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/quotaops.h>
17 #include <linux/acct.h>
18 #include <linux/capability.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 <linux/vs_base.h>
26 #include <linux/vserver/namespace.h>
27 #include <linux/vserver/xid.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
32 extern int __init init_rootfs(void);
35 extern int __init sysfs_init(void);
37 static inline int sysfs_init(void)
43 /* spinlock for vfsmount related operations, inplace of dcache_lock */
44 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
48 static struct list_head *mount_hashtable __read_mostly;
49 static int hash_mask __read_mostly, hash_bits __read_mostly;
50 static kmem_cache_t *mnt_cache __read_mostly;
51 static struct rw_semaphore namespace_sem;
54 decl_subsys(fs, NULL, NULL);
55 EXPORT_SYMBOL_GPL(fs_subsys);
57 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
60 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
61 tmp = tmp + (tmp >> hash_bits);
62 return tmp & hash_mask;
65 struct vfsmount *alloc_vfsmnt(const char *name)
67 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
69 memset(mnt, 0, sizeof(struct vfsmount));
70 atomic_set(&mnt->mnt_count, 1);
71 INIT_LIST_HEAD(&mnt->mnt_hash);
72 INIT_LIST_HEAD(&mnt->mnt_child);
73 INIT_LIST_HEAD(&mnt->mnt_mounts);
74 INIT_LIST_HEAD(&mnt->mnt_list);
75 INIT_LIST_HEAD(&mnt->mnt_expire);
76 INIT_LIST_HEAD(&mnt->mnt_share);
77 INIT_LIST_HEAD(&mnt->mnt_slave_list);
78 INIT_LIST_HEAD(&mnt->mnt_slave);
80 int size = strlen(name) + 1;
81 char *newname = kmalloc(size, GFP_KERNEL);
83 memcpy(newname, name, size);
84 mnt->mnt_devname = newname;
91 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
94 mnt->mnt_root = dget(sb->s_root);
98 EXPORT_SYMBOL(simple_set_mnt);
100 void free_vfsmnt(struct vfsmount *mnt)
102 kfree(mnt->mnt_devname);
103 kmem_cache_free(mnt_cache, mnt);
107 * find the first or last mount at @dentry on vfsmount @mnt depending on
108 * @dir. If @dir is set return the first mount else return the last mount.
110 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
113 struct list_head *head = mount_hashtable + hash(mnt, dentry);
114 struct list_head *tmp = head;
115 struct vfsmount *p, *found = NULL;
118 tmp = dir ? tmp->next : tmp->prev;
122 p = list_entry(tmp, struct vfsmount, mnt_hash);
123 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
132 * lookup_mnt increments the ref count before returning
133 * the vfsmount struct.
135 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
137 struct vfsmount *child_mnt;
138 spin_lock(&vfsmount_lock);
139 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
141 spin_unlock(&vfsmount_lock);
145 static inline int check_mnt(struct vfsmount *mnt)
147 return mnt->mnt_namespace == current->namespace;
150 static void touch_namespace(struct namespace *ns)
154 wake_up_interruptible(&ns->poll);
158 static void __touch_namespace(struct namespace *ns)
160 if (ns && ns->event != event) {
162 wake_up_interruptible(&ns->poll);
166 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
168 old_nd->dentry = mnt->mnt_mountpoint;
169 old_nd->mnt = mnt->mnt_parent;
170 mnt->mnt_parent = mnt;
171 mnt->mnt_mountpoint = mnt->mnt_root;
172 list_del_init(&mnt->mnt_child);
173 list_del_init(&mnt->mnt_hash);
174 old_nd->dentry->d_mounted--;
177 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
178 struct vfsmount *child_mnt)
180 child_mnt->mnt_parent = mntget(mnt);
181 child_mnt->mnt_mountpoint = dget(dentry);
185 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
187 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
188 list_add_tail(&mnt->mnt_hash, mount_hashtable +
189 hash(nd->mnt, nd->dentry));
190 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
194 * the caller must hold vfsmount_lock
196 static void commit_tree(struct vfsmount *mnt)
198 struct vfsmount *parent = mnt->mnt_parent;
201 struct namespace *n = parent->mnt_namespace;
203 BUG_ON(parent == mnt);
205 list_add_tail(&head, &mnt->mnt_list);
206 list_for_each_entry(m, &head, mnt_list)
207 m->mnt_namespace = n;
208 list_splice(&head, n->list.prev);
210 list_add_tail(&mnt->mnt_hash, mount_hashtable +
211 hash(parent, mnt->mnt_mountpoint));
212 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
216 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
218 struct list_head *next = p->mnt_mounts.next;
219 if (next == &p->mnt_mounts) {
223 next = p->mnt_child.next;
224 if (next != &p->mnt_parent->mnt_mounts)
229 return list_entry(next, struct vfsmount, mnt_child);
232 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
234 struct list_head *prev = p->mnt_mounts.prev;
235 while (prev != &p->mnt_mounts) {
236 p = list_entry(prev, struct vfsmount, mnt_child);
237 prev = p->mnt_mounts.prev;
