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/capability.h>
20 #include <linux/module.h>
21 #include <linux/seq_file.h>
22 #include <linux/namespace.h>
23 #include <linux/namei.h>
24 #include <linux/security.h>
25 #include <linux/mount.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 void free_vfsmnt(struct vfsmount *mnt)
93 kfree(mnt->mnt_devname);
94 kmem_cache_free(mnt_cache, mnt);
98 * find the first or last mount at @dentry on vfsmount @mnt depending on
99 * @dir. If @dir is set return the first mount else return the last mount.
101 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
104 struct list_head *head = mount_hashtable + hash(mnt, dentry);
105 struct list_head *tmp = head;
106 struct vfsmount *p, *found = NULL;
109 tmp = dir ? tmp->next : tmp->prev;
113 p = list_entry(tmp, struct vfsmount, mnt_hash);
114 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
123 * lookup_mnt increments the ref count before returning
124 * the vfsmount struct.
126 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
128 struct vfsmount *child_mnt;
129 spin_lock(&vfsmount_lock);
130 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 spin_unlock(&vfsmount_lock);
136 static inline int check_mnt(struct vfsmount *mnt)
138 return mnt->mnt_namespace == current->namespace;
141 static void touch_namespace(struct namespace *ns)
145 wake_up_interruptible(&ns->poll);
149 static void __touch_namespace(struct namespace *ns)
151 if (ns && ns->event != event) {
153 wake_up_interruptible(&ns->poll);
157 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
159 old_nd->dentry = mnt->mnt_mountpoint;
160 old_nd->mnt = mnt->mnt_parent;
161 mnt->mnt_parent = mnt;
162 mnt->mnt_mountpoint = mnt->mnt_root;
163 list_del_init(&mnt->mnt_child);
164 list_del_init(&mnt->mnt_hash);
165 old_nd->dentry->d_mounted--;
168 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
169 struct vfsmount *child_mnt)
171 child_mnt->mnt_parent = mntget(mnt);
172 child_mnt->mnt_mountpoint = dget(dentry);
176 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
178 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
179 list_add_tail(&mnt->mnt_hash, mount_hashtable +
180 hash(nd->mnt, nd->dentry));
181 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
185 * the caller must hold vfsmount_lock
187 static void commit_tree(struct vfsmount *mnt)
189 struct vfsmount *parent = mnt->mnt_parent;
192 struct namespace *n = parent->mnt_namespace;
194 BUG_ON(parent == mnt);
196 list_add_tail(&head, &mnt->mnt_list);
197 list_for_each_entry(m, &head, mnt_list)
198 m->mnt_namespace = n;
199 list_splice(&head, n->list.prev);
201 list_add_tail(&mnt->mnt_hash, mount_hashtable +
202 hash(parent, mnt->mnt_mountpoint));
203 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
207 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
209 struct list_head *next = p->mnt_mounts.next;
210 if (next == &p->mnt_mounts) {
214 next = p->mnt_child.next;
215 if (next != &p->mnt_parent->mnt_mounts)
220 return list_entry(next, struct vfsmount, mnt_child);
223 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
225 struct list_head *prev = p->mnt_mounts.prev;
226 while (prev != &p->mnt_mounts) {
227 p = list_entry(prev, struct vfsmount, mnt_child);
228 prev = p->mnt_mounts.prev;
233 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
236 struct super_block *sb = old->mnt_sb;
237 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
240 mnt->mnt_flags = old->mnt_flags;
241 atomic_inc(&sb->s_active);
243 mnt->mnt_root = dget(root);
244 mnt->mnt_mountpoint = mnt->mnt_root;
245 mnt->mnt_parent = mnt;
246 mnt->mnt_xid = old->mnt_xid;
248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
259 if (flag & CL_MAKE_SHARED)
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
274 static inline void __mntput(struct vfsmount *mnt)
276 struct super_block *sb = mnt->mnt_sb;
279 deactivate_super(sb);
282 void mntput_no_expire(struct vfsmount *mnt)
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
300 EXPORT_SYMBOL(mntput_no_expire);
302 void mnt_pin(struct vfsmount *mnt)
304 spin_lock(&vfsmount_lock);
306 spin_unlock(&vfsmount_lock);
309 EXPORT_SYMBOL(mnt_pin);
311 void mnt_unpin(struct vfsmount *mnt)
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
318 spin_unlock(&vfsmount_lock);
321 EXPORT_SYMBOL(mnt_unpin);
