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/idr.h>
19 #include <linux/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 static unsigned int m_hash_mask __read_mostly
;
31 static unsigned int m_hash_shift __read_mostly
;
32 static unsigned int mp_hash_mask __read_mostly
;
33 static unsigned int mp_hash_shift __read_mostly
;
35 static __initdata
unsigned long mhash_entries
;
36 static int __init
set_mhash_entries(char *str
)
40 mhash_entries
= simple_strtoul(str
, &str
, 0);
43 __setup("mhash_entries=", set_mhash_entries
);
45 static __initdata
unsigned long mphash_entries
;
46 static int __init
set_mphash_entries(char *str
)
50 mphash_entries
= simple_strtoul(str
, &str
, 0);
53 __setup("mphash_entries=", set_mphash_entries
);
56 static DEFINE_IDA(mnt_id_ida
);
57 static DEFINE_IDA(mnt_group_ida
);
58 static DEFINE_SPINLOCK(mnt_id_lock
);
59 static int mnt_id_start
= 0;
60 static int mnt_group_start
= 1;
62 static struct hlist_head
*mount_hashtable __read_mostly
;
63 static struct hlist_head
*mountpoint_hashtable __read_mostly
;
64 static struct kmem_cache
*mnt_cache __read_mostly
;
65 static DECLARE_RWSEM(namespace_sem
);
68 struct kobject
*fs_kobj
;
69 EXPORT_SYMBOL_GPL(fs_kobj
);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(mount_lock
);
81 static inline struct hlist_head
*m_hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
83 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
84 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
85 tmp
= tmp
+ (tmp
>> m_hash_shift
);
86 return &mount_hashtable
[tmp
& m_hash_mask
];
89 static inline struct hlist_head
*mp_hash(struct dentry
*dentry
)
91 unsigned long tmp
= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
92 tmp
= tmp
+ (tmp
>> mp_hash_shift
);
93 return &mountpoint_hashtable
[tmp
& mp_hash_mask
];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount
*mnt
)
105 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
106 spin_lock(&mnt_id_lock
);
107 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
109 mnt_id_start
= mnt
->mnt_id
+ 1;
110 spin_unlock(&mnt_id_lock
);
117 static void mnt_free_id(struct mount
*mnt
)
119 int id
= mnt
->mnt_id
;
120 spin_lock(&mnt_id_lock
);
121 ida_remove(&mnt_id_ida
, id
);
122 if (mnt_id_start
> id
)
124 spin_unlock(&mnt_id_lock
);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount
*mnt
)
136 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
139 res
= ida_get_new_above(&mnt_group_ida
,
143 mnt_group_start
= mnt
->mnt_group_id
+ 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount
*mnt
)
153 int id
= mnt
->mnt_group_id
;
154 ida_remove(&mnt_group_ida
, id
);
155 if (mnt_group_start
> id
)
156 mnt_group_start
= id
;
157 mnt
->mnt_group_id
= 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount
*mnt
, int n
)
166 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, n
);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount
*mnt
)
180 unsigned int count
= 0;
183 for_each_possible_cpu(cpu
) {
184 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_count
;
189 return mnt
->mnt_count
;
193 static struct mount
*alloc_vfsmnt(const char *name
)
195 struct mount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
199 err
= mnt_alloc_id(mnt
);
204 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
205 if (!mnt
->mnt_devname
)
210 mnt
->mnt_pcp
= alloc_percpu(struct mnt_pcp
);
212 goto out_free_devname
;
214 this_cpu_add(mnt
->mnt_pcp
->mnt_count
, 1);
217 mnt
->mnt_writers
= 0;
220 INIT_HLIST_NODE(&mnt
->mnt_hash
);
221 INIT_LIST_HEAD(&mnt
->mnt_child
);
222 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
223 INIT_LIST_HEAD(&mnt
->mnt_list
);
224 INIT_LIST_HEAD(&mnt
->mnt_expire
);
225 INIT_LIST_HEAD(&mnt
->mnt_share
);
226 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
227 INIT_LIST_HEAD(&mnt
->mnt_slave
);
228 INIT_HLIST_NODE(&mnt
->mnt_mp_list
);
229 #ifdef CONFIG_FSNOTIFY
230 INIT_HLIST_HEAD(&mnt
->mnt_fsnotify_marks
);
237 kfree(mnt
->mnt_devname
);
242 kmem_cache_free(mnt_cache
, mnt
);
247 * Most r/o checks on a fs are for operations that take
248 * discrete amounts of time, like a write() or unlink().
249 * We must keep track of when those operations start
250 * (for permission checks) and when they end, so that
251 * we can determine when writes are able to occur to
255 * __mnt_is_readonly: check whether a mount is read-only
256 * @mnt: the mount to check for its write status
258 * This shouldn't be used directly ouside of the VFS.
259 * It does not guarantee that the filesystem will stay
260 * r/w, just that it is right *now*. This can not and
261 * should not be used in place of IS_RDONLY(inode).
262 * mnt_want/drop_write() will _keep_ the filesystem
265 int __mnt_is_readonly(struct vfsmount
*mnt
)
267 if (mnt
->mnt_flags
& MNT_READONLY
)
269 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
273 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
275 static inline void mnt_inc_writers(struct mount
*mnt
)
278 this_cpu_inc(mnt
->mnt_pcp
->mnt_writers
);
284 static inline void mnt_dec_writers(struct mount
*mnt
)
287 this_cpu_dec(mnt
->mnt_pcp
->mnt_writers
);
293 static unsigned int mnt_get_writers(struct mount
*mnt
)
296 unsigned int count
= 0;
299 for_each_possible_cpu(cpu
) {
300 count
+= per_cpu_ptr(mnt
->mnt_pcp
, cpu
)->mnt_writers
;
305 return mnt
->mnt_writers
;
309 static int mnt_is_readonly(struct vfsmount
*mnt
)
311 if (mnt
->mnt_sb
->s_readonly_remount
)
313 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
315 return __mnt_is_readonly(mnt
);
319 * Most r/o & frozen checks on a fs are for operations that take discrete
320 * amounts of time, like a write() or unlink(). We must keep track of when
321 * those operations start (for permission checks) and when they end, so that we
322 * can determine when writes are able to occur to a filesystem.
325 * __mnt_want_write - get write access to a mount without freeze protection
326 * @m: the mount on which to take a write
328 * This tells the low-level filesystem that a write is about to be performed to
329 * it, and makes sure that writes are allowed (mnt it read-write) before
330 * returning success. This operation does not protect against filesystem being
331 * frozen. When the write operation is finished, __mnt_drop_write() must be
332 * called. This is effectively a refcount.
334 int __mnt_want_write(struct vfsmount
*m
)
336 struct mount
*mnt
= real_mount(m
);
340 mnt_inc_writers(mnt
);
342 * The store to mnt_inc_writers must be visible before we pass
343 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
344 * incremented count after it has set MNT_WRITE_HOLD.
347 while (ACCESS_ONCE(mnt
->mnt
.mnt_flags
) & MNT_WRITE_HOLD
)
350 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
351 * be set to match its requirements. So we must not load that until
352 * MNT_WRITE_HOLD is cleared.
355 if (mnt_is_readonly(m
)) {
356 mnt_dec_writers(mnt
);
365 * mnt_want_write - get write access to a mount
366 * @m: the mount on which to take a write
368 * This tells the low-level filesystem that a write is about to be performed to
369 * it, and makes sure that writes are allowed (mount is read-write, filesystem
370 * is not frozen) before returning success. When the write operation is
371 * finished, mnt_drop_write() must be called. This is effectively a refcount.
373 int mnt_want_write(struct vfsmount
*m
)
377 sb_start_write(m
->mnt_sb
);
378 ret
= __mnt_want_write(m
);
380 sb_end_write(m
->mnt_sb
);
383 EXPORT_SYMBOL_GPL(mnt_want_write
);
386 * mnt_clone_write - get write access to a mount
387 * @mnt: the mount on which to take a write
389 * This is effectively like mnt_want_write, except
390 * it must only be used to take an extra write reference
391 * on a mountpoint that we already know has a write reference
392 * on it. This allows some optimisation.
394 * After finished, mnt_drop_write must be called as usual to
395 * drop the reference.
397 int mnt_clone_write(struct vfsmount
*mnt
)
399 /* superblock may be r/o */
400 if (__mnt_is_readonly(mnt
))
403 mnt_inc_writers(real_mount(mnt
));
407 EXPORT_SYMBOL_GPL(mnt_clone_write
);
410 * __mnt_want_write_file - get write access to a file's mount
411 * @file: the file who's mount on which to take a write
413 * This is like __mnt_want_write, but it takes a file and can
414 * do some optimisations if the file is open for write already
416 int __mnt_want_write_file(struct file
*file
)
418 if (!(file
->f_mode
& FMODE_WRITER
))
419 return __mnt_want_write(file
->f_path
.mnt
);
421 return mnt_clone_write(file
->f_path
.mnt
);
425 * mnt_want_write_file - get write access to a file's mount
426 * @file: the file who's mount on which to take a write
428 * This is like mnt_want_write, but it takes a file and can
429 * do some optimisations if the file is open for write already
431 int mnt_want_write_file(struct file
*file
)
435 sb_start_write(file
->f_path
.mnt
->mnt_sb
);
436 ret
= __mnt_want_write_file(file
);
438 sb_end_write(file
->f_path
.mnt
->mnt_sb
);
441 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
444 * __mnt_drop_write - give up write access to a mount
445 * @mnt: the mount on which to give up write access
447 * Tells the low-level filesystem that we are done
448 * performing writes to it. Must be matched with
449 * __mnt_want_write() call above.
451 void __mnt_drop_write(struct vfsmount
*mnt
)
454 mnt_dec_writers(real_mount(mnt
));
459 * mnt_drop_write - give up write access to a mount
460 * @mnt: the mount on which to give up write access
462 * Tells the low-level filesystem that we are done performing writes to it and
463 * also allows filesystem to be frozen again. Must be matched with
464 * mnt_want_write() call above.
