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/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <linux/vs_base.h>
29 #include <linux/vs_context.h>
30 #include <linux/vs_tag.h>
31 #include <linux/vserver/space.h>
32 #include <linux/vserver/global.h>
33 #include <asm/uaccess.h>
34 #include <asm/unistd.h>
37 /* spinlock for vfsmount related operations, inplace of dcache_lock */
38 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42 static struct list_head *mount_hashtable __read_mostly;
43 static int hash_mask __read_mostly, hash_bits __read_mostly;
44 static struct kmem_cache *mnt_cache __read_mostly;
45 static struct rw_semaphore namespace_sem;
48 decl_subsys(fs, NULL, NULL);
49 EXPORT_SYMBOL_GPL(fs_subsys);
51 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
53 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
54 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
55 tmp = tmp + (tmp >> hash_bits);
56 return tmp & hash_mask;
59 struct vfsmount *alloc_vfsmnt(const char *name)
61 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
63 memset(mnt, 0, sizeof(struct vfsmount));
64 atomic_set(&mnt->mnt_count, 1);
65 INIT_LIST_HEAD(&mnt->mnt_hash);
66 INIT_LIST_HEAD(&mnt->mnt_child);
67 INIT_LIST_HEAD(&mnt->mnt_mounts);
68 INIT_LIST_HEAD(&mnt->mnt_list);
69 INIT_LIST_HEAD(&mnt->mnt_expire);
70 INIT_LIST_HEAD(&mnt->mnt_share);
71 INIT_LIST_HEAD(&mnt->mnt_slave_list);
72 INIT_LIST_HEAD(&mnt->mnt_slave);
74 int size = strlen(name) + 1;
75 char *newname = kmalloc(size, GFP_KERNEL);
77 memcpy(newname, name, size);
78 mnt->mnt_devname = newname;
85 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
88 mnt->mnt_root = dget(sb->s_root);
92 EXPORT_SYMBOL(simple_set_mnt);
94 void free_vfsmnt(struct vfsmount *mnt)
96 kfree(mnt->mnt_devname);
97 kmem_cache_free(mnt_cache, mnt);
101 * find the first or last mount at @dentry on vfsmount @mnt depending on
102 * @dir. If @dir is set return the first mount else return the last mount.
104 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
107 struct list_head *head = mount_hashtable + hash(mnt, dentry);
108 struct list_head *tmp = head;
109 struct vfsmount *p, *found = NULL;
112 tmp = dir ? tmp->next : tmp->prev;
116 p = list_entry(tmp, struct vfsmount, mnt_hash);
117 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
126 * lookup_mnt increments the ref count before returning
127 * the vfsmount struct.
129 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
131 struct vfsmount *child_mnt;
132 spin_lock(&vfsmount_lock);
133 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
135 spin_unlock(&vfsmount_lock);
139 static inline int check_mnt(struct vfsmount *mnt)
141 return mnt->mnt_ns == current->nsproxy->mnt_ns;
144 static void touch_mnt_namespace(struct mnt_namespace *ns)
148 wake_up_interruptible(&ns->poll);
152 static void __touch_mnt_namespace(struct mnt_namespace *ns)
154 if (ns && ns->event != event) {
156 wake_up_interruptible(&ns->poll);
160 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
162 old_nd->dentry = mnt->mnt_mountpoint;
163 old_nd->mnt = mnt->mnt_parent;
164 mnt->mnt_parent = mnt;
165 mnt->mnt_mountpoint = mnt->mnt_root;
166 list_del_init(&mnt->mnt_child);
167 list_del_init(&mnt->mnt_hash);
168 old_nd->dentry->d_mounted--;
171 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
172 struct vfsmount *child_mnt)
174 child_mnt->mnt_parent = mntget(mnt);
175 child_mnt->mnt_mountpoint = dget(dentry);
179 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
181 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
182 list_add_tail(&mnt->mnt_hash, mount_hashtable +
183 hash(nd->mnt, nd->dentry));
184 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
188 * the caller must hold vfsmount_lock
190 static void commit_tree(struct vfsmount *mnt)
192 struct vfsmount *parent = mnt->mnt_parent;
195 struct mnt_namespace *n = parent->mnt_ns;
197 BUG_ON(parent == mnt);
199 list_add_tail(&head, &mnt->mnt_list);
200 list_for_each_entry(m, &head, mnt_list)
202 list_splice(&head, n->list.prev);
204 list_add_tail(&mnt->mnt_hash, mount_hashtable +
205 hash(parent, mnt->mnt_mountpoint));
206 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
207 touch_mnt_namespace(n);
210 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
212 struct list_head *next = p->mnt_mounts.next;
213 if (next == &p->mnt_mounts) {
217 next = p->mnt_child.next;
218 if (next != &p->mnt_parent->mnt_mounts)
223 return list_entry(next, struct vfsmount, mnt_child);
226 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
228 struct list_head *prev = p->mnt_mounts.prev;
229 while (prev != &p->mnt_mounts) {
230 p = list_entry(prev, struct vfsmount, mnt_child);
231 prev = p->mnt_mounts.prev;
236 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
