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);
72 int size = strlen(name) + 1;
73 char *newname = kmalloc(size, GFP_KERNEL);
75 memcpy(newname, name, size);
76 mnt->mnt_devname = newname;
83 void free_vfsmnt(struct vfsmount *mnt)
85 kfree(mnt->mnt_devname);
86 kmem_cache_free(mnt_cache, mnt);
90 * Now, lookup_mnt increments the ref count before returning
91 * the vfsmount struct.
93 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
95 struct list_head *head = mount_hashtable + hash(mnt, dentry);
96 struct list_head *tmp = head;
97 struct vfsmount *p, *found = NULL;
99 spin_lock(&vfsmount_lock);
105 p = list_entry(tmp, struct vfsmount, mnt_hash);
106 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
111 spin_unlock(&vfsmount_lock);
115 static inline int check_mnt(struct vfsmount *mnt)
117 return mnt->mnt_namespace == current->namespace;
120 static void touch_namespace(struct namespace *ns)
124 wake_up_interruptible(&ns->poll);
128 static void __touch_namespace(struct namespace *ns)
130 if (ns && ns->event != event) {
132 wake_up_interruptible(&ns->poll);
136 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
138 old_nd->dentry = mnt->mnt_mountpoint;
139 old_nd->mnt = mnt->mnt_parent;
140 mnt->mnt_parent = mnt;
141 mnt->mnt_mountpoint = mnt->mnt_root;
142 list_del_init(&mnt->mnt_child);
143 list_del_init(&mnt->mnt_hash);
144 old_nd->dentry->d_mounted--;
147 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
149 mnt->mnt_parent = mntget(nd->mnt);
150 mnt->mnt_mountpoint = dget(nd->dentry);
151 list_add(&mnt->mnt_hash, mount_hashtable + hash(nd->mnt, nd->dentry));
152 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
153 nd->dentry->d_mounted++;
156 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
158 struct list_head *next = p->mnt_mounts.next;
159 if (next == &p->mnt_mounts) {
163 next = p->mnt_child.next;
164 if (next != &p->mnt_parent->mnt_mounts)
169 return list_entry(next, struct vfsmount, mnt_child);
172 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
175 struct super_block *sb = old->mnt_sb;
176 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
179 mnt->mnt_flags = old->mnt_flags;
180 atomic_inc(&sb->s_active);
182 mnt->mnt_root = dget(root);
183 mnt->mnt_mountpoint = mnt->mnt_root;
184 mnt->mnt_parent = mnt;
185 mnt->mnt_namespace = current->namespace;
187 /* stick the duplicate mount on the same expiry list
188 * as the original if that was on one */
189 if (flag & CL_EXPIRE) {
190 spin_lock(&vfsmount_lock);
191 if (!list_empty(&old->mnt_expire))
192 list_add(&mnt->mnt_expire, &old->mnt_expire);
193 spin_unlock(&vfsmount_lock);
199 static inline void __mntput(struct vfsmount *mnt)
201 struct super_block *sb = mnt->mnt_sb;
204 deactivate_super(sb);
207 void mntput_no_expire(struct vfsmount *mnt)
210 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
211 if (likely(!mnt->mnt_pinned)) {
212 spin_unlock(&vfsmount_lock);
216 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
218 spin_unlock(&vfsmount_lock);
219 acct_auto_close_mnt(mnt);
220 security_sb_umount_close(mnt);
