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_PROPAGATION) || IS_MNT_SHARED(old))
231 list_add(&mnt->mnt_share, &old->mnt_share);
232 if (flag & CL_MAKE_SHARED)
235 /* stick the duplicate mount on the same expiry list
236 * as the original if that was on one */
237 if (flag & CL_EXPIRE) {
238 spin_lock(&vfsmount_lock);
239 if (!list_empty(&old->mnt_expire))
240 list_add(&mnt->mnt_expire, &old->mnt_expire);
241 spin_unlock(&vfsmount_lock);
247 static inline void __mntput(struct vfsmount *mnt)
249 struct super_block *sb = mnt->mnt_sb;
252 deactivate_super(sb);
255 void mntput_no_expire(struct vfsmount *mnt)
258 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
259 if (likely(!mnt->mnt_pinned)) {
260 spin_unlock(&vfsmount_lock);
264 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
266 spin_unlock(&vfsmount_lock);
267 acct_auto_close_mnt(mnt);
268 security_sb_umount_close(mnt);
273 EXPORT_SYMBOL(mntput_no_expire);
275 void mnt_pin(struct vfsmount *mnt)
277 spin_lock(&vfsmount_lock);
279 spin_unlock(&vfsmount_lock);
282 EXPORT_SYMBOL(mnt_pin);
284 void mnt_unpin(struct vfsmount *mnt)
286 spin_lock(&vfsmount_lock);
287 if (mnt->mnt_pinned) {
288 atomic_inc(&mnt->mnt_count);
291 spin_unlock(&vfsmount_lock);
294 EXPORT_SYMBOL(mnt_unpin);
297 static void *m_start(struct seq_file *m, loff_t *pos)
299 struct namespace *n = m->private;
303 down_read(&namespace_sem);
304 list_for_each(p, &n->list)
306 return list_entry(p, struct vfsmount, mnt_list);
310 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
312 struct namespace *n = m->private;
313 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
315 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
318 static void m_stop(struct seq_file *m, void *v)
320 up_read(&namespace_sem);
323 static inline void mangle(struct seq_file *m, const char *s)
325 seq_escape(m, s, " \t\n\\");
328 static int show_vfsmnt(struct seq_file *m, void *v)
330 struct vfsmount *mnt = v;
332 static struct proc_fs_info {
336 { MS_SYNCHRONOUS, ",sync" },
337 { MS_DIRSYNC, ",dirsync" },
338 { MS_MANDLOCK, ",mand" },
339 { MS_NOATIME, ",noatime" },
340 { MS_NODIRATIME, ",nodiratime" },
343 static struct proc_fs_info mnt_info[] = {
344 { MNT_NOSUID, ",nosuid" },
345 { MNT_NODEV, ",nodev" },
346 { MNT_NOEXEC, ",noexec" },
349 struct proc_fs_info *fs_infop;
351 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
353 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
355 mangle(m, mnt->mnt_sb->s_type->name);
356 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
357 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
358 if (mnt->mnt_sb->s_flags & fs_infop->flag)
359 seq_puts(m, fs_infop->str);
361 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
362 if (mnt->mnt_flags & fs_infop->flag)
363 seq_puts(m, fs_infop->str);
365 if (mnt->mnt_sb->s_op->show_options)
366 err = mnt->mnt_sb->s_op->show_options(m, mnt);
367 seq_puts(m, " 0 0\n");
371 struct seq_operations mounts_op = {
379 * may_umount_tree - check if a mount tree is busy
380 * @mnt: root of mount tree
382 * This is called to check if a tree of mounts has any
383 * open files, pwds, chroots or sub mounts that are
386 int may_umount_tree(struct vfsmount *mnt)
389 int minimum_refs = 0;
392 spin_lock(&vfsmount_lock);
393 for (p = mnt; p; p = next_mnt(p, mnt)) {
394 actual_refs += atomic_read(&p->mnt_count);
397 spin_unlock(&vfsmount_lock);
399 if (actual_refs > minimum_refs)
405 EXPORT_SYMBOL(may_umount_tree);
408 * may_umount - check if a mount point is busy
409 * @mnt: root of mount
411 * This is called to check if a mount point has any
412 * open files, pwds, chroots or sub mounts. If the
413 * mount has sub mounts this will return busy
414 * regardless of whether the sub mounts are busy.
