wan/farsync: copy_from_user() to iomem is wrong
[linux-2.6/sactl.git] / fs / namespace.c
blob7953c96a20715a981032b4d7af18e428f39b3ca6
1 /*
2 * linux/fs/namespace.c
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.
8 * Heavily rewritten.
9 */
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <linux/log2.h>
29 #include <asm/uaccess.h>
30 #include <asm/unistd.h>
31 #include "pnode.h"
32 #include "internal.h"
34 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
35 #define HASH_SIZE (1UL << HASH_SHIFT)
37 /* spinlock for vfsmount related operations, inplace of dcache_lock */
38 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
40 static int event;
42 static struct list_head *mount_hashtable __read_mostly;
43 static struct kmem_cache *mnt_cache __read_mostly;
44 static struct rw_semaphore namespace_sem;
46 /* /sys/fs */
47 struct kobject *fs_kobj;
48 EXPORT_SYMBOL_GPL(fs_kobj);
50 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
52 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
53 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
54 tmp = tmp + (tmp >> HASH_SHIFT);
55 return tmp & (HASH_SIZE - 1);
58 struct vfsmount *alloc_vfsmnt(const char *name)
60 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
61 if (mnt) {
62 atomic_set(&mnt->mnt_count, 1);
63 INIT_LIST_HEAD(&mnt->mnt_hash);
64 INIT_LIST_HEAD(&mnt->mnt_child);
65 INIT_LIST_HEAD(&mnt->mnt_mounts);
66 INIT_LIST_HEAD(&mnt->mnt_list);
67 INIT_LIST_HEAD(&mnt->mnt_expire);
68 INIT_LIST_HEAD(&mnt->mnt_share);
69 INIT_LIST_HEAD(&mnt->mnt_slave_list);
70 INIT_LIST_HEAD(&mnt->mnt_slave);
71 if (name) {
72 int size = strlen(name) + 1;
73 char *newname = kmalloc(size, GFP_KERNEL);
74 if (newname) {
75 memcpy(newname, name, size);
76 mnt->mnt_devname = newname;
80 return mnt;
83 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
85 mnt->mnt_sb = sb;
86 mnt->mnt_root = dget(sb->s_root);
87 return 0;
90 EXPORT_SYMBOL(simple_set_mnt);
92 void free_vfsmnt(struct vfsmount *mnt)
94 kfree(mnt->mnt_devname);
95 kmem_cache_free(mnt_cache, mnt);
99 * find the first or last mount at @dentry on vfsmount @mnt depending on
100 * @dir. If @dir is set return the first mount else return the last mount.
102 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
103 int dir)
105 struct list_head *head = mount_hashtable + hash(mnt, dentry);
106 struct list_head *tmp = head;
107 struct vfsmount *p, *found = NULL;
109 for (;;) {
110 tmp = dir ? tmp->next : tmp->prev;
111 p = NULL;
112 if (tmp == head)
113 break;
114 p = list_entry(tmp, struct vfsmount, mnt_hash);
115 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
116 found = p;
117 break;
120 return found;
124 * lookup_mnt increments the ref count before returning
125 * the vfsmount struct.
127 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
129 struct vfsmount *child_mnt;
130 spin_lock(&vfsmount_lock);
131 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
132 mntget(child_mnt);
133 spin_unlock(&vfsmount_lock);
134 return child_mnt;
137 static inline int check_mnt(struct vfsmount *mnt)
139 return mnt->mnt_ns == current->nsproxy->mnt_ns;
142 static void touch_mnt_namespace(struct mnt_namespace *ns)
144 if (ns) {
145 ns->event = ++event;
146 wake_up_interruptible(&ns->poll);
150 static void __touch_mnt_namespace(struct mnt_namespace *ns)
152 if (ns && ns->event != event) {
153 ns->event = event;
154 wake_up_interruptible(&ns->poll);
158 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
160 old_nd->path.dentry = mnt->mnt_mountpoint;
161 old_nd->path.mnt = mnt->mnt_parent;
162 mnt->mnt_parent = mnt;
163 mnt->mnt_mountpoint = mnt->mnt_root;
164 list_del_init(&mnt->mnt_child);
165 list_del_init(&mnt->mnt_hash);
166 old_nd->path.dentry->d_mounted--;
169 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
170 struct vfsmount *child_mnt)
172 child_mnt->mnt_parent = mntget(mnt);
173 child_mnt->mnt_mountpoint = dget(dentry);
174 dentry->d_mounted++;
177 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
179 mnt_set_mountpoint(nd->path.mnt, nd->path.dentry, mnt);
180 list_add_tail(&mnt->mnt_hash, mount_hashtable +
181 hash(nd->path.mnt, nd->path.dentry));
182 list_add_tail(&mnt->mnt_child, &nd->path.mnt->mnt_mounts);
186 * the caller must hold vfsmount_lock
188 static void commit_tree(struct vfsmount *mnt)
190 struct vfsmount *parent = mnt->mnt_parent;
191 struct vfsmount *m;
192 LIST_HEAD(head);
193 struct mnt_namespace *n = parent->mnt_ns;
195 BUG_ON(parent == mnt);
197 list_add_tail(&head, &mnt->mnt_list);
198 list_for_each_entry(m, &head, mnt_list)
199 m->mnt_ns = n;
200 list_splice(&head, n->list.prev);
202 list_add_tail(&mnt->mnt_hash, mount_hashtable +
203 hash(parent, mnt->mnt_mountpoint));
204 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
205 touch_mnt_namespace(n);
208 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
210 struct list_head *next = p->mnt_mounts.next;
211 if (next == &p->mnt_mounts) {
212 while (1) {
213 if (p == root)
214 return NULL;
215 next = p->mnt_child.next;
216 if (next != &p->mnt_parent->mnt_mounts)
217 break;
218 p = p->mnt_parent;
221 return list_entry(next, struct vfsmount, mnt_child);
224 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
226 struct list_head *prev = p->mnt_mounts.prev;
227 while (prev != &p->mnt_mounts) {
228 p = list_entry(prev, struct vfsmount, mnt_child);
229 prev = p->mnt_mounts.prev;
231 return p;
234 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
235 int flag)
237 struct super_block *sb = old->mnt_sb;
238 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
240 if (mnt) {
241 mnt->mnt_flags = old->mnt_flags;
242 atomic_inc(&sb->s_active);
243 mnt->mnt_sb = sb;
244 mnt->mnt_root = dget(root);
245 mnt->mnt_mountpoint = mnt->mnt_root;
246 mnt->mnt_parent = mnt;
248 if (flag & CL_SLAVE) {
249 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
250 mnt->mnt_master = old;
251 CLEAR_MNT_SHARED(mnt);
