4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <asm/uaccess.h>
28 #include <asm/unistd.h>
31 extern int __init init_rootfs(void);
33 /* spinlock for vfsmount related operations, inplace of dcache_lock */
34 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
38 static struct list_head *mount_hashtable __read_mostly;
39 static int hash_mask __read_mostly, hash_bits __read_mostly;
40 static kmem_cache_t *mnt_cache __read_mostly;
41 static struct rw_semaphore namespace_sem;
44 decl_subsys(fs, NULL, NULL);
45 EXPORT_SYMBOL_GPL(fs_subsys);
47 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
49 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
50 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
51 tmp = tmp + (tmp >> hash_bits);
52 return tmp & hash_mask;
55 struct vfsmount *alloc_vfsmnt(const char *name)
57 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
59 memset(mnt, 0, sizeof(struct vfsmount));
60 atomic_set(&mnt->mnt_count, 1);
61 INIT_LIST_HEAD(&mnt->mnt_hash);
62 INIT_LIST_HEAD(&mnt->mnt_child);
63 INIT_LIST_HEAD(&mnt->mnt_mounts);
64 INIT_LIST_HEAD(&mnt->mnt_list);
65 INIT_LIST_HEAD(&mnt->mnt_expire);
66 INIT_LIST_HEAD(&mnt->mnt_share);
67 INIT_LIST_HEAD(&mnt->mnt_slave_list);
68 INIT_LIST_HEAD(&mnt->mnt_slave);
70 int size = strlen(name) + 1;
71 char *newname = kmalloc(size, GFP_KERNEL);
73 memcpy(newname, name, size);
74 mnt->mnt_devname = newname;
81 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
84 mnt->mnt_root = dget(sb->s_root);
88 EXPORT_SYMBOL(simple_set_mnt);
90 void free_vfsmnt(struct vfsmount *mnt)
92 kfree(mnt->mnt_devname);
93 kmem_cache_free(mnt_cache, mnt);
97 * find the first or last mount at @dentry on vfsmount @mnt depending on
98 * @dir. If @dir is set return the first mount else return the last mount.
100 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
103 struct list_head *head = mount_hashtable + hash(mnt, dentry);
104 struct list_head *tmp = head;
105 struct vfsmount *p, *found = NULL;
108 tmp = dir ? tmp->next : tmp->prev;
112 p = list_entry(tmp, struct vfsmount, mnt_hash);
113 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
122 * lookup_mnt increments the ref count before returning
123 * the vfsmount struct.
125 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
127 struct vfsmount *child_mnt;
128 spin_lock(&vfsmount_lock);
129 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
131 spin_unlock(&vfsmount_lock);
135 static inline int check_mnt(struct vfsmount *mnt)
137 return mnt->mnt_namespace == current->namespace;
140 static void touch_namespace(struct namespace *ns)
144 wake_up_interruptible(&ns->poll);
148 static void __touch_namespace(struct namespace *ns)
150 if (ns && ns->event != event) {
152 wake_up_interruptible(&ns->poll);
156 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
158 old_nd->dentry = mnt->mnt_mountpoint;
159 old_nd->mnt = mnt->mnt_parent;
160 mnt->mnt_parent = mnt;
161 mnt->mnt_mountpoint = mnt->mnt_root;
162 list_del_init(&mnt->mnt_child);
163 list_del_init(&mnt->mnt_hash);
164 old_nd->dentry->d_mounted--;
167 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
168 struct vfsmount *child_mnt)
170 child_mnt->mnt_parent = mntget(mnt);
171 child_mnt->mnt_mountpoint = dget(dentry);
175 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
177 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
178 list_add_tail(&mnt->mnt_hash, mount_hashtable +
179 hash(nd->mnt, nd->dentry));
180 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
184 * the caller must hold vfsmount_lock
186 static void commit_tree(struct vfsmount *mnt)
188 struct vfsmount *parent = mnt->mnt_parent;
191 struct namespace *n = parent->mnt_namespace;
193 BUG_ON(parent == mnt);
195 list_add_tail(&head, &mnt->mnt_list);
196 list_for_each_entry(m, &head, mnt_list)
197 m->mnt_namespace = n;
198 list_splice(&head, n->list.prev);
200 list_add_tail(&mnt->mnt_hash, mount_hashtable +
201 hash(parent, mnt->mnt_mountpoint));
202 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
206 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
208 struct list_head *next = p->mnt_mounts.next;
209 if (next == &p->mnt_mounts) {
213 next = p->mnt_child.next;
214 if (next != &p->mnt_parent->mnt_mounts)
219 return list_entry(next, struct vfsmount, mnt_child);
222 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
224 struct list_head *prev = p->mnt_mounts.prev;
225 while (prev != &p->mnt_mounts) {
226 p = list_entry(prev, struct vfsmount, mnt_child);
227 prev = p->mnt_mounts.prev;
232 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
235 struct super_block *sb = old->mnt_sb;
236 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
239 mnt->mnt_flags = old->mnt_flags;
