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/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/acct.h>
18 #include <linux/capability.h>
19 #include <linux/cpumask.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/nsproxy.h>
26 #include <linux/security.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/log2.h>
30 #include <linux/idr.h>
31 #include <linux/fs_struct.h>
32 #include <asm/uaccess.h>
33 #include <asm/unistd.h>
37 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
38 #define HASH_SIZE (1UL << HASH_SHIFT)
40 /* spinlock for vfsmount related operations, inplace of dcache_lock */
41 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
44 static DEFINE_IDA(mnt_id_ida
);
45 static DEFINE_IDA(mnt_group_ida
);
46 static int mnt_id_start
= 0;
47 static int mnt_group_start
= 1;
49 static struct list_head
*mount_hashtable __read_mostly
;
50 static struct kmem_cache
*mnt_cache __read_mostly
;
51 static struct rw_semaphore namespace_sem
;
54 struct kobject
*fs_kobj
;
55 EXPORT_SYMBOL_GPL(fs_kobj
);
57 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
59 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
60 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
61 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
62 return tmp
& (HASH_SIZE
- 1);
65 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67 /* allocation is serialized by namespace_sem */
68 static int mnt_alloc_id(struct vfsmount
*mnt
)
73 ida_pre_get(&mnt_id_ida
, GFP_KERNEL
);
74 spin_lock(&vfsmount_lock
);
75 res
= ida_get_new_above(&mnt_id_ida
, mnt_id_start
, &mnt
->mnt_id
);
77 mnt_id_start
= mnt
->mnt_id
+ 1;
78 spin_unlock(&vfsmount_lock
);
85 static void mnt_free_id(struct vfsmount
*mnt
)
88 spin_lock(&vfsmount_lock
);
89 ida_remove(&mnt_id_ida
, id
);
90 if (mnt_id_start
> id
)
92 spin_unlock(&vfsmount_lock
);
96 * Allocate a new peer group ID
98 * mnt_group_ida is protected by namespace_sem
100 static int mnt_alloc_group_id(struct vfsmount
*mnt
)
104 if (!ida_pre_get(&mnt_group_ida
, GFP_KERNEL
))
107 res
= ida_get_new_above(&mnt_group_ida
,
111 mnt_group_start
= mnt
->mnt_group_id
+ 1;
117 * Release a peer group ID
119 void mnt_release_group_id(struct vfsmount
*mnt
)
121 int id
= mnt
->mnt_group_id
;
122 ida_remove(&mnt_group_ida
, id
);
123 if (mnt_group_start
> id
)
124 mnt_group_start
= id
;
125 mnt
->mnt_group_id
= 0;
128 struct vfsmount
*alloc_vfsmnt(const char *name
)
130 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
134 err
= mnt_alloc_id(mnt
);
139 mnt
->mnt_devname
= kstrdup(name
, GFP_KERNEL
);
140 if (!mnt
->mnt_devname
)
144 atomic_set(&mnt
->mnt_count
, 1);
145 INIT_LIST_HEAD(&mnt
->mnt_hash
);
146 INIT_LIST_HEAD(&mnt
->mnt_child
);
147 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
148 INIT_LIST_HEAD(&mnt
->mnt_list
);
149 INIT_LIST_HEAD(&mnt
->mnt_expire
);
150 INIT_LIST_HEAD(&mnt
->mnt_share
);
151 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
152 INIT_LIST_HEAD(&mnt
->mnt_slave
);
154 mnt
->mnt_writers
= alloc_percpu(int);
155 if (!mnt
->mnt_writers
)
156 goto out_free_devname
;
158 mnt
->mnt_writers
= 0;
165 kfree(mnt
->mnt_devname
);
170 kmem_cache_free(mnt_cache
, mnt
);
175 * Most r/o checks on a fs are for operations that take
176 * discrete amounts of time, like a write() or unlink().
177 * We must keep track of when those operations start
178 * (for permission checks) and when they end, so that
179 * we can determine when writes are able to occur to
183 * __mnt_is_readonly: check whether a mount is read-only
184 * @mnt: the mount to check for its write status
186 * This shouldn't be used directly ouside of the VFS.
187 * It does not guarantee that the filesystem will stay
188 * r/w, just that it is right *now*. This can not and
189 * should not be used in place of IS_RDONLY(inode).
190 * mnt_want/drop_write() will _keep_ the filesystem
193 int __mnt_is_readonly(struct vfsmount
*mnt
)
195 if (mnt
->mnt_flags
& MNT_READONLY
)
197 if (mnt
->mnt_sb
->s_flags
& MS_RDONLY
)
201 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
203 static inline void inc_mnt_writers(struct vfsmount
*mnt
)
206 (*per_cpu_ptr(mnt
->mnt_writers
, smp_processor_id()))++;
212 static inline void dec_mnt_writers(struct vfsmount
*mnt
)
215 (*per_cpu_ptr(mnt
->mnt_writers
, smp_processor_id()))--;
221 static unsigned int count_mnt_writers(struct vfsmount
*mnt
)
224 unsigned int count
= 0;
227 for_each_possible_cpu(cpu
) {
228 count
+= *per_cpu_ptr(mnt
->mnt_writers
, cpu
);
233 return mnt
->mnt_writers
;
238 * Most r/o checks on a fs are for operations that take
239 * discrete amounts of time, like a write() or unlink().
240 * We must keep track of when those operations start
241 * (for permission checks) and when they end, so that
242 * we can determine when writes are able to occur to
246 * mnt_want_write - get write access to a mount
247 * @mnt: the mount on which to take a write
249 * This tells the low-level filesystem that a write is
250 * about to be performed to it, and makes sure that
251 * writes are allowed before returning success. When
252 * the write operation is finished, mnt_drop_write()
253 * must be called. This is effectively a refcount.
255 int mnt_want_write(struct vfsmount
*mnt
)
260 inc_mnt_writers(mnt
);
262 * The store to inc_mnt_writers must be visible before we pass
263 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
264 * incremented count after it has set MNT_WRITE_HOLD.
267 while (mnt
->mnt_flags
& MNT_WRITE_HOLD
)
270 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
271 * be set to match its requirements. So we must not load that until
272 * MNT_WRITE_HOLD is cleared.
275 if (__mnt_is_readonly(mnt
)) {
276 dec_mnt_writers(mnt
);
284 EXPORT_SYMBOL_GPL(mnt_want_write
);
287 * mnt_clone_write - get write access to a mount
288 * @mnt: the mount on which to take a write
290 * This is effectively like mnt_want_write, except
291 * it must only be used to take an extra write reference
292 * on a mountpoint that we already know has a write reference
293 * on it. This allows some optimisation.
295 * After finished, mnt_drop_write must be called as usual to
296 * drop the reference.