239 struct super_block *sb = old->mnt_sb;
240 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
243 mnt->mnt_flags = old->mnt_flags;
244 atomic_inc(&sb->s_active);
246 mnt->mnt_root = dget(root);
247 mnt->mnt_mountpoint = mnt->mnt_root;
248 mnt->mnt_parent = mnt;
249 mnt->mnt_tag = old->mnt_tag;
251 if (flag & CL_SLAVE) {
252 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
253 mnt->mnt_master = old;
254 CLEAR_MNT_SHARED(mnt);
256 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
257 list_add(&mnt->mnt_share, &old->mnt_share);
258 if (IS_MNT_SLAVE(old))
259 list_add(&mnt->mnt_slave, &old->mnt_slave);
260 mnt->mnt_master = old->mnt_master;
262 if (flag & CL_MAKE_SHARED)
265 /* stick the duplicate mount on the same expiry list
266 * as the original if that was on one */
267 if (flag & CL_EXPIRE) {
268 spin_lock(&vfsmount_lock);
269 if (!list_empty(&old->mnt_expire))
270 list_add(&mnt->mnt_expire, &old->mnt_expire);
271 spin_unlock(&vfsmount_lock);
277 static inline void __mntput(struct vfsmount *mnt)
279 struct super_block *sb = mnt->mnt_sb;
282 deactivate_super(sb);
285 void mntput_no_expire(struct vfsmount *mnt)
288 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
289 if (likely(!mnt->mnt_pinned)) {
290 spin_unlock(&vfsmount_lock);
294 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
296 spin_unlock(&vfsmount_lock);
297 acct_auto_close_mnt(mnt);
298 security_sb_umount_close(mnt);
303 EXPORT_SYMBOL(mntput_no_expire);
305 void mnt_pin(struct vfsmount *mnt)
307 spin_lock(&vfsmount_lock);
309 spin_unlock(&vfsmount_lock);
312 EXPORT_SYMBOL(mnt_pin);
314 void mnt_unpin(struct vfsmount *mnt)
316 spin_lock(&vfsmount_lock);
317 if (mnt->mnt_pinned) {
318 atomic_inc(&mnt->mnt_count);
321 spin_unlock(&vfsmount_lock);
324 EXPORT_SYMBOL(mnt_unpin);
327 static void *m_start(struct seq_file *m, loff_t *pos)
329 struct mnt_namespace *n = m->private;
333 down_read(&namespace_sem);
334 list_for_each(p, &n->list)
336 return list_entry(p, struct vfsmount, mnt_list);
340 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
342 struct mnt_namespace *n = m->private;
343 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
345 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
348 static void m_stop(struct seq_file *m, void *v)
350 up_read(&namespace_sem);
353 static inline void mangle(struct seq_file *m, const char *s)
355 seq_escape(m, s, " \t\n\\");
358 static int mnt_is_reachable(struct vfsmount *mnt)
360 struct vfsmount *root_mnt;
361 struct dentry *root, *point;
364 if (mnt == mnt->mnt_ns->root)
367 spin_lock(&vfsmount_lock);
368 root_mnt = current->fs->rootmnt;
369 root = current->fs->root;
372 while ((mnt != mnt->mnt_parent) && (mnt != root_mnt)) {
373 point = mnt->mnt_mountpoint;
374 mnt = mnt->mnt_parent;
377 ret = (mnt == root_mnt) && is_subdir(point, root);
379 spin_unlock(&vfsmount_lock);
384 static int show_vfsmnt(struct seq_file *m, void *v)
386 struct vfsmount *mnt = v;
388 static struct proc_fs_info {
394 { MS_RDONLY, MNT_RDONLY, "ro", "rw" },
395 { MS_SYNCHRONOUS, 0, ",sync", NULL },
396 { MS_DIRSYNC, 0, ",dirsync", NULL },
397 { MS_MANDLOCK, 0, ",mand", NULL },
398 { MS_TAGGED, 0, ",tag", NULL },
399 { MS_NOATIME, MNT_NOATIME, ",noatime", NULL },
400 { MS_NODIRATIME, MNT_NODIRATIME, ",nodiratime", NULL },
401 { MS_RELATIME, MNT_RELATIME, ",relatime", NULL },
402 { 0, MNT_NOSUID, ",nosuid", NULL },
403 { 0, MNT_NODEV, ",nodev", NULL },
404 { 0, MNT_NOEXEC, ",noexec", NULL },
407 struct proc_fs_info *p;
408 unsigned long s_flags = mnt->mnt_sb->s_flags;
409 int mnt_flags = mnt->mnt_flags;
411 if (vx_flags(VXF_HIDE_MOUNT, 0))
413 if (!mnt_is_reachable(mnt) && !vx_check(0, VS_WATCH_P))
416 if (!vx_check(0, VS_ADMIN|VS_WATCH) &&
417 mnt == current->fs->rootmnt) {
418 seq_puts(m, "/dev/root / ");
420 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
422 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
425 mangle(m, mnt->mnt_sb->s_type->name);
427 for (p = fs_info; (p->s_flag | p->mnt_flag) ; p++) {
428 if ((s_flags & p->s_flag) || (mnt_flags & p->mnt_flag)) {
430 seq_puts(m, p->set_str);
433 seq_puts(m, p->unset_str);
436 if (mnt->mnt_flags & MNT_TAGID)
437 seq_printf(m, ",tag=%d", mnt->mnt_tag);
438 if (mnt->mnt_sb->s_op->show_options)
439 err = mnt->mnt_sb->s_op->show_options(m, mnt);
440 seq_puts(m, " 0 0\n");
444 struct seq_operations mounts_op = {
451 static int show_vfsstat(struct seq_file *m, void *v)
453 struct vfsmount *mnt = v;
456 if (vx_flags(VXF_HIDE_MOUNT, 0))
458 if (!mnt_is_reachable(mnt) && !vx_check(0, VS_WATCH_P))
461 if (!vx_check(0, VS_ADMIN|VS_WATCH) &&
462 mnt == current->fs->rootmnt) {
463 seq_puts(m, "device /dev/root mounted on / ");
466 if (mnt->mnt_devname) {
467 seq_puts(m, "device ");
468 mangle(m, mnt->mnt_devname);
