4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/config.h>
12 #include <linux/syscalls.h>
13 #include <linux/slab.h>
14 #include <linux/sched.h>
15 #include <linux/smp_lock.h>
16 #include <linux/init.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/seq_file.h>
22 #include <linux/namespace.h>
23 #include <linux/namei.h>
24 #include <linux/security.h>
25 #include <linux/mount.h>
26 #include <linux/vserver/namespace.h>
27 #include <linux/vserver/xid.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
32 extern int __init init_rootfs(void);
35 extern int __init sysfs_init(void);
37 static inline int sysfs_init(void)
43 /* spinlock for vfsmount related operations, inplace of dcache_lock */
44 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
48 static struct list_head *mount_hashtable __read_mostly;
49 static int hash_mask __read_mostly, hash_bits __read_mostly;
50 static kmem_cache_t *mnt_cache __read_mostly;
51 static struct rw_semaphore namespace_sem;
54 decl_subsys(fs, NULL, NULL);
55 EXPORT_SYMBOL_GPL(fs_subsys);
57 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
60 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
61 tmp = tmp + (tmp >> hash_bits);
62 return tmp & hash_mask;
65 struct vfsmount *alloc_vfsmnt(const char *name)
67 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
69 memset(mnt, 0, sizeof(struct vfsmount));
70 atomic_set(&mnt->mnt_count, 1);
71 INIT_LIST_HEAD(&mnt->mnt_hash);
72 INIT_LIST_HEAD(&mnt->mnt_child);
73 INIT_LIST_HEAD(&mnt->mnt_mounts);
74 INIT_LIST_HEAD(&mnt->mnt_list);
75 INIT_LIST_HEAD(&mnt->mnt_expire);
76 INIT_LIST_HEAD(&mnt->mnt_share);
77 INIT_LIST_HEAD(&mnt->mnt_slave_list);
78 INIT_LIST_HEAD(&mnt->mnt_slave);
80 int size = strlen(name) + 1;
81 char *newname = kmalloc(size, GFP_KERNEL);
83 memcpy(newname, name, size);
84 mnt->mnt_devname = newname;
91 void free_vfsmnt(struct vfsmount *mnt)
93 kfree(mnt->mnt_devname);
94 kmem_cache_free(mnt_cache, mnt);
98 * find the first or last mount at @dentry on vfsmount @mnt depending on
99 * @dir. If @dir is set return the first mount else return the last mount.
101 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
104 struct list_head *head = mount_hashtable + hash(mnt, dentry);
105 struct list_head *tmp = head;
106 struct vfsmount *p, *found = NULL;
109 tmp = dir ? tmp->next : tmp->prev;
113 p = list_entry(tmp, struct vfsmount, mnt_hash);
114 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
123 * lookup_mnt increments the ref count before returning
124 * the vfsmount struct.
126 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
128 struct vfsmount *child_mnt;
129 spin_lock(&vfsmount_lock);
130 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 spin_unlock(&vfsmount_lock);
136 static inline int check_mnt(struct vfsmount *mnt)
138 return mnt->mnt_namespace == current->namespace;
141 static void touch_namespace(struct namespace *ns)
145 wake_up_interruptible(&ns->poll);
149 static void __touch_namespace(struct namespace *ns)
151 if (ns && ns->event != event) {
153 wake_up_interruptible(&ns->poll);
157 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
159 old_nd->dentry = mnt->mnt_mountpoint;
160 old_nd->mnt = mnt->mnt_parent;
161 mnt->mnt_parent = mnt;
162 mnt->mnt_mountpoint = mnt->mnt_root;
163 list_del_init(&mnt->mnt_child);
164 list_del_init(&mnt->mnt_hash);
165 old_nd->dentry->d_mounted--;
168 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
169 struct vfsmount *child_mnt)
171 child_mnt->mnt_parent = mntget(mnt);
172 child_mnt->mnt_mountpoint = dget(dentry);
176 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
178 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
179 list_add_tail(&mnt->mnt_hash, mount_hashtable +
180 hash(nd->mnt, nd->dentry));
181 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
185 * the caller must hold vfsmount_lock
187 static void commit_tree(struct vfsmount *mnt)
189 struct vfsmount *parent = mnt->mnt_parent;
192 struct namespace *n = parent->mnt_namespace;
194 BUG_ON(parent == mnt);
196 list_add_tail(&head, &mnt->mnt_list);
197 list_for_each_entry(m, &head, mnt_list)
198 m->mnt_namespace = n;
199 list_splice(&head, n->list.prev);
201 list_add_tail(&mnt->mnt_hash, mount_hashtable +
202 hash(parent, mnt->mnt_mountpoint));
203 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
207 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
209 struct list_head *next = p->mnt_mounts.next;
210 if (next == &p->mnt_mounts) {
214 next = p->mnt_child.next;
215 if (next != &p->mnt_parent->mnt_mounts)
220 return list_entry(next, struct vfsmount, mnt_child);
223 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
225 struct list_head *prev = p->mnt_mounts.prev;
226 while (prev != &p->mnt_mounts) {
227 p = list_entry(prev, struct vfsmount, mnt_child);
228 prev = p->mnt_mounts.prev;
233 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
