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
;
84 * Most r/o checks on a fs are for operations that take
85 * discrete amounts of time, like a write() or unlink().
86 * We must keep track of when those operations start
87 * (for permission checks) and when they end, so that
88 * we can determine when writes are able to occur to
92 * mnt_want_write - get write access to a mount
93 * @mnt: the mount on which to take a write
95 * This tells the low-level filesystem that a write is
96 * about to be performed to it, and makes sure that
97 * writes are allowed before returning success. When
98 * the write operation is finished, mnt_drop_write()
99 * must be called. This is effectively a refcount.
101 int mnt_want_write(struct vfsmount
*mnt
)
103 if (__mnt_is_readonly(mnt
))
107 EXPORT_SYMBOL_GPL(mnt_want_write
);
110 * mnt_drop_write - give up write access to a mount
111 * @mnt: the mount on which to give up write access
113 * Tells the low-level filesystem that we are done
114 * performing writes to it. Must be matched with
115 * mnt_want_write() call above.
117 void mnt_drop_write(struct vfsmount
*mnt
)
120 EXPORT_SYMBOL_GPL(mnt_drop_write
);
123 * __mnt_is_readonly: check whether a mount is read-only
124 * @mnt: the mount to check for its write status
126 * This shouldn't be used directly ouside of the VFS.
127 * It does not guarantee that the filesystem will stay
128 * r/w, just that it is right *now*. This can not and
129 * should not be used in place of IS_RDONLY(inode).
131 int __mnt_is_readonly(struct vfsmount
*mnt
)
133 return (mnt
->mnt_sb
->s_flags
& MS_RDONLY
);
135 EXPORT_SYMBOL_GPL(__mnt_is_readonly
);
137 int simple_set_mnt(struct vfsmount
*mnt
, struct super_block
*sb
)
140 mnt
->mnt_root
= dget(sb
->s_root
);
144 EXPORT_SYMBOL(simple_set_mnt
);
146 void free_vfsmnt(struct vfsmount
*mnt
)
148 kfree(mnt
->mnt_devname
);
149 kmem_cache_free(mnt_cache
, mnt
);
153 * find the first or last mount at @dentry on vfsmount @mnt depending on
154 * @dir. If @dir is set return the first mount else return the last mount.
156 struct vfsmount
*__lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
,
159 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
160 struct list_head
*tmp
= head
;
161 struct vfsmount
*p
, *found
= NULL
;
164 tmp
= dir
? tmp
->next
: tmp
->prev
;
168 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
169 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
178 * lookup_mnt increments the ref count before returning
179 * the vfsmount struct.
181 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
183 struct vfsmount
*child_mnt
;
184 spin_lock(&vfsmount_lock
);
185 if ((child_mnt
= __lookup_mnt(mnt
, dentry
, 1)))
187 spin_unlock(&vfsmount_lock
);
191 static inline int check_mnt(struct vfsmount
*mnt
)
193 return mnt
->mnt_ns
== current
->nsproxy
->mnt_ns
;
196 static void touch_mnt_namespace(struct mnt_namespace
*ns
)
200 wake_up_interruptible(&ns
->poll
);
204 static void __touch_mnt_namespace(struct mnt_namespace
*ns
)
206 if (ns
&& ns
->event
!= event
) {
208 wake_up_interruptible(&ns
->poll
);
212 static void detach_mnt(struct vfsmount
*mnt
, struct path
*old_path
)
214 old_path
->dentry
= mnt
->mnt_mountpoint
;
215 old_path
->mnt
= mnt
->mnt_parent
;
216 mnt
->mnt_parent
= mnt
;
217 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
218 list_del_init(&mnt
->mnt_child
);
219 list_del_init(&mnt
->mnt_hash
);
220 old_path
->dentry
->d_mounted
--;
223 void mnt_set_mountpoint(struct vfsmount
*mnt
, struct dentry
*dentry
,
224 struct vfsmount
*child_mnt
)
226 child_mnt
->mnt_parent
= mntget(mnt
);
227 child_mnt
->mnt_mountpoint
= dget(dentry
);
231 static void attach_mnt(struct vfsmount
*mnt
, struct path
*path
)
233 mnt_set_mountpoint(path
->mnt
, path
->dentry
, mnt
);
234 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
235 hash(path
->mnt
, path
->dentry
));
236 list_add_tail(&mnt
->mnt_child
, &path
->mnt
->mnt_mounts
);
240 * the caller must hold vfsmount_lock
242 static void commit_tree(struct vfsmount
*mnt
)
244 struct vfsmount
*parent
= mnt
->mnt_parent
;
247 struct mnt_namespace
*n
= parent
->mnt_ns
;
249 BUG_ON(parent
== mnt
);
251 list_add_tail(&head
, &mnt
->mnt_list
);
252 list_for_each_entry(m
, &head
, mnt_list
)
254 list_splice(&head
, n
->list
.prev
);
256 list_add_tail(&mnt
->mnt_hash
, mount_hashtable
+
257 hash(parent
, mnt
->mnt_mountpoint
));
258 list_add_tail(&mnt
->mnt_child
, &parent
->mnt_mounts
);
259 touch_mnt_namespace(n
);
262 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
264 struct list_head
*next
= p
->mnt_mounts
.next
;
265 if (next
== &p
->mnt_mounts
) {
269 next
= p
->mnt_child
.next
;
270 if (next
!= &p
->mnt_parent
->mnt_mounts
)
275 return list_entry(next
, struct vfsmount
, mnt_child
);
278 static struct vfsmount
*skip_mnt_tree(struct vfsmount
*p
)
280 struct list_head
*prev
= p
->mnt_mounts
.prev
;
281 while (prev
!= &p
->mnt_mounts
) {
282 p
= list_entry(prev
, struct vfsmount
, mnt_child
);
283 prev
= p
->mnt_mounts
.prev
;
288 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
,
291 struct super_block
*sb
= old
->mnt_sb
;
292 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
295 mnt
->mnt_flags
= old
->mnt_flags
;
296 atomic_inc(&sb
->s_active
);
298 mnt
->mnt_root
= dget(root
);
299 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
300 mnt
->mnt_parent
= mnt
;
302 if (flag
& CL_SLAVE
) {
303 list_add(&mnt
->mnt_slave
, &old
->mnt_slave_list
);
304 mnt
->mnt_master
= old
;
305 CLEAR_MNT_SHARED(mnt
);
306 } else if (!(flag
& CL_PRIVATE
)) {
307 if ((flag
& CL_PROPAGATION
) || IS_MNT_SHARED(old
))
308 list_add(&mnt
->mnt_share
, &old
->mnt_share
);
