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/vs_base.h>
27 #include <linux/vserver/namespace.h>
28 #include <linux/vserver/xid.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
33 extern int __init init_rootfs(void);
36 extern int __init sysfs_init(void);
38 static inline int sysfs_init(void)
44 /* spinlock for vfsmount related operations, inplace of dcache_lock */
45 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
49 static struct list_head *mount_hashtable __read_mostly;
50 static int hash_mask __read_mostly, hash_bits __read_mostly;
51 static kmem_cache_t *mnt_cache __read_mostly;
52 static struct rw_semaphore namespace_sem;
55 decl_subsys(fs, NULL, NULL);
56 EXPORT_SYMBOL_GPL(fs_subsys);
58 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
60 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
62 tmp = tmp + (tmp >> hash_bits);
63 return tmp & hash_mask;
66 struct vfsmount *alloc_vfsmnt(const char *name)
68 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
70 memset(mnt, 0, sizeof(struct vfsmount));
71 atomic_set(&mnt->mnt_count, 1);
72 INIT_LIST_HEAD(&mnt->mnt_hash);
73 INIT_LIST_HEAD(&mnt->mnt_child);
74 INIT_LIST_HEAD(&mnt->mnt_mounts);
75 INIT_LIST_HEAD(&mnt->mnt_list);
76 INIT_LIST_HEAD(&mnt->mnt_expire);
77 INIT_LIST_HEAD(&mnt->mnt_share);
78 INIT_LIST_HEAD(&mnt->mnt_slave_list);
79 INIT_LIST_HEAD(&mnt->mnt_slave);
81 int size = strlen(name) + 1;
82 char *newname = kmalloc(size, GFP_KERNEL);
84 memcpy(newname, name, size);
85 mnt->mnt_devname = newname;
92 void free_vfsmnt(struct vfsmount *mnt)
94 kfree(mnt->mnt_devname);
95 kmem_cache_free(mnt_cache, mnt);
99 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
102 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
105 struct list_head *head = mount_hashtable + hash(mnt, dentry);
106 struct list_head *tmp = head;
107 struct vfsmount *p, *found = NULL;
110 tmp = dir ? tmp->next : tmp->prev;
114 p = list_entry(tmp, struct vfsmount, mnt_hash);
115 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
127 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
129 struct vfsmount *child_mnt;
130 spin_lock(&vfsmount_lock);
131 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
133 spin_unlock(&vfsmount_lock);
137 static inline int check_mnt(struct vfsmount *mnt)
139 return mnt->mnt_namespace == current->namespace;
142 static void touch_namespace(struct namespace *ns)
146 wake_up_interruptible(&ns->poll);
150 static void __touch_namespace(struct namespace *ns)
152 if (ns && ns->event != event) {
154 wake_up_interruptible(&ns->poll);
158 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
160 old_nd->dentry = mnt->mnt_mountpoint;
161 old_nd->mnt = mnt->mnt_parent;
162 mnt->mnt_parent = mnt;
163 mnt->mnt_mountpoint = mnt->mnt_root;
164 list_del_init(&mnt->mnt_child);
165 list_del_init(&mnt->mnt_hash);
166 old_nd->dentry->d_mounted--;
169 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
170 struct vfsmount *child_mnt)
172 child_mnt->mnt_parent = mntget(mnt);
173 child_mnt->mnt_mountpoint = dget(dentry);
177 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
179 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
181 hash(nd->mnt, nd->dentry));
182 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
186 * the caller must hold vfsmount_lock
188 static void commit_tree(struct vfsmount *mnt)
190 struct vfsmount *parent = mnt->mnt_parent;
193 struct namespace *n = parent->mnt_namespace;
195 BUG_ON(parent == mnt);
197 list_add_tail(&head, &mnt->mnt_list);
198 list_for_each_entry(m, &head, mnt_list)
199 m->mnt_namespace = n;
200 list_splice(&head, n->list.prev);
202 list_add_tail(&mnt->mnt_hash, mount_hashtable +
203 hash(parent, mnt->mnt_mountpoint));
204 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
208 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
210 struct list_head *next = p->mnt_mounts.next;
211 if (next == &p->mnt_mounts) {
215 next = p->mnt_child.next;
216 if (next != &p->mnt_parent->mnt_mounts)
221 return list_entry(next, struct vfsmount, mnt_child);
224 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
226 struct list_head *prev = p->mnt_mounts.prev;
227 while (prev != &p->mnt_mounts) {
228 p = list_entry(prev, struct vfsmount, mnt_child);
229 prev = p->mnt_mounts.prev;
234 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
237 struct super_block *sb = old->mnt_sb;
238 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
