4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/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/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
27 #include <linux/task_work.h>
28 #include <linux/sched/task.h>
33 /* Maximum number of mounts in a mount namespace */
34 unsigned int sysctl_mount_max __read_mostly
= 100000;
36 static unsigned int m_hash_mask __read_mostly
;
37 static unsigned int m_hash_shift __read_mostly
;
38 static unsigned int mp_hash_mask __read_mostly
;
39 static unsigned int mp_hash_shift __read_mostly
;
41 static __initdata
unsigned long mhash_entries
;
42 static int __init
set_mhash_entries(char *str
)
46 mhash_entries
= simple_strtoul(str
, &str
, 0);
49 __setup("mhash_entries=", set_mhash_entries
);
51 static __initdata
unsigned long mphash_entries
;
52 static int __init
set_mphash_entries(char *str
)
56 mphash_entries
= simple_strtoul(str
, &str
, 0);
59 __setup("mphash_entries=", set_mphash_entries
);
62 static DEFINE_IDA(mnt_id_ida
);
63 static DEFINE_IDA(mnt_group_ida
);
64 static DEFINE_SPINLOCK(mnt_id_lock
);
65 static int mnt_id_start
= 0;
66 static int mnt_group_start
= 1;
68 static struct hlist_head
*mount_hashtable __read_mostly
;
69 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
70 static struct kmem_cache
*mnt_cache __read_mostly
;
71 static DECLARE_RWSEM(namespace_sem
);
74 struct kobject
*fs_kobj
;
75 EXPORT_SYMBOL_GPL(fs_kobj
);
78 * vfsmount lock may be taken for read to prevent changes to the
79 * vfsmount hash, ie. during mountpoint lookups or walking back
82 * It should be taken for write in all cases where the vfsmount
83 * tree or hash is modified or when a vfsmount structure is modified.
85 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
87 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
89 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
90 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
91 tmp
= tmp
+ (tmp
>> m_hash_shift
);
92 return &mount_hashtable
[tmp
& m_hash_mask
];
95 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
97 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
98 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
99 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
102 static int mnt_alloc_id(struct mount
*mnt
)
107 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
108 spin_lock(&mnt_id_lock
);
109 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
111 mnt_id_start
= mnt
->mnt_id
+ 1;
112 spin_unlock(&mnt_id_lock
);
119 static void mnt_free_id(struct mount
*mnt
)
121 int id
= mnt
->mnt_id
;
122 spin_lock(&mnt_id_lock
);
123 ida_remove(&mnt_id_ida
, id
);
124 if (mnt_id_start
> id
)
126 spin_unlock(&mnt_id_lock
);
130 * Allocate a new peer group ID
132 * mnt_group_ida is protected by namespace_sem
134 static int mnt_alloc_group_id(struct mount
*mnt
)
138 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
141 res
= ida_get_new_above(&mnt_group_ida
,
145 mnt_group_start
= mnt
->mnt_group_id
+ 1;
151 * Release a peer group ID
153 void mnt_release_group_id(struct mount
*mnt
)
155 int id
= mnt
->mnt_group_id
;
156 ida_remove(&mnt_group_ida
, id
);
157 if (mnt_group_start
> id
)
158 mnt_group_start
= 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 unsigned int mnt_get_count(struct mount
*mnt
)
182 unsigned int count
= 0;
185 for_each_possible_cpu(cpu
) {
186 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
191 return mnt
->mnt_count
;
195 static void drop_mountpoint(struct fs_pin
*p
)
197 struct mount
*m
= container_of(p
, struct mount
, mnt_umount
);
198 dput(m
->mnt_ex_mountpoint
);
203 static struct mount
*alloc_vfsmnt(const char *name
)
205 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
209 err
= mnt_alloc_id(mnt
);
214 mnt
->mnt_devname
= kstrdup_const(name
, GFP_KERNEL
);
215 if (!mnt
->mnt_devname
)
220 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
222 goto out_free_devname
;
224 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
227 mnt
->mnt_writers
= 0;
230 INIT_HLIST_NODE(&mnt
->mnt_hash
);
231 INIT_LIST_HEAD(&mnt
->mnt_child
);
232 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
233 INIT_LIST_HEAD(&mnt
->mnt_list
);
234 INIT_LIST_HEAD(&mnt
->mnt_expire
);
235 INIT_LIST_HEAD(&mnt
->mnt_share
);
236 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
237 INIT_LIST_HEAD(&mnt
->mnt_slave
);
238 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
239 INIT_LIST_HEAD(&mnt
->mnt_umounting
);
240 init_fs_pin(&mnt
->mnt_umount
, drop_mountpoint
);
246 kfree_const(mnt
->mnt_devname
);
251 kmem_cache_free(mnt_cache
, mnt
);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount
*mnt
)
276 if (mnt
->mnt_flags
& MNT_READONLY
)
278 if (sb_rdonly(mnt
->mnt_sb
))
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
284 static inline void mnt_inc_writers(struct mount
*mnt
)
287 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
293 static inline void mnt_dec_writers(struct mount
*mnt
)
296 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
302 static unsigned int mnt_get_writers(struct mount
*mnt
)
305 unsigned int count
= 0;
308 for_each_possible_cpu(cpu
) {
309 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
314 return mnt
->mnt_writers
;
318 static int mnt_is_readonly(struct vfsmount
*mnt
)
320 if (mnt
->mnt_sb
->s_readonly_remount
)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt
);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount
*m
)
345 struct mount
*mnt
= real_mount(m
);
349 mnt_inc_writers(mnt
);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (READ_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m
)) {
365 mnt_dec_writers(mnt
);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount
*m
)
386 sb_start_write(m
->mnt_sb
);
387 ret
= __mnt_want_write(m
);
389 sb_end_write(m
->mnt_sb
);
392 EXPORT_SYMBOL_GPL(mnt_want_write
);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount
*mnt
)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt
))
412 mnt_inc_writers(real_mount(mnt
));
416 EXPORT_SYMBOL_GPL(mnt_clone_write
);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file
*file
)
427 if (!(file
->f_mode
& FMODE_WRITER
))
428 return __mnt_want_write(file
->f_path
.mnt
);
430 return mnt_clone_write(file
->f_path
.mnt
);
434 * mnt_want_write_file_path - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 * Called by the vfs for cases when we have an open file at hand, but will do an
441 * inode operation on it (important distinction for files opened on overlayfs,
442 * since the file operations will come from the real underlying file, while
443 * inode operations come from the overlay).
445 int mnt_want_write_file_path(struct file
*file
)
449 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
450 ret
= __mnt_want_write_file(file
);
452 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
456 static inline int may_write_real(struct file
*file
)
458 struct dentry
*dentry
= file
->f_path
.dentry
;
459 struct dentry
*upperdentry
;
462 if (file
->f_mode
& FMODE_WRITER
)
466 if (likely(!(dentry
->d_flags
& DCACHE_OP_REAL
)))
469 /* File refers to upper, writable layer? */
470 upperdentry
= d_real(dentry
, NULL
, 0, D_REAL_UPPER
);
472 (file_inode(file
) == d_inode(upperdentry
) ||
473 file_inode(file
) == d_inode(dentry
)))
476 /* Lower layer: can't write to real file, sorry... */
481 * mnt_want_write_file - get write access to a file's mount
482 * @file: the file who's mount on which to take a write
484 * This is like mnt_want_write, but it takes a file and can
485 * do some optimisations if the file is open for write already
487 * Mostly called by filesystems from their ioctl operation before performing
488 * modification. On overlayfs this needs to check if the file is on a read-only
489 * lower layer and deny access in that case.
491 int mnt_want_write_file(struct file
*file
)
495 ret
= may_write_real(file
);
497 sb_start_write(file_inode(file
)->i_sb
);
498 ret
= __mnt_want_write_file(file
);
500 sb_end_write(file_inode(file
)->i_sb
);
504 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
507 * __mnt_drop_write - give up write access to a mount
508 * @mnt: the mount on which to give up write access
510 * Tells the low-level filesystem that we are done
511 * performing writes to it. Must be matched with
512 * __mnt_want_write() call above.
514 void __mnt_drop_write(struct vfsmount
*mnt
)
517 mnt_dec_writers(real_mount(mnt
));
520 EXPORT_SYMBOL_GPL(__mnt_drop_write
);
523 * mnt_drop_write - give up write access to a mount
524 * @mnt: the mount on which to give up write access
526 * Tells the low-level filesystem that we are done performing writes to it and
527 * also allows filesystem to be frozen again. Must be matched with
528 * mnt_want_write() call above.
530 void mnt_drop_write(struct vfsmount
*mnt
)
532 __mnt_drop_write(mnt
);
533 sb_end_write(mnt
->mnt_sb
);
535 EXPORT_SYMBOL_GPL(mnt_drop_write
);
537 void __mnt_drop_write_file(struct file
*file
)
539 __mnt_drop_write(file
->f_path
.mnt
);
542 void mnt_drop_write_file_path(struct file
*file
)
544 mnt_drop_write(file
->f_path
.mnt
);
547 void mnt_drop_write_file(struct file
*file
)
549 __mnt_drop_write(file
->f_path
.mnt
);
550 sb_end_write(file_inode(file
)->i_sb
);
552 EXPORT_SYMBOL(mnt_drop_write_file
);
554 static int mnt_make_readonly(struct mount
*mnt
)
559 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
561 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
562 * should be visible before we do.
567 * With writers on hold, if this value is zero, then there are
568 * definitely no active writers (although held writers may subsequently
569 * increment the count, they'll have to wait, and decrement it after
570 * seeing MNT_READONLY).