225 EXPORT_SYMBOL(mntput_no_expire);
227 void mnt_pin(struct vfsmount *mnt)
229 spin_lock(&vfsmount_lock);
231 spin_unlock(&vfsmount_lock);
234 EXPORT_SYMBOL(mnt_pin);
236 void mnt_unpin(struct vfsmount *mnt)
238 spin_lock(&vfsmount_lock);
239 if (mnt->mnt_pinned) {
240 atomic_inc(&mnt->mnt_count);
243 spin_unlock(&vfsmount_lock);
246 EXPORT_SYMBOL(mnt_unpin);
249 static void *m_start(struct seq_file *m, loff_t *pos)
251 struct namespace *n = m->private;
255 down_read(&namespace_sem);
256 list_for_each(p, &n->list)
258 return list_entry(p, struct vfsmount, mnt_list);
262 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
264 struct namespace *n = m->private;
265 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
267 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
270 static void m_stop(struct seq_file *m, void *v)
272 up_read(&namespace_sem);
275 static inline void mangle(struct seq_file *m, const char *s)
277 seq_escape(m, s, " \t\n\\");
280 static int show_vfsmnt(struct seq_file *m, void *v)
282 struct vfsmount *mnt = v;
284 static struct proc_fs_info {
288 { MS_SYNCHRONOUS, ",sync" },
289 { MS_DIRSYNC, ",dirsync" },
290 { MS_MANDLOCK, ",mand" },
291 { MS_NOATIME, ",noatime" },
292 { MS_NODIRATIME, ",nodiratime" },
295 static struct proc_fs_info mnt_info[] = {
296 { MNT_NOSUID, ",nosuid" },
297 { MNT_NODEV, ",nodev" },
298 { MNT_NOEXEC, ",noexec" },
301 struct proc_fs_info *fs_infop;
303 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
305 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
307 mangle(m, mnt->mnt_sb->s_type->name);
308 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
309 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
310 if (mnt->mnt_sb->s_flags & fs_infop->flag)
311 seq_puts(m, fs_infop->str);
313 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
314 if (mnt->mnt_flags & fs_infop->flag)
315 seq_puts(m, fs_infop->str);
317 if (mnt->mnt_sb->s_op->show_options)
318 err = mnt->mnt_sb->s_op->show_options(m, mnt);
319 seq_puts(m, " 0 0\n");
323 struct seq_operations mounts_op = {
331 * may_umount_tree - check if a mount tree is busy
332 * @mnt: root of mount tree
334 * This is called to check if a tree of mounts has any
335 * open files, pwds, chroots or sub mounts that are
338 int may_umount_tree(struct vfsmount *mnt)
341 int minimum_refs = 0;
344 spin_lock(&vfsmount_lock);
345 for (p = mnt; p; p = next_mnt(p, mnt)) {
346 actual_refs += atomic_read(&p->mnt_count);
349 spin_unlock(&vfsmount_lock);
351 if (actual_refs > minimum_refs)
357 EXPORT_SYMBOL(may_umount_tree);
360 * may_umount - check if a mount point is busy
361 * @mnt: root of mount
363 * This is called to check if a mount point has any
364 * open files, pwds, chroots or sub mounts. If the
365 * mount has sub mounts this will return busy
366 * regardless of whether the sub mounts are busy.
368 * Doesn't take quota and stuff into account. IOW, in some cases it will
369 * give false negatives. The main reason why it's here is that we need
370 * a non-destructive way to look for easily umountable filesystems.
