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/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/quotaops.h>
17 #include <linux/acct.h>
18 #include <linux/module.h>
19 #include <linux/seq_file.h>
20 #include <linux/namespace.h>
21 #include <linux/namei.h>
22 #include <linux/security.h>
23 #include <linux/mount.h>
24 #include <linux/vs_base.h>
26 #include <asm/uaccess.h>
27 #include <asm/unistd.h>
29 extern int __init init_rootfs(void);
32 extern int __init sysfs_init(void);
34 static inline int sysfs_init(void)
40 /* spinlock for vfsmount related operations, inplace of dcache_lock */
41 spinlock_t vfsmount_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
43 static struct list_head *mount_hashtable;
44 static int hash_mask, hash_bits;
45 static kmem_cache_t *mnt_cache;
47 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
49 unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
50 tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
51 tmp = tmp + (tmp >> hash_bits);
52 return tmp & hash_mask;
55 struct vfsmount *alloc_vfsmnt(const char *name)
57 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
59 memset(mnt, 0, sizeof(struct vfsmount));
60 atomic_set(&mnt->mnt_count,1);
61 INIT_LIST_HEAD(&mnt->mnt_hash);
62 INIT_LIST_HEAD(&mnt->mnt_child);
63 INIT_LIST_HEAD(&mnt->mnt_mounts);
64 INIT_LIST_HEAD(&mnt->mnt_list);
65 INIT_LIST_HEAD(&mnt->mnt_fslink);
67 int size = strlen(name)+1;
68 char *newname = kmalloc(size, GFP_KERNEL);
70 memcpy(newname, name, size);
71 mnt->mnt_devname = newname;
78 void free_vfsmnt(struct vfsmount *mnt)
80 kfree(mnt->mnt_devname);
81 kmem_cache_free(mnt_cache, mnt);
85 * Now, lookup_mnt increments the ref count before returning
86 * the vfsmount struct.
88 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
90 struct list_head * head = mount_hashtable + hash(mnt, dentry);
91 struct list_head * tmp = head;
92 struct vfsmount *p, *found = NULL;
94 spin_lock(&vfsmount_lock);
100 p = list_entry(tmp, struct vfsmount, mnt_hash);
101 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
106 spin_unlock(&vfsmount_lock);
110 EXPORT_SYMBOL(lookup_mnt);
112 static inline int check_mnt(struct vfsmount *mnt)
114 return mnt->mnt_namespace == current->namespace;
117 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
119 old_nd->dentry = mnt->mnt_mountpoint;
120 old_nd->mnt = mnt->mnt_parent;
121 mnt->mnt_parent = mnt;
122 mnt->mnt_mountpoint = mnt->mnt_root;
123 list_del_init(&mnt->mnt_child);
124 list_del_init(&mnt->mnt_hash);
125 old_nd->dentry->d_mounted--;
128 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
130 mnt->mnt_parent = mntget(nd->mnt);
131 mnt->mnt_mountpoint = dget(nd->dentry);
132 list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
133 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
134 nd->dentry->d_mounted++;
137 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
139 struct list_head *next = p->mnt_mounts.next;
140 if (next == &p->mnt_mounts) {
144 next = p->mnt_child.next;
145 if (next != &p->mnt_parent->mnt_mounts)
150 return list_entry(next, struct vfsmount, mnt_child);
153 static struct vfsmount *
154 clone_mnt(struct vfsmount *old, struct dentry *root)
156 struct super_block *sb = old->mnt_sb;
157 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
160 mnt->mnt_flags = old->mnt_flags;
161 atomic_inc(&sb->s_active);
163 mnt->mnt_root = dget(root);
164 mnt->mnt_mountpoint = mnt->mnt_root;
165 mnt->mnt_parent = mnt;
166 mnt->mnt_namespace = old->mnt_namespace;