470 seq_puts(m, "no device");
473 seq_puts(m, " mounted on ");
474 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
478 /* file system type */
479 seq_puts(m, "with fstype ");
480 mangle(m, mnt->mnt_sb->s_type->name);
482 /* optional statistics */
483 if (mnt->mnt_sb->s_op->show_stats) {
485 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
492 struct seq_operations mountstats_op = {
496 .show = show_vfsstat,
500 * may_umount_tree - check if a mount tree is busy
501 * @mnt: root of mount tree
503 * This is called to check if a tree of mounts has any
504 * open files, pwds, chroots or sub mounts that are
507 int may_umount_tree(struct vfsmount *mnt)
510 int minimum_refs = 0;
513 spin_lock(&vfsmount_lock);
514 for (p = mnt; p; p = next_mnt(p, mnt)) {
515 actual_refs += atomic_read(&p->mnt_count);
518 spin_unlock(&vfsmount_lock);
520 if (actual_refs > minimum_refs)
526 EXPORT_SYMBOL(may_umount_tree);
529 * may_umount - check if a mount point is busy
530 * @mnt: root of mount
532 * This is called to check if a mount point has any
533 * open files, pwds, chroots or sub mounts. If the
534 * mount has sub mounts this will return busy
535 * regardless of whether the sub mounts are busy.
537 * Doesn't take quota and stuff into account. IOW, in some cases it will
538 * give false negatives. The main reason why it's here is that we need
539 * a non-destructive way to look for easily umountable filesystems.
541 int may_umount(struct vfsmount *mnt)
544 spin_lock(&vfsmount_lock);
545 if (propagate_mount_busy(mnt, 2))
547 spin_unlock(&vfsmount_lock);
551 EXPORT_SYMBOL(may_umount);
553 void release_mounts(struct list_head *head)
555 struct vfsmount *mnt;
556 while (!list_empty(head)) {
557 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
558 list_del_init(&mnt->mnt_hash);
559 if (mnt->mnt_parent != mnt) {
560 struct dentry *dentry;
562 spin_lock(&vfsmount_lock);
563 dentry = mnt->mnt_mountpoint;
565 mnt->mnt_mountpoint = mnt->mnt_root;
566 mnt->mnt_parent = mnt;
567 spin_unlock(&vfsmount_lock);
575 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
579 for (p = mnt; p; p = next_mnt(p, mnt))
580 list_move(&p->mnt_hash, kill);
583 propagate_umount(kill);
585 list_for_each_entry(p, kill, mnt_hash) {
586 list_del_init(&p->mnt_expire);
587 list_del_init(&p->mnt_list);
588 __touch_mnt_namespace(p->mnt_ns);
590 list_del_init(&p->mnt_child);
591 if (p->mnt_parent != p)
592 p->mnt_mountpoint->d_mounted--;
593 change_mnt_propagation(p, MS_PRIVATE);
597 static int do_umount(struct vfsmount *mnt, int flags)
599 struct super_block *sb = mnt->mnt_sb;
601 LIST_HEAD(umount_list);
603 retval = security_sb_umount(mnt, flags);
608 * Allow userspace to request a mountpoint be expired rather than
609 * unmounting unconditionally. Unmount only happens if:
610 * (1) the mark is already set (the mark is cleared by mntput())
611 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
613 if (flags & MNT_EXPIRE) {
614 if (mnt == current->fs->rootmnt ||
615 flags & (MNT_FORCE | MNT_DETACH))
618 if (atomic_read(&mnt->mnt_count) != 2)
621 if (!xchg(&mnt->mnt_expiry_mark, 1))
626 * If we may have to abort operations to get out of this
627 * mount, and they will themselves hold resources we must
628 * allow the fs to do things. In the Unix tradition of
629 * 'Gee thats tricky lets do it in userspace' the umount_begin
630 * might fail to complete on the first run through as other tasks
631 * must return, and the like. Thats for the mount program to worry
632 * about for the moment.
636 if (sb->s_op->umount_begin)
637 sb->s_op->umount_begin(mnt, flags);
641 * No sense to grab the lock for this test, but test itself looks
642 * somewhat bogus. Suggestions for better replacement?
643 * Ho-hum... In principle, we might treat that as umount + switch
644 * to rootfs. GC would eventually take care of the old vfsmount.
645 * Actually it makes sense, especially if rootfs would contain a
646 * /reboot - static binary that would close all descriptors and
647 * call reboot(9). Then init(8) could umount root and exec /reboot.
649 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
651 * Special case for "unmounting" root ...
652 * we just try to remount it readonly.
654 down_write(&sb->s_umount);
655 if (!(sb->s_flags & MS_RDONLY)) {
658 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
661 up_write(&sb->s_umount);
665 down_write(&namespace_sem);
666 spin_lock(&vfsmount_lock);
670 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
671 if (!list_empty(&mnt->mnt_list))
672 umount_tree(mnt, 1, &umount_list);
675 spin_unlock(&vfsmount_lock);
677 security_sb_umount_busy(mnt);
678 up_write(&namespace_sem);
679 release_mounts(&umount_list);
684 * Now umount can handle mount points as well as block devices.