309 if (IS_MNT_SLAVE(old
))
310 list_add(&mnt
->mnt_slave
, &old
->mnt_slave
);
311 mnt
->mnt_master
= old
->mnt_master
;
313 if (flag
& CL_MAKE_SHARED
)
316 /* stick the duplicate mount on the same expiry list
317 * as the original if that was on one */
318 if (flag
& CL_EXPIRE
) {
319 if (!list_empty(&old
->mnt_expire
))
320 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
326 static inline void __mntput(struct vfsmount
*mnt
)
328 struct super_block
*sb
= mnt
->mnt_sb
;
331 deactivate_super(sb
);
334 void mntput_no_expire(struct vfsmount
*mnt
)
337 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
338 if (likely(!mnt
->mnt_pinned
)) {
339 spin_unlock(&vfsmount_lock
);
343 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
345 spin_unlock(&vfsmount_lock
);
346 acct_auto_close_mnt(mnt
);
347 security_sb_umount_close(mnt
);
352 EXPORT_SYMBOL(mntput_no_expire
);
354 void mnt_pin(struct vfsmount
*mnt
)
356 spin_lock(&vfsmount_lock
);
358 spin_unlock(&vfsmount_lock
);
361 EXPORT_SYMBOL(mnt_pin
);
363 void mnt_unpin(struct vfsmount
*mnt
)
365 spin_lock(&vfsmount_lock
);
366 if (mnt
->mnt_pinned
) {
367 atomic_inc(&mnt
->mnt_count
);
370 spin_unlock(&vfsmount_lock
);
373 EXPORT_SYMBOL(mnt_unpin
);
375 static inline void mangle(struct seq_file
*m
, const char *s
)
377 seq_escape(m
, s
, " \t\n\\");
381 * Simple .show_options callback for filesystems which don't want to
382 * implement more complex mount option showing.
384 * See also save_mount_options().
386 int generic_show_options(struct seq_file
*m
, struct vfsmount
*mnt
)
388 const char *options
= mnt
->mnt_sb
->s_options
;
390 if (options
!= NULL
&& options
[0]) {
397 EXPORT_SYMBOL(generic_show_options
);
400 * If filesystem uses generic_show_options(), this function should be
401 * called from the fill_super() callback.
403 * The .remount_fs callback usually needs to be handled in a special
404 * way, to make sure, that previous options are not overwritten if the
407 * Also note, that if the filesystem's .remount_fs function doesn't
408 * reset all options to their default value, but changes only newly
409 * given options, then the displayed options will not reflect reality
412 void save_mount_options(struct super_block
*sb
, char *options
)
414 kfree(sb
->s_options
);
415 sb
->s_options
= kstrdup(options
, GFP_KERNEL
);
417 EXPORT_SYMBOL(save_mount_options
);
420 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
422 struct mnt_namespace
*n
= m
->private;
424 down_read(&namespace_sem
);
425 return seq_list_start(&n
->list
, *pos
);
428 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
430 struct mnt_namespace
*n
= m
->private;
432 return seq_list_next(v
, &n
->list
, pos
);
435 static void m_stop(struct seq_file
*m
, void *v
)
437 up_read(&namespace_sem
);
440 static int show_vfsmnt(struct seq_file
*m
, void *v
)
442 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
444 static struct proc_fs_info
{
448 { MS_SYNCHRONOUS
, ",sync" },
449 { MS_DIRSYNC
, ",dirsync" },
450 { MS_MANDLOCK
, ",mand" },
453 static struct proc_fs_info mnt_info
[] = {
454 { MNT_NOSUID
, ",nosuid" },
455 { MNT_NODEV
, ",nodev" },
456 { MNT_NOEXEC
, ",noexec" },
457 { MNT_NOATIME
, ",noatime" },
458 { MNT_NODIRATIME
, ",nodiratime" },
459 { MNT_RELATIME
, ",relatime" },
462 struct proc_fs_info
*fs_infop
;
463 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
465 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
467 seq_path(m
, &mnt_path
, " \t\n\\");
469 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
470 if (mnt
->mnt_sb
->s_subtype
&& mnt
->mnt_sb
->s_subtype
[0]) {
472 mangle(m
, mnt
->mnt_sb
->s_subtype
);
474 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
475 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
476 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
477 seq_puts(m
, fs_infop
->str
);
479 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
480 if (mnt
->mnt_flags
& fs_infop
->flag
)
481 seq_puts(m
, fs_infop
->str
);
483 if (mnt
->mnt_sb
->s_op
->show_options
)
484 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
485 seq_puts(m
, " 0 0\n");
489 struct seq_operations mounts_op
= {
496 static int show_vfsstat(struct seq_file
*m
, void *v
)
498 struct vfsmount
*mnt
= list_entry(v
, struct vfsmount
, mnt_list
);
499 struct path mnt_path
= { .dentry
= mnt
->mnt_root
, .mnt
= mnt
};
503 if (mnt
->mnt_devname
) {
504 seq_puts(m
, "device ");
505 mangle(m
, mnt
->mnt_devname
);
507 seq_puts(m
, "no device");
510 seq_puts(m
, " mounted on ");
511 seq_path(m
, &mnt_path
, " \t\n\\");
514 /* file system type */
515 seq_puts(m
, "with fstype ");
516 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
518 /* optional statistics */
519 if (mnt
->mnt_sb
->s_op
->show_stats
) {
521 err
= mnt
->mnt_sb
->s_op
->show_stats(m
, mnt
);
528 struct seq_operations mountstats_op
= {
532 .show
= show_vfsstat
,
536 * may_umount_tree - check if a mount tree is busy
537 * @mnt: root of mount tree
539 * This is called to check if a tree of mounts has any
540 * open files, pwds, chroots or sub mounts that are
543 int may_umount_tree(struct vfsmount
*mnt
)
546 int minimum_refs
= 0;
549 spin_lock(&vfsmount_lock
);
550 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
551 actual_refs
+= atomic_read(&p
->mnt_count
);
554 spin_unlock(&vfsmount_lock
);
556 if (actual_refs
> minimum_refs
)
562 EXPORT_SYMBOL(may_umount_tree
);
565 * may_umount - check if a mount point is busy
566 * @mnt: root of mount
568 * This is called to check if a mount point has any
569 * open files, pwds, chroots or sub mounts. If the
570 * mount has sub mounts this will return busy
571 * regardless of whether the sub mounts are busy.