236 struct super_block *sb = old->mnt_sb;
237 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
240 mnt->mnt_flags = old->mnt_flags;
241 atomic_inc(&sb->s_active);
243 mnt->mnt_root = dget(root);
244 mnt->mnt_mountpoint = mnt->mnt_root;
245 mnt->mnt_parent = mnt;
246 mnt->mnt_xid = old->mnt_xid;
248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
259 if (flag & CL_MAKE_SHARED)
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
274 static inline void __mntput(struct vfsmount *mnt)
276 struct super_block *sb = mnt->mnt_sb;
279 deactivate_super(sb);
282 void mntput_no_expire(struct vfsmount *mnt)
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
300 EXPORT_SYMBOL(mntput_no_expire);
302 void mnt_pin(struct vfsmount *mnt)
304 spin_lock(&vfsmount_lock);
306 spin_unlock(&vfsmount_lock);
309 EXPORT_SYMBOL(mnt_pin);
311 void mnt_unpin(struct vfsmount *mnt)
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
318 spin_unlock(&vfsmount_lock);
321 EXPORT_SYMBOL(mnt_unpin);
324 static void *m_start(struct seq_file *m, loff_t *pos)
326 struct namespace *n = m->private;
330 down_read(&namespace_sem);
331 list_for_each(p, &n->list)
333 return list_entry(p, struct vfsmount, mnt_list);
337 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
339 struct namespace *n = m->private;
340 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
342 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
345 static void m_stop(struct seq_file *m, void *v)
347 up_read(&namespace_sem);
350 static inline void mangle(struct seq_file *m, const char *s)
352 seq_escape(m, s, " \t\n\\");
355 static int mnt_is_reachable(struct vfsmount *mnt)
357 struct vfsmount *root_mnt;
358 struct dentry *root, *point;
361 if (mnt == mnt->mnt_namespace->root)
364 spin_lock(&dcache_lock);
365 root_mnt = current->fs->rootmnt;
366 root = current->fs->root;
369 while ((mnt != mnt->mnt_parent) && (mnt != root_mnt)) {
370 point = mnt->mnt_mountpoint;
371 mnt = mnt->mnt_parent;
374 ret = (mnt == root_mnt) && is_subdir(point, root);
376 spin_unlock(&dcache_lock);
381 static int show_vfsmnt(struct seq_file *m, void *v)
383 struct vfsmount *mnt = v;
385 static struct proc_fs_info {
391 { MS_RDONLY, MNT_RDONLY, "ro", "rw" },
392 { MS_SYNCHRONOUS, 0, ",sync", NULL },
393 { MS_DIRSYNC, 0, ",dirsync", NULL },
394 { MS_MANDLOCK, 0, ",mand", NULL },
395 { MS_TAGXID, 0, ",tagxid", NULL },
396 { MS_NOATIME, MNT_NOATIME, ",noatime", NULL },
397 { MS_NODIRATIME, MNT_NODIRATIME, ",nodiratime", NULL },
398 { 0, MNT_NOSUID, ",nosuid", NULL },
399 { 0, MNT_NODEV, ",nodev", NULL },
400 { 0, MNT_NOEXEC, ",noexec", NULL },
403 struct proc_fs_info *p;
404 unsigned long s_flags = mnt->mnt_sb->s_flags;
405 int mnt_flags = mnt->mnt_flags;
407 if (vx_flags(VXF_HIDE_MOUNT, 0))
409 if (!mnt_is_reachable(mnt))
412 if (!vx_check(0, VX_ADMIN|VX_WATCH) &&
413 mnt == current->fs->rootmnt) {
414 seq_puts(m, "/dev/root / ");
416 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
418 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
421 mangle(m, mnt->mnt_sb->s_type->name);
423 for (p = fs_info; (p->s_flag | p->mnt_flag) ; p++) {
424 if ((s_flags & p->s_flag) || (mnt_flags & p->mnt_flag)) {
426 seq_puts(m, p->set_str);
429 seq_puts(m, p->unset_str);
432 if (mnt->mnt_flags & MNT_XID)
433 seq_printf(m, ",xid=%d", mnt->mnt_xid);
434 if (mnt->mnt_sb->s_op->show_options)
435 err = mnt->mnt_sb->s_op->show_options(m, mnt);
436 seq_puts(m, " 0 0\n");
440 struct seq_operations mounts_op = {
447 static int show_vfsstat(struct seq_file *m, void *v)
449 struct vfsmount *mnt = v;
453 if (mnt->mnt_devname) {
454 seq_puts(m, "device ");
455 mangle(m, mnt->mnt_devname);
457 seq_puts(m, "no device");
460 seq_puts(m, " mounted on ");
461 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
464 /* file system type */
465 seq_puts(m, "with fstype ");
466 mangle(m, mnt->mnt_sb->s_type->name);
468 /* optional statistics */
469 if (mnt->mnt_sb->s_op->show_stats) {
471 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
478 struct seq_operations mountstats_op = {
482 .show = show_vfsstat,
486 * may_umount_tree - check if a mount tree is busy
487 * @mnt: root of mount tree
489 * This is called to check if a tree of mounts has any
490 * open files, pwds, chroots or sub mounts that are
493 int may_umount_tree(struct vfsmount *mnt)
496 int minimum_refs = 0;
499 spin_lock(&vfsmount_lock);
500 for (p = mnt; p; p = next_mnt(p, mnt)) {
501 actual_refs += atomic_read(&p->mnt_count);
504 spin_unlock(&vfsmount_lock);
506 if (actual_refs > minimum_refs)
512 EXPORT_SYMBOL(may_umount_tree);
515 * may_umount - check if a mount point is busy
516 * @mnt: root of mount
518 * This is called to check if a mount point has any
519 * open files, pwds, chroots or sub mounts. If the
520 * mount has sub mounts this will return busy
521 * regardless of whether the sub mounts are busy.