240 atomic_inc(&sb->s_active);
242 mnt->mnt_root = dget(root);
243 mnt->mnt_mountpoint = mnt->mnt_root;
244 mnt->mnt_parent = mnt;
246 if (flag & CL_SLAVE) {
247 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
248 mnt->mnt_master = old;
249 CLEAR_MNT_SHARED(mnt);
251 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
252 list_add(&mnt->mnt_share, &old->mnt_share);
253 if (IS_MNT_SLAVE(old))
254 list_add(&mnt->mnt_slave, &old->mnt_slave);
255 mnt->mnt_master = old->mnt_master;
257 if (flag & CL_MAKE_SHARED)
260 /* stick the duplicate mount on the same expiry list
261 * as the original if that was on one */
262 if (flag & CL_EXPIRE) {
263 spin_lock(&vfsmount_lock);
264 if (!list_empty(&old->mnt_expire))
265 list_add(&mnt->mnt_expire, &old->mnt_expire);
266 spin_unlock(&vfsmount_lock);
272 static inline void __mntput(struct vfsmount *mnt)
274 struct super_block *sb = mnt->mnt_sb;
277 deactivate_super(sb);
280 void mntput_no_expire(struct vfsmount *mnt)
283 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
284 if (likely(!mnt->mnt_pinned)) {
285 spin_unlock(&vfsmount_lock);
289 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
291 spin_unlock(&vfsmount_lock);
292 acct_auto_close_mnt(mnt);
293 security_sb_umount_close(mnt);
298 EXPORT_SYMBOL(mntput_no_expire);
300 void mnt_pin(struct vfsmount *mnt)
302 spin_lock(&vfsmount_lock);
304 spin_unlock(&vfsmount_lock);
307 EXPORT_SYMBOL(mnt_pin);
309 void mnt_unpin(struct vfsmount *mnt)
311 spin_lock(&vfsmount_lock);
312 if (mnt->mnt_pinned) {
313 atomic_inc(&mnt->mnt_count);
316 spin_unlock(&vfsmount_lock);
319 EXPORT_SYMBOL(mnt_unpin);
322 static void *m_start(struct seq_file *m, loff_t *pos)
324 struct namespace *n = m->private;
328 down_read(&namespace_sem);
329 list_for_each(p, &n->list)
331 return list_entry(p, struct vfsmount, mnt_list);
335 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
337 struct namespace *n = m->private;
338 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
340 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
343 static void m_stop(struct seq_file *m, void *v)
345 up_read(&namespace_sem);
348 static inline void mangle(struct seq_file *m, const char *s)
350 seq_escape(m, s, " \t\n\\");
353 static int show_vfsmnt(struct seq_file *m, void *v)
355 struct vfsmount *mnt = v;
357 static struct proc_fs_info {
361 { MS_SYNCHRONOUS, ",sync" },
362 { MS_DIRSYNC, ",dirsync" },
363 { MS_MANDLOCK, ",mand" },
366 static struct proc_fs_info mnt_info[] = {
367 { MNT_NOSUID, ",nosuid" },
368 { MNT_NODEV, ",nodev" },
369 { MNT_NOEXEC, ",noexec" },
370 { MNT_NOATIME, ",noatime" },
371 { MNT_NODIRATIME, ",nodiratime" },
374 struct proc_fs_info *fs_infop;
376 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
378 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
380 mangle(m, mnt->mnt_sb->s_type->name);
381 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
382 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
383 if (mnt->mnt_sb->s_flags & fs_infop->flag)
384 seq_puts(m, fs_infop->str);
386 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
387 if (mnt->mnt_flags & fs_infop->flag)
388 seq_puts(m, fs_infop->str);
390 if (mnt->mnt_sb->s_op->show_options)
391 err = mnt->mnt_sb->s_op->show_options(m, mnt);
392 seq_puts(m, " 0 0\n");
396 struct seq_operations mounts_op = {
403 static int show_vfsstat(struct seq_file *m, void *v)
405 struct vfsmount *mnt = v;
409 if (mnt->mnt_devname) {
410 seq_puts(m, "device ");
411 mangle(m, mnt->mnt_devname);
413 seq_puts(m, "no device");
416 seq_puts(m, " mounted on ");
417 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
420 /* file system type */
421 seq_puts(m, "with fstype ");
422 mangle(m, mnt->mnt_sb->s_type->name);
424 /* optional statistics */
425 if (mnt->mnt_sb->s_op->show_stats) {
427 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
434 struct seq_operations mountstats_op = {
438 .show = show_vfsstat,
442 * may_umount_tree - check if a mount tree is busy
443 * @mnt: root of mount tree
445 * This is called to check if a tree of mounts has any
446 * open files, pwds, chroots or sub mounts that are
449 int may_umount_tree(struct vfsmount *mnt)
452 int minimum_refs = 0;
455 spin_lock(&vfsmount_lock);
456 for (p = mnt; p; p = next_mnt(p, mnt)) {
457 actual_refs += atomic_read(&p->mnt_count);
460 spin_unlock(&vfsmount_lock);
462 if (actual_refs > minimum_refs)
468 EXPORT_SYMBOL(may_umount_tree);
471 * may_umount - check if a mount point is busy
472 * @mnt: root of mount
474 * This is called to check if a mount point has any
475 * open files, pwds, chroots or sub mounts. If the
476 * mount has sub mounts this will return busy
477 * regardless of whether the sub mounts are busy.