252 } else if (!(flag & CL_PRIVATE)) {
253 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
254 list_add(&mnt->mnt_share, &old->mnt_share);
255 if (IS_MNT_SLAVE(old))
256 list_add(&mnt->mnt_slave, &old->mnt_slave);
257 mnt->mnt_master = old->mnt_master;
259 if (flag & CL_MAKE_SHARED)
260 set_mnt_shared(mnt);
262 /* stick the duplicate mount on the same expiry list
263 * as the original if that was on one */
264 if (flag & CL_EXPIRE) {
265 spin_lock(&vfsmount_lock);
266 if (!list_empty(&old->mnt_expire))
267 list_add(&mnt->mnt_expire, &old->mnt_expire);
268 spin_unlock(&vfsmount_lock);
271 return mnt;
274 static inline void __mntput(struct vfsmount *mnt)
276 struct super_block *sb = mnt->mnt_sb;
277 dput(mnt->mnt_root);
278 free_vfsmnt(mnt);
279 deactivate_super(sb);
282 void mntput_no_expire(struct vfsmount *mnt)
284 repeat:
285 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
286 if (likely(!mnt->mnt_pinned)) {
287 spin_unlock(&vfsmount_lock);
288 __mntput(mnt);
289 return;
291 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
292 mnt->mnt_pinned = 0;
293 spin_unlock(&vfsmount_lock);
294 acct_auto_close_mnt(mnt);
295 security_sb_umount_close(mnt);
296 goto repeat;
300 EXPORT_SYMBOL(mntput_no_expire);
302 void mnt_pin(struct vfsmount *mnt)
304 spin_lock(&vfsmount_lock);
305 mnt->mnt_pinned++;
306 spin_unlock(&vfsmount_lock);
309 EXPORT_SYMBOL(mnt_pin);
311 void mnt_unpin(struct vfsmount *mnt)
313 spin_lock(&vfsmount_lock);
314 if (mnt->mnt_pinned) {
315 atomic_inc(&mnt->mnt_count);
316 mnt->mnt_pinned--;
318 spin_unlock(&vfsmount_lock);
321 EXPORT_SYMBOL(mnt_unpin);
323 static inline void mangle(struct seq_file *m, const char *s)
325 seq_escape(m, s, " \t\n\\");
329 * Simple .show_options callback for filesystems which don't want to
330 * implement more complex mount option showing.
332 * See also save_mount_options().
334 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
336 const char *options = mnt->mnt_sb->s_options;
338 if (options != NULL && options[0]) {
339 seq_putc(m, ',');
340 mangle(m, options);
343 return 0;
345 EXPORT_SYMBOL(generic_show_options);
348 * If filesystem uses generic_show_options(), this function should be
349 * called from the fill_super() callback.
351 * The .remount_fs callback usually needs to be handled in a special
352 * way, to make sure, that previous options are not overwritten if the
353 * remount fails.
355 * Also note, that if the filesystem's .remount_fs function doesn't
356 * reset all options to their default value, but changes only newly
357 * given options, then the displayed options will not reflect reality
358 * any more.
360 void save_mount_options(struct super_block *sb, char *options)
362 kfree(sb->s_options);
363 sb->s_options = kstrdup(options, GFP_KERNEL);
365 EXPORT_SYMBOL(save_mount_options);
367 /* iterator */
368 static void *m_start(struct seq_file *m, loff_t *pos)
370 struct mnt_namespace *n = m->private;
372 down_read(&namespace_sem);
373 return seq_list_start(&n->list, *pos);
376 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
378 struct mnt_namespace *n = m->private;
380 return seq_list_next(v, &n->list, pos);
383 static void m_stop(struct seq_file *m, void *v)
385 up_read(&namespace_sem);
388 static int show_vfsmnt(struct seq_file *m, void *v)
390 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
391 int err = 0;
392 static struct proc_fs_info {
393 int flag;
394 char *str;
395 } fs_info[] = {
396 { MS_SYNCHRONOUS, ",sync" },
397 { MS_DIRSYNC, ",dirsync" },
398 { MS_MANDLOCK, ",mand" },
399 { 0, NULL }
401 static struct proc_fs_info mnt_info[] = {
402 { MNT_NOSUID, ",nosuid" },
403 { MNT_NODEV, ",nodev" },
404 { MNT_NOEXEC, ",noexec" },
405 { MNT_NOATIME, ",noatime" },
406 { MNT_NODIRATIME, ",nodiratime" },
407 { MNT_RELATIME, ",relatime" },
408 { 0, NULL }
410 struct proc_fs_info *fs_infop;
411 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
413 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
414 seq_putc(m, ' ');
415 seq_path(m, &mnt_path, " \t\n\\");
416 seq_putc(m, ' ');
417 mangle(m, mnt->mnt_sb->s_type->name);
418 if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) {
419 seq_putc(m, '.');
420 mangle(m, mnt->mnt_sb->s_subtype);
422 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
423 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
424 if (mnt->mnt_sb->s_flags & fs_infop->flag)
425 seq_puts(m, fs_infop->str);
427 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
428 if (mnt->mnt_flags & fs_infop->flag)
429 seq_puts(m, fs_infop->str);
431 if (mnt->mnt_sb->s_op->show_options)
432 err = mnt->mnt_sb->s_op->show_options(m, mnt);
433 seq_puts(m, " 0 0\n");
434 return err;
437 struct seq_operations mounts_op = {
438 .start = m_start,
439 .next = m_next,
440 .stop = m_stop,
441 .show = show_vfsmnt
444 static int show_vfsstat(struct seq_file *m, void *v)
446 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
447 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
448 int err = 0;
450 /* device */
451 if (mnt->mnt_devname) {
452 seq_puts(m, "device ");
453 mangle(m, mnt->mnt_devname);
454 } else
455 seq_puts(m, "no device");
457 /* mount point */
458 seq_puts(m, " mounted on ");
459 seq_path(m, &mnt_path, " \t\n\\");
460 seq_putc(m, ' ');
462 /* file system type */
463 seq_puts(m, "with fstype ");
464 mangle(m, mnt->mnt_sb->s_type->name);
466 /* optional statistics */
467 if (mnt->mnt_sb->s_op->show_stats) {
468 seq_putc(m, ' ');
469 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
472 seq_putc(m, '\n');
473 return err;
476 struct seq_operations mountstats_op = {
477 .start = m_start,
478 .next = m_next,
479 .stop = m_stop,
480 .show = show_vfsstat,
484 * may_umount_tree - check if a mount tree is busy
485 * @mnt: root of mount tree
487 * This is called to check if a tree of mounts has any
488 * open files, pwds, chroots or sub mounts that are
489 * busy.
