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/module.h>
20 #include <linux/seq_file.h>
21 #include <linux/namespace.h>
22 #include <linux/namei.h>
23 #include <linux/security.h>
24 #include <linux/mount.h>
25 #include <asm/uaccess.h>
26 #include <asm/unistd.h>
29 extern int __init init_rootfs(void);
31 #define CL_EXPIRE 0x01
34 extern int __init sysfs_init(void);
36 static inline int sysfs_init(void)
42 /* spinlock for vfsmount related operations, inplace of dcache_lock */
43 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
47 static struct list_head *mount_hashtable;
48 static int hash_mask __read_mostly, hash_bits __read_mostly;
49 static kmem_cache_t *mnt_cache;
50 static struct rw_semaphore namespace_sem;
52 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
54 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
55 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
56 tmp = tmp + (tmp >> hash_bits);
57 return tmp & hash_mask;
60 struct vfsmount *alloc_vfsmnt(const char *name)
62 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
64 memset(mnt, 0, sizeof(struct vfsmount));
65 atomic_set(&mnt->mnt_count, 1);
66 INIT_LIST_HEAD(&mnt->mnt_hash);
67 INIT_LIST_HEAD(&mnt->mnt_child);
68 INIT_LIST_HEAD(&mnt->mnt_mounts);
69 INIT_LIST_HEAD(&mnt->mnt_list);
70 INIT_LIST_HEAD(&mnt->mnt_expire);
71 INIT_LIST_HEAD(&mnt->mnt_share);
73 int size = strlen(name) + 1;
74 char *newname = kmalloc(size, GFP_KERNEL);
76 memcpy(newname, name, size);
77 mnt->mnt_devname = newname;
84 void free_vfsmnt(struct vfsmount *mnt)
86 kfree(mnt->mnt_devname);
87 kmem_cache_free(mnt_cache, mnt);
91 * Now, lookup_mnt increments the ref count before returning
92 * the vfsmount struct.
94 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
96 struct list_head *head = mount_hashtable + hash(mnt, dentry);
97 struct list_head *tmp = head;
98 struct vfsmount *p, *found = NULL;
100 spin_lock(&vfsmount_lock);
106 p = list_entry(tmp, struct vfsmount, mnt_hash);
107 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
112 spin_unlock(&vfsmount_lock);
116 static inline int check_mnt(struct vfsmount *mnt)
118 return mnt->mnt_namespace == current->namespace;
121 static void touch_namespace(struct namespace *ns)
125 wake_up_interruptible(&ns->poll);
129 static void __touch_namespace(struct namespace *ns)
131 if (ns && ns->event != event) {
133 wake_up_interruptible(&ns->poll);
137 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
139 old_nd->dentry = mnt->mnt_mountpoint;
140 old_nd->mnt = mnt->mnt_parent;
141 mnt->mnt_parent = mnt;
142 mnt->mnt_mountpoint = mnt->mnt_root;
143 list_del_init(&mnt->mnt_child);
144 list_del_init(&mnt->mnt_hash);
145 old_nd->dentry->d_mounted--;
148 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
150 mnt->mnt_parent = mntget(nd->mnt);
151 mnt->mnt_mountpoint = dget(nd->dentry);
152 list_add(&mnt->mnt_hash, mount_hashtable + hash(nd->mnt, nd->dentry));
153 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
154 nd->dentry->d_mounted++;
157 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
159 struct list_head *next = p->mnt_mounts.next;
160 if (next == &p->mnt_mounts) {
164 next = p->mnt_child.next;
165 if (next != &p->mnt_parent->mnt_mounts)
170 return list_entry(next, struct vfsmount, mnt_child);
173 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
176 struct super_block *sb = old->mnt_sb;
177 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
180 mnt->mnt_flags = old->mnt_flags;
181 atomic_inc(&sb->s_active);
183 mnt->mnt_root = dget(root);
184 mnt->mnt_mountpoint = mnt->mnt_root;
185 mnt->mnt_parent = mnt;