479 * Doesn't take quota and stuff into account. IOW, in some cases it will
480 * give false negatives. The main reason why it's here is that we need
481 * a non-destructive way to look for easily umountable filesystems.
483 int may_umount(struct vfsmount *mnt)
486 spin_lock(&vfsmount_lock);
487 if (propagate_mount_busy(mnt, 2))
489 spin_unlock(&vfsmount_lock);
493 EXPORT_SYMBOL(may_umount);
495 void release_mounts(struct list_head *head)
497 struct vfsmount *mnt;
498 while (!list_empty(head)) {
499 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
500 list_del_init(&mnt->mnt_hash);
501 if (mnt->mnt_parent != mnt) {
502 struct dentry *dentry;
504 spin_lock(&vfsmount_lock);
505 dentry = mnt->mnt_mountpoint;
507 mnt->mnt_mountpoint = mnt->mnt_root;
508 mnt->mnt_parent = mnt;
509 spin_unlock(&vfsmount_lock);
517 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
521 for (p = mnt; p; p = next_mnt(p, mnt))
522 list_move(&p->mnt_hash, kill);
525 propagate_umount(kill);
527 list_for_each_entry(p, kill, mnt_hash) {
528 list_del_init(&p->mnt_expire);
529 list_del_init(&p->mnt_list);
530 __touch_namespace(p->mnt_namespace);
531 p->mnt_namespace = NULL;
532 list_del_init(&p->mnt_child);
533 if (p->mnt_parent != p)
534 p->mnt_mountpoint->d_mounted--;
535 change_mnt_propagation(p, MS_PRIVATE);
539 static int do_umount(struct vfsmount *mnt, int flags)
541 struct super_block *sb = mnt->mnt_sb;
543 LIST_HEAD(umount_list);
545 retval = security_sb_umount(mnt, flags);
550 * Allow userspace to request a mountpoint be expired rather than
551 * unmounting unconditionally. Unmount only happens if:
552 * (1) the mark is already set (the mark is cleared by mntput())
553 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
555 if (flags & MNT_EXPIRE) {
556 if (mnt == current->fs->rootmnt ||
557 flags & (MNT_FORCE | MNT_DETACH))
560 if (atomic_read(&mnt->mnt_count) != 2)
563 if (!xchg(&mnt->mnt_expiry_mark, 1))
568 * If we may have to abort operations to get out of this
569 * mount, and they will themselves hold resources we must
570 * allow the fs to do things. In the Unix tradition of
571 * 'Gee thats tricky lets do it in userspace' the umount_begin
572 * might fail to complete on the first run through as other tasks
573 * must return, and the like. Thats for the mount program to worry
574 * about for the moment.
578 if (sb->s_op->umount_begin)
579 sb->s_op->umount_begin(mnt, flags);
583 * No sense to grab the lock for this test, but test itself looks
584 * somewhat bogus. Suggestions for better replacement?
585 * Ho-hum... In principle, we might treat that as umount + switch
586 * to rootfs. GC would eventually take care of the old vfsmount.
587 * Actually it makes sense, especially if rootfs would contain a
588 * /reboot - static binary that would close all descriptors and
589 * call reboot(9). Then init(8) could umount root and exec /reboot.
591 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
593 * Special case for "unmounting" root ...
594 * we just try to remount it readonly.
596 down_write(&sb->s_umount);
597 if (!(sb->s_flags & MS_RDONLY)) {
600 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
603 up_write(&sb->s_umount);
607 down_write(&namespace_sem);
608 spin_lock(&vfsmount_lock);
612 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
613 if (!list_empty(&mnt->mnt_list))
614 umount_tree(mnt, 1, &umount_list);
617 spin_unlock(&vfsmount_lock);
619 security_sb_umount_busy(mnt);
620 up_write(&namespace_sem);
621 release_mounts(&umount_list);
626 * Now umount can handle mount points as well as block devices.
627 * This is important for filesystems which use unnamed block devices.
629 * We now support a flag for forced unmount like the other 'big iron'
630 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
633 asmlinkage long sys_umount(char __user * name, int flags)
638 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
642 if (nd.dentry != nd.mnt->mnt_root)
644 if (!check_mnt(nd.mnt))
648 if (!capable(CAP_SYS_ADMIN))
651 retval = do_umount(nd.mnt, flags);
653 path_release_on_umount(&nd);
658 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
661 * The 2.0 compatible umount. No flags.