324 static void *m_start(struct seq_file *m, loff_t *pos)
326 struct namespace *n = m->private;
330 down_read(&namespace_sem);
331 list_for_each(p, &n->list)
333 return list_entry(p, struct vfsmount, mnt_list);
337 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
339 struct namespace *n = m->private;
340 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
342 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
345 static void m_stop(struct seq_file *m, void *v)
347 up_read(&namespace_sem);
350 static inline void mangle(struct seq_file *m, const char *s)
352 seq_escape(m, s, " \t\n\\");
355 static int mnt_is_reachable(struct vfsmount *mnt)
357 struct vfsmount *root_mnt;
358 struct dentry *root, *point;
361 if (mnt == mnt->mnt_namespace->root)
364 spin_lock(&dcache_lock);
365 root_mnt = current->fs->rootmnt;
366 root = current->fs->root;
369 while ((mnt != mnt->mnt_parent) && (mnt != root_mnt)) {
370 point = mnt->mnt_mountpoint;
371 mnt = mnt->mnt_parent;
374 ret = (mnt == root_mnt) && is_subdir(point, root);
376 spin_unlock(&dcache_lock);
381 static int show_vfsmnt(struct seq_file *m, void *v)
383 struct vfsmount *mnt = v;
385 static struct proc_fs_info {
391 { MS_RDONLY, MNT_RDONLY, "ro", "rw" },
392 { MS_SYNCHRONOUS, 0, ",sync", NULL },
393 { MS_DIRSYNC, 0, ",dirsync", NULL },
394 { MS_MANDLOCK, 0, ",mand", NULL },
395 { MS_TAGXID, 0, ",tagxid", NULL },
396 { MS_NOATIME, MNT_NOATIME, ",noatime", NULL },
397 { MS_NODIRATIME, MNT_NODIRATIME, ",nodiratime", NULL },
398 { 0, MNT_NOSUID, ",nosuid", NULL },
399 { 0, MNT_NODEV, ",nodev", NULL },
400 { 0, MNT_NOEXEC, ",noexec", NULL },
403 struct proc_fs_info *p;
404 unsigned long s_flags = mnt->mnt_sb->s_flags;
405 int mnt_flags = mnt->mnt_flags;
407 if (vx_flags(VXF_HIDE_MOUNT, 0))
409 if (!mnt_is_reachable(mnt))
412 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
413 mnt == current->fs->rootmnt) {
414 seq_puts(m, "/dev/root / ");
416 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
418 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
421 mangle(m, mnt->mnt_sb->s_type->name);
423 for (p = fs_info; (p->s_flag | p->mnt_flag) ; p++) {
424 if ((s_flags & p->s_flag) || (mnt_flags & p->mnt_flag)) {
426 seq_puts(m, p->set_str);
429 seq_puts(m, p->unset_str);
432 if (mnt->mnt_flags & MNT_XID)
433 seq_printf(m, ",xid=%d", mnt->mnt_xid);
434 if (mnt->mnt_sb->s_op->show_options)
435 err = mnt->mnt_sb->s_op->show_options(m, mnt);
436 seq_puts(m, " 0 0\n");
440 struct seq_operations mounts_op = {
447 static int show_vfsstat(struct seq_file *m, void *v)
449 struct vfsmount *mnt = v;
452 if (vx_flags(VXF_HIDE_MOUNT, 0))
454 if (!mnt_is_reachable(mnt) && !vx_check(0, VX_WATCH))
457 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
458 mnt == current->fs->rootmnt) {
459 seq_puts(m, "device /dev/root mounted on / ");
462 if (mnt->mnt_devname) {
463 seq_puts(m, "device ");
464 mangle(m, mnt->mnt_devname);
466 seq_puts(m, "no device");
469 seq_puts(m, " mounted on ");
470 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
474 /* file system type */
475 seq_puts(m, "with fstype ");
476 mangle(m, mnt->mnt_sb->s_type->name);
478 /* optional statistics */
479 if (mnt->mnt_sb->s_op->show_stats) {
481 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
488 struct seq_operations mountstats_op = {
492 .show = show_vfsstat,
496 * may_umount_tree - check if a mount tree is busy
497 * @mnt: root of mount tree
499 * This is called to check if a tree of mounts has any
500 * open files, pwds, chroots or sub mounts that are
503 int may_umount_tree(struct vfsmount *mnt)
506 int minimum_refs = 0;
509 spin_lock(&vfsmount_lock);
510 for (p = mnt; p; p = next_mnt(p, mnt)) {
511 actual_refs += atomic_read(&p->mnt_count);
514 spin_unlock(&vfsmount_lock);
516 if (actual_refs > minimum_refs)
522 EXPORT_SYMBOL(may_umount_tree);
525 * may_umount - check if a mount point is busy
526 * @mnt: root of mount
528 * This is called to check if a mount point has any
529 * open files, pwds, chroots or sub mounts. If the
530 * mount has sub mounts this will return busy
531 * regardless of whether the sub mounts are busy.
533 * Doesn't take quota and stuff into account. IOW, in some cases it will
534 * give false negatives. The main reason why it's here is that we need
535 * a non-destructive way to look for easily umountable filesystems.
