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);
71 int size = strlen(name) + 1;
72 char *newname = kmalloc(size, GFP_KERNEL);
74 memcpy(newname, name, size);
75 mnt->mnt_devname = newname;
82 void free_vfsmnt(struct vfsmount *mnt)
84 kfree(mnt->mnt_devname);
85 kmem_cache_free(mnt_cache, mnt);
89 * Now, lookup_mnt increments the ref count before returning
90 * the vfsmount struct.
92 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
94 struct list_head *head = mount_hashtable + hash(mnt, dentry);
95 struct list_head *tmp = head;
96 struct vfsmount *p, *found = NULL;
98 spin_lock(&vfsmount_lock);
104 p = list_entry(tmp, struct vfsmount, mnt_hash);
105 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
110 spin_unlock(&vfsmount_lock);
114 static inline int check_mnt(struct vfsmount *mnt)
116 return mnt->mnt_namespace == current->namespace;
119 static void touch_namespace(struct namespace *ns)
123 wake_up_interruptible(&ns->poll);
127 static void __touch_namespace(struct namespace *ns)
129 if (ns && ns->event != event) {
131 wake_up_interruptible(&ns->poll);
135 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
137 old_nd->dentry = mnt->mnt_mountpoint;
138 old_nd->mnt = mnt->mnt_parent;
139 mnt->mnt_parent = mnt;
140 mnt->mnt_mountpoint = mnt->mnt_root;
141 list_del_init(&mnt->mnt_child);
142 list_del_init(&mnt->mnt_hash);
143 old_nd->dentry->d_mounted--;
146 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
147 struct vfsmount *child_mnt)
149 child_mnt->mnt_parent = mntget(mnt);
150 child_mnt->mnt_mountpoint = dget(dentry);
154 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
156 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
157 list_add_tail(&mnt->mnt_hash, mount_hashtable +
158 hash(nd->mnt, nd->dentry));
159 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
163 * the caller must hold vfsmount_lock
165 static void commit_tree(struct vfsmount *mnt)
167 struct vfsmount *parent = mnt->mnt_parent;
170 struct namespace *n = parent->mnt_namespace;
172 BUG_ON(parent == mnt);
174 list_add_tail(&head, &mnt->mnt_list);
175 list_for_each_entry(m, &head, mnt_list)
176 m->mnt_namespace = n;
177 list_splice(&head, n->list.prev);
179 list_add_tail(&mnt->mnt_hash, mount_hashtable +
180 hash(parent, mnt->mnt_mountpoint));
181 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
185 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
187 struct list_head *next = p->mnt_mounts.next;
188 if (next == &p->mnt_mounts) {
192 next = p->mnt_child.next;
193 if (next != &p->mnt_parent->mnt_mounts)
198 return list_entry(next, struct vfsmount, mnt_child);
201 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
204 struct super_block *sb = old->mnt_sb;
205 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
208 mnt->mnt_flags = old->mnt_flags;
209 atomic_inc(&sb->s_active);
211 mnt->mnt_root = dget(root);
212 mnt->mnt_mountpoint = mnt->mnt_root;
213 mnt->mnt_parent = mnt;
215 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
216 list_add(&mnt->mnt_share, &old->mnt_share);
217 if (flag & CL_MAKE_SHARED)
220 /* stick the duplicate mount on the same expiry list
221 * as the original if that was on one */
222 if (flag & CL_EXPIRE) {
223 spin_lock(&vfsmount_lock);
224 if (!list_empty(&old->mnt_expire))
225 list_add(&mnt->mnt_expire, &old->mnt_expire);
226 spin_unlock(&vfsmount_lock);
232 static inline void __mntput(struct vfsmount *mnt)
234 struct super_block *sb = mnt->mnt_sb;
