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>
29 extern int __init init_rootfs(void);
32 extern int __init sysfs_init(void);
34 static inline int sysfs_init(void)
40 /* spinlock for vfsmount related operations, inplace of dcache_lock */
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
45 static struct list_head *mount_hashtable;
46 static int hash_mask __read_mostly, hash_bits __read_mostly;
47 static kmem_cache_t *mnt_cache;
48 static struct rw_semaphore namespace_sem;
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_bits);
55 return tmp & hash_mask;
58 struct vfsmount *alloc_vfsmnt(const char *name)
60 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
62 memset(mnt, 0, sizeof(struct vfsmount));
63 atomic_set(&mnt->mnt_count, 1);
64 INIT_LIST_HEAD(&mnt->mnt_hash);
65 INIT_LIST_HEAD(&mnt->mnt_child);
66 INIT_LIST_HEAD(&mnt->mnt_mounts);
67 INIT_LIST_HEAD(&mnt->mnt_list);
68 INIT_LIST_HEAD(&mnt->mnt_expire);
69 INIT_LIST_HEAD(&mnt->mnt_share);
70 INIT_LIST_HEAD(&mnt->mnt_slave_list);
71 INIT_LIST_HEAD(&mnt->mnt_slave);
73 int size = strlen(name) + 1;
74 char *newname = kmalloc(size, GFP_KERNEL);
76 memcpy(newname, name, size);
77 mnt->mnt_devname = newname;
84 void free_vfsmnt(struct vfsmount *mnt)
86 kfree(mnt->mnt_devname);
87 kmem_cache_free(mnt_cache, mnt);
91 * find the first or last mount at @dentry on vfsmount @mnt depending on
92 * @dir. If @dir is set return the first mount else return the last mount.
94 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
97 struct list_head *head = mount_hashtable + hash(mnt, dentry);
98 struct list_head *tmp = head;
99 struct vfsmount *p, *found = NULL;
102 tmp = dir ? tmp->next : tmp->prev;
106 p = list_entry(tmp, struct vfsmount, mnt_hash);
107 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
116 * lookup_mnt increments the ref count before returning
117 * the vfsmount struct.
119 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
121 struct vfsmount *child_mnt;
122 spin_lock(&vfsmount_lock);
123 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
125 spin_unlock(&vfsmount_lock);
129 static inline int check_mnt(struct vfsmount *mnt)
131 return mnt->mnt_namespace == current->namespace;
134 static void touch_namespace(struct namespace *ns)
138 wake_up_interruptible(&ns->poll);
142 static void __touch_namespace(struct namespace *ns)
144 if (ns && ns->event != event) {
146 wake_up_interruptible(&ns->poll);
150 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
152 old_nd->dentry = mnt->mnt_mountpoint;
153 old_nd->mnt = mnt->mnt_parent;
154 mnt->mnt_parent = mnt;
155 mnt->mnt_mountpoint = mnt->mnt_root;
156 list_del_init(&mnt->mnt_child);
157 list_del_init(&mnt->mnt_hash);
158 old_nd->dentry->d_mounted--;
161 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
162 struct vfsmount *child_mnt)
164 child_mnt->mnt_parent = mntget(mnt);
165 child_mnt->mnt_mountpoint = dget(dentry);
169 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
171 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
172 list_add_tail(&mnt->mnt_hash, mount_hashtable +
173 hash(nd->mnt, nd->dentry));
174 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
178 * the caller must hold vfsmount_lock
180 static void commit_tree(struct vfsmount *mnt)
182 struct vfsmount *parent = mnt->mnt_parent;
185 struct namespace *n = parent->mnt_namespace;
187 BUG_ON(parent == mnt);
189 list_add_tail(&head, &mnt->mnt_list);
190 list_for_each_entry(m, &head, mnt_list)
191 m->mnt_namespace = n;
192 list_splice(&head, n->list.prev);
194 list_add_tail(&mnt->mnt_hash, mount_hashtable +
195 hash(parent, mnt->mnt_mountpoint));
196 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
200 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
202 struct list_head *next = p->mnt_mounts.next;
203 if (next == &p->mnt_mounts) {
207 next = p->mnt_child.next;
208 if (next != &p->mnt_parent->mnt_mounts)
213 return list_entry(next, struct vfsmount, mnt_child);
216 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
219 struct super_block *sb = old->mnt_sb;
220 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
