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/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.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/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <linux/log2.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
34 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35 #define HASH_SIZE (1UL << HASH_SHIFT)
37 /* spinlock for vfsmount related operations, inplace of dcache_lock */
38 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
42 static struct list_head
*mount_hashtable __read_mostly
;
43 static struct kmem_cache
*mnt_cache __read_mostly
;
44 static struct rw_semaphore namespace_sem
;
47 struct kobject
*fs_kobj
;
48 EXPORT_SYMBOL_GPL(fs_kobj
);
50 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
52 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
53 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
54 tmp
= tmp
+ (tmp
>> HASH_SHIFT
);
55 return tmp
& (HASH_SIZE
- 1);
58 struct vfsmount
*alloc_vfsmnt(const char *name
)
60 struct vfsmount
*mnt
= kmem_cache_zalloc(mnt_cache
, GFP_KERNEL
);
62 atomic_set(&mnt
->mnt_count
, 1);
63 INIT_LIST_HEAD(&mnt
->mnt_hash
);
64 INIT_LIST_HEAD(&mnt
->mnt_child
);
65 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
66 INIT_LIST_HEAD(&mnt
->mnt_list
);
67 INIT_LIST_HEAD(&mnt
->mnt_expire
);
68 INIT_LIST_HEAD(&mnt
->mnt_share
);
69 INIT_LIST_HEAD(&mnt
->mnt_slave_list
);
70 INIT_LIST_HEAD(&mnt
->mnt_slave
);
72 int size
= strlen(name
) + 1;
73 char *newname
= kmalloc(size
, GFP_KERNEL
);
75 memcpy(newname
, name
, size
);
76 mnt
->mnt_devname
= newname
;
83 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
86 mnt
->mnt_root
= dget(sb
->s_root
);
90 EXPORT_SYMBOL(simple_set_mnt
);
92 void free_vfsmnt(struct vfsmount
*mnt
)
94 kfree(mnt
->mnt_devname
);
95 kmem_cache_free(mnt_cache
, mnt
);
99 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
102 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
105 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
106 struct list_head
*tmp
= head
;
107 struct vfsmount
*p
, *found
= NULL
;
110 tmp
= dir
? tmp
->next
: tmp
->prev
;
114 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
115 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
127 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
129 struct vfsmount
*child_mnt
;
130 spin_lock(&vfsmount_lock
);
131 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
133 spin_unlock(&vfsmount_lock
);
137 static inline int check_mnt(struct vfsmount
*mnt
)
139 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
142 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
146 wake_up_interruptible(&ns
->poll
);
150 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
152 if (ns
&& ns
->event
!= event
) {
154 wake_up_interruptible(&ns
->poll
);
158 static void detach_mnt(struct vfsmount
*mnt
, struct path
*old_path
)
160 old_path
->dentry
= mnt
->mnt_mountpoint
;
161 old_path
->mnt
= mnt
->mnt_parent
;
162 mnt
->mnt_parent
= mnt
;
163 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
164 list_del_init(&mnt
->mnt_child
);
165 list_del_init(&mnt
->mnt_hash
);
166 old_path
->dentry
->d_mounted
--;
169 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
170 struct vfsmount
*child_mnt
)
172 child_mnt
->mnt_parent
= mntget(mnt
);
173 child_mnt
->mnt_mountpoint
= dget(dentry
);
177 static void attach_mnt(struct vfsmount
*mnt
, struct path
*path
)
179 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
180 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
181 hash(path
->mnt
, path
->dentry
));
182 list_add_tail(&mnt
->mnt_child
, &path
->mnt
->mnt_mounts
);
186 * the caller must hold vfsmount_lock
188 static void commit_tree(struct vfsmount
*mnt
)
190 struct vfsmount
*parent
= mnt
->mnt_parent
;
193 struct mnt_namespace
*n
= parent
->mnt_ns
;
195 BUG_ON(parent
== mnt
);
197 list_add_tail(&head
, &mnt
->mnt_list
);
198 list_for_each_entry(m
, &head
, mnt_list
)
200 list_splice(&head
, n
->list
.prev
);
202 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
203 hash(parent
, mnt
->mnt_mountpoint
));
204 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
205 touch_mnt_namespace(n
);
208 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
210 struct list_head
*next
= p
->mnt_mounts
.next
;
211 if (next
== &p
->mnt_mounts
) {
215 next
= p
->mnt_child
.next
;
216 if (next
!= &p
->mnt_parent
->mnt_mounts
)
221 return list_entry(next
, struct vfsmount
, mnt_child
);
224 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
226 struct list_head
*prev
= p
->mnt_mounts
.prev
;
227 while (prev
!= &p
->mnt_mounts
) {
228 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
229 prev
= p
->mnt_mounts
.prev
;
234 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
237 struct super_block
*sb
= old
->mnt_sb
;
238 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
241 mnt
->mnt_flags
= old
->mnt_flags
;
242 atomic_inc(&sb
->s_active
);
244 mnt
->mnt_root
= dget(root
);
245 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
246 mnt
->mnt_parent
= mnt
;
248 if (flag
& CL_SLAVE
) {
249 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
250 mnt
->mnt_master
= old
;
251 CLEAR_MNT_SHARED(mnt
);
252 } else if (!(flag
& CL_PRIVATE
)) {
253 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
254 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
255 if (IS_MNT_SLAVE(old
))
256 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
257 mnt
->mnt_master
= old
->mnt_master
;
259 if (flag
& CL_MAKE_SHARED
)
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag
& CL_EXPIRE
) {
265 spin_lock(&vfsmount_lock
);
266 if (!list_empty(&old
->mnt_expire
))
267 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
268 spin_unlock(&vfsmount_lock
);
274 static inline void __mntput(struct vfsmount
*mnt
)
276 struct super_block
*sb
= mnt
->mnt_sb
;
279 deactivate_super(sb
);
282 void mntput_no_expire(struct vfsmount
*mnt
)
285 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
286 if (likely(!mnt
->mnt_pinned
)) {
287 spin_unlock(&vfsmount_lock
);
291 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
293 spin_unlock(&vfsmount_lock
);
294 acct_auto_close_mnt(mnt
);
295 security_sb_umount_close(mnt
);
300 EXPORT_SYMBOL(mntput_no_expire
);
302 void mnt_pin(struct vfsmount
*mnt
)
304 spin_lock(&vfsmount_lock
);
306 spin_unlock(&vfsmount_lock
);
309 EXPORT_SYMBOL(mnt_pin
);
311 void mnt_unpin(struct vfsmount
*mnt
)
313 spin_lock(&vfsmount_lock
);
314 if (mnt
->mnt_pinned
) {
315 atomic_inc(&mnt
->mnt_count
);
318 spin_unlock(&vfsmount_lock
);
321 EXPORT_SYMBOL(mnt_unpin
);
323 static inline void mangle(struct seq_file
*m
, const char *s
)
325 seq_escape(m
, s
, " \t\n\\");
329 * Simple .show_options callback for filesystems which don't want to
330 * implement more complex mount option showing.
