************************************************
2106 * | source-->| shared | private | slave | unbindable |
2110 * |**************************************************************************
2111 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2113 * |non-shared| shared (+) | private | slave (*) | invalid |
2114 * ***************************************************************************
2115 * A bind operation clones the source mount and mounts the clone on the
2116 * destination mount.
2118 * (++) the cloned mount is propagated to all the mounts in the propagation
2119 * tree of the destination mount and the cloned mount is added to
2120 * the peer group of the source mount.
2121 * (+) the cloned mount is created under the destination mount and is marked
2122 * as shared. The cloned mount is added to the peer group of the source
2124 * (+++) the mount is propagated to all the mounts in the propagation tree
2125 * of the destination mount and the cloned mount is made slave
2126 * of the same master as that of the source mount. The cloned mount
2127 * is marked as 'shared and slave'.
2128 * (*) the cloned mount is made a slave of the same master as that of the
2131 * ---------------------------------------------------------------------------
2132 * | MOVE MOUNT OPERATION |
2133 * |**************************************************************************
2134 * | source-->| shared | private | slave | unbindable |
2138 * |**************************************************************************
2139 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2141 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2142 * ***************************************************************************
2144 * (+) the mount is moved to the destination. And is then propagated to
2145 * all the mounts in the propagation tree of the destination mount.
2146 * (+*) the mount is moved to the destination.
2147 * (+++) the mount is moved to the destination and is then propagated to
2148 * all the mounts belonging to the destination mount's propagation tree.
2149 * the mount is marked as 'shared and slave'.
2150 * (*) the mount continues to be a slave at the new location.
2152 * if the source mount is a tree, the operations explained above is
2153 * applied to each mount in the tree.
2154 * Must be called without spinlocks held, since this function can sleep
2157 static int attach_recursive_mnt(struct mount
*source_mnt
,
2158 struct mount
*dest_mnt
,
2159 struct mountpoint
*dest_mp
,
2162 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2163 HLIST_HEAD(tree_list
);
2164 struct mnt_namespace
*ns
= dest_mnt
->mnt_ns
;
2165 struct mountpoint
*smp
;
2166 struct mount
*child
, *p
;
2167 struct hlist_node
*n
;
2170 /* Preallocate a mountpoint in case the new mounts need
2171 * to be tucked under other mounts.
2173 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
2175 return PTR_ERR(smp
);
2177 /* Is there space to add these mounts to the mount namespace? */
2179 err
= count_mounts(ns
, source_mnt
);
2184 if (IS_MNT_SHARED(dest_mnt
)) {
2185 err
= invent_group_ids(source_mnt
, true);
2188 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2191 goto out_cleanup_ids
;
2192 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2198 unhash_mnt(source_mnt
);
2199 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
2200 touch_mnt_namespace(source_mnt
->mnt_ns
);
2202 if (source_mnt
->mnt_ns
) {
2203 /* move from anon - the caller will destroy */
2204 list_del_init(&source_mnt
->mnt_ns
->list
);
2206 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2207 commit_tree(source_mnt
);
2210 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2212 hlist_del_init(&child
->mnt_hash
);
2213 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2214 child
->mnt_mountpoint
);
2216 mnt_change_mountpoint(child
, smp
, q
);
2217 /* Notice when we are propagating across user namespaces */
2218 if (child
->mnt_parent
->mnt_ns
->user_ns
!= user_ns
)
2219 lock_mnt_tree(child
);
2220 child
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2223 put_mountpoint(smp
);
2224 unlock_mount_hash();
2229 while (!hlist_empty(&tree_list
)) {
2230 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2231 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2232 umount_tree(child
, UMOUNT_SYNC
);
2234 unlock_mount_hash();
2235 cleanup_group_ids(source_mnt
, NULL
);
2237 ns
->pending_mounts
= 0;
2239 read_seqlock_excl(&mount_lock
);
2240 put_mountpoint(smp
);
2241 read_sequnlock_excl(&mount_lock
);
2246 static struct mountpoint
*lock_mount(struct path
*path
)
2248 struct vfsmount
*mnt
;
2249 struct dentry
*dentry
= path
->dentry
;
2251 inode_lock(dentry
->d_inode
);
2252 if (unlikely(cant_mount(dentry
))) {
2253 inode_unlock(dentry
->d_inode
);
2254 return ERR_PTR(-ENOENT
);
2257 mnt
= lookup_mnt(path
);
2259 struct mountpoint
*mp
= get_mountpoint(dentry
);
2262 inode_unlock(dentry
->d_inode
);
2268 inode_unlock(path
->dentry
->d_inode
);
2271 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2275 static void unlock_mount(struct mountpoint
*where
)
2277 struct dentry
*dentry
= where
->m_dentry
;
2279 read_seqlock_excl(&mount_lock
);
2280 put_mountpoint(where
);
2281 read_sequnlock_excl(&mount_lock
);
2284 inode_unlock(dentry
->d_inode
);
2287 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2289 if (mnt
->mnt
.mnt_sb
->s_flags
& SB_NOUSER
)
2292 if (d_is_dir(mp
->m_dentry
) !=
2293 d_is_dir(mnt
->mnt
.mnt_root
))
2296 return attach_recursive_mnt(mnt
, p
, mp
, false);
2300 * Sanity check the flags to change_mnt_propagation.
2303 static int flags_to_propagation_type(int ms_flags
)
2305 int type
= ms_flags
& ~(MS_REC
| MS_SILENT
);
2307 /* Fail if any non-propagation flags are set */
2308 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2310 /* Only one propagation flag should be set */
2311 if (!is_power_of_2(type
))
2317 * recursively change the type of the mountpoint.
2319 static int do_change_type(struct path
*path
, int ms_flags
)
2322 struct mount
*mnt
= real_mount(path
->mnt
);
2323 int recurse
= ms_flags
& MS_REC
;
2327 if (path
->dentry
!= path
->mnt
->mnt_root
)
2330 type
= flags_to_propagation_type(ms_flags
);
2335 if (type
== MS_SHARED
) {
2336 err
= invent_group_ids(mnt
, recurse
);
2342 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2343 change_mnt_propagation(m
, type
);
2344 unlock_mount_hash();
2351 static struct mount
*__do_loopback(struct path
*old_path
, int recurse
)
2353 struct mount
*mnt
= ERR_PTR(-EINVAL
), *old
= real_mount(old_path
->mnt
);
2355 if (IS_MNT_UNBINDABLE(old
))
2358 if (!check_mnt(old
) && old_path
->dentry
->d_op
!= &ns_dentry_operations
)
2361 if (!recurse
&& has_locked_children(old
, old_path
->dentry
))
2365 mnt
= copy_tree(old
, old_path
->dentry
, CL_COPY_MNT_NS_FILE
);
2367 mnt
= clone_mnt(old
, old_path
->dentry
, 0);
2370 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2376 * do loopback mount.