223 mnt->mnt_flags = old->mnt_flags;
224 atomic_inc(&sb->s_active);
226 mnt->mnt_root = dget(root);
227 mnt->mnt_mountpoint = mnt->mnt_root;
228 mnt->mnt_parent = mnt;
230 if (flag & CL_SLAVE) {
231 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
232 mnt->mnt_master = old;
233 CLEAR_MNT_SHARED(mnt);
235 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
236 list_add(&mnt->mnt_share, &old->mnt_share);
237 if (IS_MNT_SLAVE(old))
238 list_add(&mnt->mnt_slave, &old->mnt_slave);
239 mnt->mnt_master = old->mnt_master;
241 if (flag & CL_MAKE_SHARED)
244 /* stick the duplicate mount on the same expiry list
245 * as the original if that was on one */
246 if (flag & CL_EXPIRE) {
247 spin_lock(&vfsmount_lock);
248 if (!list_empty(&old->mnt_expire))
249 list_add(&mnt->mnt_expire, &old->mnt_expire);
250 spin_unlock(&vfsmount_lock);
256 static inline void __mntput(struct vfsmount *mnt)
258 struct super_block *sb = mnt->mnt_sb;
261 deactivate_super(sb);
264 void mntput_no_expire(struct vfsmount *mnt)
267 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
268 if (likely(!mnt->mnt_pinned)) {
269 spin_unlock(&vfsmount_lock);
273 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
275 spin_unlock(&vfsmount_lock);
276 acct_auto_close_mnt(mnt);
277 security_sb_umount_close(mnt);
282 EXPORT_SYMBOL(mntput_no_expire);
284 void mnt_pin(struct vfsmount *mnt)
286 spin_lock(&vfsmount_lock);
288 spin_unlock(&vfsmount_lock);
291 EXPORT_SYMBOL(mnt_pin);
293 void mnt_unpin(struct vfsmount *mnt)
295 spin_lock(&vfsmount_lock);
296 if (mnt->mnt_pinned) {
297 atomic_inc(&mnt->mnt_count);
300 spin_unlock(&vfsmount_lock);
303 EXPORT_SYMBOL(mnt_unpin);
306 static void *m_start(struct seq_file *m, loff_t *pos)
308 struct namespace *n = m->private;
312 down_read(&namespace_sem);
313 list_for_each(p, &n->list)
315 return list_entry(p, struct vfsmount, mnt_list);
319 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
321 struct namespace *n = m->private;
322 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
324 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
327 static void m_stop(struct seq_file *m, void *v)
329 up_read(&namespace_sem);
332 static inline void mangle(struct seq_file *m, const char *s)
334 seq_escape(m, s, " \t\n\\");
337 static int show_vfsmnt(struct seq_file *m, void *v)
339 struct vfsmount *mnt = v;
341 static struct proc_fs_info {
345 { MS_SYNCHRONOUS, ",sync" },
346 { MS_DIRSYNC, ",dirsync" },
347 { MS_MANDLOCK, ",mand" },
348 { MS_NOATIME, ",noatime" },
349 { MS_NODIRATIME, ",nodiratime" },
352 static struct proc_fs_info mnt_info[] = {
353 { MNT_NOSUID, ",nosuid" },
354 { MNT_NODEV, ",nodev" },
355 { MNT_NOEXEC, ",noexec" },
358 struct proc_fs_info *fs_infop;
360 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
362 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
364 mangle(m, mnt->mnt_sb->s_type->name);
365 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
366 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
367 if (mnt->mnt_sb->s_flags & fs_infop->flag)
368 seq_puts(m, fs_infop->str);
370 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
371 if (mnt->mnt_flags & fs_infop->flag)
372 seq_puts(m, fs_infop->str);
374 if (mnt->mnt_sb->s_op->show_options)
375 err = mnt->mnt_sb->s_op->show_options(m, mnt);
376 seq_puts(m, " 0 0\n");
380 struct seq_operations mounts_op = {
388 * may_umount_tree - check if a mount tree is busy
389 * @mnt: root of mount tree
391 * This is called to check if a tree of mounts has any
392 * open files, pwds, chroots or sub mounts that are
395 int may_umount_tree(struct vfsmount *mnt)
398 int minimum_refs = 0;
401 spin_lock(&vfsmount_lock);
402 for (p = mnt; p; p = next_mnt(p, mnt)) {
403 actual_refs += atomic_read(&p->mnt_count);
406 spin_unlock(&vfsmount_lock);
408 if (actual_refs > minimum_refs)
414 EXPORT_SYMBOL(may_umount_tree);
417 * may_umount - check if a mount point is busy
418 * @mnt: root of mount
420 * This is called to check if a mount point has any
421 * open files, pwds, chroots or sub mounts. If the
422 * mount has sub mounts this will return busy
423 * regardless of whether the sub mounts are busy.