663 asmlinkage long sys_oldumount(char __user * name)
665 return sys_umount(name, 0);
670 static int mount_is_safe(struct nameidata *nd)
672 if (capable(CAP_SYS_ADMIN))
676 if (S_ISLNK(nd->dentry->d_inode->i_mode))
678 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
679 if (current->uid != nd->dentry->d_inode->i_uid)
682 if (vfs_permission(nd, MAY_WRITE))
688 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
693 if (d == NULL || d == d->d_parent)
699 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
702 struct vfsmount *res, *p, *q, *r, *s;
705 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
708 res = q = clone_mnt(mnt, dentry, flag);
711 q->mnt_mountpoint = mnt->mnt_mountpoint;
714 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
715 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
718 for (s = r; s; s = next_mnt(s, r)) {
719 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
720 s = skip_mnt_tree(s);
723 while (p != s->mnt_parent) {
729 nd.dentry = p->mnt_mountpoint;
730 q = clone_mnt(p, p->mnt_root, flag);
733 spin_lock(&vfsmount_lock);
734 list_add_tail(&q->mnt_list, &res->mnt_list);
736 spin_unlock(&vfsmount_lock);
742 LIST_HEAD(umount_list);
743 spin_lock(&vfsmount_lock);
744 umount_tree(res, 0, &umount_list);
745 spin_unlock(&vfsmount_lock);
746 release_mounts(&umount_list);
752 * @source_mnt : mount tree to be attached
753 * @nd : place the mount tree @source_mnt is attached
754 * @parent_nd : if non-null, detach the source_mnt from its parent and
755 * store the parent mount and mountpoint dentry.
756 * (done when source_mnt is moved)
758 * NOTE: in the table below explains the semantics when a source mount
759 * of a given type is attached to a destination mount of a given type.
760 * ---------------------------------------------------------------------------
761 * | BIND MOUNT OPERATION |
762 * |**************************************************************************
763 * | source-->| shared | private | slave | unbindable |
767 * |**************************************************************************
768 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
770 * |non-shared| shared (+) | private | slave (*) | invalid |
771 * ***************************************************************************
772 * A bind operation clones the source mount and mounts the clone on the
775 * (++) the cloned mount is propagated to all the mounts in the propagation
776 * tree of the destination mount and the cloned mount is added to
777 * the peer group of the source mount.
778 * (+) the cloned mount is created under the destination mount and is marked
779 * as shared. The cloned mount is added to the peer group of the source
781 * (+++) the mount is propagated to all the mounts in the propagation tree
782 * of the destination mount and the cloned mount is made slave
783 * of the same master as that of the source mount. The cloned mount
784 * is marked as 'shared and slave'.
785 * (*) the cloned mount is made a slave of the same master as that of the
788 * ---------------------------------------------------------------------------
789 * | MOVE MOUNT OPERATION |
790 * |**************************************************************************
791 * | source-->| shared | private | slave | unbindable |
795 * |**************************************************************************
796 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
798 * |non-shared| shared (+*) | private | slave (*) | unbindable |
799 * ***************************************************************************
801 * (+) the mount is moved to the destination. And is then propagated to
802 * all the mounts in the propagation tree of the destination mount.
803 * (+*) the mount is moved to the destination.
804 * (+++) the mount is moved to the destination and is then propagated to
805 * all the mounts belonging to the destination mount's propagation tree.
806 * the mount is marked as 'shared and slave'.
807 * (*) the mount continues to be a slave at the new location.
809 * if the source mount is a tree, the operations explained above is
810 * applied to each mount in the tree.
811 * Must be called without spinlocks held, since this function can sleep
814 static int attach_recursive_mnt(struct vfsmount *source_mnt,
815 struct nameidata *nd, struct nameidata *parent_nd)
817 LIST_HEAD(tree_list);
818 struct vfsmount *dest_mnt = nd->mnt;
819 struct dentry *dest_dentry = nd->dentry;
820 struct vfsmount *child, *p;
822 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
825 if (IS_MNT_SHARED(dest_mnt)) {
826 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
830 spin_lock(&vfsmount_lock);
832 detach_mnt(source_mnt, parent_nd);
833 attach_mnt(source_mnt, nd);
834 touch_namespace(current->namespace);
836 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
837 commit_tree(source_mnt);
840 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
841 list_del_init(&child->mnt_hash);
844 spin_unlock(&vfsmount_lock);
848 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
851 if (mnt->mnt_sb->s_flags & MS_NOUSER)
854 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
855 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
859 mutex_lock(&nd->dentry->d_inode->i_mutex);
860 if (IS_DEADDIR(nd->dentry->d_inode))
863 err = security_sb_check_sb(mnt, nd);
868 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
869 err = attach_recursive_mnt(mnt, nd, NULL);
871 mutex_unlock(&nd->dentry->d_inode->i_mutex);
873 security_sb_post_addmount(mnt, nd);
878 * recursively change the type of the mountpoint.