491 int may_umount_tree(struct vfsmount *mnt)
493 int actual_refs = 0;
494 int minimum_refs = 0;
495 struct vfsmount *p;
497 spin_lock(&vfsmount_lock);
498 for (p = mnt; p; p = next_mnt(p, mnt)) {
499 actual_refs += atomic_read(&p->mnt_count);
500 minimum_refs += 2;
502 spin_unlock(&vfsmount_lock);
504 if (actual_refs > minimum_refs)
505 return 0;
507 return 1;
510 EXPORT_SYMBOL(may_umount_tree);
513 * may_umount - check if a mount point is busy
514 * @mnt: root of mount
516 * This is called to check if a mount point has any
517 * open files, pwds, chroots or sub mounts. If the
518 * mount has sub mounts this will return busy
519 * regardless of whether the sub mounts are busy.
521 * Doesn't take quota and stuff into account. IOW, in some cases it will
522 * give false negatives. The main reason why it's here is that we need
523 * a non-destructive way to look for easily umountable filesystems.
525 int may_umount(struct vfsmount *mnt)
527 int ret = 1;
528 spin_lock(&vfsmount_lock);
529 if (propagate_mount_busy(mnt, 2))
530 ret = 0;
531 spin_unlock(&vfsmount_lock);
532 return ret;
535 EXPORT_SYMBOL(may_umount);
537 void release_mounts(struct list_head *head)
539 struct vfsmount *mnt;
540 while (!list_empty(head)) {
541 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
542 list_del_init(&mnt->mnt_hash);
543 if (mnt->mnt_parent != mnt) {
544 struct dentry *dentry;
545 struct vfsmount *m;
546 spin_lock(&vfsmount_lock);
547 dentry = mnt->mnt_mountpoint;
548 m = mnt->mnt_parent;
549 mnt->mnt_mountpoint = mnt->mnt_root;
550 mnt->mnt_parent = mnt;
551 spin_unlock(&vfsmount_lock);
552 dput(dentry);
553 mntput(m);
555 mntput(mnt);
559 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
561 struct vfsmount *p;
563 for (p = mnt; p; p = next_mnt(p, mnt))
564 list_move(&p->mnt_hash, kill);
566 if (propagate)
567 propagate_umount(kill);
569 list_for_each_entry(p, kill, mnt_hash) {
570 list_del_init(&p->mnt_expire);
571 list_del_init(&p->mnt_list);
572 __touch_mnt_namespace(p->mnt_ns);
573 p->mnt_ns = NULL;
574 list_del_init(&p->mnt_child);
575 if (p->mnt_parent != p)
576 p->mnt_mountpoint->d_mounted--;
577 change_mnt_propagation(p, MS_PRIVATE);
581 static int do_umount(struct vfsmount *mnt, int flags)
583 struct super_block *sb = mnt->mnt_sb;
584 int retval;
585 LIST_HEAD(umount_list);
587 retval = security_sb_umount(mnt, flags);
588 if (retval)
589 return retval;
592 * Allow userspace to request a mountpoint be expired rather than
593 * unmounting unconditionally. Unmount only happens if:
594 * (1) the mark is already set (the mark is cleared by mntput())
595 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
597 if (flags & MNT_EXPIRE) {
598 if (mnt == current->fs->root.mnt ||
599 flags & (MNT_FORCE | MNT_DETACH))
600 return -EINVAL;
602 if (atomic_read(&mnt->mnt_count) != 2)
603 return -EBUSY;
605 if (!xchg(&mnt->mnt_expiry_mark, 1))
606 return -EAGAIN;
610 * If we may have to abort operations to get out of this
611 * mount, and they will themselves hold resources we must
612 * allow the fs to do things. In the Unix tradition of
613 * 'Gee thats tricky lets do it in userspace' the umount_begin
614 * might fail to complete on the first run through as other tasks
615 * must return, and the like. Thats for the mount program to worry
616 * about for the moment.
619 lock_kernel();
620 if (sb->s_op->umount_begin)
621 sb->s_op->umount_begin(mnt, flags);
622 unlock_kernel();
625 * No sense to grab the lock for this test, but test itself looks
626 * somewhat bogus. Suggestions for better replacement?
627 * Ho-hum... In principle, we might treat that as umount + switch
628 * to rootfs. GC would eventually take care of the old vfsmount.
629 * Actually it makes sense, especially if rootfs would contain a
630 * /reboot - static binary that would close all descriptors and
631 * call reboot(9). Then init(8) could umount root and exec /reboot.
633 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
635 * Special case for "unmounting" root ...
636 * we just try to remount it readonly.
638 down_write(&sb->s_umount);
639 if (!(sb->s_flags & MS_RDONLY)) {
640 lock_kernel();
641 DQUOT_OFF(sb);
642 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
643 unlock_kernel();
645 up_write(&sb->s_umount);
646 return retval;
649 down_write(&namespace_sem);
650 spin_lock(&vfsmount_lock);
651 event++;
653 retval = -EBUSY;
654 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
655 if (!list_empty(&mnt->mnt_list))
656 umount_tree(mnt, 1, &umount_list);
657 retval = 0;
659 spin_unlock(&vfsmount_lock);
660 if (retval)
661 security_sb_umount_busy(mnt);
662 up_write(&namespace_sem);
663 release_mounts(&umount_list);
664 return retval;
668 * Now umount can handle mount points as well as block devices.
669 * This is important for filesystems which use unnamed block devices.
671 * We now support a flag for forced unmount like the other 'big iron'
672 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
675 asmlinkage long sys_umount(char __user * name, int flags)
677 struct nameidata nd;
678 int retval;
680 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
681 if (retval)
682 goto out;
683 retval = -EINVAL;
684 if (nd.path.dentry != nd.path.mnt->mnt_root)
685 goto dput_and_out;
686 if (!check_mnt(nd.path.mnt))
687 goto dput_and_out;
689 retval = -EPERM;
690 if (!capable(CAP_SYS_ADMIN))
691 goto dput_and_out;
693 retval = do_umount(nd.path.mnt, flags);
694 dput_and_out:
695 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
696 dput(nd.path.dentry);
697 mntput_no_expire(nd.path.mnt);
698 out:
699 return retval;
702 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
705 * The 2.0 compatible umount. No flags.
