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/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/namei.h>
16 #include <linux/security.h>
17 #include <linux/idr.h>
18 #include <linux/acct.h> /* acct_auto_close_mnt */
19 #include <linux/ramfs.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
26 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
27 #define HASH_SIZE (1UL << HASH_SHIFT)
30 static DEFINE_IDA(mnt_id_ida);
31 static DEFINE_IDA(mnt_group_ida);
32 static DEFINE_SPINLOCK(mnt_id_lock);
33 static int mnt_id_start = 0;
34 static int mnt_group_start = 1;
36 static struct list_head *mount_hashtable __read_mostly;
37 static struct kmem_cache *mnt_cache __read_mostly;
38 static struct rw_semaphore namespace_sem;
41 struct kobject *fs_kobj;
42 EXPORT_SYMBOL_GPL(fs_kobj);
45 * vfsmount lock may be taken for read to prevent changes to the
46 * vfsmount hash, ie. during mountpoint lookups or walking back
49 * It should be taken for write in all cases where the vfsmount
50 * tree or hash is modified or when a vfsmount structure is modified.
52 DEFINE_BRLOCK(vfsmount_lock);
54 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
58 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
62 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
65 * allocation is serialized by namespace_sem, but we need the spinlock to
66 * serialize with freeing.
68 static int mnt_alloc_id(struct mount *mnt)
73 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
74 spin_lock(&mnt_id_lock);
75 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
77 mnt_id_start = mnt->mnt_id + 1;
78 spin_unlock(&mnt_id_lock);
85 static void mnt_free_id(struct mount *mnt)
88 spin_lock(&mnt_id_lock);
89 ida_remove(&mnt_id_ida, id);
90 if (mnt_id_start > id)
92 spin_unlock(&mnt_id_lock);
96 * Allocate a new peer group ID
98 * mnt_group_ida is protected by namespace_sem
100 static int mnt_alloc_group_id(struct mount *mnt)
104 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107 res = ida_get_new_above(&mnt_group_ida,
111 mnt_group_start = mnt->mnt_group_id + 1;
117 * Release a peer group ID
119 void mnt_release_group_id(struct mount *mnt)
121 int id = mnt->mnt_group_id;
122 ida_remove(&mnt_group_ida, id);
123 if (mnt_group_start > id)
124 mnt_group_start = id;
125 mnt->mnt_group_id = 0;
129 * vfsmount lock must be held for read
131 static inline void mnt_add_count(struct mount *mnt, int n)
134 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
143 * vfsmount lock must be held for write
145 unsigned int mnt_get_count(struct mount *mnt)
148 unsigned int count = 0;
151 for_each_possible_cpu(cpu) {
152 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
157 return mnt->mnt_count;
161 static struct mount *alloc_vfsmnt(const char *name)
163 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
167 err = mnt_alloc_id(mnt);
172 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
173 if (!mnt->mnt_devname)
178 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
180 goto out_free_devname;
182 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
185 mnt->mnt_writers = 0;
188 INIT_LIST_HEAD(&mnt->mnt_hash);
189 INIT_LIST_HEAD(&mnt->mnt_child);
190 INIT_LIST_HEAD(&mnt->mnt_mounts);
191 INIT_LIST_HEAD(&mnt->mnt_list);
192 INIT_LIST_HEAD(&mnt->mnt_expire);
193 INIT_LIST_HEAD(&mnt->mnt_share);
194 INIT_LIST_HEAD(&mnt->mnt_slave_list);
195 INIT_LIST_HEAD(&mnt->mnt_slave);
196 #ifdef CONFIG_FSNOTIFY
197 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
204 kfree(mnt->mnt_devname);
209 kmem_cache_free(mnt_cache, mnt);
214 * Most r/o checks on a fs are for operations that take
215 * discrete amounts of time, like a write() or unlink().
216 * We must keep track of when those operations start
217 * (for permission checks) and when they end, so that
218 * we can determine when writes are able to occur to
222 * __mnt_is_readonly: check whether a mount is read-only
223 * @mnt: the mount to check for its write status
225 * This shouldn't be used directly ouside of the VFS.
226 * It does not guarantee that the filesystem will stay
227 * r/w, just that it is right *now*. This can not and
228 * should not be used in place of IS_RDONLY(inode).
229 * mnt_want/drop_write() will _keep_ the filesystem
232 int __mnt_is_readonly(struct vfsmount *mnt)
234 if (mnt->mnt_flags & MNT_READONLY)
236 if (mnt->mnt_sb->s_flags & MS_RDONLY)
240 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
242 static inline void mnt_inc_writers(struct mount *mnt)
245 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
251 static inline void mnt_dec_writers(struct mount *mnt)
254 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
260 static unsigned int mnt_get_writers(struct mount *mnt)
263 unsigned int count = 0;
266 for_each_possible_cpu(cpu) {
267 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
272 return mnt->mnt_writers;
276 static int mnt_is_readonly(struct vfsmount *mnt)
278 if (mnt->mnt_sb->s_readonly_remount)
280 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
282 return __mnt_is_readonly(mnt);
286 * Most r/o & frozen checks on a fs are for operations that take discrete
287 * amounts of time, like a write() or unlink(). We must keep track of when
288 * those operations start (for permission checks) and when they end, so that we
289 * can determine when writes are able to occur to a filesystem.
292 * __mnt_want_write - get write access to a mount without freeze protection
293 * @m: the mount on which to take a write
295 * This tells the low-level filesystem that a write is about to be performed to
296 * it, and makes sure that writes are allowed (mnt it read-write) before
297 * returning success. This operation does not protect against filesystem being
298 * frozen. When the write operation is finished, __mnt_drop_write() must be
299 * called. This is effectively a refcount.
301 int __mnt_want_write(struct vfsmount *m)
303 struct mount *mnt = real_mount(m);
307 mnt_inc_writers(mnt);
309 * The store to mnt_inc_writers must be visible before we pass
310 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
311 * incremented count after it has set MNT_WRITE_HOLD.
314 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
317 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
318 * be set to match its requirements. So we must not load that until
319 * MNT_WRITE_HOLD is cleared.
322 if (mnt_is_readonly(m)) {
323 mnt_dec_writers(mnt);
332 * mnt_want_write - get write access to a mount
333 * @m: the mount on which to take a write
335 * This tells the low-level filesystem that a write is about to be performed to
336 * it, and makes sure that writes are allowed (mount is read-write, filesystem
337 * is not frozen) before returning success. When the write operation is
338 * finished, mnt_drop_write() must be called. This is effectively a refcount.
340 int mnt_want_write(struct vfsmount *m)
344 sb_start_write(m->mnt_sb);
345 ret = __mnt_want_write(m);
347 sb_end_write(m->mnt_sb);
350 EXPORT_SYMBOL_GPL(mnt_want_write);
353 * mnt_clone_write - get write access to a mount
354 * @mnt: the mount on which to take a write
356 * This is effectively like mnt_want_write, except
357 * it must only be used to take an extra write reference
358 * on a mountpoint that we already know has a write reference
359 * on it. This allows some optimisation.
361 * After finished, mnt_drop_write must be called as usual to
362 * drop the reference.