168 /* stick the duplicate mount on the same expiry list
169 * as the original if that was on one */
170 spin_lock(&vfsmount_lock);
171 if (!list_empty(&old->mnt_fslink))
172 list_add(&mnt->mnt_fslink, &old->mnt_fslink);
173 spin_unlock(&vfsmount_lock);
178 void __mntput(struct vfsmount *mnt)
180 struct super_block *sb = mnt->mnt_sb;
183 deactivate_super(sb);
186 EXPORT_SYMBOL(__mntput);
189 static void *m_start(struct seq_file *m, loff_t *pos)
191 struct namespace *n = m->private;
196 list_for_each(p, &n->list)
198 return list_entry(p, struct vfsmount, mnt_list);
202 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
204 struct namespace *n = m->private;
205 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
207 return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
210 static void m_stop(struct seq_file *m, void *v)
212 struct namespace *n = m->private;
216 static inline void mangle(struct seq_file *m, const char *s)
218 seq_escape(m, s, " \t\n\\");
221 static int show_vfsmnt(struct seq_file *m, void *v)
223 struct vfsmount *mnt = v;
225 static struct proc_fs_info {
229 { MS_SYNCHRONOUS, ",sync" },
230 { MS_DIRSYNC, ",dirsync" },
231 { MS_MANDLOCK, ",mand" },
232 { MS_NOATIME, ",noatime" },
233 { MS_NODIRATIME, ",nodiratime" },
234 { MS_TAGXID, ",tagxid" },
237 static struct proc_fs_info mnt_info[] = {
238 { MNT_NOSUID, ",nosuid" },
239 { MNT_NODEV, ",nodev" },
240 { MNT_NOEXEC, ",noexec" },
243 struct proc_fs_info *fs_infop;
245 if (vx_flags(VXF_HIDE_MOUNT, 0))
248 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
250 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
252 mangle(m, mnt->mnt_sb->s_type->name);
253 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
254 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
255 if (mnt->mnt_sb->s_flags & fs_infop->flag)
256 seq_puts(m, fs_infop->str);
258 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
259 if (mnt->mnt_flags & fs_infop->flag)
260 seq_puts(m, fs_infop->str);
262 if (mnt->mnt_sb->s_op->show_options)
263 err = mnt->mnt_sb->s_op->show_options(m, mnt);
264 seq_puts(m, " 0 0\n");
268 struct seq_operations mounts_op = {
276 * may_umount_tree - check if a mount tree is busy
277 * @mnt: root of mount tree
279 * This is called to check if a tree of mounts has any
280 * open files, pwds, chroots or sub mounts that are
283 int may_umount_tree(struct vfsmount *mnt)
285 struct list_head *next;
286 struct vfsmount *this_parent = mnt;
290 spin_lock(&vfsmount_lock);
291 actual_refs = atomic_read(&mnt->mnt_count);
294 next = this_parent->mnt_mounts.next;
296 while (next != &this_parent->mnt_mounts) {
297 struct vfsmount *p = list_entry(next, struct vfsmount, mnt_child);
301 actual_refs += atomic_read(&p->mnt_count);
304 if (!list_empty(&p->mnt_mounts)) {
310 if (this_parent != mnt) {
311 next = this_parent->mnt_child.next;
312 this_parent = this_parent->mnt_parent;
315 spin_unlock(&vfsmount_lock);
317 if (actual_refs > minimum_refs)
323 EXPORT_SYMBOL(may_umount_tree);
326 * may_umount - check if a mount point is busy
327 * @mnt: root of mount
329 * This is called to check if a mount point has any
330 * open files, pwds, chroots or sub mounts. If the
331 * mount has sub mounts this will return busy
332 * regardless of whether the sub mounts are busy.
334 * Doesn't take quota and stuff into account. IOW, in some cases it will
335 * give false negatives. The main reason why it's here is that we need
336 * a non-destructive way to look for easily umountable filesystems.