466 void mnt_drop_write(struct vfsmount
*mnt
)
468 __mnt_drop_write(mnt
);
469 sb_end_write(mnt
->mnt_sb
);
471 EXPORT_SYMBOL_GPL(mnt_drop_write
);
473 void __mnt_drop_write_file(struct file
*file
)
475 __mnt_drop_write(file
->f_path
.mnt
);
478 void mnt_drop_write_file(struct file
*file
)
480 mnt_drop_write(file
->f_path
.mnt
);
482 EXPORT_SYMBOL(mnt_drop_write_file
);
484 static int mnt_make_readonly(struct mount
*mnt
)
489 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
491 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
492 * should be visible before we do.
497 * With writers on hold, if this value is zero, then there are
498 * definitely no active writers (although held writers may subsequently
499 * increment the count, they'll have to wait, and decrement it after
500 * seeing MNT_READONLY).
502 * It is OK to have counter incremented on one CPU and decremented on
503 * another: the sum will add up correctly. The danger would be when we
504 * sum up each counter, if we read a counter before it is incremented,
505 * but then read another CPU's count which it has been subsequently
506 * decremented from -- we would see more decrements than we should.
507 * MNT_WRITE_HOLD protects against this scenario, because
508 * mnt_want_write first increments count, then smp_mb, then spins on
509 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
510 * we're counting up here.
512 if (mnt_get_writers(mnt
) > 0)
515 mnt
->mnt
.mnt_flags
|= MNT_READONLY
;
517 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
518 * that become unheld will see MNT_READONLY.
521 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
526 static void __mnt_unmake_readonly(struct mount
*mnt
)
529 mnt
->mnt
.mnt_flags
&= ~MNT_READONLY
;
533 int sb_prepare_remount_readonly(struct super_block
*sb
)
538 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
539 if (atomic_long_read(&sb
->s_remove_count
))
543 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
544 if (!(mnt
->mnt
.mnt_flags
& MNT_READONLY
)) {
545 mnt
->mnt
.mnt_flags
|= MNT_WRITE_HOLD
;
547 if (mnt_get_writers(mnt
) > 0) {
553 if (!err
&& atomic_long_read(&sb
->s_remove_count
))
557 sb
->s_readonly_remount
= 1;
560 list_for_each_entry(mnt
, &sb
->s_mounts
, mnt_instance
) {
561 if (mnt
->mnt
.mnt_flags
& MNT_WRITE_HOLD
)
562 mnt
->mnt
.mnt_flags
&= ~MNT_WRITE_HOLD
;
569 static void free_vfsmnt(struct mount
*mnt
)
571 kfree(mnt
->mnt_devname
);
573 free_percpu(mnt
->mnt_pcp
);
575 kmem_cache_free(mnt_cache
, mnt
);
578 static void delayed_free_vfsmnt(struct rcu_head
*head
)
580 free_vfsmnt(container_of(head
, struct mount
, mnt_rcu
));
583 /* call under rcu_read_lock */
584 bool legitimize_mnt(struct vfsmount
*bastard
, unsigned seq
)
587 if (read_seqretry(&mount_lock
, seq
))
591 mnt
= real_mount(bastard
);
592 mnt_add_count(mnt
, 1);
593 if (likely(!read_seqretry(&mount_lock
, seq
)))
595 if (bastard
->mnt_flags
& MNT_SYNC_UMOUNT
) {
596 mnt_add_count(mnt
, -1);
606 * find the first mount at @dentry on vfsmount @mnt.
607 * call under rcu_read_lock()
609 struct mount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
611 struct hlist_head
*head
= m_hash(mnt
, dentry
);
614 hlist_for_each_entry_rcu(p
, head
, mnt_hash
)
615 if (&p
->mnt_parent
->mnt
== mnt
&& p
->mnt_mountpoint
== dentry
)
621 * find the last mount at @dentry on vfsmount @mnt.
622 * mount_lock must be held.
624 struct mount
*__lookup_mnt_last(struct vfsmount
*mnt
, struct dentry
*dentry
)
626 struct mount
*p
, *res
;
627 res
= p
= __lookup_mnt(mnt
, dentry
);
630 hlist_for_each_entry_continue(p
, mnt_hash
) {
631 if (&p
->mnt_parent
->mnt
!= mnt
|| p
->mnt_mountpoint
!= dentry
)
640 * lookup_mnt - Return the first child mount mounted at path
642 * "First" means first mounted chronologically. If you create the
645 * mount /dev/sda1 /mnt
646 * mount /dev/sda2 /mnt
647 * mount /dev/sda3 /mnt
649 * Then lookup_mnt() on the base /mnt dentry in the root mount will
650 * return successively the root dentry and vfsmount of /dev/sda1, then
651 * /dev/sda2, then /dev/sda3, then NULL.
653 * lookup_mnt takes a reference to the found vfsmount.
655 struct vfsmount
*lookup_mnt(struct path
*path
)
657 struct mount
*child_mnt
;
663 seq
= read_seqbegin(&mount_lock
);
664 child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
);
665 m
= child_mnt
? &child_mnt
->mnt
: NULL
;
666 } while (!legitimize_mnt(m
, seq
));
672 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
673 * current mount namespace.
675 * The common case is dentries are not mountpoints at all and that
676 * test is handled inline. For the slow case when we are actually
677 * dealing with a mountpoint of some kind, walk through all of the
678 * mounts in the current mount namespace and test to see if the dentry
681 * The mount_hashtable is not usable in the context because we
682 * need to identify all mounts that may be in the current mount
683 * namespace not just a mount that happens to have some specified
686 bool __is_local_mountpoint(struct dentry
*dentry
)
688 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
690 bool is_covered
= false;
692 if (!d_mountpoint(dentry
))
695 down_read(&namespace_sem
);
696 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
697 is_covered
= (mnt
->mnt_mountpoint
== dentry
);
701 up_read(&namespace_sem
);
706 static struct mountpoint
*lookup_mountpoint(struct dentry
*dentry
)
708 struct hlist_head
*chain
= mp_hash(dentry
);
709 struct mountpoint
*mp
;
711 hlist_for_each_entry(mp
, chain
, m_hash
) {
712 if (mp
->m_dentry
== dentry
) {
713 /* might be worth a WARN_ON() */
714 if (d_unlinked(dentry
))
715 return ERR_PTR(-ENOENT
);
723 static struct mountpoint
*new_mountpoint(struct dentry
*dentry
)
725 struct hlist_head
*chain
= mp_hash(dentry
);
726 struct mountpoint
*mp
;
729 mp
= kmalloc(sizeof(struct mountpoint
), GFP_KERNEL
);
731 return ERR_PTR(-ENOMEM
);
733 ret
= d_set_mounted(dentry
);
739 mp
->m_dentry
= dentry
;
741 hlist_add_head(&mp
->m_hash
, chain
);
742 INIT_HLIST_HEAD(&mp
->m_list
);
746 static void put_mountpoint(struct mountpoint
*mp
)
748 if (!--mp
->m_count
) {
749 struct dentry
*dentry
= mp
->m_dentry
;
750 BUG_ON(!hlist_empty(&mp
->m_list
));
751 spin_lock(&dentry
->d_lock
);
752 dentry
->d_flags
&= ~DCACHE_MOUNTED
;
753 spin_unlock(&dentry
->d_lock
);
754 hlist_del(&mp
->m_hash
);
759 static inline int check_mnt(struct mount
*mnt
)
761 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
765 * vfsmount lock must be held for write
767 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
771 wake_up_interruptible(&ns
->poll
);
776 * vfsmount lock must be held for write
778 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
780 if (ns
&& ns
->event
!= event
) {
782 wake_up_interruptible(&ns
->poll
);
787 * vfsmount lock must be held for write
789 static void detach_mnt(struct mount
*mnt
, struct path
*old_path
)
791 old_path
->dentry
= mnt
->mnt_mountpoint
;
792 old_path
->mnt
= &mnt
->mnt_parent
->mnt
;
793 mnt
->mnt_parent
= mnt
;
794 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
795 list_del_init(&mnt
->mnt_child
);
796 hlist_del_init_rcu(&mnt
->mnt_hash
);
797 hlist_del_init(&mnt
->mnt_mp_list
);
798 put_mountpoint(mnt
->mnt_mp
);
803 * vfsmount lock must be held for write
805 void mnt_set_mountpoint(struct mount
*mnt
,
806 struct mountpoint
*mp
,
807 struct mount
*child_mnt
)
810 mnt_add_count(mnt
, 1); /* essentially, that's mntget */
811 child_mnt
->mnt_mountpoint
= dget(mp
->m_dentry
);
812 child_mnt
->mnt_parent
= mnt
;
813 child_mnt
->mnt_mp
= mp
;
814 hlist_add_head(&child_mnt
->mnt_mp_list
, &mp
->m_list
);
818 * vfsmount lock must be held for write
820 static void attach_mnt(struct mount
*mnt
,
821 struct mount
*parent
,
822 struct mountpoint
*mp
)
824 mnt_set_mountpoint(parent
, mp
, mnt
);
825 hlist_add_head_rcu(&mnt
->mnt_hash
, m_hash(&parent
->mnt
, mp
->m_dentry
));
826 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
829 static void attach_shadowed(struct mount
*mnt
,
830 struct mount
*parent
,
831 struct mount
*shadows
)
834 hlist_add_behind_rcu(&mnt
->mnt_hash
, &shadows
->mnt_hash
);
835 list_add(&mnt
->mnt_child
, &shadows
->mnt_child
);
837 hlist_add_head_rcu(&mnt
->mnt_hash
,
838 m_hash(&parent
->mnt
, mnt
->mnt_mountpoint
));
839 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
844 * vfsmount lock must be held for write
846 static void commit_tree(struct mount
*mnt
, struct mount
*shadows
)
848 struct mount
*parent
= mnt
->mnt_parent
;
851 struct mnt_namespace
*n
= parent
->mnt_ns
;
853 BUG_ON(parent
== mnt
);
855 list_add_tail(&head
, &mnt
->mnt_list
);
856 list_for_each_entry(m
, &head
, mnt_list
)
859 list_splice(&head
, n
->list
.prev
);
861 attach_shadowed(mnt
, parent
, shadows
);
862 touch_mnt_namespace(n
);
865 static struct mount
*next_mnt(struct mount
*p
, struct mount
*root
)
867 struct list_head
*next
= p
->mnt_mounts
.next
;
868 if (next
== &p
->mnt_mounts
) {
872 next
= p
->mnt_child
.next
;
873 if (next
!= &p
->mnt_parent
->mnt_mounts
)
878 return list_entry(next
, struct mount
, mnt_child
);
881 static struct mount
*skip_mnt_tree(struct mount