364 int mnt_clone_write(struct vfsmount *mnt)
366 /* superblock may be r/o */
367 if (__mnt_is_readonly(mnt))
370 mnt_inc_writers(real_mount(mnt));
374 EXPORT_SYMBOL_GPL(mnt_clone_write);
377 * __mnt_want_write_file - get write access to a file's mount
378 * @file: the file who's mount on which to take a write
380 * This is like __mnt_want_write, but it takes a file and can
381 * do some optimisations if the file is open for write already
383 int __mnt_want_write_file(struct file *file)
385 struct inode *inode = file->f_dentry->d_inode;
387 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
388 return __mnt_want_write(file->f_path.mnt);
390 return mnt_clone_write(file->f_path.mnt);
394 * mnt_want_write_file - get write access to a file's mount
395 * @file: the file who's mount on which to take a write
397 * This is like mnt_want_write, but it takes a file and can
398 * do some optimisations if the file is open for write already
400 int mnt_want_write_file(struct file *file)
404 sb_start_write(file->f_path.mnt->mnt_sb);
405 ret = __mnt_want_write_file(file);
407 sb_end_write(file->f_path.mnt->mnt_sb);
410 EXPORT_SYMBOL_GPL(mnt_want_write_file);
413 * __mnt_drop_write - give up write access to a mount
414 * @mnt: the mount on which to give up write access
416 * Tells the low-level filesystem that we are done
417 * performing writes to it. Must be matched with
418 * __mnt_want_write() call above.
420 void __mnt_drop_write(struct vfsmount *mnt)
423 mnt_dec_writers(real_mount(mnt));
428 * mnt_drop_write - give up write access to a mount
429 * @mnt: the mount on which to give up write access
431 * Tells the low-level filesystem that we are done performing writes to it and
432 * also allows filesystem to be frozen again. Must be matched with
433 * mnt_want_write() call above.
435 void mnt_drop_write(struct vfsmount *mnt)
437 __mnt_drop_write(mnt);
438 sb_end_write(mnt->mnt_sb);
440 EXPORT_SYMBOL_GPL(mnt_drop_write);
442 void __mnt_drop_write_file(struct file *file)
444 __mnt_drop_write(file->f_path.mnt);
447 void mnt_drop_write_file(struct file *file)
449 mnt_drop_write(file->f_path.mnt);
451 EXPORT_SYMBOL(mnt_drop_write_file);
453 static int mnt_make_readonly(struct mount *mnt)
457 br_write_lock(&vfsmount_lock);
458 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
460 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
461 * should be visible before we do.
466 * With writers on hold, if this value is zero, then there are
467 * definitely no active writers (although held writers may subsequently
468 * increment the count, they'll have to wait, and decrement it after
469 * seeing MNT_READONLY).
471 * It is OK to have counter incremented on one CPU and decremented on
472 * another: the sum will add up correctly. The danger would be when we
473 * sum up each counter, if we read a counter before it is incremented,
474 * but then read another CPU's count which it has been subsequently
475 * decremented from -- we would see more decrements than we should.
476 * MNT_WRITE_HOLD protects against this scenario, because
477 * mnt_want_write first increments count, then smp_mb, then spins on
478 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
479 * we're counting up here.
481 if (mnt_get_writers(mnt) > 0)
484 mnt->mnt.mnt_flags |= MNT_READONLY;
486 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
487 * that become unheld will see MNT_READONLY.
490 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
491 br_write_unlock(&vfsmount_lock);
495 static void __mnt_unmake_readonly(struct mount *mnt)
497 br_write_lock(&vfsmount_lock);
498 mnt->mnt.mnt_flags &= ~MNT_READONLY;
499 br_write_unlock(&vfsmount_lock);
502 int sb_prepare_remount_readonly(struct super_block *sb)
507 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
508 if (atomic_long_read(&sb->s_remove_count))
511 br_write_lock(&vfsmount_lock);
512 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
513 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
514 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
516 if (mnt_get_writers(mnt) > 0) {
522 if (!err && atomic_long_read(&sb->s_remove_count))
526 sb->s_readonly_remount = 1;
529 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
530 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
531 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
533 br_write_unlock(&vfsmount_lock);
538 static void free_vfsmnt(struct mount *mnt)
540 kfree(mnt->mnt_devname);
543 free_percpu(mnt->mnt_pcp);
545 kmem_cache_free(mnt_cache, mnt);
549 * find the first or last mount at @dentry on vfsmount @mnt depending on
550 * @dir. If @dir is set return the first mount else return the last mount.
551 * vfsmount_lock must be held for read or write.
553 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
556 struct list_head *head = mount_hashtable + hash(mnt, dentry);
557 struct list_head *tmp = head;
558 struct mount *p, *found = NULL;
561 tmp = dir ? tmp->next : tmp->prev;
565 p = list_entry(tmp, struct mount, mnt_hash);
566 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
575 * lookup_mnt - Return the first child mount mounted at path
577 * "First" means first mounted chronologically. If you create the
580 * mount /dev/sda1 /mnt
581 * mount /dev/sda2 /mnt
582 * mount /dev/sda3 /mnt
584 * Then lookup_mnt() on the base /mnt dentry in the root mount will
585 * return successively the root dentry and vfsmount of /dev/sda1, then
586 * /dev/sda2, then /dev/sda3, then NULL.
588 * lookup_mnt takes a reference to the found vfsmount.
590 struct vfsmount *lookup_mnt(struct path *path)
592 struct mount *child_mnt;
594 br_read_lock(&vfsmount_lock);
595 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
597 mnt_add_count(child_mnt, 1);
598 br_read_unlock(&vfsmount_lock);
599 return &child_mnt->mnt;
601 br_read_unlock(&vfsmount_lock);
606 static inline int check_mnt(struct mount *mnt)
608 return mnt->mnt_ns == current->nsproxy->mnt_ns;
612 * vfsmount lock must be held for write
614 static void touch_mnt_namespace(struct mnt_namespace *ns)
618 wake_up_interruptible(&ns->poll);
623 * vfsmount lock must be held for write
625 static void __touch_mnt_namespace(struct mnt_namespace *ns)
627 if (ns && ns->event != event) {
629 wake_up_interruptible(&ns->poll);
634 * Clear dentry's mounted state if it has no remaining mounts.
635 * vfsmount_lock must be held for write.