573 * Doesn't take quota and stuff into account. IOW, in some cases it will
574 * give false negatives. The main reason why it's here is that we need
575 * a non-destructive way to look for easily umountable filesystems.
577 int may_umount(struct vfsmount
*mnt
)
580 spin_lock(&vfsmount_lock
);
581 if (propagate_mount_busy(mnt
, 2))
583 spin_unlock(&vfsmount_lock
);
587 EXPORT_SYMBOL(may_umount
);
589 void release_mounts(struct list_head
*head
)
591 struct vfsmount
*mnt
;
592 while (!list_empty(head
)) {
593 mnt
= list_first_entry(head
, struct vfsmount
, mnt_hash
);
594 list_del_init(&mnt
->mnt_hash
);
595 if (mnt
->mnt_parent
!= mnt
) {
596 struct dentry
*dentry
;
598 spin_lock(&vfsmount_lock
);
599 dentry
= mnt
->mnt_mountpoint
;
601 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
602 mnt
->mnt_parent
= mnt
;
604 spin_unlock(&vfsmount_lock
);
612 void umount_tree(struct vfsmount
*mnt
, int propagate
, struct list_head
*kill
)
616 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
))
617 list_move(&p
->mnt_hash
, kill
);
620 propagate_umount(kill
);
622 list_for_each_entry(p
, kill
, mnt_hash
) {
623 list_del_init(&p
->mnt_expire
);
624 list_del_init(&p
->mnt_list
);
625 __touch_mnt_namespace(p
->mnt_ns
);
627 list_del_init(&p
->mnt_child
);
628 if (p
->mnt_parent
!= p
) {
629 p
->mnt_parent
->mnt_ghosts
++;
630 p
->mnt_mountpoint
->d_mounted
--;
632 change_mnt_propagation(p
, MS_PRIVATE
);
636 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
);
638 static int do_umount(struct vfsmount
*mnt
, int flags
)
640 struct super_block
*sb
= mnt
->mnt_sb
;
642 LIST_HEAD(umount_list
);
644 retval
= security_sb_umount(mnt
, flags
);
649 * Allow userspace to request a mountpoint be expired rather than
650 * unmounting unconditionally. Unmount only happens if:
651 * (1) the mark is already set (the mark is cleared by mntput())
652 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
654 if (flags
& MNT_EXPIRE
) {
655 if (mnt
== current
->fs
->root
.mnt
||
656 flags
& (MNT_FORCE
| MNT_DETACH
))
659 if (atomic_read(&mnt
->mnt_count
) != 2)
662 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
667 * If we may have to abort operations to get out of this
668 * mount, and they will themselves hold resources we must
669 * allow the fs to do things. In the Unix tradition of
670 * 'Gee thats tricky lets do it in userspace' the umount_begin
671 * might fail to complete on the first run through as other tasks
672 * must return, and the like. Thats for the mount program to worry
673 * about for the moment.
677 if (sb
->s_op
->umount_begin
)
678 sb
->s_op
->umount_begin(mnt
, flags
);
682 * No sense to grab the lock for this test, but test itself looks
683 * somewhat bogus. Suggestions for better replacement?
684 * Ho-hum... In principle, we might treat that as umount + switch
685 * to rootfs. GC would eventually take care of the old vfsmount.
686 * Actually it makes sense, especially if rootfs would contain a
687 * /reboot - static binary that would close all descriptors and
688 * call reboot(9). Then init(8) could umount root and exec /reboot.
690 if (mnt
== current
->fs
->root
.mnt
&& !(flags
& MNT_DETACH
)) {
692 * Special case for "unmounting" root ...
693 * we just try to remount it readonly.
695 down_write(&sb
->s_umount
);
696 if (!(sb
->s_flags
& MS_RDONLY
)) {
699 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
702 up_write(&sb
->s_umount
);
706 down_write(&namespace_sem
);
707 spin_lock(&vfsmount_lock
);
710 if (!(flags
& MNT_DETACH
))
711 shrink_submounts(mnt
, &umount_list
);
714 if (flags
& MNT_DETACH
|| !propagate_mount_busy(mnt
, 2)) {
715 if (!list_empty(&mnt
->mnt_list
))
716 umount_tree(mnt
, 1, &umount_list
);
719 spin_unlock(&vfsmount_lock
);
721 security_sb_umount_busy(mnt
);
722 up_write(&namespace_sem
);
723 release_mounts(&umount_list
);
728 * Now umount can handle mount points as well as block devices.
729 * This is important for filesystems which use unnamed block devices.
731 * We now support a flag for forced unmount like the other 'big iron'
732 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
735 asmlinkage
long sys_umount(char __user
* name
, int flags
)
740 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
744 if (nd
.path
.dentry
!= nd
.path
.mnt
->mnt_root
)
746 if (!check_mnt(nd
.path
.mnt
))
750 if (!capable(CAP_SYS_ADMIN
))
753 retval
= do_umount(nd
.path
.mnt
, flags
);
755 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
756 dput(nd
.path
.dentry
);
757 mntput_no_expire(nd
.path
.mnt
);
762 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
765 * The 2.0 compatible umount. No flags.