338 int may_umount(struct vfsmount *mnt)
340 if (atomic_read(&mnt->mnt_count) > 2)
345 EXPORT_SYMBOL(may_umount);
347 static inline void __umount_tree(struct vfsmount *mnt, struct list_head *kill)
349 while (!list_empty(kill)) {
350 mnt = list_entry(kill->next, struct vfsmount, mnt_list);
351 list_del_init(&mnt->mnt_list);
352 list_del_init(&mnt->mnt_fslink);
353 if (mnt->mnt_parent == mnt) {
354 spin_unlock(&vfsmount_lock);
356 struct nameidata old_nd;
357 detach_mnt(mnt, &old_nd);
358 spin_unlock(&vfsmount_lock);
359 path_release(&old_nd);
362 spin_lock(&vfsmount_lock);
366 void umount_tree(struct vfsmount *mnt)
371 for (p = mnt; p; p = next_mnt(p, mnt)) {
372 list_del(&p->mnt_list);
373 list_add(&p->mnt_list, &kill);
375 __umount_tree(mnt, &kill);
378 void umount_unused(struct vfsmount *mnt, struct fs_struct *fs)
383 for (p = mnt; p; p = next_mnt(p, mnt)) {
384 if (p == fs->rootmnt || p == fs->pwdmnt)
386 list_del(&p->mnt_list);
387 list_add(&p->mnt_list, &kill);
389 __umount_tree(mnt, &kill);
392 static int do_umount(struct vfsmount *mnt, int flags)
394 struct super_block * sb = mnt->mnt_sb;
397 retval = security_sb_umount(mnt, flags);
402 * Allow userspace to request a mountpoint be expired rather than
403 * unmounting unconditionally. Unmount only happens if:
404 * (1) the mark is already set (the mark is cleared by mntput())
405 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
407 if (flags & MNT_EXPIRE) {
408 if (mnt == current->fs->rootmnt ||
409 flags & (MNT_FORCE | MNT_DETACH))
412 if (atomic_read(&mnt->mnt_count) != 2)
415 if (!xchg(&mnt->mnt_expiry_mark, 1))
420 * If we may have to abort operations to get out of this
421 * mount, and they will themselves hold resources we must
422 * allow the fs to do things. In the Unix tradition of
423 * 'Gee thats tricky lets do it in userspace' the umount_begin
424 * might fail to complete on the first run through as other tasks
425 * must return, and the like. Thats for the mount program to worry
426 * about for the moment.
430 if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
431 sb->s_op->umount_begin(sb);
435 * No sense to grab the lock for this test, but test itself looks
436 * somewhat bogus. Suggestions for better replacement?
437 * Ho-hum... In principle, we might treat that as umount + switch
438 * to rootfs. GC would eventually take care of the old vfsmount.
439 * Actually it makes sense, especially if rootfs would contain a
440 * /reboot - static binary that would close all descriptors and
441 * call reboot(9). Then init(8) could umount root and exec /reboot.
443 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
445 * Special case for "unmounting" root ...
446 * we just try to remount it readonly.
448 down_write(&sb->s_umount);
449 if (!(sb->s_flags & MS_RDONLY)) {
451 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
454 up_write(&sb->s_umount);
458 down_write(¤t->namespace->sem);
459 spin_lock(&vfsmount_lock);
461 if (atomic_read(&sb->s_active) == 1) {
462 /* last instance - try to be smart */
463 spin_unlock(&vfsmount_lock);
468 security_sb_umount_close(mnt);
469 spin_lock(&vfsmount_lock);
472 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
473 if (!list_empty(&mnt->mnt_list))
477 spin_unlock(&vfsmount_lock);
479 security_sb_umount_busy(mnt);
480 up_write(¤t->namespace->sem);
485 * Now umount can handle mount points as well as block devices.
486 * This is important for filesystems which use unnamed block devices.
488 * We now support a flag for forced unmount like the other 'big iron'
489 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
492 asmlinkage long sys_umount(char __user * name, int flags)
497 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
501 if (nd.dentry != nd.mnt->mnt_root)
503 if (!check_mnt(nd.mnt))
507 if (!capable(CAP_SYS_ADMIN) && !vx_ccaps(VXC_SECURE_MOUNT))
510 retval = do_umount(nd.mnt, flags);
512 path_release_on_umount(&nd);
517 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
520 * The 2.0 compatible umount. No flags.