880 static int do_change_type(struct nameidata *nd, int flag)
882 struct vfsmount *m, *mnt = nd->mnt;
883 int recurse = flag & MS_REC;
884 int type = flag & ~MS_REC;
886 if (nd->dentry != nd->mnt->mnt_root)
889 down_write(&namespace_sem);
890 spin_lock(&vfsmount_lock);
891 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
892 change_mnt_propagation(m, type);
893 spin_unlock(&vfsmount_lock);
894 up_write(&namespace_sem);
901 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
903 struct nameidata old_nd;
904 struct vfsmount *mnt = NULL;
905 int err = mount_is_safe(nd);
908 if (!old_name || !*old_name)
910 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
914 down_write(&namespace_sem);
916 if (IS_MNT_UNBINDABLE(old_nd.mnt))
919 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
924 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
926 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
931 err = graft_tree(mnt, nd);
933 LIST_HEAD(umount_list);
934 spin_lock(&vfsmount_lock);
935 umount_tree(mnt, 0, &umount_list);
936 spin_unlock(&vfsmount_lock);
937 release_mounts(&umount_list);
941 up_write(&namespace_sem);
942 path_release(&old_nd);
947 * change filesystem flags. dir should be a physical root of filesystem.
948 * If you've mounted a non-root directory somewhere and want to do remount
949 * on it - tough luck.
951 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
955 struct super_block *sb = nd->mnt->mnt_sb;
957 if (!capable(CAP_SYS_ADMIN))
960 if (!check_mnt(nd->mnt))
963 if (nd->dentry != nd->mnt->mnt_root)
966 down_write(&sb->s_umount);
967 err = do_remount_sb(sb, flags, data, 0);
969 nd->mnt->mnt_flags = mnt_flags;
970 up_write(&sb->s_umount);
972 security_sb_post_remount(nd->mnt, flags, data);
976 static inline int tree_contains_unbindable(struct vfsmount *mnt)
979 for (p = mnt; p; p = next_mnt(p, mnt)) {
980 if (IS_MNT_UNBINDABLE(p))
986 static int do_move_mount(struct nameidata *nd, char *old_name)
988 struct nameidata old_nd, parent_nd;
991 if (!capable(CAP_SYS_ADMIN))
993 if (!old_name || !*old_name)
995 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
999 down_write(&namespace_sem);
1000 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1003 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1007 mutex_lock(&nd->dentry->d_inode->i_mutex);
1008 if (IS_DEADDIR(nd->dentry->d_inode))
1011 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1015 if (old_nd.dentry != old_nd.mnt->mnt_root)
1018 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1021 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1022 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1025 * Don't move a mount residing in a shared parent.
1027 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1030 * Don't move a mount tree containing unbindable mounts to a destination
1031 * mount which is shared.
1033 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1036 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1037 if (p == old_nd.mnt)
1040 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1043 spin_lock(&vfsmount_lock);
1044 /* if the mount is moved, it should no longer be expire
1046 list_del_init(&old_nd.mnt->mnt_expire);
1047 spin_unlock(&vfsmount_lock);
1049 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1051 up_write(&namespace_sem);
1053 path_release(&parent_nd);
1054 path_release(&old_nd);
1059 * create a new mount for userspace and request it to be added into the
1062 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1063 int mnt_flags, char *name, void *data)
1065 struct vfsmount *mnt;
1067 if (!type || !memchr(type, 0, PAGE_SIZE))
1070 /* we need capabilities... */
1071 if (!capable(CAP_SYS_ADMIN))
1074 mnt = do_kern_mount(type, flags, name, data);
1076 return PTR_ERR(mnt);
1078 return do_add_mount(mnt, nd, mnt_flags, NULL);
1082 * add a mount into a namespace's mount tree
1083 * - provide the option of adding the new mount to an expiration list
1085 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1086 int mnt_flags, struct list_head *fslist)
1090 down_write(&namespace_sem);
1091 /* Something was mounted here while we slept */
1092 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1095 if (!check_mnt(nd->mnt))
1098 /* Refuse the same filesystem on the same mount point */
1100 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1101 nd->mnt->mnt_root == nd->dentry)
1105 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1108 newmnt->mnt_flags = mnt_flags;
1109 if ((err = graft_tree(newmnt, nd)))
1113 /* add to the specified expiration list */
1114 spin_lock(&vfsmount_lock);
1115 list_add_tail(&newmnt->mnt_expire, fslist);
1116 spin_unlock(&vfsmount_lock);
1118 up_write(&namespace_sem);
1122 up_write(&namespace_sem);
1127 EXPORT_SYMBOL_GPL(do_add_mount);
1129 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1130 struct list_head *umounts)
1132 spin_lock(&vfsmount_lock);
1135 * Check if mount is still attached, if not, let whoever holds it deal
1138 if (mnt->mnt_parent == mnt) {
1139 spin_unlock(&vfsmount_lock);
1144 * Check that it is still dead: the count should now be 2 - as
1145 * contributed by the vfsmount parent and the mntget above
1147 if (!propagate_mount_busy(mnt, 2)) {