186 mnt->mnt_namespace = current->namespace;
188 /* stick the duplicate mount on the same expiry list
189 * as the original if that was on one */
190 if (flag & CL_EXPIRE) {
191 spin_lock(&vfsmount_lock);
192 if (!list_empty(&old->mnt_expire))
193 list_add(&mnt->mnt_expire, &old->mnt_expire);
194 spin_unlock(&vfsmount_lock);
200 static inline void __mntput(struct vfsmount *mnt)
202 struct super_block *sb = mnt->mnt_sb;
205 deactivate_super(sb);
208 void mntput_no_expire(struct vfsmount *mnt)
211 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
212 if (likely(!mnt->mnt_pinned)) {
213 spin_unlock(&vfsmount_lock);
217 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
219 spin_unlock(&vfsmount_lock);
220 acct_auto_close_mnt(mnt);
221 security_sb_umount_close(mnt);
226 EXPORT_SYMBOL(mntput_no_expire);
228 void mnt_pin(struct vfsmount *mnt)
230 spin_lock(&vfsmount_lock);
232 spin_unlock(&vfsmount_lock);
235 EXPORT_SYMBOL(mnt_pin);
237 void mnt_unpin(struct vfsmount *mnt)
239 spin_lock(&vfsmount_lock);
240 if (mnt->mnt_pinned) {
241 atomic_inc(&mnt->mnt_count);
244 spin_unlock(&vfsmount_lock);
247 EXPORT_SYMBOL(mnt_unpin);
250 static void *m_start(struct seq_file *m, loff_t *pos)
252 struct namespace *n = m->private;
256 down_read(&namespace_sem);
257 list_for_each(p, &n->list)
259 return list_entry(p, struct vfsmount, mnt_list);
263 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
265 struct namespace *n = m->private;
266 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
268 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
271 static void m_stop(struct seq_file *m, void *v)
273 up_read(&namespace_sem);
276 static inline void mangle(struct seq_file *m, const char *s)
278 seq_escape(m, s, " \t\n\\");
281 static int show_vfsmnt(struct seq_file *m, void *v)
283 struct vfsmount *mnt = v;
285 static struct proc_fs_info {
289 { MS_SYNCHRONOUS, ",sync" },
290 { MS_DIRSYNC, ",dirsync" },
291 { MS_MANDLOCK, ",mand" },
292 { MS_NOATIME, ",noatime" },
293 { MS_NODIRATIME, ",nodiratime" },
296 static struct proc_fs_info mnt_info[] = {
297 { MNT_NOSUID, ",nosuid" },
298 { MNT_NODEV, ",nodev" },
299 { MNT_NOEXEC, ",noexec" },
302 struct proc_fs_info *fs_infop;
304 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
306 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
308 mangle(m, mnt->mnt_sb->s_type->name);
309 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
310 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
311 if (mnt->mnt_sb->s_flags & fs_infop->flag)
312 seq_puts(m, fs_infop->str);
314 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
315 if (mnt->mnt_flags & fs_infop->flag)
316 seq_puts(m, fs_infop->str);
318 if (mnt->mnt_sb->s_op->show_options)
319 err = mnt->mnt_sb->s_op->show_options(m, mnt);
320 seq_puts(m, " 0 0\n");
324 struct seq_operations mounts_op = {
332 * may_umount_tree - check if a mount tree is busy
333 * @mnt: root of mount tree
335 * This is called to check if a tree of mounts has any
336 * open files, pwds, chroots or sub mounts that are
339 int may_umount_tree(struct vfsmount *mnt)
342 int minimum_refs = 0;
345 spin_lock(&vfsmount_lock);
346 for (p = mnt; p; p = next_mnt(p, mnt)) {
347 actual_refs += atomic_read(&p->mnt_count);
350 spin_unlock(&vfsmount_lock);
352 if (actual_refs > minimum_refs)
358 EXPORT_SYMBOL(may_umount_tree);
361 * may_umount - check if a mount point is busy
362 * @mnt: root of mount
364 * This is called to check if a mount point has any
365 * open files, pwds, chroots or sub mounts. If the
366 * mount has sub mounts this will return busy
367 * regardless of whether the sub mounts are busy.