298 int mnt_clone_write(struct vfsmount
*mnt
)
300 /* superblock may be r/o */
301 if (__mnt_is_readonly(mnt
))
304 inc_mnt_writers(mnt
);
308 EXPORT_SYMBOL_GPL(mnt_clone_write
);
311 * mnt_want_write_file - get write access to a file's mount
312 * @file: the file who's mount on which to take a write
314 * This is like mnt_want_write, but it takes a file and can
315 * do some optimisations if the file is open for write already
317 int mnt_want_write_file(struct file
*file
)
319 struct inode
*inode
= file
->f_dentry
->d_inode
;
320 if (!(file
->f_mode
& FMODE_WRITE
) || special_file(inode
->i_mode
))
321 return mnt_want_write(file
->f_path
.mnt
);
323 return mnt_clone_write(file
->f_path
.mnt
);
325 EXPORT_SYMBOL_GPL(mnt_want_write_file
);
328 * mnt_drop_write - give up write access to a mount
329 * @mnt: the mount on which to give up write access
331 * Tells the low-level filesystem that we are done
332 * performing writes to it. Must be matched with
333 * mnt_want_write() call above.
335 void mnt_drop_write(struct vfsmount
*mnt
)
338 dec_mnt_writers(mnt
);
341 EXPORT_SYMBOL_GPL(mnt_drop_write
);
343 static int mnt_make_readonly(struct vfsmount
*mnt
)
347 spin_lock(&vfsmount_lock
);
348 mnt
->mnt_flags
|= MNT_WRITE_HOLD
;
350 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
351 * should be visible before we do.
356 * With writers on hold, if this value is zero, then there are
357 * definitely no active writers (although held writers may subsequently
358 * increment the count, they'll have to wait, and decrement it after
359 * seeing MNT_READONLY).
361 * It is OK to have counter incremented on one CPU and decremented on
362 * another: the sum will add up correctly. The danger would be when we
363 * sum up each counter, if we read a counter before it is incremented,
364 * but then read another CPU's count which it has been subsequently
365 * decremented from -- we would see more decrements than we should.
366 * MNT_WRITE_HOLD protects against this scenario, because
367 * mnt_want_write first increments count, then smp_mb, then spins on
368 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
369 * we're counting up here.
371 if (count_mnt_writers(mnt
) > 0)
374 mnt
->mnt_flags
|= MNT_READONLY
;
376 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
377 * that become unheld will see MNT_READONLY.
380 mnt
->mnt_flags
&= ~MNT_WRITE_HOLD
;
381 spin_unlock(&vfsmount_lock
);
385 static void __mnt_unmake_readonly(struct vfsmount
*mnt
)
387 spin_lock(&vfsmount_lock
);
388 mnt
->mnt_flags
&= ~MNT_READONLY
;
389 spin_unlock(&vfsmount_lock
);
392 void simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
395 mnt
->mnt_root
= dget(sb
->s_root
);
398 EXPORT_SYMBOL(simple_set_mnt
);
400 void free_vfsmnt(struct vfsmount
*mnt
)
402 kfree(mnt
->mnt_devname
);
405 free_percpu(mnt
->mnt_writers
);
407 kmem_cache_free(mnt_cache
, mnt
);
411 * find the first or last mount at @dentry on vfsmount @mnt depending on
412 * @dir. If @dir is set return the first mount else return the last mount.
414 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
417 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
418 struct list_head
*tmp
= head
;
419 struct vfsmount
*p
, *found
= NULL
;
422 tmp
= dir
? tmp
->next
: tmp
->prev
;
426 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
427 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
436 * lookup_mnt increments the ref count before returning
437 * the vfsmount struct.
439 struct vfsmount
*lookup_mnt(struct path
*path
)
441 struct vfsmount
*child_mnt
;
442 spin_lock(&vfsmount_lock
);
443 if ((child_mnt
= __lookup_mnt(path
->mnt
, path
->dentry
, 1)))
445 spin_unlock(&vfsmount_lock
);
449 static inline int check_mnt(struct vfsmount
*mnt
)
451 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
454 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
458 wake_up_interruptible(&ns
->poll
);
462 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
464 if (ns
&& ns
->event
!= event
) {
466 wake_up_interruptible(&ns
->poll
);
470 static void detach_mnt(struct vfsmount
*mnt
, struct path
*old_path
)
472 old_path
->dentry
= mnt
->mnt_mountpoint
;
473 old_path
->mnt
= mnt
->mnt_parent
;
474 mnt
->mnt_parent
= mnt
;
475 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
476 list_del_init(&mnt
->mnt_child
);
477 list_del_init(&mnt
->mnt_hash
);
478 old_path
->dentry
->d_mounted
--;
481 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
482 struct vfsmount
*child_mnt
)
484 child_mnt
->mnt_parent
= mntget(mnt
);
485 child_mnt
->mnt_mountpoint
= dget(dentry
);
489 static void attach_mnt(struct vfsmount
*mnt
, struct path
*path
)
491 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
492 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
493 hash(path
->mnt
, path
->dentry
));
494 list_add_tail(&mnt
->mnt_child
, &path
->mnt
->mnt_mounts
);
498 * the caller must hold vfsmount_lock
500 static void commit_tree(struct vfsmount
*mnt
)
502 struct vfsmount
*parent
= mnt
->mnt_parent
;
505 struct mnt_namespace
*n
= parent
->mnt_ns
;
507 BUG_ON(parent
== mnt
);
509 list_add_tail(&head
, &mnt
->mnt_list
);
510 list_for_each_entry(m
, &head
, mnt_list
)
512 list_splice(&head
, n
->list
.prev
);
514 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
515 hash(parent
, mnt
->mnt_mountpoint
));
516 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
517 touch_mnt_namespace(n
);
520 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
522 struct list_head
*next
= p
->mnt_mounts
.next
;
523 if (next
== &p
->mnt_mounts
) {
527 next
= p
->mnt_child
.next
;
528 if (next
!= &p
->mnt_parent
->mnt_mounts
)
533 return list_entry(next
, struct vfsmount
, mnt_child
);
536 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
538 struct list_head
*prev
= p
->mnt_mounts
.prev
;
539 while (prev
!= &p
->mnt_mounts
) {
540 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
541 prev
= p
->mnt_mounts
.prev
;
546 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
549 struct super_block
*sb
= old
->mnt_sb
;
550 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
553 if (flag
& (CL_SLAVE
| CL_PRIVATE
))
554 mnt
->mnt_group_id
= 0; /* not a peer of original */
556 mnt
->mnt_group_id
= old
->mnt_group_id
;
558 if ((flag
& CL_MAKE_SHARED
) && !mnt
->mnt_group_id
) {
559 int err
= mnt_alloc_group_id(mnt
);
564 mnt
->mnt_flags
= old
->mnt_flags
;
565 atomic_inc(&sb
->s_active
);
567 mnt
->mnt_root
= dget(root
);
568 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
569 mnt
->mnt_parent
= mnt
;
571 if (flag
& CL_SLAVE
) {
572 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
573 mnt
->mnt_master
= old
;
574 CLEAR_MNT_SHARED(mnt
);
575 } else if (!(flag
& CL_PRIVATE
)) {
576 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
577 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
578 if (IS_MNT_SLAVE(old
))
579 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
580 mnt
->mnt_master
= old
->mnt_master
;
582 if (flag
& CL_MAKE_SHARED
)
585 /* stick the duplicate mount on the same expiry list
586 * as the original if that was on one */
587 if (flag
& CL_EXPIRE
) {
588 if (!list_empty(&old
->mnt_expire
))
589 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
599 static inline void __mntput(struct vfsmount
*mnt
)
601 struct super_block
*sb
= mnt
->mnt_sb
;
603 * This probably indicates that somebody messed
604 * up a mnt_want/drop_write() pair. If this
605 * happens, the filesystem was probably unable
606 * to make r/w->r/o transitions.
