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;
277 * Most r/o checks on a fs are for operations that take
278 * discrete amounts of time, like a write() or unlink().
279 * We must keep track of when those operations start
280 * (for permission checks) and when they end, so that
281 * we can determine when writes are able to occur to
285 * mnt_want_write - get write access to a mount
286 * @m: the mount on which to take a write
288 * This tells the low-level filesystem that a write is
289 * about to be performed to it, and makes sure that
290 * writes are allowed before returning success. When
291 * the write operation is finished, mnt_drop_write()
292 * must be called. This is effectively a refcount.
294 int mnt_want_write(struct vfsmount *m)
296 struct mount *mnt = real_mount(m);
300 mnt_inc_writers(mnt);
302 * The store to mnt_inc_writers must be visible before we pass
303 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
304 * incremented count after it has set MNT_WRITE_HOLD.
307 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
310 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
311 * be set to match its requirements. So we must not load that until
312 * MNT_WRITE_HOLD is cleared.
315 if (__mnt_is_readonly(m)) {
316 mnt_dec_writers(mnt);
324 EXPORT_SYMBOL_GPL(mnt_want_write);
327 * mnt_clone_write - get write access to a mount
328 * @mnt: the mount on which to take a write
330 * This is effectively like mnt_want_write, except
331 * it must only be used to take an extra write reference
332 * on a mountpoint that we already know has a write reference
333 * on it. This allows some optimisation.
335 * After finished, mnt_drop_write must be called as usual to
336 * drop the reference.
338 int mnt_clone_write(struct vfsmount *mnt)
340 /* superblock may be r/o */
341 if (__mnt_is_readonly(mnt))
344 mnt_inc_writers(real_mount(mnt));
348 EXPORT_SYMBOL_GPL(mnt_clone_write);
351 * mnt_want_write_file - get write access to a file's mount
352 * @file: the file who's mount on which to take a write
354 * This is like mnt_want_write, but it takes a file and can
355 * do some optimisations if the file is open for write already
357 int mnt_want_write_file(struct file *file)
359 struct inode *inode = file->f_dentry->d_inode;
360 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
361 return mnt_want_write(file->f_path.mnt);
363 return mnt_clone_write(file->f_path.mnt);
365 EXPORT_SYMBOL_GPL(mnt_want_write_file);
368 * mnt_drop_write - give up write access to a mount
369 * @mnt: the mount on which to give up write access
371 * Tells the low-level filesystem that we are done
372 * performing writes to it. Must be matched with
373 * mnt_want_write() call above.
375 void mnt_drop_write(struct vfsmount *mnt)
378 mnt_dec_writers(real_mount(mnt));
381 EXPORT_SYMBOL_GPL(mnt_drop_write);
383 void mnt_drop_write_file(struct file *file)
385 mnt_drop_write(file->f_path.mnt);
387 EXPORT_SYMBOL(mnt_drop_write_file);
389 static int mnt_make_readonly(struct mount *mnt)
393 br_write_lock(vfsmount_lock);
394 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
396 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
397 * should be visible before we do.
402 * With writers on hold, if this value is zero, then there are
403 * definitely no active writers (although held writers may subsequently
404 * increment the count, they'll have to wait, and decrement it after
405 * seeing MNT_READONLY).
407 * It is OK to have counter incremented on one CPU and decremented on
408 * another: the sum will add up correctly. The danger would be when we
409 * sum up each counter, if we read a counter before it is incremented,
410 * but then read another CPU's count which it has been subsequently
411 * decremented from -- we would see more decrements than we should.
412 * MNT_WRITE_HOLD protects against this scenario, because
413 * mnt_want_write first increments count, then smp_mb, then spins on
414 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
415 * we're counting up here.
417 if (mnt_get_writers(mnt) > 0)
420 mnt->mnt.mnt_flags |= MNT_READONLY;
422 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
423 * that become unheld will see MNT_READONLY.
426 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
427 br_write_unlock(vfsmount_lock);
431 static void __mnt_unmake_readonly(struct mount *mnt)
433 br_write_lock(vfsmount_lock);
434 mnt->mnt.mnt_flags &= ~MNT_READONLY;
435 br_write_unlock(vfsmount_lock);
438 static void free_vfsmnt(struct mount *mnt)
440 kfree(mnt->mnt_devname);
443 free_percpu(mnt->mnt_pcp);
445 kmem_cache_free(mnt_cache, mnt);
449 * find the first or last mount at @dentry on vfsmount @mnt depending on
450 * @dir. If @dir is set return the first mount else return the last mount.
451 * vfsmount_lock must be held for read or write.