707 asmlinkage long sys_oldumount(char __user * name)
709 return sys_umount(name, 0);
712 #endif
714 static int mount_is_safe(struct nameidata *nd)
716 if (capable(CAP_SYS_ADMIN))
717 return 0;
718 return -EPERM;
719 #ifdef notyet
720 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
721 return -EPERM;
722 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
723 if (current->uid != nd->path.dentry->d_inode->i_uid)
724 return -EPERM;
726 if (vfs_permission(nd, MAY_WRITE))
727 return -EPERM;
728 return 0;
729 #endif
732 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
734 while (1) {
735 if (d == dentry)
736 return 1;
737 if (d == NULL || d == d->d_parent)
738 return 0;
739 d = d->d_parent;
743 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
744 int flag)
746 struct vfsmount *res, *p, *q, *r, *s;
747 struct nameidata nd;
749 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
750 return NULL;
752 res = q = clone_mnt(mnt, dentry, flag);
753 if (!q)
754 goto Enomem;
755 q->mnt_mountpoint = mnt->mnt_mountpoint;
757 p = mnt;
758 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
759 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
760 continue;
762 for (s = r; s; s = next_mnt(s, r)) {
763 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
764 s = skip_mnt_tree(s);
765 continue;
767 while (p != s->mnt_parent) {
768 p = p->mnt_parent;
769 q = q->mnt_parent;
771 p = s;
772 nd.path.mnt = q;
773 nd.path.dentry = p->mnt_mountpoint;
774 q = clone_mnt(p, p->mnt_root, flag);
775 if (!q)
776 goto Enomem;
777 spin_lock(&vfsmount_lock);
778 list_add_tail(&q->mnt_list, &res->mnt_list);
779 attach_mnt(q, &nd);
780 spin_unlock(&vfsmount_lock);
783 return res;
784 Enomem:
785 if (res) {
786 LIST_HEAD(umount_list);
787 spin_lock(&vfsmount_lock);
788 umount_tree(res, 0, &umount_list);
789 spin_unlock(&vfsmount_lock);
790 release_mounts(&umount_list);
792 return NULL;
795 struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
797 struct vfsmount *tree;
798 down_read(&namespace_sem);
799 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
800 up_read(&namespace_sem);
801 return tree;
804 void drop_collected_mounts(struct vfsmount *mnt)
806 LIST_HEAD(umount_list);
807 down_read(&namespace_sem);
808 spin_lock(&vfsmount_lock);
809 umount_tree(mnt, 0, &umount_list);
810 spin_unlock(&vfsmount_lock);
811 up_read(&namespace_sem);
812 release_mounts(&umount_list);
816 * @source_mnt : mount tree to be attached
817 * @nd : place the mount tree @source_mnt is attached
818 * @parent_nd : if non-null, detach the source_mnt from its parent and
819 * store the parent mount and mountpoint dentry.
820 * (done when source_mnt is moved)
822 * NOTE: in the table below explains the semantics when a source mount
823 * of a given type is attached to a destination mount of a given type.
824 * ---------------------------------------------------------------------------
825 * | BIND MOUNT OPERATION |
826 * |**************************************************************************
827 * | source-->| shared | private | slave | unbindable |
828 * | dest | | | | |
829 * | | | | | | |
830 * | v | | | | |
831 * |**************************************************************************
832 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
833 * | | | | | |
834 * |non-shared| shared (+) | private | slave (*) | invalid |
835 * ***************************************************************************
836 * A bind operation clones the source mount and mounts the clone on the
837 * destination mount.
839 * (++) the cloned mount is propagated to all the mounts in the propagation
840 * tree of the destination mount and the cloned mount is added to
841 * the peer group of the source mount.
842 * (+) the cloned mount is created under the destination mount and is marked
843 * as shared. The cloned mount is added to the peer group of the source
844 * mount.
845 * (+++) the mount is propagated to all the mounts in the propagation tree
846 * of the destination mount and the cloned mount is made slave
847 * of the same master as that of the source mount. The cloned mount
848 * is marked as 'shared and slave'.
849 * (*) the cloned mount is made a slave of the same master as that of the
850 * source mount.
852 * ---------------------------------------------------------------------------
853 * | MOVE MOUNT OPERATION |
854 * |**************************************************************************
855 * | source-->| shared | private | slave | unbindable |
856 * | dest | | | | |
857 * | | | | | | |
858 * | v | | | | |
859 * |**************************************************************************
860 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
861 * | | | | | |
862 * |non-shared| shared (+*) | private | slave (*) | unbindable |
863 * ***************************************************************************
865 * (+) the mount is moved to the destination. And is then propagated to
866 * all the mounts in the propagation tree of the destination mount.
867 * (+*) the mount is moved to the destination.
868 * (+++) the mount is moved to the destination and is then propagated to
869 * all the mounts belonging to the destination mount's propagation tree.
870 * the mount is marked as 'shared and slave'.
871 * (*) the mount continues to be a slave at the new location.
873 * if the source mount is a tree, the operations explained above is
874 * applied to each mount in the tree.
875 * Must be called without spinlocks held, since this function can sleep
876 * in allocations.
878 static int attach_recursive_mnt(struct vfsmount *source_mnt,
879 struct nameidata *nd, struct nameidata *parent_nd)
881 LIST_HEAD(tree_list);
882 struct vfsmount *dest_mnt = nd->path.mnt;
883 struct dentry *dest_dentry = nd->path.dentry;
884 struct vfsmount *child, *p;
886 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
887 return -EINVAL;
889 if (IS_MNT_SHARED(dest_mnt)) {
890 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
891 set_mnt_shared(p);
894 spin_lock(&vfsmount_lock);
895 if (parent_nd) {
896 detach_mnt(source_mnt, parent_nd);
897 attach_mnt(source_mnt, nd);
898 touch_mnt_namespace(current->nsproxy->mnt_ns);
899 } else {
900 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
901 commit_tree(source_mnt);
904 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
905 list_del_init(&child->mnt_hash);
906 commit_tree(child);
908 spin_unlock(&vfsmount_lock);
909 return 0;
912 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
914 int err;
915 if (mnt->mnt_sb->s_flags & MS_NOUSER)
916 return -EINVAL;
918 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
919 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
920 return -ENOTDIR;
922 err = -ENOENT;
923 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
924 if (IS_DEADDIR(nd->path.dentry->d_inode))
925 goto out_unlock;
927 err = security_sb_check_sb(mnt, nd);
928 if (err)
929 goto out_unlock;
931 err = -ENOENT;
932 if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry))
933 err = attach_recursive_mnt(mnt, nd, NULL);
934 out_unlock:
935 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
936 if (!err)
937 security_sb_post_addmount(mnt, nd);
938 return err;
942 * recursively change the type of the mountpoint.
943 * noinline this do_mount helper to save do_mount stack space.
945 static noinline int do_change_type(struct nameidata *nd, int flag)
947 struct vfsmount *m, *mnt = nd->path.mnt;
948 int recurse = flag & MS_REC;
949 int type = flag & ~MS_REC;
951 if (!capable(CAP_SYS_ADMIN))
952 return -EPERM;
954 if (nd->path.dentry != nd->path.mnt->mnt_root)
955 return -EINVAL;
957 down_write(&namespace_sem);
958 spin_lock(&vfsmount_lock);
959 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
960 change_mnt_propagation(m, type);
961 spin_unlock(&vfsmount_lock);
962 up_write(&namespace_sem);
963 return 0;
967 * do loopback mount.