241 mnt->mnt_flags = old->mnt_flags;
242 atomic_inc(&sb->s_active);
244 mnt->mnt_root = dget(root);
245 mnt->mnt_mountpoint = mnt->mnt_root;
246 mnt->mnt_parent = mnt;
247 mnt->mnt_xid = old->mnt_xid;
249 if (flag & CL_SLAVE) {
250 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
251 mnt->mnt_master = old;
252 CLEAR_MNT_SHARED(mnt);
254 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
255 list_add(&mnt->mnt_share, &old->mnt_share);
256 if (IS_MNT_SLAVE(old))
257 list_add(&mnt->mnt_slave, &old->mnt_slave);
258 mnt->mnt_master = old->mnt_master;
260 if (flag & CL_MAKE_SHARED)
263 /* stick the duplicate mount on the same expiry list
264 * as the original if that was on one */
265 if (flag & CL_EXPIRE) {
266 spin_lock(&vfsmount_lock);
267 if (!list_empty(&old->mnt_expire))
268 list_add(&mnt->mnt_expire, &old->mnt_expire);
269 spin_unlock(&vfsmount_lock);
275 static inline void __mntput(struct vfsmount *mnt)
277 struct super_block *sb = mnt->mnt_sb;
280 deactivate_super(sb);
283 void mntput_no_expire(struct vfsmount *mnt)
286 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
287 if (likely(!mnt->mnt_pinned)) {
288 spin_unlock(&vfsmount_lock);
292 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
294 spin_unlock(&vfsmount_lock);
295 acct_auto_close_mnt(mnt);
296 security_sb_umount_close(mnt);
301 EXPORT_SYMBOL(mntput_no_expire);
303 void mnt_pin(struct vfsmount *mnt)
305 spin_lock(&vfsmount_lock);
307 spin_unlock(&vfsmount_lock);
310 EXPORT_SYMBOL(mnt_pin);
312 void mnt_unpin(struct vfsmount *mnt)
314 spin_lock(&vfsmount_lock);
315 if (mnt->mnt_pinned) {
316 atomic_inc(&mnt->mnt_count);
319 spin_unlock(&vfsmount_lock);
322 EXPORT_SYMBOL(mnt_unpin);
325 static void *m_start(struct seq_file *m, loff_t *pos)
327 struct namespace *n = m->private;
331 down_read(&namespace_sem);
332 list_for_each(p, &n->list)
334 return list_entry(p, struct vfsmount, mnt_list);
338 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
340 struct namespace *n = m->private;
341 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
343 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
346 static void m_stop(struct seq_file *m, void *v)
348 up_read(&namespace_sem);
351 static inline void mangle(struct seq_file *m, const char *s)
353 seq_escape(m, s, " \t\n\\");
356 static int mnt_is_reachable(struct vfsmount *mnt)
358 struct vfsmount *root_mnt;
359 struct dentry *root, *point;
362 if (mnt == mnt->mnt_namespace->root)
365 spin_lock(&dcache_lock);
366 root_mnt = current->fs->rootmnt;
367 root = current->fs->root;
370 while ((mnt != mnt->mnt_parent) && (mnt != root_mnt)) {
371 point = mnt->mnt_mountpoint;
372 mnt = mnt->mnt_parent;
375 ret = (mnt == root_mnt) && is_subdir(point, root);
377 spin_unlock(&dcache_lock);
382 static int show_vfsmnt(struct seq_file *m, void *v)
384 struct vfsmount *mnt = v;
386 static struct proc_fs_info {
392 { MS_RDONLY, MNT_RDONLY, "ro", "rw" },
393 { MS_SYNCHRONOUS, 0, ",sync", NULL },
394 { MS_DIRSYNC, 0, ",dirsync", NULL },
395 { MS_MANDLOCK, 0, ",mand", NULL },
396 { MS_TAGXID, 0, ",tagxid", NULL },
397 { MS_NOATIME, MNT_NOATIME, ",noatime", NULL },
398 { MS_NODIRATIME, MNT_NODIRATIME, ",nodiratime", NULL },
399 { 0, MNT_NOSUID, ",nosuid", NULL },
400 { 0, MNT_NODEV, ",nodev", NULL },
401 { 0, MNT_NOEXEC, ",noexec", NULL },
404 struct proc_fs_info *p;
405 unsigned long s_flags = mnt->mnt_sb->s_flags;
406 int mnt_flags = mnt->mnt_flags;
408 if (vx_flags(VXF_HIDE_MOUNT, 0))
410 if (!mnt_is_reachable(mnt))
413 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
414 mnt == current->fs->rootmnt) {
415 seq_puts(m, "/dev/root / ");
417 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
419 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
422 mangle(m, mnt->mnt_sb->s_type->name);
424 for (p = fs_info; (p->s_flag | p->mnt_flag) ; p++) {
425 if ((s_flags & p->s_flag) || (mnt_flags & p->mnt_flag)) {
427 seq_puts(m, p->set_str);
430 seq_puts(m, p->unset_str);
433 if (mnt->mnt_flags & MNT_XID)
434 seq_printf(m, ",xid=%d", mnt->mnt_xid);
435 if (mnt->mnt_sb->s_op->show_options)
436 err = mnt->mnt_sb->s_op->show_options(m, mnt);
437 seq_puts(m, " 0 0\n");
441 struct seq_operations mounts_op = {
448 static int show_vfsstat(struct seq_file *m, void *v)
450 struct vfsmount *mnt = v;
453 if (vx_flags(VXF_HIDE_MOUNT, 0))
455 if (!mnt_is_reachable(mnt) && !vx_check(0, VX_WATCH))
458 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
459 mnt == current->fs->rootmnt) {