2378 static int do_loopback(struct path
*path
, const char *old_name
,
2381 struct path old_path
;
2382 struct mount
*mnt
= NULL
, *parent
;
2383 struct mountpoint
*mp
;
2385 if (!old_name
|| !*old_name
)
2387 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2392 if (mnt_ns_loop(old_path
.dentry
))
2395 mp
= lock_mount(path
);
2401 parent
= real_mount(path
->mnt
);
2402 if (!check_mnt(parent
))
2405 mnt
= __do_loopback(&old_path
, recurse
);
2411 err
= graft_tree(mnt
, parent
, mp
);
2414 umount_tree(mnt
, UMOUNT_SYNC
);
2415 unlock_mount_hash();
2420 path_put(&old_path
);
2424 static struct file
*open_detached_copy(struct path
*path
, bool recursive
)
2426 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2427 struct mnt_namespace
*ns
= alloc_mnt_ns(user_ns
, true);
2428 struct mount
*mnt
, *p
;
2432 return ERR_CAST(ns
);
2435 mnt
= __do_loopback(path
, recursive
);
2439 return ERR_CAST(mnt
);
2443 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2448 list_add_tail(&ns
->list
, &mnt
->mnt_list
);
2450 unlock_mount_hash();
2454 path
->mnt
= &mnt
->mnt
;
2455 file
= dentry_open(path
, O_PATH
, current_cred());
2457 dissolve_on_fput(path
->mnt
);
2459 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
2463 SYSCALL_DEFINE3(open_tree
, int, dfd
, const char __user
*, filename
, unsigned, flags
)
2467 int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
2468 bool detached
= flags
& OPEN_TREE_CLONE
;
2472 BUILD_BUG_ON(OPEN_TREE_CLOEXEC
!= O_CLOEXEC
);
2474 if (flags
& ~(AT_EMPTY_PATH
| AT_NO_AUTOMOUNT
| AT_RECURSIVE
|
2475 AT_SYMLINK_NOFOLLOW
| OPEN_TREE_CLONE
|
2479 if ((flags
& (AT_RECURSIVE
| OPEN_TREE_CLONE
)) == AT_RECURSIVE
)
2482 if (flags
& AT_NO_AUTOMOUNT
)
2483 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
2484 if (flags
& AT_SYMLINK_NOFOLLOW
)
2485 lookup_flags
&= ~LOOKUP_FOLLOW
;
2486 if (flags
& AT_EMPTY_PATH
)
2487 lookup_flags
|= LOOKUP_EMPTY
;
2489 if (detached
&& !may_mount())
2492 fd
= get_unused_fd_flags(flags
& O_CLOEXEC
);
2496 error
= user_path_at(dfd
, filename
, lookup_flags
, &path
);
2497 if (unlikely(error
)) {
2498 file
= ERR_PTR(error
);
2501 file
= open_detached_copy(&path
, flags
& AT_RECURSIVE
);
2503 file
= dentry_open(&path
, O_PATH
, current_cred());
2508 return PTR_ERR(file
);
2510 fd_install(fd
, file
);
2515 * Don't allow locked mount flags to be cleared.
2517 * No locks need to be held here while testing the various MNT_LOCK
2518 * flags because those flags can never be cleared once they are set.
2520 static bool can_change_locked_flags(struct mount
*mnt
, unsigned int mnt_flags
)
2522 unsigned int fl
= mnt
->mnt
.mnt_flags
;
2524 if ((fl
& MNT_LOCK_READONLY
) &&
2525 !(mnt_flags
& MNT_READONLY
))
2528 if ((fl
& MNT_LOCK_NODEV
) &&
2529 !(mnt_flags
& MNT_NODEV
))
2532 if ((fl
& MNT_LOCK_NOSUID
) &&
2533 !(mnt_flags
& MNT_NOSUID
))
2536 if ((fl
& MNT_LOCK_NOEXEC
) &&
2537 !(mnt_flags
& MNT_NOEXEC
))
2540 if ((fl
& MNT_LOCK_ATIME
) &&
2541 ((fl
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
)))
2547 static int change_mount_ro_state(struct mount
*mnt
, unsigned int mnt_flags
)
2549 bool readonly_request
= (mnt_flags
& MNT_READONLY
);
2551 if (readonly_request
== __mnt_is_readonly(&mnt
->mnt
))
2554 if (readonly_request
)
2555 return mnt_make_readonly(mnt
);
2557 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
2561 static void set_mount_attributes(struct mount
*mnt
, unsigned int mnt_flags
)
2563 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2564 mnt
->mnt
.mnt_flags
= mnt_flags
;
2565 touch_mnt_namespace(mnt
->mnt_ns
);
2568 static void mnt_warn_timestamp_expiry(struct path
*mountpoint
, struct vfsmount
*mnt
)
2570 struct super_block
*sb
= mnt
->mnt_sb
;
2572 if (!__mnt_is_readonly(mnt
) &&
2573 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX
> sb
->s_time_max
)) {
2574 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
2575 char *mntpath
= buf
? d_path(mountpoint
, buf
, PAGE_SIZE
) : ERR_PTR(-ENOMEM
);
2578 time64_to_tm(sb
->s_time_max
, 0, &tm
);
2580 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2582 is_mounted(mnt
) ? "remounted" : "mounted",
2584 tm
.tm_year
+1900, (unsigned long long)sb
->s_time_max
);
2586 free_page((unsigned long)buf
);
2591 * Handle reconfiguration of the mountpoint only without alteration of the
2592 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2595 static int do_reconfigure_mnt(struct path
*path
, unsigned int mnt_flags
)
2597 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2598 struct mount
*mnt
= real_mount(path
->mnt
);
2601 if (!check_mnt(mnt
))
2604 if (path
->dentry
!= mnt
->mnt
.mnt_root
)
2607 if (!can_change_locked_flags(mnt
, mnt_flags
))
2611 * We're only checking whether the superblock is read-only not
2612 * changing it, so only take down_read(&sb->s_umount).
2614 down_read(&sb
->s_umount
);
2616 ret
= change_mount_ro_state(mnt
, mnt_flags
);
2618 set_mount_attributes(mnt
, mnt_flags
);
2619 unlock_mount_hash();
2620 up_read(&sb
->s_umount
);
2622 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2628 * change filesystem flags. dir should be a physical root of filesystem.
2629 * If you've mounted a non-root directory somewhere and want to do remount
2630 * on it - tough luck.
2632 static int do_remount(struct path
*path
, int ms_flags
, int sb_flags
,
2633 int mnt_flags
, void *data
)
2636 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2637 struct mount
*mnt
= real_mount(path
->mnt
);
2638 struct fs_context
*fc
;
2640 if (!check_mnt(mnt
))
2643 if (path
->dentry
!= path
->mnt
->mnt_root
)
2646 if (!can_change_locked_flags(mnt
, mnt_flags
))
2649 fc
= fs_context_for_reconfigure(path
->dentry
, sb_flags
, MS_RMT_MASK
);
2654 err
= parse_monolithic_mount_data(fc
, data
);
2656 down_write(&sb
->s_umount
);
2658 if (ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
)) {
2659 err
= reconfigure_super(fc
);
2662 set_mount_attributes(mnt
, mnt_flags
);
2663 unlock_mount_hash();
2666 up_write(&sb
->s_umount
);
2669 mnt_warn_timestamp_expiry(path
, &mnt
->mnt
);
2675 static inline int tree_contains_unbindable(struct mount
*mnt
)
2678 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2679 if (IS_MNT_UNBINDABLE(p
))
2686 * Check that there aren't references to earlier/same mount namespaces in the
2687 * specified subtree. Such references can act as pins for mount namespaces
2688 * that aren't checked by the mount-cycle checking code, thereby allowing
2689 * cycles to be made.