332 * See also save_mount_options().
334 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
336 const char *options
= mnt
->mnt_sb
->s_options
;
338 if (options
!= NULL
&& options
[0]) {
345 EXPORT_SYMBOL(generic_show_options
);
348 * If filesystem uses generic_show_options(), this function should be
349 * called from the fill_super() callback.
351 * The .remount_fs callback usually needs to be handled in a special
352 * way, to make sure, that previous options are not overwritten if the
355 * Also note, that if the filesystem's .remount_fs function doesn't
356 * reset all options to their default value, but changes only newly
357 * given options, then the displayed options will not reflect reality
360 void save_mount_options(struct super_block
*sb
, char *options
)
362 kfree(sb
->s_options
);
363 sb
->s_options
= kstrdup(options
, GFP_KERNEL
);
365 EXPORT_SYMBOL(save_mount_options
);
368 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
370 struct mnt_namespace
*n
= m
->private;
372 down_read(&namespace_sem
);
373 return seq_list_start(&n
->list
, *pos
);
376 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
378 struct mnt_namespace
*n
= m
->private;
380 return seq_list_next(v
, &n
->list
, pos
);
383 static void m_stop(struct seq_file
*m
, void *v
)
385 up_read(&namespace_sem
);
388 static int show_vfsmnt(struct seq_file
*m
, void *v
)
390 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
392 static struct proc_fs_info
{
396 { MS_SYNCHRONOUS
, ",sync" },
397 { MS_DIRSYNC
, ",dirsync" },
398 { MS_MANDLOCK
, ",mand" },
401 static struct proc_fs_info mnt_info
[] = {
402 { MNT_NOSUID
, ",nosuid" },
403 { MNT_NODEV
, ",nodev" },
404 { MNT_NOEXEC
, ",noexec" },
405 { MNT_NOATIME
, ",noatime" },
406 { MNT_NODIRATIME
, ",nodiratime" },
407 { MNT_RELATIME
, ",relatime" },
410 struct proc_fs_info
*fs_infop
;
411 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
413 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
415 seq_path(m
, &mnt_path
, " \t\n\\");
417 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
418 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
420 mangle(m
, mnt
->mnt_sb
->s_subtype
);
422 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
423 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
424 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
425 seq_puts(m
, fs_infop
->str
);
427 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
428 if (mnt
->mnt_flags
& fs_infop
->flag
)
429 seq_puts(m
, fs_infop
->str
);
431 if (mnt
->mnt_sb
->s_op
->show_options
)
432 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
433 seq_puts(m
, " 0 0\n");
437 struct seq_operations mounts_op
= {
444 static int show_vfsstat(struct seq_file
*m
, void *v
)
446 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
447 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
451 if (mnt
->mnt_devname
) {
452 seq_puts(m
, "device ");
453 mangle(m
, mnt
->mnt_devname
);
455 seq_puts(m
, "no device");
458 seq_puts(m
, " mounted on ");
459 seq_path(m
, &mnt_path
, " \t\n\\");
462 /* file system type */
463 seq_puts(m
, "with fstype ");
464 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
466 /* optional statistics */
467 if (mnt
->mnt_sb
->s_op
->show_stats
) {
469 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
476 struct seq_operations mountstats_op
= {
480 .show
= show_vfsstat
,
484 * may_umount_tree - check if a mount tree is busy
485 * @mnt: root of mount tree
487 * This is called to check if a tree of mounts has any
488 * open files, pwds, chroots or sub mounts that are
491 int may_umount_tree(struct vfsmount
*mnt
)
494 int minimum_refs
= 0;
497 spin_lock(&vfsmount_lock
);
498 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
499 actual_refs
+= atomic_read(&p
->mnt_count
);
502 spin_unlock(&vfsmount_lock
);
504 if (actual_refs
> minimum_refs
)
510 EXPORT_SYMBOL(may_umount_tree
);
513 * may_umount - check if a mount point is busy
514 * @mnt: root of mount
516 * This is called to check if a mount point has any
517 * open files, pwds, chroots or sub mounts. If the
518 * mount has sub mounts this will return busy
519 * regardless of whether the sub mounts are busy.