460 seq_puts(m, "device /dev/root mounted on / ");
463 if (mnt->mnt_devname) {
464 seq_puts(m, "device ");
465 mangle(m, mnt->mnt_devname);
467 seq_puts(m, "no device");
470 seq_puts(m, " mounted on ");
471 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
475 /* file system type */
476 seq_puts(m, "with fstype ");
477 mangle(m, mnt->mnt_sb->s_type->name);
479 /* optional statistics */
480 if (mnt->mnt_sb->s_op->show_stats) {
482 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
489 struct seq_operations mountstats_op = {
493 .show = show_vfsstat,
497 * may_umount_tree - check if a mount tree is busy
498 * @mnt: root of mount tree
500 * This is called to check if a tree of mounts has any
501 * open files, pwds, chroots or sub mounts that are
504 int may_umount_tree(struct vfsmount *mnt)
507 int minimum_refs = 0;
510 spin_lock(&vfsmount_lock);
511 for (p = mnt; p; p = next_mnt(p, mnt)) {
512 actual_refs += atomic_read(&p->mnt_count);
515 spin_unlock(&vfsmount_lock);
517 if (actual_refs > minimum_refs)
523 EXPORT_SYMBOL(may_umount_tree);
526 * may_umount - check if a mount point is busy
527 * @mnt: root of mount
529 * This is called to check if a mount point has any
530 * open files, pwds, chroots or sub mounts. If the
531 * mount has sub mounts this will return busy
532 * regardless of whether the sub mounts are busy.
534 * Doesn't take quota and stuff into account. IOW, in some cases it will
535 * give false negatives. The main reason why it's here is that we need
536 * a non-destructive way to look for easily umountable filesystems.
538 int may_umount(struct vfsmount *mnt)
541 spin_lock(&vfsmount_lock);
542 if (propagate_mount_busy(mnt, 2))
544 spin_unlock(&vfsmount_lock);
548 EXPORT_SYMBOL(may_umount);
550 void release_mounts(struct list_head *head)
552 struct vfsmount *mnt;
553 while (!list_empty(head)) {
554 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
555 list_del_init(&mnt->mnt_hash);
556 if (mnt->mnt_parent != mnt) {
557 struct dentry *dentry;
559 spin_lock(&vfsmount_lock);
560 dentry = mnt->mnt_mountpoint;
562 mnt->mnt_mountpoint = mnt->mnt_root;
563 mnt->mnt_parent = mnt;
564 spin_unlock(&vfsmount_lock);
572 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
576 for (p = mnt; p; p = next_mnt(p, mnt)) {
577 list_del(&p->mnt_hash);
578 list_add(&p->mnt_hash, kill);
582 propagate_umount(kill);
584 list_for_each_entry(p, kill, mnt_hash) {
585 list_del_init(&p->mnt_expire);
586 list_del_init(&p->mnt_list);
587 __touch_namespace(p->mnt_namespace);
588 p->mnt_namespace = NULL;
589 list_del_init(&p->mnt_child);
590 if (p->mnt_parent != p)
591 p->mnt_mountpoint->d_mounted--;
592 change_mnt_propagation(p, MS_PRIVATE);
596 static int do_umount(struct vfsmount *mnt, int flags)
598 struct super_block *sb = mnt->mnt_sb;
600 LIST_HEAD(umount_list);
602 retval = security_sb_umount(mnt, flags);
607 * Allow userspace to request a mountpoint be expired rather than
608 * unmounting unconditionally. Unmount only happens if:
609 * (1) the mark is already set (the mark is cleared by mntput())
610 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
612 if (flags & MNT_EXPIRE) {
613 if (mnt == current->fs->rootmnt ||
614 flags & (MNT_FORCE | MNT_DETACH))
617 if (atomic_read(&mnt->mnt_count) != 2)
620 if (!xchg(&mnt->mnt_expiry_mark, 1))
625 * If we may have to abort operations to get out of this
626 * mount, and they will themselves hold resources we must
627 * allow the fs to do things. In the Unix tradition of
628 * 'Gee thats tricky lets do it in userspace' the umount_begin
629 * might fail to complete on the first run through as other tasks
630 * must return, and the like. Thats for the mount program to worry
631 * about for the moment.
635 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
636 sb->s_op->umount_begin(sb);
640 * No sense to grab the lock for this test, but test itself looks
641 * somewhat bogus. Suggestions for better replacement?
642 * Ho-hum... In principle, we might treat that as umount + switch
643 * to rootfs. GC would eventually take care of the old vfsmount.
644 * Actually it makes sense, especially if rootfs would contain a
645 * /reboot - static binary that would close all descriptors and
646 * call reboot(9). Then init(8) could umount root and exec /reboot.
648 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
650 * Special case for "unmounting" root ...
651 * we just try to remount it readonly.