369 * Doesn't take quota and stuff into account. IOW, in some cases it will
370 * give false negatives. The main reason why it's here is that we need
371 * a non-destructive way to look for easily umountable filesystems.
373 int may_umount(struct vfsmount *mnt)
375 if (atomic_read(&mnt->mnt_count) > 2)
380 EXPORT_SYMBOL(may_umount);
382 static void release_mounts(struct list_head *head)
384 struct vfsmount *mnt;
385 while(!list_empty(head)) {
386 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
387 list_del_init(&mnt->mnt_hash);
388 if (mnt->mnt_parent != mnt) {
389 struct dentry *dentry;
391 spin_lock(&vfsmount_lock);
392 dentry = mnt->mnt_mountpoint;
394 mnt->mnt_mountpoint = mnt->mnt_root;
395 mnt->mnt_parent = mnt;
396 spin_unlock(&vfsmount_lock);
404 static void umount_tree(struct vfsmount *mnt, struct list_head *kill)
408 for (p = mnt; p; p = next_mnt(p, mnt)) {
409 list_del(&p->mnt_hash);
410 list_add(&p->mnt_hash, kill);
413 list_for_each_entry(p, kill, mnt_hash) {
414 list_del_init(&p->mnt_expire);
415 list_del_init(&p->mnt_list);
416 __touch_namespace(p->mnt_namespace);
417 p->mnt_namespace = NULL;
418 list_del_init(&p->mnt_child);
419 if (p->mnt_parent != p)
420 mnt->mnt_mountpoint->d_mounted--;
424 static int do_umount(struct vfsmount *mnt, int flags)
426 struct super_block *sb = mnt->mnt_sb;
428 LIST_HEAD(umount_list);
430 retval = security_sb_umount(mnt, flags);
435 * Allow userspace to request a mountpoint be expired rather than
436 * unmounting unconditionally. Unmount only happens if:
437 * (1) the mark is already set (the mark is cleared by mntput())
438 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
440 if (flags & MNT_EXPIRE) {
441 if (mnt == current->fs->rootmnt ||
442 flags & (MNT_FORCE | MNT_DETACH))
445 if (atomic_read(&mnt->mnt_count) != 2)
448 if (!xchg(&mnt->mnt_expiry_mark, 1))
453 * If we may have to abort operations to get out of this
454 * mount, and they will themselves hold resources we must
455 * allow the fs to do things. In the Unix tradition of
456 * 'Gee thats tricky lets do it in userspace' the umount_begin
457 * might fail to complete on the first run through as other tasks
458 * must return, and the like. Thats for the mount program to worry
459 * about for the moment.
463 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
464 sb->s_op->umount_begin(sb);
468 * No sense to grab the lock for this test, but test itself looks
469 * somewhat bogus. Suggestions for better replacement?
470 * Ho-hum... In principle, we might treat that as umount + switch
471 * to rootfs. GC would eventually take care of the old vfsmount.
472 * Actually it makes sense, especially if rootfs would contain a
473 * /reboot - static binary that would close all descriptors and
474 * call reboot(9). Then init(8) could umount root and exec /reboot.
476 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
478 * Special case for "unmounting" root ...
479 * we just try to remount it readonly.
481 down_write(&sb->s_umount);
482 if (!(sb->s_flags & MS_RDONLY)) {
485 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
488 up_write(&sb->s_umount);
492 down_write(&namespace_sem);
493 spin_lock(&vfsmount_lock);
497 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
498 if (!list_empty(&mnt->mnt_list))
499 umount_tree(mnt, &umount_list);
502 spin_unlock(&vfsmount_lock);
504 security_sb_umount_busy(mnt);
505 up_write(&namespace_sem);
506 release_mounts(&umount_list);
511 * Now umount can handle mount points as well as block devices.
512 * This is important for filesystems which use unnamed block devices.
514 * We now support a flag for forced unmount like the other 'big iron'
515 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
518 asmlinkage long sys_umount(char __user * name, int flags)
523 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
527 if (nd.dentry != nd.mnt->mnt_root)
529 if (!check_mnt(nd.mnt))
533 if (!capable(CAP_SYS_ADMIN))
536 retval = do_umount(nd.mnt, flags);
538 path_release_on_umount(&nd);
543 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
546 * The 2.0 compatible umount. No flags.