416 * Doesn't take quota and stuff into account. IOW, in some cases it will
417 * give false negatives. The main reason why it's here is that we need
418 * a non-destructive way to look for easily umountable filesystems.
420 int may_umount(struct vfsmount *mnt)
423 spin_lock(&vfsmount_lock);
424 if (propagate_mount_busy(mnt, 2))
426 spin_unlock(&vfsmount_lock);
430 EXPORT_SYMBOL(may_umount);
432 void release_mounts(struct list_head *head)
434 struct vfsmount *mnt;
435 while(!list_empty(head)) {
436 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
437 list_del_init(&mnt->mnt_hash);
438 if (mnt->mnt_parent != mnt) {
439 struct dentry *dentry;
441 spin_lock(&vfsmount_lock);
442 dentry = mnt->mnt_mountpoint;
444 mnt->mnt_mountpoint = mnt->mnt_root;
445 mnt->mnt_parent = mnt;
446 spin_unlock(&vfsmount_lock);
454 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
458 for (p = mnt; p; p = next_mnt(p, mnt)) {
459 list_del(&p->mnt_hash);
460 list_add(&p->mnt_hash, kill);
464 propagate_umount(kill);
466 list_for_each_entry(p, kill, mnt_hash) {
467 list_del_init(&p->mnt_expire);
468 list_del_init(&p->mnt_list);
469 __touch_namespace(p->mnt_namespace);
470 p->mnt_namespace = NULL;
471 list_del_init(&p->mnt_child);
472 if (p->mnt_parent != p)
473 mnt->mnt_mountpoint->d_mounted--;
474 change_mnt_propagation(p, MS_PRIVATE);
478 static int do_umount(struct vfsmount *mnt, int flags)
480 struct super_block *sb = mnt->mnt_sb;
482 LIST_HEAD(umount_list);
484 retval = security_sb_umount(mnt, flags);
489 * Allow userspace to request a mountpoint be expired rather than
490 * unmounting unconditionally. Unmount only happens if:
491 * (1) the mark is already set (the mark is cleared by mntput())
492 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
494 if (flags & MNT_EXPIRE) {
495 if (mnt == current->fs->rootmnt ||
496 flags & (MNT_FORCE | MNT_DETACH))
499 if (atomic_read(&mnt->mnt_count) != 2)
502 if (!xchg(&mnt->mnt_expiry_mark, 1))
507 * If we may have to abort operations to get out of this
508 * mount, and they will themselves hold resources we must
509 * allow the fs to do things. In the Unix tradition of
510 * 'Gee thats tricky lets do it in userspace' the umount_begin
511 * might fail to complete on the first run through as other tasks
512 * must return, and the like. Thats for the mount program to worry
513 * about for the moment.
517 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
518 sb->s_op->umount_begin(sb);
522 * No sense to grab the lock for this test, but test itself looks
523 * somewhat bogus. Suggestions for better replacement?
524 * Ho-hum... In principle, we might treat that as umount + switch
525 * to rootfs. GC would eventually take care of the old vfsmount.
526 * Actually it makes sense, especially if rootfs would contain a
527 * /reboot - static binary that would close all descriptors and
528 * call reboot(9). Then init(8) could umount root and exec /reboot.
530 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
532 * Special case for "unmounting" root ...
533 * we just try to remount it readonly.