653 down_write(&sb->s_umount);
654 if (!(sb->s_flags & MS_RDONLY)) {
657 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
660 up_write(&sb->s_umount);
664 down_write(&namespace_sem);
665 spin_lock(&vfsmount_lock);
669 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
670 if (!list_empty(&mnt->mnt_list))
671 umount_tree(mnt, 1, &umount_list);
674 spin_unlock(&vfsmount_lock);
676 security_sb_umount_busy(mnt);
677 up_write(&namespace_sem);
678 release_mounts(&umount_list);
683 * Now umount can handle mount points as well as block devices.
684 * This is important for filesystems which use unnamed block devices.
686 * We now support a flag for forced unmount like the other 'big iron'
687 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
690 asmlinkage long sys_umount(char __user * name, int flags)
695 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
699 if (nd.dentry != nd.mnt->mnt_root)
701 if (!check_mnt(nd.mnt))
705 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
708 retval = do_umount(nd.mnt, flags);
710 path_release_on_umount(&nd);
715 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
718 * The 2.0 compatible umount. No flags.
720 asmlinkage long sys_oldumount(char __user * name)
722 return sys_umount(name, 0);
727 static int mount_is_safe(struct nameidata *nd)
729 if (vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
733 if (S_ISLNK(nd->dentry->d_inode->i_mode))
735 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
736 if (current->uid != nd->dentry->d_inode->i_uid)
739 if (vfs_permission(nd, MAY_WRITE))
745 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
750 if (d == NULL || d == d->d_parent)
756 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
759 struct vfsmount *res, *p, *q, *r, *s;
762 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
765 res = q = clone_mnt(mnt, dentry, flag);
768 q->mnt_mountpoint = mnt->mnt_mountpoint;
771 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
772 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
775 for (s = r; s; s = next_mnt(s, r)) {
776 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
777 s = skip_mnt_tree(s);
780 while (p != s->mnt_parent) {
786 nd.dentry = p->mnt_mountpoint;
787 q = clone_mnt(p, p->mnt_root, flag);
790 spin_lock(&vfsmount_lock);
791 list_add_tail(&q->mnt_list, &res->mnt_list);
793 spin_unlock(&vfsmount_lock);
799 LIST_HEAD(umount_list);
800 spin_lock(&vfsmount_lock);
801 umount_tree(res, 0, &umount_list);
802 spin_unlock(&vfsmount_lock);
803 release_mounts(&umount_list);
809 * @source_mnt : mount tree to be attached
810 * @nd : place the mount tree @source_mnt is attached
811 * @parent_nd : if non-null, detach the source_mnt from its parent and
812 * store the parent mount and mountpoint dentry.
813 * (done when source_mnt is moved)
815 * NOTE: in the table below explains the semantics when a source mount
816 * of a given type is attached to a destination mount of a given type.
817 * ---------------------------------------------------------------------------
818 * | BIND MOUNT OPERATION |
819 * |**************************************************************************
820 * | source-->| shared | private | slave | unbindable |
824 * |**************************************************************************
825 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
827 * |non-shared| shared (+) | private | slave (*) | invalid |
828 * ***************************************************************************
829 * A bind operation clones the source mount and mounts the clone on the
832 * (++) the cloned mount is propagated to all the mounts in the propagation
833 * tree of the destination mount and the cloned mount is added to
834 * the peer group of the source mount.
835 * (+) the cloned mount is created under the destination mount and is marked
836 * as shared. The cloned mount is added to the peer group of the source
838 * (+++) the mount is propagated to all the mounts in the propagation tree
839 * of the destination mount and the cloned mount is made slave
840 * of the same master as that of the source mount. The cloned mount
841 * is marked as 'shared and slave'.
842 * (*) the cloned mount is made a slave of the same master as that of the
845 * ---------------------------------------------------------------------------
846 * | MOVE MOUNT OPERATION |
847 * |**************************************************************************
848 * | source-->| shared | private | slave | unbindable |
852 * |**************************************************************************
853 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
855 * |non-shared| shared (+*) | private | slave (*) | unbindable |
856 * ***************************************************************************
858 * (+) the mount is moved to the destination. And is then propagated to
859 * all the mounts in the propagation tree of the destination mount.
860 * (+*) the mount is moved to the destination.
861 * (+++) the mount is moved to the destination and is then propagated to
862 * all the mounts belonging to the destination mount's propagation tree.
863 * the mount is marked as 'shared and slave'.
864 * (*) the mount continues to be a slave at the new location.
866 * if the source mount is a tree, the operations explained above is
867 * applied to each mount in the tree.