968 * noinline this do_mount helper to save do_mount stack space.
970 static noinline int do_loopback(struct nameidata *nd, char *old_name,
971 int recurse)
973 struct nameidata old_nd;
974 struct vfsmount *mnt = NULL;
975 int err = mount_is_safe(nd);
976 if (err)
977 return err;
978 if (!old_name || !*old_name)
979 return -EINVAL;
980 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
981 if (err)
982 return err;
984 down_write(&namespace_sem);
985 err = -EINVAL;
986 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
987 goto out;
989 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
990 goto out;
992 err = -ENOMEM;
993 if (recurse)
994 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
995 else
996 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
998 if (!mnt)
999 goto out;
1001 err = graft_tree(mnt, nd);
1002 if (err) {
1003 LIST_HEAD(umount_list);
1004 spin_lock(&vfsmount_lock);
1005 umount_tree(mnt, 0, &umount_list);
1006 spin_unlock(&vfsmount_lock);
1007 release_mounts(&umount_list);
1010 out:
1011 up_write(&namespace_sem);
1012 path_put(&old_nd.path);
1013 return err;
1017 * change filesystem flags. dir should be a physical root of filesystem.
1018 * If you've mounted a non-root directory somewhere and want to do remount
1019 * on it - tough luck.
1020 * noinline this do_mount helper to save do_mount stack space.
1022 static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1023 void *data)
1025 int err;
1026 struct super_block *sb = nd->path.mnt->mnt_sb;
1028 if (!capable(CAP_SYS_ADMIN))
1029 return -EPERM;
1031 if (!check_mnt(nd->path.mnt))
1032 return -EINVAL;
1034 if (nd->path.dentry != nd->path.mnt->mnt_root)
1035 return -EINVAL;
1037 down_write(&sb->s_umount);
1038 err = do_remount_sb(sb, flags, data, 0);
1039 if (!err)
1040 nd->path.mnt->mnt_flags = mnt_flags;
1041 up_write(&sb->s_umount);
1042 if (!err)
1043 security_sb_post_remount(nd->path.mnt, flags, data);
1044 return err;
1047 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1049 struct vfsmount *p;
1050 for (p = mnt; p; p = next_mnt(p, mnt)) {
1051 if (IS_MNT_UNBINDABLE(p))
1052 return 1;
1054 return 0;
1058 * noinline this do_mount helper to save do_mount stack space.
1060 static noinline int do_move_mount(struct nameidata *nd, char *old_name)
1062 struct nameidata old_nd, parent_nd;
1063 struct vfsmount *p;
1064 int err = 0;
1065 if (!capable(CAP_SYS_ADMIN))
1066 return -EPERM;
1067 if (!old_name || !*old_name)
1068 return -EINVAL;
1069 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1070 if (err)
1071 return err;
1073 down_write(&namespace_sem);
1074 while (d_mountpoint(nd->path.dentry) &&
1075 follow_down(&nd->path.mnt, &nd->path.dentry))
1077 err = -EINVAL;
1078 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
1079 goto out;
1081 err = -ENOENT;
1082 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1083 if (IS_DEADDIR(nd->path.dentry->d_inode))
1084 goto out1;
1086 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
1087 goto out1;
1089 err = -EINVAL;
1090 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
1091 goto out1;
1093 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
1094 goto out1;
1096 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1097 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
1098 goto out1;
1100 * Don't move a mount residing in a shared parent.
1102 if (old_nd.path.mnt->mnt_parent &&
1103 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
1104 goto out1;
1106 * Don't move a mount tree containing unbindable mounts to a destination
1107 * mount which is shared.
1109 if (IS_MNT_SHARED(nd->path.mnt) &&
1110 tree_contains_unbindable(old_nd.path.mnt))
1111 goto out1;
1112 err = -ELOOP;
1113 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1114 if (p == old_nd.path.mnt)
1115 goto out1;
1117 err = attach_recursive_mnt(old_nd.path.mnt, nd, &parent_nd);
1118 if (err)
1119 goto out1;
1121 spin_lock(&vfsmount_lock);
1122 /* if the mount is moved, it should no longer be expire
1123 * automatically */
1124 list_del_init(&old_nd.path.mnt->mnt_expire);
1125 spin_unlock(&vfsmount_lock);
1126 out1:
1127 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
1128 out:
1129 up_write(&namespace_sem);
1130 if (!err)
1131 path_put(&parent_nd.path);
1132 path_put(&old_nd.path);
1133 return err;
1137 * create a new mount for userspace and request it to be added into the
1138 * namespace's tree
1139 * noinline this do_mount helper to save do_mount stack space.
1141 static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
1142 int mnt_flags, char *name, void *data)
1144 struct vfsmount *mnt;
1146 if (!type || !memchr(type, 0, PAGE_SIZE))
1147 return -EINVAL;
1149 /* we need capabilities... */
1150 if (!capable(CAP_SYS_ADMIN))
1151 return -EPERM;
1153 mnt = do_kern_mount(type, flags, name, data);
1154 if (IS_ERR(mnt))
1155 return PTR_ERR(mnt);
1157 return do_add_mount(mnt, nd, mnt_flags, NULL);
1161 * add a mount into a namespace's mount tree
1162 * - provide the option of adding the new mount to an expiration list
1164 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1165 int mnt_flags, struct list_head *fslist)
1167 int err;
1169 down_write(&namespace_sem);
1170 /* Something was mounted here while we slept */
1171 while (d_mountpoint(nd->path.dentry) &&
1172 follow_down(&nd->path.mnt, &nd->path.dentry))
1174 err = -EINVAL;
1175 if (!check_mnt(nd->path.mnt))
1176 goto unlock;
1178 /* Refuse the same filesystem on the same mount point */
1179 err = -EBUSY;
1180 if (nd->path.mnt->mnt_sb == newmnt->mnt_sb &&
1181 nd->path.mnt->mnt_root == nd->path.dentry)
1182 goto unlock;
1184 err = -EINVAL;
1185 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1186 goto unlock;
1188 newmnt->mnt_flags = mnt_flags;
1189 if ((err = graft_tree(newmnt, nd)))
1190 goto unlock;
1192 if (fslist) {
1193 /* add to the specified expiration list */
1194 spin_lock(&vfsmount_lock);
1195 list_add_tail(&newmnt->mnt_expire, fslist);
1196 spin_unlock(&vfsmount_lock);
1198 up_write(&namespace_sem);
1199 return 0;
1201 unlock:
1202 up_write(&namespace_sem);
1203 mntput(newmnt);
1204 return err;
1207 EXPORT_SYMBOL_GPL(do_add_mount);
1209 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1210 struct list_head *umounts)
1212 spin_lock(&vfsmount_lock);