535 down_write(&sb->s_umount);
536 if (!(sb->s_flags & MS_RDONLY)) {
539 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
542 up_write(&sb->s_umount);
546 down_write(&namespace_sem);
547 spin_lock(&vfsmount_lock);
551 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
552 if (!list_empty(&mnt->mnt_list))
553 umount_tree(mnt, 1, &umount_list);
556 spin_unlock(&vfsmount_lock);
558 security_sb_umount_busy(mnt);
559 up_write(&namespace_sem);
560 release_mounts(&umount_list);
565 * Now umount can handle mount points as well as block devices.
566 * This is important for filesystems which use unnamed block devices.
568 * We now support a flag for forced unmount like the other 'big iron'
569 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
572 asmlinkage long sys_umount(char __user * name, int flags)
577 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
581 if (nd.dentry != nd.mnt->mnt_root)
583 if (!check_mnt(nd.mnt))
587 if (!capable(CAP_SYS_ADMIN))
590 retval = do_umount(nd.mnt, flags);
592 path_release_on_umount(&nd);
597 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
600 * The 2.0 compatible umount. No flags.
602 asmlinkage long sys_oldumount(char __user * name)
604 return sys_umount(name, 0);
609 static int mount_is_safe(struct nameidata *nd)
611 if (capable(CAP_SYS_ADMIN))
615 if (S_ISLNK(nd->dentry->d_inode->i_mode))
617 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
618 if (current->uid != nd->dentry->d_inode->i_uid)
621 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
627 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
632 if (d == NULL || d == d->d_parent)
638 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
641 struct vfsmount *res, *p, *q, *r, *s;
644 res = q = clone_mnt(mnt, dentry, flag);
647 q->mnt_mountpoint = mnt->mnt_mountpoint;
650 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
651 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
654 for (s = r; s; s = next_mnt(s, r)) {
655 while (p != s->mnt_parent) {
661 nd.dentry = p->mnt_mountpoint;
662 q = clone_mnt(p, p->mnt_root, flag);
665 spin_lock(&vfsmount_lock);
666 list_add_tail(&q->mnt_list, &res->mnt_list);
668 spin_unlock(&vfsmount_lock);
674 LIST_HEAD(umount_list);
675 spin_lock(&vfsmount_lock);
676 umount_tree(res, 0, &umount_list);
677 spin_unlock(&vfsmount_lock);
678 release_mounts(&umount_list);
684 * @source_mnt : mount tree to be attached
685 * @nd : place the mount tree @source_mnt is attached
686 * @parent_nd : if non-null, detach the source_mnt from its parent and
687 * store the parent mount and mountpoint dentry.
688 * (done when source_mnt is moved)
690 * NOTE: in the table below explains the semantics when a source mount
691 * of a given type is attached to a destination mount of a given type.
692 * ---------------------------------------------
693 * | BIND MOUNT OPERATION |
694 * |********************************************
695 * | source-->| shared | private |
699 * |********************************************
700 * | shared | shared (++) | shared (+) |
702 * |non-shared| shared (+) | private |
703 * *********************************************
704 * A bind operation clones the source mount and mounts the clone on the
707 * (++) the cloned mount is propagated to all the mounts in the propagation
708 * tree of the destination mount and the cloned mount is added to
709 * the peer group of the source mount.
710 * (+) the cloned mount is created under the destination mount and is marked
711 * as shared. The cloned mount is added to the peer group of the source
713 * ---------------------------------------------
714 * | MOVE MOUNT OPERATION |
715 * |********************************************
716 * | source-->| shared | private |
720 * |********************************************
721 * | shared | shared (+) | shared (+) |
723 * |non-shared| shared (+*) | private |
724 * *********************************************
725 * (+) the mount is moved to the destination. And is then propagated to all
726 * the mounts in the propagation tree of the destination mount.
727 * (+*) the mount is moved to the destination.
729 * if the source mount is a tree, the operations explained above is
730 * applied to each mount in the tree.