767 asmlinkage
long sys_oldumount(char __user
* name
)
769 return sys_umount(name
, 0);
774 static int mount_is_safe(struct nameidata
*nd
)
776 if (capable(CAP_SYS_ADMIN
))
780 if (S_ISLNK(nd
->path
.dentry
->d_inode
->i_mode
))
782 if (nd
->path
.dentry
->d_inode
->i_mode
& S_ISVTX
) {
783 if (current
->uid
!= nd
->path
.dentry
->d_inode
->i_uid
)
786 if (vfs_permission(nd
, MAY_WRITE
))
792 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
797 if (d
== NULL
|| d
== d
->d_parent
)
803 struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
,
806 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
809 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(mnt
))
812 res
= q
= clone_mnt(mnt
, dentry
, flag
);
815 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
818 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
819 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
822 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
823 if (!(flag
& CL_COPY_ALL
) && IS_MNT_UNBINDABLE(s
)) {
824 s
= skip_mnt_tree(s
);
827 while (p
!= s
->mnt_parent
) {
833 path
.dentry
= p
->mnt_mountpoint
;
834 q
= clone_mnt(p
, p
->mnt_root
, flag
);
837 spin_lock(&vfsmount_lock
);
838 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
839 attach_mnt(q
, &path
);
840 spin_unlock(&vfsmount_lock
);
846 LIST_HEAD(umount_list
);
847 spin_lock(&vfsmount_lock
);
848 umount_tree(res
, 0, &umount_list
);
849 spin_unlock(&vfsmount_lock
);
850 release_mounts(&umount_list
);
855 struct vfsmount
*collect_mounts(struct vfsmount
*mnt
, struct dentry
*dentry
)
857 struct vfsmount
*tree
;
858 down_read(&namespace_sem
);
859 tree
= copy_tree(mnt
, dentry
, CL_COPY_ALL
| CL_PRIVATE
);
860 up_read(&namespace_sem
);
864 void drop_collected_mounts(struct vfsmount
*mnt
)
866 LIST_HEAD(umount_list
);
867 down_read(&namespace_sem
);
868 spin_lock(&vfsmount_lock
);
869 umount_tree(mnt
, 0, &umount_list
);
870 spin_unlock(&vfsmount_lock
);
871 up_read(&namespace_sem
);
872 release_mounts(&umount_list
);
876 * @source_mnt : mount tree to be attached
877 * @nd : place the mount tree @source_mnt is attached
878 * @parent_nd : if non-null, detach the source_mnt from its parent and
879 * store the parent mount and mountpoint dentry.
880 * (done when source_mnt is moved)
882 * NOTE: in the table below explains the semantics when a source mount
883 * of a given type is attached to a destination mount of a given type.
884 * ---------------------------------------------------------------------------
885 * | BIND MOUNT OPERATION |
886 * |**************************************************************************
887 * | source-->| shared | private | slave | unbindable |
891 * |**************************************************************************
892 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
894 * |non-shared| shared (+) | private | slave (*) | invalid |
895 * ***************************************************************************
896 * A bind operation clones the source mount and mounts the clone on the
899 * (++) the cloned mount is propagated to all the mounts in the propagation
900 * tree of the destination mount and the cloned mount is added to
901 * the peer group of the source mount.
902 * (+) the cloned mount is created under the destination mount and is marked
903 * as shared. The cloned mount is added to the peer group of the source
905 * (+++) the mount is propagated to all the mounts in the propagation tree
906 * of the destination mount and the cloned mount is made slave
907 * of the same master as that of the source mount. The cloned mount
908 * is marked as 'shared and slave'.
909 * (*) the cloned mount is made a slave of the same master as that of the
912 * ---------------------------------------------------------------------------
913 * | MOVE MOUNT OPERATION |
914 * |**************************************************************************
915 * | source-->| shared | private | slave | unbindable |
919 * |**************************************************************************
920 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
922 * |non-shared| shared (+*) | private | slave (*) | unbindable |
923 * ***************************************************************************
925 * (+) the mount is moved to the destination. And is then propagated to
926 * all the mounts in the propagation tree of the destination mount.
927 * (+*) the mount is moved to the destination.
928 * (+++) the mount is moved to the destination and is then propagated to
929 * all the mounts belonging to the destination mount's propagation tree.
930 * the mount is marked as 'shared and slave'.
931 * (*) the mount continues to be a slave at the new location.
933 * if the source mount is a tree, the operations explained above is
934 * applied to each mount in the tree.
935 * Must be called without spinlocks held, since this function can sleep
938 static int attach_recursive_mnt(struct vfsmount
*source_mnt
,
939 struct path
*path
, struct path
*parent_path
)
941 LIST_HEAD(tree_list
);
942 struct vfsmount
*dest_mnt
= path
->mnt
;
943 struct dentry
*dest_dentry
= path
->dentry
;
944 struct vfsmount
*child
, *p
;
946 if (propagate_mnt(dest_mnt
, dest_dentry
, source_mnt
, &tree_list
))
949 if (IS_MNT_SHARED(dest_mnt
)) {
950 for (p
= source_mnt
; p
; p
= next_mnt(p
, source_mnt
))
954 spin_lock(&vfsmount_lock
);
956 detach_mnt(source_mnt
, parent_path
);
957 attach_mnt(source_mnt
, path
);
958 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
960 mnt_set_mountpoint(dest_mnt
, dest_dentry
, source_mnt
);
961 commit_tree(source_mnt
);
964 list_for_each_entry_safe(child
, p
, &tree_list
, mnt_hash
) {
965 list_del_init(&child
->mnt_hash
);
968 spin_unlock(&vfsmount_lock
);
972 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
975 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
978 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
979 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
983 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
984 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
987 err
= security_sb_check_sb(mnt
, nd
);
992 if (IS_ROOT(nd
->path
.dentry
) || !d_unhashed(nd
->path
.dentry
))
993 err
= attach_recursive_mnt(mnt
, &nd
->path
, NULL
);
995 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
997 security_sb_post_addmount(mnt
, nd
);
1002 * recursively change the type of the mountpoint.