242 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
245 struct super_block *sb = old->mnt_sb;
246 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
249 mnt->mnt_flags = old->mnt_flags;
250 atomic_inc(&sb->s_active);
252 mnt->mnt_root = dget(root);
253 mnt->mnt_mountpoint = mnt->mnt_root;
254 mnt->mnt_parent = mnt;
255 mnt->mnt_xid = old->mnt_xid;
257 if (flag & CL_SLAVE) {
258 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
259 mnt->mnt_master = old;
260 CLEAR_MNT_SHARED(mnt);
262 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
263 list_add(&mnt->mnt_share, &old->mnt_share);
264 if (IS_MNT_SLAVE(old))
265 list_add(&mnt->mnt_slave, &old->mnt_slave);
266 mnt->mnt_master = old->mnt_master;
268 if (flag & CL_MAKE_SHARED)
271 /* stick the duplicate mount on the same expiry list
272 * as the original if that was on one */
273 if (flag & CL_EXPIRE) {
274 spin_lock(&vfsmount_lock);
275 if (!list_empty(&old->mnt_expire))
276 list_add(&mnt->mnt_expire, &old->mnt_expire);
277 spin_unlock(&vfsmount_lock);
283 static inline void __mntput(struct vfsmount *mnt)
285 struct super_block *sb = mnt->mnt_sb;
288 deactivate_super(sb);
291 void mntput_no_expire(struct vfsmount *mnt)
294 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
295 if (likely(!mnt->mnt_pinned)) {
296 spin_unlock(&vfsmount_lock);
300 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
302 spin_unlock(&vfsmount_lock);
303 acct_auto_close_mnt(mnt);
304 security_sb_umount_close(mnt);
309 EXPORT_SYMBOL(mntput_no_expire);
311 void mnt_pin(struct vfsmount *mnt)
313 spin_lock(&vfsmount_lock);
315 spin_unlock(&vfsmount_lock);
318 EXPORT_SYMBOL(mnt_pin);
320 void mnt_unpin(struct vfsmount *mnt)
322 spin_lock(&vfsmount_lock);
323 if (mnt->mnt_pinned) {
324 atomic_inc(&mnt->mnt_count);
327 spin_unlock(&vfsmount_lock);
330 EXPORT_SYMBOL(mnt_unpin);
333 static void *m_start(struct seq_file *m, loff_t *pos)
335 struct namespace *n = m->private;
339 down_read(&namespace_sem);
340 list_for_each(p, &n->list)
342 return list_entry(p, struct vfsmount, mnt_list);
346 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
348 struct namespace *n = m->private;
349 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
351 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
354 static void m_stop(struct seq_file *m, void *v)
356 up_read(&namespace_sem);
359 static inline void mangle(struct seq_file *m, const char *s)
361 seq_escape(m, s, " \t\n\\");
364 static int mnt_is_reachable(struct vfsmount *mnt)
366 struct vfsmount *root_mnt;
367 struct dentry *root, *point;
370 if (mnt == mnt->mnt_namespace->root)
373 spin_lock(&dcache_lock);
374 root_mnt = current->fs->rootmnt;
375 root = current->fs->root;
378 while ((mnt != mnt->mnt_parent) && (mnt != root_mnt)) {
379 point = mnt->mnt_mountpoint;
380 mnt = mnt->mnt_parent;
383 ret = (mnt == root_mnt) && is_subdir(point, root);
385 spin_unlock(&dcache_lock);
390 static int show_vfsmnt(struct seq_file *m, void *v)
392 struct vfsmount *mnt = v;
394 static struct proc_fs_info {
400 { MS_RDONLY, MNT_RDONLY, "ro", "rw" },
401 { MS_SYNCHRONOUS, 0, ",sync", NULL },
402 { MS_DIRSYNC, 0, ",dirsync", NULL },
403 { MS_MANDLOCK, 0, ",mand", NULL },
404 { MS_TAGXID, 0, ",tagxid", NULL },
405 { MS_NOATIME, MNT_NOATIME, ",noatime", NULL },
406 { MS_NODIRATIME, MNT_NODIRATIME, ",nodiratime", NULL },
407 { 0, MNT_NOSUID, ",nosuid", NULL },
408 { 0, MNT_NODEV, ",nodev", NULL },
409 { 0, MNT_NOEXEC, ",noexec", NULL },
412 struct proc_fs_info *p;
413 unsigned long s_flags = mnt->mnt_sb->s_flags;
414 int mnt_flags = mnt->mnt_flags;
416 if (vx_flags(VXF_HIDE_MOUNT, 0))
418 if (!mnt_is_reachable(mnt))
421 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
422 mnt == current->fs->rootmnt) {
423 seq_puts(m, "/dev/root / ");
425 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
427 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
430 mangle(m, mnt->mnt_sb->s_type->name);
432 for (p = fs_info; (p->s_flag | p->mnt_flag) ; p++) {
433 if ((s_flags & p->s_flag) || (mnt_flags & p->mnt_flag)) {
435 seq_puts(m, p->set_str);
438 seq_puts(m, p->unset_str);
441 if (mnt->mnt_flags & MNT_XID)
442 seq_printf(m, ",xid=%d", mnt->mnt_xid);
443 if (mnt->mnt_sb->s_op->show_options)
444 err = mnt->mnt_sb->s_op->show_options(m, mnt);
445 seq_puts(m, " 0 0\n");
449 struct seq_operations mounts_op = {
456 static int show_vfsstat(struct seq_file *m, void *v)
458 struct vfsmount *mnt = v;
461 if (vx_flags(VXF_HIDE_MOUNT, 0))
463 if (!mnt_is_reachable(mnt) && !vx_check(0, VX_WATCH))
466 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
467 mnt == current->fs->rootmnt) {
468 seq_puts(m, "device /dev/root mounted on / ");
471 if (mnt->mnt_devname) {
472 seq_puts(m, "device ");
473 mangle(m, mnt->mnt_devname);