1148 /* delete from the namespace */
1149 touch_namespace(mnt->mnt_namespace);
1150 list_del_init(&mnt->mnt_list);
1151 mnt->mnt_namespace = NULL;
1152 umount_tree(mnt, 1, umounts);
1153 spin_unlock(&vfsmount_lock);
1156 * Someone brought it back to life whilst we didn't have any
1157 * locks held so return it to the expiration list
1159 list_add_tail(&mnt->mnt_expire, mounts);
1160 spin_unlock(&vfsmount_lock);
1165 * go through the vfsmounts we've just consigned to the graveyard to
1166 * - check that they're still dead
1167 * - delete the vfsmount from the appropriate namespace under lock
1168 * - dispose of the corpse
1170 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1172 struct namespace *namespace;
1173 struct vfsmount *mnt;
1175 while (!list_empty(graveyard)) {
1177 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1178 list_del_init(&mnt->mnt_expire);
1180 /* don't do anything if the namespace is dead - all the
1181 * vfsmounts from it are going away anyway */
1182 namespace = mnt->mnt_namespace;
1183 if (!namespace || !namespace->root)
1185 get_namespace(namespace);
1187 spin_unlock(&vfsmount_lock);
1188 down_write(&namespace_sem);
1189 expire_mount(mnt, mounts, &umounts);
1190 up_write(&namespace_sem);
1191 release_mounts(&umounts);
1193 put_namespace(namespace);
1194 spin_lock(&vfsmount_lock);
1199 * process a list of expirable mountpoints with the intent of discarding any
1200 * mountpoints that aren't in use and haven't been touched since last we came
1203 void mark_mounts_for_expiry(struct list_head *mounts)
1205 struct vfsmount *mnt, *next;
1206 LIST_HEAD(graveyard);
1208 if (list_empty(mounts))
1211 spin_lock(&vfsmount_lock);
1213 /* extract from the expiration list every vfsmount that matches the
1214 * following criteria:
1215 * - only referenced by its parent vfsmount
1216 * - still marked for expiry (marked on the last call here; marks are
1217 * cleared by mntput())
1219 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1220 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1221 atomic_read(&mnt->mnt_count) != 1)
1225 list_move(&mnt->mnt_expire, &graveyard);
1228 expire_mount_list(&graveyard, mounts);
1230 spin_unlock(&vfsmount_lock);
1233 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1236 * Ripoff of 'select_parent()'
1238 * search the list of submounts for a given mountpoint, and move any
1239 * shrinkable submounts to the 'graveyard' list.
1241 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1243 struct vfsmount *this_parent = parent;
1244 struct list_head *next;
1248 next = this_parent->mnt_mounts.next;
1250 while (next != &this_parent->mnt_mounts) {
1251 struct list_head *tmp = next;
1252 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1255 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1258 * Descend a level if the d_mounts list is non-empty.
1260 if (!list_empty(&mnt->mnt_mounts)) {
1265 if (!propagate_mount_busy(mnt, 1)) {
1267 list_move_tail(&mnt->mnt_expire, graveyard);
1272 * All done at this level ... ascend and resume the search
1274 if (this_parent != parent) {
1275 next = this_parent->mnt_child.next;
1276 this_parent = this_parent->mnt_parent;
1283 * process a list of expirable mountpoints with the intent of discarding any
1284 * submounts of a specific parent mountpoint
1286 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1288 LIST_HEAD(graveyard);
1291 spin_lock(&vfsmount_lock);
1293 /* extract submounts of 'mountpoint' from the expiration list */
1294 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1295 expire_mount_list(&graveyard, mounts);
1297 spin_unlock(&vfsmount_lock);
1300 EXPORT_SYMBOL_GPL(shrink_submounts);
1303 * Some copy_from_user() implementations do not return the exact number of
1304 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1305 * Note that this function differs from copy_from_user() in that it will oops
1306 * on bad values of `to', rather than returning a short copy.
1308 static long exact_copy_from_user(void *to, const void __user * from,
1312 const char __user *f = from;
1315 if (!access_ok(VERIFY_READ, from, n))
1319 if (__get_user(c, f)) {
1330 int copy_mount_options(const void __user * data, unsigned long *where)
1340 if (!(page = __get_free_page(GFP_KERNEL)))
1343 /* We only care that *some* data at the address the user
1344 * gave us is valid. Just in case, we'll zero
1345 * the remainder of the page.
1347 /* copy_from_user cannot cross TASK_SIZE ! */
1348 size = TASK_SIZE - (unsigned long)data;
1349 if (size > PAGE_SIZE)
1352 i = size - exact_copy_from_user((void *)page, data, size);
1358 memset((char *)page + i, 0, PAGE_SIZE - i);
1364 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1365 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1367 * data is a (void *) that can point to any structure up to
1368 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1369 * information (or be NULL).
1371 * Pre-0.97 versions of mount() didn't have a flags word.
1372 * When the flags word was introduced its top half was required
1373 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1374 * Therefore, if this magic number is present, it carries no information
1375 * and must be discarded.