548 asmlinkage long sys_oldumount(char __user * name)
550 return sys_umount(name, 0);
555 static int mount_is_safe(struct nameidata *nd)
557 if (capable(CAP_SYS_ADMIN))
561 if (S_ISLNK(nd->dentry->d_inode->i_mode))
563 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
564 if (current->uid != nd->dentry->d_inode->i_uid)
567 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
573 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
578 if (d == NULL || d == d->d_parent)
584 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
587 struct vfsmount *res, *p, *q, *r, *s;
590 res = q = clone_mnt(mnt, dentry, flag);
593 q->mnt_mountpoint = mnt->mnt_mountpoint;
596 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
597 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
600 for (s = r; s; s = next_mnt(s, r)) {
601 while (p != s->mnt_parent) {
607 nd.dentry = p->mnt_mountpoint;
608 q = clone_mnt(p, p->mnt_root, flag);
611 spin_lock(&vfsmount_lock);
612 list_add_tail(&q->mnt_list, &res->mnt_list);
614 spin_unlock(&vfsmount_lock);
620 LIST_HEAD(umount_list);
621 spin_lock(&vfsmount_lock);
622 umount_tree(res, &umount_list);
623 spin_unlock(&vfsmount_lock);
624 release_mounts(&umount_list);
629 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
632 if (mnt->mnt_sb->s_flags & MS_NOUSER)
635 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
636 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
640 down(&nd->dentry->d_inode->i_sem);
641 if (IS_DEADDIR(nd->dentry->d_inode))
644 err = security_sb_check_sb(mnt, nd);
649 spin_lock(&vfsmount_lock);
650 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
651 struct list_head head;
654 list_add_tail(&head, &mnt->mnt_list);
655 list_splice(&head, current->namespace->list.prev);
657 touch_namespace(current->namespace);
659 spin_unlock(&vfsmount_lock);
661 up(&nd->dentry->d_inode->i_sem);
663 security_sb_post_addmount(mnt, nd);
668 * recursively change the type of the mountpoint.
670 static int do_change_type(struct nameidata *nd, int flag)
672 struct vfsmount *m, *mnt = nd->mnt;
673 int recurse = flag & MS_REC;
674 int type = flag & ~MS_REC;
676 if (nd->dentry != nd->mnt->mnt_root)
679 down_write(&namespace_sem);
680 spin_lock(&vfsmount_lock);
681 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
682 change_mnt_propagation(m, type);
683 spin_unlock(&vfsmount_lock);
684 up_write(&namespace_sem);
691 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
693 struct nameidata old_nd;
694 struct vfsmount *mnt = NULL;
695 int err = mount_is_safe(nd);
698 if (!old_name || !*old_name)
700 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
704 down_write(&namespace_sem);
706 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
711 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
713 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
718 err = graft_tree(mnt, nd);
720 LIST_HEAD(umount_list);
721 spin_lock(&vfsmount_lock);
722 umount_tree(mnt, &umount_list);
723 spin_unlock(&vfsmount_lock);
724 release_mounts(&umount_list);
728 up_write(&namespace_sem);
729 path_release(&old_nd);
734 * change filesystem flags. dir should be a physical root of filesystem.
735 * If you've mounted a non-root directory somewhere and want to do remount
736 * on it - tough luck.