521 * Doesn't take quota and stuff into account. IOW, in some cases it will
522 * give false negatives. The main reason why it's here is that we need
523 * a non-destructive way to look for easily umountable filesystems.
525 int may_umount(struct vfsmount
*mnt
)
528 spin_lock(&vfsmount_lock
);
529 if (propagate_mount_busy(mnt
, 2))
531 spin_unlock(&vfsmount_lock
);
535 EXPORT_SYMBOL(may_umount
);
537 void release_mounts(struct list_head
*head
)
539 struct vfsmount
*mnt
;
540 while (!list_empty(head
)) {
541 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
542 list_del_init(&mnt
->mnt_hash
);
543 if (mnt
->mnt_parent
!= mnt
) {
544 struct dentry
*dentry
;
546 spin_lock(&vfsmount_lock
);
547 dentry
= mnt
->mnt_mountpoint
;
549 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
550 mnt
->mnt_parent
= mnt
;
552 spin_unlock(&vfsmount_lock
);
560 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
564 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
565 list_move(&p
->mnt_hash
, kill
);
568 propagate_umount(kill
);
570 list_for_each_entry(p
, kill
, mnt_hash
) {
571 list_del_init(&p
->mnt_expire
);
572 list_del_init(&p
->mnt_list
);
573 __touch_mnt_namespace(p
->mnt_ns
);
575 list_del_init(&p
->mnt_child
);
576 if (p
->mnt_parent
!= p
) {
577 p
->mnt_parent
->mnt_ghosts
++;
578 p
->mnt_mountpoint
->d_mounted
--;
580 change_mnt_propagation(p
, MS_PRIVATE
);
584 static int do_umount(struct vfsmount
*mnt
, int flags
)
586 struct super_block
*sb
= mnt
->mnt_sb
;
588 LIST_HEAD(umount_list
);
590 retval
= security_sb_umount(mnt
, flags
);
595 * Allow userspace to request a mountpoint be expired rather than
596 * unmounting unconditionally. Unmount only happens if:
597 * (1) the mark is already set (the mark is cleared by mntput())
598 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
600 if (flags
& MNT_EXPIRE
) {
601 if (mnt
== current
->fs
->root
.mnt
||
602 flags
& (MNT_FORCE
| MNT_DETACH
))
605 if (atomic_read(&mnt
->mnt_count
) != 2)
608 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
613 * If we may have to abort operations to get out of this
614 * mount, and they will themselves hold resources we must
615 * allow the fs to do things. In the Unix tradition of
616 * 'Gee thats tricky lets do it in userspace' the umount_begin
617 * might fail to complete on the first run through as other tasks
618 * must return, and the like. Thats for the mount program to worry
619 * about for the moment.
623 if (sb
->s_op
->umount_begin
)
624 sb
->s_op
->umount_begin(mnt
, flags
);
628 * No sense to grab the lock for this test, but test itself looks
629 * somewhat bogus. Suggestions for better replacement?
630 * Ho-hum... In principle, we might treat that as umount + switch
631 * to rootfs. GC would eventually take care of the old vfsmount.
632 * Actually it makes sense, especially if rootfs would contain a
633 * /reboot - static binary that would close all descriptors and
634 * call reboot(9). Then init(8) could umount root and exec /reboot.
636 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
638 * Special case for "unmounting" root ...
639 * we just try to remount it readonly.
641 down_write(&sb
->s_umount
);
642 if (!(sb
->s_flags
& MS_RDONLY
)) {
645 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
648 up_write(&sb
->s_umount
);
652 down_write(&namespace_sem
);
653 spin_lock(&vfsmount_lock
);
657 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
658 if (!list_empty(&mnt
->mnt_list
))
659 umount_tree(mnt
, 1, &umount_list
);
662 spin_unlock(&vfsmount_lock
);
664 security_sb_umount_busy(mnt
);
665 up_write(&namespace_sem
);
666 release_mounts(&umount_list
);
671 * Now umount can handle mount points as well as block devices.
672 * This is important for filesystems which use unnamed block devices.
674 * We now support a flag for forced unmount like the other 'big iron'
675 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
678 asmlinkage
long sys_umount(char __user
* name
, int flags
)
683 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
687 if (nd
.path
.dentry
!= nd
.path
.mnt
->mnt_root
)
689 if (!check_mnt(nd
.path
.mnt
))
693 if (!capable(CAP_SYS_ADMIN
))
696 retval
= do_umount(nd
.path
.mnt
, flags
);
698 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
699 dput(nd
.path
.dentry
);
700 mntput_no_expire(nd
.path
.mnt
);
705 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
708 * The 2.0 compatible umount. No flags.