572 * It is OK to have counter incremented on one CPU and decremented on
573 * another: the sum will add up correctly. The danger would be when we
574 * sum up each counter, if we read a counter before it is incremented,
575 * but then read another CPU's count which it has been subsequently
576 * decremented from -- we would see more decrements than we should.
577 * MNT_WRITE_HOLD protects against this scenario, because
578 * mnt_want_write first increments count, then smp_mb, then spins on
579 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
580 * we're counting up here.
582 if (mnt_get_writers(mnt
) > 0)
585 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
587 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
588 * that become unheld will see MNT_READONLY.
591 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
596 static void __mnt_unmake_readonly(struct mount
*mnt
)
599 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
603 int sb_prepare_remount_readonly(struct super_block
*sb
)
608 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
609 if (atomic_long_read(&sb
->s_remove_count
))
613 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
614 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
615 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
617 if (mnt_get_writers(mnt
) > 0) {
623 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
627 sb
->s_readonly_remount
= 1;
630 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
631 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
632 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
639 static void free_vfsmnt(struct mount
*mnt
)
641 kfree_const(mnt
->mnt_devname
);
643 free_percpu(mnt
->mnt_pcp
);
645 kmem_cache_free(mnt_cache
, mnt
);
648 static void delayed_free_vfsmnt(struct rcu_head
*head
)
650 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
653 /* call under rcu_read_lock */
654 int __legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
657 if (read_seqretry(&mount_lock
, seq
))
661 mnt
= real_mount(bastard
);
662 mnt_add_count(mnt
, 1);
663 if (likely(!read_seqretry(&mount_lock
, seq
)))
665 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
666 mnt_add_count(mnt
, -1);
672 /* call under rcu_read_lock */
673 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
675 int res
= __legitimize_mnt(bastard
, seq
);
678 if (unlikely(res
< 0)) {
687 * find the first mount at @dentry on vfsmount @mnt.
688 * call under rcu_read_lock()
690 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
692 struct hlist_head
*head
= m_hash(mnt
, dentry
);
695 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
696 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
702 * lookup_mnt - Return the first child mount mounted at path
704 * "First" means first mounted chronologically. If you create the
707 * mount /dev/sda1 /mnt
708 * mount /dev/sda2 /mnt
709 * mount /dev/sda3 /mnt
711 * Then lookup_mnt() on the base /mnt dentry in the root mount will
712 * return successively the root dentry and vfsmount of /dev/sda1, then
713 * /dev/sda2, then /dev/sda3, then NULL.
715 * lookup_mnt takes a reference to the found vfsmount.
717 struct vfsmount
*lookup_mnt(const struct path
*path
)
719 struct mount
*child_mnt
;
725 seq
= read_seqbegin(&mount_lock
);
726 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
727 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
728 } while (!legitimize_mnt(m
, seq
));
734 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
735 * current mount namespace.
737 * The common case is dentries are not mountpoints at all and that
738 * test is handled inline. For the slow case when we are actually
739 * dealing with a mountpoint of some kind, walk through all of the
740 * mounts in the current mount namespace and test to see if the dentry
743 * The mount_hashtable is not usable in the context because we
744 * need to identify all mounts that may be in the current mount
745 * namespace not just a mount that happens to have some specified
748 bool __is_local_mountpoint(struct dentry
*dentry
)
750 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
752 bool is_covered
= false;
754 if (!d_mountpoint(dentry
))
757 down_read(&namespace_sem
);
758 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
759 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
763 up_read(&namespace_sem
);
768 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
770 struct hlist_head
*chain
= mp_hash(dentry
);
771 struct mountpoint
*mp
;
773 hlist_for_each_entry(mp
, chain
, m_hash
) {
774 if (mp
->m_dentry
== dentry
) {
775 /* might be worth a WARN_ON() */
776 if (d_unlinked(dentry
))
777 return ERR_PTR(-ENOENT
);
785 static struct mountpoint
*get_mountpoint(struct dentry
*dentry
)
787 struct mountpoint
*mp
, *new = NULL
;
790 if (d_mountpoint(dentry
)) {
792 read_seqlock_excl(&mount_lock
);
793 mp
= lookup_mountpoint(dentry
);
794 read_sequnlock_excl(&mount_lock
);
800 new = kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
802 return ERR_PTR(-ENOMEM
);
805 /* Exactly one processes may set d_mounted */
806 ret
= d_set_mounted(dentry
);
808 /* Someone else set d_mounted? */
812 /* The dentry is not available as a mountpoint? */
817 /* Add the new mountpoint to the hash table */
818 read_seqlock_excl(&mount_lock
);
819 new->m_dentry
= dentry
;
821 hlist_add_head(&new->m_hash
, mp_hash(dentry
));
822 INIT_HLIST_HEAD(&new->m_list
);
823 read_sequnlock_excl(&mount_lock
);
832 static void put_mountpoint(struct mountpoint
*mp
)
834 if (!--mp
->m_count
) {
835 struct dentry
*dentry
= mp
->m_dentry
;
836 BUG_ON(!hlist_empty(&mp
->m_list
));
837 spin_lock(&dentry
->d_lock
);
838 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
839 spin_unlock(&dentry
->d_lock
);
840 hlist_del(&mp
->m_hash
);
845 static inline int check_mnt(struct mount
*mnt
)
847 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
850 /* for aufs, CONFIG_AUFS_BR_FUSE */
851 int is_current_mnt_ns(struct vfsmount
*mnt
)
853 return check_mnt(real_mount(mnt
));
855 EXPORT_SYMBOL_GPL(is_current_mnt_ns
);
858 * vfsmount lock must be held for write
860 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
864 wake_up_interruptible(&ns
->poll
);
869 * vfsmount lock must be held for write
871 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
873 if (ns
&& ns
->event
!= event
) {
875 wake_up_interruptible(&ns
->poll
);
880 * vfsmount lock must be held for write
882 static void unhash_mnt(struct mount
*mnt
)
884 mnt
->mnt_parent
= mnt
;
885 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
886 list_del_init(&mnt
->mnt_child
);
887 hlist_del_init_rcu(&mnt
->mnt_hash
);
888 hlist_del_init(&mnt
->mnt_mp_list
);
889 put_mountpoint(mnt
->mnt_mp
);
894 * vfsmount lock must be held for write
896 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
898 old_path
->dentry
= mnt
->mnt_mountpoint
;
899 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
904 * vfsmount lock must be held for write
906 static void umount_mnt(struct mount
*mnt
)
908 /* old mountpoint will be dropped when we can do that */
909 mnt
->mnt_ex_mountpoint
= mnt
->mnt_mountpoint
;
914 * vfsmount lock must be held for write
916 void mnt_set_mountpoint(struct mount
*mnt
,
917 struct mountpoint
*mp
,
918 struct mount
*child_mnt
)
921 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
922 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
923 child_mnt
->mnt_parent
= mnt
;
924 child_mnt
->mnt_mp
= mp
;
925 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
928 static void __attach_mnt(struct mount
*mnt
, struct mount
*parent
)
930 hlist_add_head_rcu(&mnt
->mnt_hash
,
931 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
932 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
936 * vfsmount lock must be held for write
938 static void attach_mnt(struct mount
*mnt
,
939 struct mount
*parent
,
940 struct mountpoint
*mp
)
942 mnt_set_mountpoint(parent
, mp
, mnt
);
943 __attach_mnt(mnt
, parent
);
946 void mnt_change_mountpoint(struct mount
*parent
, struct mountpoint
*mp
, struct mount
*mnt
)
948 struct mountpoint
*old_mp
= mnt
->mnt_mp
;
949 struct dentry
*old_mountpoint
= mnt
->mnt_mountpoint
;
950 struct mount
*old_parent
= mnt
->mnt_parent
;
952 list_del_init(&mnt
->mnt_child
);
953 hlist_del_init(&mnt
->mnt_mp_list
);
954 hlist_del_init_rcu(&mnt
->mnt_hash
);
956 attach_mnt(mnt
, parent
, mp
);
958 put_mountpoint(old_mp
);
961 * Safely avoid even the suggestion this code might sleep or
962 * lock the mount hash by taking advantage of the knowledge that
963 * mnt_change_mountpoint will not release the final reference
966 * During mounting, the mount passed in as the parent mount will
967 * continue to use the old mountpoint and during unmounting, the
968 * old mountpoint will continue to exist until namespace_unlock,
969 * which happens well after mnt_change_mountpoint.