425 * Doesn't take quota and stuff into account. IOW, in some cases it will
426 * give false negatives. The main reason why it's here is that we need
427 * a non-destructive way to look for easily umountable filesystems.
429 int may_umount(struct vfsmount *mnt)
432 spin_lock(&vfsmount_lock);
433 if (propagate_mount_busy(mnt, 2))
435 spin_unlock(&vfsmount_lock);
439 EXPORT_SYMBOL(may_umount);
441 void release_mounts(struct list_head *head)
443 struct vfsmount *mnt;
444 while(!list_empty(head)) {
445 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
446 list_del_init(&mnt->mnt_hash);
447 if (mnt->mnt_parent != mnt) {
448 struct dentry *dentry;
450 spin_lock(&vfsmount_lock);
451 dentry = mnt->mnt_mountpoint;
453 mnt->mnt_mountpoint = mnt->mnt_root;
454 mnt->mnt_parent = mnt;
455 spin_unlock(&vfsmount_lock);
463 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
467 for (p = mnt; p; p = next_mnt(p, mnt)) {
468 list_del(&p->mnt_hash);
469 list_add(&p->mnt_hash, kill);
473 propagate_umount(kill);
475 list_for_each_entry(p, kill, mnt_hash) {
476 list_del_init(&p->mnt_expire);
477 list_del_init(&p->mnt_list);
478 __touch_namespace(p->mnt_namespace);
479 p->mnt_namespace = NULL;
480 list_del_init(&p->mnt_child);
481 if (p->mnt_parent != p)
482 mnt->mnt_mountpoint->d_mounted--;
483 change_mnt_propagation(p, MS_PRIVATE);
487 static int do_umount(struct vfsmount *mnt, int flags)
489 struct super_block *sb = mnt->mnt_sb;
491 LIST_HEAD(umount_list);
493 retval = security_sb_umount(mnt, flags);
498 * Allow userspace to request a mountpoint be expired rather than
499 * unmounting unconditionally. Unmount only happens if:
500 * (1) the mark is already set (the mark is cleared by mntput())
501 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
503 if (flags & MNT_EXPIRE) {
504 if (mnt == current->fs->rootmnt ||
505 flags & (MNT_FORCE | MNT_DETACH))
508 if (atomic_read(&mnt->mnt_count) != 2)
511 if (!xchg(&mnt->mnt_expiry_mark, 1))
516 * If we may have to abort operations to get out of this
517 * mount, and they will themselves hold resources we must
518 * allow the fs to do things. In the Unix tradition of
519 * 'Gee thats tricky lets do it in userspace' the umount_begin
520 * might fail to complete on the first run through as other tasks
521 * must return, and the like. Thats for the mount program to worry
522 * about for the moment.
526 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
527 sb->s_op->umount_begin(sb);
531 * No sense to grab the lock for this test, but test itself looks
532 * somewhat bogus. Suggestions for better replacement?
533 * Ho-hum... In principle, we might treat that as umount + switch
534 * to rootfs. GC would eventually take care of the old vfsmount.
535 * Actually it makes sense, especially if rootfs would contain a
536 * /reboot - static binary that would close all descriptors and
537 * call reboot(9). Then init(8) could umount root and exec /reboot.
539 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
541 * Special case for "unmounting" root ...
542 * we just try to remount it readonly.
544 down_write(&sb->s_umount);
545 if (!(sb->s_flags & MS_RDONLY)) {
548 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
551 up_write(&sb->s_umount);
555 down_write(&namespace_sem);
556 spin_lock(&vfsmount_lock);
560 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
561 if (!list_empty(&mnt->mnt_list))
562 umount_tree(mnt, 1, &umount_list);
565 spin_unlock(&vfsmount_lock);
567 security_sb_umount_busy(mnt);
568 up_write(&namespace_sem);
569 release_mounts(&umount_list);
574 * Now umount can handle mount points as well as block devices.
575 * This is important for filesystems which use unnamed block devices.
577 * We now support a flag for forced unmount like the other 'big iron'
578 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
581 asmlinkage long sys_umount(char __user * name, int flags)
586 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
590 if (nd.dentry != nd.mnt->mnt_root)
592 if (!check_mnt(nd.mnt))
596 if (!capable(CAP_SYS_ADMIN))
599 retval = do_umount(nd.mnt, flags);
601 path_release_on_umount(&nd);
606 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
609 * The 2.0 compatible umount. No flags.