*p
)
883 struct list_head
*prev
= p
->mnt_mounts
.prev
;
884 while (prev
!= &p
->mnt_mounts
) {
885 p
= list_entry(prev
, struct mount
, mnt_child
);
886 prev
= p
->mnt_mounts
.prev
;
892 vfs_kern_mount(struct file_system_type
*type
, int flags
, const char *name
, void *data
)
898 return ERR_PTR(-ENODEV
);
900 mnt
= alloc_vfsmnt(name
);
902 return ERR_PTR(-ENOMEM
);
904 if (flags
& MS_KERNMOUNT
)
905 mnt
->mnt
.mnt_flags
= MNT_INTERNAL
;
907 root
= mount_fs(type
, flags
, name
, data
);
911 return ERR_CAST(root
);
914 mnt
->mnt
.mnt_root
= root
;
915 mnt
->mnt
.mnt_sb
= root
->d_sb
;
916 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
917 mnt
->mnt_parent
= mnt
;
919 list_add_tail(&mnt
->mnt_instance
, &root
->d_sb
->s_mounts
);
923 EXPORT_SYMBOL_GPL(vfs_kern_mount
);
925 static struct mount
*clone_mnt(struct mount
*old
, struct dentry
*root
,
928 struct super_block
*sb
= old
->mnt
.mnt_sb
;
932 mnt
= alloc_vfsmnt(old
->mnt_devname
);
934 return ERR_PTR(-ENOMEM
);
936 if (flag
& (CL_SLAVE
| CL_PRIVATE
| CL_SHARED_TO_SLAVE
))
937 mnt
->mnt_group_id
= 0; /* not a peer of original */
939 mnt
->mnt_group_id
= old
->mnt_group_id
;
941 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
942 err
= mnt_alloc_group_id(mnt
);
947 mnt
->mnt
.mnt_flags
= old
->mnt
.mnt_flags
& ~(MNT_WRITE_HOLD
|MNT_MARKED
);
948 /* Don't allow unprivileged users to change mount flags */
949 if (flag
& CL_UNPRIVILEGED
) {
950 mnt
->mnt
.mnt_flags
|= MNT_LOCK_ATIME
;
952 if (mnt
->mnt
.mnt_flags
& MNT_READONLY
)
953 mnt
->mnt
.mnt_flags
|= MNT_LOCK_READONLY
;
955 if (mnt
->mnt
.mnt_flags
& MNT_NODEV
)
956 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NODEV
;
958 if (mnt
->mnt
.mnt_flags
& MNT_NOSUID
)
959 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOSUID
;
961 if (mnt
->mnt
.mnt_flags
& MNT_NOEXEC
)
962 mnt
->mnt
.mnt_flags
|= MNT_LOCK_NOEXEC
;
965 /* Don't allow unprivileged users to reveal what is under a mount */
966 if ((flag
& CL_UNPRIVILEGED
) && list_empty(&old
->mnt_expire
))
967 mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
969 atomic_inc(&sb
->s_active
);
970 mnt
->mnt
.mnt_sb
= sb
;
971 mnt
->mnt
.mnt_root
= dget(root
);
972 mnt
->mnt_mountpoint
= mnt
->mnt
.mnt_root
;
973 mnt
->mnt_parent
= mnt
;
975 list_add_tail(&mnt
->mnt_instance
, &sb
->s_mounts
);
978 if ((flag
& CL_SLAVE
) ||
979 ((flag
& CL_SHARED_TO_SLAVE
) && IS_MNT_SHARED(old
))) {
980 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
981 mnt
->mnt_master
= old
;
982 CLEAR_MNT_SHARED(mnt
);
983 } else if (!(flag
& CL_PRIVATE
)) {
984 if ((flag
& CL_MAKE_SHARED
) || IS_MNT_SHARED(old
))
985 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
986 if (IS_MNT_SLAVE(old
))
987 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
988 mnt
->mnt_master
= old
->mnt_master
;
990 if (flag
& CL_MAKE_SHARED
)
993 /* stick the duplicate mount on the same expiry list
994 * as the original if that was on one */
995 if (flag
& CL_EXPIRE
) {
996 if (!list_empty(&old
->mnt_expire
))
997 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
1005 return ERR_PTR(err
);
1008 static void cleanup_mnt(struct mount
*mnt
)
1011 * This probably indicates that somebody messed
1012 * up a mnt_want/drop_write() pair. If this
1013 * happens, the filesystem was probably unable
1014 * to make r/w->r/o transitions.
1017 * The locking used to deal with mnt_count decrement provides barriers,
1018 * so mnt_get_writers() below is safe.
1020 WARN_ON(mnt_get_writers(mnt
));
1021 if (unlikely(mnt
->mnt_pins
.first
))
1023 fsnotify_vfsmount_delete(&mnt
->mnt
);
1024 dput(mnt
->mnt
.mnt_root
);
1025 deactivate_super(mnt
->mnt
.mnt_sb
);
1027 call_rcu(&mnt
->mnt_rcu
, delayed_free_vfsmnt
);
1030 static void __cleanup_mnt(struct rcu_head
*head
)
1032 cleanup_mnt(container_of(head
, struct mount
, mnt_rcu
));
1035 static LLIST_HEAD(delayed_mntput_list
);
1036 static void delayed_mntput(struct work_struct
*unused
)
1038 struct llist_node
*node
= llist_del_all(&delayed_mntput_list
);
1039 struct llist_node
*next
;
1041 for (; node
; node
= next
) {
1042 next
= llist_next(node
);
1043 cleanup_mnt(llist_entry(node
, struct mount
, mnt_llist
));
1046 static DECLARE_DELAYED_WORK(delayed_mntput_work
, delayed_mntput
);
1048 static void mntput_no_expire(struct mount
*mnt
)
1051 mnt_add_count(mnt
, -1);
1052 if (likely(mnt
->mnt_ns
)) { /* shouldn't be the last one */
1057 if (mnt_get_count(mnt
)) {
1059 unlock_mount_hash();
1062 if (unlikely(mnt
->mnt
.mnt_flags
& MNT_DOOMED
)) {
1064 unlock_mount_hash();
1067 mnt
->mnt
.mnt_flags
|= MNT_DOOMED
;
1070 list_del(&mnt
->mnt_instance
);
1071 unlock_mount_hash();
1073 if (likely(!(mnt
->mnt
.mnt_flags
& MNT_INTERNAL
))) {
1074 struct task_struct
*task
= current
;
1075 if (likely(!(task
->flags
& PF_KTHREAD
))) {
1076 init_task_work(&mnt
->mnt_rcu
, __cleanup_mnt
);
1077 if (!task_work_add(task
, &mnt
->mnt_rcu
, true))
1080 if (llist_add(&mnt
->mnt_llist
, &delayed_mntput_list
))
1081 schedule_delayed_work(&delayed_mntput_work
, 1);
1087 void mntput(struct vfsmount
*mnt
)
1090 struct mount
*m
= real_mount(mnt
);
1091 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1092 if (unlikely(m
->mnt_expiry_mark
))
1093 m
->mnt_expiry_mark
= 0;
1094 mntput_no_expire(m
);
1097 EXPORT_SYMBOL(mntput
);
1099 struct vfsmount
*mntget(struct vfsmount
*mnt
)
1102 mnt_add_count(real_mount(mnt
), 1);
1105 EXPORT_SYMBOL(mntget
);
1107 struct vfsmount
*mnt_clone_internal(struct path
*path
)
1110 p
= clone_mnt(real_mount(path
->mnt
), path
->dentry
, CL_PRIVATE
);
1113 p
->mnt
.mnt_flags
|= MNT_INTERNAL
;
1117 static inline void mangle(struct seq_file
*m
, const char *s
)
1119 seq_escape(m
, s
, " \t\n\\");
1123 * Simple .show_options callback for filesystems which don't want to
1124 * implement more complex mount option showing.
1126 * See also save_mount_options().
1128 int generic_show_options(struct seq_file
*m
, struct dentry
*root
)
1130 const char *options
;
1133 options
= rcu_dereference(root
->d_sb
->s_options
);
1135 if (options
!= NULL
&& options
[0]) {
1143 EXPORT_SYMBOL(generic_show_options
);
1146 * If filesystem uses generic_show_options(), this function should be
1147 * called from the fill_super() callback.
1149 * The .remount_fs callback usually needs to be handled in a special
1150 * way, to make sure, that previous options are not overwritten if the
1153 * Also note, that if the filesystem's .remount_fs function doesn't
1154 * reset all options to their default value, but changes only newly
1155 * given options, then the displayed options will not reflect reality
1158 void save_mount_options(struct super_block
*sb
, char *options
)
1160 BUG_ON(sb
->s_options
);
1161 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
1163 EXPORT_SYMBOL(save_mount_options
);
1165 void replace_mount_options(struct super_block
*sb
, char *options
)
1167 char *old
= sb
->s_options
;
1168 rcu_assign_pointer(sb
->s_options
, options
);
1174 EXPORT_SYMBOL(replace_mount_options
);
1176 #ifdef CONFIG_PROC_FS
1177 /* iterator; we want it to have access to namespace_sem, thus here... */
1178 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
1180 struct proc_mounts
*p
= proc_mounts(m
);
1182 down_read(&namespace_sem
);
1183 if (p
->cached_event
== p
->ns
->event
) {
1184 void *v
= p
->cached_mount
;
1185 if (*pos
== p
->cached_index
)
1187 if (*pos
== p
->cached_index
+ 1) {
1188 v
= seq_list_next(v
, &p
->ns
->list
, &p
->cached_index
);
1189 return p
->cached_mount
= v
;
1193 p
->cached_event
= p
->ns
->event
;
1194 p
->cached_mount
= seq_list_start(&p
->ns
->list
, *pos
);
1195 p
->cached_index
= *pos
;
1196 return p
->cached_mount
;
1199 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1201 struct proc_mounts
*p
= proc_mounts(m
);
1203 p
->cached_mount
= seq_list_next(v
, &p
->ns
->list
, pos
);
1204 p
->cached_index
= *pos
;
1205 return p
->cached_mount
;
1208 static void m_stop(struct seq_file
*m
, void *v
)
1210 up_read(&namespace_sem
);
1213 static int m_show(struct seq_file
*m
, void *v
)
1215 struct proc_mounts
*p
= proc_mounts(m
);
1216 struct mount
*r
= list_entry(v
, struct mount
, mnt_list
);
1217 return p
->show(m
, &r
->mnt
);
1220 const struct seq_operations mounts_op
= {
1226 #endif /* CONFIG_PROC_FS */
1229 * may_umount_tree - check if a mount tree is busy
1230 * @mnt: root of mount tree
1232 * This is called to check if a tree of mounts has any
1233 * open files, pwds, chroots or sub mounts that are
1236 int may_umount_tree(struct vfsmount
*m
)
1238 struct mount
*mnt
= real_mount(m
);
1239 int actual_refs
= 0;
1240 int minimum_refs
= 0;
1244 /* write lock needed for mnt_get_count */
1246 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1247 actual_refs
+= mnt_get_count(p
);
1250 unlock_mount_hash();
1252 if (actual_refs
> minimum_refs
)
1258 EXPORT_SYMBOL(may_umount_tree
);
1261 * may_umount - check if a mount point is busy
1262 * @mnt: root of mount
1264 * This is called to check if a mount point has any
1265 * open files, pwds, chroots or sub mounts. If the
1266 * mount has sub mounts this will return busy
1267 * regardless of whether the sub mounts are busy.