475 seq_puts(m, "no device");
478 seq_puts(m, " mounted on ");
479 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
483 /* file system type */
484 seq_puts(m, "with fstype ");
485 mangle(m, mnt->mnt_sb->s_type->name);
487 /* optional statistics */
488 if (mnt->mnt_sb->s_op->show_stats) {
490 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
497 struct seq_operations mountstats_op = {
501 .show = show_vfsstat,
505 * may_umount_tree - check if a mount tree is busy
506 * @mnt: root of mount tree
508 * This is called to check if a tree of mounts has any
509 * open files, pwds, chroots or sub mounts that are
512 int may_umount_tree(struct vfsmount *mnt)
515 int minimum_refs = 0;
518 spin_lock(&vfsmount_lock);
519 for (p = mnt; p; p = next_mnt(p, mnt)) {
520 actual_refs += atomic_read(&p->mnt_count);
523 spin_unlock(&vfsmount_lock);
525 if (actual_refs > minimum_refs)
531 EXPORT_SYMBOL(may_umount_tree);
534 * may_umount - check if a mount point is busy
535 * @mnt: root of mount
537 * This is called to check if a mount point has any
538 * open files, pwds, chroots or sub mounts. If the
539 * mount has sub mounts this will return busy
540 * regardless of whether the sub mounts are busy.
542 * Doesn't take quota and stuff into account. IOW, in some cases it will
543 * give false negatives. The main reason why it's here is that we need
544 * a non-destructive way to look for easily umountable filesystems.
546 int may_umount(struct vfsmount *mnt)
549 spin_lock(&vfsmount_lock);
550 if (propagate_mount_busy(mnt, 2))
552 spin_unlock(&vfsmount_lock);
556 EXPORT_SYMBOL(may_umount);
558 void release_mounts(struct list_head *head)
560 struct vfsmount *mnt;
561 while (!list_empty(head)) {
562 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
563 list_del_init(&mnt->mnt_hash);
564 if (mnt->mnt_parent != mnt) {
565 struct dentry *dentry;
567 spin_lock(&vfsmount_lock);
568 dentry = mnt->mnt_mountpoint;
570 mnt->mnt_mountpoint = mnt->mnt_root;
571 mnt->mnt_parent = mnt;
572 spin_unlock(&vfsmount_lock);
580 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
584 for (p = mnt; p; p = next_mnt(p, mnt))
585 list_move(&p->mnt_hash, kill);
588 propagate_umount(kill);
590 list_for_each_entry(p, kill, mnt_hash) {
591 list_del_init(&p->mnt_expire);
592 list_del_init(&p->mnt_list);
593 __touch_namespace(p->mnt_namespace);
594 p->mnt_namespace = NULL;
595 list_del_init(&p->mnt_child);
596 if (p->mnt_parent != p)
597 p->mnt_mountpoint->d_mounted--;
598 change_mnt_propagation(p, MS_PRIVATE);
602 static int do_umount(struct vfsmount *mnt, int flags)
604 struct super_block *sb = mnt->mnt_sb;
606 LIST_HEAD(umount_list);
608 retval = security_sb_umount(mnt, flags);
613 * Allow userspace to request a mountpoint be expired rather than
614 * unmounting unconditionally. Unmount only happens if:
615 * (1) the mark is already set (the mark is cleared by mntput())
616 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
618 if (flags & MNT_EXPIRE) {
619 if (mnt == current->fs->rootmnt ||
620 flags & (MNT_FORCE | MNT_DETACH))
623 if (atomic_read(&mnt->mnt_count) != 2)
626 if (!xchg(&mnt->mnt_expiry_mark, 1))
631 * If we may have to abort operations to get out of this
632 * mount, and they will themselves hold resources we must
633 * allow the fs to do things. In the Unix tradition of
634 * 'Gee thats tricky lets do it in userspace' the umount_begin
635 * might fail to complete on the first run through as other tasks
636 * must return, and the like. Thats for the mount program to worry
637 * about for the moment.
641 if (sb->s_op->umount_begin)
642 sb->s_op->umount_begin(mnt, flags);
646 * No sense to grab the lock for this test, but test itself looks
647 * somewhat bogus. Suggestions for better replacement?
648 * Ho-hum... In principle, we might treat that as umount + switch
649 * to rootfs. GC would eventually take care of the old vfsmount.
650 * Actually it makes sense, especially if rootfs would contain a
651 * /reboot - static binary that would close all descriptors and
652 * call reboot(9). Then init(8) could umount root and exec /reboot.
654 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
656 * Special case for "unmounting" root ...
657 * we just try to remount it readonly.
659 down_write(&sb->s_umount);
660 if (!(sb->s_flags & MS_RDONLY)) {
663 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
666 up_write(&sb->s_umount);
670 down_write(&namespace_sem);
671 spin_lock(&vfsmount_lock);
675 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
676 if (!list_empty(&mnt->mnt_list))
677 umount_tree(mnt, 1, &umount_list);
680 spin_unlock(&vfsmount_lock);
682 security_sb_umount_busy(mnt);
683 up_write(&namespace_sem);
684 release_mounts(&umount_list);
689 * Now umount can handle mount points as well as block devices.