685 * This is important for filesystems which use unnamed block devices.
687 * We now support a flag for forced unmount like the other 'big iron'
688 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
691 asmlinkage long sys_umount(char __user * name, int flags)
696 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
700 if (nd.dentry != nd.mnt->mnt_root)
702 if (!check_mnt(nd.mnt))
706 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
709 retval = do_umount(nd.mnt, flags);
711 path_release_on_umount(&nd);
716 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
719 * The 2.0 compatible umount. No flags.
721 asmlinkage long sys_oldumount(char __user * name)
723 return sys_umount(name, 0);
728 static int mount_is_safe(struct nameidata *nd)
730 if (vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
734 if (S_ISLNK(nd->dentry->d_inode->i_mode))
736 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
737 if (current->uid != nd->dentry->d_inode->i_uid)
740 if (vfs_permission(nd, MAY_WRITE))
746 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
751 if (d == NULL || d == d->d_parent)
757 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
760 struct vfsmount *res, *p, *q, *r, *s;
763 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
766 res = q = clone_mnt(mnt, dentry, flag);
769 q->mnt_mountpoint = mnt->mnt_mountpoint;
772 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
773 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
776 for (s = r; s; s = next_mnt(s, r)) {
777 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
778 s = skip_mnt_tree(s);
781 while (p != s->mnt_parent) {
787 nd.dentry = p->mnt_mountpoint;
788 q = clone_mnt(p, p->mnt_root, flag);
791 spin_lock(&vfsmount_lock);
792 list_add_tail(&q->mnt_list, &res->mnt_list);
794 spin_unlock(&vfsmount_lock);
800 LIST_HEAD(umount_list);
801 spin_lock(&vfsmount_lock);
802 umount_tree(res, 0, &umount_list);
803 spin_unlock(&vfsmount_lock);
804 release_mounts(&umount_list);
810 * @source_mnt : mount tree to be attached
811 * @nd : place the mount tree @source_mnt is attached
812 * @parent_nd : if non-null, detach the source_mnt from its parent and
813 * store the parent mount and mountpoint dentry.
814 * (done when source_mnt is moved)
816 * NOTE: in the table below explains the semantics when a source mount
817 * of a given type is attached to a destination mount of a given type.
818 * ---------------------------------------------------------------------------
819 * | BIND MOUNT OPERATION |
820 * |**************************************************************************
821 * | source-->| shared | private | slave | unbindable |
825 * |**************************************************************************
826 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
828 * |non-shared| shared (+) | private | slave (*) | invalid |
829 * ***************************************************************************
830 * A bind operation clones the source mount and mounts the clone on the
833 * (++) the cloned mount is propagated to all the mounts in the propagation
834 * tree of the destination mount and the cloned mount is added to
835 * the peer group of the source mount.
836 * (+) the cloned mount is created under the destination mount and is marked
837 * as shared. The cloned mount is added to the peer group of the source
839 * (+++) the mount is propagated to all the mounts in the propagation tree
840 * of the destination mount and the cloned mount is made slave
841 * of the same master as that of the source mount. The cloned mount
842 * is marked as 'shared and slave'.
843 * (*) the cloned mount is made a slave of the same master as that of the
846 * ---------------------------------------------------------------------------
847 * | MOVE MOUNT OPERATION |
848 * |**************************************************************************
849 * | source-->| shared | private | slave | unbindable |
853 * |**************************************************************************
854 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
856 * |non-shared| shared (+*) | private | slave (*) | unbindable |
857 * ***************************************************************************
859 * (+) the mount is moved to the destination. And is then propagated to
860 * all the mounts in the propagation tree of the destination mount.
861 * (+*) the mount is moved to the destination.
862 * (+++) the mount is moved to the destination and is then propagated to
863 * all the mounts belonging to the destination mount's propagation tree.
864 * the mount is marked as 'shared and slave'.
865 * (*) the mount continues to be a slave at the new location.
867 * if the source mount is a tree, the operations explained above is
868 * applied to each mount in the tree.
869 * Must be called without spinlocks held, since this function can sleep
872 static int attach_recursive_mnt(struct vfsmount *source_mnt,
873 struct nameidata *nd, struct nameidata *parent_nd)
875 LIST_HEAD(tree_list);
876 struct vfsmount *dest_mnt = nd->mnt;
877 struct dentry *dest_dentry = nd->dentry;
878 struct vfsmount *child, *p;
880 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
883 if (IS_MNT_SHARED(dest_mnt)) {
884 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
888 spin_lock(&vfsmount_lock);
890 detach_mnt(source_mnt, parent_nd);
891 attach_mnt(source_mnt, nd);
892 touch_mnt_namespace(current->nsproxy->mnt_ns);
894 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
895 commit_tree(source_mnt);
898 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
899 list_del_init(&child->mnt_hash);
902 spin_unlock(&vfsmount_lock);
906 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
909 if (mnt->mnt_sb->s_flags & MS_NOUSER)
912 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
913 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
917 mutex_lock(&nd->dentry->d_inode->i_mutex);
918 if (IS_DEADDIR(nd->dentry->d_inode))
921 err = security_sb_check_sb(mnt, nd);
926 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
927 err = attach_recursive_mnt(mnt, nd, NULL);
929 mutex_unlock(&nd->dentry->d_inode->i_mutex);
931 security_sb_post_addmount(mnt, nd);
936 * recursively change the type of the mountpoint.