637 static void dentry_reset_mounted(struct dentry *dentry)
641 for (u = 0; u < HASH_SIZE; u++) {
644 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
645 if (p->mnt_mountpoint == dentry)
649 spin_lock(&dentry->d_lock);
650 dentry->d_flags &= ~DCACHE_MOUNTED;
651 spin_unlock(&dentry->d_lock);
655 * vfsmount lock must be held for write
657 static void detach_mnt(struct mount *mnt, struct path *old_path)
659 old_path->dentry = mnt->mnt_mountpoint;
660 old_path->mnt = &mnt->mnt_parent->mnt;
661 mnt->mnt_parent = mnt;
662 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
663 list_del_init(&mnt->mnt_child);
664 list_del_init(&mnt->mnt_hash);
665 dentry_reset_mounted(old_path->dentry);
669 * vfsmount lock must be held for write
671 void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
672 struct mount *child_mnt)
674 mnt_add_count(mnt, 1); /* essentially, that's mntget */
675 child_mnt->mnt_mountpoint = dget(dentry);
676 child_mnt->mnt_parent = mnt;
677 spin_lock(&dentry->d_lock);
678 dentry->d_flags |= DCACHE_MOUNTED;
679 spin_unlock(&dentry->d_lock);
683 * vfsmount lock must be held for write
685 static void attach_mnt(struct mount *mnt, struct path *path)
687 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
688 list_add_tail(&mnt->mnt_hash, mount_hashtable +
689 hash(path->mnt, path->dentry));
690 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
694 * vfsmount lock must be held for write
696 static void commit_tree(struct mount *mnt)
698 struct mount *parent = mnt->mnt_parent;
701 struct mnt_namespace *n = parent->mnt_ns;
703 BUG_ON(parent == mnt);
705 list_add_tail(&head, &mnt->mnt_list);
706 list_for_each_entry(m, &head, mnt_list)
709 list_splice(&head, n->list.prev);
711 list_add_tail(&mnt->mnt_hash, mount_hashtable +
712 hash(&parent->mnt, mnt->mnt_mountpoint));
713 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
714 touch_mnt_namespace(n);
717 static struct mount *next_mnt(struct mount *p, struct mount *root)
719 struct list_head *next = p->mnt_mounts.next;
720 if (next == &p->mnt_mounts) {
724 next = p->mnt_child.next;
725 if (next != &p->mnt_parent->mnt_mounts)
730 return list_entry(next, struct mount, mnt_child);
733 static struct mount *skip_mnt_tree(struct mount *p)
735 struct list_head *prev = p->mnt_mounts.prev;
736 while (prev != &p->mnt_mounts) {
737 p = list_entry(prev, struct mount, mnt_child);
738 prev = p->mnt_mounts.prev;
744 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
750 return ERR_PTR(-ENODEV);
752 mnt = alloc_vfsmnt(name);
754 return ERR_PTR(-ENOMEM);
756 if (flags & MS_KERNMOUNT)
757 mnt->mnt.mnt_flags = MNT_INTERNAL;
759 root = mount_fs(type, flags, name, data);
762 return ERR_CAST(root);
765 mnt->mnt.mnt_root = root;
766 mnt->mnt.mnt_sb = root->d_sb;
767 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
768 mnt->mnt_parent = mnt;
769 br_write_lock(&vfsmount_lock);
770 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
771 br_write_unlock(&vfsmount_lock);
774 EXPORT_SYMBOL_GPL(vfs_kern_mount);
776 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
779 struct super_block *sb = old->mnt.mnt_sb;
783 mnt = alloc_vfsmnt(old->mnt_devname);
785 return ERR_PTR(-ENOMEM);
787 if (flag & (CL_SLAVE | CL_PRIVATE))
788 mnt->mnt_group_id = 0; /* not a peer of original */
790 mnt->mnt_group_id = old->mnt_group_id;
792 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
793 err = mnt_alloc_group_id(mnt);
798 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
799 atomic_inc(&sb->s_active);
800 mnt->mnt.mnt_sb = sb;
801 mnt->mnt.mnt_root = dget(root);
802 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
803 mnt->mnt_parent = mnt;
804 br_write_lock(&vfsmount_lock);
805 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
806 br_write_unlock(&vfsmount_lock);
808 if (flag & CL_SLAVE) {
809 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
810 mnt->mnt_master = old;
811 CLEAR_MNT_SHARED(mnt);
812 } else if (!(flag & CL_PRIVATE)) {
813 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
814 list_add(&mnt->mnt_share, &old->mnt_share);
815 if (IS_MNT_SLAVE(old))
816 list_add(&mnt->mnt_slave, &old->mnt_slave);
817 mnt->mnt_master = old->mnt_master;
819 if (flag & CL_MAKE_SHARED)
822 /* stick the duplicate mount on the same expiry list
823 * as the original if that was on one */
824 if (flag & CL_EXPIRE) {
825 if (!list_empty(&old->mnt_expire))
826 list_add(&mnt->mnt_expire, &old->mnt_expire);
836 static inline void mntfree(struct mount *mnt)
838 struct vfsmount *m = &mnt->mnt;
839 struct super_block *sb = m->mnt_sb;
842 * This probably indicates that somebody messed
843 * up a mnt_want/drop_write() pair. If this
844 * happens, the filesystem was probably unable
845 * to make r/w->r/o transitions.
848 * The locking used to deal with mnt_count decrement provides barriers,
849 * so mnt_get_writers() below is safe.
851 WARN_ON(mnt_get_writers(mnt));
852 fsnotify_vfsmount_delete(m);
855 deactivate_super(sb);
858 static void mntput_no_expire(struct mount *mnt)
862 br_read_lock(&vfsmount_lock);
863 if (likely(mnt->mnt_ns)) {
864 /* shouldn't be the last one */
865 mnt_add_count(mnt, -1);
866 br_read_unlock(&vfsmount_lock);
869 br_read_unlock(&vfsmount_lock);
871 br_write_lock(&vfsmount_lock);
872 mnt_add_count(mnt, -1);
873 if (mnt_get_count(mnt)) {
874 br_write_unlock(&vfsmount_lock);
878 mnt_add_count(mnt, -1);
879 if (likely(mnt_get_count(mnt)))
881 br_write_lock(&vfsmount_lock);
883 if (unlikely(mnt->mnt_pinned)) {
884 mnt_add_count(mnt, mnt->mnt_pinned + 1);
886 br_write_unlock(&vfsmount_lock);
887 acct_auto_close_mnt(&mnt->mnt);
891 list_del(&mnt->mnt_instance);
892 br_write_unlock(&vfsmount_lock);
896 void mntput(struct vfsmount *mnt)
899 struct mount *m = real_mount(mnt);
900 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
901 if (unlikely(m->mnt_expiry_mark))
902 m->mnt_expiry_mark = 0;
906 EXPORT_SYMBOL(mntput);
908 struct vfsmount *mntget(struct vfsmount *mnt)
911 mnt_add_count(real_mount(mnt), 1);
914 EXPORT_SYMBOL(mntget);
916 void mnt_pin(struct vfsmount *mnt)
918 br_write_lock(&vfsmount_lock);
919 real_mount(mnt)->mnt_pinned++;
920 br_write_unlock(&vfsmount_lock);
922 EXPORT_SYMBOL(mnt_pin);
924 void mnt_unpin(struct vfsmount *m)
926 struct mount *mnt = real_mount(m);
927 br_write_lock(&vfsmount_lock);
928 if (mnt->mnt_pinned) {
929 mnt_add_count(mnt, 1);
932 br_write_unlock(&vfsmount_lock);
934 EXPORT_SYMBOL(mnt_unpin);
936 static inline void mangle(struct seq_file *m, const char *s)
938 seq_escape(m, s, " \t\n\\");
942 * Simple .show_options callback for filesystems which don't want to
943 * implement more complex mount option showing.
945 * See also save_mount_options().
947 int generic_show_options(struct seq_file *m, struct dentry *root)
952 options = rcu_dereference(root->d_sb->s_options);
954 if (options != NULL && options[0]) {
962 EXPORT_SYMBOL(generic_show_options);
965 * If filesystem uses generic_show_options(), this function should be
966 * called from the fill_super() callback.