1215 * Check if mount is still attached, if not, let whoever holds it deal
1216 * with the sucker
1218 if (mnt->mnt_parent == mnt) {
1219 spin_unlock(&vfsmount_lock);
1220 return;
1224 * Check that it is still dead: the count should now be 2 - as
1225 * contributed by the vfsmount parent and the mntget above
1227 if (!propagate_mount_busy(mnt, 2)) {
1228 /* delete from the namespace */
1229 touch_mnt_namespace(mnt->mnt_ns);
1230 list_del_init(&mnt->mnt_list);
1231 mnt->mnt_ns = NULL;
1232 umount_tree(mnt, 1, umounts);
1233 spin_unlock(&vfsmount_lock);
1234 } else {
1236 * Someone brought it back to life whilst we didn't have any
1237 * locks held so return it to the expiration list
1239 list_add_tail(&mnt->mnt_expire, mounts);
1240 spin_unlock(&vfsmount_lock);
1245 * go through the vfsmounts we've just consigned to the graveyard to
1246 * - check that they're still dead
1247 * - delete the vfsmount from the appropriate namespace under lock
1248 * - dispose of the corpse
1250 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1252 struct mnt_namespace *ns;
1253 struct vfsmount *mnt;
1255 while (!list_empty(graveyard)) {
1256 LIST_HEAD(umounts);
1257 mnt = list_first_entry(graveyard, struct vfsmount, mnt_expire);
1258 list_del_init(&mnt->mnt_expire);
1260 /* don't do anything if the namespace is dead - all the
1261 * vfsmounts from it are going away anyway */
1262 ns = mnt->mnt_ns;
1263 if (!ns || !ns->root)
1264 continue;
1265 get_mnt_ns(ns);
1267 spin_unlock(&vfsmount_lock);
1268 down_write(&namespace_sem);
1269 expire_mount(mnt, mounts, &umounts);
1270 up_write(&namespace_sem);
1271 release_mounts(&umounts);
1272 mntput(mnt);
1273 put_mnt_ns(ns);
1274 spin_lock(&vfsmount_lock);
1279 * process a list of expirable mountpoints with the intent of discarding any
1280 * mountpoints that aren't in use and haven't been touched since last we came
1281 * here
1283 void mark_mounts_for_expiry(struct list_head *mounts)
1285 struct vfsmount *mnt, *next;
1286 LIST_HEAD(graveyard);
1288 if (list_empty(mounts))
1289 return;
1291 spin_lock(&vfsmount_lock);
1293 /* extract from the expiration list every vfsmount that matches the
1294 * following criteria:
1295 * - only referenced by its parent vfsmount
1296 * - still marked for expiry (marked on the last call here; marks are
1297 * cleared by mntput())
1299 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1300 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1301 atomic_read(&mnt->mnt_count) != 1)
1302 continue;
1304 mntget(mnt);
1305 list_move(&mnt->mnt_expire, &graveyard);
1308 expire_mount_list(&graveyard, mounts);
1310 spin_unlock(&vfsmount_lock);
1313 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1316 * Ripoff of 'select_parent()'
1318 * search the list of submounts for a given mountpoint, and move any
1319 * shrinkable submounts to the 'graveyard' list.
1321 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1323 struct vfsmount *this_parent = parent;
1324 struct list_head *next;
1325 int found = 0;
1327 repeat:
1328 next = this_parent->mnt_mounts.next;
1329 resume:
1330 while (next != &this_parent->mnt_mounts) {
1331 struct list_head *tmp = next;
1332 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1334 next = tmp->next;
1335 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1336 continue;
1338 * Descend a level if the d_mounts list is non-empty.
1340 if (!list_empty(&mnt->mnt_mounts)) {
1341 this_parent = mnt;
1342 goto repeat;
1345 if (!propagate_mount_busy(mnt, 1)) {
1346 mntget(mnt);
1347 list_move_tail(&mnt->mnt_expire, graveyard);
1348 found++;
1352 * All done at this level ... ascend and resume the search
1354 if (this_parent != parent) {
1355 next = this_parent->mnt_child.next;
1356 this_parent = this_parent->mnt_parent;
1357 goto resume;
1359 return found;
1363 * process a list of expirable mountpoints with the intent of discarding any
1364 * submounts of a specific parent mountpoint
1366 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1368 LIST_HEAD(graveyard);
1369 int found;
1371 spin_lock(&vfsmount_lock);
1373 /* extract submounts of 'mountpoint' from the expiration list */
1374 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1375 expire_mount_list(&graveyard, mounts);
1377 spin_unlock(&vfsmount_lock);
1380 EXPORT_SYMBOL_GPL(shrink_submounts);
1383 * Some copy_from_user() implementations do not return the exact number of
1384 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1385 * Note that this function differs from copy_from_user() in that it will oops
1386 * on bad values of `to', rather than returning a short copy.
1388 static long exact_copy_from_user(void *to, const void __user * from,
1389 unsigned long n)
1391 char *t = to;
1392 const char __user *f = from;
1393 char c;
1395 if (!access_ok(VERIFY_READ, from, n))
1396 return n;
1398 while (n) {
1399 if (__get_user(c, f)) {
1400 memset(t, 0, n);
1401 break;
1403 *t++ = c;
1404 f++;
1405 n--;
1407 return n;
1410 int copy_mount_options(const void __user * data, unsigned long *where)
1412 int i;
1413 unsigned long page;
1414 unsigned long size;
1416 *where = 0;
1417 if (!data)
1418 return 0;
1420 if (!(page = __get_free_page(GFP_KERNEL)))
1421 return -ENOMEM;
1423 /* We only care that *some* data at the address the user
1424 * gave us is valid. Just in case, we'll zero
1425 * the remainder of the page.
1427 /* copy_from_user cannot cross TASK_SIZE ! */
1428 size = TASK_SIZE - (unsigned long)data;
1429 if (size > PAGE_SIZE)
1430 size = PAGE_SIZE;
1432 i = size - exact_copy_from_user((void *)page, data, size);
1433 if (!i) {
1434 free_page(page);
1435 return -EFAULT;
1437 if (i != PAGE_SIZE)
1438 memset((char *)page + i, 0, PAGE_SIZE - i);
1439 *where = page;
1440 return 0;
1444 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1445 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1447 * data is a (void *) that can point to any structure up to
1448 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1449 * information (or be NULL).
1451 * Pre-0.97 versions of mount() didn't have a flags word.
1452 * When the flags word was introduced its top half was required
1453 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1454 * Therefore, if this magic number is present, it carries no information
1455 * and must be discarded.