868 * Must be called without spinlocks held, since this function can sleep
871 static int attach_recursive_mnt(struct vfsmount *source_mnt,
872 struct nameidata *nd, struct nameidata *parent_nd)
874 LIST_HEAD(tree_list);
875 struct vfsmount *dest_mnt = nd->mnt;
876 struct dentry *dest_dentry = nd->dentry;
877 struct vfsmount *child, *p;
879 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
882 if (IS_MNT_SHARED(dest_mnt)) {
883 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
887 spin_lock(&vfsmount_lock);
889 detach_mnt(source_mnt, parent_nd);
890 attach_mnt(source_mnt, nd);
891 touch_namespace(current->namespace);
893 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
894 commit_tree(source_mnt);
897 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
898 list_del_init(&child->mnt_hash);
901 spin_unlock(&vfsmount_lock);
905 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
908 if (mnt->mnt_sb->s_flags & MS_NOUSER)
911 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
912 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
916 mutex_lock(&nd->dentry->d_inode->i_mutex);
917 if (IS_DEADDIR(nd->dentry->d_inode))
920 err = security_sb_check_sb(mnt, nd);
925 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
926 err = attach_recursive_mnt(mnt, nd, NULL);
928 mutex_unlock(&nd->dentry->d_inode->i_mutex);
930 security_sb_post_addmount(mnt, nd);
935 * recursively change the type of the mountpoint.
937 static int do_change_type(struct nameidata *nd, int flag)
939 struct vfsmount *m, *mnt = nd->mnt;
940 int recurse = flag & MS_REC;
941 int type = flag & ~MS_REC;
943 if (nd->dentry != nd->mnt->mnt_root)
946 down_write(&namespace_sem);
947 spin_lock(&vfsmount_lock);
948 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
949 change_mnt_propagation(m, type);
950 spin_unlock(&vfsmount_lock);
951 up_write(&namespace_sem);
958 static int do_loopback(struct nameidata *nd, char *old_name, xid_t xid,
959 unsigned long flags, int mnt_flags)
961 struct nameidata old_nd;
962 struct vfsmount *mnt = NULL;
963 int err = mount_is_safe(nd);
964 int recurse = flags & MS_REC;
967 if (!old_name || !*old_name)
969 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
973 down_write(&namespace_sem);
975 if (IS_MNT_UNBINDABLE(old_nd.mnt))
978 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
983 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
985 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
990 mnt->mnt_flags = mnt_flags;
991 if (flags & MS_XID) {
993 mnt->mnt_flags |= MNT_XID;
996 err = graft_tree(mnt, nd);
998 LIST_HEAD(umount_list);
999 spin_lock(&vfsmount_lock);
1000 umount_tree(mnt, 0, &umount_list);
1001 spin_unlock(&vfsmount_lock);
1002 release_mounts(&umount_list);
1004 mnt->mnt_flags = mnt_flags;
1007 up_write(&namespace_sem);
1008 path_release(&old_nd);
1013 * change filesystem flags. dir should be a physical root of filesystem.
1014 * If you've mounted a non-root directory somewhere and want to do remount
1015 * on it - tough luck.
1017 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1018 void *data, xid_t xid)
1021 struct super_block *sb = nd->mnt->mnt_sb;
1023 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_REMOUNT))
1026 if (!check_mnt(nd->mnt))
1029 if (nd->dentry != nd->mnt->mnt_root)
1032 down_write(&sb->s_umount);
1033 err = do_remount_sb(sb, flags, data, 0);
1035 nd->mnt->mnt_flags = mnt_flags;
1036 up_write(&sb->s_umount);
1038 security_sb_post_remount(nd->mnt, flags, data);
1042 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1045 for (p = mnt; p; p = next_mnt(p, mnt)) {
1046 if (IS_MNT_UNBINDABLE(p))
1052 static int do_move_mount(struct nameidata *nd, char *old_name)
1054 struct nameidata old_nd, parent_nd;
1057 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1059 if (!old_name || !*old_name)
1061 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1065 down_write(&namespace_sem);
1066 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1069 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1073 mutex_lock(&nd->dentry->d_inode->i_mutex);
1074 if (IS_DEADDIR(nd->dentry->d_inode))
1077 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1081 if (old_nd.dentry != old_nd.mnt->mnt_root)
1084 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1087 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1088 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1091 * Don't move a mount residing in a shared parent.
1093 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1096 * Don't move a mount tree containing unbindable mounts to a destination
1097 * mount which is shared.