710 asmlinkage
long sys_oldumount(char __user
* name
)
712 return sys_umount(name
, 0);
717 static int mount_is_safe(struct nameidata
*nd
)
719 if (capable(CAP_SYS_ADMIN
))
723 if (S_ISLNK(nd
->path
.dentry
->d_inode
->i_mode
))
725 if (nd
->path
.dentry
->d_inode
->i_mode
& S_ISVTX
) {
726 if (current
->uid
!= nd
->path
.dentry
->d_inode
->i_uid
)
729 if (vfs_permission(nd
, MAY_WRITE
))
735 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
740 if (d
== NULL
|| d
== d
->d_parent
)
746 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
749 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
752 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
755 res
= q
= clone_mnt(mnt
, dentry
, flag
);
758 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
761 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
762 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
765 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
766 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
767 s
= skip_mnt_tree(s
);
770 while (p
!= s
->mnt_parent
) {
776 path
.dentry
= p
->mnt_mountpoint
;
777 q
= clone_mnt(p
, p
->mnt_root
, flag
);
780 spin_lock(&vfsmount_lock
);
781 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
782 attach_mnt(q
, &path
);
783 spin_unlock(&vfsmount_lock
);
789 LIST_HEAD(umount_list
);
790 spin_lock(&vfsmount_lock
);
791 umount_tree(res
, 0, &umount_list
);
792 spin_unlock(&vfsmount_lock
);
793 release_mounts(&umount_list
);
798 struct vfsmount
*collect_mounts(struct vfsmount
*mnt
, struct dentry
*dentry
)
800 struct vfsmount
*tree
;
801 down_read(&namespace_sem
);
802 tree
= copy_tree(mnt
, dentry
, CL_COPY_ALL
| CL_PRIVATE
);
803 up_read(&namespace_sem
);
807 void drop_collected_mounts(struct vfsmount
*mnt
)
809 LIST_HEAD(umount_list
);
810 down_read(&namespace_sem
);
811 spin_lock(&vfsmount_lock
);
812 umount_tree(mnt
, 0, &umount_list
);
813 spin_unlock(&vfsmount_lock
);
814 up_read(&namespace_sem
);
815 release_mounts(&umount_list
);
819 * @source_mnt : mount tree to be attached
820 * @nd : place the mount tree @source_mnt is attached
821 * @parent_nd : if non-null, detach the source_mnt from its parent and
822 * store the parent mount and mountpoint dentry.
823 * (done when source_mnt is moved)
825 * NOTE: in the table below explains the semantics when a source mount
826 * of a given type is attached to a destination mount of a given type.
827 * ---------------------------------------------------------------------------
828 * | BIND MOUNT OPERATION |
829 * |**************************************************************************
830 * | source-->| shared | private | slave | unbindable |
834 * |**************************************************************************
835 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
837 * |non-shared| shared (+) | private | slave (*) | invalid |
838 * ***************************************************************************
839 * A bind operation clones the source mount and mounts the clone on the
842 * (++) the cloned mount is propagated to all the mounts in the propagation
843 * tree of the destination mount and the cloned mount is added to
844 * the peer group of the source mount.
845 * (+) the cloned mount is created under the destination mount and is marked
846 * as shared. The cloned mount is added to the peer group of the source
848 * (+++) the mount is propagated to all the mounts in the propagation tree
849 * of the destination mount and the cloned mount is made slave
850 * of the same master as that of the source mount. The cloned mount
851 * is marked as 'shared and slave'.
852 * (*) the cloned mount is made a slave of the same master as that of the
855 * ---------------------------------------------------------------------------
856 * | MOVE MOUNT OPERATION |
857 * |**************************************************************************
858 * | source-->| shared | private | slave | unbindable |
862 * |**************************************************************************
863 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
865 * |non-shared| shared (+*) | private | slave (*) | unbindable |
866 * ***************************************************************************
868 * (+) the mount is moved to the destination. And is then propagated to
869 * all the mounts in the propagation tree of the destination mount.
870 * (+*) the mount is moved to the destination.
871 * (+++) the mount is moved to the destination and is then propagated to
872 * all the mounts belonging to the destination mount's propagation tree.
873 * the mount is marked as 'shared and slave'.
874 * (*) the mount continues to be a slave at the new location.
876 * if the source mount is a tree, the operations explained above is
877 * applied to each mount in the tree.
878 * Must be called without spinlocks held, since this function can sleep
881 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
882 struct path
*path
, struct path
*parent_path
)
884 LIST_HEAD(tree_list
);
885 struct vfsmount
*dest_mnt
= path
->mnt
;
886 struct dentry
*dest_dentry
= path
->dentry
;
887 struct vfsmount
*child
, *p
;
889 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
892 if (IS_MNT_SHARED(dest_mnt
)) {
893 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
897 spin_lock(&vfsmount_lock
);
899 detach_mnt(source_mnt
, parent_path
);
900 attach_mnt(source_mnt
, path
);
901 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
903 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
904 commit_tree(source_mnt
);
907 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
908 list_del_init(&child
->mnt_hash
);
911 spin_unlock(&vfsmount_lock
);
915 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
918 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
921 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
922 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
926 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
927 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
930 err
= security_sb_check_sb(mnt
, nd
);
935 if (IS_ROOT(nd
->path
.dentry
) || !d_unhashed(nd
->path
.dentry
))
936 err
= attach_recursive_mnt(mnt
, &nd
->path
, NULL
);
938 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
940 security_sb_post_addmount(mnt
, nd
);
945 * recursively change the type of the mountpoint.
946 * noinline this do_mount helper to save do_mount stack space.
948 static noinline
int do_change_type(struct nameidata
*nd
, int flag
)
950 struct vfsmount
*m
, *mnt
= nd
->path
.mnt
;
951 int recurse
= flag
& MS_REC
;
952 int type
= flag
& ~MS_REC
;
954 if (!capable(CAP_SYS_ADMIN
))
957 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
960 down_write(&namespace_sem
);
961 spin_lock(&vfsmount_lock
);
962 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
963 change_mnt_propagation(m
, type
);
964 spin_unlock(&vfsmount_lock
);
965 up_write(&namespace_sem
);
971 * noinline this do_mount helper to save do_mount stack space.