971 spin_lock(&old_mountpoint
->d_lock
);
972 old_mountpoint
->d_lockref
.count
--;
973 spin_unlock(&old_mountpoint
->d_lock
);
975 mnt_add_count(old_parent
, -1);
979 * vfsmount lock must be held for write
981 static void commit_tree(struct mount
*mnt
)
983 struct mount
*parent
= mnt
->mnt_parent
;
986 struct mnt_namespace
*n
= parent
->mnt_ns
;
988 BUG_ON(parent
== mnt
);
990 list_add_tail(&head
, &mnt
->mnt_list
);
991 list_for_each_entry(m
, &head
, mnt_list
)
994 list_splice(&head
, n
->list
.prev
);
996 n
->mounts
+= n
->pending_mounts
;
997 n
->pending_mounts
= 0;
999 __attach_mnt(mnt
, parent
);
1000 touch_mnt_namespace(n
);
1003 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
1005 struct list_head
*next
= p
->mnt_mounts
.next
;
1006 if (next
== &p
->mnt_mounts
) {
1010 next
= p
->mnt_child
.next
;
1011 if (next
!= &p
->mnt_parent
->mnt_mounts
)
1016 return list_entry(next
, struct mount
, mnt_child
);
1019 static struct mount
*skip_mnt_tree(struct mount
*p
)
1021 struct list_head
*prev
= p
->mnt_mounts
.prev
;
1022 while (prev
!= &p
->mnt_mounts
) {
1023 p
= list_entry(prev
, struct mount
, mnt_child
);
1024 prev
= p
->mnt_mounts
.prev
;
1030 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
1033 struct dentry
*root
;
1036 return ERR_PTR(-ENODEV
);
1038 mnt
= alloc_vfsmnt(name
);
1040 return ERR_PTR(-ENOMEM
);
1042 if (flags
& SB_KERNMOUNT
)
1043 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
1045 root
= mount_fs(type
, flags
, name
, data
);
1049 return ERR_CAST(root
);
1052 mnt
->mnt
.mnt_root
= root
;
1053 mnt
->mnt
.mnt_sb
= root
->d_sb
;
1054 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1055 mnt
->mnt_parent
= mnt
;
1057 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
1058 unlock_mount_hash();
1061 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
1064 vfs_submount(const struct dentry
*mountpoint
, struct file_system_type
*type
,
1065 const char *name
, void *data
)
1067 /* Until it is worked out how to pass the user namespace
1068 * through from the parent mount to the submount don't support
1069 * unprivileged mounts with submounts.
1071 if (mountpoint
->d_sb
->s_user_ns
!= &init_user_ns
)
1072 return ERR_PTR(-EPERM
);
1074 return vfs_kern_mount(type
, SB_SUBMOUNT
, name
, data
);
1076 EXPORT_SYMBOL_GPL(vfs_submount
);
1078 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
1081 struct super_block
*sb
= old
->mnt
.mnt_sb
;
1085 mnt
= alloc_vfsmnt(old
->mnt_devname
);
1087 return ERR_PTR(-ENOMEM
);
1089 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
1090 mnt
->mnt_group_id
= 0; /* not a peer of original */
1092 mnt
->mnt_group_id
= old
->mnt_group_id
;
1094 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
1095 err
= mnt_alloc_group_id(mnt
);
1100 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~(MNT_WRITE_HOLD
|MNT_MARKED
);
1101 /* Don't allow unprivileged users to change mount flags */
1102 if (flag
& CL_UNPRIVILEGED
) {
1103 mnt
->mnt
.mnt_flags
|= MNT_LOCK_ATIME
;
1105 if (mnt
->mnt
.mnt_flags
& MNT_READONLY
)
1106 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
1108 if (mnt
->mnt
.mnt_flags
& MNT_NODEV
)
1109 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NODEV
;
1111 if (mnt
->mnt
.mnt_flags
& MNT_NOSUID
)
1112 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOSUID
;
1114 if (mnt
->mnt
.mnt_flags
& MNT_NOEXEC
)
1115 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOEXEC
;
1118 /* Don't allow unprivileged users to reveal what is under a mount */
1119 if ((flag
& CL_UNPRIVILEGED
) &&
1120 (!(flag
& CL_EXPIRE
) || list_empty(&old
->mnt_expire
)))
1121 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
1123 atomic_inc(&sb
->s_active
);
1124 mnt
->mnt
.mnt_sb
= sb
;
1125 mnt
->mnt
.mnt_root
= dget(root
);
1126 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
1127 mnt
->mnt_parent
= mnt
;
1129 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
1130 unlock_mount_hash();
1132 if ((flag
& CL_SLAVE
) ||
1133 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
1134 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
1135 mnt
->mnt_master
= old
;
1136 CLEAR_MNT_SHARED(mnt
);
1137 } else if (!(flag
& CL_PRIVATE
)) {
1138 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
1139 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
1140 if (IS_MNT_SLAVE(old
))
1141 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
1142 mnt
->mnt_master
= old
->mnt_master
;
1144 CLEAR_MNT_SHARED(mnt
);
1146 if (flag
& CL_MAKE_SHARED
)
1147 set_mnt_shared(mnt
);
1149 /* stick the duplicate mount on the same expiry list
1150 * as the original if that was on one */
1151 if (flag
& CL_EXPIRE
) {
1152 if (!list_empty(&old
->mnt_expire
))
1153 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1161 return ERR_PTR(err
);
1164 static void cleanup_mnt(struct mount
*mnt
)
1167 * This probably indicates that somebody messed
1168 * up a mnt_want/drop_write() pair. If this
1169 * happens, the filesystem was probably unable
1170 * to make r/w->r/o transitions.
1173 * The locking used to deal with mnt_count decrement provides barriers,
1174 * so mnt_get_writers() below is safe.
1176 WARN_ON(mnt_get_writers(mnt
));
1177 if (unlikely(mnt
->mnt_pins
.first
))
1179 fsnotify_vfsmount_delete(&mnt
->mnt
);
1180 dput(mnt
->mnt
.mnt_root
);
1181 deactivate_super(mnt
->mnt
.mnt_sb
);
1183 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1186 static void __cleanup_mnt(struct rcu_head
*head
)
1188 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1191 static LLIST_HEAD(delayed_mntput_list
);
1192 static void delayed_mntput(struct work_struct
*unused
)
1194 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1195 struct mount
*m
, *t
;
1197 llist_for_each_entry_safe(m
, t
, node
, mnt_llist
)
1200 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1202 static void mntput_no_expire(struct mount
*mnt
)
1205 mnt_add_count(mnt
, -1);
1206 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
1211 if (mnt_get_count(mnt
)) {
1213 unlock_mount_hash();
1216 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1218 unlock_mount_hash();
1221 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1224 list_del(&mnt
->mnt_instance
);
1226 if (unlikely(!list_empty(&mnt
->mnt_mounts
))) {
1227 struct mount
*p
, *tmp
;
1228 list_for_each_entry_safe(p
, tmp
, &mnt
->mnt_mounts
, mnt_child
) {
1232 unlock_mount_hash();
1234 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1235 struct task_struct
*task
= current
;
1236 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1237 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1238 if (!task_work_add(task
, &mnt
->mnt_rcu
, true))
1241 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1242 schedule_delayed_work(&delayed_mntput_work
, 1);
1248 void mntput(struct vfsmount
*mnt
)
1251 struct mount
*m
= real_mount(mnt
);
1252 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1253 if (unlikely(m
->mnt_expiry_mark
))
1254 m
->mnt_expiry_mark
= 0;
1255 mntput_no_expire(m
);
1258 EXPORT_SYMBOL(mntput
);
1260 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1263 mnt_add_count(real_mount(mnt
), 1);
1266 EXPORT_SYMBOL(mntget
);
1268 /* path_is_mountpoint() - Check if path is a mount in the current
1271 * d_mountpoint() can only be used reliably to establish if a dentry is
1272 * not mounted in any namespace and that common case is handled inline.
1273 * d_mountpoint() isn't aware of the possibility there may be multiple
1274 * mounts using a given dentry in a different namespace. This function
1275 * checks if the passed in path is a mountpoint rather than the dentry
1278 bool path_is_mountpoint(const struct path
*path
)
1283 if (!d_mountpoint(path
->dentry
))
1288 seq
= read_seqbegin(&mount_lock
);
1289 res
= __path_is_mountpoint(path
);
1290 } while (read_seqretry(&mount_lock
, seq
));
1295 EXPORT_SYMBOL(path_is_mountpoint
);
1297 struct vfsmount
*mnt_clone_internal(const struct path
*path
)
1300 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1303 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1307 #ifdef CONFIG_PROC_FS
1308 /* iterator; we want it to have access to namespace_sem, thus here... */
1309 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1311 struct proc_mounts
*p
= m
->private;
1313 down_read(&namespace_sem
);
1314 if (p
->cached_event
== p
->ns
->event
) {
1315 void *v
= p
->cached_mount
;
1316 if (*pos
== p
->cached_index
)
1318 if (*pos
== p
->cached_index
+ 1) {
1319 v
= seq_list_next(v
, &p
->ns
->list
, &p
->cached_index
);
1320 return p
->cached_mount
= v
;
1324 p
->cached_event
= p
->ns
->event
;
1325 p
->cached_mount
= seq_list_start(&p
->ns
->list
, *pos
);
1326 p
->cached_index
= *pos
;
1327 return p
->cached_mount
;
1330 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1332 struct proc_mounts
*p
= m
->private;
1334 p
->cached_mount
= seq_list_next(v
, &p
->ns
->list
, pos
);
1335 p
->cached_index
= *pos
;
1336 return p
->cached_mount
;
1339 static void m_stop(struct seq_file
*m
, void *v
)
1341 up_read(&namespace_sem
);
1344 static int m_show(struct seq_file
*m
, void *v
)
1346 struct proc_mounts
*p
= m
->private;
1347 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1348 return p
->show(m
, &r
->mnt
);
1351 const struct seq_operations mounts_op
= {
1357 #endif /* CONFIG_PROC_FS */
1360 * may_umount_tree - check if a mount tree is busy
1361 * @mnt: root of mount tree
1363 * This is called to check if a tree of mounts has any
1364 * open files, pwds, chroots or sub mounts that are
1367 int may_umount_tree(struct vfsmount
*m
)
1369 struct mount
*mnt
= real_mount(m
);
1370 int actual_refs
= 0;
1371 int minimum_refs
= 0;
1375 /* write lock needed for mnt_get_count */
1377 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1378 actual_refs
+= mnt_get_count(p
);
1381 unlock_mount_hash();
1383 if (actual_refs
> minimum_refs
)
1389 EXPORT_SYMBOL(may_umount_tree
);
1392 * may_umount - check if a mount point is busy
1393 * @mnt: root of mount
1395 * This is called to check if a mount point has any
1396 * open files, pwds, chroots or sub mounts. If the
1397 * mount has sub mounts this will return busy
1398 * regardless of whether the sub mounts are busy.