2691 static bool check_for_nsfs_mounts(struct mount
*subtree
)
2697 for (p
= subtree
; p
; p
= next_mnt(p
, subtree
))
2698 if (mnt_ns_loop(p
->mnt
.mnt_root
))
2703 unlock_mount_hash();
2707 static int do_set_group(struct path
*from_path
, struct path
*to_path
)
2709 struct mount
*from
, *to
;
2712 from
= real_mount(from_path
->mnt
);
2713 to
= real_mount(to_path
->mnt
);
2718 /* To and From must be mounted */
2719 if (!is_mounted(&from
->mnt
))
2721 if (!is_mounted(&to
->mnt
))
2725 /* We should be allowed to modify mount namespaces of both mounts */
2726 if (!ns_capable(from
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
))
2728 if (!ns_capable(to
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
))
2732 /* To and From paths should be mount roots */
2733 if (from_path
->dentry
!= from_path
->mnt
->mnt_root
)
2735 if (to_path
->dentry
!= to_path
->mnt
->mnt_root
)
2738 /* Setting sharing groups is only allowed across same superblock */
2739 if (from
->mnt
.mnt_sb
!= to
->mnt
.mnt_sb
)
2742 /* From mount root should be wider than To mount root */
2743 if (!is_subdir(to
->mnt
.mnt_root
, from
->mnt
.mnt_root
))
2746 /* From mount should not have locked children in place of To's root */
2747 if (has_locked_children(from
, to
->mnt
.mnt_root
))
2750 /* Setting sharing groups is only allowed on private mounts */
2751 if (IS_MNT_SHARED(to
) || IS_MNT_SLAVE(to
))
2754 /* From should not be private */
2755 if (!IS_MNT_SHARED(from
) && !IS_MNT_SLAVE(from
))
2758 if (IS_MNT_SLAVE(from
)) {
2759 struct mount
*m
= from
->mnt_master
;
2761 list_add(&to
->mnt_slave
, &m
->mnt_slave_list
);
2765 if (IS_MNT_SHARED(from
)) {
2766 to
->mnt_group_id
= from
->mnt_group_id
;
2767 list_add(&to
->mnt_share
, &from
->mnt_share
);
2770 unlock_mount_hash();
2779 static int do_move_mount(struct path
*old_path
, struct path
*new_path
)
2781 struct mnt_namespace
*ns
;
2784 struct mount
*parent
;
2785 struct mountpoint
*mp
, *old_mp
;
2789 mp
= lock_mount(new_path
);
2793 old
= real_mount(old_path
->mnt
);
2794 p
= real_mount(new_path
->mnt
);
2795 parent
= old
->mnt_parent
;
2796 attached
= mnt_has_parent(old
);
2797 old_mp
= old
->mnt_mp
;
2801 /* The mountpoint must be in our namespace. */
2805 /* The thing moved must be mounted... */
2806 if (!is_mounted(&old
->mnt
))
2809 /* ... and either ours or the root of anon namespace */
2810 if (!(attached
? check_mnt(old
) : is_anon_ns(ns
)))
2813 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2816 if (old_path
->dentry
!= old_path
->mnt
->mnt_root
)
2819 if (d_is_dir(new_path
->dentry
) !=
2820 d_is_dir(old_path
->dentry
))
2823 * Don't move a mount residing in a shared parent.
2825 if (attached
&& IS_MNT_SHARED(parent
))
2828 * Don't move a mount tree containing unbindable mounts to a destination
2829 * mount which is shared.
2831 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2834 if (!check_for_nsfs_mounts(old
))
2836 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2840 err
= attach_recursive_mnt(old
, real_mount(new_path
->mnt
), mp
,
2845 /* if the mount is moved, it should no longer be expire
2847 list_del_init(&old
->mnt_expire
);
2849 put_mountpoint(old_mp
);
2854 mntput_no_expire(parent
);
2861 static int do_move_mount_old(struct path
*path
, const char *old_name
)
2863 struct path old_path
;
2866 if (!old_name
|| !*old_name
)
2869 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2873 err
= do_move_mount(&old_path
, path
);
2874 path_put(&old_path
);
2879 * add a mount into a namespace's mount tree
2881 static int do_add_mount(struct mount
*newmnt
, struct mountpoint
*mp
,
2882 struct path
*path
, int mnt_flags
)
2884 struct mount
*parent
= real_mount(path
->mnt
);
2886 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2888 if (unlikely(!check_mnt(parent
))) {
2889 /* that's acceptable only for automounts done in private ns */
2890 if (!(mnt_flags
& MNT_SHRINKABLE
))
2892 /* ... and for those we'd better have mountpoint still alive */
2893 if (!parent
->mnt_ns
)
2897 /* Refuse the same filesystem on the same mount point */
2898 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2899 path
->mnt
->mnt_root
== path
->dentry
)
2902 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2905 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2906 return graft_tree(newmnt
, parent
, mp
);
2909 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
);
2912 * Create a new mount using a superblock configuration and request it
2913 * be added to the namespace tree.
2915 static int do_new_mount_fc(struct fs_context
*fc
, struct path
*mountpoint
,
2916 unsigned int mnt_flags
)
2918 struct vfsmount
*mnt
;
2919 struct mountpoint
*mp
;
2920 struct super_block
*sb
= fc
->root
->d_sb
;
2923 error
= security_sb_kern_mount(sb
);
2924 if (!error
&& mount_too_revealing(sb
, &mnt_flags
))
2927 if (unlikely(error
)) {
2932 up_write(&sb
->s_umount
);
2934 mnt
= vfs_create_mount(fc
);
2936 return PTR_ERR(mnt
);
2938 mnt_warn_timestamp_expiry(mountpoint
, mnt
);
2940 mp
= lock_mount(mountpoint
);
2945 error
= do_add_mount(real_mount(mnt
), mp
, mountpoint
, mnt_flags
);
2953 * create a new mount for userspace and request it to be added into the
2956 static int do_new_mount(struct path
*path
, const char *fstype
, int sb_flags
,
2957 int mnt_flags
, const char *name
, void *data
)
2959 struct file_system_type
*type
;
2960 struct fs_context
*fc
;
2961 const char *subtype
= NULL
;
2967 type
= get_fs_type(fstype
);
2971 if (type
->fs_flags
& FS_HAS_SUBTYPE
) {
2972 subtype
= strchr(fstype
, '.');
2976 put_filesystem(type
);
2982 fc
= fs_context_for_mount(type
, sb_flags
);
2983 put_filesystem(type
);
2988 err
= vfs_parse_fs_string(fc
, "subtype",
2989 subtype
, strlen(subtype
));
2991 err
= vfs_parse_fs_string(fc
, "source", name
, strlen(name
));
2993 err
= parse_monolithic_mount_data(fc
, data
);
2994 if (!err
&& !mount_capable(fc
))
2997 err
= vfs_get_tree(fc
);
2999 err
= do_new_mount_fc(fc
, path
, mnt_flags
);
3005 int finish_automount(struct vfsmount
*m
, struct path
*path
)
3007 struct dentry
*dentry
= path
->dentry
;
3008 struct mountpoint
*mp
;
3017 mnt
= real_mount(m
);
3018 /* The new mount record should have at least 2 refs to prevent it being
3019 * expired before we get a chance to add it
3021 BUG_ON(mnt_get_count(mnt
) < 2);
3023 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
3024 m
->mnt_root
== dentry
) {
3030 * we don't want to use lock_mount() - in this case finding something
3031 * that overmounts our mountpoint to be means "quitely drop what we've
3032 * got", not "try to mount it on top".
3034 inode_lock(dentry
->d_inode
);
3036 if (unlikely(cant_mount(dentry
))) {
3038 goto discard_locked
;
3041 if (unlikely(__lookup_mnt(path
->mnt
, dentry
))) {
3044 goto discard_locked
;
3047 mp
= get_mountpoint(dentry
);
3050 goto discard_locked
;
3053 err
= do_add_mount(mnt
, mp
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
3062 inode_unlock(dentry
->d_inode
);
3064 /* remove m from any expiration list it may be on */
3065 if (!list_empty(&mnt
->mnt_expire
)) {
3067 list_del_init(&mnt
->mnt_expire
);
3076 * mnt_set_expiry - Put a mount on an expiration list
3077 * @mnt: The mount to list.
3078 * @expiry_list: The list to add the mount to.