1457 long do_mount(char *dev_name, char *dir_name, char *type_page,
1458 unsigned long flags, void *data_page)
1460 struct nameidata nd;
1461 int retval = 0;
1462 int mnt_flags = 0;
1464 /* Discard magic */
1465 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1466 flags &= ~MS_MGC_MSK;
1468 /* Basic sanity checks */
1470 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1471 return -EINVAL;
1472 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1473 return -EINVAL;
1475 if (data_page)
1476 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1478 /* Separate the per-mountpoint flags */
1479 if (flags & MS_NOSUID)
1480 mnt_flags |= MNT_NOSUID;
1481 if (flags & MS_NODEV)
1482 mnt_flags |= MNT_NODEV;
1483 if (flags & MS_NOEXEC)
1484 mnt_flags |= MNT_NOEXEC;
1485 if (flags & MS_NOATIME)
1486 mnt_flags |= MNT_NOATIME;
1487 if (flags & MS_NODIRATIME)
1488 mnt_flags |= MNT_NODIRATIME;
1489 if (flags & MS_RELATIME)
1490 mnt_flags |= MNT_RELATIME;
1492 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1493 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
1495 /* ... and get the mountpoint */
1496 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1497 if (retval)
1498 return retval;
1500 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1501 if (retval)
1502 goto dput_out;
1504 if (flags & MS_REMOUNT)
1505 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1506 data_page);
1507 else if (flags & MS_BIND)
1508 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1509 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1510 retval = do_change_type(&nd, flags);
1511 else if (flags & MS_MOVE)
1512 retval = do_move_mount(&nd, dev_name);
1513 else
1514 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1515 dev_name, data_page);
1516 dput_out:
1517 path_put(&nd.path);
1518 return retval;
1522 * Allocate a new namespace structure and populate it with contents
1523 * copied from the namespace of the passed in task structure.
1525 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1526 struct fs_struct *fs)
1528 struct mnt_namespace *new_ns;
1529 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1530 struct vfsmount *p, *q;
1532 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1533 if (!new_ns)
1534 return ERR_PTR(-ENOMEM);
1536 atomic_set(&new_ns->count, 1);
1537 INIT_LIST_HEAD(&new_ns->list);
1538 init_waitqueue_head(&new_ns->poll);
1539 new_ns->event = 0;
1541 down_write(&namespace_sem);
1542 /* First pass: copy the tree topology */
1543 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1544 CL_COPY_ALL | CL_EXPIRE);
1545 if (!new_ns->root) {
1546 up_write(&namespace_sem);
1547 kfree(new_ns);
1548 return ERR_PTR(-ENOMEM);;
1550 spin_lock(&vfsmount_lock);
1551 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1552 spin_unlock(&vfsmount_lock);
1555 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1556 * as belonging to new namespace. We have already acquired a private
1557 * fs_struct, so tsk->fs->lock is not needed.
1559 p = mnt_ns->root;
1560 q = new_ns->root;
1561 while (p) {
1562 q->mnt_ns = new_ns;
1563 if (fs) {
1564 if (p == fs->root.mnt) {
1565 rootmnt = p;
1566 fs->root.mnt = mntget(q);
1568 if (p == fs->pwd.mnt) {
1569 pwdmnt = p;
1570 fs->pwd.mnt = mntget(q);
1572 if (p == fs->altroot.mnt) {
1573 altrootmnt = p;
1574 fs->altroot.mnt = mntget(q);
1577 p = next_mnt(p, mnt_ns->root);
1578 q = next_mnt(q, new_ns->root);
1580 up_write(&namespace_sem);
1582 if (rootmnt)
1583 mntput(rootmnt);
1584 if (pwdmnt)
1585 mntput(pwdmnt);
1586 if (altrootmnt)
1587 mntput(altrootmnt);
1589 return new_ns;
1592 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
1593 struct fs_struct *new_fs)
1595 struct mnt_namespace *new_ns;
1597 BUG_ON(!ns);
1598 get_mnt_ns(ns);
1600 if (!(flags & CLONE_NEWNS))
1601 return ns;
1603 new_ns = dup_mnt_ns(ns, new_fs);
1605 put_mnt_ns(ns);
1606 return new_ns;
1609 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1610 char __user * type, unsigned long flags,
1611 void __user * data)
1613 int retval;
1614 unsigned long data_page;
1615 unsigned long type_page;
1616 unsigned long dev_page;
1617 char *dir_page;
1619 retval = copy_mount_options(type, &type_page);
1620 if (retval < 0)
1621 return retval;
1623 dir_page = getname(dir_name);
1624 retval = PTR_ERR(dir_page);
1625 if (IS_ERR(dir_page))
1626 goto out1;
1628 retval = copy_mount_options(dev_name, &dev_page);
1629 if (retval < 0)
1630 goto out2;
1632 retval = copy_mount_options(data, &data_page);
1633 if (retval < 0)
1634 goto out3;
1636 lock_kernel();
1637 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1638 flags, (void *)data_page);
1639 unlock_kernel();
1640 free_page(data_page);
1642 out3:
1643 free_page(dev_page);
1644 out2:
1645 putname(dir_page);
1646 out1:
1647 free_page(type_page);
1648 return retval;
1652 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1653 * It can block. Requires the big lock held.
1655 void set_fs_root(struct fs_struct *fs, struct path *path)
1657 struct path old_root;
1659 write_lock(&fs->lock);
1660 old_root = fs->root;
1661 fs->root = *path;
1662 path_get(path);
1663 write_unlock(&fs->lock);
1664 if (old_root.dentry)
1665 path_put(&old_root);
1669 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1670 * It can block. Requires the big lock held.
1672 void set_fs_pwd(struct fs_struct *fs, struct path *path)
1674 struct path old_pwd;
1676 write_lock(&fs->lock);
1677 old_pwd = fs->pwd;
1678 fs->pwd = *path;
1679 path_get(path);
1680 write_unlock(&fs->lock);
1682 if (old_pwd.dentry)
1683 path_put(&old_pwd);
1686 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1688 struct task_struct *g, *p;
1689 struct fs_struct *fs;
1691 read_lock(&tasklist_lock);
1692 do_each_thread(g, p) {
1693 task_lock(p);
1694 fs = p->fs;
1695 if (fs) {
1696 atomic_inc(&fs->count);
1697 task_unlock(p);
1698 if (fs->root.dentry == old_nd->path.dentry
1699 && fs->root.mnt == old_nd->path.mnt)
1700 set_fs_root(fs, &new_nd->path);
1701 if (fs->pwd.dentry == old_nd->path.dentry
1702 && fs->pwd.mnt == old_nd->path.mnt)
1703 set_fs_pwd(fs, &new_nd->path);
1704 put_fs_struct(fs);
1705 } else
1706 task_unlock(p);
1707 } while_each_thread(g, p);
1708 read_unlock(&tasklist_lock);
1712 * pivot_root Semantics:
1713 * Moves the root file system of the current process to the directory put_old,
1714 * makes new_root as the new root file system of the current process, and sets
1715 * root/cwd of all processes which had them on the current root to new_root.