523 * Doesn't take quota and stuff into account. IOW, in some cases it will
524 * give false negatives. The main reason why it's here is that we need
525 * a non-destructive way to look for easily umountable filesystems.
527 int may_umount(struct vfsmount *mnt)
530 spin_lock(&vfsmount_lock);
531 if (propagate_mount_busy(mnt, 2))
533 spin_unlock(&vfsmount_lock);
537 EXPORT_SYMBOL(may_umount);
539 void release_mounts(struct list_head *head)
541 struct vfsmount *mnt;
542 while (!list_empty(head)) {
543 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
544 list_del_init(&mnt->mnt_hash);
545 if (mnt->mnt_parent != mnt) {
546 struct dentry *dentry;
548 spin_lock(&vfsmount_lock);
549 dentry = mnt->mnt_mountpoint;
551 mnt->mnt_mountpoint = mnt->mnt_root;
552 mnt->mnt_parent = mnt;
553 spin_unlock(&vfsmount_lock);
561 static inline void __umount_list(struct vfsmount *mnt,
562 int propagate, struct list_head *kill)
567 propagate_umount(kill);
569 list_for_each_entry(p, kill, mnt_hash) {
570 list_del_init(&p->mnt_expire);
571 list_del_init(&p->mnt_list);
572 __touch_namespace(p->mnt_namespace);
573 p->mnt_namespace = NULL;
574 list_del_init(&p->mnt_child);
575 if (p->mnt_parent != p)
576 p->mnt_mountpoint->d_mounted--;
577 change_mnt_propagation(p, MS_PRIVATE);
581 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
585 for (p = mnt; p; p = next_mnt(p, mnt)) {
586 list_del(&p->mnt_hash);
587 list_add(&p->mnt_hash, kill);
588 // p->mnt_namespace = NULL;
590 __umount_list(mnt, propagate, kill);
593 void umount_unused(struct vfsmount *mnt, struct fs_struct *fs)
598 for (p = mnt; p; p = next_mnt(p, mnt)) {
599 if (p == fs->rootmnt || p == fs->pwdmnt)
601 list_del(&p->mnt_list);
602 list_add(&p->mnt_list, &kill);
603 p->mnt_namespace = NULL;
605 __umount_list(mnt, 0, &kill);
608 static int do_umount(struct vfsmount *mnt, int flags)
610 struct super_block *sb = mnt->mnt_sb;
612 LIST_HEAD(umount_list);
614 retval = security_sb_umount(mnt, flags);
619 * Allow userspace to request a mountpoint be expired rather than
620 * unmounting unconditionally. Unmount only happens if:
621 * (1) the mark is already set (the mark is cleared by mntput())
622 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
624 if (flags & MNT_EXPIRE) {
625 if (mnt == current->fs->rootmnt ||
626 flags & (MNT_FORCE | MNT_DETACH))
629 if (atomic_read(&mnt->mnt_count) != 2)
632 if (!xchg(&mnt->mnt_expiry_mark, 1))
637 * If we may have to abort operations to get out of this
638 * mount, and they will themselves hold resources we must
639 * allow the fs to do things. In the Unix tradition of
640 * 'Gee thats tricky lets do it in userspace' the umount_begin
641 * might fail to complete on the first run through as other tasks
642 * must return, and the like. Thats for the mount program to worry
643 * about for the moment.
647 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
648 sb->s_op->umount_begin(sb);
652 * No sense to grab the lock for this test, but test itself looks
653 * somewhat bogus. Suggestions for better replacement?
654 * Ho-hum... In principle, we might treat that as umount + switch
655 * to rootfs. GC would eventually take care of the old vfsmount.
656 * Actually it makes sense, especially if rootfs would contain a
657 * /reboot - static binary that would close all descriptors and
658 * call reboot(9). Then init(8) could umount root and exec /reboot.
660 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
662 * Special case for "unmounting" root ...
663 * we just try to remount it readonly.
665 down_write(&sb->s_umount);
666 if (!(sb->s_flags & MS_RDONLY)) {
669 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
672 up_write(&sb->s_umount);
676 down_write(&namespace_sem);
677 spin_lock(&vfsmount_lock);
681 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
682 if (!list_empty(&mnt->mnt_list))
683 umount_tree(mnt, 1, &umount_list);
686 spin_unlock(&vfsmount_lock);
688 security_sb_umount_busy(mnt);
689 up_write(&namespace_sem);
690 release_mounts(&umount_list);
695 * Now umount can handle mount points as well as block devices.
