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/config.h>
12 #include <linux/syscalls.h>
13 #include <linux/slab.h>
14 #include <linux/sched.h>
15 #include <linux/smp_lock.h>
16 #include <linux/init.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/module.h>
20 #include <linux/seq_file.h>
21 #include <linux/namespace.h>
22 #include <linux/namei.h>
23 #include <linux/security.h>
24 #include <linux/mount.h>
25 #include <asm/uaccess.h>
26 #include <asm/unistd.h>
28 extern int __init
init_rootfs(void);
31 extern int __init
sysfs_init(void);
33 static inline int sysfs_init(void)
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40 __cacheline_aligned_in_smp
DEFINE_SPINLOCK(vfsmount_lock
);
44 static struct list_head
*mount_hashtable
;
45 static int hash_mask __read_mostly
, hash_bits __read_mostly
;
46 static kmem_cache_t
*mnt_cache
;
48 static inline unsigned long hash(struct vfsmount
*mnt
, struct dentry
*dentry
)
50 unsigned long tmp
= ((unsigned long)mnt
/ L1_CACHE_BYTES
);
51 tmp
+= ((unsigned long)dentry
/ L1_CACHE_BYTES
);
52 tmp
= tmp
+ (tmp
>> hash_bits
);
53 return tmp
& hash_mask
;
56 struct vfsmount
*alloc_vfsmnt(const char *name
)
58 struct vfsmount
*mnt
= kmem_cache_alloc(mnt_cache
, GFP_KERNEL
);
60 memset(mnt
, 0, sizeof(struct vfsmount
));
61 atomic_set(&mnt
->mnt_count
, 1);
62 INIT_LIST_HEAD(&mnt
->mnt_hash
);
63 INIT_LIST_HEAD(&mnt
->mnt_child
);
64 INIT_LIST_HEAD(&mnt
->mnt_mounts
);
65 INIT_LIST_HEAD(&mnt
->mnt_list
);
66 INIT_LIST_HEAD(&mnt
->mnt_expire
);
68 int size
= strlen(name
) + 1;
69 char *newname
= kmalloc(size
, GFP_KERNEL
);
71 memcpy(newname
, name
, size
);
72 mnt
->mnt_devname
= newname
;
79 void free_vfsmnt(struct vfsmount
*mnt
)
81 kfree(mnt
->mnt_devname
);
82 kmem_cache_free(mnt_cache
, mnt
);
86 * Now, lookup_mnt increments the ref count before returning
87 * the vfsmount struct.
89 struct vfsmount
*lookup_mnt(struct vfsmount
*mnt
, struct dentry
*dentry
)
91 struct list_head
*head
= mount_hashtable
+ hash(mnt
, dentry
);
92 struct list_head
*tmp
= head
;
93 struct vfsmount
*p
, *found
= NULL
;
95 spin_lock(&vfsmount_lock
);
101 p
= list_entry(tmp
, struct vfsmount
, mnt_hash
);
102 if (p
->mnt_parent
== mnt
&& p
->mnt_mountpoint
== dentry
) {
107 spin_unlock(&vfsmount_lock
);
111 static inline int check_mnt(struct vfsmount
*mnt
)
113 return mnt
->mnt_namespace
== current
->namespace;
116 static void touch_namespace(struct namespace *ns
)
120 wake_up_interruptible(&ns
->poll
);
124 static void __touch_namespace(struct namespace *ns
)
126 if (ns
&& ns
->event
!= event
) {
128 wake_up_interruptible(&ns
->poll
);
132 static void detach_mnt(struct vfsmount
*mnt
, struct nameidata
*old_nd
)
134 old_nd
->dentry
= mnt
->mnt_mountpoint
;
135 old_nd
->mnt
= mnt
->mnt_parent
;
136 mnt
->mnt_parent
= mnt
;
137 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
138 list_del_init(&mnt
->mnt_child
);
139 list_del_init(&mnt
->mnt_hash
);
140 old_nd
->dentry
->d_mounted
--;
143 static void attach_mnt(struct vfsmount
*mnt
, struct nameidata
*nd
)
145 mnt
->mnt_parent
= mntget(nd
->mnt
);
146 mnt
->mnt_mountpoint
= dget(nd
->dentry
);
147 list_add(&mnt
->mnt_hash
, mount_hashtable
+ hash(nd
->mnt
, nd
->dentry
));
148 list_add_tail(&mnt
->mnt_child
, &nd
->mnt
->mnt_mounts
);
149 nd
->dentry
->d_mounted
++;
152 static struct vfsmount
*next_mnt(struct vfsmount
*p
, struct vfsmount
*root
)
154 struct list_head
*next
= p
->mnt_mounts
.next
;
155 if (next
== &p
->mnt_mounts
) {
159 next
= p
->mnt_child
.next
;
160 if (next
!= &p
->mnt_parent
->mnt_mounts
)
165 return list_entry(next
, struct vfsmount
, mnt_child
);
168 static struct vfsmount
*clone_mnt(struct vfsmount
*old
, struct dentry
*root
)
170 struct super_block
*sb
= old
->mnt_sb
;
171 struct vfsmount
*mnt
= alloc_vfsmnt(old
->mnt_devname
);
174 mnt
->mnt_flags
= old
->mnt_flags
;
175 atomic_inc(&sb
->s_active
);
177 mnt
->mnt_root
= dget(root
);
178 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
179 mnt
->mnt_parent
= mnt
;
180 mnt
->mnt_namespace
= current
->namespace;
