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>
28 extern int __init init_rootfs(void);
31 extern int __init sysfs_init(void);
33 static inline int sysfs_init(void)
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42 static struct list_head *mount_hashtable;
43 static int hash_mask, hash_bits;
44 static kmem_cache_t *mnt_cache;
46 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
48 unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
49 tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
50 tmp = tmp + (tmp >> hash_bits);
51 return tmp & hash_mask;
54 struct vfsmount *alloc_vfsmnt(const char *name)
56 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
58 memset(mnt, 0, sizeof(struct vfsmount));
59 atomic_set(&mnt->mnt_count,1);
60 INIT_LIST_HEAD(&mnt->mnt_hash);
61 INIT_LIST_HEAD(&mnt->mnt_child);
62 INIT_LIST_HEAD(&mnt->mnt_mounts);
63 INIT_LIST_HEAD(&mnt->mnt_list);
64 INIT_LIST_HEAD(&mnt->mnt_expire);
66 int size = strlen(name)+1;
67 char *newname = kmalloc(size, GFP_KERNEL);
69 memcpy(newname, name, size);
70 mnt->mnt_devname = newname;
77 void free_vfsmnt(struct vfsmount *mnt)
79 kfree(mnt->mnt_devname);
80 kmem_cache_free(mnt_cache, mnt);
84 * Now, lookup_mnt increments the ref count before returning
85 * the vfsmount struct.
87 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
89 struct list_head * head = mount_hashtable + hash(mnt, dentry);
90 struct list_head * tmp = head;
91 struct vfsmount *p, *found = NULL;
93 spin_lock(&vfsmount_lock);
99 p = list_entry(tmp, struct vfsmount, mnt_hash);
100 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
105 spin_unlock(&vfsmount_lock);
109 static inline int check_mnt(struct vfsmount *mnt)
111 return mnt->mnt_namespace == current->namespace;
114 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
116 old_nd->dentry = mnt->mnt_mountpoint;
117 old_nd->mnt = mnt->mnt_parent;
118 mnt->mnt_parent = mnt;
119 mnt->mnt_mountpoint = mnt->mnt_root;
120 list_del_init(&mnt->mnt_child);
121 list_del_init(&mnt->mnt_hash);
122 old_nd->dentry->d_mounted--;
125 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
127 mnt->mnt_parent = mntget(nd->mnt);
128 mnt->mnt_mountpoint = dget(nd->dentry);
129 list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
130 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
131 nd->dentry->d_mounted++;
134 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
136 struct list_head *next = p->mnt_mounts.next;
137 if (next == &p->mnt_mounts) {
141 next = p->mnt_child.next;
142 if (next != &p->mnt_parent->mnt_mounts)
147 return list_entry(next, struct vfsmount, mnt_child);
150 static struct vfsmount *
151 clone_mnt(struct vfsmount *old, struct dentry *root)
153 struct super_block *sb = old->mnt_sb;
154 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
157 mnt->mnt_flags = old->mnt_flags;
158 atomic_inc(&sb->s_active);
160 mnt->mnt_root = dget(root);
161 mnt->mnt_mountpoint = mnt->mnt_root;
162 mnt->mnt_parent = mnt;
163 mnt->mnt_namespace = old->mnt_namespace;
165 /* stick the duplicate mount on the same expiry list
166 * as the original if that was on one */
167 spin_lock(&vfsmount_lock);
168 if (!list_empty(&old->mnt_expire))
169 list_add(&mnt->mnt_expire, &old->mnt_expire);
170 spin_unlock(&vfsmount_lock);
175 void __mntput(struct vfsmount *mnt)
177 struct super_block *sb = mnt->mnt_sb;
180 deactivate_super(sb);
183 EXPORT_SYMBOL(__mntput);
186 static void *m_start(struct seq_file *m, loff_t *pos)
188 struct namespace *n = m->private;
193 list_for_each(p, &n->list)
195 return list_entry(p, struct vfsmount, mnt_list);
199 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
201 struct namespace *n = m->private;
202 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
204 return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
207 static void m_stop(struct seq_file *m, void *v)
209 struct namespace *n = m->private;
213 static inline void mangle(struct seq_file *m, const char *s)
215 seq_escape(m, s, " \t\n\\");
218 static int show_vfsmnt(struct seq_file *m, void *v)
220 struct vfsmount *mnt = v;