696 * This is important for filesystems which use unnamed block devices.
698 * We now support a flag for forced unmount like the other 'big iron'
699 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
702 asmlinkage long sys_umount(char __user * name, int flags)
707 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
711 if (nd.dentry != nd.mnt->mnt_root)
713 if (!check_mnt(nd.mnt))
717 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
720 retval = do_umount(nd.mnt, flags);
722 path_release_on_umount(&nd);
727 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
730 * The 2.0 compatible umount. No flags.
732 asmlinkage long sys_oldumount(char __user * name)
734 return sys_umount(name, 0);
739 static int mount_is_safe(struct nameidata *nd)
741 if (vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
745 if (S_ISLNK(nd->dentry->d_inode->i_mode))
747 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
748 if (current->uid != nd->dentry->d_inode->i_uid)
751 if (vfs_permission(nd, MAY_WRITE))
757 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
762 if (d == NULL || d == d->d_parent)
768 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
771 struct vfsmount *res, *p, *q, *r, *s;
774 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
777 res = q = clone_mnt(mnt, dentry, flag);
780 q->mnt_mountpoint = mnt->mnt_mountpoint;
783 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
784 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
787 for (s = r; s; s = next_mnt(s, r)) {
788 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
789 s = skip_mnt_tree(s);
792 while (p != s->mnt_parent) {
798 nd.dentry = p->mnt_mountpoint;
799 q = clone_mnt(p, p->mnt_root, flag);
802 spin_lock(&vfsmount_lock);
803 list_add_tail(&q->mnt_list, &res->mnt_list);
805 spin_unlock(&vfsmount_lock);
811 LIST_HEAD(umount_list);
812 spin_lock(&vfsmount_lock);
813 umount_tree(res, 0, &umount_list);
814 spin_unlock(&vfsmount_lock);
815 release_mounts(&umount_list);
821 * @source_mnt : mount tree to be attached
822 * @nd : place the mount tree @source_mnt is attached
823 * @parent_nd : if non-null, detach the source_mnt from its parent and
824 * store the parent mount and mountpoint dentry.
825 * (done when source_mnt is moved)
827 * NOTE: in the table below explains the semantics when a source mount
828 * of a given type is attached to a destination mount of a given type.
829 * ---------------------------------------------------------------------------
830 * | BIND MOUNT OPERATION |
831 * |**************************************************************************
832 * | source-->| shared | private | slave | unbindable |
836 * |**************************************************************************
837 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
839 * |non-shared| shared (+) | private | slave (*) | invalid |
840 * ***************************************************************************
841 * A bind operation clones the source mount and mounts the clone on the
844 * (++) the cloned mount is propagated to all the mounts in the propagation
845 * tree of the destination mount and the cloned mount is added to
846 * the peer group of the source mount.
847 * (+) the cloned mount is created under the destination mount and is marked
848 * as shared. The cloned mount is added to the peer group of the source
850 * (+++) the mount is propagated to all the mounts in the propagation tree
851 * of the destination mount and the cloned mount is made slave
852 * of the same master as that of the source mount. The cloned mount
853 * is marked as 'shared and slave'.
854 * (*) the cloned mount is made a slave of the same master as that of the
857 * ---------------------------------------------------------------------------
858 * | MOVE MOUNT OPERATION |
859 * |**************************************************************************
860 * | source-->| shared | private | slave | unbindable |
864 * |**************************************************************************
865 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
867 * |non-shared| shared (+*) | private | slave (*) | unbindable |
868 * ***************************************************************************
870 * (+) the mount is moved to the destination. And is then propagated to
871 * all the mounts in the propagation tree of the destination mount.
872 * (+*) the mount is moved to the destination.
873 * (+++) the mount is moved to the destination and is then propagated to
874 * all the mounts belonging to the destination mount's propagation tree.
875 * the mount is marked as 'shared and slave'.
876 * (*) the mount continues to be a slave at the new location.
878 * if the source mount is a tree, the operations explained above is
879 * applied to each mount in the tree.
880 * Must be called without spinlocks held, since this function can sleep
883 static int attach_recursive_mnt(struct vfsmount *source_mnt,
884 struct nameidata *nd, struct nameidata *parent_nd)
886 LIST_HEAD(tree_list);
887 struct vfsmount *dest_mnt = nd->mnt;
888 struct dentry *dest_dentry = nd->dentry;
889 struct vfsmount *child, *p;
891 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
894 if (IS_MNT_SHARED(dest_mnt)) {
895 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
899 spin_lock(&vfsmount_lock);
901 detach_mnt(source_mnt, parent_nd);
902 attach_mnt(source_mnt, nd);
903 touch_namespace(current->namespace);
905 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
906 commit_tree(source_mnt);
909 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
910 list_del_init(&child->mnt_hash);
913 spin_unlock(&vfsmount_lock);
917 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
920 if (mnt->mnt_sb->s_flags & MS_NOUSER)
923 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
924 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
928 mutex_lock(&nd->dentry->d_inode->i_mutex);
929 if (IS_DEADDIR(nd->dentry->d_inode))
932 err = security_sb_check_sb(mnt, nd);
937 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
938 err = attach_recursive_mnt(mnt, nd, NULL);
940 mutex_unlock(&nd->dentry->d_inode->i_mutex);
942 security_sb_post_addmount(mnt, nd);
947 * recursively change the type of the mountpoint.