372 int may_umount(struct vfsmount *mnt)
374 if (atomic_read(&mnt->mnt_count) > 2)
379 EXPORT_SYMBOL(may_umount);
381 static void release_mounts(struct list_head *head)
383 struct vfsmount *mnt;
384 while(!list_empty(head)) {
385 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
386 list_del_init(&mnt->mnt_hash);
387 if (mnt->mnt_parent != mnt) {
388 struct dentry *dentry;
390 spin_lock(&vfsmount_lock);
391 dentry = mnt->mnt_mountpoint;
393 mnt->mnt_mountpoint = mnt->mnt_root;
394 mnt->mnt_parent = mnt;
395 spin_unlock(&vfsmount_lock);
403 static void umount_tree(struct vfsmount *mnt, struct list_head *kill)
407 for (p = mnt; p; p = next_mnt(p, mnt)) {
408 list_del(&p->mnt_hash);
409 list_add(&p->mnt_hash, kill);
412 list_for_each_entry(p, kill, mnt_hash) {
413 list_del_init(&p->mnt_expire);
414 list_del_init(&p->mnt_list);
415 __touch_namespace(p->mnt_namespace);
416 p->mnt_namespace = NULL;
417 list_del_init(&p->mnt_child);
418 if (p->mnt_parent != p)
419 mnt->mnt_mountpoint->d_mounted--;
423 static int do_umount(struct vfsmount *mnt, int flags)
425 struct super_block *sb = mnt->mnt_sb;
427 LIST_HEAD(umount_list);
429 retval = security_sb_umount(mnt, flags);
434 * Allow userspace to request a mountpoint be expired rather than
435 * unmounting unconditionally. Unmount only happens if:
436 * (1) the mark is already set (the mark is cleared by mntput())
437 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
439 if (flags & MNT_EXPIRE) {
440 if (mnt == current->fs->rootmnt ||
441 flags & (MNT_FORCE | MNT_DETACH))
444 if (atomic_read(&mnt->mnt_count) != 2)
447 if (!xchg(&mnt->mnt_expiry_mark, 1))
452 * If we may have to abort operations to get out of this
453 * mount, and they will themselves hold resources we must
454 * allow the fs to do things. In the Unix tradition of
455 * 'Gee thats tricky lets do it in userspace' the umount_begin
456 * might fail to complete on the first run through as other tasks
457 * must return, and the like. Thats for the mount program to worry
458 * about for the moment.
462 if ((flags & MNT_FORCE) && sb->s_op->umount_begin)
463 sb->s_op->umount_begin(sb);
467 * No sense to grab the lock for this test, but test itself looks
468 * somewhat bogus. Suggestions for better replacement?
469 * Ho-hum... In principle, we might treat that as umount + switch
470 * to rootfs. GC would eventually take care of the old vfsmount.
471 * Actually it makes sense, especially if rootfs would contain a
472 * /reboot - static binary that would close all descriptors and
473 * call reboot(9). Then init(8) could umount root and exec /reboot.
475 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
477 * Special case for "unmounting" root ...
478 * we just try to remount it readonly.
480 down_write(&sb->s_umount);
481 if (!(sb->s_flags & MS_RDONLY)) {
484 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
487 up_write(&sb->s_umount);
491 down_write(&namespace_sem);
492 spin_lock(&vfsmount_lock);
496 if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
497 if (!list_empty(&mnt->mnt_list))
498 umount_tree(mnt, &umount_list);
501 spin_unlock(&vfsmount_lock);
503 security_sb_umount_busy(mnt);
504 up_write(&namespace_sem);
505 release_mounts(&umount_list);
510 * Now umount can handle mount points as well as block devices.
511 * This is important for filesystems which use unnamed block devices.
513 * We now support a flag for forced unmount like the other 'big iron'
514 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
517 asmlinkage long sys_umount(char __user * name, int flags)
522 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
526 if (nd.dentry != nd.mnt->mnt_root)
528 if (!check_mnt(nd.mnt))
532 if (!capable(CAP_SYS_ADMIN))
535 retval = do_umount(nd.mnt, flags);
537 path_release_on_umount(&nd);
542 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
545 * The 2.0 compatible umount. No flags.