182 /* stick the duplicate mount on the same expiry list
183 * as the original if that was on one */
184 spin_lock(&vfsmount_lock
);
185 if (!list_empty(&old
->mnt_expire
))
186 list_add(&mnt
->mnt_expire
, &old
->mnt_expire
);
187 spin_unlock(&vfsmount_lock
);
192 static inline void __mntput(struct vfsmount
*mnt
)
194 struct super_block
*sb
= mnt
->mnt_sb
;
197 deactivate_super(sb
);
200 void mntput_no_expire(struct vfsmount
*mnt
)
203 if (atomic_dec_and_lock(&mnt
->mnt_count
, &vfsmount_lock
)) {
204 if (likely(!mnt
->mnt_pinned
)) {
205 spin_unlock(&vfsmount_lock
);
209 atomic_add(mnt
->mnt_pinned
+ 1, &mnt
->mnt_count
);
211 spin_unlock(&vfsmount_lock
);
212 acct_auto_close_mnt(mnt
);
213 security_sb_umount_close(mnt
);
218 EXPORT_SYMBOL(mntput_no_expire
);
220 void mnt_pin(struct vfsmount
*mnt
)
222 spin_lock(&vfsmount_lock
);
224 spin_unlock(&vfsmount_lock
);
227 EXPORT_SYMBOL(mnt_pin
);
229 void mnt_unpin(struct vfsmount
*mnt
)
231 spin_lock(&vfsmount_lock
);
232 if (mnt
->mnt_pinned
) {
233 atomic_inc(&mnt
->mnt_count
);
236 spin_unlock(&vfsmount_lock
);
239 EXPORT_SYMBOL(mnt_unpin
);
242 static void *m_start(struct seq_file
*m
, loff_t
*pos
)
244 struct namespace *n
= m
->private;
249 list_for_each(p
, &n
->list
)
251 return list_entry(p
, struct vfsmount
, mnt_list
);
255 static void *m_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
257 struct namespace *n
= m
->private;
258 struct list_head
*p
= ((struct vfsmount
*)v
)->mnt_list
.next
;
260 return p
== &n
->list
? NULL
: list_entry(p
, struct vfsmount
, mnt_list
);
263 static void m_stop(struct seq_file
*m
, void *v
)
265 struct namespace *n
= m
->private;
269 static inline void mangle(struct seq_file
*m
, const char *s
)
271 seq_escape(m
, s
, " \t\n\\");
274 static int show_vfsmnt(struct seq_file
*m
, void *v
)
276 struct vfsmount
*mnt
= v
;
278 static struct proc_fs_info
{
282 { MS_SYNCHRONOUS
, ",sync" },
283 { MS_DIRSYNC
, ",dirsync" },
284 { MS_MANDLOCK
, ",mand" },
285 { MS_NOATIME
, ",noatime" },
286 { MS_NODIRATIME
, ",nodiratime" },
289 static struct proc_fs_info mnt_info
[] = {
290 { MNT_NOSUID
, ",nosuid" },
291 { MNT_NODEV
, ",nodev" },
292 { MNT_NOEXEC
, ",noexec" },
295 struct proc_fs_info
*fs_infop
;
297 mangle(m
, mnt
->mnt_devname
? mnt
->mnt_devname
: "none");
299 seq_path(m
, mnt
, mnt
->mnt_root
, " \t\n\\");
301 mangle(m
, mnt
->mnt_sb
->s_type
->name
);
302 seq_puts(m
, mnt
->mnt_sb
->s_flags
& MS_RDONLY
? " ro" : " rw");
303 for (fs_infop
= fs_info
; fs_infop
->flag
; fs_infop
++) {
304 if (mnt
->mnt_sb
->s_flags
& fs_infop
->flag
)
305 seq_puts(m
, fs_infop
->str
);
307 for (fs_infop
= mnt_info
; fs_infop
->flag
; fs_infop
++) {
308 if (mnt
->mnt_flags
& fs_infop
->flag
)
309 seq_puts(m
, fs_infop
->str
);
311 if (mnt
->mnt_sb
->s_op
->show_options
)
312 err
= mnt
->mnt_sb
->s_op
->show_options(m
, mnt
);
313 seq_puts(m
, " 0 0\n");
317 struct seq_operations mounts_op
= {
325 * may_umount_tree - check if a mount tree is busy
326 * @mnt: root of mount tree
328 * This is called to check if a tree of mounts has any
329 * open files, pwds, chroots or sub mounts that are
332 int may_umount_tree(struct vfsmount
*mnt
)
334 struct list_head
*next
;
335 struct vfsmount
*this_parent
= mnt
;
339 spin_lock(&vfsmount_lock
);
340 actual_refs
= atomic_read(&mnt
->mnt_count
);
343 next
= this_parent
->mnt_mounts
.next
;
345 while (next
!= &this_parent
->mnt_mounts
) {
347 list_entry(next
, struct vfsmount
, mnt_child
);
351 actual_refs
+= atomic_read(&p
->mnt_count
);
354 if (!list_empty(&p
->mnt_mounts
)) {
360 if (this_parent
!= mnt
) {
361 next
= this_parent
->mnt_child
.next
;
362 this_parent
= this_parent
->mnt_parent
;
365 spin_unlock(&vfsmount_lock
);
367 if (actual_refs
> minimum_refs
)
373 EXPORT_SYMBOL(may_umount_tree
);
376 * may_umount - check if a mount point is busy
377 * @mnt: root of mount
379 * This is called to check if a mount point has any
380 * open files, pwds, chroots or sub mounts. If the
381 * mount has sub mounts this will return busy
382 * regardless of whether the sub mounts are busy.