690 * This is important for filesystems which use unnamed block devices.
692 * We now support a flag for forced unmount like the other 'big iron'
693 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
696 asmlinkage long sys_umount(char __user * name, int flags)
701 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
705 if (nd.dentry != nd.mnt->mnt_root)
707 if (!check_mnt(nd.mnt))
711 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
714 retval = do_umount(nd.mnt, flags);
716 path_release_on_umount(&nd);
721 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
724 * The 2.0 compatible umount. No flags.
726 asmlinkage long sys_oldumount(char __user * name)
728 return sys_umount(name, 0);
733 static int mount_is_safe(struct nameidata *nd)
735 if (vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
739 if (S_ISLNK(nd->dentry->d_inode->i_mode))
741 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
742 if (current->uid != nd->dentry->d_inode->i_uid)
745 if (vfs_permission(nd, MAY_WRITE))
751 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
756 if (d == NULL || d == d->d_parent)
762 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
765 struct vfsmount *res, *p, *q, *r, *s;
768 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
771 res = q = clone_mnt(mnt, dentry, flag);
774 q->mnt_mountpoint = mnt->mnt_mountpoint;
777 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
778 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
781 for (s = r; s; s = next_mnt(s, r)) {
782 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
783 s = skip_mnt_tree(s);
786 while (p != s->mnt_parent) {
792 nd.dentry = p->mnt_mountpoint;
793 q = clone_mnt(p, p->mnt_root, flag);
796 spin_lock(&vfsmount_lock);
797 list_add_tail(&q->mnt_list, &res->mnt_list);
799 spin_unlock(&vfsmount_lock);
805 LIST_HEAD(umount_list);
806 spin_lock(&vfsmount_lock);
807 umount_tree(res, 0, &umount_list);
808 spin_unlock(&vfsmount_lock);
809 release_mounts(&umount_list);
815 * @source_mnt : mount tree to be attached
816 * @nd : place the mount tree @source_mnt is attached
817 * @parent_nd : if non-null, detach the source_mnt from its parent and
818 * store the parent mount and mountpoint dentry.
819 * (done when source_mnt is moved)
821 * NOTE: in the table below explains the semantics when a source mount
822 * of a given type is attached to a destination mount of a given type.
823 * ---------------------------------------------------------------------------
824 * | BIND MOUNT OPERATION |
825 * |**************************************************************************
826 * | source-->| shared | private | slave | unbindable |
830 * |**************************************************************************
831 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
833 * |non-shared| shared (+) | private | slave (*) | invalid |
834 * ***************************************************************************
835 * A bind operation clones the source mount and mounts the clone on the
838 * (++) the cloned mount is propagated to all the mounts in the propagation
839 * tree of the destination mount and the cloned mount is added to
840 * the peer group of the source mount.
841 * (+) the cloned mount is created under the destination mount and is marked
842 * as shared. The cloned mount is added to the peer group of the source
844 * (+++) the mount is propagated to all the mounts in the propagation tree
845 * of the destination mount and the cloned mount is made slave
846 * of the same master as that of the source mount. The cloned mount
847 * is marked as 'shared and slave'.
848 * (*) the cloned mount is made a slave of the same master as that of the
851 * ---------------------------------------------------------------------------
852 * | MOVE MOUNT OPERATION |
853 * |**************************************************************************
854 * | source-->| shared | private | slave | unbindable |
858 * |**************************************************************************
859 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
861 * |non-shared| shared (+*) | private | slave (*) | unbindable |
862 * ***************************************************************************
864 * (+) the mount is moved to the destination. And is then propagated to
865 * all the mounts in the propagation tree of the destination mount.
866 * (+*) the mount is moved to the destination.
867 * (+++) the mount is moved to the destination and is then propagated to
868 * all the mounts belonging to the destination mount's propagation tree.
869 * the mount is marked as 'shared and slave'.
870 * (*) the mount continues to be a slave at the new location.
872 * if the source mount is a tree, the operations explained above is
873 * applied to each mount in the tree.
874 * Must be called without spinlocks held, since this function can sleep
877 static int attach_recursive_mnt(struct vfsmount *source_mnt,
878 struct nameidata *nd, struct nameidata *parent_nd)
880 LIST_HEAD(tree_list);
881 struct vfsmount *dest_mnt = nd->mnt;
882 struct dentry *dest_dentry = nd->dentry;
883 struct vfsmount *child, *p;
885 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
888 if (IS_MNT_SHARED(dest_mnt)) {
889 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
893 spin_lock(&vfsmount_lock);
895 detach_mnt(source_mnt, parent_nd);
896 attach_mnt(source_mnt, nd);
897 touch_namespace(current->namespace);
899 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
900 commit_tree(source_mnt);
903 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
904 list_del_init(&child->mnt_hash);
907 spin_unlock(&vfsmount_lock);
911 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
914 if (mnt->mnt_sb->s_flags & MS_NOUSER)
917 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
918 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
922 mutex_lock(&nd->dentry->d_inode->i_mutex);
923 if (IS_DEADDIR(nd->dentry->d_inode))
926 err = security_sb_check_sb(mnt, nd);
931 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
932 err = attach_recursive_mnt(mnt, nd, NULL);
934 mutex_unlock(&nd->dentry->d_inode->i_mutex);
936 security_sb_post_addmount(mnt, nd);
941 * recursively change the type of the mountpoint.