731 * Must be called without spinlocks held, since this function can sleep
734 static int attach_recursive_mnt(struct vfsmount *source_mnt,
735 struct nameidata *nd, struct nameidata *parent_nd)
737 LIST_HEAD(tree_list);
738 struct vfsmount *dest_mnt = nd->mnt;
739 struct dentry *dest_dentry = nd->dentry;
740 struct vfsmount *child, *p;
742 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
745 if (IS_MNT_SHARED(dest_mnt)) {
746 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
750 spin_lock(&vfsmount_lock);
752 detach_mnt(source_mnt, parent_nd);
753 attach_mnt(source_mnt, nd);
754 touch_namespace(current->namespace);
756 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
757 commit_tree(source_mnt);
760 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
761 list_del_init(&child->mnt_hash);
764 spin_unlock(&vfsmount_lock);
768 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
771 if (mnt->mnt_sb->s_flags & MS_NOUSER)
774 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
775 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
779 down(&nd->dentry->d_inode->i_sem);
780 if (IS_DEADDIR(nd->dentry->d_inode))
783 err = security_sb_check_sb(mnt, nd);
788 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
789 err = attach_recursive_mnt(mnt, nd, NULL);
791 up(&nd->dentry->d_inode->i_sem);
793 security_sb_post_addmount(mnt, nd);
798 * recursively change the type of the mountpoint.
800 static int do_change_type(struct nameidata *nd, int flag)
802 struct vfsmount *m, *mnt = nd->mnt;
803 int recurse = flag & MS_REC;
804 int type = flag & ~MS_REC;
806 if (nd->dentry != nd->mnt->mnt_root)
809 down_write(&namespace_sem);
810 spin_lock(&vfsmount_lock);
811 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
812 change_mnt_propagation(m, type);
813 spin_unlock(&vfsmount_lock);
814 up_write(&namespace_sem);
821 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
823 struct nameidata old_nd;
824 struct vfsmount *mnt = NULL;
825 int err = mount_is_safe(nd);
828 if (!old_name || !*old_name)
830 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
834 down_write(&namespace_sem);
836 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
841 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
843 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
848 err = graft_tree(mnt, nd);
850 LIST_HEAD(umount_list);
851 spin_lock(&vfsmount_lock);
852 umount_tree(mnt, 0, &umount_list);
853 spin_unlock(&vfsmount_lock);
854 release_mounts(&umount_list);
858 up_write(&namespace_sem);
859 path_release(&old_nd);
864 * change filesystem flags. dir should be a physical root of filesystem.
865 * If you've mounted a non-root directory somewhere and want to do remount
866 * on it - tough luck.
868 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
872 struct super_block *sb = nd->mnt->mnt_sb;
874 if (!capable(CAP_SYS_ADMIN))
877 if (!check_mnt(nd->mnt))
880 if (nd->dentry != nd->mnt->mnt_root)
883 down_write(&sb->s_umount);
884 err = do_remount_sb(sb, flags, data, 0);
886 nd->mnt->mnt_flags = mnt_flags;
887 up_write(&sb->s_umount);
889 security_sb_post_remount(nd->mnt, flags, data);
893 static int do_move_mount(struct nameidata *nd, char *old_name)
895 struct nameidata old_nd, parent_nd;
898 if (!capable(CAP_SYS_ADMIN))
900 if (!old_name || !*old_name)
902 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
906 down_write(&namespace_sem);
907 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
910 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
914 down(&nd->dentry->d_inode->i_sem);
915 if (IS_DEADDIR(nd->dentry->d_inode))
918 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
922 if (old_nd.dentry != old_nd.mnt->mnt_root)
925 if (old_nd.mnt == old_nd.mnt->mnt_parent)
928 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
929 S_ISDIR(old_nd.dentry->d_inode->i_mode))
932 * Don't move a mount residing in a shared parent.