523 asmlinkage long sys_oldumount(char __user * name)
525 return sys_umount(name,0);
530 static int mount_is_safe(struct nameidata *nd)
532 if (capable(CAP_SYS_ADMIN))
534 if (vx_ccaps(VXC_SECURE_MOUNT))
538 if (S_ISLNK(nd->dentry->d_inode->i_mode))
540 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
541 if (current->uid != nd->dentry->d_inode->i_uid)
544 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
551 lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
556 if (d == NULL || d == d->d_parent)
562 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
564 struct vfsmount *res, *p, *q, *r, *s;
568 res = q = clone_mnt(mnt, dentry);
571 q->mnt_mountpoint = mnt->mnt_mountpoint;
574 for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) {
575 r = list_entry(h, struct vfsmount, mnt_child);
576 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
579 for (s = r; s; s = next_mnt(s, r)) {
580 while (p != s->mnt_parent) {
586 nd.dentry = p->mnt_mountpoint;
587 q = clone_mnt(p, p->mnt_root);
590 spin_lock(&vfsmount_lock);
591 list_add_tail(&q->mnt_list, &res->mnt_list);
593 spin_unlock(&vfsmount_lock);
599 spin_lock(&vfsmount_lock);
601 spin_unlock(&vfsmount_lock);
606 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
609 if (mnt->mnt_sb->s_flags & MS_NOUSER)
612 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
613 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
617 down(&nd->dentry->d_inode->i_sem);
618 if (IS_DEADDIR(nd->dentry->d_inode))
621 err = security_sb_check_sb(mnt, nd);
626 spin_lock(&vfsmount_lock);
627 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
628 struct list_head head;
631 list_add_tail(&head, &mnt->mnt_list);
632 list_splice(&head, current->namespace->list.prev);
636 spin_unlock(&vfsmount_lock);
638 up(&nd->dentry->d_inode->i_sem);
640 security_sb_post_addmount(mnt, nd);
647 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
649 struct nameidata old_nd;
650 struct vfsmount *mnt = NULL;
651 int err = mount_is_safe(nd);
654 if (!old_name || !*old_name)
656 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
660 down_write(¤t->namespace->sem);
662 if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
665 mnt = copy_tree(old_nd.mnt, old_nd.dentry);
667 mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
671 /* stop bind mounts from expiring */
672 spin_lock(&vfsmount_lock);
673 list_del_init(&mnt->mnt_fslink);
674 spin_unlock(&vfsmount_lock);
676 err = graft_tree(mnt, nd);
678 spin_lock(&vfsmount_lock);
680 spin_unlock(&vfsmount_lock);
685 up_write(¤t->namespace->sem);
686 path_release(&old_nd);
691 * change filesystem flags. dir should be a physical root of filesystem.
692 * If you've mounted a non-root directory somewhere and want to do remount
693 * on it - tough luck.