1717 * Restrictions:
1718 * The new_root and put_old must be directories, and must not be on the
1719 * same file system as the current process root. The put_old must be
1720 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1721 * pointed to by put_old must yield the same directory as new_root. No other
1722 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1724 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1725 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1726 * in this situation.
1728 * Notes:
1729 * - we don't move root/cwd if they are not at the root (reason: if something
1730 * cared enough to change them, it's probably wrong to force them elsewhere)
1731 * - it's okay to pick a root that isn't the root of a file system, e.g.
1732 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1733 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1734 * first.
1736 asmlinkage long sys_pivot_root(const char __user * new_root,
1737 const char __user * put_old)
1739 struct vfsmount *tmp;
1740 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1741 int error;
1743 if (!capable(CAP_SYS_ADMIN))
1744 return -EPERM;
1746 lock_kernel();
1748 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1749 &new_nd);
1750 if (error)
1751 goto out0;
1752 error = -EINVAL;
1753 if (!check_mnt(new_nd.path.mnt))
1754 goto out1;
1756 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1757 if (error)
1758 goto out1;
1760 error = security_sb_pivotroot(&old_nd, &new_nd);
1761 if (error) {
1762 path_put(&old_nd.path);
1763 goto out1;
1766 read_lock(&current->fs->lock);
1767 user_nd.path = current->fs->root;
1768 path_get(&current->fs->root);
1769 read_unlock(&current->fs->lock);
1770 down_write(&namespace_sem);
1771 mutex_lock(&old_nd.path.dentry->d_inode->i_mutex);
1772 error = -EINVAL;
1773 if (IS_MNT_SHARED(old_nd.path.mnt) ||
1774 IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) ||
1775 IS_MNT_SHARED(user_nd.path.mnt->mnt_parent))
1776 goto out2;
1777 if (!check_mnt(user_nd.path.mnt))
1778 goto out2;
1779 error = -ENOENT;
1780 if (IS_DEADDIR(new_nd.path.dentry->d_inode))
1781 goto out2;
1782 if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry))
1783 goto out2;
1784 if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry))
1785 goto out2;
1786 error = -EBUSY;
1787 if (new_nd.path.mnt == user_nd.path.mnt ||
1788 old_nd.path.mnt == user_nd.path.mnt)
1789 goto out2; /* loop, on the same file system */
1790 error = -EINVAL;
1791 if (user_nd.path.mnt->mnt_root != user_nd.path.dentry)
1792 goto out2; /* not a mountpoint */
1793 if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt)
1794 goto out2; /* not attached */
1795 if (new_nd.path.mnt->mnt_root != new_nd.path.dentry)
1796 goto out2; /* not a mountpoint */
1797 if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt)
1798 goto out2; /* not attached */
1799 /* make sure we can reach put_old from new_root */
1800 tmp = old_nd.path.mnt;
1801 spin_lock(&vfsmount_lock);
1802 if (tmp != new_nd.path.mnt) {
1803 for (;;) {
1804 if (tmp->mnt_parent == tmp)
1805 goto out3; /* already mounted on put_old */
1806 if (tmp->mnt_parent == new_nd.path.mnt)
1807 break;
1808 tmp = tmp->mnt_parent;
1810 if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry))
1811 goto out3;
1812 } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry))
1813 goto out3;
1814 detach_mnt(new_nd.path.mnt, &parent_nd);
1815 detach_mnt(user_nd.path.mnt, &root_parent);
1816 /* mount old root on put_old */
1817 attach_mnt(user_nd.path.mnt, &old_nd);
1818 /* mount new_root on / */
1819 attach_mnt(new_nd.path.mnt, &root_parent);
1820 touch_mnt_namespace(current->nsproxy->mnt_ns);
1821 spin_unlock(&vfsmount_lock);
1822 chroot_fs_refs(&user_nd, &new_nd);
1823 security_sb_post_pivotroot(&user_nd, &new_nd);
1824 error = 0;
1825 path_put(&root_parent.path);
1826 path_put(&parent_nd.path);
1827 out2:
1828 mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex);
1829 up_write(&namespace_sem);
1830 path_put(&user_nd.path);
1831 path_put(&old_nd.path);
1832 out1:
1833 path_put(&new_nd.path);
1834 out0:
1835 unlock_kernel();
1836 return error;
1837 out3:
1838 spin_unlock(&vfsmount_lock);
1839 goto out2;
1842 static void __init init_mount_tree(void)
1844 struct vfsmount *mnt;
1845 struct mnt_namespace *ns;
1846 struct path root;
1848 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1849 if (IS_ERR(mnt))
1850 panic("Can't create rootfs");
1851 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
1852 if (!ns)
1853 panic("Can't allocate initial namespace");
1854 atomic_set(&ns->count, 1);
1855 INIT_LIST_HEAD(&ns->list);
1856 init_waitqueue_head(&ns->poll);
1857 ns->event = 0;
1858 list_add(&mnt->mnt_list, &ns->list);
1859 ns->root = mnt;
1860 mnt->mnt_ns = ns;
1862 init_task.nsproxy->mnt_ns = ns;
1863 get_mnt_ns(ns);
1865 root.mnt = ns->root;
1866 root.dentry = ns->root->mnt_root;
1868 set_fs_pwd(current->fs, &root);
1869 set_fs_root(current->fs, &root);
1872 void __init mnt_init(void)
1874 unsigned u;
1875 int err;
1877 init_rwsem(&namespace_sem);
1879 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1880 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1882 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1884 if (!mount_hashtable)
1885 panic("Failed to allocate mount hash table\n");
1887 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
1889 for (u = 0; u < HASH_SIZE; u++)
1890 INIT_LIST_HEAD(&mount_hashtable[u]);
1892 err = sysfs_init();
1893 if (err)
1894 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1895 __FUNCTION__, err);
1896 fs_kobj = kobject_create_and_add("fs", NULL);
1897 if (!fs_kobj)
1898 printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__);
1899 init_rootfs();
1900 init_mount_tree();
1903 void __put_mnt_ns(struct mnt_namespace *ns)
1905 struct vfsmount *root = ns->root;
1906 LIST_HEAD(umount_list);
1907 ns->root = NULL;
1908 spin_unlock(&vfsmount_lock);
1909 down_write(&namespace_sem);
1910 spin_lock(&vfsmount_lock);
1911 umount_tree(root, 0, &umount_list);
1912 spin_unlock(&vfsmount_lock);
1913 up_write(&namespace_sem);
1914 release_mounts(&umount_list);
1915 kfree(ns);