609 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
610 * provides barriers, so count_mnt_writers() below is safe. AV
612 WARN_ON(count_mnt_writers(mnt
));
615 deactivate_super(sb
);
618 void mntput_no_expire(struct vfsmount
*mnt
)
621 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
622 if (likely(!mnt
->mnt_pinned
)) {
623 spin_unlock(&vfsmount_lock
);
627 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
629 spin_unlock(&vfsmount_lock
);
630 acct_auto_close_mnt(mnt
);
631 security_sb_umount_close(mnt
);
636 EXPORT_SYMBOL(mntput_no_expire
);
638 void mnt_pin(struct vfsmount
*mnt
)
640 spin_lock(&vfsmount_lock
);
642 spin_unlock(&vfsmount_lock
);
645 EXPORT_SYMBOL(mnt_pin
);
647 void mnt_unpin(struct vfsmount
*mnt
)
649 spin_lock(&vfsmount_lock
);
650 if (mnt
->mnt_pinned
) {
651 atomic_inc(&mnt
->mnt_count
);
654 spin_unlock(&vfsmount_lock
);
657 EXPORT_SYMBOL(mnt_unpin
);
659 static inline void mangle(struct seq_file
*m
, const char *s
)
661 seq_escape(m
, s
, " \t\n\\");
665 * Simple .show_options callback for filesystems which don't want to
666 * implement more complex mount option showing.
668 * See also save_mount_options().
670 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
675 options
= rcu_dereference(mnt
->mnt_sb
->s_options
);
677 if (options
!= NULL
&& options
[0]) {
685 EXPORT_SYMBOL(generic_show_options
);
688 * If filesystem uses generic_show_options(), this function should be
689 * called from the fill_super() callback.
691 * The .remount_fs callback usually needs to be handled in a special
692 * way, to make sure, that previous options are not overwritten if the
695 * Also note, that if the filesystem's .remount_fs function doesn't
696 * reset all options to their default value, but changes only newly
697 * given options, then the displayed options will not reflect reality
700 void save_mount_options(struct super_block
*sb
, char *options
)
702 BUG_ON(sb
->s_options
);
703 rcu_assign_pointer(sb
->s_options
, kstrdup(options
, GFP_KERNEL
));
705 EXPORT_SYMBOL(save_mount_options
);
707 void replace_mount_options(struct super_block
*sb
, char *options
)
709 char *old
= sb
->s_options
;
710 rcu_assign_pointer(sb
->s_options
, options
);
716 EXPORT_SYMBOL(replace_mount_options
);
718 #ifdef CONFIG_PROC_FS
720 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
722 struct proc_mounts
*p
= m
->private;
724 down_read(&namespace_sem
);
725 return seq_list_start(&p
->ns
->list
, *pos
);
728 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
730 struct proc_mounts
*p
= m
->private;
732 return seq_list_next(v
, &p
->ns
->list
, pos
);
735 static void m_stop(struct seq_file
*m
, void *v
)
737 up_read(&namespace_sem
);
740 struct proc_fs_info
{
745 static int show_sb_opts(struct seq_file
*m
, struct super_block
*sb
)
747 static const struct proc_fs_info fs_info
[] = {
748 { MS_SYNCHRONOUS
, ",sync" },
749 { MS_DIRSYNC
, ",dirsync" },
750 { MS_MANDLOCK
, ",mand" },
753 const struct proc_fs_info
*fs_infop
;
755 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
756 if (sb
->s_flags
& fs_infop
->flag
)
757 seq_puts(m
, fs_infop
->str
);
760 return security_sb_show_options(m
, sb
);
763 static void show_mnt_opts(struct seq_file
*m
, struct vfsmount
*mnt
)
765 static const struct proc_fs_info mnt_info
[] = {
766 { MNT_NOSUID
, ",nosuid" },
767 { MNT_NODEV
, ",nodev" },
768 { MNT_NOEXEC
, ",noexec" },
769 { MNT_NOATIME
, ",noatime" },
770 { MNT_NODIRATIME
, ",nodiratime" },
771 { MNT_RELATIME
, ",relatime" },
772 { MNT_STRICTATIME
, ",strictatime" },
775 const struct proc_fs_info
*fs_infop
;
777 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
778 if (mnt
->mnt_flags
& fs_infop
->flag
)
779 seq_puts(m
, fs_infop
->str
);
783 static void show_type(struct seq_file
*m
, struct super_block
*sb
)
785 mangle(m
, sb
->s_type
->name
);
786 if (sb
->s_subtype
&& sb
->s_subtype
[0]) {
788 mangle(m
, sb
->s_subtype
);
792 static int show_vfsmnt(struct seq_file
*m
, void *v
)
794 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
796 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
798 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
800 seq_path(m
, &mnt_path
, " \t\n\\");
802 show_type(m
, mnt
->mnt_sb
);
803 seq_puts(m
, __mnt_is_readonly(mnt
) ? " ro" : " rw");
804 err
= show_sb_opts(m
, mnt
->mnt_sb
);
807 show_mnt_opts(m
, mnt
);
808 if (mnt
->mnt_sb
->s_op
->show_options
)
809 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
810 seq_puts(m
, " 0 0\n");
815 const struct seq_operations mounts_op
= {
822 static int show_mountinfo(struct seq_file
*m
, void *v
)
824 struct proc_mounts
*p
= m
->private;
825 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
826 struct super_block
*sb
= mnt
->mnt_sb
;
827 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
828 struct path root
= p
->root
;
831 seq_printf(m
, "%i %i %u:%u ", mnt
->mnt_id
, mnt
->mnt_parent
->mnt_id
,
832 MAJOR(sb
->s_dev
), MINOR(sb
->s_dev
));
833 seq_dentry(m
, mnt
->mnt_root
, " \t\n\\");
835 seq_path_root(m
, &mnt_path
, &root
, " \t\n\\");
836 if (root
.mnt
!= p
->root
.mnt
|| root
.dentry
!= p
->root
.dentry
) {
838 * Mountpoint is outside root, discard that one. Ugly,
839 * but less so than trying to do that in iterator in a
840 * race-free way (due to renames).