1400 * Doesn't take quota and stuff into account. IOW, in some cases it will
1401 * give false negatives. The main reason why it's here is that we need
1402 * a non-destructive way to look for easily umountable filesystems.
1404 int may_umount(struct vfsmount
*mnt
)
1407 down_read(&namespace_sem
);
1409 if (propagate_mount_busy(real_mount(mnt
), 2))
1411 unlock_mount_hash();
1412 up_read(&namespace_sem
);
1416 EXPORT_SYMBOL(may_umount
);
1418 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
1420 static void namespace_unlock(void)
1422 struct hlist_head head
;
1424 hlist_move_list(&unmounted
, &head
);
1426 up_write(&namespace_sem
);
1428 if (likely(hlist_empty(&head
)))
1433 group_pin_kill(&head
);
1436 static inline void namespace_lock(void)
1438 down_write(&namespace_sem
);
1441 enum umount_tree_flags
{
1443 UMOUNT_PROPAGATE
= 2,
1444 UMOUNT_CONNECTED
= 4,
1447 static bool disconnect_mount(struct mount
*mnt
, enum umount_tree_flags how
)
1449 /* Leaving mounts connected is only valid for lazy umounts */
1450 if (how
& UMOUNT_SYNC
)
1453 /* A mount without a parent has nothing to be connected to */
1454 if (!mnt_has_parent(mnt
))
1457 /* Because the reference counting rules change when mounts are
1458 * unmounted and connected, umounted mounts may not be
1459 * connected to mounted mounts.
1461 if (!(mnt
->mnt_parent
->mnt
.mnt_flags
& MNT_UMOUNT
))
1464 /* Has it been requested that the mount remain connected? */
1465 if (how
& UMOUNT_CONNECTED
)
1468 /* Is the mount locked such that it needs to remain connected? */
1469 if (IS_MNT_LOCKED(mnt
))
1472 /* By default disconnect the mount */
1477 * mount_lock must be held
1478 * namespace_sem must be held for write
1480 static void umount_tree(struct mount
*mnt
, enum umount_tree_flags how
)
1482 LIST_HEAD(tmp_list
);
1485 if (how
& UMOUNT_PROPAGATE
)
1486 propagate_mount_unlock(mnt
);
1488 /* Gather the mounts to umount */
1489 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1490 p
->mnt
.mnt_flags
|= MNT_UMOUNT
;
1491 list_move(&p
->mnt_list
, &tmp_list
);
1494 /* Hide the mounts from mnt_mounts */
1495 list_for_each_entry(p
, &tmp_list
, mnt_list
) {
1496 list_del_init(&p
->mnt_child
);
1499 /* Add propogated mounts to the tmp_list */
1500 if (how
& UMOUNT_PROPAGATE
)
1501 propagate_umount(&tmp_list
);
1503 while (!list_empty(&tmp_list
)) {
1504 struct mnt_namespace
*ns
;
1506 p
= list_first_entry(&tmp_list
, struct mount
, mnt_list
);
1507 list_del_init(&p
->mnt_expire
);
1508 list_del_init(&p
->mnt_list
);
1512 __touch_mnt_namespace(ns
);
1515 if (how
& UMOUNT_SYNC
)
1516 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1518 disconnect
= disconnect_mount(p
, how
);
1520 pin_insert_group(&p
->mnt_umount
, &p
->mnt_parent
->mnt
,
1521 disconnect
? &unmounted
: NULL
);
1522 if (mnt_has_parent(p
)) {
1523 mnt_add_count(p
->mnt_parent
, -1);
1525 /* Don't forget about p */
1526 list_add_tail(&p
->mnt_child
, &p
->mnt_parent
->mnt_mounts
);
1531 change_mnt_propagation(p
, MS_PRIVATE
);
1535 static void shrink_submounts(struct mount
*mnt
);
1537 static int do_umount(struct mount
*mnt
, int flags
)
1539 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1542 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1547 * Allow userspace to request a mountpoint be expired rather than
1548 * unmounting unconditionally. Unmount only happens if:
1549 * (1) the mark is already set (the mark is cleared by mntput())
1550 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1552 if (flags
& MNT_EXPIRE
) {
1553 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1554 flags
& (MNT_FORCE
| MNT_DETACH
))
1558 * probably don't strictly need the lock here if we examined
1559 * all race cases, but it's a slowpath.
1562 if (mnt_get_count(mnt
) != 2) {
1563 unlock_mount_hash();
1566 unlock_mount_hash();
1568 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1573 * If we may have to abort operations to get out of this
1574 * mount, and they will themselves hold resources we must
1575 * allow the fs to do things. In the Unix tradition of
1576 * 'Gee thats tricky lets do it in userspace' the umount_begin
1577 * might fail to complete on the first run through as other tasks
1578 * must return, and the like. Thats for the mount program to worry
1579 * about for the moment.
1582 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1583 sb
->s_op
->umount_begin(sb
);
1587 * No sense to grab the lock for this test, but test itself looks
1588 * somewhat bogus. Suggestions for better replacement?
1589 * Ho-hum... In principle, we might treat that as umount + switch
1590 * to rootfs. GC would eventually take care of the old vfsmount.
1591 * Actually it makes sense, especially if rootfs would contain a
1592 * /reboot - static binary that would close all descriptors and
1593 * call reboot(9). Then init(8) could umount root and exec /reboot.
1595 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1597 * Special case for "unmounting" root ...
1598 * we just try to remount it readonly.
1600 if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
1602 down_write(&sb
->s_umount
);
1604 retval
= do_remount_sb(sb
, SB_RDONLY
, NULL
, 0);
1605 up_write(&sb
->s_umount
);
1613 if (flags
& MNT_DETACH
) {
1614 if (!list_empty(&mnt
->mnt_list
))
1615 umount_tree(mnt
, UMOUNT_PROPAGATE
);
1618 shrink_submounts(mnt
);
1620 if (!propagate_mount_busy(mnt
, 2)) {
1621 if (!list_empty(&mnt
->mnt_list
))
1622 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
1626 unlock_mount_hash();
1632 * __detach_mounts - lazily unmount all mounts on the specified dentry
1634 * During unlink, rmdir, and d_drop it is possible to loose the path
1635 * to an existing mountpoint, and wind up leaking the mount.
1636 * detach_mounts allows lazily unmounting those mounts instead of
1639 * The caller may hold dentry->d_inode->i_mutex.
1641 void __detach_mounts(struct dentry
*dentry
)
1643 struct mountpoint
*mp
;
1648 mp
= lookup_mountpoint(dentry
);
1649 if (IS_ERR_OR_NULL(mp
))
1653 while (!hlist_empty(&mp
->m_list
)) {
1654 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1655 if (mnt
->mnt
.mnt_flags
& MNT_UMOUNT
) {
1656 hlist_add_head(&mnt
->mnt_umount
.s_list
, &unmounted
);
1659 else umount_tree(mnt
, UMOUNT_CONNECTED
);
1663 unlock_mount_hash();
1668 * Is the caller allowed to modify his namespace?
1670 static inline bool may_mount(void)
1672 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1675 static inline bool may_mandlock(void)
1677 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1680 return capable(CAP_SYS_ADMIN
);
1684 * Now umount can handle mount points as well as block devices.
1685 * This is important for filesystems which use unnamed block devices.
1687 * We now support a flag for forced unmount like the other 'big iron'
1688 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1691 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1696 int lookup_flags
= 0;
1698 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1704 if (!(flags
& UMOUNT_NOFOLLOW
))
1705 lookup_flags
|= LOOKUP_FOLLOW
;
1707 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1710 mnt
= real_mount(path
.mnt
);
1712 if (path
.dentry
!= path
.mnt
->mnt_root
)
1714 if (!check_mnt(mnt
))
1716 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1719 if (flags
& MNT_FORCE
&& !capable(CAP_SYS_ADMIN
))
1722 retval
= do_umount(mnt
, flags
);
1724 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1726 mntput_no_expire(mnt
);
1731 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1734 * The 2.0 compatible umount. No flags.
1736 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1738 return sys_umount(name
, 0);
1743 static bool is_mnt_ns_file(struct dentry
*dentry
)
1745 /* Is this a proxy for a mount namespace? */
1746 return dentry
->d_op
== &ns_dentry_operations
&&
1747 dentry
->d_fsdata
== &mntns_operations
;
1750 struct mnt_namespace
*to_mnt_ns(struct ns_common
*ns
)
1752 return container_of(ns
, struct mnt_namespace
, ns
);
1755 static bool mnt_ns_loop(struct dentry
*dentry
)
1757 /* Could bind mounting the mount namespace inode cause a
1758 * mount namespace loop?