1099 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1102 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1103 if (p == old_nd.mnt)
1106 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1109 spin_lock(&vfsmount_lock);
1110 /* if the mount is moved, it should no longer be expire
1112 list_del_init(&old_nd.mnt->mnt_expire);
1113 spin_unlock(&vfsmount_lock);
1115 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1117 up_write(&namespace_sem);
1119 path_release(&parent_nd);
1120 path_release(&old_nd);
1125 * create a new mount for userspace and request it to be added into the
1128 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1129 int mnt_flags, char *name, void *data)
1131 struct vfsmount *mnt;
1133 if (!type || !memchr(type, 0, PAGE_SIZE))
1136 /* we need capabilities... */
1137 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1140 mnt = do_kern_mount(type, flags, name, data);
1142 return PTR_ERR(mnt);
1144 return do_add_mount(mnt, nd, mnt_flags, NULL);
1148 * add a mount into a namespace's mount tree
1149 * - provide the option of adding the new mount to an expiration list
1151 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1152 int mnt_flags, struct list_head *fslist)
1156 down_write(&namespace_sem);
1157 /* Something was mounted here while we slept */
1158 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1161 if (!check_mnt(nd->mnt))
1164 /* Refuse the same filesystem on the same mount point */
1166 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1167 nd->mnt->mnt_root == nd->dentry)
1171 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1174 newmnt->mnt_flags = mnt_flags;
1175 if ((err = graft_tree(newmnt, nd)))
1179 /* add to the specified expiration list */
1180 spin_lock(&vfsmount_lock);
1181 list_add_tail(&newmnt->mnt_expire, fslist);
1182 spin_unlock(&vfsmount_lock);
1184 up_write(&namespace_sem);
1188 up_write(&namespace_sem);
1193 EXPORT_SYMBOL_GPL(do_add_mount);
1195 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1196 struct list_head *umounts)
1198 spin_lock(&vfsmount_lock);
1201 * Check if mount is still attached, if not, let whoever holds it deal
1204 if (mnt->mnt_parent == mnt) {
1205 spin_unlock(&vfsmount_lock);
1210 * Check that it is still dead: the count should now be 2 - as
1211 * contributed by the vfsmount parent and the mntget above
1213 if (!propagate_mount_busy(mnt, 2)) {
1214 /* delete from the namespace */
1215 touch_namespace(mnt->mnt_namespace);
1216 list_del_init(&mnt->mnt_list);
1217 mnt->mnt_namespace = NULL;
1218 umount_tree(mnt, 1, umounts);
1219 spin_unlock(&vfsmount_lock);
1222 * Someone brought it back to life whilst we didn't have any
1223 * locks held so return it to the expiration list
1225 list_add_tail(&mnt->mnt_expire, mounts);
1226 spin_unlock(&vfsmount_lock);
1231 * process a list of expirable mountpoints with the intent of discarding any
1232 * mountpoints that aren't in use and haven't been touched since last we came
1235 void mark_mounts_for_expiry(struct list_head *mounts)
1237 struct namespace *namespace;
1238 struct vfsmount *mnt, *next;
1239 LIST_HEAD(graveyard);
1241 if (list_empty(mounts))
1244 spin_lock(&vfsmount_lock);
1246 /* extract from the expiration list every vfsmount that matches the
1247 * following criteria:
1248 * - only referenced by its parent vfsmount
1249 * - still marked for expiry (marked on the last call here; marks are
1250 * cleared by mntput())
1252 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1253 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1254 atomic_read(&mnt->mnt_count) != 1)
1258 list_move(&mnt->mnt_expire, &graveyard);
1262 * go through the vfsmounts we've just consigned to the graveyard to
1263 * - check that they're still dead
1264 * - delete the vfsmount from the appropriate namespace under lock
1265 * - dispose of the corpse
1267 while (!list_empty(&graveyard)) {
1269 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1270 list_del_init(&mnt->mnt_expire);
1272 /* don't do anything if the namespace is dead - all the
1273 * vfsmounts from it are going away anyway */
1274 namespace = mnt->mnt_namespace;
1275 if (!namespace || !namespace->root)
1277 get_namespace(namespace);
1279 spin_unlock(&vfsmount_lock);
1280 down_write(&namespace_sem);
1281 expire_mount(mnt, mounts, &umounts);
1282 up_write(&namespace_sem);
1283 release_mounts(&umounts);
1285 put_namespace(namespace);
1286 spin_lock(&vfsmount_lock);
1289 spin_unlock(&vfsmount_lock);
1292 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1295 * Some copy_from_user() implementations do not return the exact number of
1296 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1297 * Note that this function differs from copy_from_user() in that it will oops
1298 * on bad values of `to', rather than returning a short copy.
1300 static long exact_copy_from_user(void *to, const void __user * from,
1304 const char __user *f = from;
1307 if (!access_ok(VERIFY_READ, from, n))
1311 if (__get_user(c, f)) {
1322 int copy_mount_options(const void __user * data, unsigned long *where)
1332 if (!(page = __get_free_page(GFP_KERNEL)))
1335 /* We only care that *some* data at the address the user
1336 * gave us is valid. Just in case, we'll zero
1337 * the remainder of the page.
1339 /* copy_from_user cannot cross TASK_SIZE ! */
1340 size = TASK_SIZE - (unsigned long)data;
1341 if (size > PAGE_SIZE)
1344 i = size - exact_copy_from_user((void *)page, data, size);
1350 memset((char *)page + i, 0, PAGE_SIZE - i);
1356 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1357 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1359 * data is a (void *) that can point to any structure up to
1360 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1361 * information (or be NULL).
1363 * Pre-0.97 versions of mount() didn't have a flags word.
1364 * When the flags word was introduced its top half was required
1365 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1366 * Therefore, if this magic number is present, it carries no information
1367 * and must be discarded.