938 static int do_change_type(struct nameidata *nd, int flag)
940 struct vfsmount *m, *mnt = nd->mnt;
941 int recurse = flag & MS_REC;
942 int type = flag & ~MS_REC;
944 if (nd->dentry != nd->mnt->mnt_root)
947 down_write(&namespace_sem);
948 spin_lock(&vfsmount_lock);
949 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
950 change_mnt_propagation(m, type);
951 spin_unlock(&vfsmount_lock);
952 up_write(&namespace_sem);
959 static int do_loopback(struct nameidata *nd, char *old_name, tag_t tag,
960 unsigned long flags, int mnt_flags)
962 struct nameidata old_nd;
963 struct vfsmount *mnt = NULL;
964 int err = mount_is_safe(nd);
965 int recurse = flags & MS_REC;
968 if (!old_name || !*old_name)
970 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
974 down_write(&namespace_sem);
976 if (IS_MNT_UNBINDABLE(old_nd.mnt))
979 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
984 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
986 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
991 mnt->mnt_flags = mnt_flags;
992 if (flags & MS_TAGID) {
994 mnt->mnt_flags |= MNT_TAGID;
997 err = graft_tree(mnt, nd);
999 LIST_HEAD(umount_list);
1000 spin_lock(&vfsmount_lock);
1001 umount_tree(mnt, 0, &umount_list);
1002 spin_unlock(&vfsmount_lock);
1003 release_mounts(&umount_list);
1005 mnt->mnt_flags = mnt_flags;
1008 up_write(&namespace_sem);
1009 path_release(&old_nd);
1014 * change filesystem flags. dir should be a physical root of filesystem.
1015 * If you've mounted a non-root directory somewhere and want to do remount
1016 * on it - tough luck.
1018 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1019 void *data, xid_t xid)
1022 struct super_block *sb = nd->mnt->mnt_sb;
1024 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_REMOUNT))
1027 if (!check_mnt(nd->mnt))
1030 if (nd->dentry != nd->mnt->mnt_root)
1033 down_write(&sb->s_umount);
1034 err = do_remount_sb(sb, flags, data, 0);
1036 nd->mnt->mnt_flags = mnt_flags;
1037 up_write(&sb->s_umount);
1039 security_sb_post_remount(nd->mnt, flags, data);
1043 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1046 for (p = mnt; p; p = next_mnt(p, mnt)) {
1047 if (IS_MNT_UNBINDABLE(p))
1053 static int do_move_mount(struct nameidata *nd, char *old_name)
1055 struct nameidata old_nd, parent_nd;
1058 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1060 if (!old_name || !*old_name)
1062 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1066 down_write(&namespace_sem);
1067 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1070 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1074 mutex_lock(&nd->dentry->d_inode->i_mutex);
1075 if (IS_DEADDIR(nd->dentry->d_inode))
1078 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1082 if (old_nd.dentry != old_nd.mnt->mnt_root)
1085 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1088 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1089 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1092 * Don't move a mount residing in a shared parent.
1094 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1097 * Don't move a mount tree containing unbindable mounts to a destination
1098 * mount which is shared.
1100 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1103 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1104 if (p == old_nd.mnt)
1107 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1110 spin_lock(&vfsmount_lock);
1111 /* if the mount is moved, it should no longer be expire
1113 list_del_init(&old_nd.mnt->mnt_expire);
1114 spin_unlock(&vfsmount_lock);
1116 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1118 up_write(&namespace_sem);
1120 path_release(&parent_nd);
1121 path_release(&old_nd);
1126 * create a new mount for userspace and request it to be added into the
1129 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1130 int mnt_flags, char *name, void *data)
1132 struct vfsmount *mnt;
1134 if (!type || !memchr(type, 0, PAGE_SIZE))
1137 /* we need capabilities... */
1138 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1141 mnt = do_kern_mount(type, flags, name, data);
1143 return PTR_ERR(mnt);
1145 return do_add_mount(mnt, nd, mnt_flags, NULL);
1149 * add a mount into a namespace's mount tree
1150 * - provide the option of adding the new mount to an expiration list
1152 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1153 int mnt_flags, struct list_head *fslist)
1157 down_write(&namespace_sem);
1158 /* Something was mounted here while we slept */
1159 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1162 if (!check_mnt(nd->mnt))
1165 /* Refuse the same filesystem on the same mount point */
1167 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1168 nd->mnt->mnt_root == nd->dentry)
1172 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1175 newmnt->mnt_flags = mnt_flags;
1176 if ((err = graft_tree(newmnt, nd)))
1180 /* add to the specified expiration list */
1181 spin_lock(&vfsmount_lock);
1182 list_add_tail(&newmnt->mnt_expire, fslist);
1183 spin_unlock(&vfsmount_lock);
1185 up_write(&namespace_sem);
1189 up_write(&namespace_sem);
1194 EXPORT_SYMBOL_GPL(do_add_mount);
1196 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1197 struct list_head *umounts)
1199 spin_lock(&vfsmount_lock);
1202 * Check if mount is still attached, if not, let whoever holds it deal