611 asmlinkage long sys_oldumount(char __user * name)
613 return sys_umount(name, 0);
618 static int mount_is_safe(struct nameidata *nd)
620 if (capable(CAP_SYS_ADMIN))
624 if (S_ISLNK(nd->dentry->d_inode->i_mode))
626 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
627 if (current->uid != nd->dentry->d_inode->i_uid)
630 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
636 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
641 if (d == NULL || d == d->d_parent)
647 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
650 struct vfsmount *res, *p, *q, *r, *s;
653 res = q = clone_mnt(mnt, dentry, flag);
656 q->mnt_mountpoint = mnt->mnt_mountpoint;
659 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
660 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
663 for (s = r; s; s = next_mnt(s, r)) {
664 while (p != s->mnt_parent) {
670 nd.dentry = p->mnt_mountpoint;
671 q = clone_mnt(p, p->mnt_root, flag);
674 spin_lock(&vfsmount_lock);
675 list_add_tail(&q->mnt_list, &res->mnt_list);
677 spin_unlock(&vfsmount_lock);
683 LIST_HEAD(umount_list);
684 spin_lock(&vfsmount_lock);
685 umount_tree(res, 0, &umount_list);
686 spin_unlock(&vfsmount_lock);
687 release_mounts(&umount_list);
693 * @source_mnt : mount tree to be attached
694 * @nd : place the mount tree @source_mnt is attached
695 * @parent_nd : if non-null, detach the source_mnt from its parent and
696 * store the parent mount and mountpoint dentry.
697 * (done when source_mnt is moved)
699 * NOTE: in the table below explains the semantics when a source mount
700 * of a given type is attached to a destination mount of a given type.
701 * -------------------------------------------------------------
702 * | BIND MOUNT OPERATION |
703 * |*************************************************************
704 * | source-->| shared | private | slave |
708 * |*************************************************************
709 * | shared | shared (++) | shared (+) | shared(+++)|
711 * |non-shared| shared (+) | private | slave (*) |
712 * **************************************************************
713 * A bind operation clones the source mount and mounts the clone on the
716 * (++) the cloned mount is propagated to all the mounts in the propagation
717 * tree of the destination mount and the cloned mount is added to
718 * the peer group of the source mount.
719 * (+) the cloned mount is created under the destination mount and is marked
720 * as shared. The cloned mount is added to the peer group of the source
722 * (+++) the mount is propagated to all the mounts in the propagation tree
723 * of the destination mount and the cloned mount is made slave
724 * of the same master as that of the source mount. The cloned mount
725 * is marked as 'shared and slave'.
726 * (*) the cloned mount is made a slave of the same master as that of the
729 * --------------------------------------------------------------
730 * | MOVE MOUNT OPERATION |
731 * |*************************************************************
732 * | source-->| shared | private | slave |
736 * |*************************************************************
737 * | shared | shared (+) | shared (+) | shared(+++) |
739 * |non-shared| shared (+*) | private | slave (*) |
740 * **************************************************************
742 * (+) the mount is moved to the destination. And is then propagated to
743 * all the mounts in the propagation tree of the destination mount.
744 * (+*) the mount is moved to the destination.
745 * (+++) the mount is moved to the destination and is then propagated to
746 * all the mounts belonging to the destination mount's propagation tree.
747 * the mount is marked as 'shared and slave'.
748 * (*) the mount continues to be a slave at the new location.
750 * if the source mount is a tree, the operations explained above is
751 * applied to each mount in the tree.
752 * Must be called without spinlocks held, since this function can sleep
755 static int attach_recursive_mnt(struct vfsmount *source_mnt,
756 struct nameidata *nd, struct nameidata *parent_nd)
758 LIST_HEAD(tree_list);
759 struct vfsmount *dest_mnt = nd->mnt;
760 struct dentry *dest_dentry = nd->dentry;
761 struct vfsmount *child, *p;
763 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
766 if (IS_MNT_SHARED(dest_mnt)) {
767 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
771 spin_lock(&vfsmount_lock);
773 detach_mnt(source_mnt, parent_nd);
774 attach_mnt(source_mnt, nd);
775 touch_namespace(current->namespace);
777 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
778 commit_tree(source_mnt);
781 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
782 list_del_init(&child->mnt_hash);
785 spin_unlock(&vfsmount_lock);
789 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
792 if (mnt->mnt_sb->s_flags & MS_NOUSER)
795 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
796 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
800 down(&nd->dentry->d_inode->i_sem);
801 if (IS_DEADDIR(nd->dentry->d_inode))
804 err = security_sb_check_sb(mnt, nd);
809 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
810 err = attach_recursive_mnt(mnt, nd, NULL);
812 up(&nd->dentry->d_inode->i_sem);
814 security_sb_post_addmount(mnt, nd);
819 * recursively change the type of the mountpoint.