3080 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
3084 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
3088 EXPORT_SYMBOL(mnt_set_expiry
);
3091 * process a list of expirable mountpoints with the intent of discarding any
3092 * mountpoints that aren't in use and haven't been touched since last we came
3095 void mark_mounts_for_expiry(struct list_head
*mounts
)
3097 struct mount
*mnt
, *next
;
3098 LIST_HEAD(graveyard
);
3100 if (list_empty(mounts
))
3106 /* extract from the expiration list every vfsmount that matches the
3107 * following criteria:
3108 * - only referenced by its parent vfsmount
3109 * - still marked for expiry (marked on the last call here; marks are
3110 * cleared by mntput())
3112 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
3113 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
3114 propagate_mount_busy(mnt
, 1))
3116 list_move(&mnt
->mnt_expire
, &graveyard
);
3118 while (!list_empty(&graveyard
)) {
3119 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
3120 touch_mnt_namespace(mnt
->mnt_ns
);
3121 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3123 unlock_mount_hash();
3127 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
3130 * Ripoff of 'select_parent()'
3132 * search the list of submounts for a given mountpoint, and move any
3133 * shrinkable submounts to the 'graveyard' list.
3135 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
3137 struct mount
*this_parent
= parent
;
3138 struct list_head
*next
;
3142 next
= this_parent
->mnt_mounts
.next
;
3144 while (next
!= &this_parent
->mnt_mounts
) {
3145 struct list_head
*tmp
= next
;
3146 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
3149 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
3152 * Descend a level if the d_mounts list is non-empty.
3154 if (!list_empty(&mnt
->mnt_mounts
)) {
3159 if (!propagate_mount_busy(mnt
, 1)) {
3160 list_move_tail(&mnt
->mnt_expire
, graveyard
);
3165 * All done at this level ... ascend and resume the search
3167 if (this_parent
!= parent
) {
3168 next
= this_parent
->mnt_child
.next
;
3169 this_parent
= this_parent
->mnt_parent
;
3176 * process a list of expirable mountpoints with the intent of discarding any
3177 * submounts of a specific parent mountpoint
3179 * mount_lock must be held for write
3181 static void shrink_submounts(struct mount
*mnt
)
3183 LIST_HEAD(graveyard
);
3186 /* extract submounts of 'mountpoint' from the expiration list */
3187 while (select_submounts(mnt
, &graveyard
)) {
3188 while (!list_empty(&graveyard
)) {
3189 m
= list_first_entry(&graveyard
, struct mount
,
3191 touch_mnt_namespace(m
->mnt_ns
);
3192 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
3197 static void *copy_mount_options(const void __user
* data
)
3200 unsigned left
, offset
;
3205 copy
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
3207 return ERR_PTR(-ENOMEM
);
3209 left
= copy_from_user(copy
, data
, PAGE_SIZE
);
3212 * Not all architectures have an exact copy_from_user(). Resort to
3215 offset
= PAGE_SIZE
- left
;
3218 if (get_user(c
, (const char __user
*)data
+ offset
))
3225 if (left
== PAGE_SIZE
) {
3227 return ERR_PTR(-EFAULT
);
3233 static char *copy_mount_string(const void __user
*data
)
3235 return data
? strndup_user(data
, PATH_MAX
) : NULL
;
3239 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3240 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3242 * data is a (void *) that can point to any structure up to
3243 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3244 * information (or be NULL).
3246 * Pre-0.97 versions of mount() didn't have a flags word.
3247 * When the flags word was introduced its top half was required
3248 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3249 * Therefore, if this magic number is present, it carries no information
3250 * and must be discarded.
3252 int path_mount(const char *dev_name
, struct path
*path
,
3253 const char *type_page
, unsigned long flags
, void *data_page
)
3255 unsigned int mnt_flags
= 0, sb_flags
;
3259 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
3260 flags
&= ~MS_MGC_MSK
;
3262 /* Basic sanity checks */
3264 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
3266 if (flags
& MS_NOUSER
)
3269 ret
= security_sb_mount(dev_name
, path
, type_page
, flags
, data_page
);
3274 if (flags
& SB_MANDLOCK
)
3277 /* Default to relatime unless overriden */
3278 if (!(flags
& MS_NOATIME
))
3279 mnt_flags
|= MNT_RELATIME
;
3281 /* Separate the per-mountpoint flags */
3282 if (flags
& MS_NOSUID
)
3283 mnt_flags
|= MNT_NOSUID
;
3284 if (flags
& MS_NODEV
)
3285 mnt_flags
|= MNT_NODEV
;
3286 if (flags
& MS_NOEXEC
)
3287 mnt_flags
|= MNT_NOEXEC
;
3288 if (flags
& MS_NOATIME
)
3289 mnt_flags
|= MNT_NOATIME
;
3290 if (flags
& MS_NODIRATIME
)
3291 mnt_flags
|= MNT_NODIRATIME
;
3292 if (flags
& MS_STRICTATIME
)
3293 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
3294 if (flags
& MS_RDONLY
)
3295 mnt_flags
|= MNT_READONLY
;
3296 if (flags
& MS_NOSYMFOLLOW
)
3297 mnt_flags
|= MNT_NOSYMFOLLOW
;
3299 /* The default atime for remount is preservation */
3300 if ((flags
& MS_REMOUNT
) &&
3301 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
3302 MS_STRICTATIME
)) == 0)) {
3303 mnt_flags
&= ~MNT_ATIME_MASK
;
3304 mnt_flags
|= path
->mnt
->mnt_flags
& MNT_ATIME_MASK
;
3307 sb_flags
= flags
& (SB_RDONLY
|
3316 if ((flags
& (MS_REMOUNT
| MS_BIND
)) == (MS_REMOUNT
| MS_BIND
))
3317 return do_reconfigure_mnt(path
, mnt_flags
);
3318 if (flags
& MS_REMOUNT
)
3319 return do_remount(path
, flags
, sb_flags
, mnt_flags
, data_page
);
3320 if (flags
& MS_BIND
)
3321 return do_loopback(path
, dev_name
, flags
& MS_REC
);
3322 if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
3323 return do_change_type(path
, flags
);
3324 if (flags
& MS_MOVE
)
3325 return do_move_mount_old(path
, dev_name
);
3327 return do_new_mount(path
, type_page
, sb_flags
, mnt_flags
, dev_name
,
3331 long do_mount(const char *dev_name
, const char __user
*dir_name
,
3332 const char *type_page
, unsigned long flags
, void *data_page
)
3337 ret
= user_path_at(AT_FDCWD
, dir_name
, LOOKUP_FOLLOW
, &path
);
3340 ret
= path_mount(dev_name
, &path
, type_page
, flags
, data_page
);
3345 static struct ucounts
*inc_mnt_namespaces(struct user_namespace
*ns
)
3347 return inc_ucount(ns
, current_euid(), UCOUNT_MNT_NAMESPACES
);
3350 static void dec_mnt_namespaces(struct ucounts
*ucounts
)
3352 dec_ucount(ucounts
, UCOUNT_MNT_NAMESPACES
);
3355 static void free_mnt_ns(struct mnt_namespace
*ns
)
3357 if (!is_anon_ns(ns
))
3358 ns_free_inum(&ns
->ns
);
3359 dec_mnt_namespaces(ns
->ucounts
);
3360 put_user_ns(ns
->user_ns
);
3365 * Assign a sequence number so we can detect when we attempt to bind
3366 * mount a reference to an older mount namespace into the current
3367 * mount namespace, preventing reference counting loops. A 64bit
3368 * number incrementing at 10Ghz will take 12,427 years to wrap which
3369 * is effectively never, so we can ignore the possibility.