943 static int do_change_type(struct nameidata *nd, int flag)
945 struct vfsmount *m, *mnt = nd->mnt;
946 int recurse = flag & MS_REC;
947 int type = flag & ~MS_REC;
949 if (nd->dentry != nd->mnt->mnt_root)
952 down_write(&namespace_sem);
953 spin_lock(&vfsmount_lock);
954 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
955 change_mnt_propagation(m, type);
956 spin_unlock(&vfsmount_lock);
957 up_write(&namespace_sem);
964 static int do_loopback(struct nameidata *nd, char *old_name, xid_t xid,
965 unsigned long flags, int mnt_flags)
967 struct nameidata old_nd;
968 struct vfsmount *mnt = NULL;
969 int err = mount_is_safe(nd);
970 int recurse = flags & MS_REC;
973 if (!old_name || !*old_name)
975 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
979 down_write(&namespace_sem);
981 if (IS_MNT_UNBINDABLE(old_nd.mnt))
984 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
989 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
991 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
996 mnt->mnt_flags = mnt_flags;
997 if (flags & MS_XID) {
999 mnt->mnt_flags |= MNT_XID;
1002 err = graft_tree(mnt, nd);
1004 LIST_HEAD(umount_list);
1005 spin_lock(&vfsmount_lock);
1006 umount_tree(mnt, 0, &umount_list);
1007 spin_unlock(&vfsmount_lock);
1008 release_mounts(&umount_list);
1010 mnt->mnt_flags = mnt_flags;
1013 up_write(&namespace_sem);
1014 path_release(&old_nd);
1019 * change filesystem flags. dir should be a physical root of filesystem.
1020 * If you've mounted a non-root directory somewhere and want to do remount
1021 * on it - tough luck.
1023 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1024 void *data, xid_t xid)
1027 struct super_block *sb = nd->mnt->mnt_sb;
1029 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_REMOUNT))
1032 if (!check_mnt(nd->mnt))
1035 if (nd->dentry != nd->mnt->mnt_root)
1038 down_write(&sb->s_umount);
1039 err = do_remount_sb(sb, flags, data, 0);
1041 nd->mnt->mnt_flags = mnt_flags;
1042 up_write(&sb->s_umount);
1044 security_sb_post_remount(nd->mnt, flags, data);
1048 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1051 for (p = mnt; p; p = next_mnt(p, mnt)) {
1052 if (IS_MNT_UNBINDABLE(p))
1058 static int do_move_mount(struct nameidata *nd, char *old_name)
1060 struct nameidata old_nd, parent_nd;
1063 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1065 if (!old_name || !*old_name)
1067 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1071 down_write(&namespace_sem);
1072 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1075 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1079 mutex_lock(&nd->dentry->d_inode->i_mutex);
1080 if (IS_DEADDIR(nd->dentry->d_inode))
1083 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1087 if (old_nd.dentry != old_nd.mnt->mnt_root)
1090 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1093 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1094 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1097 * Don't move a mount residing in a shared parent.
1099 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1102 * Don't move a mount tree containing unbindable mounts to a destination
1103 * mount which is shared.
1105 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1108 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1109 if (p == old_nd.mnt)
1112 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1115 spin_lock(&vfsmount_lock);
1116 /* if the mount is moved, it should no longer be expire
1118 list_del_init(&old_nd.mnt->mnt_expire);
1119 spin_unlock(&vfsmount_lock);
1121 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1123 up_write(&namespace_sem);
1125 path_release(&parent_nd);
1126 path_release(&old_nd);
1131 * create a new mount for userspace and request it to be added into the
1134 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1135 int mnt_flags, char *name, void *data)
1137 struct vfsmount *mnt;
1139 if (!type || !memchr(type, 0, PAGE_SIZE))
1142 /* we need capabilities... */
1143 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1146 mnt = do_kern_mount(type, flags, name, data);
1148 return PTR_ERR(mnt);
1150 return do_add_mount(mnt, nd, mnt_flags, NULL);
1154 * add a mount into a namespace's mount tree
1155 * - provide the option of adding the new mount to an expiration list
1157 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1158 int mnt_flags, struct list_head *fslist)
1162 down_write(&namespace_sem);
1163 /* Something was mounted here while we slept */
1164 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1167 if (!check_mnt(nd->mnt))
1170 /* Refuse the same filesystem on the same mount point */
1172 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1173 nd->mnt->mnt_root == nd->dentry)
1177 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1180 newmnt->mnt_flags = mnt_flags;
1181 if ((err = graft_tree(newmnt, nd)))
1185 /* add to the specified expiration list */
1186 spin_lock(&vfsmount_lock);
1187 list_add_tail(&newmnt->mnt_expire, fslist);
1188 spin_unlock(&vfsmount_lock);
1190 up_write(&namespace_sem);