973 static noinline
int do_loopback(struct nameidata
*nd
, char *old_name
,
976 struct nameidata old_nd
;
977 struct vfsmount
*mnt
= NULL
;
978 int err
= mount_is_safe(nd
);
981 if (!old_name
|| !*old_name
)
983 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
987 down_write(&namespace_sem
);
989 if (IS_MNT_UNBINDABLE(old_nd
.path
.mnt
))
992 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
997 mnt
= copy_tree(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
999 mnt
= clone_mnt(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1004 err
= graft_tree(mnt
, nd
);
1006 LIST_HEAD(umount_list
);
1007 spin_lock(&vfsmount_lock
);
1008 umount_tree(mnt
, 0, &umount_list
);
1009 spin_unlock(&vfsmount_lock
);
1010 release_mounts(&umount_list
);
1014 up_write(&namespace_sem
);
1015 path_put(&old_nd
.path
);
1020 * change filesystem flags. dir should be a physical root of filesystem.
1021 * If you've mounted a non-root directory somewhere and want to do remount
1022 * on it - tough luck.
1023 * noinline this do_mount helper to save do_mount stack space.
1025 static noinline
int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
1029 struct super_block
*sb
= nd
->path
.mnt
->mnt_sb
;
1031 if (!capable(CAP_SYS_ADMIN
))
1034 if (!check_mnt(nd
->path
.mnt
))
1037 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1040 down_write(&sb
->s_umount
);
1041 err
= do_remount_sb(sb
, flags
, data
, 0);
1043 nd
->path
.mnt
->mnt_flags
= mnt_flags
;
1044 up_write(&sb
->s_umount
);
1046 security_sb_post_remount(nd
->path
.mnt
, flags
, data
);
1050 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1053 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1054 if (IS_MNT_UNBINDABLE(p
))
1061 * noinline this do_mount helper to save do_mount stack space.
1063 static noinline
int do_move_mount(struct nameidata
*nd
, char *old_name
)
1065 struct nameidata old_nd
;
1066 struct path parent_path
;
1069 if (!capable(CAP_SYS_ADMIN
))
1071 if (!old_name
|| !*old_name
)
1073 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1077 down_write(&namespace_sem
);
1078 while (d_mountpoint(nd
->path
.dentry
) &&
1079 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1082 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1086 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1087 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
1090 if (!IS_ROOT(nd
->path
.dentry
) && d_unhashed(nd
->path
.dentry
))
1094 if (old_nd
.path
.dentry
!= old_nd
.path
.mnt
->mnt_root
)
1097 if (old_nd
.path
.mnt
== old_nd
.path
.mnt
->mnt_parent
)
1100 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
1101 S_ISDIR(old_nd
.path
.dentry
->d_inode
->i_mode
))
1104 * Don't move a mount residing in a shared parent.
1106 if (old_nd
.path
.mnt
->mnt_parent
&&
1107 IS_MNT_SHARED(old_nd
.path
.mnt
->mnt_parent
))
1110 * Don't move a mount tree containing unbindable mounts to a destination
1111 * mount which is shared.
1113 if (IS_MNT_SHARED(nd
->path
.mnt
) &&
1114 tree_contains_unbindable(old_nd
.path
.mnt
))
1117 for (p
= nd
->path
.mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1118 if (p
== old_nd
.path
.mnt
)
1121 err
= attach_recursive_mnt(old_nd
.path
.mnt
, &nd
->path
, &parent_path
);
1125 spin_lock(&vfsmount_lock
);
1126 /* if the mount is moved, it should no longer be expire
1128 list_del_init(&old_nd
.path
.mnt
->mnt_expire
);
1129 spin_unlock(&vfsmount_lock
);
1131 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1133 up_write(&namespace_sem
);
1135 path_put(&parent_path
);
1136 path_put(&old_nd
.path
);
1141 * create a new mount for userspace and request it to be added into the
1143 * noinline this do_mount helper to save do_mount stack space.
1145 static noinline
int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1146 int mnt_flags
, char *name
, void *data
)
1148 struct vfsmount
*mnt
;
1150 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1153 /* we need capabilities... */
1154 if (!capable(CAP_SYS_ADMIN
))
1157 mnt
= do_kern_mount(type
, flags
, name
, data
);
1159 return PTR_ERR(mnt
);
1161 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1165 * add a mount into a namespace's mount tree
1166 * - provide the option of adding the new mount to an expiration list
1168 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1169 int mnt_flags
, struct list_head
*fslist
)
1173 down_write(&namespace_sem
);
1174 /* Something was mounted here while we slept */
1175 while (d_mountpoint(nd
->path
.dentry
) &&
1176 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1179 if (!check_mnt(nd
->path
.mnt
))
1182 /* Refuse the same filesystem on the same mount point */
1184 if (nd
->path
.mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1185 nd
->path
.mnt
->mnt_root
== nd
->path
.dentry
)
1189 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1192 newmnt
->mnt_flags
= mnt_flags
;
1193 if ((err
= graft_tree(newmnt
, nd
)))
1197 /* add to the specified expiration list */
1198 spin_lock(&vfsmount_lock
);
1199 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1200 spin_unlock(&vfsmount_lock
);
1202 up_write(&namespace_sem
);
1206 up_write(&namespace_sem
);
1211 EXPORT_SYMBOL_GPL(do_add_mount
);
1213 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
1214 struct list_head
*umounts
)
1216 spin_lock(&vfsmount_lock
);
1219 * Check if mount is still attached, if not, let whoever holds it deal
1222 if (mnt
->mnt_parent
== mnt