1760 struct mnt_namespace
*mnt_ns
;
1761 if (!is_mnt_ns_file(dentry
))
1764 mnt_ns
= to_mnt_ns(get_proc_ns(dentry
->d_inode
));
1765 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1768 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1771 struct mount
*res
, *p
, *q
, *r
, *parent
;
1773 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1774 return ERR_PTR(-EINVAL
);
1776 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1777 return ERR_PTR(-EINVAL
);
1779 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1783 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1786 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1788 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1791 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1792 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1793 IS_MNT_UNBINDABLE(s
)) {
1794 s
= skip_mnt_tree(s
);
1797 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1798 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1799 s
= skip_mnt_tree(s
);
1802 while (p
!= s
->mnt_parent
) {
1808 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1812 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1813 attach_mnt(q
, parent
, p
->mnt_mp
);
1814 unlock_mount_hash();
1821 umount_tree(res
, UMOUNT_SYNC
);
1822 unlock_mount_hash();
1827 /* Caller should check returned pointer for errors */
1829 struct vfsmount
*collect_mounts(const struct path
*path
)
1833 if (!check_mnt(real_mount(path
->mnt
)))
1834 tree
= ERR_PTR(-EINVAL
);
1836 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1837 CL_COPY_ALL
| CL_PRIVATE
);
1840 return ERR_CAST(tree
);
1844 void drop_collected_mounts(struct vfsmount
*mnt
)
1848 umount_tree(real_mount(mnt
), UMOUNT_SYNC
);
1849 unlock_mount_hash();
1854 * clone_private_mount - create a private clone of a path
1856 * This creates a new vfsmount, which will be the clone of @path. The new will
1857 * not be attached anywhere in the namespace and will be private (i.e. changes
1858 * to the originating mount won't be propagated into this).
1860 * Release with mntput().
1862 struct vfsmount
*clone_private_mount(const struct path
*path
)
1864 struct mount
*old_mnt
= real_mount(path
->mnt
);
1865 struct mount
*new_mnt
;
1867 if (IS_MNT_UNBINDABLE(old_mnt
))
1868 return ERR_PTR(-EINVAL
);
1870 new_mnt
= clone_mnt(old_mnt
, path
->dentry
, CL_PRIVATE
);
1871 if (IS_ERR(new_mnt
))
1872 return ERR_CAST(new_mnt
);
1874 return &new_mnt
->mnt
;
1876 EXPORT_SYMBOL_GPL(clone_private_mount
);
1878 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1879 struct vfsmount
*root
)
1882 int res
= f(root
, arg
);
1885 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1886 res
= f(&mnt
->mnt
, arg
);
1892 EXPORT_SYMBOL_GPL(iterate_mounts
);
1894 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1898 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1899 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1900 mnt_release_group_id(p
);
1904 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1908 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1909 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1910 int err
= mnt_alloc_group_id(p
);
1912 cleanup_group_ids(mnt
, p
);
1921 int count_mounts(struct mnt_namespace
*ns
, struct mount
*mnt
)
1923 unsigned int max
= READ_ONCE(sysctl_mount_max
);
1924 unsigned int mounts
= 0, old
, pending
, sum
;
1927 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1931 pending
= ns
->pending_mounts
;
1932 sum
= old
+ pending
;
1936 (mounts
> (max
- sum
)))
1939 ns
->pending_mounts
= pending
+ mounts
;
1944 * @source_mnt : mount tree to be attached
1945 * @nd : place the mount tree @source_mnt is attached
1946 * @parent_nd : if non-null, detach the source_mnt from its parent and
1947 * store the parent mount and mountpoint dentry.
1948 * (done when source_mnt is moved)
1950 * NOTE: in the table below explains the semantics when a source mount
1951 * of a given type is attached to a destination mount of a given type.
1952 * ---------------------------------------------------------------------------
1953 * | BIND MOUNT OPERATION |
1954 * |**************************************************************************
1955 * | source-->| shared | private | slave | unbindable |
1959 * |**************************************************************************
1960 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1962 * |non-shared| shared (+) | private | slave (*) | invalid |
1963 * ***************************************************************************
1964 * A bind operation clones the source mount and mounts the clone on the
1965 * destination mount.
1967 * (++) the cloned mount is propagated to all the mounts in the propagation
1968 * tree of the destination mount and the cloned mount is added to
1969 * the peer group of the source mount.
1970 * (+) the cloned mount is created under the destination mount and is marked
1971 * as shared. The cloned mount is added to the peer group of the source
1973 * (+++) the mount is propagated to all the mounts in the propagation tree
1974 * of the destination mount and the cloned mount is made slave
1975 * of the same master as that of the source mount. The cloned mount
1976 * is marked as 'shared and slave'.
1977 * (*) the cloned mount is made a slave of the same master as that of the
1980 * ---------------------------------------------------------------------------
1981 * | MOVE MOUNT OPERATION |
1982 * |**************************************************************************
1983 * | source-->| shared | private | slave | unbindable |
1987 * |**************************************************************************
1988 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1990 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1991 * ***************************************************************************
1993 * (+) the mount is moved to the destination. And is then propagated to
1994 * all the mounts in the propagation tree of the destination mount.
1995 * (+*) the mount is moved to the destination.
1996 * (+++) the mount is moved to the destination and is then propagated to
1997 * all the mounts belonging to the destination mount's propagation tree.
1998 * the mount is marked as 'shared and slave'.
1999 * (*) the mount continues to be a slave at the new location.
2001 * if the source mount is a tree, the operations explained above is
2002 * applied to each mount in the tree.
2003 * Must be called without spinlocks held, since this function can sleep
2006 static int attach_recursive_mnt(struct mount
*source_mnt
,
2007 struct mount
*dest_mnt
,
2008 struct mountpoint
*dest_mp
,
2009 struct path
*parent_path
)
2011 HLIST_HEAD(tree_list
);
2012 struct mnt_namespace
*ns
= dest_mnt
->mnt_ns
;
2013 struct mountpoint
*smp
;
2014 struct mount
*child
, *p
;
2015 struct hlist_node
*n
;
2018 /* Preallocate a mountpoint in case the new mounts need
2019 * to be tucked under other mounts.
2021 smp
= get_mountpoint(source_mnt
->mnt
.mnt_root
);
2023 return PTR_ERR(smp
);
2025 /* Is there space to add these mounts to the mount namespace? */
2027 err
= count_mounts(ns
, source_mnt
);
2032 if (IS_MNT_SHARED(dest_mnt
)) {
2033 err
= invent_group_ids(source_mnt
, true);
2036 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
2039 goto out_cleanup_ids
;
2040 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
2046 detach_mnt(source_mnt
, parent_path
);
2047 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
2048 touch_mnt_namespace(source_mnt
->mnt_ns
);
2050 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
2051 commit_tree(source_mnt
);
2054 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
2056 hlist_del_init(&child
->mnt_hash
);
2057 q
= __lookup_mnt(&child
->mnt_parent
->mnt
,
2058 child
->mnt_mountpoint
);
2060 mnt_change_mountpoint(child
, smp
, q
);
2063 put_mountpoint(smp
);
2064 unlock_mount_hash();
2069 while (!hlist_empty(&tree_list
)) {
2070 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
2071 child
->mnt_parent
->mnt_ns
->pending_mounts
= 0;
2072 umount_tree(child
, UMOUNT_SYNC
);
2074 unlock_mount_hash();
2075 cleanup_group_ids(source_mnt
, NULL
);
2077 ns
->pending_mounts
= 0;
2079 read_seqlock_excl(&mount_lock
);
2080 put_mountpoint(smp
);
2081 read_sequnlock_excl(&mount_lock
);
2086 static struct mountpoint
*lock_mount(struct path
*path
)
2088 struct vfsmount
*mnt
;
2089 struct dentry
*dentry
= path
->dentry
;
2091 inode_lock(dentry
->d_inode
);
2092 if (unlikely(cant_mount(dentry
))) {
2093 inode_unlock(dentry
->d_inode
);
2094 return ERR_PTR(-ENOENT
);
2097 mnt
= lookup_mnt(path
);
2099 struct mountpoint
*mp
= get_mountpoint(dentry
);
2102 inode_unlock(dentry
->d_inode
);
2108 inode_unlock(path
->dentry
->d_inode
);
2111 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
2115 static void unlock_mount(struct mountpoint
*where
)
2117 struct dentry
*dentry
= where
->m_dentry
;
2119 read_seqlock_excl(&mount_lock
);
2120 put_mountpoint(where
);
2121 read_sequnlock_excl(&mount_lock
);
2124 inode_unlock(dentry
->d_inode
);
2127 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
2129 if (mnt
->mnt
.mnt_sb
->s_flags
& SB_NOUSER
)
2132 if (d_is_dir(mp
->m_dentry
) !=
2133 d_is_dir(mnt
->mnt
.mnt_root
))
2136 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
2140 * Sanity check the flags to change_mnt_propagation.
2143 static int flags_to_propagation_type(int ms_flags
)
2145 int type
= ms_flags
& ~(MS_REC
| MS_SILENT
);
2147 /* Fail if any non-propagation flags are set */
2148 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2150 /* Only one propagation flag should be set */
2151 if (!is_power_of_2(type
))
2157 * recursively change the type of the mountpoint.
2159 static int do_change_type(struct path
*path
, int ms_flags
)
2162 struct mount
*mnt
= real_mount(path
->mnt
);
2163 int recurse
= ms_flags
& MS_REC
;
2167 if (path
->dentry
!= path
->mnt
->mnt_root
)
2170 type
= flags_to_propagation_type(ms_flags
);
2175 if (type
== MS_SHARED
) {
2176 err
= invent_group_ids(mnt
, recurse
);
2182 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
2183 change_mnt_propagation(m
, type
);
2184 unlock_mount_hash();
2191 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
2193 struct mount
*child
;
2194 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
2195 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
2198 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
2205 * do loopback mount.