547 asmlinkage long sys_oldumount(char __user * name)
549 return sys_umount(name, 0);
554 static int mount_is_safe(struct nameidata *nd)
556 if (capable(CAP_SYS_ADMIN))
560 if (S_ISLNK(nd->dentry->d_inode->i_mode))
562 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
563 if (current->uid != nd->dentry->d_inode->i_uid)
566 if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
572 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
577 if (d == NULL || d == d->d_parent)
583 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
586 struct vfsmount *res, *p, *q, *r, *s;
589 res = q = clone_mnt(mnt, dentry, flag);
592 q->mnt_mountpoint = mnt->mnt_mountpoint;
595 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
596 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
599 for (s = r; s; s = next_mnt(s, r)) {
600 while (p != s->mnt_parent) {
606 nd.dentry = p->mnt_mountpoint;
607 q = clone_mnt(p, p->mnt_root, flag);
610 spin_lock(&vfsmount_lock);
611 list_add_tail(&q->mnt_list, &res->mnt_list);
613 spin_unlock(&vfsmount_lock);
619 LIST_HEAD(umount_list);
620 spin_lock(&vfsmount_lock);
621 umount_tree(res, &umount_list);
622 spin_unlock(&vfsmount_lock);
623 release_mounts(&umount_list);
628 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
631 if (mnt->mnt_sb->s_flags & MS_NOUSER)
634 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
635 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
639 down(&nd->dentry->d_inode->i_sem);
640 if (IS_DEADDIR(nd->dentry->d_inode))
643 err = security_sb_check_sb(mnt, nd);
648 spin_lock(&vfsmount_lock);
649 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
650 struct list_head head;
653 list_add_tail(&head, &mnt->mnt_list);
654 list_splice(&head, current->namespace->list.prev);
656 touch_namespace(current->namespace);
658 spin_unlock(&vfsmount_lock);
660 up(&nd->dentry->d_inode->i_sem);
662 security_sb_post_addmount(mnt, nd);
667 * recursively change the type of the mountpoint.
669 static int do_change_type(struct nameidata *nd, int flag)
671 struct vfsmount *m, *mnt = nd->mnt;
672 int recurse = flag & MS_REC;
673 int type = flag & ~MS_REC;
675 if (nd->dentry != nd->mnt->mnt_root)
678 down_write(&namespace_sem);
679 spin_lock(&vfsmount_lock);
680 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
681 change_mnt_propagation(m, type);
682 spin_unlock(&vfsmount_lock);
683 up_write(&namespace_sem);
690 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
692 struct nameidata old_nd;
693 struct vfsmount *mnt = NULL;
694 int err = mount_is_safe(nd);
697 if (!old_name || !*old_name)
699 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
703 down_write(&namespace_sem);
705 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
710 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
712 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
717 err = graft_tree(mnt, nd);
719 LIST_HEAD(umount_list);
720 spin_lock(&vfsmount_lock);
721 umount_tree(mnt, &umount_list);
722 spin_unlock(&vfsmount_lock);
723 release_mounts(&umount_list);
727 up_write(&namespace_sem);
728 path_release(&old_nd);
733 * change filesystem flags. dir should be a physical root of filesystem.
734 * If you've mounted a non-root directory somewhere and want to do remount
735 * on it - tough luck.
737 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
741 struct super_block *sb = nd->mnt->mnt_sb;
743 if (!capable(CAP_SYS_ADMIN))
746 if (!check_mnt(nd->mnt))
749 if (nd->dentry != nd->mnt->mnt_root)
752 down_write(&sb->s_umount);
753 err = do_remount_sb(sb, flags, data, 0);
755 nd->mnt->mnt_flags = mnt_flags;
756 up_write(&sb->s_umount);
758 security_sb_post_remount(nd->mnt, flags, data);
762 static int do_move_mount(struct nameidata *nd, char *old_name)
764 struct nameidata old_nd, parent_nd;
767 if (!capable(CAP_SYS_ADMIN))
769 if (!old_name || !*old_name)
771 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
775 down_write(&namespace_sem);
776 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
779 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
783 down(&nd->dentry->d_inode->i_sem);
784 if (IS_DEADDIR(nd->dentry->d_inode))
787 spin_lock(&vfsmount_lock);
788 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
792 if (old_nd.dentry != old_nd.mnt->mnt_root)
795 if (old_nd.mnt == old_nd.mnt->mnt_parent)
798 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
799 S_ISDIR(old_nd.dentry->d_inode->i_mode))
803 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
808 detach_mnt(old_nd.mnt, &parent_nd);