537 int may_umount(struct vfsmount *mnt)
540 spin_lock(&vfsmount_lock);
541 if (propagate_mount_busy(mnt, 2))
543 spin_unlock(&vfsmount_lock);
547 EXPORT_SYMBOL(may_umount);
549 void release_mounts(struct list_head *head)
551 struct vfsmount *mnt;
552 while (!list_empty(head)) {
553 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
554 list_del_init(&mnt->mnt_hash);
555 if (mnt->mnt_parent != mnt) {
556 struct dentry *dentry;
558 spin_lock(&vfsmount_lock);
559 dentry = mnt->mnt_mountpoint;
561 mnt->mnt_mountpoint = mnt->mnt_root;
562 mnt->mnt_parent = mnt;
563 spin_unlock(&vfsmount_lock);
571 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
575 for (p = mnt; p; p = next_mnt(p, mnt)) {
576 list_del(&p->mnt_hash);
577 list_add(&p->mnt_hash, kill);
581 propagate_umount(kill);
583 list_for_each_entry(p, kill, mnt_hash) {
584 list_del_init(&p->mnt_expire);
585 list_del_init(&p->mnt_list);
586 __touch_namespace(p->mnt_namespace);
587 p->mnt_namespace = NULL;
588 list_del_init(&p->mnt_child);
589 if (p->mnt_parent != p)
590 p->mnt_mountpoint->d_mounted--;
591 change_mnt_propagation(p, MS_PRIVATE);
595 static int do_umount(struct vfsmount *mnt, int flags)
597 struct super_block *sb = mnt->mnt_sb;
599 LIST_HEAD(umount_list);
601 retval = security_sb_umount(mnt, flags);
606 * Allow userspace to request a mountpoint be expired rather than
607 * unmounting unconditionally. Unmount only happens if:
608 * (1) the mark is already set (the mark is cleared by mntput())
609 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
611 if (flags & MNT_EXPIRE) {
612 if (mnt == current->fs->rootmnt ||
613 flags & (MNT_FORCE | MNT_DETACH))
616 if (atomic_read(&mnt->mnt_count) != 2)
619 if (!xchg(&mnt->mnt_expiry_mark, 1))
624 * If we may have to abort operations to get out of this
625 * mount, and they will themselves hold resources we must
626 * allow the fs to do things. In the Unix tradition of
627 * 'Gee thats tricky lets do it in userspace' the umount_begin
628 * might fail to complete on the first run through as other tasks
629 * must return, and the like. Thats for the mount program to worry
630 * about for the moment.
634 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
635 sb->s_op->umount_begin(sb);
639 * No sense to grab the lock for this test, but test itself looks
640 * somewhat bogus. Suggestions for better replacement?
641 * Ho-hum... In principle, we might treat that as umount + switch
642 * to rootfs. GC would eventually take care of the old vfsmount.
643 * Actually it makes sense, especially if rootfs would contain a
644 * /reboot - static binary that would close all descriptors and
645 * call reboot(9). Then init(8) could umount root and exec /reboot.
647 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
649 * Special case for "unmounting" root ...
650 * we just try to remount it readonly.
652 down_write(&sb->s_umount);
653 if (!(sb->s_flags & MS_RDONLY)) {
656 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
659 up_write(&sb->s_umount);
663 down_write(&namespace_sem);
664 spin_lock(&vfsmount_lock);
668 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
669 if (!list_empty(&mnt->mnt_list))
670 umount_tree(mnt, 1, &umount_list);
673 spin_unlock(&vfsmount_lock);
675 security_sb_umount_busy(mnt);
676 up_write(&namespace_sem);
677 release_mounts(&umount_list);
682 * Now umount can handle mount points as well as block devices.
683 * This is important for filesystems which use unnamed block devices.
685 * We now support a flag for forced unmount like the other 'big iron'
686 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
689 asmlinkage long sys_umount(char __user * name, int flags)
694 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
698 if (nd.dentry != nd.mnt->mnt_root)
700 if (!check_mnt(nd.mnt))
704 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
707 retval = do_umount(nd.mnt, flags);
709 path_release_on_umount(&nd);
714 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
717 * The 2.0 compatible umount. No flags.