2207 static int do_loopback(struct path
*path
, const char *old_name
,
2210 struct path old_path
;
2211 struct mount
*mnt
= NULL
, *old
, *parent
;
2212 struct mountpoint
*mp
;
2214 if (!old_name
|| !*old_name
)
2216 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
2221 if (mnt_ns_loop(old_path
.dentry
))
2224 mp
= lock_mount(path
);
2229 old
= real_mount(old_path
.mnt
);
2230 parent
= real_mount(path
->mnt
);
2233 if (IS_MNT_UNBINDABLE(old
))
2236 if (!check_mnt(parent
))
2239 if (!check_mnt(old
) && old_path
.dentry
->d_op
!= &ns_dentry_operations
)
2242 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
2246 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
2248 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
2255 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2257 err
= graft_tree(mnt
, parent
, mp
);
2260 umount_tree(mnt
, UMOUNT_SYNC
);
2261 unlock_mount_hash();
2266 path_put(&old_path
);
2270 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
2273 int readonly_request
= 0;
2275 if (ms_flags
& MS_RDONLY
)
2276 readonly_request
= 1;
2277 if (readonly_request
== __mnt_is_readonly(mnt
))
2280 if (readonly_request
)
2281 error
= mnt_make_readonly(real_mount(mnt
));
2283 __mnt_unmake_readonly(real_mount(mnt
));
2288 * change filesystem flags. dir should be a physical root of filesystem.
2289 * If you've mounted a non-root directory somewhere and want to do remount
2290 * on it - tough luck.
2292 static int do_remount(struct path
*path
, int ms_flags
, int sb_flags
,
2293 int mnt_flags
, void *data
)
2296 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2297 struct mount
*mnt
= real_mount(path
->mnt
);
2299 if (!check_mnt(mnt
))
2302 if (path
->dentry
!= path
->mnt
->mnt_root
)
2305 /* Don't allow changing of locked mnt flags.
2307 * No locks need to be held here while testing the various
2308 * MNT_LOCK flags because those flags can never be cleared
2309 * once they are set.
2311 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_READONLY
) &&
2312 !(mnt_flags
& MNT_READONLY
)) {
2315 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NODEV
) &&
2316 !(mnt_flags
& MNT_NODEV
)) {
2319 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOSUID
) &&
2320 !(mnt_flags
& MNT_NOSUID
)) {
2323 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOEXEC
) &&
2324 !(mnt_flags
& MNT_NOEXEC
)) {
2327 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_ATIME
) &&
2328 ((mnt
->mnt
.mnt_flags
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
))) {
2332 err
= security_sb_remount(sb
, data
);
2336 down_write(&sb
->s_umount
);
2337 if (ms_flags
& MS_BIND
)
2338 err
= change_mount_flags(path
->mnt
, ms_flags
);
2339 else if (!ns_capable(sb
->s_user_ns
, CAP_SYS_ADMIN
))
2342 err
= do_remount_sb(sb
, sb_flags
, data
, 0);
2345 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2346 mnt
->mnt
.mnt_flags
= mnt_flags
;
2347 touch_mnt_namespace(mnt
->mnt_ns
);
2348 unlock_mount_hash();
2350 up_write(&sb
->s_umount
);
2354 static inline int tree_contains_unbindable(struct mount
*mnt
)
2357 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2358 if (IS_MNT_UNBINDABLE(p
))
2364 static int do_move_mount(struct path
*path
, const char *old_name
)
2366 struct path old_path
, parent_path
;
2369 struct mountpoint
*mp
;
2371 if (!old_name
|| !*old_name
)
2373 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2377 mp
= lock_mount(path
);
2382 old
= real_mount(old_path
.mnt
);
2383 p
= real_mount(path
->mnt
);
2386 if (!check_mnt(p
) || !check_mnt(old
))
2389 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2393 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
2396 if (!mnt_has_parent(old
))
2399 if (d_is_dir(path
->dentry
) !=
2400 d_is_dir(old_path
.dentry
))
2403 * Don't move a mount residing in a shared parent.
2405 if (IS_MNT_SHARED(old
->mnt_parent
))
2408 * Don't move a mount tree containing unbindable mounts to a destination
2409 * mount which is shared.
2411 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2414 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2418 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2422 /* if the mount is moved, it should no longer be expire
2424 list_del_init(&old
->mnt_expire
);
2429 path_put(&parent_path
);
2430 path_put(&old_path
);
2434 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2437 const char *subtype
= strchr(fstype
, '.');
2446 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2448 if (!mnt
->mnt_sb
->s_subtype
)
2454 return ERR_PTR(err
);
2458 * add a mount into a namespace's mount tree
2460 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2462 struct mountpoint
*mp
;
2463 struct mount
*parent
;
2466 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2468 mp
= lock_mount(path
);
2472 parent
= real_mount(path
->mnt
);
2474 if (unlikely(!check_mnt(parent
))) {
2475 /* that's acceptable only for automounts done in private ns */
2476 if (!(mnt_flags
& MNT_SHRINKABLE
))
2478 /* ... and for those we'd better have mountpoint still alive */
2479 if (!parent
->mnt_ns
)
2483 /* Refuse the same filesystem on the same mount point */
2485 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2486 path
->mnt
->mnt_root
== path
->dentry
)
2490 if (d_is_symlink(newmnt
->mnt
.mnt_root
))
2493 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2494 err
= graft_tree(newmnt
, parent
, mp
);
2501 static bool mount_too_revealing(struct vfsmount
*mnt
, int *new_mnt_flags
);
2504 * create a new mount for userspace and request it to be added into the
2507 static int do_new_mount(struct path
*path
, const char *fstype
, int sb_flags
,
2508 int mnt_flags
, const char *name
, void *data
)
2510 struct file_system_type
*type
;
2511 struct vfsmount
*mnt
;
2517 type
= get_fs_type(fstype
);
2521 mnt
= vfs_kern_mount(type
, sb_flags
, name
, data
);
2522 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2523 !mnt
->mnt_sb
->s_subtype
)
2524 mnt
= fs_set_subtype(mnt
, fstype
);
2526 put_filesystem(type
);
2528 return PTR_ERR(mnt
);
2530 if (mount_too_revealing(mnt
, &mnt_flags
)) {
2535 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2541 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2543 struct mount
*mnt
= real_mount(m
);
2545 /* The new mount record should have at least 2 refs to prevent it being
2546 * expired before we get a chance to add it
2548 BUG_ON(mnt_get_count(mnt
) < 2);
2550 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2551 m
->mnt_root
== path
->dentry
) {
2556 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2560 /* remove m from any expiration list it may be on */
2561 if (!list_empty(&mnt
->mnt_expire
)) {
2563 list_del_init(&mnt
->mnt_expire
);
2572 * mnt_set_expiry - Put a mount on an expiration list
2573 * @mnt: The mount to list.
2574 * @expiry_list: The list to add the mount to.
2576 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2580 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2584 EXPORT_SYMBOL(mnt_set_expiry
);
2587 * process a list of expirable mountpoints with the intent of discarding any
2588 * mountpoints that aren't in use and haven't been touched since last we came
2591 void mark_mounts_for_expiry(struct list_head
*mounts
)
2593 struct mount
*mnt
, *next
;
2594 LIST_HEAD(graveyard
);
2596 if (list_empty(mounts
))
2602 /* extract from the expiration list every vfsmount that matches the
2603 * following criteria:
2604 * - only referenced by its parent vfsmount
2605 * - still marked for expiry (marked on the last call here; marks are
2606 * cleared by mntput())
2608 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2609 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2610 propagate_mount_busy(mnt
, 1))
2612 list_move(&mnt
->mnt_expire
, &graveyard
);
2614 while (!list_empty(&graveyard
)) {
2615 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2616 touch_mnt_namespace(mnt
->mnt_ns
);
2617 umount_tree(mnt
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2619 unlock_mount_hash();
2623 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2626 * Ripoff of 'select_parent()'
2628 * search the list of submounts for a given mountpoint, and move any
2629 * shrinkable submounts to the 'graveyard' list.
2631 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2633 struct mount
*this_parent
= parent
;
2634 struct list_head
*next
;
2638 next
= this_parent
->mnt_mounts
.next
;
2640 while (next
!= &this_parent
->mnt_mounts
) {
2641 struct list_head
*tmp
= next
;
2642 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2645 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2648 * Descend a level if the d_mounts list is non-empty.
2650 if (!list_empty(&mnt
->mnt_mounts
)) {
2655 if (!propagate_mount_busy(mnt
, 1)) {
2656 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2661 * All done at this level ... ascend and resume the search
2663 if (this_parent
!= parent
) {
2664 next
= this_parent
->mnt_child
.next
;
2665 this_parent
= this_parent
->mnt_parent
;
2672 * process a list of expirable mountpoints with the intent of discarding any
2673 * submounts of a specific parent mountpoint
2675 * mount_lock must be held for write
2677 static void shrink_submounts(struct mount
*mnt
)
2679 LIST_HEAD(graveyard
);
2682 /* extract submounts of 'mountpoint' from the expiration list */
2683 while (select_submounts(mnt
, &graveyard
)) {
2684 while (!list_empty(&graveyard
)) {
2685 m
= list_first_entry(&graveyard
, struct mount
,
2687 touch_mnt_namespace(m
->mnt_ns
);
2688 umount_tree(m
, UMOUNT_PROPAGATE
|UMOUNT_SYNC
);
2694 * Some copy_from_user() implementations do not return the exact number of
2695 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2696 * Note that this function differs from copy_from_user() in that it will oops
2697 * on bad values of `to', rather than returning a short copy.
2699 static long exact_copy_from_user(void *to
, const void __user
* from
,
2703 const char __user
*f
= from
;
2706 if (!access_ok(VERIFY_READ
, from
, n
))
2710 if (__get_user(c
, f
)) {
2721 void *copy_mount_options(const void __user
* data
)
2730 copy
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
2732 return ERR_PTR(-ENOMEM
);
2734 /* We only care that *some* data at the address the user
2735 * gave us is valid. Just in case, we'll zero
2736 * the remainder of the page.