222 static struct proc_fs_info {
226 { MS_SYNCHRONOUS, ",sync" },
227 { MS_DIRSYNC, ",dirsync" },
228 { MS_MANDLOCK, ",mand" },
229 { MS_NOATIME, ",noatime" },
230 { MS_NODIRATIME, ",nodiratime" },
233 static struct proc_fs_info mnt_info[] = {
234 { MNT_NOSUID, ",nosuid" },
235 { MNT_NODEV, ",nodev" },
236 { MNT_NOEXEC, ",noexec" },
239 struct proc_fs_info *fs_infop;
241 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
243 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
245 mangle(m, mnt->mnt_sb->s_type->name);
246 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
247 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
248 if (mnt->mnt_sb->s_flags & fs_infop->flag)
249 seq_puts(m, fs_infop->str);
251 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
252 if (mnt->mnt_flags & fs_infop->flag)
253 seq_puts(m, fs_infop->str);
255 if (mnt->mnt_sb->s_op->show_options)
256 err = mnt->mnt_sb->s_op->show_options(m, mnt);
257 seq_puts(m, " 0 0\n");
261 struct seq_operations mounts_op = {
269 * may_umount_tree - check if a mount tree is busy
270 * @mnt: root of mount tree
272 * This is called to check if a tree of mounts has any
273 * open files, pwds, chroots or sub mounts that are
276 int may_umount_tree(struct vfsmount *mnt)
278 struct list_head *next;
279 struct vfsmount *this_parent = mnt;
283 spin_lock(&vfsmount_lock);
284 actual_refs = atomic_read(&mnt->mnt_count);
287 next = this_parent->mnt_mounts.next;
289 while (next != &this_parent->mnt_mounts) {
290 struct vfsmount *p = list_entry(next, struct vfsmount, mnt_child);
294 actual_refs += atomic_read(&p->mnt_count);
297 if (!list_empty(&p->mnt_mounts)) {
303 if (this_parent != mnt) {
304 next = this_parent->mnt_child.next;
305 this_parent = this_parent->mnt_parent;
308 spin_unlock(&vfsmount_lock);
310 if (actual_refs > minimum_refs)
316 EXPORT_SYMBOL(may_umount_tree);
319 * may_umount - check if a mount point is busy
320 * @mnt: root of mount
322 * This is called to check if a mount point has any
323 * open files, pwds, chroots or sub mounts. If the
324 * mount has sub mounts this will return busy
325 * regardless of whether the sub mounts are busy.
327 * Doesn't take quota and stuff into account. IOW, in some cases it will
328 * give false negatives. The main reason why it's here is that we need
329 * a non-destructive way to look for easily umountable filesystems.
331 int may_umount(struct vfsmount *mnt)
333 if (atomic_read(&mnt->mnt_count) > 2)
338 EXPORT_SYMBOL(may_umount);
340 static void umount_tree(struct vfsmount *mnt)
345 for (p = mnt; p; p = next_mnt(p, mnt)) {
346 list_del(&p->mnt_list);
347 list_add(&p->mnt_list, &kill);
348 p->mnt_namespace = NULL;
351 while (!list_empty(&kill)) {
352 mnt = list_entry(kill.next, struct vfsmount, mnt_list);
353 list_del_init(&mnt->mnt_list);
354 list_del_init(&mnt->mnt_expire);
355 if (mnt->mnt_parent == mnt) {
356 spin_unlock(&vfsmount_lock);
358 struct nameidata old_nd;
359 detach_mnt(mnt, &old_nd);
360 spin_unlock(&vfsmount_lock);
361 path_release(&old_nd);
364 spin_lock(&vfsmount_lock);
368 static int do_umount(struct vfsmount *mnt, int flags)
370 struct super_block * sb = mnt->mnt_sb;
373 retval = security_sb_umount(mnt, flags);
378 * Allow userspace to request a mountpoint be expired rather than
379 * unmounting unconditionally. Unmount only happens if:
380 * (1) the mark is already set (the mark is cleared by mntput())
381 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
383 if (flags & MNT_EXPIRE) {
384 if (mnt == current->fs->rootmnt ||
385 flags & (MNT_FORCE | MNT_DETACH))
388 if (atomic_read(&mnt->mnt_count) != 2)
391 if (!xchg(&mnt->mnt_expiry_mark, 1))
396 * If we may have to abort operations to get out of this
397 * mount, and they will themselves hold resources we must
398 * allow the fs to do things. In the Unix tradition of
399 * 'Gee thats tricky lets do it in userspace' the umount_begin
400 * might fail to complete on the first run through as other tasks
401 * must return, and the like. Thats for the mount program to worry
402 * about for the moment.