949 static int do_change_type(struct nameidata *nd, int flag)
951 struct vfsmount *m, *mnt = nd->mnt;
952 int recurse = flag & MS_REC;
953 int type = flag & ~MS_REC;
955 if (nd->dentry != nd->mnt->mnt_root)
958 down_write(&namespace_sem);
959 spin_lock(&vfsmount_lock);
960 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
961 change_mnt_propagation(m, type);
962 spin_unlock(&vfsmount_lock);
963 up_write(&namespace_sem);
970 static int do_loopback(struct nameidata *nd, char *old_name, xid_t xid,
971 unsigned long flags, int mnt_flags)
973 struct nameidata old_nd;
974 struct vfsmount *mnt = NULL;
975 int err = mount_is_safe(nd);
976 int recurse = flags & MS_REC;
979 if (!old_name || !*old_name)
981 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
985 down_write(&namespace_sem);
987 if (IS_MNT_UNBINDABLE(old_nd.mnt))
990 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
995 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
997 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
1002 mnt->mnt_flags = mnt_flags;
1003 if (flags & MS_XID) {
1005 mnt->mnt_flags |= MNT_XID;
1008 err = graft_tree(mnt, nd);
1010 LIST_HEAD(umount_list);
1011 spin_lock(&vfsmount_lock);
1012 umount_tree(mnt, 0, &umount_list);
1013 spin_unlock(&vfsmount_lock);
1014 release_mounts(&umount_list);
1016 mnt->mnt_flags = mnt_flags;
1019 up_write(&namespace_sem);
1020 path_release(&old_nd);
1025 * change filesystem flags. dir should be a physical root of filesystem.
1026 * If you've mounted a non-root directory somewhere and want to do remount
1027 * on it - tough luck.
1029 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1030 void *data, xid_t xid)
1033 struct super_block *sb = nd->mnt->mnt_sb;
1035 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_REMOUNT))
1038 if (!check_mnt(nd->mnt))
1041 if (nd->dentry != nd->mnt->mnt_root)
1044 down_write(&sb->s_umount);
1045 err = do_remount_sb(sb, flags, data, 0);
1047 nd->mnt->mnt_flags = mnt_flags;
1048 up_write(&sb->s_umount);
1050 security_sb_post_remount(nd->mnt, flags, data);
1054 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1057 for (p = mnt; p; p = next_mnt(p, mnt)) {
1058 if (IS_MNT_UNBINDABLE(p))
1064 static int do_move_mount(struct nameidata *nd, char *old_name)
1066 struct nameidata old_nd, parent_nd;
1069 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1071 if (!old_name || !*old_name)
1073 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1077 down_write(&namespace_sem);
1078 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1081 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1085 mutex_lock(&nd->dentry->d_inode->i_mutex);
1086 if (IS_DEADDIR(nd->dentry->d_inode))
1089 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1093 if (old_nd.dentry != old_nd.mnt->mnt_root)
1096 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1099 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1100 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1103 * Don't move a mount residing in a shared parent.
1105 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1108 * Don't move a mount tree containing unbindable mounts to a destination
1109 * mount which is shared.
1111 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1114 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1115 if (p == old_nd.mnt)
1118 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1121 spin_lock(&vfsmount_lock);
1122 /* if the mount is moved, it should no longer be expire
1124 list_del_init(&old_nd.mnt->mnt_expire);
1125 spin_unlock(&vfsmount_lock);
1127 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1129 up_write(&namespace_sem);
1131 path_release(&parent_nd);
1132 path_release(&old_nd);
1137 * create a new mount for userspace and request it to be added into the
1140 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1141 int mnt_flags, char *name, void *data)
1143 struct vfsmount *mnt;
1145 if (!type || !memchr(type, 0, PAGE_SIZE))
1148 /* we need capabilities... */
1149 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT))
1152 mnt = do_kern_mount(type, flags, name, data);
1154 return PTR_ERR(mnt);
1156 return do_add_mount(mnt, nd, mnt_flags, NULL);
1160 * add a mount into a namespace's mount tree
1161 * - provide the option of adding the new mount to an expiration list
1163 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1164 int mnt_flags, struct list_head *fslist)
1168 down_write(&namespace_sem);
1169 /* Something was mounted here while we slept */
1170 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1173 if (!check_mnt(nd->mnt))
1176 /* Refuse the same filesystem on the same mount point */