237 deactivate_super(sb);
240 void mntput_no_expire(struct vfsmount *mnt)
243 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
244 if (likely(!mnt->mnt_pinned)) {
245 spin_unlock(&vfsmount_lock);
249 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
251 spin_unlock(&vfsmount_lock);
252 acct_auto_close_mnt(mnt);
253 security_sb_umount_close(mnt);
258 EXPORT_SYMBOL(mntput_no_expire);
260 void mnt_pin(struct vfsmount *mnt)
262 spin_lock(&vfsmount_lock);
264 spin_unlock(&vfsmount_lock);
267 EXPORT_SYMBOL(mnt_pin);
269 void mnt_unpin(struct vfsmount *mnt)
271 spin_lock(&vfsmount_lock);
272 if (mnt->mnt_pinned) {
273 atomic_inc(&mnt->mnt_count);
276 spin_unlock(&vfsmount_lock);
279 EXPORT_SYMBOL(mnt_unpin);
282 static void *m_start(struct seq_file *m, loff_t *pos)
284 struct namespace *n = m->private;
288 down_read(&namespace_sem);
289 list_for_each(p, &n->list)
291 return list_entry(p, struct vfsmount, mnt_list);
295 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
297 struct namespace *n = m->private;
298 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
300 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
303 static void m_stop(struct seq_file *m, void *v)
305 up_read(&namespace_sem);
308 static inline void mangle(struct seq_file *m, const char *s)
310 seq_escape(m, s, " \t\n\\");
313 static int show_vfsmnt(struct seq_file *m, void *v)
315 struct vfsmount *mnt = v;
317 static struct proc_fs_info {
321 { MS_SYNCHRONOUS, ",sync" },
322 { MS_DIRSYNC, ",dirsync" },
323 { MS_MANDLOCK, ",mand" },
324 { MS_NOATIME, ",noatime" },
325 { MS_NODIRATIME, ",nodiratime" },
328 static struct proc_fs_info mnt_info[] = {
329 { MNT_NOSUID, ",nosuid" },
330 { MNT_NODEV, ",nodev" },
331 { MNT_NOEXEC, ",noexec" },
334 struct proc_fs_info *fs_infop;
336 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
338 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
340 mangle(m, mnt->mnt_sb->s_type->name);
341 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
342 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
343 if (mnt->mnt_sb->s_flags & fs_infop->flag)
344 seq_puts(m, fs_infop->str);
346 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
347 if (mnt->mnt_flags & fs_infop->flag)
348 seq_puts(m, fs_infop->str);
350 if (mnt->mnt_sb->s_op->show_options)
351 err = mnt->mnt_sb->s_op->show_options(m, mnt);
352 seq_puts(m, " 0 0\n");
356 struct seq_operations mounts_op = {
364 * may_umount_tree - check if a mount tree is busy
365 * @mnt: root of mount tree
367 * This is called to check if a tree of mounts has any
368 * open files, pwds, chroots or sub mounts that are
371 int may_umount_tree(struct vfsmount *mnt)
374 int minimum_refs = 0;
377 spin_lock(&vfsmount_lock);
378 for (p = mnt; p; p = next_mnt(p, mnt)) {
379 actual_refs += atomic_read(&p->mnt_count);
382 spin_unlock(&vfsmount_lock);
384 if (actual_refs > minimum_refs)
390 EXPORT_SYMBOL(may_umount_tree);
393 * may_umount - check if a mount point is busy
394 * @mnt: root of mount
396 * This is called to check if a mount point has any
397 * open files, pwds, chroots or sub mounts. If the
398 * mount has sub mounts this will return busy
399 * regardless of whether the sub mounts are busy.
401 * Doesn't take quota and stuff into account. IOW, in some cases it will
402 * give false negatives. The main reason why it's here is that we need
403 * a non-destructive way to look for easily umountable filesystems.