1369 long do_mount(char *dev_name, char *dir_name, char *type_page,
1370 unsigned long flags, void *data_page)
1372 struct nameidata nd;
1378 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1379 flags &= ~MS_MGC_MSK;
1381 /* Basic sanity checks */
1383 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1385 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1389 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1391 retval = vx_parse_xid(data_page, &xid, 1);
1393 mnt_flags |= MNT_XID;
1394 /* bind and re-mounts get xid flag */
1395 if (flags & (MS_BIND|MS_REMOUNT))
1399 /* Separate the per-mountpoint flags */
1400 if (flags & MS_RDONLY)
1401 mnt_flags |= MNT_RDONLY;
1402 if (flags & MS_NOSUID)
1403 mnt_flags |= MNT_NOSUID;
1404 if (flags & MS_NODEV)
1405 mnt_flags |= MNT_NODEV;
1406 if (flags & MS_NOEXEC)
1407 mnt_flags |= MNT_NOEXEC;
1408 if (flags & MS_NOATIME)
1409 mnt_flags |= MNT_NOATIME;
1410 if (flags & MS_NODIRATIME)
1411 mnt_flags |= MNT_NODIRATIME;
1413 if (!capable(CAP_SYS_ADMIN))
1414 mnt_flags |= MNT_NODEV;
1415 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1416 MS_NOATIME | MS_NODIRATIME);
1418 /* ... and get the mountpoint */
1419 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1423 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1427 if (flags & MS_REMOUNT)
1428 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1430 else if (flags & MS_BIND)
1431 retval = do_loopback(&nd, dev_name, xid, flags, mnt_flags);
1432 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1433 retval = do_change_type(&nd, flags);
1434 else if (flags & MS_MOVE)
1435 retval = do_move_mount(&nd, dev_name);
1437 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1438 dev_name, data_page);
1445 * Allocate a new namespace structure and populate it with contents
1446 * copied from the namespace of the passed in task structure.
1448 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1450 struct namespace *namespace = tsk->namespace;
1451 struct namespace *new_ns;
1452 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1453 struct vfsmount *p, *q;
1455 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1459 atomic_set(&new_ns->count, 1);
1460 INIT_LIST_HEAD(&new_ns->list);
1461 init_waitqueue_head(&new_ns->poll);
1464 down_write(&namespace_sem);
1465 /* First pass: copy the tree topology */
1466 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1467 CL_COPY_ALL | CL_EXPIRE);
1468 if (!new_ns->root) {
1469 up_write(&namespace_sem);
1473 spin_lock(&vfsmount_lock);
1474 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1475 spin_unlock(&vfsmount_lock);
1478 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1479 * as belonging to new namespace. We have already acquired a private
1480 * fs_struct, so tsk->fs->lock is not needed.
1482 p = namespace->root;
1485 q->mnt_namespace = new_ns;
1487 if (p == fs->rootmnt) {
1489 fs->rootmnt = mntget(q);
1491 if (p == fs->pwdmnt) {
1493 fs->pwdmnt = mntget(q);
1495 if (p == fs->altrootmnt) {
1497 fs->altrootmnt = mntget(q);
1500 p = next_mnt(p, namespace->root);
1501 q = next_mnt(q, new_ns->root);
1503 up_write(&namespace_sem);
1515 int copy_namespace(int flags, struct task_struct *tsk)
1517 struct namespace *namespace = tsk->namespace;
1518 struct namespace *new_ns;
1524 get_namespace(namespace);
1526 if (!(flags & CLONE_NEWNS))
1529 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT)) {
1534 new_ns = dup_namespace(tsk, tsk->fs);
1540 tsk->namespace = new_ns;
1543 put_namespace(namespace);
1547 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1548 char __user * type, unsigned long flags,
1552 unsigned long data_page;
1553 unsigned long type_page;
1554 unsigned long dev_page;
1557 retval = copy_mount_options(type, &type_page);
1561 dir_page = getname(dir_name);
1562 retval = PTR_ERR(dir_page);
1563 if (IS_ERR(dir_page))
1566 retval = copy_mount_options(dev_name, &dev_page);
1570 retval = copy_mount_options(data, &data_page);
1575 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1576 flags, (void *)data_page);
1578 free_page(data_page);
1581 free_page(dev_page);
1585 free_page(type_page);
1590 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1591 * It can block. Requires the big lock held.
1593 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1594 struct dentry *dentry)
1596 struct dentry *old_root;
1597 struct vfsmount *old_rootmnt;
1598 write_lock(&fs->lock);
1599 old_root = fs->root;
1600 old_rootmnt = fs->rootmnt;
1601 fs->rootmnt = mntget(mnt);
1602 fs->root = dget(dentry);
1603 write_unlock(&fs->lock);
1606 mntput(old_rootmnt);
1610 EXPORT_SYMBOL_GPL(set_fs_root);
1613 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1614 * It can block. Requires the big lock held.
1616 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1617 struct dentry *dentry)
1619 struct dentry *old_pwd;
1620 struct vfsmount *old_pwdmnt;
1622 write_lock(&fs->lock);
1624 old_pwdmnt = fs->pwdmnt;
1625 fs->pwdmnt = mntget(mnt);
1626 fs->pwd = dget(dentry);
1627 write_unlock(&fs->lock);
1635 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1637 struct task_struct *g, *p;
1638 struct fs_struct *fs;
1640 read_lock(&tasklist_lock);
1641 do_each_thread(g, p) {
1645 atomic_inc(&fs->count);
1647 if (fs->root == old_nd->dentry
1648 && fs->rootmnt == old_nd->mnt)
1649 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1650 if (fs->pwd == old_nd->dentry
1651 && fs->pwdmnt == old_nd->mnt)
1652 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1656 } while_each_thread(g, p);
1657 read_unlock(&tasklist_lock);
1661 * pivot_root Semantics:
1662 * Moves the root file system of the current process to the directory put_old,
1663 * makes new_root as the new root file system of the current process, and sets
1664 * root/cwd of all processes which had them on the current root to new_root.