) {
1223 spin_unlock(&vfsmount_lock
);
1228 * Check that it is still dead: the count should now be 2 - as
1229 * contributed by the vfsmount parent and the mntget above
1231 if (!propagate_mount_busy(mnt
, 2)) {
1232 /* delete from the namespace */
1233 touch_mnt_namespace(mnt
->mnt_ns
);
1234 list_del_init(&mnt
->mnt_list
);
1236 umount_tree(mnt
, 1, umounts
);
1237 spin_unlock(&vfsmount_lock
);
1240 * Someone brought it back to life whilst we didn't have any
1241 * locks held so return it to the expiration list
1243 list_add_tail(&mnt
->mnt_expire
, mounts
);
1244 spin_unlock(&vfsmount_lock
);
1249 * go through the vfsmounts we've just consigned to the graveyard to
1250 * - check that they're still dead
1251 * - delete the vfsmount from the appropriate namespace under lock
1252 * - dispose of the corpse
1254 static void expire_mount_list(struct list_head
*graveyard
, struct list_head
*mounts
)
1256 struct mnt_namespace
*ns
;
1257 struct vfsmount
*mnt
;
1259 while (!list_empty(graveyard
)) {
1261 mnt
= list_first_entry(graveyard
, struct vfsmount
, mnt_expire
);
1262 list_del_init(&mnt
->mnt_expire
);
1264 /* don't do anything if the namespace is dead - all the
1265 * vfsmounts from it are going away anyway */
1267 if (!ns
|| !ns
->root
)
1271 spin_unlock(&vfsmount_lock
);
1272 down_write(&namespace_sem
);
1273 expire_mount(mnt
, mounts
, &umounts
);
1274 up_write(&namespace_sem
);
1275 release_mounts(&umounts
);
1278 spin_lock(&vfsmount_lock
);
1283 * process a list of expirable mountpoints with the intent of discarding any
1284 * mountpoints that aren't in use and haven't been touched since last we came
1287 void mark_mounts_for_expiry(struct list_head
*mounts
)
1289 struct vfsmount
*mnt
, *next
;
1290 LIST_HEAD(graveyard
);
1292 if (list_empty(mounts
))
1295 spin_lock(&vfsmount_lock
);
1297 /* extract from the expiration list every vfsmount that matches the
1298 * following criteria:
1299 * - only referenced by its parent vfsmount
1300 * - still marked for expiry (marked on the last call here; marks are
1301 * cleared by mntput())
1303 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1304 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1305 atomic_read(&mnt
->mnt_count
) != 1)
1309 list_move(&mnt
->mnt_expire
, &graveyard
);
1312 expire_mount_list(&graveyard
, mounts
);
1314 spin_unlock(&vfsmount_lock
);
1317 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1320 * Ripoff of 'select_parent()'
1322 * search the list of submounts for a given mountpoint, and move any
1323 * shrinkable submounts to the 'graveyard' list.
1325 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1327 struct vfsmount
*this_parent
= parent
;
1328 struct list_head
*next
;
1332 next
= this_parent
->mnt_mounts
.next
;
1334 while (next
!= &this_parent
->mnt_mounts
) {
1335 struct list_head
*tmp
= next
;
1336 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1339 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1342 * Descend a level if the d_mounts list is non-empty.
1344 if (!list_empty(&mnt
->mnt_mounts
)) {
1349 if (!propagate_mount_busy(mnt
, 1)) {
1351 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1356 * All done at this level ... ascend and resume the search
1358 if (this_parent
!= parent
) {
1359 next
= this_parent
->mnt_child
.next
;
1360 this_parent
= this_parent
->mnt_parent
;
1367 * process a list of expirable mountpoints with the intent of discarding any
1368 * submounts of a specific parent mountpoint
1370 void shrink_submounts(struct vfsmount
*mountpoint
, struct list_head
*mounts
)
1372 LIST_HEAD(graveyard
);
1375 spin_lock(&vfsmount_lock
);
1377 /* extract submounts of 'mountpoint' from the expiration list */
1378 while ((found
= select_submounts(mountpoint
, &graveyard
)) != 0)
1379 expire_mount_list(&graveyard
, mounts
);
1381 spin_unlock(&vfsmount_lock
);
1384 EXPORT_SYMBOL_GPL(shrink_submounts
);
1387 * Some copy_from_user() implementations do not return the exact number of
1388 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1389 * Note that this function differs from copy_from_user() in that it will oops
1390 * on bad values of `to', rather than returning a short copy.
1392 static long exact_copy_from_user(void *to
, const void __user
* from
,
1396 const char __user
*f
= from
;
1399 if (!access_ok(VERIFY_READ
, from
, n
))
1403 if (__get_user(c
, f
)) {
1414 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1424 if (!(page
= __get_free_page(GFP_KERNEL
)))
1427 /* We only care that *some* data at the address the user
1428 * gave us is valid. Just in case, we'll zero
1429 * the remainder of the page.
1431 /* copy_from_user cannot cross TASK_SIZE ! */
1432 size
= TASK_SIZE
- (unsigned long)data
;
1433 if (size
> PAGE_SIZE
)
1436 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1442 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1448 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1449 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1451 * data is a (void *) that can point to any structure up to
1452 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1453 * information (or be NULL).
1455 * Pre-0.97 versions of mount() didn't have a flags word.
1456 * When the flags word was introduced its top half was required
1457 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1458 * Therefore, if this magic number is present, it carries no information
1459 * and must be discarded.