934 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
937 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
941 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
944 spin_lock(&vfsmount_lock);
945 /* if the mount is moved, it should no longer be expire
947 list_del_init(&old_nd.mnt->mnt_expire);
948 spin_unlock(&vfsmount_lock);
950 up(&nd->dentry->d_inode->i_sem);
952 up_write(&namespace_sem);
954 path_release(&parent_nd);
955 path_release(&old_nd);
960 * create a new mount for userspace and request it to be added into the
963 static int do_new_mount(struct nameidata *nd, char *type, int flags,
964 int mnt_flags, char *name, void *data)
966 struct vfsmount *mnt;
968 if (!type || !memchr(type, 0, PAGE_SIZE))
971 /* we need capabilities... */
972 if (!capable(CAP_SYS_ADMIN))
975 mnt = do_kern_mount(type, flags, name, data);
979 return do_add_mount(mnt, nd, mnt_flags, NULL);
983 * add a mount into a namespace's mount tree
984 * - provide the option of adding the new mount to an expiration list
986 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
987 int mnt_flags, struct list_head *fslist)
991 down_write(&namespace_sem);
992 /* Something was mounted here while we slept */
993 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
996 if (!check_mnt(nd->mnt))
999 /* Refuse the same filesystem on the same mount point */
1001 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1002 nd->mnt->mnt_root == nd->dentry)
1006 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1009 newmnt->mnt_flags = mnt_flags;
1010 if ((err = graft_tree(newmnt, nd)))
1014 /* add to the specified expiration list */
1015 spin_lock(&vfsmount_lock);
1016 list_add_tail(&newmnt->mnt_expire, fslist);
1017 spin_unlock(&vfsmount_lock);
1019 up_write(&namespace_sem);
1023 up_write(&namespace_sem);
1028 EXPORT_SYMBOL_GPL(do_add_mount);
1030 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1031 struct list_head *umounts)
1033 spin_lock(&vfsmount_lock);
1036 * Check if mount is still attached, if not, let whoever holds it deal
1039 if (mnt->mnt_parent == mnt) {
1040 spin_unlock(&vfsmount_lock);
1045 * Check that it is still dead: the count should now be 2 - as
1046 * contributed by the vfsmount parent and the mntget above
1048 if (!propagate_mount_busy(mnt, 2)) {
1049 /* delete from the namespace */
1050 touch_namespace(mnt->mnt_namespace);
1051 list_del_init(&mnt->mnt_list);
1052 mnt->mnt_namespace = NULL;
1053 umount_tree(mnt, 1, umounts);
1054 spin_unlock(&vfsmount_lock);
1057 * Someone brought it back to life whilst we didn't have any
1058 * locks held so return it to the expiration list
1060 list_add_tail(&mnt->mnt_expire, mounts);
1061 spin_unlock(&vfsmount_lock);
1066 * process a list of expirable mountpoints with the intent of discarding any
1067 * mountpoints that aren't in use and haven't been touched since last we came
1070 void mark_mounts_for_expiry(struct list_head *mounts)
1072 struct namespace *namespace;
1073 struct vfsmount *mnt, *next;
1074 LIST_HEAD(graveyard);
1076 if (list_empty(mounts))
1079 spin_lock(&vfsmount_lock);
1081 /* extract from the expiration list every vfsmount that matches the
1082 * following criteria:
1083 * - only referenced by its parent vfsmount
1084 * - still marked for expiry (marked on the last call here; marks are
1085 * cleared by mntput())
1087 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1088 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1089 atomic_read(&mnt->mnt_count) != 1)
1093 list_move(&mnt->mnt_expire, &graveyard);
1097 * go through the vfsmounts we've just consigned to the graveyard to
1098 * - check that they're still dead
1099 * - delete the vfsmount from the appropriate namespace under lock
1100 * - dispose of the corpse
1102 while (!list_empty(&graveyard)) {
1104 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1105 list_del_init(&mnt->mnt_expire);
1107 /* don't do anything if the namespace is dead - all the
1108 * vfsmounts from it are going away anyway */
1109 namespace = mnt->mnt_namespace;
1110 if (!namespace || !namespace->root)
1112 get_namespace(namespace);
1114 spin_unlock(&vfsmount_lock);
1115 down_write(&namespace_sem);
1116 expire_mount(mnt, mounts, &umounts);
1117 up_write(&namespace_sem);
1118 release_mounts(&umounts);
1120 put_namespace(namespace);
1121 spin_lock(&vfsmount_lock);
1124 spin_unlock(&vfsmount_lock);
1127 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1130 * Some copy_from_user() implementations do not return the exact number of
1131 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1132 * Note that this function differs from copy_from_user() in that it will oops
1133 * on bad values of `to', rather than returning a short copy.