968 * The .remount_fs callback usually needs to be handled in a special
969 * way, to make sure, that previous options are not overwritten if the
972 * Also note, that if the filesystem's .remount_fs function doesn't
973 * reset all options to their default value, but changes only newly
974 * given options, then the displayed options will not reflect reality
977 void save_mount_options(struct super_block *sb, char *options)
979 BUG_ON(sb->s_options);
980 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
982 EXPORT_SYMBOL(save_mount_options);
984 void replace_mount_options(struct super_block *sb, char *options)
986 char *old = sb->s_options;
987 rcu_assign_pointer(sb->s_options, options);
993 EXPORT_SYMBOL(replace_mount_options);
995 #ifdef CONFIG_PROC_FS
996 /* iterator; we want it to have access to namespace_sem, thus here... */
997 static void *m_start(struct seq_file *m, loff_t *pos)
999 struct proc_mounts *p = proc_mounts(m);
1001 down_read(&namespace_sem);
1002 return seq_list_start(&p->ns->list, *pos);
1005 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1007 struct proc_mounts *p = proc_mounts(m);
1009 return seq_list_next(v, &p->ns->list, pos);
1012 static void m_stop(struct seq_file *m, void *v)
1014 up_read(&namespace_sem);
1017 static int m_show(struct seq_file *m, void *v)
1019 struct proc_mounts *p = proc_mounts(m);
1020 struct mount *r = list_entry(v, struct mount, mnt_list);
1021 return p->show(m, &r->mnt);
1024 const struct seq_operations mounts_op = {
1030 #endif /* CONFIG_PROC_FS */
1033 * may_umount_tree - check if a mount tree is busy
1034 * @mnt: root of mount tree
1036 * This is called to check if a tree of mounts has any
1037 * open files, pwds, chroots or sub mounts that are
1040 int may_umount_tree(struct vfsmount *m)
1042 struct mount *mnt = real_mount(m);
1043 int actual_refs = 0;
1044 int minimum_refs = 0;
1048 /* write lock needed for mnt_get_count */
1049 br_write_lock(&vfsmount_lock);
1050 for (p = mnt; p; p = next_mnt(p, mnt)) {
1051 actual_refs += mnt_get_count(p);
1054 br_write_unlock(&vfsmount_lock);
1056 if (actual_refs > minimum_refs)
1062 EXPORT_SYMBOL(may_umount_tree);
1065 * may_umount - check if a mount point is busy
1066 * @mnt: root of mount
1068 * This is called to check if a mount point has any
1069 * open files, pwds, chroots or sub mounts. If the
1070 * mount has sub mounts this will return busy
1071 * regardless of whether the sub mounts are busy.
1073 * Doesn't take quota and stuff into account. IOW, in some cases it will
1074 * give false negatives. The main reason why it's here is that we need
1075 * a non-destructive way to look for easily umountable filesystems.
1077 int may_umount(struct vfsmount *mnt)
1080 down_read(&namespace_sem);
1081 br_write_lock(&vfsmount_lock);
1082 if (propagate_mount_busy(real_mount(mnt), 2))
1084 br_write_unlock(&vfsmount_lock);
1085 up_read(&namespace_sem);
1089 EXPORT_SYMBOL(may_umount);
1091 void release_mounts(struct list_head *head)
1094 while (!list_empty(head)) {
1095 mnt = list_first_entry(head, struct mount, mnt_hash);
1096 list_del_init(&mnt->mnt_hash);
1097 if (mnt_has_parent(mnt)) {
1098 struct dentry *dentry;
1101 br_write_lock(&vfsmount_lock);
1102 dentry = mnt->mnt_mountpoint;
1103 m = mnt->mnt_parent;
1104 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1105 mnt->mnt_parent = mnt;
1107 br_write_unlock(&vfsmount_lock);
1116 * vfsmount lock must be held for write
1117 * namespace_sem must be held for write
1119 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1121 LIST_HEAD(tmp_list);
1124 for (p = mnt; p; p = next_mnt(p, mnt))
1125 list_move(&p->mnt_hash, &tmp_list);
1128 propagate_umount(&tmp_list);
1130 list_for_each_entry(p, &tmp_list, mnt_hash) {
1131 list_del_init(&p->mnt_expire);
1132 list_del_init(&p->mnt_list);
1133 __touch_mnt_namespace(p->mnt_ns);
1135 list_del_init(&p->mnt_child);
1136 if (mnt_has_parent(p)) {
1137 p->mnt_parent->mnt_ghosts++;
1138 dentry_reset_mounted(p->mnt_mountpoint);
1140 change_mnt_propagation(p, MS_PRIVATE);
1142 list_splice(&tmp_list, kill);
1145 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1147 static int do_umount(struct mount *mnt, int flags)
1149 struct super_block *sb = mnt->mnt.mnt_sb;
1151 LIST_HEAD(umount_list);
1153 retval = security_sb_umount(&mnt->mnt, flags);
1158 * Allow userspace to request a mountpoint be expired rather than
1159 * unmounting unconditionally. Unmount only happens if:
1160 * (1) the mark is already set (the mark is cleared by mntput())
1161 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1163 if (flags & MNT_EXPIRE) {
1164 if (&mnt->mnt == current->fs->root.mnt ||
1165 flags & (MNT_FORCE | MNT_DETACH))
1169 * probably don't strictly need the lock here if we examined
1170 * all race cases, but it's a slowpath.
1172 br_write_lock(&vfsmount_lock);
1173 if (mnt_get_count(mnt) != 2) {
1174 br_write_unlock(&vfsmount_lock);
1177 br_write_unlock(&vfsmount_lock);
1179 if (!xchg(&mnt->mnt_expiry_mark, 1))
1184 * If we may have to abort operations to get out of this
1185 * mount, and they will themselves hold resources we must
1186 * allow the fs to do things. In the Unix tradition of
1187 * 'Gee thats tricky lets do it in userspace' the umount_begin
1188 * might fail to complete on the first run through as other tasks
1189 * must return, and the like. Thats for the mount program to worry
1190 * about for the moment.
1193 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1194 sb->s_op->umount_begin(sb);
1198 * No sense to grab the lock for this test, but test itself looks
1199 * somewhat bogus. Suggestions for better replacement?
1200 * Ho-hum... In principle, we might treat that as umount + switch
1201 * to rootfs. GC would eventually take care of the old vfsmount.
1202 * Actually it makes sense, especially if rootfs would contain a
1203 * /reboot - static binary that would close all descriptors and
1204 * call reboot(9). Then init(8) could umount root and exec /reboot.
1206 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1208 * Special case for "unmounting" root ...
1209 * we just try to remount it readonly.
1211 down_write(&sb->s_umount);
1212 if (!(sb->s_flags & MS_RDONLY))
1213 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1214 up_write(&sb->s_umount);
1218 down_write(&namespace_sem);
1219 br_write_lock(&vfsmount_lock);
1222 if (!(flags & MNT_DETACH))
1223 shrink_submounts(mnt, &umount_list);
1226 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1227 if (!list_empty(&mnt->mnt_list))
1228 umount_tree(mnt, 1, &umount_list);
1231 br_write_unlock(&vfsmount_lock);
1232 up_write(&namespace_sem);
1233 release_mounts(&umount_list);
1238 * Now umount can handle mount points as well as block devices.
1239 * This is important for filesystems which use unnamed block devices.