738 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
742 struct super_block *sb = nd->mnt->mnt_sb;
744 if (!capable(CAP_SYS_ADMIN))
747 if (!check_mnt(nd->mnt))
750 if (nd->dentry != nd->mnt->mnt_root)
753 down_write(&sb->s_umount);
754 err = do_remount_sb(sb, flags, data, 0);
756 nd->mnt->mnt_flags = mnt_flags;
757 up_write(&sb->s_umount);
759 security_sb_post_remount(nd->mnt, flags, data);
763 static int do_move_mount(struct nameidata *nd, char *old_name)
765 struct nameidata old_nd, parent_nd;
768 if (!capable(CAP_SYS_ADMIN))
770 if (!old_name || !*old_name)
772 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
776 down_write(&namespace_sem);
777 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
780 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
784 down(&nd->dentry->d_inode->i_sem);
785 if (IS_DEADDIR(nd->dentry->d_inode))
788 spin_lock(&vfsmount_lock);
789 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
793 if (old_nd.dentry != old_nd.mnt->mnt_root)
796 if (old_nd.mnt == old_nd.mnt->mnt_parent)
799 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
800 S_ISDIR(old_nd.dentry->d_inode->i_mode))
804 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
809 detach_mnt(old_nd.mnt, &parent_nd);
810 attach_mnt(old_nd.mnt, nd);
811 touch_namespace(current->namespace);
813 /* if the mount is moved, it should no longer be expire
815 list_del_init(&old_nd.mnt->mnt_expire);
817 spin_unlock(&vfsmount_lock);
819 up(&nd->dentry->d_inode->i_sem);
821 up_write(&namespace_sem);
823 path_release(&parent_nd);
824 path_release(&old_nd);
829 * create a new mount for userspace and request it to be added into the
832 static int do_new_mount(struct nameidata *nd, char *type, int flags,
833 int mnt_flags, char *name, void *data)
835 struct vfsmount *mnt;
837 if (!type || !memchr(type, 0, PAGE_SIZE))
840 /* we need capabilities... */
841 if (!capable(CAP_SYS_ADMIN))
844 mnt = do_kern_mount(type, flags, name, data);
848 return do_add_mount(mnt, nd, mnt_flags, NULL);
852 * add a mount into a namespace's mount tree
853 * - provide the option of adding the new mount to an expiration list
855 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
856 int mnt_flags, struct list_head *fslist)
860 down_write(&namespace_sem);
861 /* Something was mounted here while we slept */
862 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
865 if (!check_mnt(nd->mnt))
868 /* Refuse the same filesystem on the same mount point */
870 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
871 nd->mnt->mnt_root == nd->dentry)
875 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
878 newmnt->mnt_flags = mnt_flags;
879 if ((err = graft_tree(newmnt, nd)))
883 /* add to the specified expiration list */
884 spin_lock(&vfsmount_lock);
885 list_add_tail(&newmnt->mnt_expire, fslist);
886 spin_unlock(&vfsmount_lock);
888 up_write(&namespace_sem);
892 up_write(&namespace_sem);
897 EXPORT_SYMBOL_GPL(do_add_mount);
899 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
900 struct list_head *umounts)
902 spin_lock(&vfsmount_lock);
905 * Check if mount is still attached, if not, let whoever holds it deal
908 if (mnt->mnt_parent == mnt) {
909 spin_unlock(&vfsmount_lock);
914 * Check that it is still dead: the count should now be 2 - as
915 * contributed by the vfsmount parent and the mntget above
917 if (atomic_read(&mnt->mnt_count) == 2) {
918 /* delete from the namespace */
919 touch_namespace(mnt->mnt_namespace);
920 list_del_init(&mnt->mnt_list);
921 mnt->mnt_namespace = NULL;
922 umount_tree(mnt, umounts);
923 spin_unlock(&vfsmount_lock);
926 * Someone brought it back to life whilst we didn't have any
927 * locks held so return it to the expiration list
929 list_add_tail(&mnt->mnt_expire, mounts);
930 spin_unlock(&vfsmount_lock);
935 * process a list of expirable mountpoints with the intent of discarding any
936 * mountpoints that aren't in use and haven't been touched since last we came
939 void mark_mounts_for_expiry(struct list_head *mounts)
941 struct namespace *namespace;
942 struct vfsmount *mnt, *next;
943 LIST_HEAD(graveyard);
945 if (list_empty(mounts))
948 spin_lock(&vfsmount_lock);
950 /* extract from the expiration list every vfsmount that matches the
951 * following criteria:
952 * - only referenced by its parent vfsmount
953 * - still marked for expiry (marked on the last call here; marks are
954 * cleared by mntput())
956 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
957 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
958 atomic_read(&mnt->mnt_count) != 1)
962 list_move(&mnt->mnt_expire, &graveyard);
966 * go through the vfsmounts we've just consigned to the graveyard to
967 * - check that they're still dead
968 * - delete the vfsmount from the appropriate namespace under lock
969 * - dispose of the corpse
971 while (!list_empty(&graveyard)) {
973 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
974 list_del_init(&mnt->mnt_expire);
976 /* don't do anything if the namespace is dead - all the
977 * vfsmounts from it are going away anyway */
978 namespace = mnt->mnt_namespace;
979 if (!namespace || !namespace->root)
981 get_namespace(namespace);
983 spin_unlock(&vfsmount_lock);
984 down_write(&namespace_sem);
985 expire_mount(mnt, mounts, &umounts);
986 up_write(&namespace_sem);
987 release_mounts(&umounts);
989 put_namespace(namespace);
990 spin_lock(&vfsmount_lock);
993 spin_unlock(&vfsmount_lock);
996 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
999 * Some copy_from_user() implementations do not return the exact number of
1000 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1001 * Note that this function differs from copy_from_user() in that it will oops
1002 * on bad values of `to', rather than returning a short copy.