1194 up_write(&namespace_sem);
1199 EXPORT_SYMBOL_GPL(do_add_mount);
1201 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1202 struct list_head *umounts)
1204 spin_lock(&vfsmount_lock);
1207 * Check if mount is still attached, if not, let whoever holds it deal
1210 if (mnt->mnt_parent == mnt) {
1211 spin_unlock(&vfsmount_lock);
1216 * Check that it is still dead: the count should now be 2 - as
1217 * contributed by the vfsmount parent and the mntget above
1219 if (!propagate_mount_busy(mnt, 2)) {
1220 /* delete from the namespace */
1221 touch_namespace(mnt->mnt_namespace);
1222 list_del_init(&mnt->mnt_list);
1223 mnt->mnt_namespace = NULL;
1224 umount_tree(mnt, 1, umounts);
1225 spin_unlock(&vfsmount_lock);
1228 * Someone brought it back to life whilst we didn't have any
1229 * locks held so return it to the expiration list
1231 list_add_tail(&mnt->mnt_expire, mounts);
1232 spin_unlock(&vfsmount_lock);
1237 * go through the vfsmounts we've just consigned to the graveyard to
1238 * - check that they're still dead
1239 * - delete the vfsmount from the appropriate namespace under lock
1240 * - dispose of the corpse
1242 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1244 struct namespace *namespace;
1245 struct vfsmount *mnt;
1247 while (!list_empty(graveyard)) {
1249 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1250 list_del_init(&mnt->mnt_expire);
1252 /* don't do anything if the namespace is dead - all the
1253 * vfsmounts from it are going away anyway */
1254 namespace = mnt->mnt_namespace;
1255 if (!namespace || !namespace->root)
1257 get_namespace(namespace);
1259 spin_unlock(&vfsmount_lock);
1260 down_write(&namespace_sem);
1261 expire_mount(mnt, mounts, &umounts);
1262 up_write(&namespace_sem);
1263 release_mounts(&umounts);
1265 put_namespace(namespace);
1266 spin_lock(&vfsmount_lock);
1271 * process a list of expirable mountpoints with the intent of discarding any
1272 * mountpoints that aren't in use and haven't been touched since last we came
1275 void mark_mounts_for_expiry(struct list_head *mounts)
1277 struct vfsmount *mnt, *next;
1278 LIST_HEAD(graveyard);
1280 if (list_empty(mounts))
1283 spin_lock(&vfsmount_lock);
1285 /* extract from the expiration list every vfsmount that matches the
1286 * following criteria:
1287 * - only referenced by its parent vfsmount
1288 * - still marked for expiry (marked on the last call here; marks are
1289 * cleared by mntput())
1291 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1292 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1293 atomic_read(&mnt->mnt_count) != 1)
1297 list_move(&mnt->mnt_expire, &graveyard);
1300 expire_mount_list(&graveyard, mounts);
1302 spin_unlock(&vfsmount_lock);
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)) {
1339 list_move_tail(&mnt->mnt_expire, graveyard);
1344 * All done at this level ... ascend and resume the search
1346 if (this_parent != parent) {
1347 next = this_parent->mnt_child.next;
1348 this_parent = this_parent->mnt_parent;
1355 * process a list of expirable mountpoints with the intent of discarding any
1356 * submounts of a specific parent mountpoint
1358 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1360 LIST_HEAD(graveyard);
1363 spin_lock(&vfsmount_lock);
1365 /* extract submounts of 'mountpoint' from the expiration list */
1366 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1367 expire_mount_list(&graveyard, mounts);
1369 spin_unlock(&vfsmount_lock);
1372 EXPORT_SYMBOL_GPL(shrink_submounts);
1375 * Some copy_from_user() implementations do not return the exact number of
1376 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1377 * Note that this function differs from copy_from_user() in that it will oops
1378 * on bad values of `to', rather than returning a short copy.
1380 static long exact_copy_from_user(void *to, const void __user * from,
1384 const char __user *f = from;
1387 if (!access_ok(VERIFY_READ, from, n))
1391 if (__get_user(c, f)) {
1402 int copy_mount_options(const void __user * data, unsigned long *where)
1412 if (!(page = __get_free_page(GFP_KERNEL)))
1415 /* We only care that *some* data at the address the user
1416 * gave us is valid. Just in case, we'll zero
1417 * the remainder of the page.
1419 /* copy_from_user cannot cross TASK_SIZE ! */
1420 size = TASK_SIZE - (unsigned long)data;
1421 if (size > PAGE_SIZE)
1424 i = size - exact_copy_from_user((void *)page, data, size);
1430 memset((char *)page + i, 0, PAGE_SIZE - i);
1436 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1437 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1439 * data is a (void *) that can point to any structure up to
1440 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1441 * information (or be NULL).
1443 * Pre-0.97 versions of mount() didn't have a flags word.
1444 * When the flags word was introduced its top half was required
1445 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1446 * Therefore, if this magic number is present, it carries no information
1447 * and must be discarded.