821 static int do_change_type(struct nameidata *nd, int flag)
823 struct vfsmount *m, *mnt = nd->mnt;
824 int recurse = flag & MS_REC;
825 int type = flag & ~MS_REC;
827 if (nd->dentry != nd->mnt->mnt_root)
830 down_write(&namespace_sem);
831 spin_lock(&vfsmount_lock);
832 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
833 change_mnt_propagation(m, type);
834 spin_unlock(&vfsmount_lock);
835 up_write(&namespace_sem);
842 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
844 struct nameidata old_nd;
845 struct vfsmount *mnt = NULL;
846 int err = mount_is_safe(nd);
849 if (!old_name || !*old_name)
851 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
855 down_write(&namespace_sem);
857 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
862 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
864 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
869 err = graft_tree(mnt, nd);
871 LIST_HEAD(umount_list);
872 spin_lock(&vfsmount_lock);
873 umount_tree(mnt, 0, &umount_list);
874 spin_unlock(&vfsmount_lock);
875 release_mounts(&umount_list);
879 up_write(&namespace_sem);
880 path_release(&old_nd);
885 * change filesystem flags. dir should be a physical root of filesystem.
886 * If you've mounted a non-root directory somewhere and want to do remount
887 * on it - tough luck.
889 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
893 struct super_block *sb = nd->mnt->mnt_sb;
895 if (!capable(CAP_SYS_ADMIN))
898 if (!check_mnt(nd->mnt))
901 if (nd->dentry != nd->mnt->mnt_root)
904 down_write(&sb->s_umount);
905 err = do_remount_sb(sb, flags, data, 0);
907 nd->mnt->mnt_flags = mnt_flags;
908 up_write(&sb->s_umount);
910 security_sb_post_remount(nd->mnt, flags, data);
914 static int do_move_mount(struct nameidata *nd, char *old_name)
916 struct nameidata old_nd, parent_nd;
919 if (!capable(CAP_SYS_ADMIN))
921 if (!old_name || !*old_name)
923 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
927 down_write(&namespace_sem);
928 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
931 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
935 down(&nd->dentry->d_inode->i_sem);
936 if (IS_DEADDIR(nd->dentry->d_inode))
939 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
943 if (old_nd.dentry != old_nd.mnt->mnt_root)
946 if (old_nd.mnt == old_nd.mnt->mnt_parent)
949 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
950 S_ISDIR(old_nd.dentry->d_inode->i_mode))
953 * Don't move a mount residing in a shared parent.
955 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
958 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
962 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
965 spin_lock(&vfsmount_lock);
966 /* if the mount is moved, it should no longer be expire
968 list_del_init(&old_nd.mnt->mnt_expire);
969 spin_unlock(&vfsmount_lock);
971 up(&nd->dentry->d_inode->i_sem);
973 up_write(&namespace_sem);
975 path_release(&parent_nd);
976 path_release(&old_nd);
981 * create a new mount for userspace and request it to be added into the
984 static int do_new_mount(struct nameidata *nd, char *type, int flags,
985 int mnt_flags, char *name, void *data)
987 struct vfsmount *mnt;
989 if (!type || !memchr(type, 0, PAGE_SIZE))
992 /* we need capabilities... */
993 if (!capable(CAP_SYS_ADMIN))
996 mnt = do_kern_mount(type, flags, name, data);
1000 return do_add_mount(mnt, nd, mnt_flags, NULL);
1004 * add a mount into a namespace's mount tree
1005 * - provide the option of adding the new mount to an expiration list
1007 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1008 int mnt_flags, struct list_head *fslist)
1012 down_write(&namespace_sem);
1013 /* Something was mounted here while we slept */
1014 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1017 if (!check_mnt(nd->mnt))
1020 /* Refuse the same filesystem on the same mount point */
1022 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1023 nd->mnt->mnt_root == nd->dentry)
1027 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1030 newmnt->mnt_flags = mnt_flags;
1031 if ((err = graft_tree(newmnt, nd)))
1035 /* add to the specified expiration list */
1036 spin_lock(&vfsmount_lock);
1037 list_add_tail(&newmnt->mnt_expire, fslist);
1038 spin_unlock(&vfsmount_lock);