844 seq_puts(m
, mnt
->mnt_flags
& MNT_READONLY
? " ro" : " rw");
845 show_mnt_opts(m
, mnt
);
847 /* Tagged fields ("foo:X" or "bar") */
848 if (IS_MNT_SHARED(mnt
))
849 seq_printf(m
, " shared:%i", mnt
->mnt_group_id
);
850 if (IS_MNT_SLAVE(mnt
)) {
851 int master
= mnt
->mnt_master
->mnt_group_id
;
852 int dom
= get_dominating_id(mnt
, &p
->root
);
853 seq_printf(m
, " master:%i", master
);
854 if (dom
&& dom
!= master
)
855 seq_printf(m
, " propagate_from:%i", dom
);
857 if (IS_MNT_UNBINDABLE(mnt
))
858 seq_puts(m
, " unbindable");
860 /* Filesystem specific data */
864 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
865 seq_puts(m
, sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
866 err
= show_sb_opts(m
, sb
);
869 if (sb
->s_op
->show_options
)
870 err
= sb
->s_op
->show_options(m
, mnt
);
876 const struct seq_operations mountinfo_op
= {
880 .show
= show_mountinfo
,
883 static int show_vfsstat(struct seq_file
*m
, void *v
)
885 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
886 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
890 if (mnt
->mnt_devname
) {
891 seq_puts(m
, "device ");
892 mangle(m
, mnt
->mnt_devname
);
894 seq_puts(m
, "no device");
897 seq_puts(m
, " mounted on ");
898 seq_path(m
, &mnt_path
, " \t\n\\");
901 /* file system type */
902 seq_puts(m
, "with fstype ");
903 show_type(m
, mnt
->mnt_sb
);
905 /* optional statistics */
906 if (mnt
->mnt_sb
->s_op
->show_stats
) {
908 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
915 const struct seq_operations mountstats_op
= {
919 .show
= show_vfsstat
,
921 #endif /* CONFIG_PROC_FS */
924 * may_umount_tree - check if a mount tree is busy
925 * @mnt: root of mount tree
927 * This is called to check if a tree of mounts has any
928 * open files, pwds, chroots or sub mounts that are
931 int may_umount_tree(struct vfsmount
*mnt
)
934 int minimum_refs
= 0;
937 spin_lock(&vfsmount_lock
);
938 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
939 actual_refs
+= atomic_read(&p
->mnt_count
);
942 spin_unlock(&vfsmount_lock
);
944 if (actual_refs
> minimum_refs
)
950 EXPORT_SYMBOL(may_umount_tree
);
953 * may_umount - check if a mount point is busy
954 * @mnt: root of mount
956 * This is called to check if a mount point has any
957 * open files, pwds, chroots or sub mounts. If the
958 * mount has sub mounts this will return busy
959 * regardless of whether the sub mounts are busy.
961 * Doesn't take quota and stuff into account. IOW, in some cases it will
962 * give false negatives. The main reason why it's here is that we need
963 * a non-destructive way to look for easily umountable filesystems.
965 int may_umount(struct vfsmount
*mnt
)
968 spin_lock(&vfsmount_lock
);
969 if (propagate_mount_busy(mnt
, 2))
971 spin_unlock(&vfsmount_lock
);
975 EXPORT_SYMBOL(may_umount
);
977 void release_mounts(struct list_head
*head
)
979 struct vfsmount
*mnt
;
980 while (!list_empty(head
)) {
981 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
982 list_del_init(&mnt
->mnt_hash
);
983 if (mnt
->mnt_parent
!= mnt
) {
984 struct dentry
*dentry
;
986 spin_lock(&vfsmount_lock
);
987 dentry
= mnt
->mnt_mountpoint
;
989 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
990 mnt
->mnt_parent
= mnt
;
992 spin_unlock(&vfsmount_lock
);
1000 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
1004 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
1005 list_move(&p
->mnt_hash
, kill
);
1008 propagate_umount(kill
);
1010 list_for_each_entry(p
, kill
, mnt_hash
) {
1011 list_del_init(&p
->mnt_expire
);
1012 list_del_init(&p
->mnt_list
);
1013 __touch_mnt_namespace(p
->mnt_ns
);
1015 list_del_init(&p
->mnt_child
);
1016 if (p
->mnt_parent
!= p
) {
1017 p
->mnt_parent
->mnt_ghosts
++;
1018 p
->mnt_mountpoint
->d_mounted
--;
1020 change_mnt_propagation(p
, MS_PRIVATE
);
1024 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
);
1026 static int do_umount(struct vfsmount
*mnt
, int flags
)
1028 struct super_block
*sb
= mnt
->mnt_sb
;
1030 LIST_HEAD(umount_list
);
1032 retval
= security_sb_umount(mnt
, flags
);
1037 * Allow userspace to request a mountpoint be expired rather than
1038 * unmounting unconditionally. Unmount only happens if:
1039 * (1) the mark is already set (the mark is cleared by mntput())
1040 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1042 if (flags
& MNT_EXPIRE
) {
1043 if (mnt
== current
->fs
->root
.mnt
||
1044 flags
& (MNT_FORCE
| MNT_DETACH
))
1047 if (atomic_read(&mnt
->mnt_count
) != 2)
1050 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
1055 * If we may have to abort operations to get out of this
1056 * mount, and they will themselves hold resources we must
1057 * allow the fs to do things. In the Unix tradition of
1058 * 'Gee thats tricky lets do it in userspace' the umount_begin
1059 * might fail to complete on the first run through as other tasks
1060 * must return, and the like. Thats for the mount program to worry
1061 * about for the moment.
1064 if (flags
& MNT_FORCE
&& sb
->s_op
->umount_begin
) {
1065 sb
->s_op
->umount_begin(sb
);
1069 * No sense to grab the lock for this test, but test itself looks
1070 * somewhat bogus. Suggestions for better replacement?
1071 * Ho-hum... In principle, we might treat that as umount + switch
1072 * to rootfs. GC would eventually take care of the old vfsmount.
1073 * Actually it makes sense, especially if rootfs would contain a
1074 * /reboot - static binary that would close all descriptors and
1075 * call reboot(9). Then init(8) could umount root and exec /reboot.
1077 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
1079 * Special case for "unmounting" root ...
1080 * we just try to remount it readonly.
1082 down_write(&sb
->s_umount
);
1083 if (!(sb
->s_flags
& MS_RDONLY
))
1084 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
1085 up_write(&sb
->s_umount
);
1089 down_write(&namespace_sem
);
1090 spin_lock(&vfsmount_lock
);
1093 if (!(flags
& MNT_DETACH
))
1094 shrink_submounts(mnt
, &umount_list
);
1097 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
1098 if (!list_empty(&mnt
->mnt_list
))
1099 umount_tree(mnt
, 1, &umount_list
);
1102 spin_unlock(&vfsmount_lock
);
1104 security_sb_umount_busy(mnt
);
1105 up_write(&namespace_sem
);
1106 release_mounts(&umount_list
);
1111 * Now umount can handle mount points as well as block devices.
1112 * This is important for filesystems which use unnamed block devices.
1114 * We now support a flag for forced unmount like the other 'big iron'
1115 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1118 SYSCALL_DEFINE2(umount
, char __user
*, name
, int, flags
)
1123 retval
= user_path(name
, &path
);
1127 if (path
.dentry
!= path
.mnt
->mnt_root
)
1129 if (!check_mnt(path
.mnt
))
1133 if (!capable(CAP_SYS_ADMIN
))
1136 retval
= do_umount(path
.mnt
, flags
);
1138 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1140 mntput_no_expire(path
.mnt
);
1145 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1148 * The 2.0 compatible umount. No flags.