405 int may_umount(struct vfsmount *mnt)
407 if (atomic_read(&mnt->mnt_count) > 2)
412 EXPORT_SYMBOL(may_umount);
414 void release_mounts(struct list_head *head)
416 struct vfsmount *mnt;
417 while(!list_empty(head)) {
418 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
419 list_del_init(&mnt->mnt_hash);
420 if (mnt->mnt_parent != mnt) {
421 struct dentry *dentry;
423 spin_lock(&vfsmount_lock);
424 dentry = mnt->mnt_mountpoint;
426 mnt->mnt_mountpoint = mnt->mnt_root;
427 mnt->mnt_parent = mnt;
428 spin_unlock(&vfsmount_lock);
436 void umount_tree(struct vfsmount *mnt, struct list_head *kill)
440 for (p = mnt; p; p = next_mnt(p, mnt)) {
441 list_del(&p->mnt_hash);
442 list_add(&p->mnt_hash, kill);
445 list_for_each_entry(p, kill, mnt_hash) {
446 list_del_init(&p->mnt_expire);
447 list_del_init(&p->mnt_list);
448 __touch_namespace(p->mnt_namespace);
449 p->mnt_namespace = NULL;
450 list_del_init(&p->mnt_child);
451 if (p->mnt_parent != p)
452 mnt->mnt_mountpoint->d_mounted--;
456 static int do_umount(struct vfsmount *mnt, int flags)
458 struct super_block *sb = mnt->mnt_sb;
460 LIST_HEAD(umount_list);
462 retval = security_sb_umount(mnt, flags);
467 * Allow userspace to request a mountpoint be expired rather than
468 * unmounting unconditionally. Unmount only happens if:
469 * (1) the mark is already set (the mark is cleared by mntput())
470 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
472 if (flags & MNT_EXPIRE) {
473 if (mnt == current->fs->rootmnt ||
474 flags & (MNT_FORCE | MNT_DETACH))
477 if (atomic_read(&mnt->mnt_count) != 2)
480 if (!xchg(&mnt->mnt_expiry_mark, 1))
485 * If we may have to abort operations to get out of this
486 * mount, and they will themselves hold resources we must
487 * allow the fs to do things. In the Unix tradition of
488 * 'Gee thats tricky lets do it in userspace' the umount_begin
489 * might fail to complete on the first run through as other tasks
490 * must return, and the like. Thats for the mount program to worry
491 * about for the moment.
495 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
496 sb->s_op->umount_begin(sb);
500 * No sense to grab the lock for this test, but test itself looks
501 * somewhat bogus. Suggestions for better replacement?
502 * Ho-hum... In principle, we might treat that as umount + switch
503 * to rootfs. GC would eventually take care of the old vfsmount.
504 * Actually it makes sense, especially if rootfs would contain a
505 * /reboot - static binary that would close all descriptors and
506 * call reboot(9). Then init(8) could umount root and exec /reboot.
508 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
510 * Special case for "unmounting" root ...
511 * we just try to remount it readonly.
513 down_write(&sb->s_umount);
514 if (!(sb->s_flags & MS_RDONLY)) {
517 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
520 up_write(&sb->s_umount);
524 down_write(&namespace_sem);
525 spin_lock(&vfsmount_lock);
529 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
530 if (!list_empty(&mnt->mnt_list))
531 umount_tree(mnt, &umount_list);
534 spin_unlock(&vfsmount_lock);
536 security_sb_umount_busy(mnt);
537 up_write(&namespace_sem);
538 release_mounts(&umount_list);
543 * Now umount can handle mount points as well as block devices.
544 * This is important for filesystems which use unnamed block devices.
546 * We now support a flag for forced unmount like the other 'big iron'
547 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
550 asmlinkage long sys_umount(char __user * name, int flags)
555 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
559 if (nd.dentry != nd.mnt->mnt_root)
561 if (!check_mnt(nd.mnt))
565 if (!capable(CAP_SYS_ADMIN))
568 retval = do_umount(nd.mnt, flags);
570 path_release_on_umount(&nd);
575 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
578 * The 2.0 compatible umount. No flags.
580 asmlinkage long sys_oldumount(char __user * name)
582 return sys_umount(name, 0);
587 static int mount_is_safe(struct nameidata *nd)
589 if (capable(CAP_SYS_ADMIN))
593 if (S_ISLNK(nd->dentry->d_inode->i_mode))
595 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
596 if (current->uid != nd->dentry->d_inode->i_uid)
599 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
605 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
610 if (d == NULL || d == d->d_parent)
616 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
619 struct vfsmount *res, *p, *q, *r, *s;
622 res = q = clone_mnt(mnt, dentry, flag);
625 q->mnt_mountpoint = mnt->mnt_mountpoint;
628 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
629 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
632 for (s = r; s; s = next_mnt(s, r)) {
633 while (p != s->mnt_parent) {
639 nd.dentry = p->mnt_mountpoint;
640 q = clone_mnt(p, p->mnt_root, flag);
643 spin_lock(&vfsmount_lock);
644 list_add_tail(&q->mnt_list, &res->mnt_list);
646 spin_unlock(&vfsmount_lock);
652 LIST_HEAD(umount_list);
653 spin_lock(&vfsmount_lock);
654 umount_tree(res, &umount_list);
655 spin_unlock(&vfsmount_lock);
656 release_mounts(&umount_list);
662 * @source_mnt : mount tree to be attached
663 * @nd : place the mount tree @source_mnt is attached
665 * NOTE: in the table below explains the semantics when a source mount
666 * of a given type is attached to a destination mount of a given type.