1003 * noinline this do_mount helper to save do_mount stack space.
1005 static noinline
int do_change_type(struct nameidata
*nd
, int flag
)
1007 struct vfsmount
*m
, *mnt
= nd
->path
.mnt
;
1008 int recurse
= flag
& MS_REC
;
1009 int type
= flag
& ~MS_REC
;
1011 if (!capable(CAP_SYS_ADMIN
))
1014 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1017 down_write(&namespace_sem
);
1018 spin_lock(&vfsmount_lock
);
1019 for (m
= mnt
; m
; m
= (recurse
? next_mnt(m
, mnt
) : NULL
))
1020 change_mnt_propagation(m
, type
);
1021 spin_unlock(&vfsmount_lock
);
1022 up_write(&namespace_sem
);
1027 * do loopback mount.
1028 * noinline this do_mount helper to save do_mount stack space.
1030 static noinline
int do_loopback(struct nameidata
*nd
, char *old_name
,
1033 struct nameidata old_nd
;
1034 struct vfsmount
*mnt
= NULL
;
1035 int err
= mount_is_safe(nd
);
1038 if (!old_name
|| !*old_name
)
1040 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1044 down_write(&namespace_sem
);
1046 if (IS_MNT_UNBINDABLE(old_nd
.path
.mnt
))
1049 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1054 mnt
= copy_tree(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1056 mnt
= clone_mnt(old_nd
.path
.mnt
, old_nd
.path
.dentry
, 0);
1061 err
= graft_tree(mnt
, nd
);
1063 LIST_HEAD(umount_list
);
1064 spin_lock(&vfsmount_lock
);
1065 umount_tree(mnt
, 0, &umount_list
);
1066 spin_unlock(&vfsmount_lock
);
1067 release_mounts(&umount_list
);
1071 up_write(&namespace_sem
);
1072 path_put(&old_nd
.path
);
1077 * change filesystem flags. dir should be a physical root of filesystem.
1078 * If you've mounted a non-root directory somewhere and want to do remount
1079 * on it - tough luck.
1080 * noinline this do_mount helper to save do_mount stack space.
1082 static noinline
int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
1086 struct super_block
*sb
= nd
->path
.mnt
->mnt_sb
;
1088 if (!capable(CAP_SYS_ADMIN
))
1091 if (!check_mnt(nd
->path
.mnt
))
1094 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
)
1097 down_write(&sb
->s_umount
);
1098 err
= do_remount_sb(sb
, flags
, data
, 0);
1100 nd
->path
.mnt
->mnt_flags
= mnt_flags
;
1101 up_write(&sb
->s_umount
);
1103 security_sb_post_remount(nd
->path
.mnt
, flags
, data
);
1107 static inline int tree_contains_unbindable(struct vfsmount
*mnt
)
1110 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
1111 if (IS_MNT_UNBINDABLE(p
))
1118 * noinline this do_mount helper to save do_mount stack space.
1120 static noinline
int do_move_mount(struct nameidata
*nd
, char *old_name
)
1122 struct nameidata old_nd
;
1123 struct path parent_path
;
1126 if (!capable(CAP_SYS_ADMIN
))
1128 if (!old_name
|| !*old_name
)
1130 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
1134 down_write(&namespace_sem
);
1135 while (d_mountpoint(nd
->path
.dentry
) &&
1136 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1139 if (!check_mnt(nd
->path
.mnt
) || !check_mnt(old_nd
.path
.mnt
))
1143 mutex_lock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1144 if (IS_DEADDIR(nd
->path
.dentry
->d_inode
))
1147 if (!IS_ROOT(nd
->path
.dentry
) && d_unhashed(nd
->path
.dentry
))
1151 if (old_nd
.path
.dentry
!= old_nd
.path
.mnt
->mnt_root
)
1154 if (old_nd
.path
.mnt
== old_nd
.path
.mnt
->mnt_parent
)
1157 if (S_ISDIR(nd
->path
.dentry
->d_inode
->i_mode
) !=
1158 S_ISDIR(old_nd
.path
.dentry
->d_inode
->i_mode
))
1161 * Don't move a mount residing in a shared parent.
1163 if (old_nd
.path
.mnt
->mnt_parent
&&
1164 IS_MNT_SHARED(old_nd
.path
.mnt
->mnt_parent
))
1167 * Don't move a mount tree containing unbindable mounts to a destination
1168 * mount which is shared.
1170 if (IS_MNT_SHARED(nd
->path
.mnt
) &&
1171 tree_contains_unbindable(old_nd
.path
.mnt
))
1174 for (p
= nd
->path
.mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
1175 if (p
== old_nd
.path
.mnt
)
1178 err
= attach_recursive_mnt(old_nd
.path
.mnt
, &nd
->path
, &parent_path
);
1182 /* if the mount is moved, it should no longer be expire
1184 list_del_init(&old_nd
.path
.mnt
->mnt_expire
);
1186 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
1188 up_write(&namespace_sem
);
1190 path_put(&parent_path
);
1191 path_put(&old_nd
.path
);
1196 * create a new mount for userspace and request it to be added into the
1198 * noinline this do_mount helper to save do_mount stack space.