1461 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1462 unsigned long flags
, void *data_page
)
1464 struct nameidata nd
;
1469 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1470 flags
&= ~MS_MGC_MSK
;
1472 /* Basic sanity checks */
1474 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1476 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1480 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1482 /* Separate the per-mountpoint flags */
1483 if (flags
& MS_NOSUID
)
1484 mnt_flags
|= MNT_NOSUID
;
1485 if (flags
& MS_NODEV
)
1486 mnt_flags
|= MNT_NODEV
;
1487 if (flags
& MS_NOEXEC
)
1488 mnt_flags
|= MNT_NOEXEC
;
1489 if (flags
& MS_NOATIME
)
1490 mnt_flags
|= MNT_NOATIME
;
1491 if (flags
& MS_NODIRATIME
)
1492 mnt_flags
|= MNT_NODIRATIME
;
1493 if (flags
& MS_RELATIME
)
1494 mnt_flags
|= MNT_RELATIME
;
1496 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1497 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
);
1499 /* ... and get the mountpoint */
1500 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1504 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1508 if (flags
& MS_REMOUNT
)
1509 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1511 else if (flags
& MS_BIND
)
1512 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1513 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1514 retval
= do_change_type(&nd
, flags
);
1515 else if (flags
& MS_MOVE
)
1516 retval
= do_move_mount(&nd
, dev_name
);
1518 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1519 dev_name
, data_page
);
1526 * Allocate a new namespace structure and populate it with contents
1527 * copied from the namespace of the passed in task structure.
1529 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1530 struct fs_struct
*fs
)
1532 struct mnt_namespace
*new_ns
;
1533 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1534 struct vfsmount
*p
, *q
;
1536 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1538 return ERR_PTR(-ENOMEM
);
1540 atomic_set(&new_ns
->count
, 1);
1541 INIT_LIST_HEAD(&new_ns
->list
);
1542 init_waitqueue_head(&new_ns
->poll
);
1545 down_write(&namespace_sem
);
1546 /* First pass: copy the tree topology */
1547 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1548 CL_COPY_ALL
| CL_EXPIRE
);
1549 if (!new_ns
->root
) {
1550 up_write(&namespace_sem
);
1552 return ERR_PTR(-ENOMEM
);;
1554 spin_lock(&vfsmount_lock
);
1555 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1556 spin_unlock(&vfsmount_lock
);
1559 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1560 * as belonging to new namespace. We have already acquired a private
1561 * fs_struct, so tsk->fs->lock is not needed.
1568 if (p
== fs
->root
.mnt
) {
1570 fs
->root
.mnt
= mntget(q
);
1572 if (p
== fs
->pwd
.mnt
) {
1574 fs
->pwd
.mnt
= mntget(q
);
1576 if (p
== fs
->altroot
.mnt
) {
1578 fs
->altroot
.mnt
= mntget(q
);
1581 p
= next_mnt(p
, mnt_ns
->root
);
1582 q
= next_mnt(q
, new_ns
->root
);
1584 up_write(&namespace_sem
);
1596 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
1597 struct fs_struct
*new_fs
)
1599 struct mnt_namespace
*new_ns
;
1604 if (!(flags
& CLONE_NEWNS
))
1607 new_ns
= dup_mnt_ns(ns
, new_fs
);
1613 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1614 char __user
* type
, unsigned long flags
,
1618 unsigned long data_page
;
1619 unsigned long type_page
;
1620 unsigned long dev_page
;
1623 retval
= copy_mount_options(type
, &type_page
);
1627 dir_page
= getname(dir_name
);
1628 retval
= PTR_ERR(dir_page
);
1629 if (IS_ERR(dir_page
))
1632 retval
= copy_mount_options(dev_name
, &dev_page
);
1636 retval
= copy_mount_options(data
, &data_page
);
1641 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1642 flags
, (void *)data_page
);
1644 free_page(data_page
);
1647 free_page(dev_page
);
1651 free_page(type_page
);
1656 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1657 * It can block. Requires the big lock held.
1659 void set_fs_root(struct fs_struct
*fs
, struct path
*path
)
1661 struct path old_root
;
1663 write_lock(&fs
->lock
);
1664 old_root
= fs
->root
;
1667 write_unlock(&fs
->lock
);
1668 if (old_root
.dentry
)
1669 path_put(&old_root
);
1673 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1674 * It can block. Requires the big lock held.
1676 void set_fs_pwd(struct fs_struct
*fs
, struct path
*path
)
1678 struct path old_pwd
;
1680 write_lock(&fs
->lock
);
1684 write_unlock(&fs
->lock
);
1690 static void chroot_fs_refs(struct path
*old_root
, struct path
*new_root
)
1692 struct task_struct
*g
, *p
;
1693 struct fs_struct
*fs
;
1695 read_lock(&tasklist_lock
);
1696 do_each_thread(g
, p
) {
1700 atomic_inc(&fs
->count
);
1702 if (fs
->root
.dentry
== old_root
->dentry
1703 && fs
->root
.mnt
== old_root
->mnt
)
1704 set_fs_root(fs
, new_root
);
1705 if (fs
->pwd
.dentry
== old_root
->dentry
1706 && fs
->pwd
.mnt
== old_root
->mnt
)
1707 set_fs_pwd(fs
, new_root
);
1711 } while_each_thread(g
, p
);
1712 read_unlock(&tasklist_lock
);
1716 * pivot_root Semantics:
1717 * Moves the root file system of the current process to the directory put_old,
1718 * makes new_root as the new root file system of the current process, and sets
1719 * root/cwd of all processes which had them on the current root to new_root.
1722 * The new_root and put_old must be directories, and must not be on the
1723 * same file system as the current process root. The put_old must be
1724 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1725 * pointed to by put_old must yield the same directory as new_root. No other
1726 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1728 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1729 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1730 * in this situation.
1733 * - we don't move root/cwd if they are not at the root (reason: if something
1734 * cared enough to change them, it's probably wrong to force them elsewhere)
1735 * - it's okay to pick a root that isn't the root of a file system, e.g.