1150 SYSCALL_DEFINE1(oldumount
, char __user
*, name
)
1152 return sys_umount(name
, 0);
1157 static int mount_is_safe(struct path
*path
)
1159 if (capable(CAP_SYS_ADMIN
))
1163 if (S_ISLNK(path
->dentry
->d_inode
->i_mode
))
1165 if (path
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
1166 if (current_uid() != path
->dentry
->d_inode
->i_uid
)
1169 if (inode_permission(path
->dentry
->d_inode
, MAY_WRITE
))
1175 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
1178 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
1181 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
1184 res
= q
= clone_mnt(mnt
, dentry
, flag
);
1187 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
1190 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
1191 if (!is_subdir(r
->mnt_mountpoint
, dentry
))
1194 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
1195 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
1196 s
= skip_mnt_tree(s
);
1199 while (p
!= s
->mnt_parent
) {
1205 path
.dentry
= p
->mnt_mountpoint
;
1206 q
= clone_mnt(p
, p
->mnt_root
, flag
);
1209 spin_lock(&vfsmount_lock
);
1210 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
1211 attach_mnt(q
, &path
);
1212 spin_unlock(&vfsmount_lock
);
1218 LIST_HEAD(umount_list
);
1219 spin_lock(&vfsmount_lock
);
1220 umount_tree(res
, 0, &umount_list
);
1221 spin_unlock(&vfsmount_lock
);
1222 release_mounts(&umount_list
);
1227 struct vfsmount
*collect_mounts(struct path
*path
)
1229 struct vfsmount
*tree
;
1230 down_write(&namespace_sem
);
1231 tree
= copy_tree(path
->mnt
, path
->dentry
, CL_COPY_ALL
| CL_PRIVATE
);
1232 up_write(&namespace_sem
);
1236 void drop_collected_mounts(struct vfsmount
*mnt
)
1238 LIST_HEAD(umount_list
);
1239 down_write(&namespace_sem
);
1240 spin_lock(&vfsmount_lock
);
1241 umount_tree(mnt
, 0, &umount_list
);
1242 spin_unlock(&vfsmount_lock
);
1243 up_write(&namespace_sem
);
1244 release_mounts(&umount_list
);
1247 static void cleanup_group_ids(struct vfsmount
*mnt
, struct vfsmount
*end
)
1251 for (p
= mnt
; p
!= end
; p
= next_mnt(p
, mnt
)) {
1252 if (p
->mnt_group_id
&& !IS_MNT_SHARED(p
))
1253 mnt_release_group_id(p
);
1257 static int invent_group_ids(struct vfsmount
*mnt
, bool recurse
)
1261 for (p
= mnt
; p
; p
= recurse
? next_mnt(p
, mnt
) : NULL
) {
1262 if (!p
->mnt_group_id
&& !IS_MNT_SHARED(p
)) {
1263 int err
= mnt_alloc_group_id(p
);
1265 cleanup_group_ids(mnt
, p
);
1275 * @source_mnt : mount tree to be attached
1276 * @nd : place the mount tree @source_mnt is attached
1277 * @parent_nd : if non-null, detach the source_mnt from its parent and
1278 * store the parent mount and mountpoint dentry.
1279 * (done when source_mnt is moved)
1281 * NOTE: in the table below explains the semantics when a source mount
1282 * of a given type is attached to a destination mount of a given type.
1283 * ---------------------------------------------------------------------------
1284 * | BIND MOUNT OPERATION |
1285 * |**************************************************************************
1286 * | source-->| shared | private | slave | unbindable |
1290 * |**************************************************************************
1291 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1293 * |non-shared| shared (+) | private | slave (*) | invalid |
1294 * ***************************************************************************
1295 * A bind operation clones the source mount and mounts the clone on the
1296 * destination mount.
1298 * (++) the cloned mount is propagated to all the mounts in the propagation
1299 * tree of the destination mount and the cloned mount is added to
1300 * the peer group of the source mount.
1301 * (+) the cloned mount is created under the destination mount and is marked
1302 * as shared. The cloned mount is added to the peer group of the source
1304 * (+++) the mount is propagated to all the mounts in the propagation tree
1305 * of the destination mount and the cloned mount is made slave
1306 * of the same master as that of the source mount. The cloned mount
1307 * is marked as 'shared and slave'.
1308 * (*) the cloned mount is made a slave of the same master as that of the
1311 * ---------------------------------------------------------------------------
1312 * | MOVE MOUNT OPERATION |
1313 * |**************************************************************************
1314 * | source-->| shared | private | slave | unbindable |
1318 * |**************************************************************************
1319 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1321 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1322 * ***************************************************************************
1324 * (+) the mount is moved to the destination. And is then propagated to
1325 * all the mounts in the propagation tree of the destination mount.
1326 * (+*) the mount is moved to the destination.
1327 * (+++) the mount is moved to the destination and is then propagated to
1328 * all the mounts belonging to the destination mount's propagation tree.
1329 * the mount is marked as 'shared and slave'.
1330 * (*) the mount continues to be a slave at the new location.
1332 * if the source mount is a tree, the operations explained above is
1333 * applied to each mount in the tree.
1334 * Must be called without spinlocks held, since this function can sleep
1337 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
1338 struct path
*path
, struct path
*parent_path
)
1340 LIST_HEAD(tree_list
);
1341 struct vfsmount
*dest_mnt
= path
->mnt
;
1342 struct dentry
*dest_dentry
= path
->dentry
;
1343 struct vfsmount
*child
, *p
;
1346 if (IS_MNT_SHARED(dest_mnt
)) {
1347 err
= invent_group_ids(source_mnt
, true);
1351 err
= propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
);
1353 goto out_cleanup_ids
;
1355 if (IS_MNT_SHARED(dest_mnt
)) {
1356 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
1360 spin_lock(&vfsmount_lock
);
1362 detach_mnt(source_mnt
, parent_path
);
1363 attach_mnt(source_mnt
, path
);
1364 touch_mnt_namespace(parent_path
->mnt
->mnt_ns
);
1366 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
1367 commit_tree(source_mnt
);
1370 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
1371 list_del_init(&child
->mnt_hash
);
1374 spin_unlock(&vfsmount_lock
);
1378 if (IS_MNT_SHARED(dest_mnt
))
1379 cleanup_group_ids(source_mnt
, NULL
);
1384 static int graft_tree(struct vfsmount
*mnt
, struct path
*path
)
1387 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
1390 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1391 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
1395 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1396 if (IS_DEADDIR(path
->dentry
->d_inode
))
1399 err
= security_sb_check_sb(mnt
, path
);
1404 if (!d_unlinked(path
->dentry
))
1405 err
= attach_recursive_mnt(mnt
, path
, NULL
);
1407 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1409 security_sb_post_addmount(mnt
, path
);
1414 * recursively change the type of the mountpoint.
1416 static int do_change_type(struct path
*path
, int flag
)
1418 struct vfsmount
*m
, *mnt
= path
->mnt
;
1419 int recurse
= flag
& MS_REC
;
1420 int type
= flag
& ~MS_REC
;
1423 if (!capable(CAP_SYS_ADMIN
))
1426 if (path
->dentry
!= path
->mnt
->mnt_root
)
1429 down_write(&namespace_sem
);
1430 if (type
== MS_SHARED
) {
1431 err
= invent_group_ids(mnt
, recurse
);
1436 spin_lock(&vfsmount_lock
);
1437 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1438 change_mnt_propagation(m
, type
);
1439 spin_unlock(&vfsmount_lock
);
1442 up_write(&namespace_sem
);
1447 * do loopback mount.