1200 static noinline
int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
1201 int mnt_flags
, char *name
, void *data
)
1203 struct vfsmount
*mnt
;
1205 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
1208 /* we need capabilities... */
1209 if (!capable(CAP_SYS_ADMIN
))
1212 mnt
= do_kern_mount(type
, flags
, name
, data
);
1214 return PTR_ERR(mnt
);
1216 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
1220 * add a mount into a namespace's mount tree
1221 * - provide the option of adding the new mount to an expiration list
1223 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
1224 int mnt_flags
, struct list_head
*fslist
)
1228 down_write(&namespace_sem
);
1229 /* Something was mounted here while we slept */
1230 while (d_mountpoint(nd
->path
.dentry
) &&
1231 follow_down(&nd
->path
.mnt
, &nd
->path
.dentry
))
1234 if (!check_mnt(nd
->path
.mnt
))
1237 /* Refuse the same filesystem on the same mount point */
1239 if (nd
->path
.mnt
->mnt_sb
== newmnt
->mnt_sb
&&
1240 nd
->path
.mnt
->mnt_root
== nd
->path
.dentry
)
1244 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
1247 newmnt
->mnt_flags
= mnt_flags
;
1248 if ((err
= graft_tree(newmnt
, nd
)))
1251 if (fslist
) /* add to the specified expiration list */
1252 list_add_tail(&newmnt
->mnt_expire
, fslist
);
1254 up_write(&namespace_sem
);
1258 up_write(&namespace_sem
);
1263 EXPORT_SYMBOL_GPL(do_add_mount
);
1266 * process a list of expirable mountpoints with the intent of discarding any
1267 * mountpoints that aren't in use and haven't been touched since last we came
1270 void mark_mounts_for_expiry(struct list_head
*mounts
)
1272 struct vfsmount
*mnt
, *next
;
1273 LIST_HEAD(graveyard
);
1276 if (list_empty(mounts
))
1279 down_write(&namespace_sem
);
1280 spin_lock(&vfsmount_lock
);
1282 /* extract from the expiration list every vfsmount that matches the
1283 * following criteria:
1284 * - only referenced by its parent vfsmount
1285 * - still marked for expiry (marked on the last call here; marks are
1286 * cleared by mntput())
1288 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
1289 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
1290 propagate_mount_busy(mnt
, 1))
1292 list_move(&mnt
->mnt_expire
, &graveyard
);
1294 while (!list_empty(&graveyard
)) {
1295 mnt
= list_first_entry(&graveyard
, struct vfsmount
, mnt_expire
);
1296 touch_mnt_namespace(mnt
->mnt_ns
);
1297 umount_tree(mnt
, 1, &umounts
);
1299 spin_unlock(&vfsmount_lock
);
1300 up_write(&namespace_sem
);
1302 release_mounts(&umounts
);
1305 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
1308 * Ripoff of 'select_parent()'
1310 * search the list of submounts for a given mountpoint, and move any
1311 * shrinkable submounts to the 'graveyard' list.
1313 static int select_submounts(struct vfsmount
*parent
, struct list_head
*graveyard
)
1315 struct vfsmount
*this_parent
= parent
;
1316 struct list_head
*next
;
1320 next
= this_parent
->mnt_mounts
.next
;
1322 while (next
!= &this_parent
->mnt_mounts
) {
1323 struct list_head
*tmp
= next
;
1324 struct vfsmount
*mnt
= list_entry(tmp
, struct vfsmount
, mnt_child
);
1327 if (!(mnt
->mnt_flags
& MNT_SHRINKABLE
))
1330 * Descend a level if the d_mounts list is non-empty.
1332 if (!list_empty(&mnt
->mnt_mounts
)) {
1337 if (!propagate_mount_busy(mnt
, 1)) {
1338 list_move_tail(&mnt
->mnt_expire
, graveyard
);
1343 * All done at this level ... ascend and resume the search
1345 if (this_parent
!= parent
) {
1346 next
= this_parent
->mnt_child
.next
;
1347 this_parent
= this_parent
->mnt_parent
;
1354 * process a list of expirable mountpoints with the intent of discarding any
1355 * submounts of a specific parent mountpoint
1357 static void shrink_submounts(struct vfsmount
*mnt
, struct list_head
*umounts
)
1359 LIST_HEAD(graveyard
);
1362 /* extract submounts of 'mountpoint' from the expiration list */
1363 while (select_submounts(mnt
, &graveyard
)) {
1364 while (!list_empty(&graveyard
)) {
1365 m
= list_first_entry(&graveyard
, struct vfsmount
,
1367 touch_mnt_namespace(mnt
->mnt_ns
);
1368 umount_tree(mnt
, 1, umounts
);
1374 * Some copy_from_user() implementations do not return the exact number of
1375 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1376 * Note that this function differs from copy_from_user() in that it will oops
1377 * on bad values of `to', rather than returning a short copy.
1379 static long exact_copy_from_user(void *to
, const void __user
* from
,
1383 const char __user
*f
= from
;
1386 if (!access_ok(VERIFY_READ
, from
, n
))
1390 if (__get_user(c
, f
)) {
1401 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1411 if (!(page
= __get_free_page(GFP_KERNEL
)))
1414 /* We only care that *some* data at the address the user
1415 * gave us is valid. Just in case, we'll zero
1416 * the remainder of the page.
1418 /* copy_from_user cannot cross TASK_SIZE ! */
1419 size
= TASK_SIZE
- (unsigned long)data
;
1420 if (size
> PAGE_SIZE
)
1423 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1429 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1435 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1436 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1438 * data is a (void *) that can point to any structure up to
1439 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1440 * information (or be NULL).
1442 * Pre-0.97 versions of mount() didn't have a flags word.
1443 * When the flags word was introduced its top half was required
1444 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1445 * Therefore, if this magic number is present, it carries no information
1446 * and must be discarded.