719 asmlinkage long sys_oldumount(char __user * name)
721 return sys_umount(name, 0);
726 static int mount_is_safe(struct nameidata *nd)
728 if (vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
732 if (S_ISLNK(nd->dentry->d_inode->i_mode))
734 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
735 if (current->uid != nd->dentry->d_inode->i_uid)
738 if (vfs_permission(nd, MAY_WRITE))
744 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
749 if (d == NULL || d == d->d_parent)
755 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
758 struct vfsmount *res, *p, *q, *r, *s;
761 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
764 res = q = clone_mnt(mnt, dentry, flag);
767 q->mnt_mountpoint = mnt->mnt_mountpoint;
770 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
771 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
774 for (s = r; s; s = next_mnt(s, r)) {
775 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
776 s = skip_mnt_tree(s);
779 while (p != s->mnt_parent) {
785 nd.dentry = p->mnt_mountpoint;
786 q = clone_mnt(p, p->mnt_root, flag);
789 spin_lock(&vfsmount_lock);
790 list_add_tail(&q->mnt_list, &res->mnt_list);
792 spin_unlock(&vfsmount_lock);
798 LIST_HEAD(umount_list);
799 spin_lock(&vfsmount_lock);
800 umount_tree(res, 0, &umount_list);
801 spin_unlock(&vfsmount_lock);
802 release_mounts(&umount_list);
808 * @source_mnt : mount tree to be attached
809 * @nd : place the mount tree @source_mnt is attached
810 * @parent_nd : if non-null, detach the source_mnt from its parent and
811 * store the parent mount and mountpoint dentry.
812 * (done when source_mnt is moved)
814 * NOTE: in the table below explains the semantics when a source mount
815 * of a given type is attached to a destination mount of a given type.
816 * ---------------------------------------------------------------------------
817 * | BIND MOUNT OPERATION |
818 * |**************************************************************************
819 * | source-->| shared | private | slave | unbindable |
823 * |**************************************************************************
824 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
826 * |non-shared| shared (+) | private | slave (*) | invalid |
827 * ***************************************************************************
828 * A bind operation clones the source mount and mounts the clone on the
831 * (++) the cloned mount is propagated to all the mounts in the propagation
832 * tree of the destination mount and the cloned mount is added to
833 * the peer group of the source mount.
834 * (+) the cloned mount is created under the destination mount and is marked
835 * as shared. The cloned mount is added to the peer group of the source
837 * (+++) the mount is propagated to all the mounts in the propagation tree
838 * of the destination mount and the cloned mount is made slave
839 * of the same master as that of the source mount. The cloned mount
840 * is marked as 'shared and slave'.
841 * (*) the cloned mount is made a slave of the same master as that of the
844 * ---------------------------------------------------------------------------
845 * | MOVE MOUNT OPERATION |
846 * |**************************************************************************
847 * | source-->| shared | private | slave | unbindable |
851 * |**************************************************************************
852 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
854 * |non-shared| shared (+*) | private | slave (*) | unbindable |
855 * ***************************************************************************
857 * (+) the mount is moved to the destination. And is then propagated to
858 * all the mounts in the propagation tree of the destination mount.
859 * (+*) the mount is moved to the destination.
860 * (+++) the mount is moved to the destination and is then propagated to
861 * all the mounts belonging to the destination mount's propagation tree.
862 * the mount is marked as 'shared and slave'.
863 * (*) the mount continues to be a slave at the new location.
865 * if the source mount is a tree, the operations explained above is
866 * applied to each mount in the tree.
867 * Must be called without spinlocks held, since this function can sleep
870 static int attach_recursive_mnt(struct vfsmount *source_mnt,
871 struct nameidata *nd, struct nameidata *parent_nd)
873 LIST_HEAD(tree_list);
874 struct vfsmount *dest_mnt = nd->mnt;
875 struct dentry *dest_dentry = nd->dentry;
876 struct vfsmount *child, *p;
878 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
881 if (IS_MNT_SHARED(dest_mnt)) {
882 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
886 spin_lock(&vfsmount_lock);
888 detach_mnt(source_mnt, parent_nd);
889 attach_mnt(source_mnt, nd);
890 touch_namespace(current->namespace);
892 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
893 commit_tree(source_mnt);
896 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
897 list_del_init(&child->mnt_hash);
900 spin_unlock(&vfsmount_lock);
904 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
907 if (mnt->mnt_sb->s_flags & MS_NOUSER)
910 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
911 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
915 mutex_lock(&nd->dentry->d_inode->i_mutex);
916 if (IS_DEADDIR(nd->dentry->d_inode))
919 err = security_sb_check_sb(mnt, nd);
924 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
925 err = attach_recursive_mnt(mnt, nd, NULL);
927 mutex_unlock(&nd->dentry->d_inode->i_mutex);
929 security_sb_post_addmount(mnt, nd);
934 * recursively change the type of the mountpoint.