809 attach_mnt(old_nd.mnt, nd);
810 touch_namespace(current->namespace);
812 /* if the mount is moved, it should no longer be expire
814 list_del_init(&old_nd.mnt->mnt_expire);
816 spin_unlock(&vfsmount_lock);
818 up(&nd->dentry->d_inode->i_sem);
820 up_write(&namespace_sem);
822 path_release(&parent_nd);
823 path_release(&old_nd);
828 * create a new mount for userspace and request it to be added into the
831 static int do_new_mount(struct nameidata *nd, char *type, int flags,
832 int mnt_flags, char *name, void *data)
834 struct vfsmount *mnt;
836 if (!type || !memchr(type, 0, PAGE_SIZE))
839 /* we need capabilities... */
840 if (!capable(CAP_SYS_ADMIN))
843 mnt = do_kern_mount(type, flags, name, data);
847 return do_add_mount(mnt, nd, mnt_flags, NULL);
851 * add a mount into a namespace's mount tree
852 * - provide the option of adding the new mount to an expiration list
854 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
855 int mnt_flags, struct list_head *fslist)
859 down_write(&namespace_sem);
860 /* Something was mounted here while we slept */
861 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
864 if (!check_mnt(nd->mnt))
867 /* Refuse the same filesystem on the same mount point */
869 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
870 nd->mnt->mnt_root == nd->dentry)
874 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
877 newmnt->mnt_flags = mnt_flags;
878 if ((err = graft_tree(newmnt, nd)))
882 /* add to the specified expiration list */
883 spin_lock(&vfsmount_lock);
884 list_add_tail(&newmnt->mnt_expire, fslist);
885 spin_unlock(&vfsmount_lock);
887 up_write(&namespace_sem);
891 up_write(&namespace_sem);
896 EXPORT_SYMBOL_GPL(do_add_mount);
898 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
899 struct list_head *umounts)
901 spin_lock(&vfsmount_lock);
904 * Check if mount is still attached, if not, let whoever holds it deal
907 if (mnt->mnt_parent == mnt) {
908 spin_unlock(&vfsmount_lock);
913 * Check that it is still dead: the count should now be 2 - as
914 * contributed by the vfsmount parent and the mntget above
916 if (atomic_read(&mnt->mnt_count) == 2) {
917 /* delete from the namespace */
918 touch_namespace(mnt->mnt_namespace);
919 list_del_init(&mnt->mnt_list);
920 mnt->mnt_namespace = NULL;
921 umount_tree(mnt, umounts);
922 spin_unlock(&vfsmount_lock);
925 * Someone brought it back to life whilst we didn't have any
926 * locks held so return it to the expiration list
928 list_add_tail(&mnt->mnt_expire, mounts);
929 spin_unlock(&vfsmount_lock);
934 * process a list of expirable mountpoints with the intent of discarding any
935 * mountpoints that aren't in use and haven't been touched since last we came
938 void mark_mounts_for_expiry(struct list_head *mounts)
940 struct namespace *namespace;
941 struct vfsmount *mnt, *next;
942 LIST_HEAD(graveyard);
944 if (list_empty(mounts))
947 spin_lock(&vfsmount_lock);
949 /* extract from the expiration list every vfsmount that matches the
950 * following criteria:
951 * - only referenced by its parent vfsmount
952 * - still marked for expiry (marked on the last call here; marks are
953 * cleared by mntput())
955 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
956 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
957 atomic_read(&mnt->mnt_count) != 1)
961 list_move(&mnt->mnt_expire, &graveyard);
965 * go through the vfsmounts we've just consigned to the graveyard to
966 * - check that they're still dead
967 * - delete the vfsmount from the appropriate namespace under lock
968 * - dispose of the corpse
970 while (!list_empty(&graveyard)) {
972 mnt = list_entry(graveyard.next, struct vfsmount, mnt_expire);
973 list_del_init(&mnt->mnt_expire);
975 /* don't do anything if the namespace is dead - all the
976 * vfsmounts from it are going away anyway */
977 namespace = mnt->mnt_namespace;
978 if (!namespace || !namespace->root)
980 get_namespace(namespace);
982 spin_unlock(&vfsmount_lock);
983 down_write(&namespace_sem);
984 expire_mount(mnt, mounts, &umounts);
985 up_write(&namespace_sem);
986 release_mounts(&umounts);
988 put_namespace(namespace);
989 spin_lock(&vfsmount_lock);
992 spin_unlock(&vfsmount_lock);
995 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
998 * Some copy_from_user() implementations do not return the exact number of
999 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1000 * Note that this function differs from copy_from_user() in that it will oops
1001 * on bad values of `to', rather than returning a short copy.