3371 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
3373 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
, bool anon
)
3375 struct mnt_namespace
*new_ns
;
3376 struct ucounts
*ucounts
;
3379 ucounts
= inc_mnt_namespaces(user_ns
);
3381 return ERR_PTR(-ENOSPC
);
3383 new_ns
= kzalloc(sizeof(struct mnt_namespace
), GFP_KERNEL_ACCOUNT
);
3385 dec_mnt_namespaces(ucounts
);
3386 return ERR_PTR(-ENOMEM
);
3389 ret
= ns_alloc_inum(&new_ns
->ns
);
3392 dec_mnt_namespaces(ucounts
);
3393 return ERR_PTR(ret
);
3396 new_ns
->ns
.ops
= &mntns_operations
;
3398 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
3399 refcount_set(&new_ns
->ns
.count
, 1);
3400 INIT_LIST_HEAD(&new_ns
->list
);
3401 init_waitqueue_head(&new_ns
->poll
);
3402 spin_lock_init(&new_ns
->ns_lock
);
3403 new_ns
->user_ns
= get_user_ns(user_ns
);
3404 new_ns
->ucounts
= ucounts
;
3409 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
3410 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
3412 struct mnt_namespace
*new_ns
;
3413 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
3414 struct mount
*p
, *q
;
3421 if (likely(!(flags
& CLONE_NEWNS
))) {
3428 new_ns
= alloc_mnt_ns(user_ns
, false);
3433 /* First pass: copy the tree topology */
3434 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
3435 if (user_ns
!= ns
->user_ns
)
3436 copy_flags
|= CL_SHARED_TO_SLAVE
;
3437 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
3440 free_mnt_ns(new_ns
);
3441 return ERR_CAST(new);
3443 if (user_ns
!= ns
->user_ns
) {
3446 unlock_mount_hash();
3449 list_add_tail(&new_ns
->list
, &new->mnt_list
);
3452 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3453 * as belonging to new namespace. We have already acquired a private
3454 * fs_struct, so tsk->fs->lock is not needed.
3462 if (&p
->mnt
== new_fs
->root
.mnt
) {
3463 new_fs
->root
.mnt
= mntget(&q
->mnt
);
3466 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
3467 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
3471 p
= next_mnt(p
, old
);
3472 q
= next_mnt(q
, new);
3475 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
3476 p
= next_mnt(p
, old
);
3488 struct dentry
*mount_subtree(struct vfsmount
*m
, const char *name
)
3490 struct mount
*mnt
= real_mount(m
);
3491 struct mnt_namespace
*ns
;
3492 struct super_block
*s
;
3496 ns
= alloc_mnt_ns(&init_user_ns
, true);
3499 return ERR_CAST(ns
);
3504 list_add(&mnt
->mnt_list
, &ns
->list
);
3506 err
= vfs_path_lookup(m
->mnt_root
, m
,
3507 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
3512 return ERR_PTR(err
);
3514 /* trade a vfsmount reference for active sb one */
3515 s
= path
.mnt
->mnt_sb
;
3516 atomic_inc(&s
->s_active
);
3518 /* lock the sucker */
3519 down_write(&s
->s_umount
);
3520 /* ... and return the root of (sub)tree on it */
3523 EXPORT_SYMBOL(mount_subtree
);
3525 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
3526 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3533 kernel_type
= copy_mount_string(type
);
3534 ret
= PTR_ERR(kernel_type
);
3535 if (IS_ERR(kernel_type
))
3538 kernel_dev
= copy_mount_string(dev_name
);
3539 ret
= PTR_ERR(kernel_dev
);
3540 if (IS_ERR(kernel_dev
))
3543 options
= copy_mount_options(data
);
3544 ret
= PTR_ERR(options
);
3545 if (IS_ERR(options
))
3548 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
, options
);
3559 #define FSMOUNT_VALID_FLAGS \
3560 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3561 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3562 MOUNT_ATTR_NOSYMFOLLOW)
3564 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3566 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3567 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3569 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags
)
3571 unsigned int mnt_flags
= 0;
3573 if (attr_flags
& MOUNT_ATTR_RDONLY
)
3574 mnt_flags
|= MNT_READONLY
;
3575 if (attr_flags
& MOUNT_ATTR_NOSUID
)
3576 mnt_flags
|= MNT_NOSUID
;
3577 if (attr_flags
& MOUNT_ATTR_NODEV
)
3578 mnt_flags
|= MNT_NODEV
;
3579 if (attr_flags
& MOUNT_ATTR_NOEXEC
)
3580 mnt_flags
|= MNT_NOEXEC
;
3581 if (attr_flags
& MOUNT_ATTR_NODIRATIME
)
3582 mnt_flags
|= MNT_NODIRATIME
;
3583 if (attr_flags
& MOUNT_ATTR_NOSYMFOLLOW
)
3584 mnt_flags
|= MNT_NOSYMFOLLOW
;
3590 * Create a kernel mount representation for a new, prepared superblock
3591 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3593 SYSCALL_DEFINE3(fsmount
, int, fs_fd
, unsigned int, flags
,
3594 unsigned int, attr_flags
)
3596 struct mnt_namespace
*ns
;
3597 struct fs_context
*fc
;
3599 struct path newmount
;
3602 unsigned int mnt_flags
= 0;
3608 if ((flags
& ~(FSMOUNT_CLOEXEC
)) != 0)
3611 if (attr_flags
& ~FSMOUNT_VALID_FLAGS
)
3614 mnt_flags
= attr_flags_to_mnt_flags(attr_flags
);
3616 switch (attr_flags
& MOUNT_ATTR__ATIME
) {
3617 case MOUNT_ATTR_STRICTATIME
:
3619 case MOUNT_ATTR_NOATIME
:
3620 mnt_flags
|= MNT_NOATIME
;
3622 case MOUNT_ATTR_RELATIME
:
3623 mnt_flags
|= MNT_RELATIME
;
3634 if (f
.file
->f_op
!= &fscontext_fops
)
3637 fc
= f
.file
->private_data
;
3639 ret
= mutex_lock_interruptible(&fc
->uapi_mutex
);
3643 /* There must be a valid superblock or we can't mount it */
3649 if (mount_too_revealing(fc
->root
->d_sb
, &mnt_flags
)) {
3650 pr_warn("VFS: Mount too revealing\n");
3655 if (fc
->phase
!= FS_CONTEXT_AWAITING_MOUNT
)
3658 if (fc
->sb_flags
& SB_MANDLOCK
)
3661 newmount
.mnt
= vfs_create_mount(fc
);
3662 if (IS_ERR(newmount
.mnt
)) {
3663 ret
= PTR_ERR(newmount
.mnt
);
3666 newmount
.dentry
= dget(fc
->root
);
3667 newmount
.mnt
->mnt_flags
= mnt_flags
;
3669 /* We've done the mount bit - now move the file context into more or
3670 * less the same state as if we'd done an fspick(). We don't want to
3671 * do any memory allocation or anything like that at this point as we
3672 * don't want to have to handle any errors incurred.
3674 vfs_clean_context(fc
);
3676 ns
= alloc_mnt_ns(current
->nsproxy
->mnt_ns
->user_ns
, true);
3681 mnt
= real_mount(newmount
.mnt
);
3685 list_add(&mnt
->mnt_list
, &ns
->list
);
3686 mntget(newmount
.mnt
);
3688 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3689 * it, not just simply put it.
3691 file
= dentry_open(&newmount
, O_PATH
, fc
->cred
);
3693 dissolve_on_fput(newmount
.mnt
);
3694 ret
= PTR_ERR(file
);
3697 file
->f_mode
|= FMODE_NEED_UNMOUNT
;
3699 ret
= get_unused_fd_flags((flags
& FSMOUNT_CLOEXEC
) ? O_CLOEXEC
: 0);
3701 fd_install(ret
, file
);
3706 path_put(&newmount
);
3708 mutex_unlock(&fc
->uapi_mutex
);
3715 * Move a mount from one place to another. In combination with
3716 * fsopen()/fsmount() this is used to install a new mount and in combination
3717 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3720 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3722 SYSCALL_DEFINE5(move_mount
,
3723 int, from_dfd
, const char __user
*, from_pathname
,
3724 int, to_dfd
, const char __user
*, to_pathname
,
3725 unsigned int, flags
)
3727 struct path from_path
, to_path
;
3728 unsigned int lflags
;
3734 if (flags
& ~MOVE_MOUNT__MASK
)
3737 /* If someone gives a pathname, they aren't permitted to move
3738 * from an fd that requires unmount as we can't get at the flag
3739 * to clear it afterwards.