384 * Doesn't take quota and stuff into account. IOW, in some cases it will
385 * give false negatives. The main reason why it's here is that we need
386 * a non-destructive way to look for easily umountable filesystems.
388 int may_umount(struct vfsmount
*mnt
)
390 if (atomic_read(&mnt
->mnt_count
) > 2)
395 EXPORT_SYMBOL(may_umount
);
397 static void release_mounts(struct list_head
*head
)
399 struct vfsmount
*mnt
;
400 while(!list_empty(head
)) {
401 mnt
= list_entry(head
->next
, struct vfsmount
, mnt_hash
);
402 list_del_init(&mnt
->mnt_hash
);
403 if (mnt
->mnt_parent
!= mnt
) {
404 struct dentry
*dentry
;
406 spin_lock(&vfsmount_lock
);
407 dentry
= mnt
->mnt_mountpoint
;
409 mnt
->mnt_mountpoint
= mnt
->mnt_root
;
410 mnt
->mnt_parent
= mnt
;
411 spin_unlock(&vfsmount_lock
);
419 static void umount_tree(struct vfsmount
*mnt
, struct list_head
*kill
)
423 for (p
= mnt
; p
; p
= next_mnt(p
, mnt
)) {
424 list_del(&p
->mnt_hash
);
425 list_add(&p
->mnt_hash
, kill
);
428 list_for_each_entry(p
, kill
, mnt_hash
) {
429 list_del_init(&p
->mnt_expire
);
430 list_del_init(&p
->mnt_list
);
431 __touch_namespace(p
->mnt_namespace
);
432 p
->mnt_namespace
= NULL
;
433 list_del_init(&p
->mnt_child
);
434 if (p
->mnt_parent
!= p
)
435 mnt
->mnt_mountpoint
->d_mounted
--;
439 static int do_umount(struct vfsmount
*mnt
, int flags
)
441 struct super_block
*sb
= mnt
->mnt_sb
;
443 LIST_HEAD(umount_list
);
445 retval
= security_sb_umount(mnt
, flags
);
450 * Allow userspace to request a mountpoint be expired rather than
451 * unmounting unconditionally. Unmount only happens if:
452 * (1) the mark is already set (the mark is cleared by mntput())
453 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
455 if (flags
& MNT_EXPIRE
) {
456 if (mnt
== current
->fs
->rootmnt
||
457 flags
& (MNT_FORCE
| MNT_DETACH
))
460 if (atomic_read(&mnt
->mnt_count
) != 2)
463 if (!xchg(&mnt
->mnt_expiry_mark
, 1))
468 * If we may have to abort operations to get out of this
469 * mount, and they will themselves hold resources we must
470 * allow the fs to do things. In the Unix tradition of
471 * 'Gee thats tricky lets do it in userspace' the umount_begin
472 * might fail to complete on the first run through as other tasks
473 * must return, and the like. Thats for the mount program to worry
474 * about for the moment.
478 if ((flags
& MNT_FORCE
) && sb
->s_op
->umount_begin
)
479 sb
->s_op
->umount_begin(sb
);
483 * No sense to grab the lock for this test, but test itself looks
484 * somewhat bogus. Suggestions for better replacement?
485 * Ho-hum... In principle, we might treat that as umount + switch
486 * to rootfs. GC would eventually take care of the old vfsmount.
487 * Actually it makes sense, especially if rootfs would contain a
488 * /reboot - static binary that would close all descriptors and
489 * call reboot(9). Then init(8) could umount root and exec /reboot.
491 if (mnt
== current
->fs
->rootmnt
&& !(flags
& MNT_DETACH
)) {
493 * Special case for "unmounting" root ...
494 * we just try to remount it readonly.
496 down_write(&sb
->s_umount
);
497 if (!(sb
->s_flags
& MS_RDONLY
)) {
500 retval
= do_remount_sb(sb
, MS_RDONLY
, NULL
, 0);
503 up_write(&sb
->s_umount
);
507 down_write(¤t
->namespace->sem
);
508 spin_lock(&vfsmount_lock
);
512 if (atomic_read(&mnt
->mnt_count
) == 2 || flags
& MNT_DETACH
) {
513 if (!list_empty(&mnt
->mnt_list
))
514 umount_tree(mnt
, &umount_list
);
517 spin_unlock(&vfsmount_lock
);
519 security_sb_umount_busy(mnt
);
520 up_write(¤t
->namespace->sem
);
521 release_mounts(&umount_list
);
526 * Now umount can handle mount points as well as block devices.
527 * This is important for filesystems which use unnamed block devices.
529 * We now support a flag for forced unmount like the other 'big iron'
530 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
533 asmlinkage
long sys_umount(char __user
* name
, int flags
)
538 retval
= __user_walk(name
, LOOKUP_FOLLOW
, &nd
);
542 if (nd
.dentry
!= nd
.mnt
->mnt_root
)
544 if (!check_mnt(nd
.mnt
))
548 if (!capable(CAP_SYS_ADMIN
))
551 retval
= do_umount(nd
.mnt
, flags
);
553 path_release_on_umount(&nd
);
558 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
561 * The 2.0 compatible umount. No flags.