667 * ---------------------------------------------
668 * | BIND MOUNT OPERATION |
669 * |********************************************
670 * | source-->| shared | private |
674 * |********************************************
675 * | shared | shared (++) | shared (+) |
677 * |non-shared| shared (+) | private |
678 * *********************************************
679 * A bind operation clones the source mount and mounts the clone on the
682 * (++) the cloned mount is propagated to all the mounts in the propagation
683 * tree of the destination mount and the cloned mount is added to
684 * the peer group of the source mount.
685 * (+) the cloned mount is created under the destination mount and is marked
686 * as shared. The cloned mount is added to the peer group of the source
689 * if the source mount is a tree, the operations explained above is
690 * applied to each mount in the tree.
691 * Must be called without spinlocks held, since this function can sleep
694 static int attach_recursive_mnt(struct vfsmount *source_mnt,
695 struct nameidata *nd)
697 LIST_HEAD(tree_list);
698 struct vfsmount *dest_mnt = nd->mnt;
699 struct dentry *dest_dentry = nd->dentry;
700 struct vfsmount *child, *p;
702 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
705 if (IS_MNT_SHARED(dest_mnt)) {
706 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
710 spin_lock(&vfsmount_lock);
711 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
712 commit_tree(source_mnt);
714 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
715 list_del_init(&child->mnt_hash);
718 spin_unlock(&vfsmount_lock);
722 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
725 if (mnt->mnt_sb->s_flags & MS_NOUSER)
728 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
729 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
733 down(&nd->dentry->d_inode->i_sem);
734 if (IS_DEADDIR(nd->dentry->d_inode))
737 err = security_sb_check_sb(mnt, nd);
742 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
743 err = attach_recursive_mnt(mnt, nd);
745 up(&nd->dentry->d_inode->i_sem);
747 security_sb_post_addmount(mnt, nd);
752 * recursively change the type of the mountpoint.
754 static int do_change_type(struct nameidata *nd, int flag)
756 struct vfsmount *m, *mnt = nd->mnt;
757 int recurse = flag & MS_REC;
758 int type = flag & ~MS_REC;
760 if (nd->dentry != nd->mnt->mnt_root)
763 down_write(&namespace_sem);
764 spin_lock(&vfsmount_lock);
765 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
766 change_mnt_propagation(m, type);
767 spin_unlock(&vfsmount_lock);
768 up_write(&namespace_sem);
775 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
777 struct nameidata old_nd;
778 struct vfsmount *mnt = NULL;
779 int err = mount_is_safe(nd);
782 if (!old_name || !*old_name)
784 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
788 down_write(&namespace_sem);
790 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
795 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
797 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
802 err = graft_tree(mnt, nd);
804 LIST_HEAD(umount_list);
805 spin_lock(&vfsmount_lock);
806 umount_tree(mnt, &umount_list);
807 spin_unlock(&vfsmount_lock);
808 release_mounts(&umount_list);
812 up_write(&namespace_sem);
813 path_release(&old_nd);
818 * change filesystem flags. dir should be a physical root of filesystem.
819 * If you've mounted a non-root directory somewhere and want to do remount
820 * on it - tough luck.