406 if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
407 sb->s_op->umount_begin(sb);
411 * No sense to grab the lock for this test, but test itself looks
412 * somewhat bogus. Suggestions for better replacement?
413 * Ho-hum... In principle, we might treat that as umount + switch
414 * to rootfs. GC would eventually take care of the old vfsmount.
415 * Actually it makes sense, especially if rootfs would contain a
416 * /reboot - static binary that would close all descriptors and
417 * call reboot(9). Then init(8) could umount root and exec /reboot.
419 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
421 * Special case for "unmounting" root ...
422 * we just try to remount it readonly.
424 down_write(&sb->s_umount);
425 if (!(sb->s_flags & MS_RDONLY)) {
428 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
431 up_write(&sb->s_umount);
435 down_write(¤t->namespace->sem);
436 spin_lock(&vfsmount_lock);
438 if (atomic_read(&sb->s_active) == 1) {
439 /* last instance - try to be smart */
440 spin_unlock(&vfsmount_lock);
445 security_sb_umount_close(mnt);
446 spin_lock(&vfsmount_lock);
449 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
450 if (!list_empty(&mnt->mnt_list))
454 spin_unlock(&vfsmount_lock);
456 security_sb_umount_busy(mnt);
457 up_write(¤t->namespace->sem);
462 * Now umount can handle mount points as well as block devices.
463 * This is important for filesystems which use unnamed block devices.
465 * We now support a flag for forced unmount like the other 'big iron'
466 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
469 asmlinkage long sys_umount(char __user * name, int flags)
474 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
478 if (nd.dentry != nd.mnt->mnt_root)
480 if (!check_mnt(nd.mnt))
484 if (!capable(CAP_SYS_ADMIN))
487 retval = do_umount(nd.mnt, flags);
489 path_release_on_umount(&nd);
494 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
497 * The 2.0 compatible umount. No flags.
500 asmlinkage long sys_oldumount(char __user * name)
502 return sys_umount(name,0);
507 static int mount_is_safe(struct nameidata *nd)
509 if (capable(CAP_SYS_ADMIN))
513 if (S_ISLNK(nd->dentry->d_inode->i_mode))
515 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
516 if (current->uid != nd->dentry->d_inode->i_uid)
519 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
526 lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
531 if (d == NULL || d == d->d_parent)
537 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
539 struct vfsmount *res, *p, *q, *r, *s;
543 res = q = clone_mnt(mnt, dentry);
546 q->mnt_mountpoint = mnt->mnt_mountpoint;
549 for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) {
550 r = list_entry(h, struct vfsmount, mnt_child);
551 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
554 for (s = r; s; s = next_mnt(s, r)) {
555 while (p != s->mnt_parent) {
561 nd.dentry = p->mnt_mountpoint;
562 q = clone_mnt(p, p->mnt_root);
565 spin_lock(&vfsmount_lock);
566 list_add_tail(&q->mnt_list, &res->mnt_list);
568 spin_unlock(&vfsmount_lock);
574 spin_lock(&vfsmount_lock);
576 spin_unlock(&vfsmount_lock);
581 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
584 if (mnt->mnt_sb->s_flags & MS_NOUSER)
587 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
588 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
592 down(&nd->dentry->d_inode->i_sem);
593 if (IS_DEADDIR(nd->dentry->d_inode))
596 err = security_sb_check_sb(mnt, nd);
601 spin_lock(&vfsmount_lock);
602 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
603 struct list_head head;
606 list_add_tail(&head, &mnt->mnt_list);
607 list_splice(&head, current->namespace->list.prev);
611 spin_unlock(&vfsmount_lock);
613 up(&nd->dentry->d_inode->i_sem);
615 security_sb_post_addmount(mnt, nd);
622 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
624 struct nameidata old_nd;
625 struct vfsmount *mnt = NULL;
626 int err = mount_is_safe(nd);
629 if (!old_name || !*old_name)
631 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
635 down_write(¤t->namespace->sem);
637 if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
640 mnt = copy_tree(old_nd.mnt, old_nd.dentry);
642 mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
646 /* stop bind mounts from expiring */
647 spin_lock(&vfsmount_lock);
648 list_del_init(&mnt->mnt_expire);
649 spin_unlock(&vfsmount_lock);
651 err = graft_tree(mnt, nd);
653 spin_lock(&vfsmount_lock);
655 spin_unlock(&vfsmount_lock);
660 up_write(¤t->namespace->sem);
661 path_release(&old_nd);
666 * change filesystem flags. dir should be a physical root of filesystem.