563 asmlinkage
long sys_oldumount(char __user
* name
)
565 return sys_umount(name
, 0);
570 static int mount_is_safe(struct nameidata
*nd
)
572 if (capable(CAP_SYS_ADMIN
))
576 if (S_ISLNK(nd
->dentry
->d_inode
->i_mode
))
578 if (nd
->dentry
->d_inode
->i_mode
& S_ISVTX
) {
579 if (current
->uid
!= nd
->dentry
->d_inode
->i_uid
)
582 if (permission(nd
->dentry
->d_inode
, MAY_WRITE
, nd
))
588 static int lives_below_in_same_fs(struct dentry
*d
, struct dentry
*dentry
)
593 if (d
== NULL
|| d
== d
->d_parent
)
599 static struct vfsmount
*copy_tree(struct vfsmount
*mnt
, struct dentry
*dentry
)
601 struct vfsmount
*res
, *p
, *q
, *r
, *s
;
604 res
= q
= clone_mnt(mnt
, dentry
);
607 q
->mnt_mountpoint
= mnt
->mnt_mountpoint
;
610 list_for_each_entry(r
, &mnt
->mnt_mounts
, mnt_child
) {
611 if (!lives_below_in_same_fs(r
->mnt_mountpoint
, dentry
))
614 for (s
= r
; s
; s
= next_mnt(s
, r
)) {
615 while (p
!= s
->mnt_parent
) {
621 nd
.dentry
= p
->mnt_mountpoint
;
622 q
= clone_mnt(p
, p
->mnt_root
);
625 spin_lock(&vfsmount_lock
);
626 list_add_tail(&q
->mnt_list
, &res
->mnt_list
);
628 spin_unlock(&vfsmount_lock
);
634 LIST_HEAD(umount_list
);
635 spin_lock(&vfsmount_lock
);
636 umount_tree(res
, &umount_list
);
637 spin_unlock(&vfsmount_lock
);
638 release_mounts(&umount_list
);
643 static int graft_tree(struct vfsmount
*mnt
, struct nameidata
*nd
)
646 if (mnt
->mnt_sb
->s_flags
& MS_NOUSER
)
649 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
650 S_ISDIR(mnt
->mnt_root
->d_inode
->i_mode
))
654 down(&nd
->dentry
->d_inode
->i_sem
);
655 if (IS_DEADDIR(nd
->dentry
->d_inode
))
658 err
= security_sb_check_sb(mnt
, nd
);
663 spin_lock(&vfsmount_lock
);
664 if (IS_ROOT(nd
->dentry
) || !d_unhashed(nd
->dentry
)) {
665 struct list_head head
;
668 list_add_tail(&head
, &mnt
->mnt_list
);
669 list_splice(&head
, current
->namespace->list
.prev
);
671 touch_namespace(current
->namespace);
673 spin_unlock(&vfsmount_lock
);
675 up(&nd
->dentry
->d_inode
->i_sem
);
677 security_sb_post_addmount(mnt
, nd
);
684 static int do_loopback(struct nameidata
*nd
, char *old_name
, int recurse
)
686 struct nameidata old_nd
;
687 struct vfsmount
*mnt
= NULL
;
688 int err
= mount_is_safe(nd
);
691 if (!old_name
|| !*old_name
)
693 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
697 down_write(¤t
->namespace->sem
);
699 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
704 mnt
= copy_tree(old_nd
.mnt
, old_nd
.dentry
);
706 mnt
= clone_mnt(old_nd
.mnt
, old_nd
.dentry
);
711 /* stop bind mounts from expiring */
712 spin_lock(&vfsmount_lock
);
713 list_del_init(&mnt
->mnt_expire
);
714 spin_unlock(&vfsmount_lock
);
716 err
= graft_tree(mnt
, nd
);
718 LIST_HEAD(umount_list
);
719 spin_lock(&vfsmount_lock
);
720 umount_tree(mnt
, &umount_list
);
721 spin_unlock(&vfsmount_lock
);
722 release_mounts(&umount_list
);
726 up_write(¤t
->namespace->sem
);
727 path_release(&old_nd
);
732 * change filesystem flags. dir should be a physical root of filesystem.
733 * If you've mounted a non-root directory somewhere and want to do remount
734 * on it - tough luck.