1449 static int do_loopback(struct path
*path
, char *old_name
,
1452 struct path old_path
;
1453 struct vfsmount
*mnt
= NULL
;
1454 int err
= mount_is_safe(path
);
1457 if (!old_name
|| !*old_name
)
1459 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1463 down_write(&namespace_sem
);
1465 if (IS_MNT_UNBINDABLE(old_path
.mnt
))
1468 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1473 mnt
= copy_tree(old_path
.mnt
, old_path
.dentry
, 0);
1475 mnt
= clone_mnt(old_path
.mnt
, old_path
.dentry
, 0);
1480 err
= graft_tree(mnt
, path
);
1482 LIST_HEAD(umount_list
);
1483 spin_lock(&vfsmount_lock
);
1484 umount_tree(mnt
, 0, &umount_list
);
1485 spin_unlock(&vfsmount_lock
);
1486 release_mounts(&umount_list
);
1490 up_write(&namespace_sem
);
1491 path_put(&old_path
);
1495 static int change_mount_flags(struct vfsmount
*mnt
, int ms_flags
)
1498 int readonly_request
= 0;
1500 if (ms_flags
& MS_RDONLY
)
1501 readonly_request
= 1;
1502 if (readonly_request
== __mnt_is_readonly(mnt
))
1505 if (readonly_request
)
1506 error
= mnt_make_readonly(mnt
);
1508 __mnt_unmake_readonly(mnt
);
1513 * change filesystem flags. dir should be a physical root of filesystem.
1514 * If you've mounted a non-root directory somewhere and want to do remount
1515 * on it - tough luck.
1517 static int do_remount(struct path
*path
, int flags
, int mnt_flags
,
1521 struct super_block
*sb
= path
->mnt
->mnt_sb
;
1523 if (!capable(CAP_SYS_ADMIN
))
1526 if (!check_mnt(path
->mnt
))
1529 if (path
->dentry
!= path
->mnt
->mnt_root
)
1532 down_write(&sb
->s_umount
);
1533 if (flags
& MS_BIND
)
1534 err
= change_mount_flags(path
->mnt
, flags
);
1536 err
= do_remount_sb(sb
, flags
, data
, 0);
1538 path
->mnt
->mnt_flags
= mnt_flags
;
1539 up_write(&sb
->s_umount
);
1541 security_sb_post_remount(path
->mnt
, flags
, data
);
1543 spin_lock(&vfsmount_lock
);
1544 touch_mnt_namespace(path
->mnt
->mnt_ns
);
1545 spin_unlock(&vfsmount_lock
);
1550 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1553 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1554 if (IS_MNT_UNBINDABLE(p
))
1560 static int do_move_mount(struct path
*path
, char *old_name
)
1562 struct path old_path
, parent_path
;
1565 if (!capable(CAP_SYS_ADMIN
))
1567 if (!old_name
|| !*old_name
)
1569 err
= kern_path(old_name
, LOOKUP_FOLLOW
, &old_path
);
1573 down_write(&namespace_sem
);
1574 while (d_mountpoint(path
->dentry
) &&
1578 if (!check_mnt(path
->mnt
) || !check_mnt(old_path
.mnt
))
1582 mutex_lock(&path
->dentry
->d_inode
->i_mutex
);
1583 if (IS_DEADDIR(path
->dentry
->d_inode
))
1586 if (d_unlinked(path
->dentry
))
1590 if (old_path
.dentry
!= old_path
.mnt
->mnt_root
)
1593 if (old_path
.mnt
== old_path
.mnt
->mnt_parent
)
1596 if (S_ISDIR(path
->dentry
->d_inode
->i_mode
) !=
1597 S_ISDIR(old_path
.dentry
->d_inode
->i_mode
))
1600 * Don't move a mount residing in a shared parent.
1602 if (old_path
.mnt
->mnt_parent
&&
1603 IS_MNT_SHARED(old_path
.mnt
->mnt_parent
))
1606 * Don't move a mount tree containing unbindable mounts to a destination
1607 * mount which is shared.
1609 if (IS_MNT_SHARED(path
->mnt
) &&
1610 tree_contains_unbindable(old_path
.mnt
))
1613 for (p
= path
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1614 if (p
== old_path
.mnt
)
1617 err
= attach_recursive_mnt(old_path
.mnt
, path
, &parent_path
);
1621 /* if the mount is moved, it should no longer be expire
1623 list_del_init(&old_path
.mnt
->mnt_expire
);
1625 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
1627 up_write(&namespace_sem
);
1629 path_put(&parent_path
);
1630 path_put(&old_path
);
1635 * create a new mount for userspace and request it to be added into the
1638 static int do_new_mount(struct path
*path
, char *type
, int flags
,
1639 int mnt_flags
, char *name
, void *data
)
1641 struct vfsmount
*mnt
;
1646 /* we need capabilities... */
1647 if (!capable(CAP_SYS_ADMIN
))
1651 mnt
= do_kern_mount(type
, flags
, name
, data
);
1654 return PTR_ERR(mnt
);
1656 return do_add_mount(mnt
, path
, mnt_flags
, NULL
);
1660 * add a mount into a namespace's mount tree
1661 * - provide the option of adding the new mount to an expiration list
1663 int do_add_mount(struct vfsmount
*newmnt
, struct path
*path
,
1664 int mnt_flags
, struct list_head
*fslist
)
1668 down_write(&namespace_sem
);
1669 /* Something was mounted here while we slept */
1670 while (d_mountpoint(path
->dentry
) &&
1674 if (!(mnt_flags
& MNT_SHRINKABLE
) && !check_mnt(path
->mnt
))
1677 /* Refuse the same filesystem on the same mount point */
1679 if (path
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1680 path
->mnt
->mnt_root
== path
->dentry
)
1684 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1687 newmnt
->mnt_flags
= mnt_flags
;
1688 if ((err
= graft_tree(newmnt
, path
)))
1691 if (fslist
) /* add to the specified expiration list */
1692 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1694 up_write(&namespace_sem
);
1698 up_write(&namespace_sem
);
1703 EXPORT_SYMBOL_GPL(do_add_mount
);
1706 * process a list of expirable mountpoints with the intent of discarding any
1707 * mountpoints that aren't in use and haven't been touched since last we came
1710 void mark_mounts_for_expiry(struct list_head
*mounts
)
1712 struct vfsmount
*mnt
, *next
;
1713 LIST_HEAD(graveyard
);
1716 if (list_empty(mounts
))
1719 down_write(&namespace_sem
);
1720 spin_lock(&vfsmount_lock
);
1722 /* extract from the expiration list every vfsmount that matches the
1723 * following criteria:
1724 * - only referenced by its parent vfsmount
1725 * - still marked for expiry (marked on the last call here; marks are
1726 * cleared by mntput())
1728 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1729 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1730 propagate_mount_busy(mnt
, 1))
1732 list_move(&mnt
->mnt_expire
, &graveyard
);
1734 while (!list_empty(&graveyard
)) {
1735 mnt
= list_first_entry(&graveyard
, struct vfsmount
, mnt_expire
);
1736 touch_mnt_namespace(mnt
->mnt_ns
);
1737 umount_tree(mnt
, 1, &umounts
);
1739 spin_unlock(&vfsmount_lock
);
1740 up_write(&namespace_sem
);
1742 release_mounts(&umounts
);
1745 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1748 * Ripoff of 'select_parent()'
1750 * search the list of submounts for a given mountpoint, and move any
1751 * shrinkable submounts to the 'graveyard' list.