1205 if (mnt->mnt_parent == mnt) {
1206 spin_unlock(&vfsmount_lock);
1211 * Check that it is still dead: the count should now be 2 - as
1212 * contributed by the vfsmount parent and the mntget above
1214 if (!propagate_mount_busy(mnt, 2)) {
1215 /* delete from the namespace */
1216 touch_mnt_namespace(mnt->mnt_ns);
1217 list_del_init(&mnt->mnt_list);
1219 umount_tree(mnt, 1, umounts);
1220 spin_unlock(&vfsmount_lock);
1223 * Someone brought it back to life whilst we didn't have any
1224 * locks held so return it to the expiration list
1226 list_add_tail(&mnt->mnt_expire, mounts);
1227 spin_unlock(&vfsmount_lock);
1232 * go through the vfsmounts we've just consigned to the graveyard to
1233 * - check that they're still dead
1234 * - delete the vfsmount from the appropriate namespace under lock
1235 * - dispose of the corpse
1237 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1239 struct mnt_namespace *ns;
1240 struct vfsmount *mnt;
1242 while (!list_empty(graveyard)) {
1244 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1245 list_del_init(&mnt->mnt_expire);
1247 /* don't do anything if the namespace is dead - all the
1248 * vfsmounts from it are going away anyway */
1250 if (!ns || !ns->root)
1254 spin_unlock(&vfsmount_lock);
1255 down_write(&namespace_sem);
1256 expire_mount(mnt, mounts, &umounts);
1257 up_write(&namespace_sem);
1258 release_mounts(&umounts);
1261 spin_lock(&vfsmount_lock);
1266 * process a list of expirable mountpoints with the intent of discarding any
1267 * mountpoints that aren't in use and haven't been touched since last we came
1270 void mark_mounts_for_expiry(struct list_head *mounts)
1272 struct vfsmount *mnt, *next;
1273 LIST_HEAD(graveyard);
1275 if (list_empty(mounts))
1278 spin_lock(&vfsmount_lock);
1280 /* extract from the expiration list every vfsmount that matches the
1281 * following criteria:
1282 * - only referenced by its parent vfsmount
1283 * - still marked for expiry (marked on the last call here; marks are
1284 * cleared by mntput())
1286 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1287 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1288 atomic_read(&mnt->mnt_count) != 1)
1292 list_move(&mnt->mnt_expire, &graveyard);
1295 expire_mount_list(&graveyard, mounts);
1297 spin_unlock(&vfsmount_lock);
1300 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1303 * Ripoff of 'select_parent()'
1305 * search the list of submounts for a given mountpoint, and move any
1306 * shrinkable submounts to the 'graveyard' list.
1308 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1310 struct vfsmount *this_parent = parent;
1311 struct list_head *next;
1315 next = this_parent->mnt_mounts.next;
1317 while (next != &this_parent->mnt_mounts) {
1318 struct list_head *tmp = next;
1319 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1322 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1325 * Descend a level if the d_mounts list is non-empty.
1327 if (!list_empty(&mnt->mnt_mounts)) {
1332 if (!propagate_mount_busy(mnt, 1)) {
1334 list_move_tail(&mnt->mnt_expire, graveyard);
1339 * All done at this level ... ascend and resume the search
1341 if (this_parent != parent) {
1342 next = this_parent->mnt_child.next;
1343 this_parent = this_parent->mnt_parent;
1350 * process a list of expirable mountpoints with the intent of discarding any
1351 * submounts of a specific parent mountpoint
1353 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1355 LIST_HEAD(graveyard);
1358 spin_lock(&vfsmount_lock);
1360 /* extract submounts of 'mountpoint' from the expiration list */
1361 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1362 expire_mount_list(&graveyard, mounts);
1364 spin_unlock(&vfsmount_lock);
1367 EXPORT_SYMBOL_GPL(shrink_submounts);
1370 * Some copy_from_user() implementations do not return the exact number of
1371 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1372 * Note that this function differs from copy_from_user() in that it will oops
1373 * on bad values of `to', rather than returning a short copy.
1375 static long exact_copy_from_user(void *to, const void __user * from,
1379 const char __user *f = from;
1382 if (!access_ok(VERIFY_READ, from, n))
1386 if (__get_user(c, f)) {
1397 int copy_mount_options(const void __user * data, unsigned long *where)
1407 if (!(page = __get_free_page(GFP_KERNEL)))
1410 /* We only care that *some* data at the address the user
1411 * gave us is valid. Just in case, we'll zero
1412 * the remainder of the page.
1414 /* copy_from_user cannot cross TASK_SIZE ! */
1415 size = TASK_SIZE - (unsigned long)data;
1416 if (size > PAGE_SIZE)
1419 i = size - exact_copy_from_user((void *)page, data, size);
1425 memset((char *)page + i, 0, PAGE_SIZE - i);
1431 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1432 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1434 * data is a (void *) that can point to any structure up to
1435 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1436 * information (or be NULL).
1438 * Pre-0.97 versions of mount() didn't have a flags word.
1439 * When the flags word was introduced its top half was required
1440 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1441 * Therefore, if this magic number is present, it carries no information
1442 * and must be discarded.