1003 static long exact_copy_from_user(void *to, const void __user * from,
1007 const char __user *f = from;
1010 if (!access_ok(VERIFY_READ, from, n))
1014 if (__get_user(c, f)) {
1025 int copy_mount_options(const void __user * data, unsigned long *where)
1035 if (!(page = __get_free_page(GFP_KERNEL)))
1038 /* We only care that *some* data at the address the user
1039 * gave us is valid. Just in case, we'll zero
1040 * the remainder of the page.
1042 /* copy_from_user cannot cross TASK_SIZE ! */
1043 size = TASK_SIZE - (unsigned long)data;
1044 if (size > PAGE_SIZE)
1047 i = size - exact_copy_from_user((void *)page, data, size);
1053 memset((char *)page + i, 0, PAGE_SIZE - i);
1059 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1060 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1062 * data is a (void *) that can point to any structure up to
1063 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1064 * information (or be NULL).
1066 * Pre-0.97 versions of mount() didn't have a flags word.
1067 * When the flags word was introduced its top half was required
1068 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1069 * Therefore, if this magic number is present, it carries no information
1070 * and must be discarded.
1072 long do_mount(char *dev_name, char *dir_name, char *type_page,
1073 unsigned long flags, void *data_page)
1075 struct nameidata nd;
1080 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1081 flags &= ~MS_MGC_MSK;
1083 /* Basic sanity checks */
1085 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1087 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1091 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1093 /* Separate the per-mountpoint flags */
1094 if (flags & MS_NOSUID)
1095 mnt_flags |= MNT_NOSUID;
1096 if (flags & MS_NODEV)
1097 mnt_flags |= MNT_NODEV;
1098 if (flags & MS_NOEXEC)
1099 mnt_flags |= MNT_NOEXEC;
1100 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE);
1102 /* ... and get the mountpoint */
1103 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1107 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1111 if (flags & MS_REMOUNT)
1112 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1114 else if (flags & MS_BIND)
1115 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1116 else if (flags & MS_PRIVATE)
1117 retval = do_change_type(&nd, flags);
1118 else if (flags & MS_MOVE)
1119 retval = do_move_mount(&nd, dev_name);
1121 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1122 dev_name, data_page);
1128 int copy_namespace(int flags, struct task_struct *tsk)
1130 struct namespace *namespace = tsk->namespace;
1131 struct namespace *new_ns;
1132 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1133 struct fs_struct *fs = tsk->fs;
1134 struct vfsmount *p, *q;
1139 get_namespace(namespace);
1141 if (!(flags & CLONE_NEWNS))
1144 if (!capable(CAP_SYS_ADMIN)) {
1145 put_namespace(namespace);
1149 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1153 atomic_set(&new_ns->count, 1);
1154 INIT_LIST_HEAD(&new_ns->list);
1155 init_waitqueue_head(&new_ns->poll);
1158 down_write(&namespace_sem);
1159 /* First pass: copy the tree topology */
1160 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1162 if (!new_ns->root) {
1163 up_write(&namespace_sem);
1167 spin_lock(&vfsmount_lock);
1168 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1169 spin_unlock(&vfsmount_lock);
1172 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1173 * as belonging to new namespace. We have already acquired a private
1174 * fs_struct, so tsk->fs->lock is not needed.