1241 * We now support a flag for forced unmount like the other 'big iron'
1242 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1245 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1250 int lookup_flags = 0;
1252 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1255 if (!(flags & UMOUNT_NOFOLLOW))
1256 lookup_flags |= LOOKUP_FOLLOW;
1258 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1261 mnt = real_mount(path.mnt);
1263 if (path.dentry != path.mnt->mnt_root)
1265 if (!check_mnt(mnt))
1269 if (!capable(CAP_SYS_ADMIN))
1272 retval = do_umount(mnt, flags);
1274 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1276 mntput_no_expire(mnt);
1281 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1284 * The 2.0 compatible umount. No flags.
1286 SYSCALL_DEFINE1(oldumount, char __user *, name)
1288 return sys_umount(name, 0);
1293 static int mount_is_safe(struct path *path)
1295 if (capable(CAP_SYS_ADMIN))
1299 if (S_ISLNK(path->dentry->d_inode->i_mode))
1301 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1302 if (current_uid() != path->dentry->d_inode->i_uid)
1305 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1311 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1314 struct mount *res, *p, *q, *r;
1317 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1318 return ERR_PTR(-EINVAL);
1320 res = q = clone_mnt(mnt, dentry, flag);
1324 q->mnt_mountpoint = mnt->mnt_mountpoint;
1327 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1329 if (!is_subdir(r->mnt_mountpoint, dentry))
1332 for (s = r; s; s = next_mnt(s, r)) {
1333 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1334 s = skip_mnt_tree(s);
1337 while (p != s->mnt_parent) {
1343 path.dentry = p->mnt_mountpoint;
1344 q = clone_mnt(p, p->mnt.mnt_root, flag);
1347 br_write_lock(&vfsmount_lock);
1348 list_add_tail(&q->mnt_list, &res->mnt_list);
1349 attach_mnt(q, &path);
1350 br_write_unlock(&vfsmount_lock);
1356 LIST_HEAD(umount_list);
1357 br_write_lock(&vfsmount_lock);
1358 umount_tree(res, 0, &umount_list);
1359 br_write_unlock(&vfsmount_lock);
1360 release_mounts(&umount_list);
1365 /* Caller should check returned pointer for errors */
1367 struct vfsmount *collect_mounts(struct path *path)
1370 down_write(&namespace_sem);
1371 tree = copy_tree(real_mount(path->mnt), path->dentry,
1372 CL_COPY_ALL | CL_PRIVATE);
1373 up_write(&namespace_sem);
1379 void drop_collected_mounts(struct vfsmount *mnt)
1381 LIST_HEAD(umount_list);
1382 down_write(&namespace_sem);
1383 br_write_lock(&vfsmount_lock);
1384 umount_tree(real_mount(mnt), 0, &umount_list);
1385 br_write_unlock(&vfsmount_lock);
1386 up_write(&namespace_sem);
1387 release_mounts(&umount_list);
1390 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1391 struct vfsmount *root)
1394 int res = f(root, arg);
1397 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1398 res = f(&mnt->mnt, arg);
1405 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1409 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1410 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1411 mnt_release_group_id(p);
1415 static int invent_group_ids(struct mount *mnt, bool recurse)
1419 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1420 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1421 int err = mnt_alloc_group_id(p);
1423 cleanup_group_ids(mnt, p);
1433 * @source_mnt : mount tree to be attached
1434 * @nd : place the mount tree @source_mnt is attached
1435 * @parent_nd : if non-null, detach the source_mnt from its parent and
1436 * store the parent mount and mountpoint dentry.
1437 * (done when source_mnt is moved)
1439 * NOTE: in the table below explains the semantics when a source mount
1440 * of a given type is attached to a destination mount of a given type.
1441 * ---------------------------------------------------------------------------
1442 * | BIND MOUNT OPERATION |
1443 * |**************************************************************************
1444 * | source-->| shared | private | slave | unbindable |
1448 * |**************************************************************************
1449 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1451 * |non-shared| shared (+) | private | slave (*) | invalid |
1452 * ***************************************************************************
1453 * A bind operation clones the source mount and mounts the clone on the
1454 * destination mount.
1456 * (++) the cloned mount is propagated to all the mounts in the propagation
1457 * tree of the destination mount and the cloned mount is added to
1458 * the peer group of the source mount.
1459 * (+) the cloned mount is created under the destination mount and is marked
1460 * as shared. The cloned mount is added to the peer group of the source
1462 * (+++) the mount is propagated to all the mounts in the propagation tree
1463 * of the destination mount and the cloned mount is made slave
1464 * of the same master as that of the source mount. The cloned mount
1465 * is marked as 'shared and slave'.
1466 * (*) the cloned mount is made a slave of the same master as that of the
1469 * ---------------------------------------------------------------------------
1470 * | MOVE MOUNT OPERATION |
1471 * |**************************************************************************
1472 * | source-->| shared | private | slave | unbindable |
1476 * |**************************************************************************
1477 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1479 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1480 * ***************************************************************************
1482 * (+) the mount is moved to the destination. And is then propagated to
1483 * all the mounts in the propagation tree of the destination mount.
1484 * (+*) the mount is moved to the destination.
1485 * (+++) the mount is moved to the destination and is then propagated to
1486 * all the mounts belonging to the destination mount's propagation tree.
1487 * the mount is marked as 'shared and slave'.
1488 * (*) the mount continues to be a slave at the new location.
1490 * if the source mount is a tree, the operations explained above is
1491 * applied to each mount in the tree.
1492 * Must be called without spinlocks held, since this function can sleep
1495 static int attach_recursive_mnt(struct mount *source_mnt,
1496 struct path *path, struct path *parent_path)
1498 LIST_HEAD(tree_list);
1499 struct mount *dest_mnt = real_mount(path->mnt);
1500 struct dentry *dest_dentry = path->dentry;
1501 struct mount *child, *p;
1504 if (IS_MNT_SHARED(dest_mnt)) {
1505 err = invent_group_ids(source_mnt, true);
1509 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1511 goto out_cleanup_ids;
1513 br_write_lock(&vfsmount_lock);
1515 if (IS_MNT_SHARED(dest_mnt)) {
1516 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1520 detach_mnt(source_mnt, parent_path);
1521 attach_mnt(source_mnt, path);
1522 touch_mnt_namespace(source_mnt->mnt_ns);
1524 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1525 commit_tree(source_mnt);
1528 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1529 list_del_init(&child->mnt_hash);
1532 br_write_unlock(&vfsmount_lock);
1537 if (IS_MNT_SHARED(dest_mnt))
1538 cleanup_group_ids(source_mnt, NULL);
1543 static int lock_mount(struct path *path)
1545 struct vfsmount *mnt;
1547 mutex_lock(&path->dentry->d_inode->i_mutex);
1548 if (unlikely(cant_mount(path->dentry))) {
1549 mutex_unlock(&path->dentry->d_inode->i_mutex);
1552 down_write(&namespace_sem);
1553 mnt = lookup_mnt(path);
1556 up_write(&namespace_sem);
1557 mutex_unlock(&path->dentry->d_inode->i_mutex);
1560 path->dentry = dget(mnt->mnt_root);
1564 static void unlock_mount(struct path *path)
1566 up_write(&namespace_sem);
1567 mutex_unlock(&path->dentry->d_inode->i_mutex);
1570 static int graft_tree(struct mount *mnt, struct path *path)
1572 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1575 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1576 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1579 if (d_unlinked(path->dentry))
1582 return attach_recursive_mnt(mnt, path, NULL);
1586 * Sanity check the flags to change_mnt_propagation.