1736 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1737 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1740 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1741 const char __user
* put_old
)
1743 struct vfsmount
*tmp
;
1744 struct nameidata new_nd
, old_nd
, user_nd
;
1745 struct path parent_path
, root_parent
;
1748 if (!capable(CAP_SYS_ADMIN
))
1753 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1758 if (!check_mnt(new_nd
.path
.mnt
))
1761 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1765 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1767 path_put(&old_nd
.path
);
1771 read_lock(¤t
->fs
->lock
);
1772 user_nd
.path
= current
->fs
->root
;
1773 path_get(¤t
->fs
->root
);
1774 read_unlock(¤t
->fs
->lock
);
1775 down_write(&namespace_sem
);
1776 mutex_lock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
1778 if (IS_MNT_SHARED(old_nd
.path
.mnt
) ||
1779 IS_MNT_SHARED(new_nd
.path
.mnt
->mnt_parent
) ||
1780 IS_MNT_SHARED(user_nd
.path
.mnt
->mnt_parent
))
1782 if (!check_mnt(user_nd
.path
.mnt
))
1785 if (IS_DEADDIR(new_nd
.path
.dentry
->d_inode
))
1787 if (d_unhashed(new_nd
.path
.dentry
) && !IS_ROOT(new_nd
.path
.dentry
))
1789 if (d_unhashed(old_nd
.path
.dentry
) && !IS_ROOT(old_nd
.path
.dentry
))
1792 if (new_nd
.path
.mnt
== user_nd
.path
.mnt
||
1793 old_nd
.path
.mnt
== user_nd
.path
.mnt
)
1794 goto out2
; /* loop, on the same file system */
1796 if (user_nd
.path
.mnt
->mnt_root
!= user_nd
.path
.dentry
)
1797 goto out2
; /* not a mountpoint */
1798 if (user_nd
.path
.mnt
->mnt_parent
== user_nd
.path
.mnt
)
1799 goto out2
; /* not attached */
1800 if (new_nd
.path
.mnt
->mnt_root
!= new_nd
.path
.dentry
)
1801 goto out2
; /* not a mountpoint */
1802 if (new_nd
.path
.mnt
->mnt_parent
== new_nd
.path
.mnt
)
1803 goto out2
; /* not attached */
1804 /* make sure we can reach put_old from new_root */
1805 tmp
= old_nd
.path
.mnt
;
1806 spin_lock(&vfsmount_lock
);
1807 if (tmp
!= new_nd
.path
.mnt
) {
1809 if (tmp
->mnt_parent
== tmp
)
1810 goto out3
; /* already mounted on put_old */
1811 if (tmp
->mnt_parent
== new_nd
.path
.mnt
)
1813 tmp
= tmp
->mnt_parent
;
1815 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.path
.dentry
))
1817 } else if (!is_subdir(old_nd
.path
.dentry
, new_nd
.path
.dentry
))
1819 detach_mnt(new_nd
.path
.mnt
, &parent_path
);
1820 detach_mnt(user_nd
.path
.mnt
, &root_parent
);
1821 /* mount old root on put_old */
1822 attach_mnt(user_nd
.path
.mnt
, &old_nd
.path
);
1823 /* mount new_root on / */
1824 attach_mnt(new_nd
.path
.mnt
, &root_parent
);
1825 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1826 spin_unlock(&vfsmount_lock
);
1827 chroot_fs_refs(&user_nd
.path
, &new_nd
.path
);
1828 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1830 path_put(&root_parent
);
1831 path_put(&parent_path
);
1833 mutex_unlock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
1834 up_write(&namespace_sem
);
1835 path_put(&user_nd
.path
);
1836 path_put(&old_nd
.path
);
1838 path_put(&new_nd
.path
);
1843 spin_unlock(&vfsmount_lock
);
1847 static void __init
init_mount_tree(void)
1849 struct vfsmount
*mnt
;
1850 struct mnt_namespace
*ns
;
1853 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1855 panic("Can't create rootfs");
1856 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1858 panic("Can't allocate initial namespace");
1859 atomic_set(&ns
->count
, 1);
1860 INIT_LIST_HEAD(&ns
->list
);
1861 init_waitqueue_head(&ns
->poll
);
1863 list_add(&mnt
->mnt_list
, &ns
->list
);
1867 init_task
.nsproxy
->mnt_ns
= ns
;
1870 root
.mnt
= ns
->root
;
1871 root
.dentry
= ns
->root
->mnt_root
;
1873 set_fs_pwd(current
->fs
, &root
);
1874 set_fs_root(current
->fs
, &root
);
1877 void __init
mnt_init(void)
1882 init_rwsem(&namespace_sem
);
1884 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1885 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1887 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1889 if (!mount_hashtable
)
1890 panic("Failed to allocate mount hash table\n");
1892 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
1894 for (u
= 0; u
< HASH_SIZE
; u
++)
1895 INIT_LIST_HEAD(&mount_hashtable
[u
]);
1899 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1901 fs_kobj
= kobject_create_and_add("fs", NULL
);
1903 printk(KERN_WARNING
"%s: kobj create error\n", __FUNCTION__
);
1908 void __put_mnt_ns(struct mnt_namespace
*ns
)
1910 struct vfsmount
*root
= ns
->root
;
1911 LIST_HEAD(umount_list
);
1913 spin_unlock(&vfsmount_lock
);
1914 down_write(&namespace_sem
);
1915 spin_lock(&vfsmount_lock
);
1916 umount_tree(root
, 0, &umount_list
);
1917 spin_unlock(&vfsmount_lock
);
1918 up_write(&namespace_sem
);
1919 release_mounts(&umount_list
);