1448 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1449 unsigned long flags
, void *data_page
)
1451 struct nameidata nd
;
1456 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1457 flags
&= ~MS_MGC_MSK
;
1459 /* Basic sanity checks */
1461 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1463 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1467 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1469 /* Separate the per-mountpoint flags */
1470 if (flags
& MS_NOSUID
)
1471 mnt_flags
|= MNT_NOSUID
;
1472 if (flags
& MS_NODEV
)
1473 mnt_flags
|= MNT_NODEV
;
1474 if (flags
& MS_NOEXEC
)
1475 mnt_flags
|= MNT_NOEXEC
;
1476 if (flags
& MS_NOATIME
)
1477 mnt_flags
|= MNT_NOATIME
;
1478 if (flags
& MS_NODIRATIME
)
1479 mnt_flags
|= MNT_NODIRATIME
;
1480 if (flags
& MS_RELATIME
)
1481 mnt_flags
|= MNT_RELATIME
;
1483 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
|
1484 MS_NOATIME
| MS_NODIRATIME
| MS_RELATIME
| MS_KERNMOUNT
);
1486 /* ... and get the mountpoint */
1487 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1491 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1495 if (flags
& MS_REMOUNT
)
1496 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1498 else if (flags
& MS_BIND
)
1499 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1500 else if (flags
& (MS_SHARED
| MS_PRIVATE
| MS_SLAVE
| MS_UNBINDABLE
))
1501 retval
= do_change_type(&nd
, flags
);
1502 else if (flags
& MS_MOVE
)
1503 retval
= do_move_mount(&nd
, dev_name
);
1505 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1506 dev_name
, data_page
);
1513 * Allocate a new namespace structure and populate it with contents
1514 * copied from the namespace of the passed in task structure.
1516 static struct mnt_namespace
*dup_mnt_ns(struct mnt_namespace
*mnt_ns
,
1517 struct fs_struct
*fs
)
1519 struct mnt_namespace
*new_ns
;
1520 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1521 struct vfsmount
*p
, *q
;
1523 new_ns
= kmalloc(sizeof(struct mnt_namespace
), GFP_KERNEL
);
1525 return ERR_PTR(-ENOMEM
);
1527 atomic_set(&new_ns
->count
, 1);
1528 INIT_LIST_HEAD(&new_ns
->list
);
1529 init_waitqueue_head(&new_ns
->poll
);
1532 down_write(&namespace_sem
);
1533 /* First pass: copy the tree topology */
1534 new_ns
->root
= copy_tree(mnt_ns
->root
, mnt_ns
->root
->mnt_root
,
1535 CL_COPY_ALL
| CL_EXPIRE
);
1536 if (!new_ns
->root
) {
1537 up_write(&namespace_sem
);
1539 return ERR_PTR(-ENOMEM
);;
1541 spin_lock(&vfsmount_lock
);
1542 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1543 spin_unlock(&vfsmount_lock
);
1546 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1547 * as belonging to new namespace. We have already acquired a private
1548 * fs_struct, so tsk->fs->lock is not needed.
1555 if (p
== fs
->root
.mnt
) {
1557 fs
->root
.mnt
= mntget(q
);
1559 if (p
== fs
->pwd
.mnt
) {
1561 fs
->pwd
.mnt
= mntget(q
);
1563 if (p
== fs
->altroot
.mnt
) {
1565 fs
->altroot
.mnt
= mntget(q
);
1568 p
= next_mnt(p
, mnt_ns
->root
);
1569 q
= next_mnt(q
, new_ns
->root
);
1571 up_write(&namespace_sem
);
1583 struct mnt_namespace
*copy_mnt_ns(unsigned long flags
, struct mnt_namespace
*ns
,
1584 struct fs_struct
*new_fs
)
1586 struct mnt_namespace
*new_ns
;
1591 if (!(flags
& CLONE_NEWNS
))
1594 new_ns
= dup_mnt_ns(ns
, new_fs
);
1600 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1601 char __user
* type
, unsigned long flags
,
1605 unsigned long data_page
;
1606 unsigned long type_page
;
1607 unsigned long dev_page
;
1610 retval
= copy_mount_options(type
, &type_page
);
1614 dir_page
= getname(dir_name
);
1615 retval
= PTR_ERR(dir_page
);
1616 if (IS_ERR(dir_page
))
1619 retval
= copy_mount_options(dev_name
, &dev_page
);
1623 retval
= copy_mount_options(data
, &data_page
);
1628 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1629 flags
, (void *)data_page
);
1631 free_page(data_page
);
1634 free_page(dev_page
);
1638 free_page(type_page
);
1643 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1644 * It can block. Requires the big lock held.
1646 void set_fs_root(struct fs_struct
*fs
, struct path
*path
)
1648 struct path old_root
;
1650 write_lock(&fs
->lock
);
1651 old_root
= fs
->root
;
1654 write_unlock(&fs
->lock
);
1655 if (old_root
.dentry
)
1656 path_put(&old_root
);
1660 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1661 * It can block. Requires the big lock held.
1663 void set_fs_pwd(struct fs_struct
*fs
, struct path
*path
)
1665 struct path old_pwd
;
1667 write_lock(&fs
->lock
);
1671 write_unlock(&fs
->lock
);
1677 static void chroot_fs_refs(struct path
*old_root
, struct path
*new_root
)
1679 struct task_struct
*g
, *p
;
1680 struct fs_struct
*fs
;
1682 read_lock(&tasklist_lock
);
1683 do_each_thread(g
, p
) {
1687 atomic_inc(&fs
->count
);
1689 if (fs
->root
.dentry
== old_root
->dentry
1690 && fs
->root
.mnt
== old_root
->mnt
)
1691 set_fs_root(fs
, new_root
);
1692 if (fs
->pwd
.dentry
== old_root
->dentry
1693 && fs
->pwd
.mnt
== old_root
->mnt
)
1694 set_fs_pwd(fs
, new_root
);
1698 } while_each_thread(g
, p
);
1699 read_unlock(&tasklist_lock
);
1703 * pivot_root Semantics:
1704 * Moves the root file system of the current process to the directory put_old,
1705 * makes new_root as the new root file system of the current process, and sets
1706 * root/cwd of all processes which had them on the current root to new_root.
1709 * The new_root and put_old must be directories, and must not be on the
1710 * same file system as the current process root. The put_old must be
1711 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1712 * pointed to by put_old must yield the same directory as new_root. No other
1713 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1715 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1716 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1717 * in this situation.
1720 * - we don't move root/cwd if they are not at the root (reason: if something
1721 * cared enough to change them, it's probably wrong to force them elsewhere)
1722 * - it's okay to pick a root that isn't the root of a file system, e.g.