667 * If you've mounted a non-root directory somewhere and want to do remount
668 * on it - tough luck.
671 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
675 struct super_block * sb = nd->mnt->mnt_sb;
677 if (!capable(CAP_SYS_ADMIN))
680 if (!check_mnt(nd->mnt))
683 if (nd->dentry != nd->mnt->mnt_root)
686 down_write(&sb->s_umount);
687 err = do_remount_sb(sb, flags, data, 0);
689 nd->mnt->mnt_flags=mnt_flags;
690 up_write(&sb->s_umount);
692 security_sb_post_remount(nd->mnt, flags, data);
696 static int do_move_mount(struct nameidata *nd, char *old_name)
698 struct nameidata old_nd, parent_nd;
701 if (!capable(CAP_SYS_ADMIN))
703 if (!old_name || !*old_name)
705 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
709 down_write(¤t->namespace->sem);
710 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
713 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
717 down(&nd->dentry->d_inode->i_sem);
718 if (IS_DEADDIR(nd->dentry->d_inode))
721 spin_lock(&vfsmount_lock);
722 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
726 if (old_nd.dentry != old_nd.mnt->mnt_root)
729 if (old_nd.mnt == old_nd.mnt->mnt_parent)
732 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
733 S_ISDIR(old_nd.dentry->d_inode->i_mode))
737 for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
742 detach_mnt(old_nd.mnt, &parent_nd);
743 attach_mnt(old_nd.mnt, nd);
745 /* if the mount is moved, it should no longer be expire
747 list_del_init(&old_nd.mnt->mnt_expire);
749 spin_unlock(&vfsmount_lock);
751 up(&nd->dentry->d_inode->i_sem);
753 up_write(¤t->namespace->sem);
755 path_release(&parent_nd);
756 path_release(&old_nd);
761 * create a new mount for userspace and request it to be added into the
764 static int do_new_mount(struct nameidata *nd, char *type, int flags,
765 int mnt_flags, char *name, void *data)
767 struct vfsmount *mnt;
769 if (!type || !memchr(type, 0, PAGE_SIZE))
772 /* we need capabilities... */
773 if (!capable(CAP_SYS_ADMIN))
776 mnt = do_kern_mount(type, flags, name, data);
780 return do_add_mount(mnt, nd, mnt_flags, NULL);
784 * add a mount into a namespace's mount tree
785 * - provide the option of adding the new mount to an expiration list
787 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
788 int mnt_flags, struct list_head *fslist)
792 down_write(¤t->namespace->sem);
793 /* Something was mounted here while we slept */
794 while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
797 if (!check_mnt(nd->mnt))
800 /* Refuse the same filesystem on the same mount point */
802 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
803 nd->mnt->mnt_root == nd->dentry)
807 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
810 newmnt->mnt_flags = mnt_flags;
811 newmnt->mnt_namespace = current->namespace;
812 err = graft_tree(newmnt, nd);
814 if (err == 0 && fslist) {
815 /* add to the specified expiration list */
816 spin_lock(&vfsmount_lock);
817 list_add_tail(&newmnt->mnt_expire, fslist);
818 spin_unlock(&vfsmount_lock);
822 up_write(¤t->namespace->sem);
827 EXPORT_SYMBOL_GPL(do_add_mount);
829 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts)
831 spin_lock(&vfsmount_lock);
834 * Check if mount is still attached, if not, let whoever holds it deal
837 if (mnt->mnt_parent == mnt) {
838 spin_unlock(&vfsmount_lock);
843 * Check that it is still dead: the count should now be 2 - as
844 * contributed by the vfsmount parent and the mntget above
846 if (atomic_read(&mnt->mnt_count) == 2) {
847 struct nameidata old_nd;
849 /* delete from the namespace */
850 list_del_init(&mnt->mnt_list);
851 mnt->mnt_namespace = NULL;
852 detach_mnt(mnt, &old_nd);