3742 if (flags
& MOVE_MOUNT_F_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3743 if (flags
& MOVE_MOUNT_F_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3744 if (flags
& MOVE_MOUNT_F_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3746 ret
= user_path_at(from_dfd
, from_pathname
, lflags
, &from_path
);
3751 if (flags
& MOVE_MOUNT_T_SYMLINKS
) lflags
|= LOOKUP_FOLLOW
;
3752 if (flags
& MOVE_MOUNT_T_AUTOMOUNTS
) lflags
|= LOOKUP_AUTOMOUNT
;
3753 if (flags
& MOVE_MOUNT_T_EMPTY_PATH
) lflags
|= LOOKUP_EMPTY
;
3755 ret
= user_path_at(to_dfd
, to_pathname
, lflags
, &to_path
);
3759 ret
= security_move_mount(&from_path
, &to_path
);
3763 if (flags
& MOVE_MOUNT_SET_GROUP
)
3764 ret
= do_set_group(&from_path
, &to_path
);
3766 ret
= do_move_mount(&from_path
, &to_path
);
3771 path_put(&from_path
);
3776 * Return true if path is reachable from root
3778 * namespace_sem or mount_lock is held
3780 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
3781 const struct path
*root
)
3783 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
3784 dentry
= mnt
->mnt_mountpoint
;
3785 mnt
= mnt
->mnt_parent
;
3787 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
3790 bool path_is_under(const struct path
*path1
, const struct path
*path2
)
3793 read_seqlock_excl(&mount_lock
);
3794 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
3795 read_sequnlock_excl(&mount_lock
);
3798 EXPORT_SYMBOL(path_is_under
);
3801 * pivot_root Semantics:
3802 * Moves the root file system of the current process to the directory put_old,
3803 * makes new_root as the new root file system of the current process, and sets
3804 * root/cwd of all processes which had them on the current root to new_root.
3807 * The new_root and put_old must be directories, and must not be on the
3808 * same file system as the current process root. The put_old must be
3809 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3810 * pointed to by put_old must yield the same directory as new_root. No other
3811 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3813 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3814 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3815 * in this situation.
3818 * - we don't move root/cwd if they are not at the root (reason: if something
3819 * cared enough to change them, it's probably wrong to force them elsewhere)
3820 * - it's okay to pick a root that isn't the root of a file system, e.g.
3821 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3822 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3825 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3826 const char __user
*, put_old
)
3828 struct path
new, old
, root
;
3829 struct mount
*new_mnt
, *root_mnt
, *old_mnt
, *root_parent
, *ex_parent
;
3830 struct mountpoint
*old_mp
, *root_mp
;
3836 error
= user_path_at(AT_FDCWD
, new_root
,
3837 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &new);
3841 error
= user_path_at(AT_FDCWD
, put_old
,
3842 LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old
);
3846 error
= security_sb_pivotroot(&old
, &new);
3850 get_fs_root(current
->fs
, &root
);
3851 old_mp
= lock_mount(&old
);
3852 error
= PTR_ERR(old_mp
);
3857 new_mnt
= real_mount(new.mnt
);
3858 root_mnt
= real_mount(root
.mnt
);
3859 old_mnt
= real_mount(old
.mnt
);
3860 ex_parent
= new_mnt
->mnt_parent
;
3861 root_parent
= root_mnt
->mnt_parent
;
3862 if (IS_MNT_SHARED(old_mnt
) ||
3863 IS_MNT_SHARED(ex_parent
) ||
3864 IS_MNT_SHARED(root_parent
))
3866 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3868 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3871 if (d_unlinked(new.dentry
))
3874 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3875 goto out4
; /* loop, on the same file system */
3877 if (root
.mnt
->mnt_root
!= root
.dentry
)
3878 goto out4
; /* not a mountpoint */
3879 if (!mnt_has_parent(root_mnt
))
3880 goto out4
; /* not attached */
3881 if (new.mnt
->mnt_root
!= new.dentry
)
3882 goto out4
; /* not a mountpoint */
3883 if (!mnt_has_parent(new_mnt
))
3884 goto out4
; /* not attached */
3885 /* make sure we can reach put_old from new_root */
3886 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3888 /* make certain new is below the root */
3889 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3892 umount_mnt(new_mnt
);
3893 root_mp
= unhash_mnt(root_mnt
); /* we'll need its mountpoint */
3894 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3895 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3896 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3898 /* mount old root on put_old */
3899 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3900 /* mount new_root on / */
3901 attach_mnt(new_mnt
, root_parent
, root_mp
);
3902 mnt_add_count(root_parent
, -1);
3903 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3904 /* A moved mount should not expire automatically */
3905 list_del_init(&new_mnt
->mnt_expire
);
3906 put_mountpoint(root_mp
);
3907 unlock_mount_hash();
3908 chroot_fs_refs(&root
, &new);
3911 unlock_mount(old_mp
);
3913 mntput_no_expire(ex_parent
);
3924 static unsigned int recalc_flags(struct mount_kattr
*kattr
, struct mount
*mnt
)
3926 unsigned int flags
= mnt
->mnt
.mnt_flags
;
3928 /* flags to clear */
3929 flags
&= ~kattr
->attr_clr
;
3930 /* flags to raise */
3931 flags
|= kattr
->attr_set
;
3936 static int can_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
3938 struct vfsmount
*m
= &mnt
->mnt
;
3940 if (!kattr
->mnt_userns
)
3944 * Once a mount has been idmapped we don't allow it to change its
3945 * mapping. It makes things simpler and callers can just create
3946 * another bind-mount they can idmap if they want to.
3948 if (mnt_user_ns(m
) != &init_user_ns
)
3951 /* The underlying filesystem doesn't support idmapped mounts yet. */
3952 if (!(m
->mnt_sb
->s_type
->fs_flags
& FS_ALLOW_IDMAP
))
3955 /* Don't yet support filesystem mountable in user namespaces. */
3956 if (m
->mnt_sb
->s_user_ns
!= &init_user_ns
)
3959 /* We're not controlling the superblock. */
3960 if (!capable(CAP_SYS_ADMIN
))
3963 /* Mount has already been visible in the filesystem hierarchy. */
3964 if (!is_anon_ns(mnt
->mnt_ns
))
3970 static struct mount
*mount_setattr_prepare(struct mount_kattr
*kattr
,
3971 struct mount
*mnt
, int *err
)
3973 struct mount
*m
= mnt
, *last
= NULL
;
3975 if (!is_mounted(&m
->mnt
)) {
3980 if (!(mnt_has_parent(m
) ? check_mnt(m
) : is_anon_ns(m
->mnt_ns
))) {
3988 flags
= recalc_flags(kattr
, m
);
3989 if (!can_change_locked_flags(m
, flags
)) {
3994 *err
= can_idmap_mount(kattr
, m
);
4000 if ((kattr
->attr_set
& MNT_READONLY
) &&
4001 !(m
->mnt
.mnt_flags
& MNT_READONLY
)) {
4002 *err
= mnt_hold_writers(m
);
4006 } while (kattr
->recurse
&& (m
= next_mnt(m
, mnt
)));
4012 static void do_idmap_mount(const struct mount_kattr
*kattr
, struct mount
*mnt
)
4014 struct user_namespace
*mnt_userns
;
4016 if (!kattr
->mnt_userns
)
4019 mnt_userns
= get_user_ns(kattr
->mnt_userns
);
4020 /* Pairs with smp_load_acquire() in mnt_user_ns(). */
4021 smp_store_release(&mnt
->mnt
.mnt_userns
, mnt_userns
);
4024 static void mount_setattr_commit(struct mount_kattr
*kattr
,
4025 struct mount
*mnt
, struct mount
*last
,
4028 struct mount
*m
= mnt
;
4034 do_idmap_mount(kattr
, m
);
4035 flags
= recalc_flags(kattr
, m
);
4036 WRITE_ONCE(m
->mnt
.mnt_flags
, flags
);
4040 * We either set MNT_READONLY above so make it visible
4041 * before ~MNT_WRITE_HOLD or we failed to recursively
4042 * apply mount options.