1178 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1179 nd->mnt->mnt_root == nd->dentry)
1183 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1186 newmnt->mnt_flags = mnt_flags;
1187 if ((err = graft_tree(newmnt, nd)))
1191 /* add to the specified expiration list */
1192 spin_lock(&vfsmount_lock);
1193 list_add_tail(&newmnt->mnt_expire, fslist);
1194 spin_unlock(&vfsmount_lock);
1196 up_write(&namespace_sem);
1200 up_write(&namespace_sem);
1205 EXPORT_SYMBOL_GPL(do_add_mount);
1207 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1208 struct list_head *umounts)
1210 spin_lock(&vfsmount_lock);
1213 * Check if mount is still attached, if not, let whoever holds it deal
1216 if (mnt->mnt_parent == mnt) {
1217 spin_unlock(&vfsmount_lock);
1222 * Check that it is still dead: the count should now be 2 - as
1223 * contributed by the vfsmount parent and the mntget above
1225 if (!propagate_mount_busy(mnt, 2)) {
1226 /* delete from the namespace */
1227 touch_namespace(mnt->mnt_namespace);
1228 list_del_init(&mnt->mnt_list);
1229 mnt->mnt_namespace = NULL;
1230 umount_tree(mnt, 1, umounts);
1231 spin_unlock(&vfsmount_lock);
1234 * Someone brought it back to life whilst we didn't have any
1235 * locks held so return it to the expiration list
1237 list_add_tail(&mnt->mnt_expire, mounts);
1238 spin_unlock(&vfsmount_lock);
1243 * process a list of expirable mountpoints with the intent of discarding any
1244 * mountpoints that aren't in use and haven't been touched since last we came
1247 void mark_mounts_for_expiry(struct list_head *mounts)
1249 struct namespace *namespace;
1250 struct vfsmount *mnt, *next;
1251 LIST_HEAD(graveyard);
1253 if (list_empty(mounts))
1256 spin_lock(&vfsmount_lock);
1258 /* extract from the expiration list every vfsmount that matches the
1259 * following criteria:
1260 * - only referenced by its parent vfsmount
1261 * - still marked for expiry (marked on the last call here; marks are
1262 * cleared by mntput())
1264 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1265 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1266 atomic_read(&mnt->mnt_count) != 1)
1270 list_move(&mnt->mnt_expire, &graveyard);
1274 * go through the vfsmounts we've just consigned to the graveyard to
1275 * - check that they're still dead
1276 * - delete the vfsmount from the appropriate namespace under lock
1277 * - dispose of the corpse
1279 while (!list_empty(&graveyard)) {
1281 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
1282 list_del_init(&mnt->mnt_expire);
1284 /* don't do anything if the namespace is dead - all the
1285 * vfsmounts from it are going away anyway */
1286 namespace = mnt->mnt_namespace;
1287 if (!namespace || !namespace->root)
1289 get_namespace(namespace);
1291 spin_unlock(&vfsmount_lock);
1292 down_write(&namespace_sem);
1293 expire_mount(mnt, mounts, &umounts);
1294 up_write(&namespace_sem);
1295 release_mounts(&umounts);
1297 put_namespace(namespace);
1298 spin_lock(&vfsmount_lock);
1301 spin_unlock(&vfsmount_lock);
1304 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1307 * Some copy_from_user() implementations do not return the exact number of
1308 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1309 * Note that this function differs from copy_from_user() in that it will oops
1310 * on bad values of `to', rather than returning a short copy.
1312 static long exact_copy_from_user(void *to, const void __user * from,
1316 const char __user *f = from;
1319 if (!access_ok(VERIFY_READ, from, n))
1323 if (__get_user(c, f)) {
1334 int copy_mount_options(const void __user * data, unsigned long *where)
1344 if (!(page = __get_free_page(GFP_KERNEL)))
1347 /* We only care that *some* data at the address the user
1348 * gave us is valid. Just in case, we'll zero
1349 * the remainder of the page.
1351 /* copy_from_user cannot cross TASK_SIZE ! */
1352 size = TASK_SIZE - (unsigned long)data;
1353 if (size > PAGE_SIZE)
1356 i = size - exact_copy_from_user((void *)page, data, size);
1362 memset((char *)page + i, 0, PAGE_SIZE - i);
1368 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1369 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1371 * data is a (void *) that can point to any structure up to
1372 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1373 * information (or be NULL).
1375 * Pre-0.97 versions of mount() didn't have a flags word.
1376 * When the flags word was introduced its top half was required
1377 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1378 * Therefore, if this magic number is present, it carries no information
1379 * and must be discarded.