822 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
826 struct super_block *sb = nd->mnt->mnt_sb;
828 if (!capable(CAP_SYS_ADMIN))
831 if (!check_mnt(nd->mnt))
834 if (nd->dentry != nd->mnt->mnt_root)
837 down_write(&sb->s_umount);
838 err = do_remount_sb(sb, flags, data, 0);
840 nd->mnt->mnt_flags = mnt_flags;
841 up_write(&sb->s_umount);
843 security_sb_post_remount(nd->mnt, flags, data);
847 static int do_move_mount(struct nameidata *nd, char *old_name)
849 struct nameidata old_nd, parent_nd;
852 if (!capable(CAP_SYS_ADMIN))
854 if (!old_name || !*old_name)
856 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
860 down_write(&namespace_sem);
861 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
864 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
868 down(&nd->dentry->d_inode->i_sem);
869 if (IS_DEADDIR(nd->dentry->d_inode))
872 spin_lock(&vfsmount_lock);
873 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
877 if (old_nd.dentry != old_nd.mnt->mnt_root)
880 if (old_nd.mnt == old_nd.mnt->mnt_parent)
883 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
884 S_ISDIR(old_nd.dentry->d_inode->i_mode))
888 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
893 detach_mnt(old_nd.mnt, &parent_nd);
894 attach_mnt(old_nd.mnt, nd);
895 touch_namespace(current->namespace);
897 /* if the mount is moved, it should no longer be expire
899 list_del_init(&old_nd.mnt->mnt_expire);
901 spin_unlock(&vfsmount_lock);
903 up(&nd->dentry->d_inode->i_sem);
905 up_write(&namespace_sem);
907 path_release(&parent_nd);
908 path_release(&old_nd);
913 * create a new mount for userspace and request it to be added into the
916 static int do_new_mount(struct nameidata *nd, char *type, int flags,
917 int mnt_flags, char *name, void *data)
919 struct vfsmount *mnt;
921 if (!type || !memchr(type, 0, PAGE_SIZE))
924 /* we need capabilities... */
925 if (!capable(CAP_SYS_ADMIN))
928 mnt = do_kern_mount(type, flags, name, data);
932 return do_add_mount(mnt, nd, mnt_flags, NULL);
936 * add a mount into a namespace's mount tree
937 * - provide the option of adding the new mount to an expiration list
939 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
940 int mnt_flags, struct list_head *fslist)
944 down_write(&namespace_sem);
945 /* Something was mounted here while we slept */
946 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
949 if (!check_mnt(nd->mnt))
952 /* Refuse the same filesystem on the same mount point */
954 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
955 nd->mnt->mnt_root == nd->dentry)
959 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
962 newmnt->mnt_flags = mnt_flags;
963 if ((err = graft_tree(newmnt, nd)))
967 /* add to the specified expiration list */
968 spin_lock(&vfsmount_lock);
969 list_add_tail(&newmnt->mnt_expire, fslist);
970 spin_unlock(&vfsmount_lock);
972 up_write(&namespace_sem);
976 up_write(&namespace_sem);
981 EXPORT_SYMBOL_GPL(do_add_mount);
983 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
984 struct list_head *umounts)
986 spin_lock(&vfsmount_lock);
989 * Check if mount is still attached, if not, let whoever holds it deal
992 if (mnt->mnt_parent == mnt) {
993 spin_unlock(&vfsmount_lock);
998 * Check that it is still dead: the count should now be 2 - as
999 * contributed by the vfsmount parent and the mntget above
1001 if (atomic_read(&mnt->mnt_count) == 2) {
1002 /* delete from the namespace */
1003 touch_namespace(mnt->mnt_namespace);
1004 list_del_init(&mnt->mnt_list);
1005 mnt->mnt_namespace = NULL;
1006 umount_tree(mnt, umounts);
1007 spin_unlock(&vfsmount_lock);
1010 * Someone brought it back to life whilst we didn't have any
1011 * locks held so return it to the expiration list
1013 list_add_tail(&mnt->mnt_expire, mounts);
1014 spin_unlock(&vfsmount_lock);
1019 * process a list of expirable mountpoints with the intent of discarding any
1020 * mountpoints that aren't in use and haven't been touched since last we came
1023 void mark_mounts_for_expiry(struct list_head *mounts)
1025 struct namespace *namespace;
1026 struct vfsmount *mnt, *next;
1027 LIST_HEAD(graveyard);
1029 if (list_empty(mounts))
1032 spin_lock(&vfsmount_lock);
1034 /* extract from the expiration list every vfsmount that matches the
1035 * following criteria:
1036 * - only referenced by its parent vfsmount
1037 * - still marked for expiry (marked on the last call here; marks are
1038 * cleared by mntput())
1040 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1041 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1042 atomic_read(&mnt->mnt_count) != 1)
1046 list_move(&mnt->mnt_expire, &graveyard);
1050 * go through the vfsmounts we've just consigned to the graveyard to