1449 long do_mount(char *dev_name, char *dir_name, char *type_page,
1450 unsigned long flags, void *data_page)
1452 struct nameidata nd;
1458 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1459 flags &= ~MS_MGC_MSK;
1461 /* Basic sanity checks */
1463 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1465 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1469 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1471 retval = vx_parse_xid(data_page, &xid, 1);
1473 mnt_flags |= MNT_XID;
1474 /* bind and re-mounts get xid flag */
1475 if (flags & (MS_BIND|MS_REMOUNT))
1479 /* Separate the per-mountpoint flags */
1480 if (flags & MS_RDONLY)
1481 mnt_flags |= MNT_RDONLY;
1482 if (flags & MS_NOSUID)
1483 mnt_flags |= MNT_NOSUID;
1484 if (flags & MS_NODEV)
1485 mnt_flags |= MNT_NODEV;
1486 if (flags & MS_NOEXEC)
1487 mnt_flags |= MNT_NOEXEC;
1488 if (flags & MS_NOATIME)
1489 mnt_flags |= MNT_NOATIME;
1490 if (flags & MS_NODIRATIME)
1491 mnt_flags |= MNT_NODIRATIME;
1493 if (!capable(CAP_SYS_ADMIN))
1494 mnt_flags |= MNT_NODEV;
1495 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1496 MS_NOATIME | MS_NODIRATIME);
1498 /* ... and get the mountpoint */
1499 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1503 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1507 if (flags & MS_REMOUNT)
1508 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1510 else if (flags & MS_BIND)
1511 retval = do_loopback(&nd, dev_name, xid, flags, mnt_flags);
1512 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1513 retval = do_change_type(&nd, flags);
1514 else if (flags & MS_MOVE)
1515 retval = do_move_mount(&nd, dev_name);
1517 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1518 dev_name, data_page);
1525 * Allocate a new namespace structure and populate it with contents
1526 * copied from the namespace of the passed in task structure.
1528 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1530 struct namespace *namespace = tsk->namespace;
1531 struct namespace *new_ns;
1532 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1533 struct vfsmount *p, *q;
1535 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1539 atomic_set(&new_ns->count, 1);
1540 INIT_LIST_HEAD(&new_ns->list);
1541 init_waitqueue_head(&new_ns->poll);
1544 down_write(&namespace_sem);
1545 /* First pass: copy the tree topology */
1546 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1547 CL_COPY_ALL | CL_EXPIRE);
1548 if (!new_ns->root) {
1549 up_write(&namespace_sem);
1553 spin_lock(&vfsmount_lock);
1554 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1555 spin_unlock(&vfsmount_lock);
1558 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1559 * as belonging to new namespace. We have already acquired a private
1560 * fs_struct, so tsk->fs->lock is not needed.
1562 p = namespace->root;
1565 q->mnt_namespace = new_ns;
1567 if (p == fs->rootmnt) {
1569 fs->rootmnt = mntget(q);
1571 if (p == fs->pwdmnt) {
1573 fs->pwdmnt = mntget(q);
1575 if (p == fs->altrootmnt) {
1577 fs->altrootmnt = mntget(q);
1580 p = next_mnt(p, namespace->root);
1581 q = next_mnt(q, new_ns->root);
1583 up_write(&namespace_sem);
1595 int copy_namespace(int flags, struct task_struct *tsk)
1597 struct namespace *namespace = tsk->namespace;
1598 struct namespace *new_ns;
1604 get_namespace(namespace);
1606 if (!(flags & CLONE_NEWNS))
1609 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT)) {
1614 new_ns = dup_namespace(tsk, tsk->fs);
1620 tsk->namespace = new_ns;
1623 put_namespace(namespace);
1627 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1628 char __user * type, unsigned long flags,
1632 unsigned long data_page;
1633 unsigned long type_page;
1634 unsigned long dev_page;
1637 retval = copy_mount_options(type, &type_page);
1641 dir_page = getname(dir_name);
1642 retval = PTR_ERR(dir_page);
1643 if (IS_ERR(dir_page))
1646 retval = copy_mount_options(dev_name, &dev_page);
1650 retval = copy_mount_options(data, &data_page);
1655 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1656 flags, (void *)data_page);
1658 free_page(data_page);
1661 free_page(dev_page);
1665 free_page(type_page);
1670 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1671 * It can block. Requires the big lock held.
1673 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1674 struct dentry *dentry)
1676 struct dentry *old_root;
1677 struct vfsmount *old_rootmnt;
1678 write_lock(&fs->lock);
1679 old_root = fs->root;
1680 old_rootmnt = fs->rootmnt;
1681 fs->rootmnt = mntget(mnt);
1682 fs->root = dget(dentry);
1683 write_unlock(&fs->lock);
1686 mntput(old_rootmnt);
1690 EXPORT_SYMBOL_GPL(set_fs_root);
1693 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1694 * It can block. Requires the big lock held.
1696 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1697 struct dentry *dentry)
1699 struct dentry *old_pwd;
1700 struct vfsmount *old_pwdmnt;
1702 write_lock(&fs->lock);
1704 old_pwdmnt = fs->pwdmnt;
1705 fs->pwdmnt = mntget(mnt);
1706 fs->pwd = dget(dentry);
1707 write_unlock(&fs->lock);
1715 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1717 struct task_struct *g, *p;
1718 struct fs_struct *fs;
1720 read_lock(&tasklist_lock);
1721 do_each_thread(g, p) {
1725 atomic_inc(&fs->count);
1727 if (fs->root == old_nd->dentry
1728 && fs->rootmnt == old_nd->mnt)
1729 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1730 if (fs->pwd == old_nd->dentry
1731 && fs->pwdmnt == old_nd->mnt)
1732 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1736 } while_each_thread(g, p);
1737 read_unlock(&tasklist_lock);
1741 * pivot_root Semantics:
1742 * Moves the root file system of the current process to the directory put_old,
1743 * makes new_root as the new root file system of the current process, and sets
1744 * root/cwd of all processes which had them on the current root to new_root.