1135 static long exact_copy_from_user(void *to, const void __user * from,
1139 const char __user *f = from;
1142 if (!access_ok(VERIFY_READ, from, n))
1146 if (__get_user(c, f)) {
1157 int copy_mount_options(const void __user * data, unsigned long *where)
1167 if (!(page = __get_free_page(GFP_KERNEL)))
1170 /* We only care that *some* data at the address the user
1171 * gave us is valid. Just in case, we'll zero
1172 * the remainder of the page.
1174 /* copy_from_user cannot cross TASK_SIZE ! */
1175 size = TASK_SIZE - (unsigned long)data;
1176 if (size > PAGE_SIZE)
1179 i = size - exact_copy_from_user((void *)page, data, size);
1185 memset((char *)page + i, 0, PAGE_SIZE - i);
1191 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1192 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1194 * data is a (void *) that can point to any structure up to
1195 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1196 * information (or be NULL).
1198 * Pre-0.97 versions of mount() didn't have a flags word.
1199 * When the flags word was introduced its top half was required
1200 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1201 * Therefore, if this magic number is present, it carries no information
1202 * and must be discarded.
1204 long do_mount(char *dev_name, char *dir_name, char *type_page,
1205 unsigned long flags, void *data_page)
1207 struct nameidata nd;
1212 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1213 flags &= ~MS_MGC_MSK;
1215 /* Basic sanity checks */
1217 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1219 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1223 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1225 /* Separate the per-mountpoint flags */
1226 if (flags & MS_NOSUID)
1227 mnt_flags |= MNT_NOSUID;
1228 if (flags & MS_NODEV)
1229 mnt_flags |= MNT_NODEV;
1230 if (flags & MS_NOEXEC)
1231 mnt_flags |= MNT_NOEXEC;
1232 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE);
1234 /* ... and get the mountpoint */
1235 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1239 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1243 if (flags & MS_REMOUNT)
1244 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1246 else if (flags & MS_BIND)
1247 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1248 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE))
1249 retval = do_change_type(&nd, flags);
1250 else if (flags & MS_MOVE)
1251 retval = do_move_mount(&nd, dev_name);
1253 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1254 dev_name, data_page);
1260 int copy_namespace(int flags, struct task_struct *tsk)
1262 struct namespace *namespace = tsk->namespace;
1263 struct namespace *new_ns;
1264 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1265 struct fs_struct *fs = tsk->fs;
1266 struct vfsmount *p, *q;
1271 get_namespace(namespace);
1273 if (!(flags & CLONE_NEWNS))
1276 if (!capable(CAP_SYS_ADMIN)) {
1277 put_namespace(namespace);
1281 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1285 atomic_set(&new_ns->count, 1);
1286 INIT_LIST_HEAD(&new_ns->list);
1287 init_waitqueue_head(&new_ns->poll);
1290 down_write(&namespace_sem);
1291 /* First pass: copy the tree topology */
1292 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1294 if (!new_ns->root) {
1295 up_write(&namespace_sem);
1299 spin_lock(&vfsmount_lock);
1300 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1301 spin_unlock(&vfsmount_lock);
1304 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1305 * as belonging to new namespace. We have already acquired a private
1306 * fs_struct, so tsk->fs->lock is not needed.