1753 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1755 struct vfsmount
*this_parent
= parent
;
1756 struct list_head
*next
;
1760 next
= this_parent
->mnt_mounts
.next
;
1762 while (next
!= &this_parent
->mnt_mounts
) {
1763 struct list_head
*tmp
= next
;
1764 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1767 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1770 * Descend a level if the d_mounts list is non-empty.
1772 if (!list_empty(&mnt
->mnt_mounts
)) {
1777 if (!propagate_mount_busy(mnt
, 1)) {
1778 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1783 * All done at this level ... ascend and resume the search
1785 if (this_parent
!= parent
) {
1786 next
= this_parent
->mnt_child
.next
;
1787 this_parent
= this_parent
->mnt_parent
;
1794 * process a list of expirable mountpoints with the intent of discarding any
1795 * submounts of a specific parent mountpoint
1797 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
)
1799 LIST_HEAD(graveyard
);
1802 /* extract submounts of 'mountpoint' from the expiration list */
1803 while (select_submounts(mnt
, &graveyard
)) {
1804 while (!list_empty(&graveyard
)) {
1805 m
= list_first_entry(&graveyard
, struct vfsmount
,
1807 touch_mnt_namespace(m
->mnt_ns
);
1808 umount_tree(m
, 1, umounts
);
1814 * Some copy_from_user() implementations do not return the exact number of
1815 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1816 * Note that this function differs from copy_from_user() in that it will oops
1817 * on bad values of `to', rather than returning a short copy.
1819 static long exact_copy_from_user(void *to
, const void __user
* from
,
1823 const char __user
*f
= from
;
1826 if (!access_ok(VERIFY_READ
, from
, n
))
1830 if (__get_user(c
, f
)) {
1841 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1851 if (!(page
= __get_free_page(GFP_KERNEL
)))
1854 /* We only care that *some* data at the address the user
1855 * gave us is valid. Just in case, we'll zero
1856 * the remainder of the page.
1858 /* copy_from_user cannot cross TASK_SIZE ! */
1859 size
= TASK_SIZE
- (unsigned long)data
;
1860 if (size
> PAGE_SIZE
)
1863 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1869 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1874 int copy_mount_string(const void __user
*data
, char **where
)
1883 tmp
= strndup_user(data
, PAGE_SIZE
);
1885 return PTR_ERR(tmp
);
1892 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1893 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1895 * data is a (void *) that can point to any structure up to
1896 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1897 * information (or be NULL).
1899 * Pre-0.97 versions of mount() didn't have a flags word.
1900 * When the flags word was introduced its top half was required
1901 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1902 * Therefore, if this magic number is present, it carries no information
1903 * and must be discarded.
1905 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1906 unsigned long flags
, void *data_page
)
1913 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1914 flags
&= ~MS_MGC_MSK
;
1916 /* Basic sanity checks */
1918 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1922 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1924 /* Default to relatime unless overriden */
1925 if (!(flags
& MS_NOATIME
))
1926 mnt_flags
|= MNT_RELATIME
;
1928 /* Separate the per-mountpoint flags */
1929 if (flags
& MS_NOSUID
)
1930 mnt_flags
|= MNT_NOSUID
;
1931 if (flags
& MS_NODEV
)
1932 mnt_flags
|= MNT_NODEV
;
1933 if (flags
& MS_NOEXEC
)
1934 mnt_flags
|= MNT_NOEXEC
;
1935 if (flags
& MS_NOATIME
)
1936 mnt_flags
|= MNT_NOATIME
;
1937 if (flags
& MS_NODIRATIME
)
1938 mnt_flags
|= MNT_NODIRATIME
;
1939 if (flags
& MS_STRICTATIME
)
1940 mnt_flags
&= ~(MNT_RELATIME
| MNT_NOATIME
);
1941 if (flags
& MS_RDONLY
)
1942 mnt_flags
|= MNT_READONLY
;
1944 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1945 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
|
1948 /* ... and get the mountpoint */
1949 retval
= kern_path(dir_name
, LOOKUP_FOLLOW
, &path
);
1953 retval
= security_sb_mount(dev_name
, &path
,
1954 type_page
, flags
, data_page
);
1958 if (flags
& MS_REMOUNT
)
1959 retval
= do_remount(&path
, flags
& ~MS_REMOUNT
, mnt_flags
,
1961 else if (flags
& MS_BIND
)
1962 retval
= do_loopback(&path
, dev_name
, flags
& MS_REC
);
1963 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1964 retval
= do_change_type(&path
, flags
);
1965 else if (flags
& MS_MOVE
)
1966 retval
= do_move_mount(&path
, dev_name
);
1968 retval
= do_new_mount(&path
, type_page
, flags
, mnt_flags
,
1969 dev_name
, data_page
);
1975 static struct mnt_namespace
*alloc_mnt_ns(void)
1977 struct mnt_namespace
*new_ns
;
1979 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1981 return ERR_PTR(-ENOMEM
);
1982 atomic_set(&new_ns
->count
, 1);
1983 new_ns
->root
= NULL
;
1984 INIT_LIST_HEAD(&new_ns
->list
);
1985 init_waitqueue_head(&new_ns
->poll
);
1991 * Allocate a new namespace structure and populate it with contents
1992 * copied from the namespace of the passed in task structure.
1994 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1995 struct fs_struct
*fs
)
1997 struct mnt_namespace
*new_ns
;
1998 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
;
1999 struct vfsmount
*p
, *q
;
2001 new_ns
= alloc_mnt_ns();
2005 down_write(&namespace_sem
);
2006 /* First pass: copy the tree topology */
2007 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
2008 CL_COPY_ALL
| CL_EXPIRE
);
2009 if (!new_ns
->root
) {
2010 up_write(&namespace_sem
);
2012 return ERR_PTR(-ENOMEM
);
2014 spin_lock(&vfsmount_lock
);
2015 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2016 spin_unlock(&vfsmount_lock
);
2019 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2020 * as belonging to new namespace. We have already acquired a private
2021 * fs_struct, so tsk->fs->lock is not needed.