1051 * - check that they're still dead
1052 * - delete the vfsmount from the appropriate namespace under lock
1053 * - dispose of the corpse
1055 while (!list_empty(&graveyard)) {
1057 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1058 list_del_init(&mnt->mnt_expire);
1060 /* don't do anything if the namespace is dead - all the
1061 * vfsmounts from it are going away anyway */
1062 namespace = mnt->mnt_namespace;
1063 if (!namespace || !namespace->root)
1065 get_namespace(namespace);
1067 spin_unlock(&vfsmount_lock);
1068 down_write(&namespace_sem);
1069 expire_mount(mnt, mounts, &umounts);
1070 up_write(&namespace_sem);
1071 release_mounts(&umounts);
1073 put_namespace(namespace);
1074 spin_lock(&vfsmount_lock);
1077 spin_unlock(&vfsmount_lock);
1080 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1083 * Some copy_from_user() implementations do not return the exact number of
1084 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1085 * Note that this function differs from copy_from_user() in that it will oops
1086 * on bad values of `to', rather than returning a short copy.
1088 static long exact_copy_from_user(void *to, const void __user * from,
1092 const char __user *f = from;
1095 if (!access_ok(VERIFY_READ, from, n))
1099 if (__get_user(c, f)) {
1110 int copy_mount_options(const void __user * data, unsigned long *where)
1120 if (!(page = __get_free_page(GFP_KERNEL)))
1123 /* We only care that *some* data at the address the user
1124 * gave us is valid. Just in case, we'll zero
1125 * the remainder of the page.
1127 /* copy_from_user cannot cross TASK_SIZE ! */
1128 size = TASK_SIZE - (unsigned long)data;
1129 if (size > PAGE_SIZE)
1132 i = size - exact_copy_from_user((void *)page, data, size);
1138 memset((char *)page + i, 0, PAGE_SIZE - i);
1144 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1145 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1147 * data is a (void *) that can point to any structure up to
1148 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1149 * information (or be NULL).
1151 * Pre-0.97 versions of mount() didn't have a flags word.
1152 * When the flags word was introduced its top half was required
1153 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1154 * Therefore, if this magic number is present, it carries no information
1155 * and must be discarded.
1157 long do_mount(char *dev_name, char *dir_name, char *type_page,
1158 unsigned long flags, void *data_page)
1160 struct nameidata nd;
1165 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1166 flags &= ~MS_MGC_MSK;
1168 /* Basic sanity checks */
1170 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1172 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1176 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1178 /* Separate the per-mountpoint flags */
1179 if (flags & MS_NOSUID)
1180 mnt_flags |= MNT_NOSUID;
1181 if (flags & MS_NODEV)
1182 mnt_flags |= MNT_NODEV;
1183 if (flags & MS_NOEXEC)
1184 mnt_flags |= MNT_NOEXEC;
1185 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE);
1187 /* ... and get the mountpoint */
1188 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1192 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1196 if (flags & MS_REMOUNT)
1197 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1199 else if (flags & MS_BIND)
1200 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1201 else if (flags & (MS_SHARED | MS_PRIVATE))
1202 retval = do_change_type(&nd, flags);
1203 else if (flags & MS_MOVE)
1204 retval = do_move_mount(&nd, dev_name);
1206 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1207 dev_name, data_page);
1213 int copy_namespace(int flags, struct task_struct *tsk)
1215 struct namespace *namespace = tsk->namespace;
1216 struct namespace *new_ns;
1217 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1218 struct fs_struct *fs = tsk->fs;
1219 struct vfsmount *p, *q;
1224 get_namespace(namespace);
1226 if (!(flags & CLONE_NEWNS))
1229 if (!capable(CAP_SYS_ADMIN)) {
1230 put_namespace(namespace);
1234 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1238 atomic_set(&new_ns->count, 1);
1239 INIT_LIST_HEAD(&new_ns->list);
1240 init_waitqueue_head(&new_ns->poll);
1243 down_write(&namespace_sem);
1244 /* First pass: copy the tree topology */
1245 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1247 if (!new_ns->root) {
1248 up_write(&namespace_sem);
1252 spin_lock(&vfsmount_lock);
1253 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1254 spin_unlock(&vfsmount_lock);
1257 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1258 * as belonging to new namespace. We have already acquired a private
1259 * fs_struct, so tsk->fs->lock is not needed.