1308 p = namespace->root;
1311 q->mnt_namespace = new_ns;
1313 if (p == fs->rootmnt) {
1315 fs->rootmnt = mntget(q);
1317 if (p == fs->pwdmnt) {
1319 fs->pwdmnt = mntget(q);
1321 if (p == fs->altrootmnt) {
1323 fs->altrootmnt = mntget(q);
1326 p = next_mnt(p, namespace->root);
1327 q = next_mnt(q, new_ns->root);
1329 up_write(&namespace_sem);
1331 tsk->namespace = new_ns;
1340 put_namespace(namespace);
1344 put_namespace(namespace);
1348 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1349 char __user * type, unsigned long flags,
1353 unsigned long data_page;
1354 unsigned long type_page;
1355 unsigned long dev_page;
1358 retval = copy_mount_options(type, &type_page);
1362 dir_page = getname(dir_name);
1363 retval = PTR_ERR(dir_page);
1364 if (IS_ERR(dir_page))
1367 retval = copy_mount_options(dev_name, &dev_page);
1371 retval = copy_mount_options(data, &data_page);
1376 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1377 flags, (void *)data_page);
1379 free_page(data_page);
1382 free_page(dev_page);
1386 free_page(type_page);
1391 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1392 * It can block. Requires the big lock held.
1394 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1395 struct dentry *dentry)
1397 struct dentry *old_root;
1398 struct vfsmount *old_rootmnt;
1399 write_lock(&fs->lock);
1400 old_root = fs->root;
1401 old_rootmnt = fs->rootmnt;
1402 fs->rootmnt = mntget(mnt);
1403 fs->root = dget(dentry);
1404 write_unlock(&fs->lock);
1407 mntput(old_rootmnt);
1412 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1413 * It can block. Requires the big lock held.
1415 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1416 struct dentry *dentry)
1418 struct dentry *old_pwd;
1419 struct vfsmount *old_pwdmnt;
1421 write_lock(&fs->lock);
1423 old_pwdmnt = fs->pwdmnt;
1424 fs->pwdmnt = mntget(mnt);
1425 fs->pwd = dget(dentry);
1426 write_unlock(&fs->lock);
1434 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1436 struct task_struct *g, *p;
1437 struct fs_struct *fs;
1439 read_lock(&tasklist_lock);
1440 do_each_thread(g, p) {
1444 atomic_inc(&fs->count);
1446 if (fs->root == old_nd->dentry
1447 && fs->rootmnt == old_nd->mnt)
1448 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1449 if (fs->pwd == old_nd->dentry
1450 && fs->pwdmnt == old_nd->mnt)
1451 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1455 } while_each_thread(g, p);
1456 read_unlock(&tasklist_lock);
1460 * pivot_root Semantics:
1461 * Moves the root file system of the current process to the directory put_old,
1462 * makes new_root as the new root file system of the current process, and sets
1463 * root/cwd of all processes which had them on the current root to new_root.
1466 * The new_root and put_old must be directories, and must not be on the
1467 * same file system as the current process root. The put_old must be
1468 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1469 * pointed to by put_old must yield the same directory as new_root. No other
1470 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1473 * - we don't move root/cwd if they are not at the root (reason: if something
1474 * cared enough to change them, it's probably wrong to force them elsewhere)
1475 * - it's okay to pick a root that isn't the root of a file system, e.g.