736 static int do_remount(struct nameidata
*nd
, int flags
, int mnt_flags
,
740 struct super_block
*sb
= nd
->mnt
->mnt_sb
;
742 if (!capable(CAP_SYS_ADMIN
))
745 if (!check_mnt(nd
->mnt
))
748 if (nd
->dentry
!= nd
->mnt
->mnt_root
)
751 down_write(&sb
->s_umount
);
752 err
= do_remount_sb(sb
, flags
, data
, 0);
754 nd
->mnt
->mnt_flags
= mnt_flags
;
755 up_write(&sb
->s_umount
);
757 security_sb_post_remount(nd
->mnt
, flags
, data
);
761 static int do_move_mount(struct nameidata
*nd
, char *old_name
)
763 struct nameidata old_nd
, parent_nd
;
766 if (!capable(CAP_SYS_ADMIN
))
768 if (!old_name
|| !*old_name
)
770 err
= path_lookup(old_name
, LOOKUP_FOLLOW
, &old_nd
);
774 down_write(¤t
->namespace->sem
);
775 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
778 if (!check_mnt(nd
->mnt
) || !check_mnt(old_nd
.mnt
))
782 down(&nd
->dentry
->d_inode
->i_sem
);
783 if (IS_DEADDIR(nd
->dentry
->d_inode
))
786 spin_lock(&vfsmount_lock
);
787 if (!IS_ROOT(nd
->dentry
) && d_unhashed(nd
->dentry
))
791 if (old_nd
.dentry
!= old_nd
.mnt
->mnt_root
)
794 if (old_nd
.mnt
== old_nd
.mnt
->mnt_parent
)
797 if (S_ISDIR(nd
->dentry
->d_inode
->i_mode
) !=
798 S_ISDIR(old_nd
.dentry
->d_inode
->i_mode
))
802 for (p
= nd
->mnt
; p
->mnt_parent
!= p
; p
= p
->mnt_parent
)
807 detach_mnt(old_nd
.mnt
, &parent_nd
);
808 attach_mnt(old_nd
.mnt
, nd
);
809 touch_namespace(current
->namespace);
811 /* if the mount is moved, it should no longer be expire
813 list_del_init(&old_nd
.mnt
->mnt_expire
);
815 spin_unlock(&vfsmount_lock
);
817 up(&nd
->dentry
->d_inode
->i_sem
);
819 up_write(¤t
->namespace->sem
);
821 path_release(&parent_nd
);
822 path_release(&old_nd
);
827 * create a new mount for userspace and request it to be added into the
830 static int do_new_mount(struct nameidata
*nd
, char *type
, int flags
,
831 int mnt_flags
, char *name
, void *data
)
833 struct vfsmount
*mnt
;
835 if (!type
|| !memchr(type
, 0, PAGE_SIZE
))
838 /* we need capabilities... */
839 if (!capable(CAP_SYS_ADMIN
))
842 mnt
= do_kern_mount(type
, flags
, name
, data
);
846 return do_add_mount(mnt
, nd
, mnt_flags
, NULL
);
850 * add a mount into a namespace's mount tree
851 * - provide the option of adding the new mount to an expiration list
853 int do_add_mount(struct vfsmount
*newmnt
, struct nameidata
*nd
,
854 int mnt_flags
, struct list_head
*fslist
)
858 down_write(¤t
->namespace->sem
);
859 /* Something was mounted here while we slept */
860 while (d_mountpoint(nd
->dentry
) && follow_down(&nd
->mnt
, &nd
->dentry
))
863 if (!check_mnt(nd
->mnt
))
866 /* Refuse the same filesystem on the same mount point */
868 if (nd
->mnt
->mnt_sb
== newmnt
->mnt_sb
&&
869 nd
->mnt
->mnt_root
== nd
->dentry
)
873 if (S_ISLNK(newmnt
->mnt_root
->d_inode
->i_mode
))
876 newmnt
->mnt_flags
= mnt_flags
;
877 if ((err
= graft_tree(newmnt
, nd
)))
881 /* add to the specified expiration list */
882 spin_lock(&vfsmount_lock
);
883 list_add_tail(&newmnt
->mnt_expire
, fslist
);
884 spin_unlock(&vfsmount_lock
);
886 up_write(¤t
->namespace->sem
);
890 up_write(¤t
->namespace->sem
);
895 EXPORT_SYMBOL_GPL(do_add_mount
);
897 static void expire_mount(struct vfsmount
*mnt
, struct list_head
*mounts
,
898 struct list_head
*umounts
)
900 spin_lock(&vfsmount_lock
);
903 * Check if mount is still attached, if not, let whoever holds it deal
906 if (mnt
->mnt_parent
== mnt
) {
907 spin_unlock(&vfsmount_lock
);
912 * Check that it is still dead: the count should now be 2 - as
913 * contributed by the vfsmount parent and the mntget above
915 if (atomic_read(&mnt
->mnt_count
) == 2) {
916 /* delete from the namespace */
917 touch_namespace(mnt
->mnt_namespace
);
918 list_del_init(&mnt
->mnt_list
);
919 mnt
->mnt_namespace
= NULL
;
920 umount_tree(mnt
, umounts
);
921 spin_unlock(&vfsmount_lock
);
924 * Someone brought it back to life whilst we didn't have any
925 * locks held so return it to the expiration list
927 list_add_tail(&mnt
->mnt_expire
, mounts
);
928 spin_unlock(&vfsmount_lock
);
933 * process a list of expirable mountpoints with the intent of discarding any
934 * mountpoints that aren't in use and haven't been touched since last we came
937 void mark_mounts_for_expiry(struct list_head
*mounts
)
939 struct namespace *namespace;
940 struct vfsmount
*mnt
, *next
;
941 LIST_HEAD(graveyard
);
943 if (list_empty(mounts
))
946 spin_lock(&vfsmount_lock
);
948 /* extract from the expiration list every vfsmount that matches the
949 * following criteria:
950 * - only referenced by its parent vfsmount
951 * - still marked for expiry (marked on the last call here; marks are
952 * cleared by mntput())
954 list_for_each_entry_safe(mnt
, next
, mounts
, mnt_expire
) {
955 if (!xchg(&mnt
->mnt_expiry_mark
, 1) ||
956 atomic_read(&mnt
->mnt_count
) != 1)
960 list_move(&mnt
->mnt_expire
, &graveyard
);
964 * go through the vfsmounts we've just consigned to the graveyard to
965 * - check that they're still dead
966 * - delete the vfsmount from the appropriate namespace under lock
967 * - dispose of the corpse
969 while (!list_empty(&graveyard
)) {
971 mnt
= list_entry(graveyard
.next
, struct vfsmount
, mnt_expire
);
972 list_del_init(&mnt
->mnt_expire
);
974 /* don't do anything if the namespace is dead - all the
975 * vfsmounts from it are going away anyway */
976 namespace = mnt
->mnt_namespace
;
977 if (!namespace || !namespace->root
)
979 get_namespace(namespace);
981 spin_unlock(&vfsmount_lock
);
982 down_write(&namespace->sem
);
983 expire_mount(mnt
, mounts
, &umounts
);
984 up_write(&namespace->sem
);
985 release_mounts(&umounts
);
987 put_namespace(namespace);
988 spin_lock(&vfsmount_lock
);
991 spin_unlock(&vfsmount_lock
);
994 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry
);
997 * Some copy_from_user() implementations do not return the exact number of
998 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
999 * Note that this function differs from copy_from_user() in that it will oops
1000 * on bad values of `to', rather than returning a short copy.