1667 * The new_root and put_old must be directories, and must not be on the
1668 * same file system as the current process root. The put_old must be
1669 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1670 * pointed to by put_old must yield the same directory as new_root. No other
1671 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1673 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1674 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1675 * in this situation.
1678 * - we don't move root/cwd if they are not at the root (reason: if something
1679 * cared enough to change them, it's probably wrong to force them elsewhere)
1680 * - it's okay to pick a root that isn't the root of a file system, e.g.
1681 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1682 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1685 asmlinkage long sys_pivot_root(const char __user * new_root,
1686 const char __user * put_old)
1688 struct vfsmount *tmp;
1689 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1692 if (!capable(CAP_SYS_ADMIN))
1697 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1702 if (!check_mnt(new_nd.mnt))
1705 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1709 error = security_sb_pivotroot(&old_nd, &new_nd);
1711 path_release(&old_nd);
1715 read_lock(¤t->fs->lock);
1716 user_nd.mnt = mntget(current->fs->rootmnt);
1717 user_nd.dentry = dget(current->fs->root);
1718 read_unlock(¤t->fs->lock);
1719 down_write(&namespace_sem);
1720 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1722 if (IS_MNT_SHARED(old_nd.mnt) ||
1723 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1724 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1726 if (!check_mnt(user_nd.mnt))
1729 if (IS_DEADDIR(new_nd.dentry->d_inode))
1731 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1733 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1736 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1737 goto out2; /* loop, on the same file system */
1739 if (user_nd.mnt->mnt_root != user_nd.dentry)
1740 goto out2; /* not a mountpoint */
1741 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1742 goto out2; /* not attached */
1743 if (new_nd.mnt->mnt_root != new_nd.dentry)
1744 goto out2; /* not a mountpoint */
1745 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1746 goto out2; /* not attached */
1747 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1748 spin_lock(&vfsmount_lock);
1749 if (tmp != new_nd.mnt) {
1751 if (tmp->mnt_parent == tmp)
1752 goto out3; /* already mounted on put_old */
1753 if (tmp->mnt_parent == new_nd.mnt)
1755 tmp = tmp->mnt_parent;
1757 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1759 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1761 detach_mnt(new_nd.mnt, &parent_nd);
1762 detach_mnt(user_nd.mnt, &root_parent);
1763 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1764 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1765 touch_namespace(current->namespace);
1766 spin_unlock(&vfsmount_lock);
1767 chroot_fs_refs(&user_nd, &new_nd);
1768 security_sb_post_pivotroot(&user_nd, &new_nd);
1770 path_release(&root_parent);
1771 path_release(&parent_nd);
1773 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1774 up_write(&namespace_sem);
1775 path_release(&user_nd);
1776 path_release(&old_nd);
1778 path_release(&new_nd);
1783 spin_unlock(&vfsmount_lock);
1787 static void __init init_mount_tree(void)
1789 struct vfsmount *mnt;
1790 struct namespace *namespace;
1791 struct task_struct *g, *p;
1793 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1795 panic("Can't create rootfs");
1796 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1798 panic("Can't allocate initial namespace");
1799 atomic_set(&namespace->count, 1);
1800 INIT_LIST_HEAD(&namespace->list);
1801 init_waitqueue_head(&namespace->poll);
1802 namespace->event = 0;
1803 list_add(&mnt->mnt_list, &namespace->list);
1804 namespace->root = mnt;
1805 mnt->mnt_namespace = namespace;
1807 init_task.namespace = namespace;
1808 read_lock(&tasklist_lock);
1809 do_each_thread(g, p) {
1810 get_namespace(namespace);
1811 p->namespace = namespace;
1812 } while_each_thread(g, p);
1813 read_unlock(&tasklist_lock);
1815 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1816 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1819 void __init mnt_init(unsigned long mempages)
1821 struct list_head *d;
1822 unsigned int nr_hash;
1825 init_rwsem(&namespace_sem);
1827 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1828 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1830 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1832 if (!mount_hashtable)
1833 panic("Failed to allocate mount hash table\n");
1836 * Find the power-of-two list-heads that can fit into the allocation..
1837 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1840 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1844 } while ((nr_hash >> hash_bits) != 0);
1848 * Re-calculate the actual number of entries and the mask
1849 * from the number of bits we can fit.
1851 nr_hash = 1UL << hash_bits;
1852 hash_mask = nr_hash - 1;
1854 printk("Mount-cache hash table entries: %d\n", nr_hash);
1856 /* And initialize the newly allocated array */
1857 d = mount_hashtable;
1865 subsystem_register(&fs_subsys);
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);