936 static int do_change_type(struct nameidata *nd, int flag)
938 struct vfsmount *m, *mnt = nd->mnt;
939 int recurse = flag & MS_REC;
940 int type = flag & ~MS_REC;
942 if (nd->dentry != nd->mnt->mnt_root)
945 down_write(&namespace_sem);
946 spin_lock(&vfsmount_lock);
947 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
948 change_mnt_propagation(m, type);
949 spin_unlock(&vfsmount_lock);
950 up_write(&namespace_sem);
957 static int do_loopback(struct nameidata *nd, char *old_name, xid_t xid,
958 unsigned long flags, int mnt_flags)
960 struct nameidata old_nd;
961 struct vfsmount *mnt = NULL;
962 int err = mount_is_safe(nd);
963 int recurse = flags & MS_REC;
966 if (!old_name || !*old_name)
968 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
972 down_write(&namespace_sem);
974 if (IS_MNT_UNBINDABLE(old_nd.mnt))
977 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
982 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
984 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
989 mnt->mnt_flags = mnt_flags;
990 if (flags & MS_XID) {
992 mnt->mnt_flags |= MNT_XID;
995 err = graft_tree(mnt, nd);
997 LIST_HEAD(umount_list);
998 spin_lock(&vfsmount_lock);
999 umount_tree(mnt, 0, &umount_list);
1000 spin_unlock(&vfsmount_lock);
1001 release_mounts(&umount_list);
1003 mnt->mnt_flags = mnt_flags;
1006 up_write(&namespace_sem);
1007 path_release(&old_nd);
1012 * change filesystem flags. dir should be a physical root of filesystem.
1013 * If you've mounted a non-root directory somewhere and want to do remount
1014 * on it - tough luck.
1016 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1017 void *data, xid_t xid)
1020 struct super_block *sb = nd->mnt->mnt_sb;
1022 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_REMOUNT))
1025 if (!check_mnt(nd->mnt))
1028 if (nd->dentry != nd->mnt->mnt_root)
1031 down_write(&sb->s_umount);
1032 err = do_remount_sb(sb, flags, data, 0);
1034 nd->mnt->mnt_flags = mnt_flags;
1035 up_write(&sb->s_umount);
1037 security_sb_post_remount(nd->mnt, flags, data);
1041 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1044 for (p = mnt; p; p = next_mnt(p, mnt)) {
1045 if (IS_MNT_UNBINDABLE(p))
1051 static int do_move_mount(struct nameidata *nd, char *old_name)
1053 struct nameidata old_nd, parent_nd;
1056 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1058 if (!old_name || !*old_name)
1060 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1064 down_write(&namespace_sem);
1065 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1068 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1072 mutex_lock(&nd->dentry->d_inode->i_mutex);
1073 if (IS_DEADDIR(nd->dentry->d_inode))
1076 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1080 if (old_nd.dentry != old_nd.mnt->mnt_root)
1083 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1086 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1087 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1090 * Don't move a mount residing in a shared parent.
1092 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1095 * Don't move a mount tree containing unbindable mounts to a destination
1096 * mount which is shared.
1098 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1101 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1102 if (p == old_nd.mnt)
1105 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1108 spin_lock(&vfsmount_lock);
1109 /* if the mount is moved, it should no longer be expire
1111 list_del_init(&old_nd.mnt->mnt_expire);
1112 spin_unlock(&vfsmount_lock);
1114 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1116 up_write(&namespace_sem);
1118 path_release(&parent_nd);
1119 path_release(&old_nd);
1124 * create a new mount for userspace and request it to be added into the
1127 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1128 int mnt_flags, char *name, void *data)
1130 struct vfsmount *mnt;
1132 if (!type || !memchr(type, 0, PAGE_SIZE))
1135 /* we need capabilities... */
1136 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1139 mnt = do_kern_mount(type, flags, name, data);
1141 return PTR_ERR(mnt);
1143 return do_add_mount(mnt, nd, mnt_flags, NULL);
1147 * add a mount into a namespace's mount tree
1148 * - provide the option of adding the new mount to an expiration list
1150 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1151 int mnt_flags, struct list_head *fslist)
1155 down_write(&namespace_sem);
1156 /* Something was mounted here while we slept */
1157 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1160 if (!check_mnt(nd->mnt))
1163 /* Refuse the same filesystem on the same mount point */
1165 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1166 nd->mnt->mnt_root == nd->dentry)
1170 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1173 newmnt->mnt_flags = mnt_flags;
1174 if ((err = graft_tree(newmnt, nd)))
1178 /* add to the specified expiration list */
1179 spin_lock(&vfsmount_lock);
1180 list_add_tail(&newmnt->mnt_expire, fslist);
1181 spin_unlock(&vfsmount_lock);
1183 up_write(&namespace_sem);
1187 up_write(&namespace_sem);
1192 EXPORT_SYMBOL_GPL(do_add_mount);
1194 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1195 struct list_head *umounts)
1197 spin_lock(&vfsmount_lock);
1200 * Check if mount is still attached, if not, let whoever holds it deal
1203 if (mnt->mnt_parent == mnt) {
1204 spin_unlock(&vfsmount_lock);
1209 * Check that it is still dead: the count should now be 2 - as
1210 * contributed by the vfsmount parent and the mntget above
1212 if (!propagate_mount_busy(mnt, 2)) {