1002 static long exact_copy_from_user(void *to
, const void __user
* from
,
1006 const char __user
*f
= from
;
1009 if (!access_ok(VERIFY_READ
, from
, n
))
1013 if (__get_user(c
, f
)) {
1024 int copy_mount_options(const void __user
* data
, unsigned long *where
)
1034 if (!(page
= __get_free_page(GFP_KERNEL
)))
1037 /* We only care that *some* data at the address the user
1038 * gave us is valid. Just in case, we'll zero
1039 * the remainder of the page.
1041 /* copy_from_user cannot cross TASK_SIZE ! */
1042 size
= TASK_SIZE
- (unsigned long)data
;
1043 if (size
> PAGE_SIZE
)
1046 i
= size
- exact_copy_from_user((void *)page
, data
, size
);
1052 memset((char *)page
+ i
, 0, PAGE_SIZE
- i
);
1058 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1059 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1061 * data is a (void *) that can point to any structure up to
1062 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1063 * information (or be NULL).
1065 * Pre-0.97 versions of mount() didn't have a flags word.
1066 * When the flags word was introduced its top half was required
1067 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1068 * Therefore, if this magic number is present, it carries no information
1069 * and must be discarded.
1071 long do_mount(char *dev_name
, char *dir_name
, char *type_page
,
1072 unsigned long flags
, void *data_page
)
1074 struct nameidata nd
;
1079 if ((flags
& MS_MGC_MSK
) == MS_MGC_VAL
)
1080 flags
&= ~MS_MGC_MSK
;
1082 /* Basic sanity checks */
1084 if (!dir_name
|| !*dir_name
|| !memchr(dir_name
, 0, PAGE_SIZE
))
1086 if (dev_name
&& !memchr(dev_name
, 0, PAGE_SIZE
))
1090 ((char *)data_page
)[PAGE_SIZE
- 1] = 0;
1092 /* Separate the per-mountpoint flags */
1093 if (flags
& MS_NOSUID
)
1094 mnt_flags
|= MNT_NOSUID
;
1095 if (flags
& MS_NODEV
)
1096 mnt_flags
|= MNT_NODEV
;
1097 if (flags
& MS_NOEXEC
)
1098 mnt_flags
|= MNT_NOEXEC
;
1099 flags
&= ~(MS_NOSUID
| MS_NOEXEC
| MS_NODEV
| MS_ACTIVE
);
1101 /* ... and get the mountpoint */
1102 retval
= path_lookup(dir_name
, LOOKUP_FOLLOW
, &nd
);
1106 retval
= security_sb_mount(dev_name
, &nd
, type_page
, flags
, data_page
);
1110 if (flags
& MS_REMOUNT
)
1111 retval
= do_remount(&nd
, flags
& ~MS_REMOUNT
, mnt_flags
,
1113 else if (flags
& MS_BIND
)
1114 retval
= do_loopback(&nd
, dev_name
, flags
& MS_REC
);
1115 else if (flags
& MS_MOVE
)
1116 retval
= do_move_mount(&nd
, dev_name
);
1118 retval
= do_new_mount(&nd
, type_page
, flags
, mnt_flags
,
1119 dev_name
, data_page
);
1125 int copy_namespace(int flags
, struct task_struct
*tsk
)
1127 struct namespace *namespace = tsk
->namespace;
1128 struct namespace *new_ns
;
1129 struct vfsmount
*rootmnt
= NULL
, *pwdmnt
= NULL
, *altrootmnt
= NULL
;
1130 struct fs_struct
*fs
= tsk
->fs
;
1131 struct vfsmount
*p
, *q
;
1136 get_namespace(namespace);
1138 if (!(flags
& CLONE_NEWNS
))
1141 if (!capable(CAP_SYS_ADMIN
)) {
1142 put_namespace(namespace);
1146 new_ns
= kmalloc(sizeof(struct namespace), GFP_KERNEL
);
1150 atomic_set(&new_ns
->count
, 1);
1151 init_rwsem(&new_ns
->sem
);
1152 INIT_LIST_HEAD(&new_ns
->list
);
1153 init_waitqueue_head(&new_ns
->poll
);
1156 down_write(&tsk
->namespace->sem
);
1157 /* First pass: copy the tree topology */
1158 new_ns
->root
= copy_tree(namespace->root
, namespace->root
->mnt_root
);
1159 if (!new_ns
->root
) {
1160 up_write(&tsk
->namespace->sem
);
1164 spin_lock(&vfsmount_lock
);
1165 list_add_tail(&new_ns
->list
, &new_ns
->root
->mnt_list
);
1166 spin_unlock(&vfsmount_lock
);
1169 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1170 * as belonging to new namespace. We have already acquired a private
1171 * fs_struct, so tsk->fs->lock is not needed.