1589 static int flags_to_propagation_type(int flags)
1591 int type = flags & ~(MS_REC | MS_SILENT);
1593 /* Fail if any non-propagation flags are set */
1594 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1596 /* Only one propagation flag should be set */
1597 if (!is_power_of_2(type))
1603 * recursively change the type of the mountpoint.
1605 static int do_change_type(struct path *path, int flag)
1608 struct mount *mnt = real_mount(path->mnt);
1609 int recurse = flag & MS_REC;
1613 if (!capable(CAP_SYS_ADMIN))
1616 if (path->dentry != path->mnt->mnt_root)
1619 type = flags_to_propagation_type(flag);
1623 down_write(&namespace_sem);
1624 if (type == MS_SHARED) {
1625 err = invent_group_ids(mnt, recurse);
1630 br_write_lock(&vfsmount_lock);
1631 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1632 change_mnt_propagation(m, type);
1633 br_write_unlock(&vfsmount_lock);
1636 up_write(&namespace_sem);
1641 * do loopback mount.
1643 static int do_loopback(struct path *path, char *old_name,
1646 LIST_HEAD(umount_list);
1647 struct path old_path;
1648 struct mount *mnt = NULL, *old;
1649 int err = mount_is_safe(path);
1652 if (!old_name || !*old_name)
1654 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1658 err = lock_mount(path);
1662 old = real_mount(old_path.mnt);
1665 if (IS_MNT_UNBINDABLE(old))
1668 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1672 mnt = copy_tree(old, old_path.dentry, 0);
1674 mnt = clone_mnt(old, old_path.dentry, 0);
1681 err = graft_tree(mnt, path);
1683 br_write_lock(&vfsmount_lock);
1684 umount_tree(mnt, 0, &umount_list);
1685 br_write_unlock(&vfsmount_lock);
1689 release_mounts(&umount_list);
1691 path_put(&old_path);
1695 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1698 int readonly_request = 0;
1700 if (ms_flags & MS_RDONLY)
1701 readonly_request = 1;
1702 if (readonly_request == __mnt_is_readonly(mnt))
1705 if (readonly_request)
1706 error = mnt_make_readonly(real_mount(mnt));
1708 __mnt_unmake_readonly(real_mount(mnt));
1713 * change filesystem flags. dir should be a physical root of filesystem.
1714 * If you've mounted a non-root directory somewhere and want to do remount
1715 * on it - tough luck.
1717 static int do_remount(struct path *path, int flags, int mnt_flags,
1721 struct super_block *sb = path->mnt->mnt_sb;
1722 struct mount *mnt = real_mount(path->mnt);
1724 if (!capable(CAP_SYS_ADMIN))
1727 if (!check_mnt(mnt))
1730 if (path->dentry != path->mnt->mnt_root)
1733 err = security_sb_remount(sb, data);
1737 down_write(&sb->s_umount);
1738 if (flags & MS_BIND)
1739 err = change_mount_flags(path->mnt, flags);
1741 err = do_remount_sb(sb, flags, data, 0);
1743 br_write_lock(&vfsmount_lock);
1744 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1745 mnt->mnt.mnt_flags = mnt_flags;
1746 br_write_unlock(&vfsmount_lock);
1748 up_write(&sb->s_umount);
1750 br_write_lock(&vfsmount_lock);
1751 touch_mnt_namespace(mnt->mnt_ns);
1752 br_write_unlock(&vfsmount_lock);
1757 static inline int tree_contains_unbindable(struct mount *mnt)
1760 for (p = mnt; p; p = next_mnt(p, mnt)) {
1761 if (IS_MNT_UNBINDABLE(p))
1767 static int do_move_mount(struct path *path, char *old_name)
1769 struct path old_path, parent_path;
1773 if (!capable(CAP_SYS_ADMIN))
1775 if (!old_name || !*old_name)
1777 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1781 err = lock_mount(path);
1785 old = real_mount(old_path.mnt);
1786 p = real_mount(path->mnt);
1789 if (!check_mnt(p) || !check_mnt(old))
1792 if (d_unlinked(path->dentry))
1796 if (old_path.dentry != old_path.mnt->mnt_root)
1799 if (!mnt_has_parent(old))
1802 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1803 S_ISDIR(old_path.dentry->d_inode->i_mode))
1806 * Don't move a mount residing in a shared parent.
1808 if (IS_MNT_SHARED(old->mnt_parent))
1811 * Don't move a mount tree containing unbindable mounts to a destination
1812 * mount which is shared.
1814 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1817 for (; mnt_has_parent(p); p = p->mnt_parent)
1821 err = attach_recursive_mnt(old, path, &parent_path);
1825 /* if the mount is moved, it should no longer be expire
1827 list_del_init(&old->mnt_expire);
1832 path_put(&parent_path);
1833 path_put(&old_path);
1837 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1840 const char *subtype = strchr(fstype, '.');
1849 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1851 if (!mnt->mnt_sb->s_subtype)
1857 return ERR_PTR(err);
1860 static struct vfsmount *
1861 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1863 struct file_system_type *type = get_fs_type(fstype);
1864 struct vfsmount *mnt;
1866 return ERR_PTR(-ENODEV);
1867 mnt = vfs_kern_mount(type, flags, name, data);
1868 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1869 !mnt->mnt_sb->s_subtype)
1870 mnt = fs_set_subtype(mnt, fstype);
1871 put_filesystem(type);
1876 * add a mount into a namespace's mount tree
1878 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1882 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1884 err = lock_mount(path);
1889 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1892 /* Refuse the same filesystem on the same mount point */
1894 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1895 path->mnt->mnt_root == path->dentry)
1899 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1902 newmnt->mnt.mnt_flags = mnt_flags;
1903 err = graft_tree(newmnt, path);
1911 * create a new mount for userspace and request it to be added into the
1914 static int do_new_mount(struct path *path, char *type, int flags,
1915 int mnt_flags, char *name, void *data)
1917 struct vfsmount *mnt;
1923 /* we need capabilities... */
1924 if (!capable(CAP_SYS_ADMIN))
1927 mnt = do_kern_mount(type, flags, name, data);
1929 return PTR_ERR(mnt);
1931 err = do_add_mount(real_mount(mnt), path, mnt_flags);
1937 int finish_automount(struct vfsmount *m, struct path *path)
1939 struct mount *mnt = real_mount(m);
1941 /* The new mount record should have at least 2 refs to prevent it being
1942 * expired before we get a chance to add it
1944 BUG_ON(mnt_get_count(mnt) < 2);
1946 if (m->mnt_sb == path->mnt->mnt_sb &&
1947 m->mnt_root == path->dentry) {
1952 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1956 /* remove m from any expiration list it may be on */
1957 if (!list_empty(&mnt->mnt_expire)) {
1958 down_write(&namespace_sem);
1959 br_write_lock(&vfsmount_lock);
1960 list_del_init(&mnt->mnt_expire);
1961 br_write_unlock(&vfsmount_lock);
1962 up_write(&namespace_sem);
1970 * mnt_set_expiry - Put a mount on an expiration list
1971 * @mnt: The mount to list.
1972 * @expiry_list: The list to add the mount to.