4044 if ((kattr
->attr_set
& MNT_READONLY
) &&
4045 (m
->mnt
.mnt_flags
& MNT_WRITE_HOLD
))
4046 mnt_unhold_writers(m
);
4048 if (!err
&& kattr
->propagation
)
4049 change_mnt_propagation(m
, kattr
->propagation
);
4052 * On failure, only cleanup until we found the first mount
4053 * we failed to handle.
4055 if (err
&& m
== last
)
4057 } while (kattr
->recurse
&& (m
= next_mnt(m
, mnt
)));
4060 touch_mnt_namespace(mnt
->mnt_ns
);
4063 static int do_mount_setattr(struct path
*path
, struct mount_kattr
*kattr
)
4065 struct mount
*mnt
= real_mount(path
->mnt
), *last
= NULL
;
4068 if (path
->dentry
!= mnt
->mnt
.mnt_root
)
4071 if (kattr
->propagation
) {
4073 * Only take namespace_lock() if we're actually changing
4077 if (kattr
->propagation
== MS_SHARED
) {
4078 err
= invent_group_ids(mnt
, kattr
->recurse
);
4089 * Get the mount tree in a shape where we can change mount
4090 * properties without failure.
4092 last
= mount_setattr_prepare(kattr
, mnt
, &err
);
4093 if (last
) /* Commit all changes or revert to the old state. */
4094 mount_setattr_commit(kattr
, mnt
, last
, err
);
4096 unlock_mount_hash();
4098 if (kattr
->propagation
) {
4101 cleanup_group_ids(mnt
, NULL
);
4107 static int build_mount_idmapped(const struct mount_attr
*attr
, size_t usize
,
4108 struct mount_kattr
*kattr
, unsigned int flags
)
4111 struct ns_common
*ns
;
4112 struct user_namespace
*mnt_userns
;
4115 if (!((attr
->attr_set
| attr
->attr_clr
) & MOUNT_ATTR_IDMAP
))
4119 * We currently do not support clearing an idmapped mount. If this ever
4120 * is a use-case we can revisit this but for now let's keep it simple
4123 if (attr
->attr_clr
& MOUNT_ATTR_IDMAP
)
4126 if (attr
->userns_fd
> INT_MAX
)
4129 file
= fget(attr
->userns_fd
);
4133 if (!proc_ns_file(file
)) {
4138 ns
= get_proc_ns(file_inode(file
));
4139 if (ns
->ops
->type
!= CLONE_NEWUSER
) {
4145 * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4146 * This is simpler than just having to treat NULL as unmapped. Users
4147 * wanting to idmap a mount to init_user_ns can just use a namespace
4148 * with an identity mapping.
4150 mnt_userns
= container_of(ns
, struct user_namespace
, ns
);
4151 if (mnt_userns
== &init_user_ns
) {
4155 kattr
->mnt_userns
= get_user_ns(mnt_userns
);
4162 static int build_mount_kattr(const struct mount_attr
*attr
, size_t usize
,
4163 struct mount_kattr
*kattr
, unsigned int flags
)
4165 unsigned int lookup_flags
= LOOKUP_AUTOMOUNT
| LOOKUP_FOLLOW
;
4167 if (flags
& AT_NO_AUTOMOUNT
)
4168 lookup_flags
&= ~LOOKUP_AUTOMOUNT
;
4169 if (flags
& AT_SYMLINK_NOFOLLOW
)
4170 lookup_flags
&= ~LOOKUP_FOLLOW
;
4171 if (flags
& AT_EMPTY_PATH
)
4172 lookup_flags
|= LOOKUP_EMPTY
;
4174 *kattr
= (struct mount_kattr
) {
4175 .lookup_flags
= lookup_flags
,
4176 .recurse
= !!(flags
& AT_RECURSIVE
),
4179 if (attr
->propagation
& ~MOUNT_SETATTR_PROPAGATION_FLAGS
)
4181 if (hweight32(attr
->propagation
& MOUNT_SETATTR_PROPAGATION_FLAGS
) > 1)
4183 kattr
->propagation
= attr
->propagation
;
4185 if ((attr
->attr_set
| attr
->attr_clr
) & ~MOUNT_SETATTR_VALID_FLAGS
)
4188 kattr
->attr_set
= attr_flags_to_mnt_flags(attr
->attr_set
);
4189 kattr
->attr_clr
= attr_flags_to_mnt_flags(attr
->attr_clr
);
4192 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4193 * users wanting to transition to a different atime setting cannot
4194 * simply specify the atime setting in @attr_set, but must also
4195 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4196 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4197 * @attr_clr and that @attr_set can't have any atime bits set if
4198 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4200 if (attr
->attr_clr
& MOUNT_ATTR__ATIME
) {
4201 if ((attr
->attr_clr
& MOUNT_ATTR__ATIME
) != MOUNT_ATTR__ATIME
)
4205 * Clear all previous time settings as they are mutually
4208 kattr
->attr_clr
|= MNT_RELATIME
| MNT_NOATIME
;
4209 switch (attr
->attr_set
& MOUNT_ATTR__ATIME
) {
4210 case MOUNT_ATTR_RELATIME
:
4211 kattr
->attr_set
|= MNT_RELATIME
;
4213 case MOUNT_ATTR_NOATIME
:
4214 kattr
->attr_set
|= MNT_NOATIME
;
4216 case MOUNT_ATTR_STRICTATIME
:
4222 if (attr
->attr_set
& MOUNT_ATTR__ATIME
)
4226 return build_mount_idmapped(attr
, usize
, kattr
, flags
);
4229 static void finish_mount_kattr(struct mount_kattr
*kattr
)
4231 put_user_ns(kattr
->mnt_userns
);
4232 kattr
->mnt_userns
= NULL
;
4235 SYSCALL_DEFINE5(mount_setattr
, int, dfd
, const char __user
*, path
,
4236 unsigned int, flags
, struct mount_attr __user
*, uattr
,
4241 struct mount_attr attr
;
4242 struct mount_kattr kattr
;
4244 BUILD_BUG_ON(sizeof(struct mount_attr
) != MOUNT_ATTR_SIZE_VER0
);
4246 if (flags
& ~(AT_EMPTY_PATH
|
4248 AT_SYMLINK_NOFOLLOW
|
4252 if (unlikely(usize
> PAGE_SIZE
))
4254 if (unlikely(usize
< MOUNT_ATTR_SIZE_VER0
))
4260 err
= copy_struct_from_user(&attr
, sizeof(attr
), uattr
, usize
);
4264 /* Don't bother walking through the mounts if this is a nop. */
4265 if (attr
.attr_set
== 0 &&
4266 attr
.attr_clr
== 0 &&
4267 attr
.propagation
== 0)
4270 err
= build_mount_kattr(&attr
, usize
, &kattr
, flags
);
4274 err
= user_path_at(dfd
, path
, kattr
.lookup_flags
, &target
);
4276 err
= do_mount_setattr(&target
, &kattr
);
4279 finish_mount_kattr(&kattr
);
4283 static void __init
init_mount_tree(void)
4285 struct vfsmount
*mnt
;
4287 struct mnt_namespace
*ns
;
4290 mnt
= vfs_kern_mount(&rootfs_fs_type
, 0, "rootfs", NULL
);
4292 panic("Can't create rootfs");
4294 ns
= alloc_mnt_ns(&init_user_ns
, false);
4296 panic("Can't allocate initial namespace");
4297 m
= real_mount(mnt
);
4301 list_add(&m
->mnt_list
, &ns
->list
);
4302 init_task
.nsproxy
->mnt_ns
= ns
;
4306 root
.dentry
= mnt
->mnt_root
;
4307 mnt
->mnt_flags
|= MNT_LOCKED
;
4309 set_fs_pwd(current
->fs
, &root
);
4310 set_fs_root(current
->fs
, &root
);
4313 void __init
mnt_init(void)
4317 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
4318 0, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
|SLAB_ACCOUNT
, NULL
);
4320 mount_hashtable
= alloc_large_system_hash("Mount-cache",
4321 sizeof(struct hlist_head
),
4324 &m_hash_shift
, &m_hash_mask
, 0, 0);