696 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
700 struct super_block * sb = nd->mnt->mnt_sb;
702 if (!capable(CAP_SYS_ADMIN))
705 if (!check_mnt(nd->mnt))
708 if (nd->dentry != nd->mnt->mnt_root)
711 down_write(&sb->s_umount);
712 err = do_remount_sb(sb, flags, data, 0);
714 nd->mnt->mnt_flags=mnt_flags;
715 up_write(&sb->s_umount);
717 security_sb_post_remount(nd->mnt, flags, data);
721 static int do_move_mount(struct nameidata *nd, char *old_name)
723 struct nameidata old_nd, parent_nd;
726 if (!capable(CAP_SYS_ADMIN))
728 if (!old_name || !*old_name)
730 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
734 down_write(¤t->namespace->sem);
735 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
738 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
742 down(&nd->dentry->d_inode->i_sem);
743 if (IS_DEADDIR(nd->dentry->d_inode))
746 spin_lock(&vfsmount_lock);
747 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
751 if (old_nd.dentry != old_nd.mnt->mnt_root)
754 if (old_nd.mnt == old_nd.mnt->mnt_parent)
757 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
758 S_ISDIR(old_nd.dentry->d_inode->i_mode))
762 for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
767 detach_mnt(old_nd.mnt, &parent_nd);
768 attach_mnt(old_nd.mnt, nd);
770 /* if the mount is moved, it should no longer be expire
772 list_del_init(&old_nd.mnt->mnt_fslink);
774 spin_unlock(&vfsmount_lock);
776 up(&nd->dentry->d_inode->i_sem);
778 up_write(¤t->namespace->sem);
780 path_release(&parent_nd);
781 path_release(&old_nd);
786 * create a new mount for userspace and request it to be added into the
789 static int do_new_mount(struct nameidata *nd, char *type, int flags,
790 int mnt_flags, char *name, void *data)
792 struct vfsmount *mnt;
794 if (!type || !memchr(type, 0, PAGE_SIZE))
797 /* we need capabilities... */
798 if (!capable(CAP_SYS_ADMIN) && !vx_ccaps(VXC_SECURE_MOUNT))
801 mnt = do_kern_mount(type, flags, name, data);
805 return do_add_mount(mnt, nd, mnt_flags, NULL);
809 * add a mount into a namespace's mount tree
810 * - provide the option of adding the new mount to an expiration list
812 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
813 int mnt_flags, struct list_head *fslist)
817 down_write(¤t->namespace->sem);
818 /* Something was mounted here while we slept */
819 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
822 if (!check_mnt(nd->mnt))
825 /* Refuse the same filesystem on the same mount point */
827 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
828 nd->mnt->mnt_root == nd->dentry)
832 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
835 newmnt->mnt_flags = mnt_flags;
836 err = graft_tree(newmnt, nd);
838 if (err == 0 && fslist) {
839 /* add to the specified expiration list */
840 spin_lock(&vfsmount_lock);
841 list_add_tail(&newmnt->mnt_fslink, fslist);
842 spin_unlock(&vfsmount_lock);
846 up_write(¤t->namespace->sem);
851 EXPORT_SYMBOL_GPL(do_add_mount);
854 * process a list of expirable mountpoints with the intent of discarding any
855 * mountpoints that aren't in use and haven't been touched since last we came
858 void mark_mounts_for_expiry(struct list_head *mounts)
860 struct namespace *namespace;
861 struct vfsmount *mnt, *next;
862 LIST_HEAD(graveyard);
864 if (list_empty(mounts))
867 spin_lock(&vfsmount_lock);
869 /* extract from the expiration list every vfsmount that matches the
870 * following criteria:
871 * - only referenced by its parent vfsmount
872 * - still marked for expiry (marked on the last call here; marks are
873 * cleared by mntput())
875 list_for_each_entry_safe(mnt, next, mounts, mnt_fslink) {
876 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
877 atomic_read(&mnt->mnt_count) != 1)
881 list_move(&mnt->mnt_fslink, &graveyard);
885 * go through the vfsmounts we've just consigned to the graveyard to
886 * - check that they're still dead
887 * - delete the vfsmount from the appropriate namespace under lock
888 * - dispose of the corpse
890 while (!list_empty(&graveyard)) {
891 mnt = list_entry(graveyard.next, struct vfsmount, mnt_fslink);
892 list_del_init(&mnt->mnt_fslink);
894 /* don't do anything if the namespace is dead - all the
895 * vfsmounts from it are going away anyway */
896 namespace = mnt->mnt_namespace;
897 if (!namespace || atomic_read(&namespace->count) <= 0)
899 get_namespace(namespace);
901 spin_unlock(&vfsmount_lock);
902 down_write(&namespace->sem);
903 spin_lock(&vfsmount_lock);
905 /* check that it is still dead: the count should now be 2 - as