1974 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1976 down_write(&namespace_sem);
1977 br_write_lock(&vfsmount_lock);
1979 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1981 br_write_unlock(&vfsmount_lock);
1982 up_write(&namespace_sem);
1984 EXPORT_SYMBOL(mnt_set_expiry);
1987 * process a list of expirable mountpoints with the intent of discarding any
1988 * mountpoints that aren't in use and haven't been touched since last we came
1991 void mark_mounts_for_expiry(struct list_head *mounts)
1993 struct mount *mnt, *next;
1994 LIST_HEAD(graveyard);
1997 if (list_empty(mounts))
2000 down_write(&namespace_sem);
2001 br_write_lock(&vfsmount_lock);
2003 /* extract from the expiration list every vfsmount that matches the
2004 * following criteria:
2005 * - only referenced by its parent vfsmount
2006 * - still marked for expiry (marked on the last call here; marks are
2007 * cleared by mntput())
2009 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2010 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2011 propagate_mount_busy(mnt, 1))
2013 list_move(&mnt->mnt_expire, &graveyard);
2015 while (!list_empty(&graveyard)) {
2016 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2017 touch_mnt_namespace(mnt->mnt_ns);
2018 umount_tree(mnt, 1, &umounts);
2020 br_write_unlock(&vfsmount_lock);
2021 up_write(&namespace_sem);
2023 release_mounts(&umounts);
2026 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2029 * Ripoff of 'select_parent()'
2031 * search the list of submounts for a given mountpoint, and move any
2032 * shrinkable submounts to the 'graveyard' list.
2034 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2036 struct mount *this_parent = parent;
2037 struct list_head *next;
2041 next = this_parent->mnt_mounts.next;
2043 while (next != &this_parent->mnt_mounts) {
2044 struct list_head *tmp = next;
2045 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2048 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2051 * Descend a level if the d_mounts list is non-empty.
2053 if (!list_empty(&mnt->mnt_mounts)) {
2058 if (!propagate_mount_busy(mnt, 1)) {
2059 list_move_tail(&mnt->mnt_expire, graveyard);
2064 * All done at this level ... ascend and resume the search
2066 if (this_parent != parent) {
2067 next = this_parent->mnt_child.next;
2068 this_parent = this_parent->mnt_parent;
2075 * process a list of expirable mountpoints with the intent of discarding any
2076 * submounts of a specific parent mountpoint
2078 * vfsmount_lock must be held for write
2080 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2082 LIST_HEAD(graveyard);
2085 /* extract submounts of 'mountpoint' from the expiration list */
2086 while (select_submounts(mnt, &graveyard)) {
2087 while (!list_empty(&graveyard)) {
2088 m = list_first_entry(&graveyard, struct mount,
2090 touch_mnt_namespace(m->mnt_ns);
2091 umount_tree(m, 1, umounts);
2097 * Some copy_from_user() implementations do not return the exact number of
2098 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2099 * Note that this function differs from copy_from_user() in that it will oops
2100 * on bad values of `to', rather than returning a short copy.
2102 static long exact_copy_from_user(void *to, const void __user * from,
2106 const char __user *f = from;
2109 if (!access_ok(VERIFY_READ, from, n))
2113 if (__get_user(c, f)) {
2124 int copy_mount_options(const void __user * data, unsigned long *where)
2134 if (!(page = __get_free_page(GFP_KERNEL)))
2137 /* We only care that *some* data at the address the user
2138 * gave us is valid. Just in case, we'll zero
2139 * the remainder of the page.
2141 /* copy_from_user cannot cross TASK_SIZE ! */
2142 size = TASK_SIZE - (unsigned long)data;
2143 if (size > PAGE_SIZE)
2146 i = size - exact_copy_from_user((void *)page, data, size);
2152 memset((char *)page + i, 0, PAGE_SIZE - i);
2157 int copy_mount_string(const void __user *data, char **where)
2166 tmp = strndup_user(data, PAGE_SIZE);
2168 return PTR_ERR(tmp);
2175 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2176 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2178 * data is a (void *) that can point to any structure up to
2179 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2180 * information (or be NULL).
2182 * Pre-0.97 versions of mount() didn't have a flags word.
2183 * When the flags word was introduced its top half was required
2184 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2185 * Therefore, if this magic number is present, it carries no information
2186 * and must be discarded.
2188 long do_mount(char *dev_name, char *dir_name, char *type_page,
2189 unsigned long flags, void *data_page)
2196 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2197 flags &= ~MS_MGC_MSK;
2199 /* Basic sanity checks */
2201 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2205 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2207 /* ... and get the mountpoint */
2208 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2212 retval = security_sb_mount(dev_name, &path,
2213 type_page, flags, data_page);
2217 /* Default to relatime unless overriden */
2218 if (!(flags & MS_NOATIME))
2219 mnt_flags |= MNT_RELATIME;
2221 /* Separate the per-mountpoint flags */
2222 if (flags & MS_NOSUID)
2223 mnt_flags |= MNT_NOSUID;
2224 if (flags & MS_NODEV)
2225 mnt_flags |= MNT_NODEV;
2226 if (flags & MS_NOEXEC)
2227 mnt_flags |= MNT_NOEXEC;
2228 if (flags & MS_NOATIME)
2229 mnt_flags |= MNT_NOATIME;
2230 if (flags & MS_NODIRATIME)
2231 mnt_flags |= MNT_NODIRATIME;
2232 if (flags & MS_STRICTATIME)
2233 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2234 if (flags & MS_RDONLY)
2235 mnt_flags |= MNT_READONLY;
2237 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2238 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2241 if (flags & MS_REMOUNT)
2242 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2244 else if (flags & MS_BIND)
2245 retval = do_loopback(&path, dev_name, flags & MS_REC);
2246 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2247 retval = do_change_type(&path, flags);
2248 else if (flags & MS_MOVE)
2249 retval = do_move_mount(&path, dev_name);
2251 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2252 dev_name, data_page);
2258 static struct mnt_namespace *alloc_mnt_ns(void)
2260 struct mnt_namespace *new_ns;
2262 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2264 return ERR_PTR(-ENOMEM);
2265 atomic_set(&new_ns->count, 1);
2266 new_ns->root = NULL;
2267 INIT_LIST_HEAD(&new_ns->list);
2268 init_waitqueue_head(&new_ns->poll);
2274 * Allocate a new namespace structure and populate it with contents
2275 * copied from the namespace of the passed in task structure.
2277 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2278 struct fs_struct *fs)
2280 struct mnt_namespace *new_ns;
2281 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2282 struct mount *p, *q;
2283 struct mount *old = mnt_ns->root;
2286 new_ns = alloc_mnt_ns();
2290 down_write(&namespace_sem);
2291 /* First pass: copy the tree topology */
2292 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2294 up_write(&namespace_sem);
2296 return ERR_CAST(new);
2299 br_write_lock(&vfsmount_lock);
2300 list_add_tail(&new_ns->list, &new->mnt_list);
2301 br_write_unlock(&vfsmount_lock);
2304 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2305 * as belonging to new namespace. We have already acquired a private
2306 * fs_struct, so tsk->fs->lock is not needed.