2028 if (p
== fs
->root
.mnt
) {
2030 fs
->root
.mnt
= mntget(q
);
2032 if (p
== fs
->pwd
.mnt
) {
2034 fs
->pwd
.mnt
= mntget(q
);
2037 p
= next_mnt(p
, mnt_ns
->root
);
2038 q
= next_mnt(q
, new_ns
->root
);
2040 up_write(&namespace_sem
);
2050 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
2051 struct fs_struct
*new_fs
)
2053 struct mnt_namespace
*new_ns
;
2058 if (!(flags
& CLONE_NEWNS
))
2061 new_ns
= dup_mnt_ns(ns
, new_fs
);
2068 * create_mnt_ns - creates a private namespace and adds a root filesystem
2069 * @mnt: pointer to the new root filesystem mountpoint
2071 struct mnt_namespace
*create_mnt_ns(struct vfsmount
*mnt
)
2073 struct mnt_namespace
*new_ns
;
2075 new_ns
= alloc_mnt_ns();
2076 if (!IS_ERR(new_ns
)) {
2077 mnt
->mnt_ns
= new_ns
;
2079 list_add(&new_ns
->list
, &new_ns
->root
->mnt_list
);
2083 EXPORT_SYMBOL(create_mnt_ns
);
2085 SYSCALL_DEFINE5(mount
, char __user
*, dev_name
, char __user
*, dir_name
,
2086 char __user
*, type
, unsigned long, flags
, void __user
*, data
)
2092 unsigned long data_page
;
2094 ret
= copy_mount_string(type
, &kernel_type
);
2098 kernel_dir
= getname(dir_name
);
2099 if (IS_ERR(kernel_dir
)) {
2100 ret
= PTR_ERR(kernel_dir
);
2104 ret
= copy_mount_string(dev_name
, &kernel_dev
);
2108 ret
= copy_mount_options(data
, &data_page
);
2112 ret
= do_mount(kernel_dev
, kernel_dir
, kernel_type
, flags
,
2113 (void *) data_page
);
2115 free_page(data_page
);
2119 putname(kernel_dir
);
2127 * pivot_root Semantics:
2128 * Moves the root file system of the current process to the directory put_old,
2129 * makes new_root as the new root file system of the current process, and sets
2130 * root/cwd of all processes which had them on the current root to new_root.
2133 * The new_root and put_old must be directories, and must not be on the
2134 * same file system as the current process root. The put_old must be
2135 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2136 * pointed to by put_old must yield the same directory as new_root. No other
2137 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2139 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2140 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2141 * in this situation.
2144 * - we don't move root/cwd if they are not at the root (reason: if something
2145 * cared enough to change them, it's probably wrong to force them elsewhere)
2146 * - it's okay to pick a root that isn't the root of a file system, e.g.
2147 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2148 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2151 SYSCALL_DEFINE2(pivot_root
, const char __user
*, new_root
,
2152 const char __user
*, put_old
)
2154 struct vfsmount
*tmp
;
2155 struct path
new, old
, parent_path
, root_parent
, root
;
2158 if (!capable(CAP_SYS_ADMIN
))
2161 error
= user_path_dir(new_root
, &new);
2165 if (!check_mnt(new.mnt
))
2168 error
= user_path_dir(put_old
, &old
);
2172 error
= security_sb_pivotroot(&old
, &new);
2178 read_lock(¤t
->fs
->lock
);
2179 root
= current
->fs
->root
;
2180 path_get(¤t
->fs
->root
);
2181 read_unlock(¤t
->fs
->lock
);
2182 down_write(&namespace_sem
);
2183 mutex_lock(&old
.dentry
->d_inode
->i_mutex
);
2185 if (IS_MNT_SHARED(old
.mnt
) ||
2186 IS_MNT_SHARED(new.mnt
->mnt_parent
) ||
2187 IS_MNT_SHARED(root
.mnt
->mnt_parent
))
2189 if (!check_mnt(root
.mnt
))
2192 if (IS_DEADDIR(new.dentry
->d_inode
))
2194 if (d_unlinked(new.dentry
))
2196 if (d_unlinked(old
.dentry
))
2199 if (new.mnt
== root
.mnt
||
2200 old
.mnt
== root
.mnt
)
2201 goto out2
; /* loop, on the same file system */
2203 if (root
.mnt
->mnt_root
!= root
.dentry
)
2204 goto out2
; /* not a mountpoint */
2205 if (root
.mnt
->mnt_parent
== root
.mnt
)
2206 goto out2
; /* not attached */
2207 if (new.mnt
->mnt_root
!= new.dentry
)
2208 goto out2
; /* not a mountpoint */
2209 if (new.mnt
->mnt_parent
== new.mnt
)
2210 goto out2
; /* not attached */
2211 /* make sure we can reach put_old from new_root */
2213 spin_lock(&vfsmount_lock
);
2214 if (tmp
!= new.mnt
) {
2216 if (tmp
->mnt_parent
== tmp
)
2217 goto out3
; /* already mounted on put_old */
2218 if (tmp
->mnt_parent
== new.mnt
)
2220 tmp
= tmp
->mnt_parent
;
2222 if (!is_subdir(tmp
->mnt_mountpoint
, new.dentry
))
2224 } else if (!is_subdir(old
.dentry
, new.dentry
))
2226 detach_mnt(new.mnt
, &parent_path
);
2227 detach_mnt(root
.mnt
, &root_parent
);
2228 /* mount old root on put_old */
2229 attach_mnt(root
.mnt
, &old
);
2230 /* mount new_root on / */
2231 attach_mnt(new.mnt
, &root_parent
);
2232 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
2233 spin_unlock(&vfsmount_lock
);
2234 chroot_fs_refs(&root
, &new);
2235 security_sb_post_pivotroot(&root
, &new);
2237 path_put(&root_parent
);
2238 path_put(&parent_path
);
2240 mutex_unlock(&old
.dentry
->d_inode
->i_mutex
);
2241 up_write(&namespace_sem
);
2249 spin_unlock(&vfsmount_lock
);
2253 static void __init
init_mount_tree(void)
2255 struct vfsmount
*mnt
;
2256 struct mnt_namespace
*ns
;
2259 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
2261 panic("Can't create rootfs");
2262 ns
= create_mnt_ns(mnt
);
2264 panic("Can't allocate initial namespace");
2266 init_task
.nsproxy
->mnt_ns
= ns
;
2269 root
.mnt
= ns
->root
;
2270 root
.dentry
= ns
->root
->mnt_root
;
2272 set_fs_pwd(current
->fs
, &root
);
2273 set_fs_root(current
->fs
, &root
);
2276 void __init
mnt_init(void)
2281 init_rwsem(&namespace_sem
);
2283 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
2284 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
2286 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
2288 if (!mount_hashtable
)
2289 panic("Failed to allocate mount hash table\n");
2291 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
2293 for (u
= 0; u
< HASH_SIZE
; u
++)
2294 INIT_LIST_HEAD(&mount_hashtable
[u
]);
2298 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
2300 fs_kobj
= kobject_create_and_add("fs", NULL
);
2302 printk(KERN_WARNING
"%s: kobj create error\n", __func__
);
2307 void put_mnt_ns(struct mnt_namespace
*ns
)
2309 struct vfsmount
*root
;
2310 LIST_HEAD(umount_list
);
2312 if (!atomic_dec_and_lock(&ns
->count
, &vfsmount_lock
))
2316 spin_unlock(&vfsmount_lock
);
2317 down_write(&namespace_sem
);
2318 spin_lock(&vfsmount_lock
);
2319 umount_tree(root
, 0, &umount_list
);
2320 spin_unlock(&vfsmount_lock
);
2321 up_write(&namespace_sem
);
2322 release_mounts(&umount_list
);
2325 EXPORT_SYMBOL(put_mnt_ns
);