1004 static long exact_copy_from_user(void *to, const void __user * from,
1008 const char __user *f = from;
1011 if (!access_ok(VERIFY_READ, from, n))
1015 if (__get_user(c, f)) {
1026 int copy_mount_options(const void __user * data, unsigned long *where)
1036 if (!(page = __get_free_page(GFP_KERNEL)))
1039 /* We only care that *some* data at the address the user
1040 * gave us is valid. Just in case, we'll zero
1041 * the remainder of the page.
1043 /* copy_from_user cannot cross TASK_SIZE ! */
1044 size = TASK_SIZE - (unsigned long)data;
1045 if (size > PAGE_SIZE)
1048 i = size - exact_copy_from_user((void *)page, data, size);
1054 memset((char *)page + i, 0, PAGE_SIZE - i);
1060 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1061 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1063 * data is a (void *) that can point to any structure up to
1064 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1065 * information (or be NULL).
1067 * Pre-0.97 versions of mount() didn't have a flags word.
1068 * When the flags word was introduced its top half was required
1069 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1070 * Therefore, if this magic number is present, it carries no information
1071 * and must be discarded.
1073 long do_mount(char *dev_name, char *dir_name, char *type_page,
1074 unsigned long flags, void *data_page)
1076 struct nameidata nd;
1081 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1082 flags &= ~MS_MGC_MSK;
1084 /* Basic sanity checks */
1086 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1088 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1092 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1094 /* Separate the per-mountpoint flags */
1095 if (flags & MS_NOSUID)
1096 mnt_flags |= MNT_NOSUID;
1097 if (flags & MS_NODEV)
1098 mnt_flags |= MNT_NODEV;
1099 if (flags & MS_NOEXEC)
1100 mnt_flags |= MNT_NOEXEC;
1101 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE);
1103 /* ... and get the mountpoint */
1104 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1108 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1112 if (flags & MS_REMOUNT)
1113 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1115 else if (flags & MS_BIND)
1116 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1117 else if (flags & (MS_SHARED | MS_PRIVATE))
1118 retval = do_change_type(&nd, flags);
1119 else if (flags & MS_MOVE)
1120 retval = do_move_mount(&nd, dev_name);
1122 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1123 dev_name, data_page);
1129 int copy_namespace(int flags, struct task_struct *tsk)
1131 struct namespace *namespace = tsk->namespace;
1132 struct namespace *new_ns;
1133 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1134 struct fs_struct *fs = tsk->fs;
1135 struct vfsmount *p, *q;
1140 get_namespace(namespace);
1142 if (!(flags & CLONE_NEWNS))
1145 if (!capable(CAP_SYS_ADMIN)) {
1146 put_namespace(namespace);
1150 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1154 atomic_set(&new_ns->count, 1);
1155 INIT_LIST_HEAD(&new_ns->list);
1156 init_waitqueue_head(&new_ns->poll);
1159 down_write(&namespace_sem);
1160 /* First pass: copy the tree topology */
1161 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1163 if (!new_ns->root) {
1164 up_write(&namespace_sem);
1168 spin_lock(&vfsmount_lock);
1169 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1170 spin_unlock(&vfsmount_lock);
1173 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1174 * as belonging to new namespace. We have already acquired a private
1175 * fs_struct, so tsk->fs->lock is not needed.