2738 /* copy_from_user cannot cross TASK_SIZE ! */
2739 size
= TASK_SIZE
- (unsigned long)data
;
2740 if (size
> PAGE_SIZE
)
2743 i
= size
- exact_copy_from_user(copy
, data
, size
);
2746 return ERR_PTR(-EFAULT
);
2749 memset(copy
+ i
, 0, PAGE_SIZE
- i
);
2753 char *copy_mount_string(const void __user
*data
)
2755 return data
? strndup_user(data
, PAGE_SIZE
) : NULL
;
2759 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2760 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2762 * data is a (void *) that can point to any structure up to
2763 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2764 * information (or be NULL).
2766 * Pre-0.97 versions of mount() didn't have a flags word.
2767 * When the flags word was introduced its top half was required
2768 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2769 * Therefore, if this magic number is present, it carries no information
2770 * and must be discarded.
2772 long do_mount(const char *dev_name
, const char __user
*dir_name
,
2773 const char *type_page
, unsigned long flags
, void *data_page
)
2776 unsigned int mnt_flags
= 0, sb_flags
;
2780 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2781 flags
&= ~MS_MGC_MSK
;
2783 /* Basic sanity checks */
2785 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2787 if (flags
& MS_NOUSER
)
2790 /* ... and get the mountpoint */
2791 retval
= user_path(dir_name
, &path
);
2795 retval
= security_sb_mount(dev_name
, &path
,
2796 type_page
, flags
, data_page
);
2797 if (!retval
&& !may_mount())
2799 if (!retval
&& (flags
& SB_MANDLOCK
) && !may_mandlock())
2804 /* Default to relatime unless overriden */
2805 if (!(flags
& MS_NOATIME
))
2806 mnt_flags
|= MNT_RELATIME
;
2808 /* Separate the per-mountpoint flags */
2809 if (flags
& MS_NOSUID
)
2810 mnt_flags
|= MNT_NOSUID
;
2811 if (flags
& MS_NODEV
)
2812 mnt_flags
|= MNT_NODEV
;
2813 if (flags
& MS_NOEXEC
)
2814 mnt_flags
|= MNT_NOEXEC
;
2815 if (flags
& MS_NOATIME
)
2816 mnt_flags
|= MNT_NOATIME
;
2817 if (flags
& MS_NODIRATIME
)
2818 mnt_flags
|= MNT_NODIRATIME
;
2819 if (flags
& MS_STRICTATIME
)
2820 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2821 if (flags
& SB_RDONLY
)
2822 mnt_flags
|= MNT_READONLY
;
2824 /* The default atime for remount is preservation */
2825 if ((flags
& MS_REMOUNT
) &&
2826 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
2827 MS_STRICTATIME
)) == 0)) {
2828 mnt_flags
&= ~MNT_ATIME_MASK
;
2829 mnt_flags
|= path
.mnt
->mnt_flags
& MNT_ATIME_MASK
;
2832 sb_flags
= flags
& (SB_RDONLY
|
2841 if (flags
& MS_REMOUNT
)
2842 retval
= do_remount(&path
, flags
, sb_flags
, mnt_flags
,
2844 else if (flags
& MS_BIND
)
2845 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2846 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2847 retval
= do_change_type(&path
, flags
);
2848 else if (flags
& MS_MOVE
)
2849 retval
= do_move_mount(&path
, dev_name
);
2851 retval
= do_new_mount(&path
, type_page
, sb_flags
, mnt_flags
,
2852 dev_name
, data_page
);
2858 static struct ucounts
*inc_mnt_namespaces(struct user_namespace
*ns
)
2860 return inc_ucount(ns
, current_euid(), UCOUNT_MNT_NAMESPACES
);
2863 static void dec_mnt_namespaces(struct ucounts
*ucounts
)
2865 dec_ucount(ucounts
, UCOUNT_MNT_NAMESPACES
);
2868 static void free_mnt_ns(struct mnt_namespace
*ns
)
2870 ns_free_inum(&ns
->ns
);
2871 dec_mnt_namespaces(ns
->ucounts
);
2872 put_user_ns(ns
->user_ns
);
2877 * Assign a sequence number so we can detect when we attempt to bind
2878 * mount a reference to an older mount namespace into the current
2879 * mount namespace, preventing reference counting loops. A 64bit
2880 * number incrementing at 10Ghz will take 12,427 years to wrap which
2881 * is effectively never, so we can ignore the possibility.
2883 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2885 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2887 struct mnt_namespace
*new_ns
;
2888 struct ucounts
*ucounts
;
2891 ucounts
= inc_mnt_namespaces(user_ns
);
2893 return ERR_PTR(-ENOSPC
);
2895 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2897 dec_mnt_namespaces(ucounts
);
2898 return ERR_PTR(-ENOMEM
);
2900 ret
= ns_alloc_inum(&new_ns
->ns
);
2903 dec_mnt_namespaces(ucounts
);
2904 return ERR_PTR(ret
);
2906 new_ns
->ns
.ops
= &mntns_operations
;
2907 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2908 atomic_set(&new_ns
->count
, 1);
2909 new_ns
->root
= NULL
;
2910 INIT_LIST_HEAD(&new_ns
->list
);
2911 init_waitqueue_head(&new_ns
->poll
);
2913 new_ns
->user_ns
= get_user_ns(user_ns
);
2914 new_ns
->ucounts
= ucounts
;
2916 new_ns
->pending_mounts
= 0;
2921 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2922 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2924 struct mnt_namespace
*new_ns
;
2925 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2926 struct mount
*p
, *q
;
2933 if (likely(!(flags
& CLONE_NEWNS
))) {
2940 new_ns
= alloc_mnt_ns(user_ns
);
2945 /* First pass: copy the tree topology */
2946 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2947 if (user_ns
!= ns
->user_ns
)
2948 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2949 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2952 free_mnt_ns(new_ns
);
2953 return ERR_CAST(new);
2956 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2959 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2960 * as belonging to new namespace. We have already acquired a private
2961 * fs_struct, so tsk->fs->lock is not needed.
2969 if (&p
->mnt
== new_fs
->root
.mnt
) {
2970 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2973 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2974 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2978 p
= next_mnt(p
, old
);
2979 q
= next_mnt(q
, new);
2982 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2983 p
= next_mnt(p
, old
);
2996 * create_mnt_ns - creates a private namespace and adds a root filesystem
2997 * @mnt: pointer to the new root filesystem mountpoint
2999 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
3001 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
3002 if (!IS_ERR(new_ns
)) {
3003 struct mount
*mnt
= real_mount(m
);
3004 mnt
->mnt_ns
= new_ns
;
3007 list_add(&mnt
->mnt_list
, &new_ns
->list
);
3014 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
3016 struct mnt_namespace
*ns
;
3017 struct super_block
*s
;
3021 ns
= create_mnt_ns(mnt
);
3023 return ERR_CAST(ns
);
3025 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
3026 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
3031 return ERR_PTR(err
);
3033 /* trade a vfsmount reference for active sb one */
3034 s
= path
.mnt
->mnt_sb
;
3035 atomic_inc(&s
->s_active
);
3037 /* lock the sucker */
3038 down_write(&s
->s_umount
);
3039 /* ... and return the root of (sub)tree on it */
3042 EXPORT_SYMBOL(mount_subtree
);
3044 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
3045 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
3052 kernel_type
= copy_mount_string(type
);
3053 ret
= PTR_ERR(kernel_type
);
3054 if (IS_ERR(kernel_type
))
3057 kernel_dev
= copy_mount_string(dev_name
);
3058 ret
= PTR_ERR(kernel_dev
);
3059 if (IS_ERR(kernel_dev
))
3062 options
= copy_mount_options(data
);
3063 ret
= PTR_ERR(options
);
3064 if (IS_ERR(options
))
3067 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
, options
);
3079 * Return true if path is reachable from root
3081 * namespace_sem or mount_lock is held
3083 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
3084 const struct path
*root
)
3086 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
3087 dentry
= mnt
->mnt_mountpoint
;
3088 mnt
= mnt
->mnt_parent
;
3090 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
3093 bool path_is_under(const struct path
*path1
, const struct path
*path2
)
3096 read_seqlock_excl(&mount_lock
);
3097 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
3098 read_sequnlock_excl(&mount_lock
);
3101 EXPORT_SYMBOL(path_is_under
);
3104 * pivot_root Semantics:
3105 * Moves the root file system of the current process to the directory put_old,
3106 * makes new_root as the new root file system of the current process, and sets
3107 * root/cwd of all processes which had them on the current root to new_root.
3110 * The new_root and put_old must be directories, and must not be on the
3111 * same file system as the current process root. The put_old must be
3112 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3113 * pointed to by put_old must yield the same directory as new_root. No other
3114 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3116 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3117 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3118 * in this situation.
3121 * - we don't move root/cwd if they are not at the root (reason: if something
3122 * cared enough to change them, it's probably wrong to force them elsewhere)
3123 * - it's okay to pick a root that isn't the root of a file system, e.g.