1261 p = namespace->root;
1264 q->mnt_namespace = new_ns;
1266 if (p == fs->rootmnt) {
1268 fs->rootmnt = mntget(q);
1270 if (p == fs->pwdmnt) {
1272 fs->pwdmnt = mntget(q);
1274 if (p == fs->altrootmnt) {
1276 fs->altrootmnt = mntget(q);
1279 p = next_mnt(p, namespace->root);
1280 q = next_mnt(q, new_ns->root);
1282 up_write(&namespace_sem);
1284 tsk->namespace = new_ns;
1293 put_namespace(namespace);
1297 put_namespace(namespace);
1301 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1302 char __user * type, unsigned long flags,
1306 unsigned long data_page;
1307 unsigned long type_page;
1308 unsigned long dev_page;
1311 retval = copy_mount_options(type, &type_page);
1315 dir_page = getname(dir_name);
1316 retval = PTR_ERR(dir_page);
1317 if (IS_ERR(dir_page))
1320 retval = copy_mount_options(dev_name, &dev_page);
1324 retval = copy_mount_options(data, &data_page);
1329 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1330 flags, (void *)data_page);
1332 free_page(data_page);
1335 free_page(dev_page);
1339 free_page(type_page);
1344 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1345 * It can block. Requires the big lock held.
1347 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1348 struct dentry *dentry)
1350 struct dentry *old_root;
1351 struct vfsmount *old_rootmnt;
1352 write_lock(&fs->lock);
1353 old_root = fs->root;
1354 old_rootmnt = fs->rootmnt;
1355 fs->rootmnt = mntget(mnt);
1356 fs->root = dget(dentry);
1357 write_unlock(&fs->lock);
1360 mntput(old_rootmnt);
1365 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1366 * It can block. Requires the big lock held.
1368 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1369 struct dentry *dentry)
1371 struct dentry *old_pwd;
1372 struct vfsmount *old_pwdmnt;
1374 write_lock(&fs->lock);
1376 old_pwdmnt = fs->pwdmnt;
1377 fs->pwdmnt = mntget(mnt);
1378 fs->pwd = dget(dentry);
1379 write_unlock(&fs->lock);
1387 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1389 struct task_struct *g, *p;
1390 struct fs_struct *fs;
1392 read_lock(&tasklist_lock);
1393 do_each_thread(g, p) {
1397 atomic_inc(&fs->count);
1399 if (fs->root == old_nd->dentry
1400 && fs->rootmnt == old_nd->mnt)
1401 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1402 if (fs->pwd == old_nd->dentry
1403 && fs->pwdmnt == old_nd->mnt)
1404 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1408 } while_each_thread(g, p);
1409 read_unlock(&tasklist_lock);
1413 * pivot_root Semantics:
1414 * Moves the root file system of the current process to the directory put_old,
1415 * makes new_root as the new root file system of the current process, and sets
1416 * root/cwd of all processes which had them on the current root to new_root.
1419 * The new_root and put_old must be directories, and must not be on the
1420 * same file system as the current process root. The put_old must be
1421 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1422 * pointed to by put_old must yield the same directory as new_root. No other
1423 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1426 * - we don't move root/cwd if they are not at the root (reason: if something
1427 * cared enough to change them, it's probably wrong to force them elsewhere)
1428 * - it's okay to pick a root that isn't the root of a file system, e.g.