1747 * The new_root and put_old must be directories, and must not be on the
1748 * same file system as the current process root. The put_old must be
1749 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1750 * pointed to by put_old must yield the same directory as new_root. No other
1751 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1753 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1754 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1755 * in this situation.
1758 * - we don't move root/cwd if they are not at the root (reason: if something
1759 * cared enough to change them, it's probably wrong to force them elsewhere)
1760 * - it's okay to pick a root that isn't the root of a file system, e.g.
1761 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1762 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1765 asmlinkage long sys_pivot_root(const char __user * new_root,
1766 const char __user * put_old)
1768 struct vfsmount *tmp;
1769 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1772 if (!capable(CAP_SYS_ADMIN))
1777 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1782 if (!check_mnt(new_nd.mnt))
1785 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1789 error = security_sb_pivotroot(&old_nd, &new_nd);
1791 path_release(&old_nd);
1795 read_lock(¤t->fs->lock);
1796 user_nd.mnt = mntget(current->fs->rootmnt);
1797 user_nd.dentry = dget(current->fs->root);
1798 read_unlock(¤t->fs->lock);
1799 down_write(&namespace_sem);
1800 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1802 if (IS_MNT_SHARED(old_nd.mnt) ||
1803 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1804 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1806 if (!check_mnt(user_nd.mnt))
1809 if (IS_DEADDIR(new_nd.dentry->d_inode))
1811 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1813 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1816 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1817 goto out2; /* loop, on the same file system */
1819 if (user_nd.mnt->mnt_root != user_nd.dentry)
1820 goto out2; /* not a mountpoint */
1821 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1822 goto out2; /* not attached */
1823 if (new_nd.mnt->mnt_root != new_nd.dentry)
1824 goto out2; /* not a mountpoint */
1825 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1826 goto out2; /* not attached */
1827 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1828 spin_lock(&vfsmount_lock);
1829 if (tmp != new_nd.mnt) {
1831 if (tmp->mnt_parent == tmp)
1832 goto out3; /* already mounted on put_old */
1833 if (tmp->mnt_parent == new_nd.mnt)
1835 tmp = tmp->mnt_parent;
1837 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1839 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1841 detach_mnt(new_nd.mnt, &parent_nd);
1842 detach_mnt(user_nd.mnt, &root_parent);
1843 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1844 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1845 touch_namespace(current->namespace);
1846 spin_unlock(&vfsmount_lock);
1847 chroot_fs_refs(&user_nd, &new_nd);
1848 security_sb_post_pivotroot(&user_nd, &new_nd);
1850 path_release(&root_parent);
1851 path_release(&parent_nd);
1853 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1854 up_write(&namespace_sem);
1855 path_release(&user_nd);
1856 path_release(&old_nd);
1858 path_release(&new_nd);
1863 spin_unlock(&vfsmount_lock);
1867 static void __init init_mount_tree(void)
1869 struct vfsmount *mnt;
1870 struct namespace *namespace;
1871 struct task_struct *g, *p;
1873 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1875 panic("Can't create rootfs");
1876 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1878 panic("Can't allocate initial namespace");
1879 atomic_set(&namespace->count, 1);
1880 INIT_LIST_HEAD(&namespace->list);
1881 init_waitqueue_head(&namespace->poll);
1882 namespace->event = 0;
1883 list_add(&mnt->mnt_list, &namespace->list);
1884 namespace->root = mnt;
1885 mnt->mnt_namespace = namespace;
1887 init_task.namespace = namespace;
1888 read_lock(&tasklist_lock);
1889 do_each_thread(g, p) {
1890 get_namespace(namespace);
1891 p->namespace = namespace;
1892 } while_each_thread(g, p);
1893 read_unlock(&tasklist_lock);
1895 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1896 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1899 void __init mnt_init(unsigned long mempages)
1901 struct list_head *d;
1902 unsigned int nr_hash;
1905 init_rwsem(&namespace_sem);
1907 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1908 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1910 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1912 if (!mount_hashtable)
1913 panic("Failed to allocate mount hash table\n");
1916 * Find the power-of-two list-heads that can fit into the allocation..
1917 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1920 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1924 } while ((nr_hash >> hash_bits) != 0);
1928 * Re-calculate the actual number of entries and the mask
1929 * from the number of bits we can fit.
1931 nr_hash = 1UL << hash_bits;
1932 hash_mask = nr_hash - 1;
1934 printk("Mount-cache hash table entries: %d\n", nr_hash);
1936 /* And initialize the newly allocated array */
1937 d = mount_hashtable;
1945 subsystem_register(&fs_subsys);
1950 void __put_namespace(struct namespace *namespace)
1952 struct vfsmount *root = namespace->root;
1953 LIST_HEAD(umount_list);
1954 namespace->root = NULL;
1955 spin_unlock(&vfsmount_lock);
1956 down_write(&namespace_sem);
1957 spin_lock(&vfsmount_lock);
1958 umount_tree(root, 0, &umount_list);
1959 spin_unlock(&vfsmount_lock);
1960 up_write(&namespace_sem);
1961 release_mounts(&umount_list);