1177 p = namespace->root;
1180 q->mnt_namespace = new_ns;
1182 if (p == fs->rootmnt) {
1184 fs->rootmnt = mntget(q);
1186 if (p == fs->pwdmnt) {
1188 fs->pwdmnt = mntget(q);
1190 if (p == fs->altrootmnt) {
1192 fs->altrootmnt = mntget(q);
1195 p = next_mnt(p, namespace->root);
1196 q = next_mnt(q, new_ns->root);
1198 up_write(&namespace_sem);
1200 tsk->namespace = new_ns;
1209 put_namespace(namespace);
1213 put_namespace(namespace);
1217 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1218 char __user * type, unsigned long flags,
1222 unsigned long data_page;
1223 unsigned long type_page;
1224 unsigned long dev_page;
1227 retval = copy_mount_options(type, &type_page);
1231 dir_page = getname(dir_name);
1232 retval = PTR_ERR(dir_page);
1233 if (IS_ERR(dir_page))
1236 retval = copy_mount_options(dev_name, &dev_page);
1240 retval = copy_mount_options(data, &data_page);
1245 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1246 flags, (void *)data_page);
1248 free_page(data_page);
1251 free_page(dev_page);
1255 free_page(type_page);
1260 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1261 * It can block. Requires the big lock held.
1263 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1264 struct dentry *dentry)
1266 struct dentry *old_root;
1267 struct vfsmount *old_rootmnt;
1268 write_lock(&fs->lock);
1269 old_root = fs->root;
1270 old_rootmnt = fs->rootmnt;
1271 fs->rootmnt = mntget(mnt);
1272 fs->root = dget(dentry);
1273 write_unlock(&fs->lock);
1276 mntput(old_rootmnt);
1281 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1282 * It can block. Requires the big lock held.
1284 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1285 struct dentry *dentry)
1287 struct dentry *old_pwd;
1288 struct vfsmount *old_pwdmnt;
1290 write_lock(&fs->lock);
1292 old_pwdmnt = fs->pwdmnt;
1293 fs->pwdmnt = mntget(mnt);
1294 fs->pwd = dget(dentry);
1295 write_unlock(&fs->lock);
1303 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1305 struct task_struct *g, *p;
1306 struct fs_struct *fs;
1308 read_lock(&tasklist_lock);
1309 do_each_thread(g, p) {
1313 atomic_inc(&fs->count);
1315 if (fs->root == old_nd->dentry
1316 && fs->rootmnt == old_nd->mnt)
1317 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1318 if (fs->pwd == old_nd->dentry
1319 && fs->pwdmnt == old_nd->mnt)
1320 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1324 } while_each_thread(g, p);
1325 read_unlock(&tasklist_lock);
1329 * pivot_root Semantics:
1330 * Moves the root file system of the current process to the directory put_old,
1331 * makes new_root as the new root file system of the current process, and sets
1332 * root/cwd of all processes which had them on the current root to new_root.
1335 * The new_root and put_old must be directories, and must not be on the
1336 * same file system as the current process root. The put_old must be
1337 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1338 * pointed to by put_old must yield the same directory as new_root. No other
1339 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1342 * - we don't move root/cwd if they are not at the root (reason: if something
1343 * cared enough to change them, it's probably wrong to force them elsewhere)
1344 * - it's okay to pick a root that isn't the root of a file system, e.g.