853 spin_unlock(&vfsmount_lock);
854 path_release(&old_nd);
857 * Now lay it to rest if this was the last ref on the superblock
859 if (atomic_read(&mnt->mnt_sb->s_active) == 1) {
860 /* last instance - try to be smart */
862 DQUOT_OFF(mnt->mnt_sb);
863 acct_auto_close(mnt->mnt_sb);
869 * Someone brought it back to life whilst we didn't have any
870 * locks held so return it to the expiration list
872 list_add_tail(&mnt->mnt_expire, mounts);
873 spin_unlock(&vfsmount_lock);
878 * process a list of expirable mountpoints with the intent of discarding any
879 * mountpoints that aren't in use and haven't been touched since last we came
882 void mark_mounts_for_expiry(struct list_head *mounts)
884 struct namespace *namespace;
885 struct vfsmount *mnt, *next;
886 LIST_HEAD(graveyard);
888 if (list_empty(mounts))
891 spin_lock(&vfsmount_lock);
893 /* extract from the expiration list every vfsmount that matches the
894 * following criteria:
895 * - only referenced by its parent vfsmount
896 * - still marked for expiry (marked on the last call here; marks are
897 * cleared by mntput())
899 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
900 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
901 atomic_read(&mnt->mnt_count) != 1)
905 list_move(&mnt->mnt_expire, &graveyard);
909 * go through the vfsmounts we've just consigned to the graveyard to
910 * - check that they're still dead
911 * - delete the vfsmount from the appropriate namespace under lock
912 * - dispose of the corpse
914 while (!list_empty(&graveyard)) {
915 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
916 list_del_init(&mnt->mnt_expire);
918 /* don't do anything if the namespace is dead - all the
919 * vfsmounts from it are going away anyway */
920 namespace = mnt->mnt_namespace;
921 if (!namespace || !namespace->root)
923 get_namespace(namespace);
925 spin_unlock(&vfsmount_lock);
926 down_write(&namespace->sem);
927 expire_mount(mnt, mounts);
928 up_write(&namespace->sem);
931 put_namespace(namespace);
933 spin_lock(&vfsmount_lock);
936 spin_unlock(&vfsmount_lock);
939 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
942 * Some copy_from_user() implementations do not return the exact number of
943 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
944 * Note that this function differs from copy_from_user() in that it will oops
945 * on bad values of `to', rather than returning a short copy.
948 exact_copy_from_user(void *to, const void __user *from, unsigned long n)
951 const char __user *f = from;
954 if (!access_ok(VERIFY_READ, from, n))
958 if (__get_user(c, f)) {
969 int copy_mount_options(const void __user *data, unsigned long *where)
979 if (!(page = __get_free_page(GFP_KERNEL)))
982 /* We only care that *some* data at the address the user
983 * gave us is valid. Just in case, we'll zero
984 * the remainder of the page.
986 /* copy_from_user cannot cross TASK_SIZE ! */
987 size = TASK_SIZE - (unsigned long)data;
988 if (size > PAGE_SIZE)
991 i = size - exact_copy_from_user((void *)page, data, size);
997 memset((char *)page + i, 0, PAGE_SIZE - i);
1003 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1004 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1006 * data is a (void *) that can point to any structure up to
1007 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1008 * information (or be NULL).
1010 * Pre-0.97 versions of mount() didn't have a flags word.
1011 * When the flags word was introduced its top half was required
1012 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1013 * Therefore, if this magic number is present, it carries no information
1014 * and must be discarded.