1173 p
= namespace->root
;
1176 q
->mnt_namespace
= new_ns
;
1178 if (p
== fs
->rootmnt
) {
1180 fs
->rootmnt
= mntget(q
);
1182 if (p
== fs
->pwdmnt
) {
1184 fs
->pwdmnt
= mntget(q
);
1186 if (p
== fs
->altrootmnt
) {
1188 fs
->altrootmnt
= mntget(q
);
1191 p
= next_mnt(p
, namespace->root
);
1192 q
= next_mnt(q
, new_ns
->root
);
1194 up_write(&tsk
->namespace->sem
);
1196 tsk
->namespace = new_ns
;
1205 put_namespace(namespace);
1209 put_namespace(namespace);
1213 asmlinkage
long sys_mount(char __user
* dev_name
, char __user
* dir_name
,
1214 char __user
* type
, unsigned long flags
,
1218 unsigned long data_page
;
1219 unsigned long type_page
;
1220 unsigned long dev_page
;
1223 retval
= copy_mount_options(type
, &type_page
);
1227 dir_page
= getname(dir_name
);
1228 retval
= PTR_ERR(dir_page
);
1229 if (IS_ERR(dir_page
))
1232 retval
= copy_mount_options(dev_name
, &dev_page
);
1236 retval
= copy_mount_options(data
, &data_page
);
1241 retval
= do_mount((char *)dev_page
, dir_page
, (char *)type_page
,
1242 flags
, (void *)data_page
);
1244 free_page(data_page
);
1247 free_page(dev_page
);
1251 free_page(type_page
);
1256 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1257 * It can block. Requires the big lock held.
1259 void set_fs_root(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1260 struct dentry
*dentry
)
1262 struct dentry
*old_root
;
1263 struct vfsmount
*old_rootmnt
;
1264 write_lock(&fs
->lock
);
1265 old_root
= fs
->root
;
1266 old_rootmnt
= fs
->rootmnt
;
1267 fs
->rootmnt
= mntget(mnt
);
1268 fs
->root
= dget(dentry
);
1269 write_unlock(&fs
->lock
);
1272 mntput(old_rootmnt
);
1277 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1278 * It can block. Requires the big lock held.
1280 void set_fs_pwd(struct fs_struct
*fs
, struct vfsmount
*mnt
,
1281 struct dentry
*dentry
)
1283 struct dentry
*old_pwd
;
1284 struct vfsmount
*old_pwdmnt
;
1286 write_lock(&fs
->lock
);
1288 old_pwdmnt
= fs
->pwdmnt
;
1289 fs
->pwdmnt
= mntget(mnt
);
1290 fs
->pwd
= dget(dentry
);
1291 write_unlock(&fs
->lock
);
1299 static void chroot_fs_refs(struct nameidata
*old_nd
, struct nameidata
*new_nd
)
1301 struct task_struct
*g
, *p
;
1302 struct fs_struct
*fs
;
1304 read_lock(&tasklist_lock
);
1305 do_each_thread(g
, p
) {
1309 atomic_inc(&fs
->count
);
1311 if (fs
->root
== old_nd
->dentry
1312 && fs
->rootmnt
== old_nd
->mnt
)
1313 set_fs_root(fs
, new_nd
->mnt
, new_nd
->dentry
);
1314 if (fs
->pwd
== old_nd
->dentry
1315 && fs
->pwdmnt
== old_nd
->mnt
)
1316 set_fs_pwd(fs
, new_nd
->mnt
, new_nd
->dentry
);
1320 } while_each_thread(g
, p
);
1321 read_unlock(&tasklist_lock
);
1325 * pivot_root Semantics:
1326 * Moves the root file system of the current process to the directory put_old,
1327 * makes new_root as the new root file system of the current process, and sets
1328 * root/cwd of all processes which had them on the current root to new_root.
1331 * The new_root and put_old must be directories, and must not be on the
1332 * same file system as the current process root. The put_old must be
1333 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1334 * pointed to by put_old must yield the same directory as new_root. No other
1335 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1338 * - we don't move root/cwd if they are not at the root (reason: if something
1339 * cared enough to change them, it's probably wrong to force them elsewhere)
1340 * - it's okay to pick a root that isn't the root of a file system, e.g.