1444 long do_mount(char *dev_name, char *dir_name, char *type_page,
1445 unsigned long flags, void *data_page)
1447 struct nameidata nd;
1453 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1454 flags &= ~MS_MGC_MSK;
1456 /* Basic sanity checks */
1458 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1460 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1464 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1466 #ifdef CONFIG_PROPAGATE
1467 retval = dx_parse_tag(data_page, &tag, 1);
1469 mnt_flags |= MNT_TAGID;
1470 /* bind and re-mounts get the tag flag */
1471 if (flags & (MS_BIND|MS_REMOUNT))
1476 /* Separate the per-mountpoint flags */
1477 if (flags & MS_RDONLY)
1478 mnt_flags |= MNT_RDONLY;
1479 if (flags & MS_NOSUID)
1480 mnt_flags |= MNT_NOSUID;
1481 if (flags & MS_NODEV)
1482 mnt_flags |= MNT_NODEV;
1483 if (flags & MS_NOEXEC)
1484 mnt_flags |= MNT_NOEXEC;
1485 if (flags & MS_NOATIME)
1486 mnt_flags |= MNT_NOATIME;
1487 if (flags & MS_NODIRATIME)
1488 mnt_flags |= MNT_NODIRATIME;
1489 if (flags & MS_RELATIME)
1490 mnt_flags |= MNT_RELATIME;
1492 if (!capable(CAP_SYS_ADMIN))
1493 mnt_flags |= MNT_NODEV;
1494 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1495 MS_NOATIME | MS_NODIRATIME | MS_RELATIME);
1497 /* ... and get the mountpoint */
1498 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1502 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1506 if (flags & MS_REMOUNT)
1507 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1509 else if (flags & MS_BIND)
1510 retval = do_loopback(&nd, dev_name, tag, flags, mnt_flags);
1511 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1512 retval = do_change_type(&nd, flags);
1513 else if (flags & MS_MOVE)
1514 retval = do_move_mount(&nd, dev_name);
1516 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1517 dev_name, data_page);
1524 * Allocate a new namespace structure and populate it with contents
1525 * copied from the namespace of the passed in task structure.
1527 struct mnt_namespace *dup_mnt_ns(struct task_struct *tsk,
1528 struct fs_struct *fs)
1530 struct mnt_namespace *mnt_ns = tsk->nsproxy->mnt_ns;
1531 struct mnt_namespace *new_ns;
1532 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1533 struct vfsmount *p, *q;
1535 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1539 atomic_set(&new_ns->count, 1);
1540 INIT_LIST_HEAD(&new_ns->list);
1541 init_waitqueue_head(&new_ns->poll);
1544 down_write(&namespace_sem);
1545 /* First pass: copy the tree topology */
1546 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1547 CL_COPY_ALL | CL_EXPIRE);
1548 if (!new_ns->root) {
1549 up_write(&namespace_sem);
1553 spin_lock(&vfsmount_lock);
1554 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1555 spin_unlock(&vfsmount_lock);
1558 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1559 * as belonging to new namespace. We have already acquired a private
1560 * fs_struct, so tsk->fs->lock is not needed.
1567 if (p == fs->rootmnt) {
1569 fs->rootmnt = mntget(q);
1571 if (p == fs->pwdmnt) {
1573 fs->pwdmnt = mntget(q);
1575 if (p == fs->altrootmnt) {
1577 fs->altrootmnt = mntget(q);
1580 p = next_mnt(p, mnt_ns->root);
1581 q = next_mnt(q, new_ns->root);
1583 up_write(&namespace_sem);
1584 atomic_inc(&vs_global_mnt_ns);
1596 int copy_mnt_ns(int flags, struct task_struct *tsk)
1598 struct mnt_namespace *ns = tsk->nsproxy->mnt_ns;
1599 struct mnt_namespace *new_ns;
1607 if (!(flags & CLONE_NEWNS))
1610 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT)) {
1615 new_ns = dup_mnt_ns(tsk, tsk->fs);
1621 tsk->nsproxy->mnt_ns = new_ns;
1628 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1629 char __user * type, unsigned long flags,
1633 unsigned long data_page;
1634 unsigned long type_page;
1635 unsigned long dev_page;
1638 retval = copy_mount_options(type, &type_page);
1642 dir_page = getname(dir_name);
1643 retval = PTR_ERR(dir_page);
1644 if (IS_ERR(dir_page))
1647 retval = copy_mount_options(dev_name, &dev_page);
1651 retval = copy_mount_options(data, &data_page);
1656 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1657 flags, (void *)data_page);
1659 free_page(data_page);
1662 free_page(dev_page);
1666 free_page(type_page);
1671 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1672 * It can block. Requires the big lock held.
1674 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1675 struct dentry *dentry)
1677 struct dentry *old_root;
1678 struct vfsmount *old_rootmnt;
1679 write_lock(&fs->lock);
1680 old_root = fs->root;
1681 old_rootmnt = fs->rootmnt;
1682 fs->rootmnt = mntget(mnt);
1683 fs->root = dget(dentry);
1684 write_unlock(&fs->lock);
1687 mntput(old_rootmnt);
1692 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1693 * It can block. Requires the big lock held.
1695 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1696 struct dentry *dentry)
1698 struct dentry *old_pwd;
1699 struct vfsmount *old_pwdmnt;
1701 write_lock(&fs->lock);
1703 old_pwdmnt = fs->pwdmnt;
1704 fs->pwdmnt = mntget(mnt);
1705 fs->pwd = dget(dentry);
1706 write_unlock(&fs->lock);
1714 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1716 struct task_struct *g, *p;
1717 struct fs_struct *fs;
1719 read_lock(&tasklist_lock);
1720 do_each_thread(g, p) {
1724 atomic_inc(&fs->count);
1726 if (fs->root == old_nd->dentry
1727 && fs->rootmnt == old_nd->mnt)
1728 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1729 if (fs->pwd == old_nd->dentry
1730 && fs->pwdmnt == old_nd->mnt)
1731 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1735 } while_each_thread(g, p);
1736 read_unlock(&tasklist_lock);
1740 * pivot_root Semantics:
1741 * Moves the root file system of the current process to the directory put_old,
1742 * makes new_root as the new root file system of the current process, and sets
1743 * root/cwd of all processes which had them on the current root to new_root.