1016 long do_mount(char * dev_name, char * dir_name, char *type_page,
1017 unsigned long flags, void *data_page)
1019 struct nameidata nd;
1024 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1025 flags &= ~MS_MGC_MSK;
1027 /* Basic sanity checks */
1029 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1031 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1035 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1037 /* Separate the per-mountpoint flags */
1038 if (flags & MS_NOSUID)
1039 mnt_flags |= MNT_NOSUID;
1040 if (flags & MS_NODEV)
1041 mnt_flags |= MNT_NODEV;
1042 if (flags & MS_NOEXEC)
1043 mnt_flags |= MNT_NOEXEC;
1044 flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_ACTIVE);
1046 /* ... and get the mountpoint */
1047 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1051 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1055 if (flags & MS_REMOUNT)
1056 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1058 else if (flags & MS_BIND)
1059 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1060 else if (flags & MS_MOVE)
1061 retval = do_move_mount(&nd, dev_name);
1063 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1064 dev_name, data_page);
1070 int copy_namespace(int flags, struct task_struct *tsk)
1072 struct namespace *namespace = tsk->namespace;
1073 struct namespace *new_ns;
1074 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1075 struct fs_struct *fs = tsk->fs;
1076 struct vfsmount *p, *q;
1081 get_namespace(namespace);
1083 if (!(flags & CLONE_NEWNS))
1086 if (!capable(CAP_SYS_ADMIN)) {
1087 put_namespace(namespace);
1091 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1095 atomic_set(&new_ns->count, 1);
1096 init_rwsem(&new_ns->sem);
1097 INIT_LIST_HEAD(&new_ns->list);
1099 down_write(&tsk->namespace->sem);
1100 /* First pass: copy the tree topology */
1101 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
1102 if (!new_ns->root) {
1103 up_write(&tsk->namespace->sem);
1107 spin_lock(&vfsmount_lock);
1108 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1109 spin_unlock(&vfsmount_lock);
1112 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1113 * as belonging to new namespace. We have already acquired a private
1114 * fs_struct, so tsk->fs->lock is not needed.
1116 p = namespace->root;
1119 q->mnt_namespace = new_ns;
1121 if (p == fs->rootmnt) {
1123 fs->rootmnt = mntget(q);
1125 if (p == fs->pwdmnt) {
1127 fs->pwdmnt = mntget(q);
1129 if (p == fs->altrootmnt) {
1131 fs->altrootmnt = mntget(q);
1134 p = next_mnt(p, namespace->root);
1135 q = next_mnt(q, new_ns->root);
1137 up_write(&tsk->namespace->sem);
1139 tsk->namespace = new_ns;
1148 put_namespace(namespace);
1152 put_namespace(namespace);
1156 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1157 char __user * type, unsigned long flags,
1161 unsigned long data_page;
1162 unsigned long type_page;
1163 unsigned long dev_page;
1166 retval = copy_mount_options (type, &type_page);
1170 dir_page = getname(dir_name);
1171 retval = PTR_ERR(dir_page);
1172 if (IS_ERR(dir_page))
1175 retval = copy_mount_options (dev_name, &dev_page);
1179 retval = copy_mount_options (data, &data_page);
1184 retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
1185 flags, (void*)data_page);
1187 free_page(data_page);
1190 free_page(dev_page);
1194 free_page(type_page);
1199 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1200 * It can block. Requires the big lock held.
1202 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1203 struct dentry *dentry)
1205 struct dentry *old_root;
1206 struct vfsmount *old_rootmnt;
1207 write_lock(&fs->lock);
1208 old_root = fs->root;
1209 old_rootmnt = fs->rootmnt;
1210 fs->rootmnt = mntget(mnt);
1211 fs->root = dget(dentry);
1212 write_unlock(&fs->lock);
1215 mntput(old_rootmnt);
1220 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1221 * It can block. Requires the big lock held.
1223 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1224 struct dentry *dentry)
1226 struct dentry *old_pwd;
1227 struct vfsmount *old_pwdmnt;
1229 write_lock(&fs->lock);
1231 old_pwdmnt = fs->pwdmnt;
1232 fs->pwdmnt = mntget(mnt);
1233 fs->pwd = dget(dentry);
1234 write_unlock(&fs->lock);
1242 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1244 struct task_struct *g, *p;
1245 struct fs_struct *fs;
1247 read_lock(&tasklist_lock);
1248 do_each_thread(g, p) {
1252 atomic_inc(&fs->count);
1254 if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
1255 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1256 if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
1257 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1261 } while_each_thread(g, p);
1262 read_unlock(&tasklist_lock);
1266 * pivot_root Semantics:
1267 * Moves the root file system of the current process to the directory put_old,
1268 * makes new_root as the new root file system of the current process, and sets
1269 * root/cwd of all processes which had them on the current root to new_root.
1272 * The new_root and put_old must be directories, and must not be on the
1273 * same file system as the current process root. The put_old must be
1274 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1275 * pointed to by put_old must yield the same directory as new_root. No other
1276 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1279 * - we don't move root/cwd if they are not at the root (reason: if something
1280 * cared enough to change them, it's probably wrong to force them elsewhere)
1281 * - it's okay to pick a root that isn't the root of a file system, e.g.