1176 p = namespace->root;
1179 q->mnt_namespace = new_ns;
1181 if (p == fs->rootmnt) {
1183 fs->rootmnt = mntget(q);
1185 if (p == fs->pwdmnt) {
1187 fs->pwdmnt = mntget(q);
1189 if (p == fs->altrootmnt) {
1191 fs->altrootmnt = mntget(q);
1194 p = next_mnt(p, namespace->root);
1195 q = next_mnt(q, new_ns->root);
1197 up_write(&namespace_sem);
1199 tsk->namespace = new_ns;
1208 put_namespace(namespace);
1212 put_namespace(namespace);
1216 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1217 char __user * type, unsigned long flags,
1221 unsigned long data_page;
1222 unsigned long type_page;
1223 unsigned long dev_page;
1226 retval = copy_mount_options(type, &type_page);
1230 dir_page = getname(dir_name);
1231 retval = PTR_ERR(dir_page);
1232 if (IS_ERR(dir_page))
1235 retval = copy_mount_options(dev_name, &dev_page);
1239 retval = copy_mount_options(data, &data_page);
1244 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1245 flags, (void *)data_page);
1247 free_page(data_page);
1250 free_page(dev_page);
1254 free_page(type_page);
1259 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1260 * It can block. Requires the big lock held.
1262 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1263 struct dentry *dentry)
1265 struct dentry *old_root;
1266 struct vfsmount *old_rootmnt;
1267 write_lock(&fs->lock);
1268 old_root = fs->root;
1269 old_rootmnt = fs->rootmnt;
1270 fs->rootmnt = mntget(mnt);
1271 fs->root = dget(dentry);
1272 write_unlock(&fs->lock);
1275 mntput(old_rootmnt);
1280 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1281 * It can block. Requires the big lock held.
1283 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1284 struct dentry *dentry)
1286 struct dentry *old_pwd;
1287 struct vfsmount *old_pwdmnt;
1289 write_lock(&fs->lock);
1291 old_pwdmnt = fs->pwdmnt;
1292 fs->pwdmnt = mntget(mnt);
1293 fs->pwd = dget(dentry);
1294 write_unlock(&fs->lock);
1302 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1304 struct task_struct *g, *p;
1305 struct fs_struct *fs;
1307 read_lock(&tasklist_lock);
1308 do_each_thread(g, p) {
1312 atomic_inc(&fs->count);
1314 if (fs->root == old_nd->dentry
1315 && fs->rootmnt == old_nd->mnt)
1316 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1317 if (fs->pwd == old_nd->dentry
1318 && fs->pwdmnt == old_nd->mnt)
1319 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1323 } while_each_thread(g, p);
1324 read_unlock(&tasklist_lock);
1328 * pivot_root Semantics:
1329 * Moves the root file system of the current process to the directory put_old,
1330 * makes new_root as the new root file system of the current process, and sets
1331 * root/cwd of all processes which had them on the current root to new_root.
1334 * The new_root and put_old must be directories, and must not be on the
1335 * same file system as the current process root. The put_old must be
1336 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1337 * pointed to by put_old must yield the same directory as new_root. No other
1338 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1341 * - we don't move root/cwd if they are not at the root (reason: if something
1342 * cared enough to change them, it's probably wrong to force them elsewhere)
1343 * - it's okay to pick a root that isn't the root of a file system, e.g.