1269 * Doesn't take quota and stuff into account. IOW, in some cases it will
1270 * give false negatives. The main reason why it's here is that we need
1271 * a non-destructive way to look for easily umountable filesystems.
1273 int may_umount(struct vfsmount
*mnt
)
1276 down_read(&namespace_sem
);
1278 if (propagate_mount_busy(real_mount(mnt
), 2))
1280 unlock_mount_hash();
1281 up_read(&namespace_sem
);
1285 EXPORT_SYMBOL(may_umount
);
1287 static HLIST_HEAD(unmounted
); /* protected by namespace_sem */
1289 static void namespace_unlock(void)
1292 struct hlist_head head
= unmounted
;
1294 if (likely(hlist_empty(&head
))) {
1295 up_write(&namespace_sem
);
1299 head
.first
->pprev
= &head
.first
;
1300 INIT_HLIST_HEAD(&unmounted
);
1302 /* undo decrements we'd done in umount_tree() */
1303 hlist_for_each_entry(mnt
, &head
, mnt_hash
)
1304 if (mnt
->mnt_ex_mountpoint
.mnt
)
1305 mntget(mnt
->mnt_ex_mountpoint
.mnt
);
1307 up_write(&namespace_sem
);
1311 while (!hlist_empty(&head
)) {
1312 mnt
= hlist_entry(head
.first
, struct mount
, mnt_hash
);
1313 hlist_del_init(&mnt
->mnt_hash
);
1314 if (mnt
->mnt_ex_mountpoint
.mnt
)
1315 path_put(&mnt
->mnt_ex_mountpoint
);
1320 static inline void namespace_lock(void)
1322 down_write(&namespace_sem
);
1326 * mount_lock must be held
1327 * namespace_sem must be held for write
1328 * how = 0 => just this tree, don't propagate
1329 * how = 1 => propagate; we know that nobody else has reference to any victims
1330 * how = 2 => lazy umount
1332 void umount_tree(struct mount
*mnt
, int how
)
1334 HLIST_HEAD(tmp_list
);
1336 struct mount
*last
= NULL
;
1338 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1339 hlist_del_init_rcu(&p
->mnt_hash
);
1340 hlist_add_head(&p
->mnt_hash
, &tmp_list
);
1343 hlist_for_each_entry(p
, &tmp_list
, mnt_hash
)
1344 list_del_init(&p
->mnt_child
);
1347 propagate_umount(&tmp_list
);
1349 hlist_for_each_entry(p
, &tmp_list
, mnt_hash
) {
1350 list_del_init(&p
->mnt_expire
);
1351 list_del_init(&p
->mnt_list
);
1352 __touch_mnt_namespace(p
->mnt_ns
);
1355 p
->mnt
.mnt_flags
|= MNT_SYNC_UMOUNT
;
1356 if (mnt_has_parent(p
)) {
1357 hlist_del_init(&p
->mnt_mp_list
);
1358 put_mountpoint(p
->mnt_mp
);
1359 mnt_add_count(p
->mnt_parent
, -1);
1360 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1361 p
->mnt_ex_mountpoint
.dentry
= p
->mnt_mountpoint
;
1362 p
->mnt_ex_mountpoint
.mnt
= &p
->mnt_parent
->mnt
;
1363 p
->mnt_mountpoint
= p
->mnt
.mnt_root
;
1367 change_mnt_propagation(p
, MS_PRIVATE
);
1371 last
->mnt_hash
.next
= unmounted
.first
;
1372 unmounted
.first
= tmp_list
.first
;
1373 unmounted
.first
->pprev
= &unmounted
.first
;
1377 static void shrink_submounts(struct mount
*mnt
);
1379 static int do_umount(struct mount
*mnt
, int flags
)
1381 struct super_block
*sb
= mnt
->mnt
.mnt_sb
;
1384 retval
= security_sb_umount(&mnt
->mnt
, flags
);
1389 * Allow userspace to request a mountpoint be expired rather than
1390 * unmounting unconditionally. Unmount only happens if:
1391 * (1) the mark is already set (the mark is cleared by mntput())
1392 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1394 if (flags
& MNT_EXPIRE
) {
1395 if (&mnt
->mnt
== current
->fs
->root
.mnt
||
1396 flags
& (MNT_FORCE
| MNT_DETACH
))
1400 * probably don't strictly need the lock here if we examined
1401 * all race cases, but it's a slowpath.
1404 if (mnt_get_count(mnt
) != 2) {
1405 unlock_mount_hash();
1408 unlock_mount_hash();
1410 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1415 * If we may have to abort operations to get out of this
1416 * mount, and they will themselves hold resources we must
1417 * allow the fs to do things. In the Unix tradition of
1418 * 'Gee thats tricky lets do it in userspace' the umount_begin
1419 * might fail to complete on the first run through as other tasks
1420 * must return, and the like. Thats for the mount program to worry
1421 * about for the moment.
1424 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1425 sb
->s_op
->umount_begin(sb
);
1429 * No sense to grab the lock for this test, but test itself looks
1430 * somewhat bogus. Suggestions for better replacement?
1431 * Ho-hum... In principle, we might treat that as umount + switch
1432 * to rootfs. GC would eventually take care of the old vfsmount.
1433 * Actually it makes sense, especially if rootfs would contain a
1434 * /reboot - static binary that would close all descriptors and
1435 * call reboot(9). Then init(8) could umount root and exec /reboot.
1437 if (&mnt
->mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1439 * Special case for "unmounting" root ...
1440 * we just try to remount it readonly.
1442 if (!capable(CAP_SYS_ADMIN
))
1444 down_write(&sb
->s_umount
);
1445 if (!(sb
->s_flags
& MS_RDONLY
))
1446 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1447 up_write(&sb
->s_umount
);
1455 if (flags
& MNT_DETACH
) {
1456 if (!list_empty(&mnt
->mnt_list
))
1457 umount_tree(mnt
, 2);
1460 shrink_submounts(mnt
);
1462 if (!propagate_mount_busy(mnt
, 2)) {
1463 if (!list_empty(&mnt
->mnt_list
))
1464 umount_tree(mnt
, 1);
1468 unlock_mount_hash();
1474 * __detach_mounts - lazily unmount all mounts on the specified dentry
1476 * During unlink, rmdir, and d_drop it is possible to loose the path
1477 * to an existing mountpoint, and wind up leaking the mount.
1478 * detach_mounts allows lazily unmounting those mounts instead of
1481 * The caller may hold dentry->d_inode->i_mutex.
1483 void __detach_mounts(struct dentry
*dentry
)
1485 struct mountpoint
*mp
;
1489 mp
= lookup_mountpoint(dentry
);
1494 while (!hlist_empty(&mp
->m_list
)) {
1495 mnt
= hlist_entry(mp
->m_list
.first
, struct mount
, mnt_mp_list
);
1496 umount_tree(mnt
, 2);
1498 unlock_mount_hash();
1505 * Is the caller allowed to modify his namespace?
1507 static inline bool may_mount(void)
1509 return ns_capable(current
->nsproxy
->mnt_ns
->user_ns
, CAP_SYS_ADMIN
);
1513 * Now umount can handle mount points as well as block devices.
1514 * This is important for filesystems which use unnamed block devices.
1516 * We now support a flag for forced unmount like the other 'big iron'
1517 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1520 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1525 int lookup_flags
= 0;
1527 if (flags
& ~(MNT_FORCE
| MNT_DETACH
| MNT_EXPIRE
| UMOUNT_NOFOLLOW
))
1533 if (!(flags
& UMOUNT_NOFOLLOW
))
1534 lookup_flags
|= LOOKUP_FOLLOW
;
1536 retval
= user_path_mountpoint_at(AT_FDCWD
, name
, lookup_flags
, &path
);
1539 mnt
= real_mount(path
.mnt
);
1541 if (path
.dentry
!= path
.mnt
->mnt_root
)
1543 if (!check_mnt(mnt
))
1545 if (mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
1548 retval
= do_umount(mnt
, flags
);
1550 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1552 mntput_no_expire(mnt
);
1557 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1560 * The 2.0 compatible umount. No flags.
1562 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1564 return sys_umount(name
, 0);
1569 static bool is_mnt_ns_file(struct dentry
*dentry
)
1571 /* Is this a proxy for a mount namespace? */
1572 struct inode
*inode
= dentry
->d_inode
;
1575 if (!proc_ns_inode(inode
))
1578 ei
= get_proc_ns(inode
);
1579 if (ei
->ns_ops
!= &mntns_operations
)
1585 static bool mnt_ns_loop(struct dentry
*dentry
)
1587 /* Could bind mounting the mount namespace inode cause a
1588 * mount namespace loop?