1040 up_write(&namespace_sem);
1044 up_write(&namespace_sem);
1049 EXPORT_SYMBOL_GPL(do_add_mount);
1051 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1052 struct list_head *umounts)
1054 spin_lock(&vfsmount_lock);
1057 * Check if mount is still attached, if not, let whoever holds it deal
1060 if (mnt->mnt_parent == mnt) {
1061 spin_unlock(&vfsmount_lock);
1066 * Check that it is still dead: the count should now be 2 - as
1067 * contributed by the vfsmount parent and the mntget above
1069 if (!propagate_mount_busy(mnt, 2)) {
1070 /* delete from the namespace */
1071 touch_namespace(mnt->mnt_namespace);
1072 list_del_init(&mnt->mnt_list);
1073 mnt->mnt_namespace = NULL;
1074 umount_tree(mnt, 1, umounts);
1075 spin_unlock(&vfsmount_lock);
1078 * Someone brought it back to life whilst we didn't have any
1079 * locks held so return it to the expiration list
1081 list_add_tail(&mnt->mnt_expire, mounts);
1082 spin_unlock(&vfsmount_lock);
1087 * process a list of expirable mountpoints with the intent of discarding any
1088 * mountpoints that aren't in use and haven't been touched since last we came
1091 void mark_mounts_for_expiry(struct list_head *mounts)
1093 struct namespace *namespace;
1094 struct vfsmount *mnt, *next;
1095 LIST_HEAD(graveyard);
1097 if (list_empty(mounts))
1100 spin_lock(&vfsmount_lock);
1102 /* extract from the expiration list every vfsmount that matches the
1103 * following criteria:
1104 * - only referenced by its parent vfsmount
1105 * - still marked for expiry (marked on the last call here; marks are
1106 * cleared by mntput())
1108 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1109 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1110 atomic_read(&mnt->mnt_count) != 1)
1114 list_move(&mnt->mnt_expire, &graveyard);
1118 * go through the vfsmounts we've just consigned to the graveyard to
1119 * - check that they're still dead
1120 * - delete the vfsmount from the appropriate namespace under lock
1121 * - dispose of the corpse
1123 while (!list_empty(&graveyard)) {
1125 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1126 list_del_init(&mnt->mnt_expire);
1128 /* don't do anything if the namespace is dead - all the
1129 * vfsmounts from it are going away anyway */
1130 namespace = mnt->mnt_namespace;
1131 if (!namespace || !namespace->root)
1133 get_namespace(namespace);
1135 spin_unlock(&vfsmount_lock);
1136 down_write(&namespace_sem);
1137 expire_mount(mnt, mounts, &umounts);
1138 up_write(&namespace_sem);
1139 release_mounts(&umounts);
1141 put_namespace(namespace);
1142 spin_lock(&vfsmount_lock);
1145 spin_unlock(&vfsmount_lock);
1148 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1151 * Some copy_from_user() implementations do not return the exact number of
1152 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1153 * Note that this function differs from copy_from_user() in that it will oops
1154 * on bad values of `to', rather than returning a short copy.
1156 static long exact_copy_from_user(void *to, const void __user * from,
1160 const char __user *f = from;
1163 if (!access_ok(VERIFY_READ, from, n))
1167 if (__get_user(c, f)) {
1178 int copy_mount_options(const void __user * data, unsigned long *where)
1188 if (!(page = __get_free_page(GFP_KERNEL)))
1191 /* We only care that *some* data at the address the user
1192 * gave us is valid. Just in case, we'll zero
1193 * the remainder of the page.
1195 /* copy_from_user cannot cross TASK_SIZE ! */
1196 size = TASK_SIZE - (unsigned long)data;
1197 if (size > PAGE_SIZE)
1200 i = size - exact_copy_from_user((void *)page, data, size);
1206 memset((char *)page + i, 0, PAGE_SIZE - i);
1212 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1213 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1215 * data is a (void *) that can point to any structure up to
1216 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1217 * information (or be NULL).
1219 * Pre-0.97 versions of mount() didn't have a flags word.
1220 * When the flags word was introduced its top half was required
1221 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1222 * Therefore, if this magic number is present, it carries no information
1223 * and must be discarded.