1213 /* delete from the namespace */
1214 touch_namespace(mnt->mnt_namespace);
1215 list_del_init(&mnt->mnt_list);
1216 mnt->mnt_namespace = NULL;
1217 umount_tree(mnt, 1, umounts);
1218 spin_unlock(&vfsmount_lock);
1221 * Someone brought it back to life whilst we didn't have any
1222 * locks held so return it to the expiration list
1224 list_add_tail(&mnt->mnt_expire, mounts);
1225 spin_unlock(&vfsmount_lock);
1230 * process a list of expirable mountpoints with the intent of discarding any
1231 * mountpoints that aren't in use and haven't been touched since last we came
1234 void mark_mounts_for_expiry(struct list_head *mounts)
1236 struct namespace *namespace;
1237 struct vfsmount *mnt, *next;
1238 LIST_HEAD(graveyard);
1240 if (list_empty(mounts))
1243 spin_lock(&vfsmount_lock);
1245 /* extract from the expiration list every vfsmount that matches the
1246 * following criteria:
1247 * - only referenced by its parent vfsmount
1248 * - still marked for expiry (marked on the last call here; marks are
1249 * cleared by mntput())
1251 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1252 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1253 atomic_read(&mnt->mnt_count) != 1)
1257 list_move(&mnt->mnt_expire, &graveyard);
1261 * go through the vfsmounts we've just consigned to the graveyard to
1262 * - check that they're still dead
1263 * - delete the vfsmount from the appropriate namespace under lock
1264 * - dispose of the corpse
1266 while (!list_empty(&graveyard)) {
1268 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1269 list_del_init(&mnt->mnt_expire);
1271 /* don't do anything if the namespace is dead - all the
1272 * vfsmounts from it are going away anyway */
1273 namespace = mnt->mnt_namespace;
1274 if (!namespace || !namespace->root)
1276 get_namespace(namespace);
1278 spin_unlock(&vfsmount_lock);
1279 down_write(&namespace_sem);
1280 expire_mount(mnt, mounts, &umounts);
1281 up_write(&namespace_sem);
1282 release_mounts(&umounts);
1284 put_namespace(namespace);
1285 spin_lock(&vfsmount_lock);
1288 spin_unlock(&vfsmount_lock);
1291 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1294 * Some copy_from_user() implementations do not return the exact number of
1295 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1296 * Note that this function differs from copy_from_user() in that it will oops
1297 * on bad values of `to', rather than returning a short copy.
1299 static long exact_copy_from_user(void *to, const void __user * from,
1303 const char __user *f = from;
1306 if (!access_ok(VERIFY_READ, from, n))
1310 if (__get_user(c, f)) {
1321 int copy_mount_options(const void __user * data, unsigned long *where)
1331 if (!(page = __get_free_page(GFP_KERNEL)))
1334 /* We only care that *some* data at the address the user
1335 * gave us is valid. Just in case, we'll zero
1336 * the remainder of the page.
1338 /* copy_from_user cannot cross TASK_SIZE ! */
1339 size = TASK_SIZE - (unsigned long)data;
1340 if (size > PAGE_SIZE)
1343 i = size - exact_copy_from_user((void *)page, data, size);
1349 memset((char *)page + i, 0, PAGE_SIZE - i);
1355 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1356 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1358 * data is a (void *) that can point to any structure up to
1359 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1360 * information (or be NULL).
1362 * Pre-0.97 versions of mount() didn't have a flags word.
1363 * When the flags word was introduced its top half was required
1364 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1365 * Therefore, if this magic number is present, it carries no information
1366 * and must be discarded.
1368 long do_mount(char *dev_name, char *dir_name, char *type_page,
1369 unsigned long flags, void *data_page)
1371 struct nameidata nd;
1377 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1378 flags &= ~MS_MGC_MSK;
1380 /* Basic sanity checks */
1382 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1384 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1388 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1390 retval = vx_parse_xid(data_page, &xid, 1);
1392 mnt_flags |= MNT_XID;
1393 /* bind and re-mounts get xid flag */
1394 if (flags & (MS_BIND|MS_REMOUNT))
1398 /* Separate the per-mountpoint flags */
1399 if (flags & MS_RDONLY)
1400 mnt_flags |= MNT_RDONLY;
1401 if (flags & MS_NOSUID)
1402 mnt_flags |= MNT_NOSUID;
1403 if (flags & MS_NODEV)
1404 mnt_flags |= MNT_NODEV;
1405 if (flags & MS_NOEXEC)
1406 mnt_flags |= MNT_NOEXEC;
1407 if (flags & MS_NOATIME)
1408 mnt_flags |= MNT_NOATIME;
1409 if (flags & MS_NODIRATIME)
1410 mnt_flags |= MNT_NODIRATIME;
1412 if (!capable(CAP_SYS_ADMIN))
1413 mnt_flags |= MNT_NODEV;
1414 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1415 MS_NOATIME | MS_NODIRATIME);
1417 /* ... and get the mountpoint */
1418 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1422 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1426 if (flags & MS_REMOUNT)
1427 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1429 else if (flags & MS_BIND)
1430 retval = do_loopback(&nd, dev_name, xid, flags, mnt_flags);
1431 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1432 retval = do_change_type(&nd, flags);
1433 else if (flags & MS_MOVE)
1434 retval = do_move_mount(&nd, dev_name);
1436 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1437 dev_name, data_page);
1444 * Allocate a new namespace structure and populate it with contents
1445 * copied from the namespace of the passed in task structure.