1590 struct mnt_namespace
*mnt_ns
;
1591 if (!is_mnt_ns_file(dentry
))
1594 mnt_ns
= get_proc_ns(dentry
->d_inode
)->ns
;
1595 return current
->nsproxy
->mnt_ns
->seq
>= mnt_ns
->seq
;
1598 struct mount
*copy_tree(struct mount
*mnt
, struct dentry
*dentry
,
1601 struct mount
*res
, *p
, *q
, *r
, *parent
;
1603 if (!(flag
& CL_COPY_UNBINDABLE
) && IS_MNT_UNBINDABLE(mnt
))
1604 return ERR_PTR(-EINVAL
);
1606 if (!(flag
& CL_COPY_MNT_NS_FILE
) && is_mnt_ns_file(dentry
))
1607 return ERR_PTR(-EINVAL
);
1609 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1613 q
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
1614 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1617 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1619 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1622 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1623 struct mount
*t
= NULL
;
1624 if (!(flag
& CL_COPY_UNBINDABLE
) &&
1625 IS_MNT_UNBINDABLE(s
)) {
1626 s
= skip_mnt_tree(s
);
1629 if (!(flag
& CL_COPY_MNT_NS_FILE
) &&
1630 is_mnt_ns_file(s
->mnt
.mnt_root
)) {
1631 s
= skip_mnt_tree(s
);
1634 while (p
!= s
->mnt_parent
) {
1640 q
= clone_mnt(p
, p
->mnt
.mnt_root
, flag
);
1644 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1645 mnt_set_mountpoint(parent
, p
->mnt_mp
, q
);
1646 if (!list_empty(&parent
->mnt_mounts
)) {
1647 t
= list_last_entry(&parent
->mnt_mounts
,
1648 struct mount
, mnt_child
);
1649 if (t
->mnt_mp
!= p
->mnt_mp
)
1652 attach_shadowed(q
, parent
, t
);
1653 unlock_mount_hash();
1660 umount_tree(res
, 0);
1661 unlock_mount_hash();
1666 /* Caller should check returned pointer for errors */
1668 struct vfsmount
*collect_mounts(struct path
*path
)
1672 tree
= copy_tree(real_mount(path
->mnt
), path
->dentry
,
1673 CL_COPY_ALL
| CL_PRIVATE
);
1676 return ERR_CAST(tree
);
1680 void drop_collected_mounts(struct vfsmount
*mnt
)
1684 umount_tree(real_mount(mnt
), 0);
1685 unlock_mount_hash();
1689 int iterate_mounts(int (*f
)(struct vfsmount
*, void *), void *arg
,
1690 struct vfsmount
*root
)
1693 int res
= f(root
, arg
);
1696 list_for_each_entry(mnt
, &real_mount(root
)->mnt_list
, mnt_list
) {
1697 res
= f(&mnt
->mnt
, arg
);
1704 static void cleanup_group_ids(struct mount
*mnt
, struct mount
*end
)
1708 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1709 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1710 mnt_release_group_id(p
);
1714 static int invent_group_ids(struct mount
*mnt
, bool recurse
)
1718 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1719 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1720 int err
= mnt_alloc_group_id(p
);
1722 cleanup_group_ids(mnt
, p
);
1732 * @source_mnt : mount tree to be attached
1733 * @nd : place the mount tree @source_mnt is attached
1734 * @parent_nd : if non-null, detach the source_mnt from its parent and
1735 * store the parent mount and mountpoint dentry.
1736 * (done when source_mnt is moved)
1738 * NOTE: in the table below explains the semantics when a source mount
1739 * of a given type is attached to a destination mount of a given type.
1740 * ---------------------------------------------------------------------------
1741 * | BIND MOUNT OPERATION |
1742 * |**************************************************************************
1743 * | source-->| shared | private | slave | unbindable |
1747 * |**************************************************************************
1748 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1750 * |non-shared| shared (+) | private | slave (*) | invalid |
1751 * ***************************************************************************
1752 * A bind operation clones the source mount and mounts the clone on the
1753 * destination mount.
1755 * (++) the cloned mount is propagated to all the mounts in the propagation
1756 * tree of the destination mount and the cloned mount is added to
1757 * the peer group of the source mount.
1758 * (+) the cloned mount is created under the destination mount and is marked
1759 * as shared. The cloned mount is added to the peer group of the source
1761 * (+++) the mount is propagated to all the mounts in the propagation tree
1762 * of the destination mount and the cloned mount is made slave
1763 * of the same master as that of the source mount. The cloned mount
1764 * is marked as 'shared and slave'.
1765 * (*) the cloned mount is made a slave of the same master as that of the
1768 * ---------------------------------------------------------------------------
1769 * | MOVE MOUNT OPERATION |
1770 * |**************************************************************************
1771 * | source-->| shared | private | slave | unbindable |
1775 * |**************************************************************************
1776 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1778 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1779 * ***************************************************************************
1781 * (+) the mount is moved to the destination. And is then propagated to
1782 * all the mounts in the propagation tree of the destination mount.
1783 * (+*) the mount is moved to the destination.
1784 * (+++) the mount is moved to the destination and is then propagated to
1785 * all the mounts belonging to the destination mount's propagation tree.
1786 * the mount is marked as 'shared and slave'.
1787 * (*) the mount continues to be a slave at the new location.
1789 * if the source mount is a tree, the operations explained above is
1790 * applied to each mount in the tree.
1791 * Must be called without spinlocks held, since this function can sleep
1794 static int attach_recursive_mnt(struct mount
*source_mnt
,
1795 struct mount
*dest_mnt
,
1796 struct mountpoint
*dest_mp
,
1797 struct path
*parent_path
)
1799 HLIST_HEAD(tree_list
);
1800 struct mount
*child
, *p
;
1801 struct hlist_node
*n
;
1804 if (IS_MNT_SHARED(dest_mnt
)) {
1805 err
= invent_group_ids(source_mnt
, true);
1808 err
= propagate_mnt(dest_mnt
, dest_mp
, source_mnt
, &tree_list
);
1811 goto out_cleanup_ids
;
1812 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1818 detach_mnt(source_mnt
, parent_path
);
1819 attach_mnt(source_mnt
, dest_mnt
, dest_mp
);
1820 touch_mnt_namespace(source_mnt
->mnt_ns
);
1822 mnt_set_mountpoint(dest_mnt
, dest_mp
, source_mnt
);
1823 commit_tree(source_mnt
, NULL
);
1826 hlist_for_each_entry_safe(child
, n
, &tree_list
, mnt_hash
) {
1828 hlist_del_init(&child
->mnt_hash
);
1829 q
= __lookup_mnt_last(&child
->mnt_parent
->mnt
,
1830 child
->mnt_mountpoint
);
1831 commit_tree(child
, q
);
1833 unlock_mount_hash();
1838 while (!hlist_empty(&tree_list
)) {
1839 child
= hlist_entry(tree_list
.first
, struct mount
, mnt_hash
);
1840 umount_tree(child
, 0);
1842 unlock_mount_hash();
1843 cleanup_group_ids(source_mnt
, NULL
);
1848 static struct mountpoint
*lock_mount(struct path
*path
)
1850 struct vfsmount
*mnt
;
1851 struct dentry
*dentry
= path
->dentry
;
1853 mutex_lock(&dentry
->d_inode
->i_mutex
);
1854 if (unlikely(cant_mount(dentry
))) {
1855 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1856 return ERR_PTR(-ENOENT
);
1859 mnt
= lookup_mnt(path
);
1861 struct mountpoint
*mp
= lookup_mountpoint(dentry
);
1863 mp
= new_mountpoint(dentry
);
1866 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1872 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1875 dentry
= path
->dentry
= dget(mnt
->mnt_root
);
1879 static void unlock_mount(struct mountpoint
*where
)
1881 struct dentry
*dentry
= where
->m_dentry
;
1882 put_mountpoint(where
);
1884 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1887 static int graft_tree(struct mount
*mnt
, struct mount
*p
, struct mountpoint
*mp
)
1889 if (mnt
->mnt
.mnt_sb
->s_flags
& MS_NOUSER
)
1892 if (S_ISDIR(mp
->m_dentry
->d_inode
->i_mode
) !=
1893 S_ISDIR(mnt
->mnt
.mnt_root
->d_inode
->i_mode
))
1896 return attach_recursive_mnt(mnt
, p
, mp
, NULL
);
1900 * Sanity check the flags to change_mnt_propagation.
1903 static int flags_to_propagation_type(int flags
)
1905 int type
= flags
& ~(MS_REC
| MS_SILENT
);
1907 /* Fail if any non-propagation flags are set */
1908 if (type
& ~(MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1910 /* Only one propagation flag should be set */
1911 if (!is_power_of_2(type
))
1917 * recursively change the type of the mountpoint.
1919 static int do_change_type(struct path
*path
, int flag
)
1922 struct mount
*mnt
= real_mount(path
->mnt
);
1923 int recurse
= flag
& MS_REC
;
1927 if (path
->dentry
!= path
->mnt
->mnt_root
)
1930 type
= flags_to_propagation_type(flag
);
1935 if (type
== MS_SHARED
) {
1936 err
= invent_group_ids(mnt
, recurse
);
1942 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1943 change_mnt_propagation(m
, type
);
1944 unlock_mount_hash();
1951 static bool has_locked_children(struct mount
*mnt
, struct dentry
*dentry
)
1953 struct mount
*child
;
1954 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
1955 if (!is_subdir(child
->mnt_mountpoint
, dentry
))
1958 if (child
->mnt
.mnt_flags
& MNT_LOCKED
)
1965 * do loopback mount.