1345 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1346 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1349 asmlinkage long sys_pivot_root(const char __user * new_root,
1350 const char __user * put_old)
1352 struct vfsmount *tmp;
1353 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1356 if (!capable(CAP_SYS_ADMIN))
1361 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1366 if (!check_mnt(new_nd.mnt))
1369 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1373 error = security_sb_pivotroot(&old_nd, &new_nd);
1375 path_release(&old_nd);
1379 read_lock(¤t->fs->lock);
1380 user_nd.mnt = mntget(current->fs->rootmnt);
1381 user_nd.dentry = dget(current->fs->root);
1382 read_unlock(¤t->fs->lock);
1383 down_write(&namespace_sem);
1384 down(&old_nd.dentry->d_inode->i_sem);
1386 if (!check_mnt(user_nd.mnt))
1389 if (IS_DEADDIR(new_nd.dentry->d_inode))
1391 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1393 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1396 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1397 goto out2; /* loop, on the same file system */
1399 if (user_nd.mnt->mnt_root != user_nd.dentry)
1400 goto out2; /* not a mountpoint */
1401 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1402 goto out2; /* not attached */
1403 if (new_nd.mnt->mnt_root != new_nd.dentry)
1404 goto out2; /* not a mountpoint */
1405 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1406 goto out2; /* not attached */
1407 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1408 spin_lock(&vfsmount_lock);
1409 if (tmp != new_nd.mnt) {
1411 if (tmp->mnt_parent == tmp)
1412 goto out3; /* already mounted on put_old */
1413 if (tmp->mnt_parent == new_nd.mnt)
1415 tmp = tmp->mnt_parent;
1417 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1419 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1421 detach_mnt(new_nd.mnt, &parent_nd);
1422 detach_mnt(user_nd.mnt, &root_parent);
1423 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1424 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1425 touch_namespace(current->namespace);
1426 spin_unlock(&vfsmount_lock);
1427 chroot_fs_refs(&user_nd, &new_nd);
1428 security_sb_post_pivotroot(&user_nd, &new_nd);
1430 path_release(&root_parent);
1431 path_release(&parent_nd);
1433 up(&old_nd.dentry->d_inode->i_sem);
1434 up_write(&namespace_sem);
1435 path_release(&user_nd);
1436 path_release(&old_nd);
1438 path_release(&new_nd);
1443 spin_unlock(&vfsmount_lock);
1447 static void __init init_mount_tree(void)
1449 struct vfsmount *mnt;
1450 struct namespace *namespace;
1451 struct task_struct *g, *p;
1453 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1455 panic("Can't create rootfs");
1456 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1458 panic("Can't allocate initial namespace");
1459 atomic_set(&namespace->count, 1);
1460 INIT_LIST_HEAD(&namespace->list);
1461 init_waitqueue_head(&namespace->poll);
1462 namespace->event = 0;
1463 list_add(&mnt->mnt_list, &namespace->list);
1464 namespace->root = mnt;
1465 mnt->mnt_namespace = namespace;
1467 init_task.namespace = namespace;
1468 read_lock(&tasklist_lock);
1469 do_each_thread(g, p) {
1470 get_namespace(namespace);
1471 p->namespace = namespace;
1472 } while_each_thread(g, p);
1473 read_unlock(&tasklist_lock);
1475 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1476 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1479 void __init mnt_init(unsigned long mempages)
1481 struct list_head *d;
1482 unsigned int nr_hash;
1485 init_rwsem(&namespace_sem);
1487 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1488 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1490 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1492 if (!mount_hashtable)
1493 panic("Failed to allocate mount hash table\n");
1496 * Find the power-of-two list-heads that can fit into the allocation..
1497 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1500 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1504 } while ((nr_hash >> hash_bits) != 0);
1508 * Re-calculate the actual number of entries and the mask
1509 * from the number of bits we can fit.
1511 nr_hash = 1UL << hash_bits;
1512 hash_mask = nr_hash - 1;
1514 printk("Mount-cache hash table entries: %d\n", nr_hash);
1516 /* And initialize the newly allocated array */
1517 d = mount_hashtable;
1529 void __put_namespace(struct namespace *namespace)
1531 struct vfsmount *root = namespace->root;
1532 LIST_HEAD(umount_list);
1533 namespace->root = NULL;
1534 spin_unlock(&vfsmount_lock);
1535 down_write(&namespace_sem);
1536 spin_lock(&vfsmount_lock);
1537 umount_tree(root, &umount_list);
1538 spin_unlock(&vfsmount_lock);
1539 up_write(&namespace_sem);
1540 release_mounts(&umount_list);