1377 long do_mount(char *dev_name, char *dir_name, char *type_page,
1378 unsigned long flags, void *data_page)
1380 struct nameidata nd;
1385 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1386 flags &= ~MS_MGC_MSK;
1388 /* Basic sanity checks */
1390 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1392 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1396 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1398 /* Separate the per-mountpoint flags */
1399 if (flags & MS_NOSUID)
1400 mnt_flags |= MNT_NOSUID;
1401 if (flags & MS_NODEV)
1402 mnt_flags |= MNT_NODEV;
1403 if (flags & MS_NOEXEC)
1404 mnt_flags |= MNT_NOEXEC;
1405 if (flags & MS_NOATIME)
1406 mnt_flags |= MNT_NOATIME;
1407 if (flags & MS_NODIRATIME)
1408 mnt_flags |= MNT_NODIRATIME;
1410 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1411 MS_NOATIME | MS_NODIRATIME);
1413 /* ... and get the mountpoint */
1414 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1418 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1422 if (flags & MS_REMOUNT)
1423 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1425 else if (flags & MS_BIND)
1426 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1427 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1428 retval = do_change_type(&nd, flags);
1429 else if (flags & MS_MOVE)
1430 retval = do_move_mount(&nd, dev_name);
1432 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1433 dev_name, data_page);
1440 * Allocate a new namespace structure and populate it with contents
1441 * copied from the namespace of the passed in task structure.
1443 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1445 struct namespace *namespace = tsk->namespace;
1446 struct namespace *new_ns;
1447 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1448 struct vfsmount *p, *q;
1450 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1454 atomic_set(&new_ns->count, 1);
1455 INIT_LIST_HEAD(&new_ns->list);
1456 init_waitqueue_head(&new_ns->poll);
1459 down_write(&namespace_sem);
1460 /* First pass: copy the tree topology */
1461 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1462 CL_COPY_ALL | CL_EXPIRE);
1463 if (!new_ns->root) {
1464 up_write(&namespace_sem);
1468 spin_lock(&vfsmount_lock);
1469 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1470 spin_unlock(&vfsmount_lock);
1473 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1474 * as belonging to new namespace. We have already acquired a private
1475 * fs_struct, so tsk->fs->lock is not needed.
1477 p = namespace->root;
1480 q->mnt_namespace = new_ns;
1482 if (p == fs->rootmnt) {
1484 fs->rootmnt = mntget(q);
1486 if (p == fs->pwdmnt) {
1488 fs->pwdmnt = mntget(q);
1490 if (p == fs->altrootmnt) {
1492 fs->altrootmnt = mntget(q);
1495 p = next_mnt(p, namespace->root);
1496 q = next_mnt(q, new_ns->root);
1498 up_write(&namespace_sem);
1510 int copy_namespace(int flags, struct task_struct *tsk)
1512 struct namespace *namespace = tsk->namespace;
1513 struct namespace *new_ns;
1519 get_namespace(namespace);
1521 if (!(flags & CLONE_NEWNS))
1524 if (!capable(CAP_SYS_ADMIN)) {
1529 new_ns = dup_namespace(tsk, tsk->fs);
1535 tsk->namespace = new_ns;
1538 put_namespace(namespace);
1542 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1543 char __user * type, unsigned long flags,
1547 unsigned long data_page;
1548 unsigned long type_page;
1549 unsigned long dev_page;
1552 retval = copy_mount_options(type, &type_page);
1556 dir_page = getname(dir_name);
1557 retval = PTR_ERR(dir_page);
1558 if (IS_ERR(dir_page))
1561 retval = copy_mount_options(dev_name, &dev_page);
1565 retval = copy_mount_options(data, &data_page);
1570 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1571 flags, (void *)data_page);
1573 free_page(data_page);
1576 free_page(dev_page);
1580 free_page(type_page);
1585 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1586 * It can block. Requires the big lock held.
1588 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1589 struct dentry *dentry)
1591 struct dentry *old_root;
1592 struct vfsmount *old_rootmnt;
1593 write_lock(&fs->lock);
1594 old_root = fs->root;
1595 old_rootmnt = fs->rootmnt;
1596 fs->rootmnt = mntget(mnt);
1597 fs->root = dget(dentry);
1598 write_unlock(&fs->lock);
1601 mntput(old_rootmnt);
1606 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1607 * It can block. Requires the big lock held.
1609 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1610 struct dentry *dentry)
1612 struct dentry *old_pwd;
1613 struct vfsmount *old_pwdmnt;
1615 write_lock(&fs->lock);
1617 old_pwdmnt = fs->pwdmnt;
1618 fs->pwdmnt = mntget(mnt);
1619 fs->pwd = dget(dentry);
1620 write_unlock(&fs->lock);
1628 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1630 struct task_struct *g, *p;
1631 struct fs_struct *fs;
1633 read_lock(&tasklist_lock);
1634 do_each_thread(g, p) {
1638 atomic_inc(&fs->count);
1640 if (fs->root == old_nd->dentry
1641 && fs->rootmnt == old_nd->mnt)
1642 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1643 if (fs->pwd == old_nd->dentry
1644 && fs->pwdmnt == old_nd->mnt)
1645 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1649 } while_each_thread(g, p);
1650 read_unlock(&tasklist_lock);
1654 * pivot_root Semantics:
1655 * Moves the root file system of the current process to the directory put_old,
1656 * makes new_root as the new root file system of the current process, and sets
1657 * root/cwd of all processes which had them on the current root to new_root.