1381 long do_mount(char *dev_name, char *dir_name, char *type_page,
1382 unsigned long flags, void *data_page)
1384 struct nameidata nd;
1390 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1391 flags &= ~MS_MGC_MSK;
1393 /* Basic sanity checks */
1395 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1397 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1401 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1403 retval = vx_parse_xid(data_page, &xid, 1);
1405 mnt_flags |= MNT_XID;
1406 /* bind and re-mounts get xid flag */
1407 if (flags & (MS_BIND|MS_REMOUNT))
1411 /* Separate the per-mountpoint flags */
1412 if (flags & MS_RDONLY)
1413 mnt_flags |= MNT_RDONLY;
1414 if (flags & MS_NOSUID)
1415 mnt_flags |= MNT_NOSUID;
1416 if (flags & MS_NODEV)
1417 mnt_flags |= MNT_NODEV;
1418 if (flags & MS_NOEXEC)
1419 mnt_flags |= MNT_NOEXEC;
1420 if (flags & MS_NOATIME)
1421 mnt_flags |= MNT_NOATIME;
1422 if (flags & MS_NODIRATIME)
1423 mnt_flags |= MNT_NODIRATIME;
1425 if (vx_ccaps(VXC_SECURE_MOUNT))
1426 mnt_flags |= MNT_NODEV;
1427 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1428 MS_NOATIME | MS_NODIRATIME);
1430 /* ... and get the mountpoint */
1431 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1435 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1439 if (flags & MS_REMOUNT)
1440 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1442 else if (flags & MS_BIND)
1443 retval = do_loopback(&nd, dev_name, xid, flags, mnt_flags);
1444 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1445 retval = do_change_type(&nd, flags);
1446 else if (flags & MS_MOVE)
1447 retval = do_move_mount(&nd, dev_name);
1449 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1450 dev_name, data_page);
1457 * Allocate a new namespace structure and populate it with contents
1458 * copied from the namespace of the passed in task structure.
1460 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1462 struct namespace *namespace = tsk->namespace;
1463 struct namespace *new_ns;
1464 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1465 struct vfsmount *p, *q;
1467 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1471 atomic_set(&new_ns->count, 1);
1472 INIT_LIST_HEAD(&new_ns->list);
1473 init_waitqueue_head(&new_ns->poll);
1476 down_write(&namespace_sem);
1477 /* First pass: copy the tree topology */
1478 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1479 CL_COPY_ALL | CL_EXPIRE);
1480 if (!new_ns->root) {
1481 up_write(&namespace_sem);
1485 spin_lock(&vfsmount_lock);
1486 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1487 spin_unlock(&vfsmount_lock);
1490 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1491 * as belonging to new namespace. We have already acquired a private
1492 * fs_struct, so tsk->fs->lock is not needed.
1494 p = namespace->root;
1497 q->mnt_namespace = new_ns;
1499 if (p == fs->rootmnt) {
1501 fs->rootmnt = mntget(q);
1503 if (p == fs->pwdmnt) {
1505 fs->pwdmnt = mntget(q);
1507 if (p == fs->altrootmnt) {
1509 fs->altrootmnt = mntget(q);
1512 p = next_mnt(p, namespace->root);
1513 q = next_mnt(q, new_ns->root);
1515 up_write(&namespace_sem);
1527 int copy_namespace(int flags, struct task_struct *tsk)
1529 struct namespace *namespace = tsk->namespace;
1530 struct namespace *new_ns;
1536 get_namespace(namespace);
1538 if (!(flags & CLONE_NEWNS))
1541 if (!vx_capable(CAP_SYS_ADMIN, VXC_SECURE_MOUNT)) {
1546 new_ns = dup_namespace(tsk, tsk->fs);
1552 tsk->namespace = new_ns;
1555 put_namespace(namespace);
1559 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1560 char __user * type, unsigned long flags,
1564 unsigned long data_page;
1565 unsigned long type_page;
1566 unsigned long dev_page;
1569 retval = copy_mount_options(type, &type_page);
1573 dir_page = getname(dir_name);
1574 retval = PTR_ERR(dir_page);
1575 if (IS_ERR(dir_page))
1578 retval = copy_mount_options(dev_name, &dev_page);
1582 retval = copy_mount_options(data, &data_page);
1587 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1588 flags, (void *)data_page);
1590 free_page(data_page);
1593 free_page(dev_page);
1597 free_page(type_page);
1602 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1603 * It can block. Requires the big lock held.
1605 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1606 struct dentry *dentry)
1608 struct dentry *old_root;
1609 struct vfsmount *old_rootmnt;
1610 write_lock(&fs->lock);
1611 old_root = fs->root;
1612 old_rootmnt = fs->rootmnt;
1613 fs->rootmnt = mntget(mnt);
1614 fs->root = dget(dentry);
1615 write_unlock(&fs->lock);
1618 mntput(old_rootmnt);
1622 EXPORT_SYMBOL_GPL(set_fs_root);
1625 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1626 * It can block. Requires the big lock held.
1628 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1629 struct dentry *dentry)
1631 struct dentry *old_pwd;
1632 struct vfsmount *old_pwdmnt;
1634 write_lock(&fs->lock);
1636 old_pwdmnt = fs->pwdmnt;
1637 fs->pwdmnt = mntget(mnt);
1638 fs->pwd = dget(dentry);
1639 write_unlock(&fs->lock);
1647 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1649 struct task_struct *g, *p;
1650 struct fs_struct *fs;
1652 read_lock(&tasklist_lock);
1653 do_each_thread(g, p) {
1657 atomic_inc(&fs->count);
1659 if (fs->root == old_nd->dentry
1660 && fs->rootmnt == old_nd->mnt)
1661 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1662 if (fs->pwd == old_nd->dentry
1663 && fs->pwdmnt == old_nd->mnt)
1664 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1668 } while_each_thread(g, p);
1669 read_unlock(&tasklist_lock);
1673 * pivot_root Semantics:
1674 * Moves the root file system of the current process to the directory put_old,
1675 * makes new_root as the new root file system of the current process, and sets
1676 * root/cwd of all processes which had them on the current root to new_root.