1967 static int do_loopback(struct path
*path
, const char *old_name
,
1970 struct path old_path
;
1971 struct mount
*mnt
= NULL
, *old
, *parent
;
1972 struct mountpoint
*mp
;
1974 if (!old_name
|| !*old_name
)
1976 err
= kern_path(old_name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &old_path
);
1981 if (mnt_ns_loop(old_path
.dentry
))
1984 mp
= lock_mount(path
);
1989 old
= real_mount(old_path
.mnt
);
1990 parent
= real_mount(path
->mnt
);
1993 if (IS_MNT_UNBINDABLE(old
))
1996 if (!check_mnt(parent
) || !check_mnt(old
))
1999 if (!recurse
&& has_locked_children(old
, old_path
.dentry
))
2003 mnt
= copy_tree(old
, old_path
.dentry
, CL_COPY_MNT_NS_FILE
);
2005 mnt
= clone_mnt(old
, old_path
.dentry
, 0);
2012 mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2014 err
= graft_tree(mnt
, parent
, mp
);
2017 umount_tree(mnt
, 0);
2018 unlock_mount_hash();
2023 path_put(&old_path
);
2027 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
2030 int readonly_request
= 0;
2032 if (ms_flags
& MS_RDONLY
)
2033 readonly_request
= 1;
2034 if (readonly_request
== __mnt_is_readonly(mnt
))
2037 if (readonly_request
)
2038 error
= mnt_make_readonly(real_mount(mnt
));
2040 __mnt_unmake_readonly(real_mount(mnt
));
2045 * change filesystem flags. dir should be a physical root of filesystem.
2046 * If you've mounted a non-root directory somewhere and want to do remount
2047 * on it - tough luck.
2049 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
2053 struct super_block
*sb
= path
->mnt
->mnt_sb
;
2054 struct mount
*mnt
= real_mount(path
->mnt
);
2056 if (!check_mnt(mnt
))
2059 if (path
->dentry
!= path
->mnt
->mnt_root
)
2062 /* Don't allow changing of locked mnt flags.
2064 * No locks need to be held here while testing the various
2065 * MNT_LOCK flags because those flags can never be cleared
2066 * once they are set.
2068 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_READONLY
) &&
2069 !(mnt_flags
& MNT_READONLY
)) {
2072 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NODEV
) &&
2073 !(mnt_flags
& MNT_NODEV
)) {
2076 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOSUID
) &&
2077 !(mnt_flags
& MNT_NOSUID
)) {
2080 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_NOEXEC
) &&
2081 !(mnt_flags
& MNT_NOEXEC
)) {
2084 if ((mnt
->mnt
.mnt_flags
& MNT_LOCK_ATIME
) &&
2085 ((mnt
->mnt
.mnt_flags
& MNT_ATIME_MASK
) != (mnt_flags
& MNT_ATIME_MASK
))) {
2089 err
= security_sb_remount(sb
, data
);
2093 down_write(&sb
->s_umount
);
2094 if (flags
& MS_BIND
)
2095 err
= change_mount_flags(path
->mnt
, flags
);
2096 else if (!capable(CAP_SYS_ADMIN
))
2099 err
= do_remount_sb(sb
, flags
, data
, 0);
2102 mnt_flags
|= mnt
->mnt
.mnt_flags
& ~MNT_USER_SETTABLE_MASK
;
2103 mnt
->mnt
.mnt_flags
= mnt_flags
;
2104 touch_mnt_namespace(mnt
->mnt_ns
);
2105 unlock_mount_hash();
2107 up_write(&sb
->s_umount
);
2111 static inline int tree_contains_unbindable(struct mount
*mnt
)
2114 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
2115 if (IS_MNT_UNBINDABLE(p
))
2121 static int do_move_mount(struct path
*path
, const char *old_name
)
2123 struct path old_path
, parent_path
;
2126 struct mountpoint
*mp
;
2128 if (!old_name
|| !*old_name
)
2130 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
2134 mp
= lock_mount(path
);
2139 old
= real_mount(old_path
.mnt
);
2140 p
= real_mount(path
->mnt
);
2143 if (!check_mnt(p
) || !check_mnt(old
))
2146 if (old
->mnt
.mnt_flags
& MNT_LOCKED
)
2150 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
2153 if (!mnt_has_parent(old
))
2156 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
2157 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
2160 * Don't move a mount residing in a shared parent.
2162 if (IS_MNT_SHARED(old
->mnt_parent
))
2165 * Don't move a mount tree containing unbindable mounts to a destination
2166 * mount which is shared.
2168 if (IS_MNT_SHARED(p
) && tree_contains_unbindable(old
))
2171 for (; mnt_has_parent(p
); p
= p
->mnt_parent
)
2175 err
= attach_recursive_mnt(old
, real_mount(path
->mnt
), mp
, &parent_path
);
2179 /* if the mount is moved, it should no longer be expire
2181 list_del_init(&old
->mnt_expire
);
2186 path_put(&parent_path
);
2187 path_put(&old_path
);
2191 static struct vfsmount
*fs_set_subtype(struct vfsmount
*mnt
, const char *fstype
)
2194 const char *subtype
= strchr(fstype
, '.');
2203 mnt
->mnt_sb
->s_subtype
= kstrdup(subtype
, GFP_KERNEL
);
2205 if (!mnt
->mnt_sb
->s_subtype
)
2211 return ERR_PTR(err
);
2215 * add a mount into a namespace's mount tree
2217 static int do_add_mount(struct mount
*newmnt
, struct path
*path
, int mnt_flags
)
2219 struct mountpoint
*mp
;
2220 struct mount
*parent
;
2223 mnt_flags
&= ~MNT_INTERNAL_FLAGS
;
2225 mp
= lock_mount(path
);
2229 parent
= real_mount(path
->mnt
);
2231 if (unlikely(!check_mnt(parent
))) {
2232 /* that's acceptable only for automounts done in private ns */
2233 if (!(mnt_flags
& MNT_SHRINKABLE
))
2235 /* ... and for those we'd better have mountpoint still alive */
2236 if (!parent
->mnt_ns
)
2240 /* Refuse the same filesystem on the same mount point */
2242 if (path
->mnt
->mnt_sb
== newmnt
->mnt
.mnt_sb
&&
2243 path
->mnt
->mnt_root
== path
->dentry
)
2247 if (S_ISLNK(newmnt
->mnt
.mnt_root
->d_inode
->i_mode
))
2250 newmnt
->mnt
.mnt_flags
= mnt_flags
;
2251 err
= graft_tree(newmnt
, parent
, mp
);
2259 * create a new mount for userspace and request it to be added into the
2262 static int do_new_mount(struct path
*path
, const char *fstype
, int flags
,
2263 int mnt_flags
, const char *name
, void *data
)
2265 struct file_system_type
*type
;
2266 struct user_namespace
*user_ns
= current
->nsproxy
->mnt_ns
->user_ns
;
2267 struct vfsmount
*mnt
;
2273 type
= get_fs_type(fstype
);
2277 if (user_ns
!= &init_user_ns
) {
2278 if (!(type
->fs_flags
& FS_USERNS_MOUNT
)) {
2279 put_filesystem(type
);
2282 /* Only in special cases allow devices from mounts
2283 * created outside the initial user namespace.
2285 if (!(type
->fs_flags
& FS_USERNS_DEV_MOUNT
)) {
2287 mnt_flags
|= MNT_NODEV
| MNT_LOCK_NODEV
;
2291 mnt
= vfs_kern_mount(type
, flags
, name
, data
);
2292 if (!IS_ERR(mnt
) && (type
->fs_flags
& FS_HAS_SUBTYPE
) &&
2293 !mnt
->mnt_sb
->s_subtype
)
2294 mnt
= fs_set_subtype(mnt
, fstype
);
2296 put_filesystem(type
);
2298 return PTR_ERR(mnt
);
2300 err
= do_add_mount(real_mount(mnt
), path
, mnt_flags
);
2306 int finish_automount(struct vfsmount
*m
, struct path
*path
)
2308 struct mount
*mnt
= real_mount(m
);
2310 /* The new mount record should have at least 2 refs to prevent it being
2311 * expired before we get a chance to add it
2313 BUG_ON(mnt_get_count(mnt
) < 2);
2315 if (m
->mnt_sb
== path
->mnt
->mnt_sb
&&
2316 m
->mnt_root
== path
->dentry
) {
2321 err
= do_add_mount(mnt
, path
, path
->mnt
->mnt_flags
| MNT_SHRINKABLE
);
2325 /* remove m from any expiration list it may be on */
2326 if (!list_empty(&mnt
->mnt_expire
)) {
2328 list_del_init(&mnt
->mnt_expire
);
2337 * mnt_set_expiry - Put a mount on an expiration list
2338 * @mnt: The mount to list.
2339 * @expiry_list: The list to add the mount to.
2341 void mnt_set_expiry(struct vfsmount
*mnt
, struct list_head
*expiry_list
)
2345 list_add_tail(&real_mount(mnt
)->mnt_expire
, expiry_list
);
2349 EXPORT_SYMBOL(mnt_set_expiry
);
2352 * process a list of expirable mountpoints with the intent of discarding any
2353 * mountpoints that aren't in use and haven't been touched since last we came
2356 void mark_mounts_for_expiry(struct list_head
*mounts
)
2358 struct mount
*mnt
, *next
;
2359 LIST_HEAD(graveyard
);
2361 if (list_empty(mounts
))
2367 /* extract from the expiration list every vfsmount that matches the
2368 * following criteria:
2369 * - only referenced by its parent vfsmount
2370 * - still marked for expiry (marked on the last call here; marks are
2371 * cleared by mntput())
2373 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
2374 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
2375 propagate_mount_busy(mnt
, 1))
2377 list_move(&mnt
->mnt_expire
, &graveyard
);
2379 while (!list_empty(&graveyard
)) {
2380 mnt
= list_first_entry(&graveyard
, struct mount
, mnt_expire
);
2381 touch_mnt_namespace(mnt
->mnt_ns
);
2382 umount_tree(mnt
, 1);
2384 unlock_mount_hash();
2388 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
2391 * Ripoff of 'select_parent()'
2393 * search the list of submounts for a given mountpoint, and move any
2394 * shrinkable submounts to the 'graveyard' list.
2396 static int select_submounts(struct mount
*parent
, struct list_head
*graveyard
)
2398 struct mount
*this_parent
= parent
;
2399 struct list_head
*next
;
2403 next
= this_parent
->mnt_mounts
.next
;
2405 while (next
!= &this_parent
->mnt_mounts
) {
2406 struct list_head
*tmp
= next
;
2407 struct mount
*mnt
= list_entry(tmp
, struct mount
, mnt_child
);
2410 if (!(mnt
->mnt
.mnt_flags
& MNT_SHRINKABLE
))
2413 * Descend a level if the d_mounts list is non-empty.