1660 * The new_root and put_old must be directories, and must not be on the
1661 * same file system as the current process root. The put_old must be
1662 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1663 * pointed to by put_old must yield the same directory as new_root. No other
1664 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1666 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1667 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1668 * in this situation.
1671 * - we don't move root/cwd if they are not at the root (reason: if something
1672 * cared enough to change them, it's probably wrong to force them elsewhere)
1673 * - it's okay to pick a root that isn't the root of a file system, e.g.
1674 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1675 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1678 asmlinkage long sys_pivot_root(const char __user * new_root,
1679 const char __user * put_old)
1681 struct vfsmount *tmp;
1682 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1685 if (!capable(CAP_SYS_ADMIN))
1690 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1695 if (!check_mnt(new_nd.mnt))
1698 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1702 error = security_sb_pivotroot(&old_nd, &new_nd);
1704 path_release(&old_nd);
1708 read_lock(¤t->fs->lock);
1709 user_nd.mnt = mntget(current->fs->rootmnt);
1710 user_nd.dentry = dget(current->fs->root);
1711 read_unlock(¤t->fs->lock);
1712 down_write(&namespace_sem);
1713 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1715 if (IS_MNT_SHARED(old_nd.mnt) ||
1716 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1717 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1719 if (!check_mnt(user_nd.mnt))
1722 if (IS_DEADDIR(new_nd.dentry->d_inode))
1724 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1726 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1729 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1730 goto out2; /* loop, on the same file system */
1732 if (user_nd.mnt->mnt_root != user_nd.dentry)
1733 goto out2; /* not a mountpoint */
1734 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1735 goto out2; /* not attached */
1736 if (new_nd.mnt->mnt_root != new_nd.dentry)
1737 goto out2; /* not a mountpoint */
1738 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1739 goto out2; /* not attached */
1740 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1741 spin_lock(&vfsmount_lock);
1742 if (tmp != new_nd.mnt) {
1744 if (tmp->mnt_parent == tmp)
1745 goto out3; /* already mounted on put_old */
1746 if (tmp->mnt_parent == new_nd.mnt)
1748 tmp = tmp->mnt_parent;
1750 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1752 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1754 detach_mnt(new_nd.mnt, &parent_nd);
1755 detach_mnt(user_nd.mnt, &root_parent);
1756 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1757 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1758 touch_namespace(current->namespace);
1759 spin_unlock(&vfsmount_lock);
1760 chroot_fs_refs(&user_nd, &new_nd);
1761 security_sb_post_pivotroot(&user_nd, &new_nd);
1763 path_release(&root_parent);
1764 path_release(&parent_nd);
1766 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1767 up_write(&namespace_sem);
1768 path_release(&user_nd);
1769 path_release(&old_nd);
1771 path_release(&new_nd);
1776 spin_unlock(&vfsmount_lock);
1780 static void __init init_mount_tree(void)
1782 struct vfsmount *mnt;
1783 struct namespace *namespace;
1784 struct task_struct *g, *p;
1786 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1788 panic("Can't create rootfs");
1789 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1791 panic("Can't allocate initial namespace");
1792 atomic_set(&namespace->count, 1);
1793 INIT_LIST_HEAD(&namespace->list);
1794 init_waitqueue_head(&namespace->poll);
1795 namespace->event = 0;
1796 list_add(&mnt->mnt_list, &namespace->list);
1797 namespace->root = mnt;
1798 mnt->mnt_namespace = namespace;
1800 init_task.namespace = namespace;
1801 read_lock(&tasklist_lock);
1802 do_each_thread(g, p) {
1803 get_namespace(namespace);
1804 p->namespace = namespace;
1805 } while_each_thread(g, p);
1806 read_unlock(&tasklist_lock);
1808 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1809 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1812 void __init mnt_init(unsigned long mempages)
1814 struct list_head *d;
1815 unsigned int nr_hash;
1819 init_rwsem(&namespace_sem);
1821 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1822 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1824 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1826 if (!mount_hashtable)
1827 panic("Failed to allocate mount hash table\n");
1830 * Find the power-of-two list-heads that can fit into the allocation..
1831 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1834 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1838 } while ((nr_hash >> hash_bits) != 0);
1842 * Re-calculate the actual number of entries and the mask
1843 * from the number of bits we can fit.
1845 nr_hash = 1UL << hash_bits;
1846 hash_mask = nr_hash - 1;
1848 printk("Mount-cache hash table entries: %d\n", nr_hash);
1850 /* And initialize the newly allocated array */
1851 d = mount_hashtable;
1860 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1862 err = subsystem_register(&fs_subsys);
1864 printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1870 void __put_namespace(struct namespace *namespace)
1872 struct vfsmount *root = namespace->root;
1873 LIST_HEAD(umount_list);
1874 namespace->root = NULL;
1875 spin_unlock(&vfsmount_lock);
1876 down_write(&namespace_sem);
1877 spin_lock(&vfsmount_lock);
1878 umount_tree(root, 0, &umount_list);
1879 spin_unlock(&vfsmount_lock);
1880 up_write(&namespace_sem);
1881 release_mounts(&umount_list);