1225 long do_mount(char *dev_name, char *dir_name, char *type_page,
1226 unsigned long flags, void *data_page)
1228 struct nameidata nd;
1233 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1234 flags &= ~MS_MGC_MSK;
1236 /* Basic sanity checks */
1238 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1240 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1244 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1246 /* Separate the per-mountpoint flags */
1247 if (flags & MS_NOSUID)
1248 mnt_flags |= MNT_NOSUID;
1249 if (flags & MS_NODEV)
1250 mnt_flags |= MNT_NODEV;
1251 if (flags & MS_NOEXEC)
1252 mnt_flags |= MNT_NOEXEC;
1253 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE);
1255 /* ... and get the mountpoint */
1256 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1260 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1264 if (flags & MS_REMOUNT)
1265 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1267 else if (flags & MS_BIND)
1268 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1269 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE))
1270 retval = do_change_type(&nd, flags);
1271 else if (flags & MS_MOVE)
1272 retval = do_move_mount(&nd, dev_name);
1274 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1275 dev_name, data_page);
1281 int copy_namespace(int flags, struct task_struct *tsk)
1283 struct namespace *namespace = tsk->namespace;
1284 struct namespace *new_ns;
1285 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1286 struct fs_struct *fs = tsk->fs;
1287 struct vfsmount *p, *q;
1292 get_namespace(namespace);
1294 if (!(flags & CLONE_NEWNS))
1297 if (!capable(CAP_SYS_ADMIN)) {
1298 put_namespace(namespace);
1302 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1306 atomic_set(&new_ns->count, 1);
1307 INIT_LIST_HEAD(&new_ns->list);
1308 init_waitqueue_head(&new_ns->poll);
1311 down_write(&namespace_sem);
1312 /* First pass: copy the tree topology */
1313 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1315 if (!new_ns->root) {
1316 up_write(&namespace_sem);
1320 spin_lock(&vfsmount_lock);
1321 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1322 spin_unlock(&vfsmount_lock);
1325 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1326 * as belonging to new namespace. We have already acquired a private
1327 * fs_struct, so tsk->fs->lock is not needed.
1329 p = namespace->root;
1332 q->mnt_namespace = new_ns;
1334 if (p == fs->rootmnt) {
1336 fs->rootmnt = mntget(q);
1338 if (p == fs->pwdmnt) {
1340 fs->pwdmnt = mntget(q);
1342 if (p == fs->altrootmnt) {
1344 fs->altrootmnt = mntget(q);
1347 p = next_mnt(p, namespace->root);
1348 q = next_mnt(q, new_ns->root);
1350 up_write(&namespace_sem);
1352 tsk->namespace = new_ns;
1361 put_namespace(namespace);
1365 put_namespace(namespace);
1369 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1370 char __user * type, unsigned long flags,
1374 unsigned long data_page;
1375 unsigned long type_page;
1376 unsigned long dev_page;
1379 retval = copy_mount_options(type, &type_page);
1383 dir_page = getname(dir_name);
1384 retval = PTR_ERR(dir_page);
1385 if (IS_ERR(dir_page))
1388 retval = copy_mount_options(dev_name, &dev_page);
1392 retval = copy_mount_options(data, &data_page);
1397 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1398 flags, (void *)data_page);
1400 free_page(data_page);
1403 free_page(dev_page);
1407 free_page(type_page);
1412 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1413 * It can block. Requires the big lock held.
1415 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1416 struct dentry *dentry)
1418 struct dentry *old_root;
1419 struct vfsmount *old_rootmnt;
1420 write_lock(&fs->lock);
1421 old_root = fs->root;
1422 old_rootmnt = fs->rootmnt;
1423 fs->rootmnt = mntget(mnt);
1424 fs->root = dget(dentry);
1425 write_unlock(&fs->lock);
1428 mntput(old_rootmnt);
1433 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1434 * It can block. Requires the big lock held.
1436 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1437 struct dentry *dentry)
1439 struct dentry *old_pwd;
1440 struct vfsmount *old_pwdmnt;
1442 write_lock(&fs->lock);
1444 old_pwdmnt = fs->pwdmnt;
1445 fs->pwdmnt = mntget(mnt);
1446 fs->pwd = dget(dentry);
1447 write_unlock(&fs->lock);
1455 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1457 struct task_struct *g, *p;
1458 struct fs_struct *fs;
1460 read_lock(&tasklist_lock);
1461 do_each_thread(g, p) {
1465 atomic_inc(&fs->count);
1467 if (fs->root == old_nd->dentry
1468 && fs->rootmnt == old_nd->mnt)
1469 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1470 if (fs->pwd == old_nd->dentry
1471 && fs->pwdmnt == old_nd->mnt)
1472 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1476 } while_each_thread(g, p);
1477 read_unlock(&tasklist_lock);
1481 * pivot_root Semantics:
1482 * Moves the root file system of the current process to the directory put_old,
1483 * makes new_root as the new root file system of the current process, and sets
1484 * root/cwd of all processes which had them on the current root to new_root.
1487 * The new_root and put_old must be directories, and must not be on the
1488 * same file system as the current process root. The put_old must be
1489 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1490 * pointed to by put_old must yield the same directory as new_root. No other
1491 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1494 * - we don't move root/cwd if they are not at the root (reason: if something
1495 * cared enough to change them, it's probably wrong to force them elsewhere)
1496 * - it's okay to pick a root that isn't the root of a file system, e.g.