2415 if (!list_empty(&mnt
->mnt_mounts
)) {
2420 if (!propagate_mount_busy(mnt
, 1)) {
2421 list_move_tail(&mnt
->mnt_expire
, graveyard
);
2426 * All done at this level ... ascend and resume the search
2428 if (this_parent
!= parent
) {
2429 next
= this_parent
->mnt_child
.next
;
2430 this_parent
= this_parent
->mnt_parent
;
2437 * process a list of expirable mountpoints with the intent of discarding any
2438 * submounts of a specific parent mountpoint
2440 * mount_lock must be held for write
2442 static void shrink_submounts(struct mount
*mnt
)
2444 LIST_HEAD(graveyard
);
2447 /* extract submounts of 'mountpoint' from the expiration list */
2448 while (select_submounts(mnt
, &graveyard
)) {
2449 while (!list_empty(&graveyard
)) {
2450 m
= list_first_entry(&graveyard
, struct mount
,
2452 touch_mnt_namespace(m
->mnt_ns
);
2459 * Some copy_from_user() implementations do not return the exact number of
2460 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2461 * Note that this function differs from copy_from_user() in that it will oops
2462 * on bad values of `to', rather than returning a short copy.
2464 static long exact_copy_from_user(void *to
, const void __user
* from
,
2468 const char __user
*f
= from
;
2471 if (!access_ok(VERIFY_READ
, from
, n
))
2475 if (__get_user(c
, f
)) {
2486 int copy_mount_options(const void __user
* data
, unsigned long *where
)
2496 if (!(page
= __get_free_page(GFP_KERNEL
)))
2499 /* We only care that *some* data at the address the user
2500 * gave us is valid. Just in case, we'll zero
2501 * the remainder of the page.
2503 /* copy_from_user cannot cross TASK_SIZE ! */
2504 size
= TASK_SIZE
- (unsigned long)data
;
2505 if (size
> PAGE_SIZE
)
2508 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
2514 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
2519 char *copy_mount_string(const void __user
*data
)
2521 return data
? strndup_user(data
, PAGE_SIZE
) : NULL
;
2525 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2526 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2528 * data is a (void *) that can point to any structure up to
2529 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2530 * information (or be NULL).
2532 * Pre-0.97 versions of mount() didn't have a flags word.
2533 * When the flags word was introduced its top half was required
2534 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2535 * Therefore, if this magic number is present, it carries no information
2536 * and must be discarded.
2538 long do_mount(const char *dev_name
, const char __user
*dir_name
,
2539 const char *type_page
, unsigned long flags
, void *data_page
)
2546 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
2547 flags
&= ~MS_MGC_MSK
;
2549 /* Basic sanity checks */
2551 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
2553 /* ... and get the mountpoint */
2554 retval
= user_path(dir_name
, &path
);
2558 retval
= security_sb_mount(dev_name
, &path
,
2559 type_page
, flags
, data_page
);
2560 if (!retval
&& !may_mount())
2565 /* Default to relatime unless overriden */
2566 if (!(flags
& MS_NOATIME
))
2567 mnt_flags
|= MNT_RELATIME
;
2569 /* Separate the per-mountpoint flags */
2570 if (flags
& MS_NOSUID
)
2571 mnt_flags
|= MNT_NOSUID
;
2572 if (flags
& MS_NODEV
)
2573 mnt_flags
|= MNT_NODEV
;
2574 if (flags
& MS_NOEXEC
)
2575 mnt_flags
|= MNT_NOEXEC
;
2576 if (flags
& MS_NOATIME
)
2577 mnt_flags
|= MNT_NOATIME
;
2578 if (flags
& MS_NODIRATIME
)
2579 mnt_flags
|= MNT_NODIRATIME
;
2580 if (flags
& MS_STRICTATIME
)
2581 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
2582 if (flags
& MS_RDONLY
)
2583 mnt_flags
|= MNT_READONLY
;
2585 /* The default atime for remount is preservation */
2586 if ((flags
& MS_REMOUNT
) &&
2587 ((flags
& (MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
|
2588 MS_STRICTATIME
)) == 0)) {
2589 mnt_flags
&= ~MNT_ATIME_MASK
;
2590 mnt_flags
|= path
.mnt
->mnt_flags
& MNT_ATIME_MASK
;
2593 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
| MS_BORN
|
2594 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
2597 if (flags
& MS_REMOUNT
)
2598 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
2600 else if (flags
& MS_BIND
)
2601 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
2602 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
2603 retval
= do_change_type(&path
, flags
);
2604 else if (flags
& MS_MOVE
)
2605 retval
= do_move_mount(&path
, dev_name
);
2607 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
2608 dev_name
, data_page
);
2614 static void free_mnt_ns(struct mnt_namespace
*ns
)
2616 proc_free_inum(ns
->proc_inum
);
2617 put_user_ns(ns
->user_ns
);
2622 * Assign a sequence number so we can detect when we attempt to bind
2623 * mount a reference to an older mount namespace into the current
2624 * mount namespace, preventing reference counting loops. A 64bit
2625 * number incrementing at 10Ghz will take 12,427 years to wrap which
2626 * is effectively never, so we can ignore the possibility.
2628 static atomic64_t mnt_ns_seq
= ATOMIC64_INIT(1);
2630 static struct mnt_namespace
*alloc_mnt_ns(struct user_namespace
*user_ns
)
2632 struct mnt_namespace
*new_ns
;
2635 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
2637 return ERR_PTR(-ENOMEM
);
2638 ret
= proc_alloc_inum(&new_ns
->proc_inum
);
2641 return ERR_PTR(ret
);
2643 new_ns
->seq
= atomic64_add_return(1, &mnt_ns_seq
);
2644 atomic_set(&new_ns
->count
, 1);
2645 new_ns
->root
= NULL
;
2646 INIT_LIST_HEAD(&new_ns
->list
);
2647 init_waitqueue_head(&new_ns
->poll
);
2649 new_ns
->user_ns
= get_user_ns(user_ns
);
2653 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2654 struct user_namespace
*user_ns
, struct fs_struct
*new_fs
)
2656 struct mnt_namespace
*new_ns
;
2657 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
2658 struct mount
*p
, *q
;
2665 if (likely(!(flags
& CLONE_NEWNS
))) {
2672 new_ns
= alloc_mnt_ns(user_ns
);
2677 /* First pass: copy the tree topology */
2678 copy_flags
= CL_COPY_UNBINDABLE
| CL_EXPIRE
;
2679 if (user_ns
!= ns
->user_ns
)
2680 copy_flags
|= CL_SHARED_TO_SLAVE
| CL_UNPRIVILEGED
;
2681 new = copy_tree(old
, old
->mnt
.mnt_root
, copy_flags
);
2684 free_mnt_ns(new_ns
);
2685 return ERR_CAST(new);
2688 list_add_tail(&new_ns
->list
, &new->mnt_list
);
2691 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2692 * as belonging to new namespace. We have already acquired a private
2693 * fs_struct, so tsk->fs->lock is not needed.
2700 if (&p
->mnt
== new_fs
->root
.mnt
) {
2701 new_fs
->root
.mnt
= mntget(&q
->mnt
);
2704 if (&p
->mnt
== new_fs
->pwd
.mnt
) {
2705 new_fs
->pwd
.mnt
= mntget(&q
->mnt
);
2709 p
= next_mnt(p
, old
);
2710 q
= next_mnt(q
, new);
2713 while (p
->mnt
.mnt_root
!= q
->mnt
.mnt_root
)
2714 p
= next_mnt(p
, old
);
2727 * create_mnt_ns - creates a private namespace and adds a root filesystem
2728 * @mnt: pointer to the new root filesystem mountpoint
2730 static struct mnt_namespace
*create_mnt_ns(struct vfsmount
*m
)
2732 struct mnt_namespace
*new_ns
= alloc_mnt_ns(&init_user_ns
);
2733 if (!IS_ERR(new_ns
)) {
2734 struct mount
*mnt
= real_mount(m
);
2735 mnt
->mnt_ns
= new_ns
;
2737 list_add(&mnt
->mnt_list
, &new_ns
->list
);
2744 struct dentry
*mount_subtree(struct vfsmount
*mnt
, const char *name
)
2746 struct mnt_namespace
*ns
;
2747 struct super_block
*s
;
2751 ns
= create_mnt_ns(mnt
);
2753 return ERR_CAST(ns
);
2755 err
= vfs_path_lookup(mnt
->mnt_root
, mnt
,
2756 name
, LOOKUP_FOLLOW
|LOOKUP_AUTOMOUNT
, &path
);
2761 return ERR_PTR(err
);
2763 /* trade a vfsmount reference for active sb one */
2764 s
= path
.mnt
->mnt_sb
;
2765 atomic_inc(&s
->s_active
);
2767 /* lock the sucker */
2768 down_write(&s
->s_umount
);
2769 /* ... and return the root of (sub)tree on it */
2772 EXPORT_SYMBOL(mount_subtree
);
2774 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2775 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2780 unsigned long data_page
;
2782 kernel_type
= copy_mount_string(type
);
2783 ret
= PTR_ERR(kernel_type
);
2784 if (IS_ERR(kernel_type
))
2787 kernel_dev
= copy_mount_string(dev_name
);
2788 ret
= PTR_ERR(kernel_dev
);
2789 if (IS_ERR(kernel_dev
))
2792 ret
= copy_mount_options(data
, &data_page
);
2796 ret
= do_mount(kernel_dev
, dir_name
, kernel_type
, flags
,
2797 (void *) data_page
);
2799 free_page(data_page
);
2809 * Return true if path is reachable from root
2811 * namespace_sem or mount_lock is held
2813 bool is_path_reachable(struct mount
*mnt
, struct dentry
*dentry
,
2814 const struct path
*root
)
2816 while (&mnt
->mnt
!= root
->mnt
&& mnt_has_parent(mnt
)) {
2817 dentry
= mnt
->mnt_mountpoint
;
2818 mnt
= mnt
->mnt_parent
;
2820 return &mnt
->mnt
== root
->mnt
&& is_subdir(dentry
, root
->dentry
);
2823 int path_is_under(struct path
*path1
, struct path
*path2
)
2826 read_seqlock_excl(&mount_lock
);
2827 res
= is_path_reachable(real_mount(path1
->mnt
), path1
->dentry
, path2
);
2828 read_sequnlock_excl(&mount_lock
);
2831 EXPORT_SYMBOL(path_is_under
);
2834 * pivot_root Semantics:
2835 * Moves the root file system of the current process to the directory put_old,
2836 * makes new_root as the new root file system of the current process, and sets
2837 * root/cwd of all processes which had them on the current root to new_root.