3124 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3125 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3128 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
3129 const char __user
*, put_old
)
3131 struct path
new, old
, parent_path
, root_parent
, root
;
3132 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
3133 struct mountpoint
*old_mp
, *root_mp
;
3139 error
= user_path_dir(new_root
, &new);
3143 error
= user_path_dir(put_old
, &old
);
3147 error
= security_sb_pivotroot(&old
, &new);
3151 get_fs_root(current
->fs
, &root
);
3152 old_mp
= lock_mount(&old
);
3153 error
= PTR_ERR(old_mp
);
3158 new_mnt
= real_mount(new.mnt
);
3159 root_mnt
= real_mount(root
.mnt
);
3160 old_mnt
= real_mount(old
.mnt
);
3161 if (IS_MNT_SHARED(old_mnt
) ||
3162 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
3163 IS_MNT_SHARED(root_mnt
->mnt_parent
))
3165 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
3167 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
3170 if (d_unlinked(new.dentry
))
3173 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
3174 goto out4
; /* loop, on the same file system */
3176 if (root
.mnt
->mnt_root
!= root
.dentry
)
3177 goto out4
; /* not a mountpoint */
3178 if (!mnt_has_parent(root_mnt
))
3179 goto out4
; /* not attached */
3180 root_mp
= root_mnt
->mnt_mp
;
3181 if (new.mnt
->mnt_root
!= new.dentry
)
3182 goto out4
; /* not a mountpoint */
3183 if (!mnt_has_parent(new_mnt
))
3184 goto out4
; /* not attached */
3185 /* make sure we can reach put_old from new_root */
3186 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
3188 /* make certain new is below the root */
3189 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
3191 root_mp
->m_count
++; /* pin it so it won't go away */
3193 detach_mnt(new_mnt
, &parent_path
);
3194 detach_mnt(root_mnt
, &root_parent
);
3195 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
3196 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
3197 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
3199 /* mount old root on put_old */
3200 attach_mnt(root_mnt
, old_mnt
, old_mp
);
3201 /* mount new_root on / */
3202 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
3203 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
3204 /* A moved mount should not expire automatically */
3205 list_del_init(&new_mnt
->mnt_expire
);
3206 put_mountpoint(root_mp
);
3207 unlock_mount_hash();
3208 chroot_fs_refs(&root
, &new);
3211 unlock_mount(old_mp
);
3213 path_put(&root_parent
);
3214 path_put(&parent_path
);
3226 static void __init
init_mount_tree(void)
3228 struct vfsmount
*mnt
;
3229 struct mnt_namespace
*ns
;
3231 struct file_system_type
*type
;
3233 type
= get_fs_type("rootfs");
3235 panic("Can't find rootfs type");
3236 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
3237 put_filesystem(type
);
3239 panic("Can't create rootfs");
3241 ns
= create_mnt_ns(mnt
);
3243 panic("Can't allocate initial namespace");
3245 init_task
.nsproxy
->mnt_ns
= ns
;
3249 root
.dentry
= mnt
->mnt_root
;
3250 mnt
->mnt_flags
|= MNT_LOCKED
;
3252 set_fs_pwd(current
->fs
, &root
);
3253 set_fs_root(current
->fs
, &root
);
3256 void __init
mnt_init(void)
3260 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
3261 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
3263 mount_hashtable
= alloc_large_system_hash("Mount-cache",
3264 sizeof(struct hlist_head
),
3267 &m_hash_shift
, &m_hash_mask
, 0, 0);
3268 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
3269 sizeof(struct hlist_head
),
3272 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
3274 if (!mount_hashtable
|| !mountpoint_hashtable
)
3275 panic("Failed to allocate mount hash table\n");
3281 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
3283 fs_kobj
= kobject_create_and_add("fs", NULL
);
3285 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
3290 void put_mnt_ns(struct mnt_namespace
*ns
)
3292 if (!atomic_dec_and_test(&ns
->count
))
3294 drop_collected_mounts(&ns
->root
->mnt
);
3298 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
3300 struct vfsmount
*mnt
;
3301 mnt
= vfs_kern_mount(type
, SB_KERNMOUNT
, type
->name
, data
);
3304 * it is a longterm mount, don't release mnt until
3305 * we unmount before file sys is unregistered
3307 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
3311 EXPORT_SYMBOL_GPL(kern_mount_data
);
3313 void kern_unmount(struct vfsmount
*mnt
)
3315 /* release long term mount so mount point can be released */
3316 if (!IS_ERR_OR_NULL(mnt
)) {
3317 real_mount(mnt
)->mnt_ns
= NULL
;
3318 synchronize_rcu(); /* yecchhh... */
3322 EXPORT_SYMBOL(kern_unmount
);
3324 bool our_mnt(struct vfsmount
*mnt
)
3326 return check_mnt(real_mount(mnt
));
3329 bool current_chrooted(void)
3331 /* Does the current process have a non-standard root */
3332 struct path ns_root
;
3333 struct path fs_root
;
3336 /* Find the namespace root */
3337 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
3338 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
3340 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
3343 get_fs_root(current
->fs
, &fs_root
);
3345 chrooted
= !path_equal(&fs_root
, &ns_root
);
3353 static bool mnt_already_visible(struct mnt_namespace
*ns
, struct vfsmount
*new,
3356 int new_flags
= *new_mnt_flags
;
3358 bool visible
= false;
3360 down_read(&namespace_sem
);
3361 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
3362 struct mount
*child
;
3365 if (mnt
->mnt
.mnt_sb
->s_type
!= new->mnt_sb
->s_type
)
3368 /* This mount is not fully visible if it's root directory
3369 * is not the root directory of the filesystem.
3371 if (mnt
->mnt
.mnt_root
!= mnt
->mnt
.mnt_sb
->s_root
)
3374 /* A local view of the mount flags */
3375 mnt_flags
= mnt
->mnt
.mnt_flags
;
3377 /* Don't miss readonly hidden in the superblock flags */
3378 if (sb_rdonly(mnt
->mnt
.mnt_sb
))
3379 mnt_flags
|= MNT_LOCK_READONLY
;
3381 /* Verify the mount flags are equal to or more permissive
3382 * than the proposed new mount.
3384 if ((mnt_flags
& MNT_LOCK_READONLY
) &&
3385 !(new_flags
& MNT_READONLY
))
3387 if ((mnt_flags
& MNT_LOCK_ATIME
) &&
3388 ((mnt_flags
& MNT_ATIME_MASK
) != (new_flags
& MNT_ATIME_MASK
)))
3391 /* This mount is not fully visible if there are any
3392 * locked child mounts that cover anything except for
3393 * empty directories.
3395 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
3396 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
3397 /* Only worry about locked mounts */
3398 if (!(child
->mnt
.mnt_flags
& MNT_LOCKED
))
3400 /* Is the directory permanetly empty? */
3401 if (!is_empty_dir_inode(inode
))
3404 /* Preserve the locked attributes */
3405 *new_mnt_flags
|= mnt_flags
& (MNT_LOCK_READONLY
| \
3412 up_read(&namespace_sem
);
3416 static bool mount_too_revealing(struct vfsmount
*mnt
, int *new_mnt_flags
)
3418 const unsigned long required_iflags
= SB_I_NOEXEC
| SB_I_NODEV
;
3419 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
3420 unsigned long s_iflags
;
3422 if (ns
->user_ns
== &init_user_ns
)
3425 /* Can this filesystem be too revealing? */
3426 s_iflags
= mnt
->mnt_sb
->s_iflags
;
3427 if (!(s_iflags
& SB_I_USERNS_VISIBLE
))
3430 if ((s_iflags
& required_iflags
) != required_iflags
) {
3431 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3436 return !mnt_already_visible(ns
, mnt
, new_mnt_flags
);
3439 bool mnt_may_suid(struct vfsmount
*mnt
)
3442 * Foreign mounts (accessed via fchdir or through /proc
3443 * symlinks) are always treated as if they are nosuid. This
3444 * prevents namespaces from trusting potentially unsafe
3445 * suid/sgid bits, file caps, or security labels that originate
3446 * in other namespaces.
3448 return !(mnt
->mnt_flags
& MNT_NOSUID
) && check_mnt(real_mount(mnt
)) &&
3449 current_in_userns(mnt
->mnt_sb
->s_user_ns
);
3452 static struct ns_common
*mntns_get(struct task_struct
*task
)
3454 struct ns_common
*ns
= NULL
;
3455 struct nsproxy
*nsproxy
;
3458 nsproxy
= task
->nsproxy
;
3460 ns
= &nsproxy
->mnt_ns
->ns
;
3461 get_mnt_ns(to_mnt_ns(ns
));
3468 static void mntns_put(struct ns_common
*ns
)
3470 put_mnt_ns(to_mnt_ns(ns
));
3473 static int mntns_install(struct nsproxy
*nsproxy
, struct ns_common
*ns
)
3475 struct fs_struct
*fs
= current
->fs
;
3476 struct mnt_namespace
*mnt_ns
= to_mnt_ns(ns
), *old_mnt_ns
;
3480 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3481 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
3482 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
3489 old_mnt_ns
= nsproxy
->mnt_ns
;
3490 nsproxy
->mnt_ns
= mnt_ns
;
3493 err
= vfs_path_lookup(mnt_ns
->root
->mnt
.mnt_root
, &mnt_ns
->root
->mnt
,
3494 "/", LOOKUP_DOWN
, &root
);
3496 /* revert to old namespace */
3497 nsproxy
->mnt_ns
= old_mnt_ns
;
3502 put_mnt_ns(old_mnt_ns
);
3504 /* Update the pwd and root */
3505 set_fs_pwd(fs
, &root
);
3506 set_fs_root(fs
, &root
);
3512 static struct user_namespace
*mntns_owner(struct ns_common
*ns
)
3514 return to_mnt_ns(ns
)->user_ns
;
3517 const struct proc_ns_operations mntns_operations
= {
3519 .type
= CLONE_NEWNS
,
3522 .install
= mntns_install
,
3523 .owner
= mntns_owner
,