906 * contributed by the vfsmount parent and the mntget above */
907 if (atomic_read(&mnt->mnt_count) == 2) {
908 struct vfsmount *xdmnt;
909 struct dentry *xdentry;
911 /* delete from the namespace */
912 list_del_init(&mnt->mnt_list);
913 list_del_init(&mnt->mnt_child);
914 list_del_init(&mnt->mnt_hash);
915 mnt->mnt_mountpoint->d_mounted--;
917 xdentry = mnt->mnt_mountpoint;
918 mnt->mnt_mountpoint = mnt->mnt_root;
919 xdmnt = mnt->mnt_parent;
920 mnt->mnt_parent = mnt;
922 spin_unlock(&vfsmount_lock);
927 /* now lay it to rest if this was the last ref on the
929 if (atomic_read(&mnt->mnt_sb->s_active) == 1) {
930 /* last instance - try to be smart */
932 DQUOT_OFF(mnt->mnt_sb);
933 acct_auto_close(mnt->mnt_sb);
939 /* someone brought it back to life whilst we didn't
940 * have any locks held so return it to the expiration
942 list_add_tail(&mnt->mnt_fslink, mounts);
943 spin_unlock(&vfsmount_lock);
946 up_write(&namespace->sem);
949 put_namespace(namespace);
951 spin_lock(&vfsmount_lock);
954 spin_unlock(&vfsmount_lock);
957 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
959 int copy_mount_options (const void __user *data, unsigned long *where)
969 if (!(page = __get_free_page(GFP_KERNEL)))
972 /* We only care that *some* data at the address the user
973 * gave us is valid. Just in case, we'll zero
974 * the remainder of the page.
976 /* copy_from_user cannot cross TASK_SIZE ! */
977 size = TASK_SIZE - (unsigned long)data;
978 if (size > PAGE_SIZE)
981 i = size - copy_from_user((void *)page, data, size);
987 memset((char *)page + i, 0, PAGE_SIZE - i);
993 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
994 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
996 * data is a (void *) that can point to any structure up to
997 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
998 * information (or be NULL).
1000 * Pre-0.97 versions of mount() didn't have a flags word.
1001 * When the flags word was introduced its top half was required
1002 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1003 * Therefore, if this magic number is present, it carries no information
1004 * and must be discarded.
1006 long do_mount(char * dev_name, char * dir_name, char *type_page,
1007 unsigned long flags, void *data_page)
1009 struct nameidata nd;
1014 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1015 flags &= ~MS_MGC_MSK;
1017 /* Basic sanity checks */
1019 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1021 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1025 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1027 /* Separate the per-mountpoint flags */
1028 if (flags & MS_NOSUID)
1029 mnt_flags |= MNT_NOSUID;
1030 if (flags & MS_NODEV)
1031 mnt_flags |= MNT_NODEV;
1032 if (flags & MS_NOEXEC)
1033 mnt_flags |= MNT_NOEXEC;
1034 flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_ACTIVE);
1036 if (vx_ccaps(VXC_SECURE_MOUNT))
1037 mnt_flags |= MNT_NODEV;
1039 /* ... and get the mountpoint */
1040 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1044 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1048 if (flags & MS_REMOUNT)
1049 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1051 else if (flags & MS_BIND)
1052 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1053 else if (flags & MS_MOVE)
1054 retval = do_move_mount(&nd, dev_name);
1056 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1057 dev_name, data_page);
1063 int copy_namespace(int flags, struct task_struct *tsk)
1065 struct namespace *namespace = tsk->namespace;
1066 struct namespace *new_ns;
1067 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1068 struct fs_struct *fs = tsk->fs;
1069 struct vfsmount *p, *q;
1074 get_namespace(namespace);
1076 if (!(flags & CLONE_NEWNS))
1079 if (!capable(CAP_SYS_ADMIN)) {
1080 put_namespace(namespace);
1084 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1088 atomic_set(&new_ns->count, 1);
1089 init_rwsem(&new_ns->sem);
1090 INIT_LIST_HEAD(&new_ns->list);
1092 down_write(&tsk->namespace->sem);
1093 /* First pass: copy the tree topology */
1094 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
1095 if (!new_ns->root) {
1096 up_write(&tsk->namespace->sem);
1100 spin_lock(&vfsmount_lock);
1101 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1102 spin_unlock(&vfsmount_lock);
1105 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1106 * as belonging to new namespace. We have already acquired a private
1107 * fs_struct, so tsk->fs->lock is not needed.