2840 * The new_root and put_old must be directories, and must not be on the
2841 * same file system as the current process root. The put_old must be
2842 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2843 * pointed to by put_old must yield the same directory as new_root. No other
2844 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2846 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2847 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2848 * in this situation.
2851 * - we don't move root/cwd if they are not at the root (reason: if something
2852 * cared enough to change them, it's probably wrong to force them elsewhere)
2853 * - it's okay to pick a root that isn't the root of a file system, e.g.
2854 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2855 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2858 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2859 const char __user
*, put_old
)
2861 struct path
new, old
, parent_path
, root_parent
, root
;
2862 struct mount
*new_mnt
, *root_mnt
, *old_mnt
;
2863 struct mountpoint
*old_mp
, *root_mp
;
2869 error
= user_path_dir(new_root
, &new);
2873 error
= user_path_dir(put_old
, &old
);
2877 error
= security_sb_pivotroot(&old
, &new);
2881 get_fs_root(current
->fs
, &root
);
2882 old_mp
= lock_mount(&old
);
2883 error
= PTR_ERR(old_mp
);
2888 new_mnt
= real_mount(new.mnt
);
2889 root_mnt
= real_mount(root
.mnt
);
2890 old_mnt
= real_mount(old
.mnt
);
2891 if (IS_MNT_SHARED(old_mnt
) ||
2892 IS_MNT_SHARED(new_mnt
->mnt_parent
) ||
2893 IS_MNT_SHARED(root_mnt
->mnt_parent
))
2895 if (!check_mnt(root_mnt
) || !check_mnt(new_mnt
))
2897 if (new_mnt
->mnt
.mnt_flags
& MNT_LOCKED
)
2900 if (d_unlinked(new.dentry
))
2903 if (new_mnt
== root_mnt
|| old_mnt
== root_mnt
)
2904 goto out4
; /* loop, on the same file system */
2906 if (root
.mnt
->mnt_root
!= root
.dentry
)
2907 goto out4
; /* not a mountpoint */
2908 if (!mnt_has_parent(root_mnt
))
2909 goto out4
; /* not attached */
2910 root_mp
= root_mnt
->mnt_mp
;
2911 if (new.mnt
->mnt_root
!= new.dentry
)
2912 goto out4
; /* not a mountpoint */
2913 if (!mnt_has_parent(new_mnt
))
2914 goto out4
; /* not attached */
2915 /* make sure we can reach put_old from new_root */
2916 if (!is_path_reachable(old_mnt
, old
.dentry
, &new))
2918 /* make certain new is below the root */
2919 if (!is_path_reachable(new_mnt
, new.dentry
, &root
))
2921 root_mp
->m_count
++; /* pin it so it won't go away */
2923 detach_mnt(new_mnt
, &parent_path
);
2924 detach_mnt(root_mnt
, &root_parent
);
2925 if (root_mnt
->mnt
.mnt_flags
& MNT_LOCKED
) {
2926 new_mnt
->mnt
.mnt_flags
|= MNT_LOCKED
;
2927 root_mnt
->mnt
.mnt_flags
&= ~MNT_LOCKED
;
2929 /* mount old root on put_old */
2930 attach_mnt(root_mnt
, old_mnt
, old_mp
);
2931 /* mount new_root on / */
2932 attach_mnt(new_mnt
, real_mount(root_parent
.mnt
), root_mp
);
2933 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2934 unlock_mount_hash();
2935 chroot_fs_refs(&root
, &new);
2936 put_mountpoint(root_mp
);
2939 unlock_mount(old_mp
);
2941 path_put(&root_parent
);
2942 path_put(&parent_path
);
2954 static void __init
init_mount_tree(void)
2956 struct vfsmount
*mnt
;
2957 struct mnt_namespace
*ns
;
2959 struct file_system_type
*type
;
2961 type
= get_fs_type("rootfs");
2963 panic("Can't find rootfs type");
2964 mnt
= vfs_kern_mount(type
, 0, "rootfs", NULL
);
2965 put_filesystem(type
);
2967 panic("Can't create rootfs");
2969 ns
= create_mnt_ns(mnt
);
2971 panic("Can't allocate initial namespace");
2973 init_task
.nsproxy
->mnt_ns
= ns
;
2977 root
.dentry
= mnt
->mnt_root
;
2979 set_fs_pwd(current
->fs
, &root
);
2980 set_fs_root(current
->fs
, &root
);
2983 void __init
mnt_init(void)
2988 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct mount
),
2989 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2991 mount_hashtable
= alloc_large_system_hash("Mount-cache",
2992 sizeof(struct hlist_head
),
2995 &m_hash_shift
, &m_hash_mask
, 0, 0);
2996 mountpoint_hashtable
= alloc_large_system_hash("Mountpoint-cache",
2997 sizeof(struct hlist_head
),
3000 &mp_hash_shift
, &mp_hash_mask
, 0, 0);
3002 if (!mount_hashtable
|| !mountpoint_hashtable
)
3003 panic("Failed to allocate mount hash table\n");
3005 for (u
= 0; u
<= m_hash_mask
; u
++)
3006 INIT_HLIST_HEAD(&mount_hashtable
[u
]);
3007 for (u
= 0; u
<= mp_hash_mask
; u
++)
3008 INIT_HLIST_HEAD(&mountpoint_hashtable
[u
]);
3014 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
3016 fs_kobj
= kobject_create_and_add("fs", NULL
);
3018 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
3023 void put_mnt_ns(struct mnt_namespace
*ns
)
3025 if (!atomic_dec_and_test(&ns
->count
))
3027 drop_collected_mounts(&ns
->root
->mnt
);
3031 struct vfsmount
*kern_mount_data(struct file_system_type
*type
, void *data
)
3033 struct vfsmount
*mnt
;
3034 mnt
= vfs_kern_mount(type
, MS_KERNMOUNT
, type
->name
, data
);
3037 * it is a longterm mount, don't release mnt until
3038 * we unmount before file sys is unregistered
3040 real_mount(mnt
)->mnt_ns
= MNT_NS_INTERNAL
;
3044 EXPORT_SYMBOL_GPL(kern_mount_data
);
3046 void kern_unmount(struct vfsmount
*mnt
)
3048 /* release long term mount so mount point can be released */
3049 if (!IS_ERR_OR_NULL(mnt
)) {
3050 real_mount(mnt
)->mnt_ns
= NULL
;
3051 synchronize_rcu(); /* yecchhh... */
3055 EXPORT_SYMBOL(kern_unmount
);
3057 bool our_mnt(struct vfsmount
*mnt
)
3059 return check_mnt(real_mount(mnt
));
3062 bool current_chrooted(void)
3064 /* Does the current process have a non-standard root */
3065 struct path ns_root
;
3066 struct path fs_root
;
3069 /* Find the namespace root */
3070 ns_root
.mnt
= ¤t
->nsproxy
->mnt_ns
->root
->mnt
;
3071 ns_root
.dentry
= ns_root
.mnt
->mnt_root
;
3073 while (d_mountpoint(ns_root
.dentry
) && follow_down_one(&ns_root
))
3076 get_fs_root(current
->fs
, &fs_root
);
3078 chrooted
= !path_equal(&fs_root
, &ns_root
);
3086 bool fs_fully_visible(struct file_system_type
*type
)
3088 struct mnt_namespace
*ns
= current
->nsproxy
->mnt_ns
;
3090 bool visible
= false;
3095 down_read(&namespace_sem
);
3096 list_for_each_entry(mnt
, &ns
->list
, mnt_list
) {
3097 struct mount
*child
;
3098 if (mnt
->mnt
.mnt_sb
->s_type
!= type
)
3101 /* This mount is not fully visible if there are any child mounts
3102 * that cover anything except for empty directories.
3104 list_for_each_entry(child
, &mnt
->mnt_mounts
, mnt_child
) {
3105 struct inode
*inode
= child
->mnt_mountpoint
->d_inode
;
3106 if (!S_ISDIR(inode
->i_mode
))
3108 if (inode
->i_nlink
> 2)
3116 up_read(&namespace_sem
);
3120 static void *mntns_get(struct task_struct
*task
)
3122 struct mnt_namespace
*ns
= NULL
;
3123 struct nsproxy
*nsproxy
;
3126 nsproxy
= task
->nsproxy
;
3128 ns
= nsproxy
->mnt_ns
;
3136 static void mntns_put(void *ns
)
3141 static int mntns_install(struct nsproxy
*nsproxy
, void *ns
)
3143 struct fs_struct
*fs
= current
->fs
;
3144 struct mnt_namespace
*mnt_ns
= ns
;
3147 if (!ns_capable(mnt_ns
->user_ns
, CAP_SYS_ADMIN
) ||
3148 !ns_capable(current_user_ns(), CAP_SYS_CHROOT
) ||
3149 !ns_capable(current_user_ns(), CAP_SYS_ADMIN
))
3156 put_mnt_ns(nsproxy
->mnt_ns
);
3157 nsproxy
->mnt_ns
= mnt_ns
;
3160 root
.mnt
= &mnt_ns
->root
->mnt
;
3161 root
.dentry
= mnt_ns
->root
->mnt
.mnt_root
;
3163 while(d_mountpoint(root
.dentry
) && follow_down_one(&root
))
3166 /* Update the pwd and root */
3167 set_fs_pwd(fs
, &root
);
3168 set_fs_root(fs
, &root
);
3174 static unsigned int mntns_inum(void *ns
)
3176 struct mnt_namespace
*mnt_ns
= ns
;
3177 return mnt_ns
->proc_inum
;
3180 const struct proc_ns_operations mntns_operations
= {
3182 .type
= CLONE_NEWNS
,
3185 .install
= mntns_install
,