2313 if (&p->mnt == fs->root.mnt) {
2314 fs->root.mnt = mntget(&q->mnt);
2317 if (&p->mnt == fs->pwd.mnt) {
2318 fs->pwd.mnt = mntget(&q->mnt);
2322 p = next_mnt(p, old);
2323 q = next_mnt(q, new);
2325 up_write(&namespace_sem);
2335 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2336 struct fs_struct *new_fs)
2338 struct mnt_namespace *new_ns;
2343 if (!(flags & CLONE_NEWNS))
2346 new_ns = dup_mnt_ns(ns, new_fs);
2353 * create_mnt_ns - creates a private namespace and adds a root filesystem
2354 * @mnt: pointer to the new root filesystem mountpoint
2356 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2358 struct mnt_namespace *new_ns = alloc_mnt_ns();
2359 if (!IS_ERR(new_ns)) {
2360 struct mount *mnt = real_mount(m);
2361 mnt->mnt_ns = new_ns;
2363 list_add(&new_ns->list, &mnt->mnt_list);
2370 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2372 struct mnt_namespace *ns;
2373 struct super_block *s;
2377 ns = create_mnt_ns(mnt);
2379 return ERR_CAST(ns);
2381 err = vfs_path_lookup(mnt->mnt_root, mnt,
2382 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2387 return ERR_PTR(err);
2389 /* trade a vfsmount reference for active sb one */
2390 s = path.mnt->mnt_sb;
2391 atomic_inc(&s->s_active);
2393 /* lock the sucker */
2394 down_write(&s->s_umount);
2395 /* ... and return the root of (sub)tree on it */
2398 EXPORT_SYMBOL(mount_subtree);
2400 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2401 char __user *, type, unsigned long, flags, void __user *, data)
2407 unsigned long data_page;
2409 ret = copy_mount_string(type, &kernel_type);
2413 kernel_dir = getname(dir_name);
2414 if (IS_ERR(kernel_dir)) {
2415 ret = PTR_ERR(kernel_dir);
2419 ret = copy_mount_string(dev_name, &kernel_dev);
2423 ret = copy_mount_options(data, &data_page);
2427 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2428 (void *) data_page);
2430 free_page(data_page);
2434 putname(kernel_dir);
2442 * Return true if path is reachable from root
2444 * namespace_sem or vfsmount_lock is held
2446 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2447 const struct path *root)
2449 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2450 dentry = mnt->mnt_mountpoint;
2451 mnt = mnt->mnt_parent;
2453 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2456 int path_is_under(struct path *path1, struct path *path2)
2459 br_read_lock(&vfsmount_lock);
2460 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2461 br_read_unlock(&vfsmount_lock);
2464 EXPORT_SYMBOL(path_is_under);
2467 * pivot_root Semantics:
2468 * Moves the root file system of the current process to the directory put_old,
2469 * makes new_root as the new root file system of the current process, and sets
2470 * root/cwd of all processes which had them on the current root to new_root.
2473 * The new_root and put_old must be directories, and must not be on the
2474 * same file system as the current process root. The put_old must be
2475 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2476 * pointed to by put_old must yield the same directory as new_root. No other
2477 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2479 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2480 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2481 * in this situation.
2484 * - we don't move root/cwd if they are not at the root (reason: if something
2485 * cared enough to change them, it's probably wrong to force them elsewhere)
2486 * - it's okay to pick a root that isn't the root of a file system, e.g.
2487 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2488 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2491 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2492 const char __user *, put_old)
2494 struct path new, old, parent_path, root_parent, root;
2495 struct mount *new_mnt, *root_mnt;
2498 if (!capable(CAP_SYS_ADMIN))
2501 error = user_path_dir(new_root, &new);
2505 error = user_path_dir(put_old, &old);
2509 error = security_sb_pivotroot(&old, &new);
2513 get_fs_root(current->fs, &root);
2514 error = lock_mount(&old);
2519 new_mnt = real_mount(new.mnt);
2520 root_mnt = real_mount(root.mnt);
2521 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2522 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2523 IS_MNT_SHARED(root_mnt->mnt_parent))
2525 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2528 if (d_unlinked(new.dentry))
2530 if (d_unlinked(old.dentry))
2533 if (new.mnt == root.mnt ||
2534 old.mnt == root.mnt)
2535 goto out4; /* loop, on the same file system */
2537 if (root.mnt->mnt_root != root.dentry)
2538 goto out4; /* not a mountpoint */
2539 if (!mnt_has_parent(root_mnt))
2540 goto out4; /* not attached */
2541 if (new.mnt->mnt_root != new.dentry)
2542 goto out4; /* not a mountpoint */
2543 if (!mnt_has_parent(new_mnt))
2544 goto out4; /* not attached */
2545 /* make sure we can reach put_old from new_root */
2546 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2548 br_write_lock(&vfsmount_lock);
2549 detach_mnt(new_mnt, &parent_path);
2550 detach_mnt(root_mnt, &root_parent);
2551 /* mount old root on put_old */
2552 attach_mnt(root_mnt, &old);
2553 /* mount new_root on / */
2554 attach_mnt(new_mnt, &root_parent);
2555 touch_mnt_namespace(current->nsproxy->mnt_ns);
2556 br_write_unlock(&vfsmount_lock);
2557 chroot_fs_refs(&root, &new);
2562 path_put(&root_parent);
2563 path_put(&parent_path);
2575 static void __init init_mount_tree(void)
2577 struct vfsmount *mnt;
2578 struct mnt_namespace *ns;
2581 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2583 panic("Can't create rootfs");
2585 ns = create_mnt_ns(mnt);
2587 panic("Can't allocate initial namespace");
2589 init_task.nsproxy->mnt_ns = ns;
2593 root.dentry = mnt->mnt_root;
2595 set_fs_pwd(current->fs, &root);
2596 set_fs_root(current->fs, &root);
2599 void __init mnt_init(void)
2604 init_rwsem(&namespace_sem);
2606 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2607 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2609 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2611 if (!mount_hashtable)
2612 panic("Failed to allocate mount hash table\n");
2614 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2616 for (u = 0; u < HASH_SIZE; u++)
2617 INIT_LIST_HEAD(&mount_hashtable[u]);
2619 br_lock_init(&vfsmount_lock);
2623 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2625 fs_kobj = kobject_create_and_add("fs", NULL);
2627 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2632 void put_mnt_ns(struct mnt_namespace *ns)
2634 LIST_HEAD(umount_list);
2636 if (!atomic_dec_and_test(&ns->count))
2638 down_write(&namespace_sem);
2639 br_write_lock(&vfsmount_lock);
2640 umount_tree(ns->root, 0, &umount_list);
2641 br_write_unlock(&vfsmount_lock);
2642 up_write(&namespace_sem);
2643 release_mounts(&umount_list);
2647 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2649 struct vfsmount *mnt;
2650 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2653 * it is a longterm mount, don't release mnt until
2654 * we unmount before file sys is unregistered
2656 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2660 EXPORT_SYMBOL_GPL(kern_mount_data);
2662 void kern_unmount(struct vfsmount *mnt)
2664 /* release long term mount so mount point can be released */
2665 if (!IS_ERR_OR_NULL(mnt)) {
2666 br_write_lock(&vfsmount_lock);
2667 real_mount(mnt)->mnt_ns = NULL;
2668 br_write_unlock(&vfsmount_lock);
2672 EXPORT_SYMBOL(kern_unmount);
2674 bool our_mnt(struct vfsmount *mnt)
2676 return check_mnt(real_mount(mnt));