1723 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1724 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1727 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1728 const char __user
* put_old
)
1730 struct vfsmount
*tmp
;
1731 struct nameidata new_nd
, old_nd
, user_nd
;
1732 struct path parent_path
, root_parent
;
1735 if (!capable(CAP_SYS_ADMIN
))
1740 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1745 if (!check_mnt(new_nd
.path
.mnt
))
1748 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1752 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1754 path_put(&old_nd
.path
);
1758 read_lock(¤t
->fs
->lock
);
1759 user_nd
.path
= current
->fs
->root
;
1760 path_get(¤t
->fs
->root
);
1761 read_unlock(¤t
->fs
->lock
);
1762 down_write(&namespace_sem
);
1763 mutex_lock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
1765 if (IS_MNT_SHARED(old_nd
.path
.mnt
) ||
1766 IS_MNT_SHARED(new_nd
.path
.mnt
->mnt_parent
) ||
1767 IS_MNT_SHARED(user_nd
.path
.mnt
->mnt_parent
))
1769 if (!check_mnt(user_nd
.path
.mnt
))
1772 if (IS_DEADDIR(new_nd
.path
.dentry
->d_inode
))
1774 if (d_unhashed(new_nd
.path
.dentry
) && !IS_ROOT(new_nd
.path
.dentry
))
1776 if (d_unhashed(old_nd
.path
.dentry
) && !IS_ROOT(old_nd
.path
.dentry
))
1779 if (new_nd
.path
.mnt
== user_nd
.path
.mnt
||
1780 old_nd
.path
.mnt
== user_nd
.path
.mnt
)
1781 goto out2
; /* loop, on the same file system */
1783 if (user_nd
.path
.mnt
->mnt_root
!= user_nd
.path
.dentry
)
1784 goto out2
; /* not a mountpoint */
1785 if (user_nd
.path
.mnt
->mnt_parent
== user_nd
.path
.mnt
)
1786 goto out2
; /* not attached */
1787 if (new_nd
.path
.mnt
->mnt_root
!= new_nd
.path
.dentry
)
1788 goto out2
; /* not a mountpoint */
1789 if (new_nd
.path
.mnt
->mnt_parent
== new_nd
.path
.mnt
)
1790 goto out2
; /* not attached */
1791 /* make sure we can reach put_old from new_root */
1792 tmp
= old_nd
.path
.mnt
;
1793 spin_lock(&vfsmount_lock
);
1794 if (tmp
!= new_nd
.path
.mnt
) {
1796 if (tmp
->mnt_parent
== tmp
)
1797 goto out3
; /* already mounted on put_old */
1798 if (tmp
->mnt_parent
== new_nd
.path
.mnt
)
1800 tmp
= tmp
->mnt_parent
;
1802 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.path
.dentry
))
1804 } else if (!is_subdir(old_nd
.path
.dentry
, new_nd
.path
.dentry
))
1806 detach_mnt(new_nd
.path
.mnt
, &parent_path
);
1807 detach_mnt(user_nd
.path
.mnt
, &root_parent
);
1808 /* mount old root on put_old */
1809 attach_mnt(user_nd
.path
.mnt
, &old_nd
.path
);
1810 /* mount new_root on / */
1811 attach_mnt(new_nd
.path
.mnt
, &root_parent
);
1812 touch_mnt_namespace(current
->nsproxy
->mnt_ns
);
1813 spin_unlock(&vfsmount_lock
);
1814 chroot_fs_refs(&user_nd
.path
, &new_nd
.path
);
1815 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1817 path_put(&root_parent
);
1818 path_put(&parent_path
);
1820 mutex_unlock(&old_nd
.path
.dentry
->d_inode
->i_mutex
);
1821 up_write(&namespace_sem
);
1822 path_put(&user_nd
.path
);
1823 path_put(&old_nd
.path
);
1825 path_put(&new_nd
.path
);
1830 spin_unlock(&vfsmount_lock
);
1834 static void __init
init_mount_tree(void)
1836 struct vfsmount
*mnt
;
1837 struct mnt_namespace
*ns
;
1840 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1842 panic("Can't create rootfs");
1843 ns
= kmalloc(sizeof(*ns
), GFP_KERNEL
);
1845 panic("Can't allocate initial namespace");
1846 atomic_set(&ns
->count
, 1);
1847 INIT_LIST_HEAD(&ns
->list
);
1848 init_waitqueue_head(&ns
->poll
);
1850 list_add(&mnt
->mnt_list
, &ns
->list
);
1854 init_task
.nsproxy
->mnt_ns
= ns
;
1857 root
.mnt
= ns
->root
;
1858 root
.dentry
= ns
->root
->mnt_root
;
1860 set_fs_pwd(current
->fs
, &root
);
1861 set_fs_root(current
->fs
, &root
);
1864 void __init
mnt_init(void)
1869 init_rwsem(&namespace_sem
);
1871 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1872 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
);
1874 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1876 if (!mount_hashtable
)
1877 panic("Failed to allocate mount hash table\n");
1879 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE
);
1881 for (u
= 0; u
< HASH_SIZE
; u
++)
1882 INIT_LIST_HEAD(&mount_hashtable
[u
]);
1886 printk(KERN_WARNING
"%s: sysfs_init error: %d\n",
1888 fs_kobj
= kobject_create_and_add("fs", NULL
);
1890 printk(KERN_WARNING
"%s: kobj create error\n", __FUNCTION__
);
1895 void __put_mnt_ns(struct mnt_namespace
*ns
)
1897 struct vfsmount
*root
= ns
->root
;
1898 LIST_HEAD(umount_list
);
1900 spin_unlock(&vfsmount_lock
);
1901 down_write(&namespace_sem
);
1902 spin_lock(&vfsmount_lock
);
1903 umount_tree(root
, 0, &umount_list
);
1904 spin_unlock(&vfsmount_lock
);
1905 up_write(&namespace_sem
);
1906 release_mounts(&umount_list
);