1447 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1449 struct namespace *namespace = tsk->namespace;
1450 struct namespace *new_ns;
1451 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1452 struct vfsmount *p, *q;
1454 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1458 atomic_set(&new_ns->count, 1);
1459 INIT_LIST_HEAD(&new_ns->list);
1460 init_waitqueue_head(&new_ns->poll);
1463 down_write(&namespace_sem);
1464 /* First pass: copy the tree topology */
1465 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1466 CL_COPY_ALL | CL_EXPIRE);
1467 if (!new_ns->root) {
1468 up_write(&namespace_sem);
1472 spin_lock(&vfsmount_lock);
1473 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1474 spin_unlock(&vfsmount_lock);
1477 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1478 * as belonging to new namespace. We have already acquired a private
1479 * fs_struct, so tsk->fs->lock is not needed.
1481 p = namespace->root;
1484 q->mnt_namespace = new_ns;
1486 if (p == fs->rootmnt) {
1488 fs->rootmnt = mntget(q);
1490 if (p == fs->pwdmnt) {
1492 fs->pwdmnt = mntget(q);
1494 if (p == fs->altrootmnt) {
1496 fs->altrootmnt = mntget(q);
1499 p = next_mnt(p, namespace->root);
1500 q = next_mnt(q, new_ns->root);
1502 up_write(&namespace_sem);
1514 int copy_namespace(int flags, struct task_struct *tsk)
1516 struct namespace *namespace = tsk->namespace;
1517 struct namespace *new_ns;
1523 get_namespace(namespace);
1525 if (!(flags & CLONE_NEWNS))
1528 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT)) {
1533 new_ns = dup_namespace(tsk, tsk->fs);
1539 tsk->namespace = new_ns;
1542 put_namespace(namespace);
1546 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1547 char __user * type, unsigned long flags,
1551 unsigned long data_page;
1552 unsigned long type_page;
1553 unsigned long dev_page;
1556 retval = copy_mount_options(type, &type_page);
1560 dir_page = getname(dir_name);
1561 retval = PTR_ERR(dir_page);
1562 if (IS_ERR(dir_page))
1565 retval = copy_mount_options(dev_name, &dev_page);
1569 retval = copy_mount_options(data, &data_page);
1574 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1575 flags, (void *)data_page);
1577 free_page(data_page);
1580 free_page(dev_page);
1584 free_page(type_page);
1589 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1590 * It can block. Requires the big lock held.
1592 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1593 struct dentry *dentry)
1595 struct dentry *old_root;
1596 struct vfsmount *old_rootmnt;
1597 write_lock(&fs->lock);
1598 old_root = fs->root;
1599 old_rootmnt = fs->rootmnt;
1600 fs->rootmnt = mntget(mnt);
1601 fs->root = dget(dentry);
1602 write_unlock(&fs->lock);
1605 mntput(old_rootmnt);
1609 EXPORT_SYMBOL_GPL(set_fs_root);
1612 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1613 * It can block. Requires the big lock held.
1615 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1616 struct dentry *dentry)
1618 struct dentry *old_pwd;
1619 struct vfsmount *old_pwdmnt;
1621 write_lock(&fs->lock);
1623 old_pwdmnt = fs->pwdmnt;
1624 fs->pwdmnt = mntget(mnt);
1625 fs->pwd = dget(dentry);
1626 write_unlock(&fs->lock);
1634 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1636 struct task_struct *g, *p;
1637 struct fs_struct *fs;
1639 read_lock(&tasklist_lock);
1640 do_each_thread(g, p) {
1644 atomic_inc(&fs->count);
1646 if (fs->root == old_nd->dentry
1647 && fs->rootmnt == old_nd->mnt)
1648 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1649 if (fs->pwd == old_nd->dentry
1650 && fs->pwdmnt == old_nd->mnt)
1651 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1655 } while_each_thread(g, p);
1656 read_unlock(&tasklist_lock);
1660 * pivot_root Semantics:
1661 * Moves the root file system of the current process to the directory put_old,
1662 * makes new_root as the new root file system of the current process, and sets
1663 * root/cwd of all processes which had them on the current root to new_root.
1666 * The new_root and put_old must be directories, and must not be on the
1667 * same file system as the current process root. The put_old must be
1668 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1669 * pointed to by put_old must yield the same directory as new_root. No other
1670 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1672 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1673 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1674 * in this situation.