1344 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1345 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1348 asmlinkage long sys_pivot_root(const char __user * new_root,
1349 const char __user * put_old)
1351 struct vfsmount *tmp;
1352 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1355 if (!capable(CAP_SYS_ADMIN))
1360 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1365 if (!check_mnt(new_nd.mnt))
1368 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1372 error = security_sb_pivotroot(&old_nd, &new_nd);
1374 path_release(&old_nd);
1378 read_lock(¤t->fs->lock);
1379 user_nd.mnt = mntget(current->fs->rootmnt);
1380 user_nd.dentry = dget(current->fs->root);
1381 read_unlock(¤t->fs->lock);
1382 down_write(&namespace_sem);
1383 down(&old_nd.dentry->d_inode->i_sem);
1385 if (!check_mnt(user_nd.mnt))
1388 if (IS_DEADDIR(new_nd.dentry->d_inode))
1390 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1392 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1395 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1396 goto out2; /* loop, on the same file system */
1398 if (user_nd.mnt->mnt_root != user_nd.dentry)
1399 goto out2; /* not a mountpoint */
1400 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1401 goto out2; /* not attached */
1402 if (new_nd.mnt->mnt_root != new_nd.dentry)
1403 goto out2; /* not a mountpoint */
1404 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1405 goto out2; /* not attached */
1406 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1407 spin_lock(&vfsmount_lock);
1408 if (tmp != new_nd.mnt) {
1410 if (tmp->mnt_parent == tmp)
1411 goto out3; /* already mounted on put_old */
1412 if (tmp->mnt_parent == new_nd.mnt)
1414 tmp = tmp->mnt_parent;
1416 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1418 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1420 detach_mnt(new_nd.mnt, &parent_nd);
1421 detach_mnt(user_nd.mnt, &root_parent);
1422 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1423 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1424 touch_namespace(current->namespace);
1425 spin_unlock(&vfsmount_lock);
1426 chroot_fs_refs(&user_nd, &new_nd);
1427 security_sb_post_pivotroot(&user_nd, &new_nd);
1429 path_release(&root_parent);
1430 path_release(&parent_nd);
1432 up(&old_nd.dentry->d_inode->i_sem);
1433 up_write(&namespace_sem);
1434 path_release(&user_nd);
1435 path_release(&old_nd);
1437 path_release(&new_nd);
1442 spin_unlock(&vfsmount_lock);
1446 static void __init init_mount_tree(void)
1448 struct vfsmount *mnt;
1449 struct namespace *namespace;
1450 struct task_struct *g, *p;
1452 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1454 panic("Can't create rootfs");
1455 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1457 panic("Can't allocate initial namespace");
1458 atomic_set(&namespace->count, 1);
1459 INIT_LIST_HEAD(&namespace->list);
1460 init_waitqueue_head(&namespace->poll);
1461 namespace->event = 0;
1462 list_add(&mnt->mnt_list, &namespace->list);
1463 namespace->root = mnt;
1464 mnt->mnt_namespace = namespace;
1466 init_task.namespace = namespace;
1467 read_lock(&tasklist_lock);
1468 do_each_thread(g, p) {
1469 get_namespace(namespace);
1470 p->namespace = namespace;
1471 } while_each_thread(g, p);
1472 read_unlock(&tasklist_lock);
1474 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1475 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1478 void __init mnt_init(unsigned long mempages)
1480 struct list_head *d;
1481 unsigned int nr_hash;
1484 init_rwsem(&namespace_sem);
1486 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1487 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1489 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1491 if (!mount_hashtable)
1492 panic("Failed to allocate mount hash table\n");
1495 * Find the power-of-two list-heads that can fit into the allocation..
1496 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1499 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1503 } while ((nr_hash >> hash_bits) != 0);
1507 * Re-calculate the actual number of entries and the mask
1508 * from the number of bits we can fit.
1510 nr_hash = 1UL << hash_bits;
1511 hash_mask = nr_hash - 1;
1513 printk("Mount-cache hash table entries: %d\n", nr_hash);
1515 /* And initialize the newly allocated array */
1516 d = mount_hashtable;
1528 void __put_namespace(struct namespace *namespace)
1530 struct vfsmount *root = namespace->root;
1531 LIST_HEAD(umount_list);
1532 namespace->root = NULL;
1533 spin_unlock(&vfsmount_lock);
1534 down_write(&namespace_sem);
1535 spin_lock(&vfsmount_lock);
1536 umount_tree(root, &umount_list);
1537 spin_unlock(&vfsmount_lock);
1538 up_write(&namespace_sem);
1539 release_mounts(&umount_list);