1497 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1498 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1501 asmlinkage long sys_pivot_root(const char __user * new_root,
1502 const char __user * put_old)
1504 struct vfsmount *tmp;
1505 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1508 if (!capable(CAP_SYS_ADMIN))
1513 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1518 if (!check_mnt(new_nd.mnt))
1521 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1525 error = security_sb_pivotroot(&old_nd, &new_nd);
1527 path_release(&old_nd);
1531 read_lock(¤t->fs->lock);
1532 user_nd.mnt = mntget(current->fs->rootmnt);
1533 user_nd.dentry = dget(current->fs->root);
1534 read_unlock(¤t->fs->lock);
1535 down_write(&namespace_sem);
1536 down(&old_nd.dentry->d_inode->i_sem);
1538 if (IS_MNT_SHARED(old_nd.mnt) ||
1539 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1540 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1542 if (!check_mnt(user_nd.mnt))
1545 if (IS_DEADDIR(new_nd.dentry->d_inode))
1547 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1549 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1552 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1553 goto out2; /* loop, on the same file system */
1555 if (user_nd.mnt->mnt_root != user_nd.dentry)
1556 goto out2; /* not a mountpoint */
1557 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1558 goto out2; /* not attached */
1559 if (new_nd.mnt->mnt_root != new_nd.dentry)
1560 goto out2; /* not a mountpoint */
1561 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1562 goto out2; /* not attached */
1563 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1564 spin_lock(&vfsmount_lock);
1565 if (tmp != new_nd.mnt) {
1567 if (tmp->mnt_parent == tmp)
1568 goto out3; /* already mounted on put_old */
1569 if (tmp->mnt_parent == new_nd.mnt)
1571 tmp = tmp->mnt_parent;
1573 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1575 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1577 detach_mnt(new_nd.mnt, &parent_nd);
1578 detach_mnt(user_nd.mnt, &root_parent);
1579 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1580 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1581 touch_namespace(current->namespace);
1582 spin_unlock(&vfsmount_lock);
1583 chroot_fs_refs(&user_nd, &new_nd);
1584 security_sb_post_pivotroot(&user_nd, &new_nd);
1586 path_release(&root_parent);
1587 path_release(&parent_nd);
1589 up(&old_nd.dentry->d_inode->i_sem);
1590 up_write(&namespace_sem);
1591 path_release(&user_nd);
1592 path_release(&old_nd);
1594 path_release(&new_nd);
1599 spin_unlock(&vfsmount_lock);
1603 static void __init init_mount_tree(void)
1605 struct vfsmount *mnt;
1606 struct namespace *namespace;
1607 struct task_struct *g, *p;
1609 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1611 panic("Can't create rootfs");
1612 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1614 panic("Can't allocate initial namespace");
1615 atomic_set(&namespace->count, 1);
1616 INIT_LIST_HEAD(&namespace->list);
1617 init_waitqueue_head(&namespace->poll);
1618 namespace->event = 0;
1619 list_add(&mnt->mnt_list, &namespace->list);
1620 namespace->root = mnt;
1621 mnt->mnt_namespace = namespace;
1623 init_task.namespace = namespace;
1624 read_lock(&tasklist_lock);
1625 do_each_thread(g, p) {
1626 get_namespace(namespace);
1627 p->namespace = namespace;
1628 } while_each_thread(g, p);
1629 read_unlock(&tasklist_lock);
1631 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1632 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1635 void __init mnt_init(unsigned long mempages)
1637 struct list_head *d;
1638 unsigned int nr_hash;
1641 init_rwsem(&namespace_sem);
1643 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1644 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1646 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1648 if (!mount_hashtable)
1649 panic("Failed to allocate mount hash table\n");
1652 * Find the power-of-two list-heads that can fit into the allocation..
1653 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1656 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1660 } while ((nr_hash >> hash_bits) != 0);
1664 * Re-calculate the actual number of entries and the mask
1665 * from the number of bits we can fit.
1667 nr_hash = 1UL << hash_bits;
1668 hash_mask = nr_hash - 1;
1670 printk("Mount-cache hash table entries: %d\n", nr_hash);
1672 /* And initialize the newly allocated array */
1673 d = mount_hashtable;
1685 void __put_namespace(struct namespace *namespace)
1687 struct vfsmount *root = namespace->root;
1688 LIST_HEAD(umount_list);
1689 namespace->root = NULL;
1690 spin_unlock(&vfsmount_lock);
1691 down_write(&namespace_sem);
1692 spin_lock(&vfsmount_lock);
1693 umount_tree(root, 0, &umount_list);
1694 spin_unlock(&vfsmount_lock);
1695 up_write(&namespace_sem);
1696 release_mounts(&umount_list);