1677 * - we don't move root/cwd if they are not at the root (reason: if something
1678 * cared enough to change them, it's probably wrong to force them elsewhere)
1679 * - it's okay to pick a root that isn't the root of a file system, e.g.
1680 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1681 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1684 asmlinkage long sys_pivot_root(const char __user * new_root,
1685 const char __user * put_old)
1687 struct vfsmount *tmp;
1688 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1691 if (!capable(CAP_SYS_ADMIN))
1696 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1701 if (!check_mnt(new_nd.mnt))
1704 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1708 error = security_sb_pivotroot(&old_nd, &new_nd);
1710 path_release(&old_nd);
1714 read_lock(¤t->fs->lock);
1715 user_nd.mnt = mntget(current->fs->rootmnt);
1716 user_nd.dentry = dget(current->fs->root);
1717 read_unlock(¤t->fs->lock);
1718 down_write(&namespace_sem);
1719 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1721 if (IS_MNT_SHARED(old_nd.mnt) ||
1722 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1723 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1725 if (!check_mnt(user_nd.mnt))
1728 if (IS_DEADDIR(new_nd.dentry->d_inode))
1730 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1732 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1735 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1736 goto out2; /* loop, on the same file system */
1738 if (user_nd.mnt->mnt_root != user_nd.dentry)
1739 goto out2; /* not a mountpoint */
1740 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1741 goto out2; /* not attached */
1742 if (new_nd.mnt->mnt_root != new_nd.dentry)
1743 goto out2; /* not a mountpoint */
1744 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1745 goto out2; /* not attached */
1746 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1747 spin_lock(&vfsmount_lock);
1748 if (tmp != new_nd.mnt) {
1750 if (tmp->mnt_parent == tmp)
1751 goto out3; /* already mounted on put_old */
1752 if (tmp->mnt_parent == new_nd.mnt)
1754 tmp = tmp->mnt_parent;
1756 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1758 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1760 detach_mnt(new_nd.mnt, &parent_nd);
1761 detach_mnt(user_nd.mnt, &root_parent);
1762 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1763 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1764 touch_namespace(current->namespace);
1765 spin_unlock(&vfsmount_lock);
1766 chroot_fs_refs(&user_nd, &new_nd);
1767 security_sb_post_pivotroot(&user_nd, &new_nd);
1769 path_release(&root_parent);
1770 path_release(&parent_nd);
1772 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1773 up_write(&namespace_sem);
1774 path_release(&user_nd);
1775 path_release(&old_nd);
1777 path_release(&new_nd);
1782 spin_unlock(&vfsmount_lock);
1786 static void __init init_mount_tree(void)
1788 struct vfsmount *mnt;
1789 struct namespace *namespace;
1790 struct task_struct *g, *p;
1792 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1794 panic("Can't create rootfs");
1795 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1797 panic("Can't allocate initial namespace");
1798 atomic_set(&namespace->count, 1);
1799 INIT_LIST_HEAD(&namespace->list);
1800 init_waitqueue_head(&namespace->poll);
1801 namespace->event = 0;
1802 list_add(&mnt->mnt_list, &namespace->list);
1803 namespace->root = mnt;
1804 mnt->mnt_namespace = namespace;
1806 init_task.namespace = namespace;
1807 read_lock(&tasklist_lock);
1808 do_each_thread(g, p) {
1809 get_namespace(namespace);
1810 p->namespace = namespace;
1811 } while_each_thread(g, p);
1812 read_unlock(&tasklist_lock);
1814 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1815 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1818 void __init mnt_init(unsigned long mempages)
1820 struct list_head *d;
1821 unsigned int nr_hash;
1824 init_rwsem(&namespace_sem);
1826 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1827 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1829 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1831 if (!mount_hashtable)
1832 panic("Failed to allocate mount hash table\n");
1835 * Find the power-of-two list-heads that can fit into the allocation..
1836 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1839 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1843 } while ((nr_hash >> hash_bits) != 0);
1847 * Re-calculate the actual number of entries and the mask
1848 * from the number of bits we can fit.
1850 nr_hash = 1UL << hash_bits;
1851 hash_mask = nr_hash - 1;
1853 printk("Mount-cache hash table entries: %d\n", nr_hash);
1855 /* And initialize the newly allocated array */
1856 d = mount_hashtable;
1864 subsystem_register(&fs_subsys);
1869 void __put_namespace(struct namespace *namespace)
1871 struct vfsmount *root = namespace->root;
1872 LIST_HEAD(umount_list);
1873 namespace->root = NULL;
1874 spin_unlock(&vfsmount_lock);
1875 down_write(&namespace_sem);
1876 spin_lock(&vfsmount_lock);
1877 umount_tree(root, 0, &umount_list);
1878 spin_unlock(&vfsmount_lock);
1879 up_write(&namespace_sem);
1880 release_mounts(&umount_list);