1429 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1430 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1433 asmlinkage long sys_pivot_root(const char __user * new_root,
1434 const char __user * put_old)
1436 struct vfsmount *tmp;
1437 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1440 if (!capable(CAP_SYS_ADMIN))
1445 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1450 if (!check_mnt(new_nd.mnt))
1453 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1457 error = security_sb_pivotroot(&old_nd, &new_nd);
1459 path_release(&old_nd);
1463 read_lock(¤t->fs->lock);
1464 user_nd.mnt = mntget(current->fs->rootmnt);
1465 user_nd.dentry = dget(current->fs->root);
1466 read_unlock(¤t->fs->lock);
1467 down_write(&namespace_sem);
1468 down(&old_nd.dentry->d_inode->i_sem);
1470 if (!check_mnt(user_nd.mnt))
1473 if (IS_DEADDIR(new_nd.dentry->d_inode))
1475 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1477 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1480 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1481 goto out2; /* loop, on the same file system */
1483 if (user_nd.mnt->mnt_root != user_nd.dentry)
1484 goto out2; /* not a mountpoint */
1485 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1486 goto out2; /* not attached */
1487 if (new_nd.mnt->mnt_root != new_nd.dentry)
1488 goto out2; /* not a mountpoint */
1489 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1490 goto out2; /* not attached */
1491 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1492 spin_lock(&vfsmount_lock);
1493 if (tmp != new_nd.mnt) {
1495 if (tmp->mnt_parent == tmp)
1496 goto out3; /* already mounted on put_old */
1497 if (tmp->mnt_parent == new_nd.mnt)
1499 tmp = tmp->mnt_parent;
1501 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1503 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1505 detach_mnt(new_nd.mnt, &parent_nd);
1506 detach_mnt(user_nd.mnt, &root_parent);
1507 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1508 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1509 touch_namespace(current->namespace);
1510 spin_unlock(&vfsmount_lock);
1511 chroot_fs_refs(&user_nd, &new_nd);
1512 security_sb_post_pivotroot(&user_nd, &new_nd);
1514 path_release(&root_parent);
1515 path_release(&parent_nd);
1517 up(&old_nd.dentry->d_inode->i_sem);
1518 up_write(&namespace_sem);
1519 path_release(&user_nd);
1520 path_release(&old_nd);
1522 path_release(&new_nd);
1527 spin_unlock(&vfsmount_lock);
1531 static void __init init_mount_tree(void)
1533 struct vfsmount *mnt;
1534 struct namespace *namespace;
1535 struct task_struct *g, *p;
1537 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1539 panic("Can't create rootfs");
1540 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1542 panic("Can't allocate initial namespace");
1543 atomic_set(&namespace->count, 1);
1544 INIT_LIST_HEAD(&namespace->list);
1545 init_waitqueue_head(&namespace->poll);
1546 namespace->event = 0;
1547 list_add(&mnt->mnt_list, &namespace->list);
1548 namespace->root = mnt;
1549 mnt->mnt_namespace = namespace;
1551 init_task.namespace = namespace;
1552 read_lock(&tasklist_lock);
1553 do_each_thread(g, p) {
1554 get_namespace(namespace);
1555 p->namespace = namespace;
1556 } while_each_thread(g, p);
1557 read_unlock(&tasklist_lock);
1559 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1560 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1563 void __init mnt_init(unsigned long mempages)
1565 struct list_head *d;
1566 unsigned int nr_hash;
1569 init_rwsem(&namespace_sem);
1571 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1572 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1574 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1576 if (!mount_hashtable)
1577 panic("Failed to allocate mount hash table\n");
1580 * Find the power-of-two list-heads that can fit into the allocation..
1581 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1584 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1588 } while ((nr_hash >> hash_bits) != 0);
1592 * Re-calculate the actual number of entries and the mask
1593 * from the number of bits we can fit.
1595 nr_hash = 1UL << hash_bits;
1596 hash_mask = nr_hash - 1;
1598 printk("Mount-cache hash table entries: %d\n", nr_hash);
1600 /* And initialize the newly allocated array */
1601 d = mount_hashtable;
1613 void __put_namespace(struct namespace *namespace)
1615 struct vfsmount *root = namespace->root;
1616 LIST_HEAD(umount_list);
1617 namespace->root = NULL;
1618 spin_unlock(&vfsmount_lock);
1619 down_write(&namespace_sem);
1620 spin_lock(&vfsmount_lock);
1621 umount_tree(root, &umount_list);
1622 spin_unlock(&vfsmount_lock);
1623 up_write(&namespace_sem);
1624 release_mounts(&umount_list);