1109 p = namespace->root;
1112 q->mnt_namespace = new_ns;
1114 if (p == fs->rootmnt) {
1116 fs->rootmnt = mntget(q);
1118 if (p == fs->pwdmnt) {
1120 fs->pwdmnt = mntget(q);
1122 if (p == fs->altrootmnt) {
1124 fs->altrootmnt = mntget(q);
1127 p = next_mnt(p, namespace->root);
1128 q = next_mnt(q, new_ns->root);
1130 up_write(&tsk->namespace->sem);
1132 tsk->namespace = new_ns;
1141 put_namespace(namespace);
1145 put_namespace(namespace);
1149 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1150 char __user * type, unsigned long flags,
1154 unsigned long data_page;
1155 unsigned long type_page;
1156 unsigned long dev_page;
1159 retval = copy_mount_options (type, &type_page);
1163 dir_page = getname(dir_name);
1164 retval = PTR_ERR(dir_page);
1165 if (IS_ERR(dir_page))
1168 retval = copy_mount_options (dev_name, &dev_page);
1172 retval = copy_mount_options (data, &data_page);
1177 retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
1178 flags, (void*)data_page);
1180 free_page(data_page);
1183 free_page(dev_page);
1187 free_page(type_page);
1192 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1193 * It can block. Requires the big lock held.
1195 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1196 struct dentry *dentry)
1198 struct dentry *old_root;
1199 struct vfsmount *old_rootmnt;
1200 write_lock(&fs->lock);
1201 old_root = fs->root;
1202 old_rootmnt = fs->rootmnt;
1203 fs->rootmnt = mntget(mnt);
1204 fs->root = dget(dentry);
1205 write_unlock(&fs->lock);
1208 mntput(old_rootmnt);
1212 EXPORT_SYMBOL(set_fs_root);
1215 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1216 * It can block. Requires the big lock held.
1218 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1219 struct dentry *dentry)
1221 struct dentry *old_pwd;
1222 struct vfsmount *old_pwdmnt;
1224 write_lock(&fs->lock);
1226 old_pwdmnt = fs->pwdmnt;
1227 fs->pwdmnt = mntget(mnt);
1228 fs->pwd = dget(dentry);
1229 write_unlock(&fs->lock);
1237 EXPORT_SYMBOL(set_fs_pwd);
1239 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1241 struct task_struct *g, *p;
1242 struct fs_struct *fs;
1244 read_lock(&tasklist_lock);
1245 do_each_thread(g, p) {
1249 atomic_inc(&fs->count);
1251 if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
1252 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1253 if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
1254 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1258 } while_each_thread(g, p);
1259 read_unlock(&tasklist_lock);
1263 * Moves the current root to put_root, and sets root/cwd of all processes
1264 * which had them on the old root to new_root.
1267 * - we don't move root/cwd if they are not at the root (reason: if something
1268 * cared enough to change them, it's probably wrong to force them elsewhere)
1269 * - it's okay to pick a root that isn't the root of a file system, e.g.