1341 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1342 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1345 asmlinkage
long sys_pivot_root(const char __user
* new_root
,
1346 const char __user
* put_old
)
1348 struct vfsmount
*tmp
;
1349 struct nameidata new_nd
, old_nd
, parent_nd
, root_parent
, user_nd
;
1352 if (!capable(CAP_SYS_ADMIN
))
1357 error
= __user_walk(new_root
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
,
1362 if (!check_mnt(new_nd
.mnt
))
1365 error
= __user_walk(put_old
, LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
, &old_nd
);
1369 error
= security_sb_pivotroot(&old_nd
, &new_nd
);
1371 path_release(&old_nd
);
1375 read_lock(¤t
->fs
->lock
);
1376 user_nd
.mnt
= mntget(current
->fs
->rootmnt
);
1377 user_nd
.dentry
= dget(current
->fs
->root
);
1378 read_unlock(¤t
->fs
->lock
);
1379 down_write(¤t
->namespace->sem
);
1380 down(&old_nd
.dentry
->d_inode
->i_sem
);
1382 if (!check_mnt(user_nd
.mnt
))
1385 if (IS_DEADDIR(new_nd
.dentry
->d_inode
))
1387 if (d_unhashed(new_nd
.dentry
) && !IS_ROOT(new_nd
.dentry
))
1389 if (d_unhashed(old_nd
.dentry
) && !IS_ROOT(old_nd
.dentry
))
1392 if (new_nd
.mnt
== user_nd
.mnt
|| old_nd
.mnt
== user_nd
.mnt
)
1393 goto out2
; /* loop, on the same file system */
1395 if (user_nd
.mnt
->mnt_root
!= user_nd
.dentry
)
1396 goto out2
; /* not a mountpoint */
1397 if (user_nd
.mnt
->mnt_parent
== user_nd
.mnt
)
1398 goto out2
; /* not attached */
1399 if (new_nd
.mnt
->mnt_root
!= new_nd
.dentry
)
1400 goto out2
; /* not a mountpoint */
1401 if (new_nd
.mnt
->mnt_parent
== new_nd
.mnt
)
1402 goto out2
; /* not attached */
1403 tmp
= old_nd
.mnt
; /* make sure we can reach put_old from new_root */
1404 spin_lock(&vfsmount_lock
);
1405 if (tmp
!= new_nd
.mnt
) {
1407 if (tmp
->mnt_parent
== tmp
)
1408 goto out3
; /* already mounted on put_old */
1409 if (tmp
->mnt_parent
== new_nd
.mnt
)
1411 tmp
= tmp
->mnt_parent
;
1413 if (!is_subdir(tmp
->mnt_mountpoint
, new_nd
.dentry
))
1415 } else if (!is_subdir(old_nd
.dentry
, new_nd
.dentry
))
1417 detach_mnt(new_nd
.mnt
, &parent_nd
);
1418 detach_mnt(user_nd
.mnt
, &root_parent
);
1419 attach_mnt(user_nd
.mnt
, &old_nd
); /* mount old root on put_old */
1420 attach_mnt(new_nd
.mnt
, &root_parent
); /* mount new_root on / */
1421 touch_namespace(current
->namespace);
1422 spin_unlock(&vfsmount_lock
);
1423 chroot_fs_refs(&user_nd
, &new_nd
);
1424 security_sb_post_pivotroot(&user_nd
, &new_nd
);
1426 path_release(&root_parent
);
1427 path_release(&parent_nd
);
1429 up(&old_nd
.dentry
->d_inode
->i_sem
);
1430 up_write(¤t
->namespace->sem
);
1431 path_release(&user_nd
);
1432 path_release(&old_nd
);
1434 path_release(&new_nd
);
1439 spin_unlock(&vfsmount_lock
);
1443 static void __init
init_mount_tree(void)
1445 struct vfsmount
*mnt
;
1446 struct namespace *namespace;
1447 struct task_struct
*g
, *p
;
1449 mnt
= do_kern_mount("rootfs", 0, "rootfs", NULL
);
1451 panic("Can't create rootfs");
1452 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL
);
1454 panic("Can't allocate initial namespace");
1455 atomic_set(&namespace->count
, 1);
1456 INIT_LIST_HEAD(&namespace->list
);
1457 init_rwsem(&namespace->sem
);
1458 init_waitqueue_head(&namespace->poll
);
1459 namespace->event
= 0;
1460 list_add(&mnt
->mnt_list
, &namespace->list
);
1461 namespace->root
= mnt
;
1462 mnt
->mnt_namespace
= namespace;
1464 init_task
.namespace = namespace;
1465 read_lock(&tasklist_lock
);
1466 do_each_thread(g
, p
) {
1467 get_namespace(namespace);
1468 p
->namespace = namespace;
1469 } while_each_thread(g
, p
);
1470 read_unlock(&tasklist_lock
);
1472 set_fs_pwd(current
->fs
, namespace->root
, namespace->root
->mnt_root
);
1473 set_fs_root(current
->fs
, namespace->root
, namespace->root
->mnt_root
);
1476 void __init
mnt_init(unsigned long mempages
)
1478 struct list_head
*d
;
1479 unsigned int nr_hash
;
1482 mnt_cache
= kmem_cache_create("mnt_cache", sizeof(struct vfsmount
),
1483 0, SLAB_HWCACHE_ALIGN
| SLAB_PANIC
, NULL
, NULL
);
1485 mount_hashtable
= (struct list_head
*)__get_free_page(GFP_ATOMIC
);
1487 if (!mount_hashtable
)
1488 panic("Failed to allocate mount hash table\n");
1491 * Find the power-of-two list-heads that can fit into the allocation..
1492 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1495 nr_hash
= PAGE_SIZE
/ sizeof(struct list_head
);
1499 } while ((nr_hash
>> hash_bits
) != 0);
1503 * Re-calculate the actual number of entries and the mask
1504 * from the number of bits we can fit.
1506 nr_hash
= 1UL << hash_bits
;
1507 hash_mask
= nr_hash
- 1;
1509 printk("Mount-cache hash table entries: %d\n", nr_hash
);
1511 /* And initialize the newly allocated array */
1512 d
= mount_hashtable
;
1524 void __put_namespace(struct namespace *namespace)
1526 struct vfsmount
*root
= namespace->root
;
1527 LIST_HEAD(umount_list
);
1528 namespace->root
= NULL
;
1529 spin_unlock(&vfsmount_lock
);
1530 down_write(&namespace->sem
);
1531 spin_lock(&vfsmount_lock
);
1532 umount_tree(root
, &umount_list
);
1533 spin_unlock(&vfsmount_lock
);
1534 up_write(&namespace->sem
);
1535 release_mounts(&umount_list
);