1282 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1283 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1287 asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
1289 struct vfsmount *tmp;
1290 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1293 if (!capable(CAP_SYS_ADMIN))
1298 error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
1302 if (!check_mnt(new_nd.mnt))
1305 error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
1309 error = security_sb_pivotroot(&old_nd, &new_nd);
1311 path_release(&old_nd);
1315 read_lock(¤t->fs->lock);
1316 user_nd.mnt = mntget(current->fs->rootmnt);
1317 user_nd.dentry = dget(current->fs->root);
1318 read_unlock(¤t->fs->lock);
1319 down_write(¤t->namespace->sem);
1320 down(&old_nd.dentry->d_inode->i_sem);
1322 if (!check_mnt(user_nd.mnt))
1325 if (IS_DEADDIR(new_nd.dentry->d_inode))
1327 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1329 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1332 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1333 goto out2; /* loop, on the same file system */
1335 if (user_nd.mnt->mnt_root != user_nd.dentry)
1336 goto out2; /* not a mountpoint */
1337 if (new_nd.mnt->mnt_root != new_nd.dentry)
1338 goto out2; /* not a mountpoint */
1339 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1340 spin_lock(&vfsmount_lock);
1341 if (tmp != new_nd.mnt) {
1343 if (tmp->mnt_parent == tmp)
1344 goto out3; /* already mounted on put_old */
1345 if (tmp->mnt_parent == new_nd.mnt)
1347 tmp = tmp->mnt_parent;
1349 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1351 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1353 detach_mnt(new_nd.mnt, &parent_nd);
1354 detach_mnt(user_nd.mnt, &root_parent);
1355 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1356 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1357 spin_unlock(&vfsmount_lock);
1358 chroot_fs_refs(&user_nd, &new_nd);
1359 security_sb_post_pivotroot(&user_nd, &new_nd);
1361 path_release(&root_parent);
1362 path_release(&parent_nd);
1364 up(&old_nd.dentry->d_inode->i_sem);
1365 up_write(¤t->namespace->sem);
1366 path_release(&user_nd);
1367 path_release(&old_nd);
1369 path_release(&new_nd);
1374 spin_unlock(&vfsmount_lock);
1378 static void __init init_mount_tree(void)
1380 struct vfsmount *mnt;
1381 struct namespace *namespace;
1382 struct task_struct *g, *p;
1384 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1386 panic("Can't create rootfs");
1387 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1389 panic("Can't allocate initial namespace");
1390 atomic_set(&namespace->count, 1);
1391 INIT_LIST_HEAD(&namespace->list);
1392 init_rwsem(&namespace->sem);
1393 list_add(&mnt->mnt_list, &namespace->list);
1394 namespace->root = mnt;
1395 mnt->mnt_namespace = namespace;
1397 init_task.namespace = namespace;
1398 read_lock(&tasklist_lock);
1399 do_each_thread(g, p) {
1400 get_namespace(namespace);
1401 p->namespace = namespace;
1402 } while_each_thread(g, p);
1403 read_unlock(&tasklist_lock);
1405 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1406 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1409 void __init mnt_init(unsigned long mempages)
1411 struct list_head *d;
1412 unsigned int nr_hash;
1415 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1416 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1418 mount_hashtable = (struct list_head *)
1419 __get_free_page(GFP_ATOMIC);
1421 if (!mount_hashtable)
1422 panic("Failed to allocate mount hash table\n");
1425 * Find the power-of-two list-heads that can fit into the allocation..
1426 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1429 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1433 } while ((nr_hash >> hash_bits) != 0);
1437 * Re-calculate the actual number of entries and the mask
1438 * from the number of bits we can fit.
1440 nr_hash = 1UL << hash_bits;
1441 hash_mask = nr_hash-1;
1443 printk("Mount-cache hash table entries: %d\n", nr_hash);
1445 /* And initialize the newly allocated array */
1446 d = mount_hashtable;
1458 void __put_namespace(struct namespace *namespace)
1460 struct vfsmount *root = namespace->root;
1461 namespace->root = NULL;
1462 spin_unlock(&vfsmount_lock);
1463 down_write(&namespace->sem);
1464 spin_lock(&vfsmount_lock);
1466 spin_unlock(&vfsmount_lock);
1467 up_write(&namespace->sem);