1270 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1271 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1275 asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
1277 struct vfsmount *tmp;
1278 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1281 if (!capable(CAP_SYS_ADMIN))
1286 error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
1290 if (!check_mnt(new_nd.mnt))
1293 error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
1297 error = security_sb_pivotroot(&old_nd, &new_nd);
1299 path_release(&old_nd);
1303 read_lock(¤t->fs->lock);
1304 user_nd.mnt = mntget(current->fs->rootmnt);
1305 user_nd.dentry = dget(current->fs->root);
1306 read_unlock(¤t->fs->lock);
1307 down_write(¤t->namespace->sem);
1308 down(&old_nd.dentry->d_inode->i_sem);
1310 if (!check_mnt(user_nd.mnt))
1313 if (IS_DEADDIR(new_nd.dentry->d_inode))
1315 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1317 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1320 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1321 goto out2; /* loop */
1323 if (user_nd.mnt->mnt_root != user_nd.dentry)
1325 if (new_nd.mnt->mnt_root != new_nd.dentry)
1326 goto out2; /* not a mountpoint */
1327 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1328 spin_lock(&vfsmount_lock);
1329 if (tmp != new_nd.mnt) {
1331 if (tmp->mnt_parent == tmp)
1333 if (tmp->mnt_parent == new_nd.mnt)
1335 tmp = tmp->mnt_parent;
1337 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1339 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1341 detach_mnt(new_nd.mnt, &parent_nd);
1342 detach_mnt(user_nd.mnt, &root_parent);
1343 attach_mnt(user_nd.mnt, &old_nd);
1344 attach_mnt(new_nd.mnt, &root_parent);
1345 spin_unlock(&vfsmount_lock);
1346 chroot_fs_refs(&user_nd, &new_nd);
1347 security_sb_post_pivotroot(&user_nd, &new_nd);
1349 path_release(&root_parent);
1350 path_release(&parent_nd);
1352 up(&old_nd.dentry->d_inode->i_sem);
1353 up_write(¤t->namespace->sem);
1354 path_release(&user_nd);
1355 path_release(&old_nd);
1357 path_release(&new_nd);
1362 spin_unlock(&vfsmount_lock);
1366 static void __init init_mount_tree(void)
1368 struct vfsmount *mnt;
1369 struct namespace *namespace;
1370 struct task_struct *g, *p;
1372 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1374 panic("Can't create rootfs");
1375 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1377 panic("Can't allocate initial namespace");
1378 atomic_set(&namespace->count, 1);
1379 INIT_LIST_HEAD(&namespace->list);
1380 init_rwsem(&namespace->sem);
1381 list_add(&mnt->mnt_list, &namespace->list);
1382 namespace->root = mnt;
1383 mnt->mnt_namespace = namespace;
1385 init_task.namespace = namespace;
1386 read_lock(&tasklist_lock);
1387 do_each_thread(g, p) {
1388 get_namespace(namespace);
1389 p->namespace = namespace;
1390 } while_each_thread(g, p);
1391 read_unlock(&tasklist_lock);
1393 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1394 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1397 void __init mnt_init(unsigned long mempages)
1399 struct list_head *d;
1400 unsigned long order;
1401 unsigned int nr_hash;
1404 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1405 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1408 mount_hashtable = (struct list_head *)
1409 __get_free_pages(GFP_ATOMIC, order);
1411 if (!mount_hashtable)
1412 panic("Failed to allocate mount hash table\n");
1415 * Find the power-of-two list-heads that can fit into the allocation..
1416 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1419 nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct list_head);
1423 } while ((nr_hash >> hash_bits) != 0);
1427 * Re-calculate the actual number of entries and the mask
1428 * from the number of bits we can fit.
1430 nr_hash = 1UL << hash_bits;
1431 hash_mask = nr_hash-1;
1433 printk("Mount-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1434 nr_hash, order, (PAGE_SIZE << order));
1436 /* And initialize the newly allocated array */
1437 d = mount_hashtable;
1449 void __put_namespace(struct namespace *namespace)
1451 struct vfsmount *mnt;
1453 down_write(&namespace->sem);
1454 spin_lock(&vfsmount_lock);
1456 list_for_each_entry(mnt, &namespace->list, mnt_list) {
1457 mnt->mnt_namespace = NULL;
1460 umount_tree(namespace->root);
1461 spin_unlock(&vfsmount_lock);
1462 up_write(&namespace->sem);