1679 * The new_root and put_old must be directories, and must not be on the
1680 * same file system as the current process root. The put_old must be
1681 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1682 * pointed to by put_old must yield the same directory as new_root. No other
1683 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1685 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1686 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1687 * in this situation.
1690 * - we don't move root/cwd if they are not at the root (reason: if something
1691 * cared enough to change them, it's probably wrong to force them elsewhere)
1692 * - it's okay to pick a root that isn't the root of a file system, e.g.
1693 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1694 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1697 asmlinkage long sys_pivot_root(const char __user * new_root,
1698 const char __user * put_old)
1700 struct vfsmount *tmp;
1701 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1704 if (!capable(CAP_SYS_ADMIN))
1709 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1714 if (!check_mnt(new_nd.mnt))
1717 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1721 error = security_sb_pivotroot(&old_nd, &new_nd);
1723 path_release(&old_nd);
1727 read_lock(¤t->fs->lock);
1728 user_nd.mnt = mntget(current->fs->rootmnt);
1729 user_nd.dentry = dget(current->fs->root);
1730 read_unlock(¤t->fs->lock);
1731 down_write(&namespace_sem);
1732 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1734 if (IS_MNT_SHARED(old_nd.mnt) ||
1735 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1736 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1738 if (!check_mnt(user_nd.mnt))
1741 if (IS_DEADDIR(new_nd.dentry->d_inode))
1743 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1745 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1748 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1749 goto out2; /* loop, on the same file system */
1751 if (user_nd.mnt->mnt_root != user_nd.dentry)
1752 goto out2; /* not a mountpoint */
1753 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1754 goto out2; /* not attached */
1755 if (new_nd.mnt->mnt_root != new_nd.dentry)
1756 goto out2; /* not a mountpoint */
1757 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1758 goto out2; /* not attached */
1759 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1760 spin_lock(&vfsmount_lock);
1761 if (tmp != new_nd.mnt) {
1763 if (tmp->mnt_parent == tmp)
1764 goto out3; /* already mounted on put_old */
1765 if (tmp->mnt_parent == new_nd.mnt)
1767 tmp = tmp->mnt_parent;
1769 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1771 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1773 detach_mnt(new_nd.mnt, &parent_nd);
1774 detach_mnt(user_nd.mnt, &root_parent);
1775 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1776 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1777 touch_namespace(current->namespace);
1778 spin_unlock(&vfsmount_lock);
1779 chroot_fs_refs(&user_nd, &new_nd);
1780 security_sb_post_pivotroot(&user_nd, &new_nd);
1782 path_release(&root_parent);
1783 path_release(&parent_nd);
1785 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1786 up_write(&namespace_sem);
1787 path_release(&user_nd);
1788 path_release(&old_nd);
1790 path_release(&new_nd);
1795 spin_unlock(&vfsmount_lock);
1799 static void __init init_mount_tree(void)
1801 struct vfsmount *mnt;
1802 struct namespace *namespace;
1803 struct task_struct *g, *p;
1805 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1807 panic("Can't create rootfs");
1808 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1810 panic("Can't allocate initial namespace");
1811 atomic_set(&namespace->count, 1);
1812 INIT_LIST_HEAD(&namespace->list);
1813 init_waitqueue_head(&namespace->poll);
1814 namespace->event = 0;
1815 list_add(&mnt->mnt_list, &namespace->list);
1816 namespace->root = mnt;
1817 mnt->mnt_namespace = namespace;
1819 init_task.namespace = namespace;
1820 read_lock(&tasklist_lock);
1821 do_each_thread(g, p) {
1822 get_namespace(namespace);
1823 p->namespace = namespace;
1824 } while_each_thread(g, p);
1825 read_unlock(&tasklist_lock);
1827 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1828 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1831 void __init mnt_init(unsigned long mempages)
1833 struct list_head *d;
1834 unsigned int nr_hash;
1837 init_rwsem(&namespace_sem);
1839 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1840 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1842 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1844 if (!mount_hashtable)
1845 panic("Failed to allocate mount hash table\n");
1848 * Find the power-of-two list-heads that can fit into the allocation..
1849 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1852 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1856 } while ((nr_hash >> hash_bits) != 0);
1860 * Re-calculate the actual number of entries and the mask
1861 * from the number of bits we can fit.
1863 nr_hash = 1UL << hash_bits;
1864 hash_mask = nr_hash - 1;
1866 printk("Mount-cache hash table entries: %d\n", nr_hash);
1868 /* And initialize the newly allocated array */
1869 d = mount_hashtable;
1877 subsystem_register(&fs_subsys);
1882 void __put_namespace(struct namespace *namespace)
1884 struct vfsmount *root = namespace->root;
1885 LIST_HEAD(umount_list);
1886 namespace->root = NULL;
1887 spin_unlock(&vfsmount_lock);
1888 down_write(&namespace_sem);
1889 spin_lock(&vfsmount_lock);
1890 umount_tree(root, 0, &umount_list);
1891 spin_unlock(&vfsmount_lock);
1892 up_write(&namespace_sem);
1893 release_mounts(&umount_list);