1746 * The new_root and put_old must be directories, and must not be on the
1747 * same file system as the current process root. The put_old must be
1748 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1749 * pointed to by put_old must yield the same directory as new_root. No other
1750 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1752 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1753 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1754 * in this situation.
1757 * - we don't move root/cwd if they are not at the root (reason: if something
1758 * cared enough to change them, it's probably wrong to force them elsewhere)
1759 * - it's okay to pick a root that isn't the root of a file system, e.g.
1760 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1761 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1764 asmlinkage long sys_pivot_root(const char __user * new_root,
1765 const char __user * put_old)
1767 struct vfsmount *tmp;
1768 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1771 if (!capable(CAP_SYS_ADMIN))
1776 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1781 if (!check_mnt(new_nd.mnt))
1784 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1788 error = security_sb_pivotroot(&old_nd, &new_nd);
1790 path_release(&old_nd);
1794 read_lock(¤t->fs->lock);
1795 user_nd.mnt = mntget(current->fs->rootmnt);
1796 user_nd.dentry = dget(current->fs->root);
1797 read_unlock(¤t->fs->lock);
1798 down_write(&namespace_sem);
1799 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1801 if (IS_MNT_SHARED(old_nd.mnt) ||
1802 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1803 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1805 if (!check_mnt(user_nd.mnt))
1808 if (IS_DEADDIR(new_nd.dentry->d_inode))
1810 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1812 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1815 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1816 goto out2; /* loop, on the same file system */
1818 if (user_nd.mnt->mnt_root != user_nd.dentry)
1819 goto out2; /* not a mountpoint */
1820 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1821 goto out2; /* not attached */
1822 if (new_nd.mnt->mnt_root != new_nd.dentry)
1823 goto out2; /* not a mountpoint */
1824 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1825 goto out2; /* not attached */
1826 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1827 spin_lock(&vfsmount_lock);
1828 if (tmp != new_nd.mnt) {
1830 if (tmp->mnt_parent == tmp)
1831 goto out3; /* already mounted on put_old */
1832 if (tmp->mnt_parent == new_nd.mnt)
1834 tmp = tmp->mnt_parent;
1836 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1838 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1840 detach_mnt(new_nd.mnt, &parent_nd);
1841 detach_mnt(user_nd.mnt, &root_parent);
1842 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1843 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1844 touch_mnt_namespace(current->nsproxy->mnt_ns);
1845 spin_unlock(&vfsmount_lock);
1846 chroot_fs_refs(&user_nd, &new_nd);
1847 security_sb_post_pivotroot(&user_nd, &new_nd);
1849 path_release(&root_parent);
1850 path_release(&parent_nd);
1852 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1853 up_write(&namespace_sem);
1854 path_release(&user_nd);
1855 path_release(&old_nd);
1857 path_release(&new_nd);
1862 spin_unlock(&vfsmount_lock);
1866 static void __init init_mount_tree(void)
1868 struct vfsmount *mnt;
1869 struct mnt_namespace *ns;
1871 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1873 panic("Can't create rootfs");
1874 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1876 panic("Can't allocate initial namespace");
1877 atomic_set(&ns->count, 1);
1878 INIT_LIST_HEAD(&ns->list);
1879 init_waitqueue_head(&ns->poll);
1881 list_add(&mnt->mnt_list, &ns->list);
1885 init_task.nsproxy->mnt_ns = ns;
1888 set_fs_pwd(current->fs, ns->root, ns->root->mnt_root);
1889 set_fs_root(current->fs, ns->root, ns->root->mnt_root);
1892 void __init mnt_init(unsigned long mempages)
1894 struct list_head *d;
1895 unsigned int nr_hash;
1899 init_rwsem(&namespace_sem);
1901 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1902 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1904 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1906 if (!mount_hashtable)
1907 panic("Failed to allocate mount hash table\n");
1910 * Find the power-of-two list-heads that can fit into the allocation..
1911 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1914 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1918 } while ((nr_hash >> hash_bits) != 0);
1922 * Re-calculate the actual number of entries and the mask
1923 * from the number of bits we can fit.
1925 nr_hash = 1UL << hash_bits;
1926 hash_mask = nr_hash - 1;
1928 printk("Mount-cache hash table entries: %d\n", nr_hash);
1930 /* And initialize the newly allocated array */
1931 d = mount_hashtable;
1940 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1942 err = subsystem_register(&fs_subsys);
1944 printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1950 void __put_mnt_ns(struct mnt_namespace *ns)
1952 struct vfsmount *root = ns->root;
1953 LIST_HEAD(umount_list);
1955 spin_unlock(&vfsmount_lock);
1956 down_write(&namespace_sem);
1957 spin_lock(&vfsmount_lock);
1958 umount_tree(root, 0, &umount_list);
1959 spin_unlock(&vfsmount_lock);
1960 up_write(&namespace_sem);
1961 release_mounts(&umount_list);
1962 atomic_dec(&vs_global_mnt_ns);