4325 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
4326 sizeof(struct hlist_head
),
4329 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
4331 if (!mount_hashtable
|| !mountpoint_hashtable
)
4332 panic("Failed to allocate mount hash table\n");
4338 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
4340 fs_kobj
= kobject_create_and_add("fs", NULL
);
4342 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
4348 void put_mnt_ns(struct mnt_namespace
*ns
)
4350 if (!refcount_dec_and_test(&ns
->ns
.count
))
4352 drop_collected_mounts(&ns
->root
->mnt
);
4356 struct vfsmount
*kern_mount(struct file_system_type
*type
)
4358 struct vfsmount
*mnt
;
4359 mnt
= vfs_kern_mount(type
, SB_KERNMOUNT
, type
->name
, NULL
);
4362 * it is a longterm mount, don't release mnt until
4363 * we unmount before file sys is unregistered
4365 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
4369 EXPORT_SYMBOL_GPL(kern_mount
);
4371 void kern_unmount(struct vfsmount
*mnt
)
4373 /* release long term mount so mount point can be released */
4374 if (!IS_ERR_OR_NULL(mnt
)) {
4375 real_mount(mnt
)->mnt_ns
= NULL
;
4376 synchronize_rcu(); /* yecchhh... */
4380 EXPORT_SYMBOL(kern_unmount
);
4382 void kern_unmount_array(struct vfsmount
*mnt
[], unsigned int num
)
4386 for (i
= 0; i
< num
; i
++)
4388 real_mount(mnt
[i
])->mnt_ns
= NULL
;
4389 synchronize_rcu_expedited();
4390 for (i
= 0; i
< num
; i
++)
4393 EXPORT_SYMBOL(kern_unmount_array
);
4395 bool our_mnt(struct vfsmount
*mnt
)
4397 return check_mnt(real_mount(mnt
));
4400 bool current_chrooted(void)
4402 /* Does the current process have a non-standard root */
4403 struct path ns_root
;
4404 struct path fs_root
;
4407 /* Find the namespace root */
4408 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
4409 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
4411 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
4414 get_fs_root(current
->fs
, &fs_root
);
4416 chrooted
= !path_equal(&fs_root
, &ns_root
);
4424 static bool mnt_already_visible(struct mnt_namespace
*ns
,
4425 const struct super_block
*sb
,
4428 int new_flags
= *new_mnt_flags
;
4430 bool visible
= false;
4432 down_read(&namespace_sem
);
4434 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
4435 struct mount
*child
;
4438 if (mnt_is_cursor(mnt
))
4441 if (mnt
->mnt
.mnt_sb
->s_type
!= sb
->s_type
)
4444 /* This mount is not fully visible if it's root directory
4445 * is not the root directory of the filesystem.
4447 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
4450 /* A local view of the mount flags */
4451 mnt_flags
= mnt
->mnt
.mnt_flags
;
4453 /* Don't miss readonly hidden in the superblock flags */
4454 if (sb_rdonly(mnt
->mnt
.mnt_sb
))
4455 mnt_flags
|= MNT_LOCK_READONLY
;
4457 /* Verify the mount flags are equal to or more permissive
4458 * than the proposed new mount.
4460 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
4461 !(new_flags
& MNT_READONLY
))
4463 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
4464 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
4467 /* This mount is not fully visible if there are any
4468 * locked child mounts that cover anything except for
4469 * empty directories.
4471 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
4472 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
4473 /* Only worry about locked mounts */
4474 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
4476 /* Is the directory permanetly empty? */
4477 if (!is_empty_dir_inode(inode
))
4480 /* Preserve the locked attributes */
4481 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
4489 up_read(&namespace_sem
);
4493 static bool mount_too_revealing(const struct super_block
*sb
, int *new_mnt_flags
)
4495 const unsigned long required_iflags
= SB_I_NOEXEC
| SB_I_NODEV
;
4496 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
4497 unsigned long s_iflags
;
4499 if (ns
->user_ns
== &init_user_ns
)
4502 /* Can this filesystem be too revealing? */
4503 s_iflags
= sb
->s_iflags
;
4504 if (!(s_iflags
& SB_I_USERNS_VISIBLE
))
4507 if ((s_iflags
& required_iflags
) != required_iflags
) {
4508 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4513 return !mnt_already_visible(ns
, sb
, new_mnt_flags
);
4516 bool mnt_may_suid(struct vfsmount
*mnt
)
4519 * Foreign mounts (accessed via fchdir or through /proc
4520 * symlinks) are always treated as if they are nosuid. This
4521 * prevents namespaces from trusting potentially unsafe
4522 * suid/sgid bits, file caps, or security labels that originate
4523 * in other namespaces.
4525 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
4526 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
4529 static struct ns_common
*mntns_get(struct task_struct
*task
)
4531 struct ns_common
*ns
= NULL
;
4532 struct nsproxy
*nsproxy
;
4535 nsproxy
= task
->nsproxy
;
4537 ns
= &nsproxy
->mnt_ns
->ns
;
4538 get_mnt_ns(to_mnt_ns(ns
));
4545 static void mntns_put(struct ns_common
*ns
)
4547 put_mnt_ns(to_mnt_ns(ns
));
4550 static int mntns_install(struct nsset
*nsset
, struct ns_common
*ns
)
4552 struct nsproxy
*nsproxy
= nsset
->nsproxy
;
4553 struct fs_struct
*fs
= nsset
->fs
;
4554 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
), *old_mnt_ns
;
4555 struct user_namespace
*user_ns
= nsset
->cred
->user_ns
;
4559 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
4560 !ns_capable(user_ns
, CAP_SYS_CHROOT
) ||
4561 !ns_capable(user_ns
, CAP_SYS_ADMIN
))
4564 if (is_anon_ns(mnt_ns
))
4571 old_mnt_ns
= nsproxy
->mnt_ns
;
4572 nsproxy
->mnt_ns
= mnt_ns
;
4575 err
= vfs_path_lookup(mnt_ns
->root
->mnt
.mnt_root
, &mnt_ns
->root
->mnt
,
4576 "/", LOOKUP_DOWN
, &root
);
4578 /* revert to old namespace */
4579 nsproxy
->mnt_ns
= old_mnt_ns
;
4584 put_mnt_ns(old_mnt_ns
);
4586 /* Update the pwd and root */
4587 set_fs_pwd(fs
, &root
);
4588 set_fs_root(fs
, &root
);
4594 static struct user_namespace
*mntns_owner(struct ns_common
*ns
)
4596 return to_mnt_ns(ns
)->user_ns
;
4599 const struct proc_ns_operations mntns_operations
= {
4601 .type
= CLONE_NEWNS
,
4604 .install
= mntns_install
,
4605 .owner
= mntns_owner
,