1476 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1477 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1480 asmlinkage long sys_pivot_root(const char __user * new_root,
1481 const char __user * put_old)
1483 struct vfsmount *tmp;
1484 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1487 if (!capable(CAP_SYS_ADMIN))
1492 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1497 if (!check_mnt(new_nd.mnt))
1500 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1504 error = security_sb_pivotroot(&old_nd, &new_nd);
1506 path_release(&old_nd);
1510 read_lock(¤t->fs->lock);
1511 user_nd.mnt = mntget(current->fs->rootmnt);
1512 user_nd.dentry = dget(current->fs->root);
1513 read_unlock(¤t->fs->lock);
1514 down_write(&namespace_sem);
1515 down(&old_nd.dentry->d_inode->i_sem);
1517 if (IS_MNT_SHARED(old_nd.mnt) ||
1518 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1519 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1521 if (!check_mnt(user_nd.mnt))
1524 if (IS_DEADDIR(new_nd.dentry->d_inode))
1526 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1528 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1531 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1532 goto out2; /* loop, on the same file system */
1534 if (user_nd.mnt->mnt_root != user_nd.dentry)
1535 goto out2; /* not a mountpoint */
1536 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1537 goto out2; /* not attached */
1538 if (new_nd.mnt->mnt_root != new_nd.dentry)
1539 goto out2; /* not a mountpoint */
1540 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1541 goto out2; /* not attached */
1542 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1543 spin_lock(&vfsmount_lock);
1544 if (tmp != new_nd.mnt) {
1546 if (tmp->mnt_parent == tmp)
1547 goto out3; /* already mounted on put_old */
1548 if (tmp->mnt_parent == new_nd.mnt)
1550 tmp = tmp->mnt_parent;
1552 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1554 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1556 detach_mnt(new_nd.mnt, &parent_nd);
1557 detach_mnt(user_nd.mnt, &root_parent);
1558 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1559 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1560 touch_namespace(current->namespace);
1561 spin_unlock(&vfsmount_lock);
1562 chroot_fs_refs(&user_nd, &new_nd);
1563 security_sb_post_pivotroot(&user_nd, &new_nd);
1565 path_release(&root_parent);
1566 path_release(&parent_nd);
1568 up(&old_nd.dentry->d_inode->i_sem);
1569 up_write(&namespace_sem);
1570 path_release(&user_nd);
1571 path_release(&old_nd);
1573 path_release(&new_nd);
1578 spin_unlock(&vfsmount_lock);
1582 static void __init init_mount_tree(void)
1584 struct vfsmount *mnt;
1585 struct namespace *namespace;
1586 struct task_struct *g, *p;
1588 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1590 panic("Can't create rootfs");
1591 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1593 panic("Can't allocate initial namespace");
1594 atomic_set(&namespace->count, 1);
1595 INIT_LIST_HEAD(&namespace->list);
1596 init_waitqueue_head(&namespace->poll);
1597 namespace->event = 0;
1598 list_add(&mnt->mnt_list, &namespace->list);
1599 namespace->root = mnt;
1600 mnt->mnt_namespace = namespace;
1602 init_task.namespace = namespace;
1603 read_lock(&tasklist_lock);
1604 do_each_thread(g, p) {
1605 get_namespace(namespace);
1606 p->namespace = namespace;
1607 } while_each_thread(g, p);
1608 read_unlock(&tasklist_lock);
1610 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1611 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1614 void __init mnt_init(unsigned long mempages)
1616 struct list_head *d;
1617 unsigned int nr_hash;
1620 init_rwsem(&namespace_sem);
1622 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1623 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1625 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1627 if (!mount_hashtable)
1628 panic("Failed to allocate mount hash table\n");
1631 * Find the power-of-two list-heads that can fit into the allocation..
1632 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1635 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1639 } while ((nr_hash >> hash_bits) != 0);
1643 * Re-calculate the actual number of entries and the mask
1644 * from the number of bits we can fit.
1646 nr_hash = 1UL << hash_bits;
1647 hash_mask = nr_hash - 1;
1649 printk("Mount-cache hash table entries: %d\n", nr_hash);
1651 /* And initialize the newly allocated array */
1652 d = mount_hashtable;
1664 void __put_namespace(struct namespace *namespace)
1666 struct vfsmount *root = namespace->root;
1667 LIST_HEAD(umount_list);
1668 namespace->root = NULL;
1669 spin_unlock(&vfsmount_lock);
1670 down_write(&namespace_sem);
1671 spin_lock(&vfsmount_lock);
1672 umount_tree(root, 0, &umount_list);
1673 spin_unlock(&vfsmount_lock);
1674 up_write(&namespace_sem);
1675 release_mounts(&umount_list);