453 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
456 struct list_head *head = mount_hashtable + hash(mnt, dentry);
457 struct list_head *tmp = head;
458 struct mount *p, *found = NULL;
461 tmp = dir ? tmp->next : tmp->prev;
465 p = list_entry(tmp, struct mount, mnt_hash);
466 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
475 * lookup_mnt increments the ref count before returning
476 * the vfsmount struct.
478 struct vfsmount *lookup_mnt(struct path *path)
480 struct mount *child_mnt;
482 br_read_lock(vfsmount_lock);
483 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
485 mnt_add_count(child_mnt, 1);
486 br_read_unlock(vfsmount_lock);
487 return &child_mnt->mnt;
489 br_read_unlock(vfsmount_lock);
494 static inline int check_mnt(struct mount *mnt)
496 return mnt->mnt_ns == current->nsproxy->mnt_ns;
500 * vfsmount lock must be held for write
502 static void touch_mnt_namespace(struct mnt_namespace *ns)
506 wake_up_interruptible(&ns->poll);
511 * vfsmount lock must be held for write
513 static void __touch_mnt_namespace(struct mnt_namespace *ns)
515 if (ns && ns->event != event) {
517 wake_up_interruptible(&ns->poll);
522 * Clear dentry's mounted state if it has no remaining mounts.
523 * vfsmount_lock must be held for write.
525 static void dentry_reset_mounted(struct dentry *dentry)
529 for (u = 0; u < HASH_SIZE; u++) {
532 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
533 if (p->mnt_mountpoint == dentry)
537 spin_lock(&dentry->d_lock);
538 dentry->d_flags &= ~DCACHE_MOUNTED;
539 spin_unlock(&dentry->d_lock);
543 * vfsmount lock must be held for write
545 static void detach_mnt(struct mount *mnt, struct path *old_path)
547 old_path->dentry = mnt->mnt_mountpoint;
548 old_path->mnt = &mnt->mnt_parent->mnt;
549 mnt->mnt_parent = mnt;
550 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
551 list_del_init(&mnt->mnt_child);
552 list_del_init(&mnt->mnt_hash);
553 dentry_reset_mounted(old_path->dentry);
557 * vfsmount lock must be held for write
559 void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
560 struct mount *child_mnt)
562 mnt_add_count(mnt, 1); /* essentially, that's mntget */
563 child_mnt->mnt_mountpoint = dget(dentry);
564 child_mnt->mnt_parent = mnt;
565 spin_lock(&dentry->d_lock);
566 dentry->d_flags |= DCACHE_MOUNTED;
567 spin_unlock(&dentry->d_lock);
571 * vfsmount lock must be held for write
573 static void attach_mnt(struct mount *mnt, struct path *path)
575 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
576 list_add_tail(&mnt->mnt_hash, mount_hashtable +
577 hash(path->mnt, path->dentry));
578 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
581 static inline void __mnt_make_longterm(struct mount *mnt)
584 atomic_inc(&mnt->mnt_longterm);
588 /* needs vfsmount lock for write */
589 static inline void __mnt_make_shortterm(struct mount *mnt)
592 atomic_dec(&mnt->mnt_longterm);
597 * vfsmount lock must be held for write
599 static void commit_tree(struct mount *mnt)
601 struct mount *parent = mnt->mnt_parent;
604 struct mnt_namespace *n = parent->mnt_ns;
606 BUG_ON(parent == mnt);
608 list_add_tail(&head, &mnt->mnt_list);
609 list_for_each_entry(m, &head, mnt_list) {
611 __mnt_make_longterm(m);
614 list_splice(&head, n->list.prev);
616 list_add_tail(&mnt->mnt_hash, mount_hashtable +
617 hash(&parent->mnt, mnt->mnt_mountpoint));
618 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
619 touch_mnt_namespace(n);
622 static struct mount *next_mnt(struct mount *p, struct mount *root)
624 struct list_head *next = p->mnt_mounts.next;
625 if (next == &p->mnt_mounts) {
629 next = p->mnt_child.next;
630 if (next != &p->mnt_parent->mnt_mounts)
635 return list_entry(next, struct mount, mnt_child);
638 static struct mount *skip_mnt_tree(struct mount *p)
640 struct list_head *prev = p->mnt_mounts.prev;
641 while (prev != &p->mnt_mounts) {
642 p = list_entry(prev, struct mount, mnt_child);
643 prev = p->mnt_mounts.prev;
649 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
655 return ERR_PTR(-ENODEV);
657 mnt = alloc_vfsmnt(name);
659 return ERR_PTR(-ENOMEM);
661 if (flags & MS_KERNMOUNT)
662 mnt->mnt.mnt_flags = MNT_INTERNAL;
664 root = mount_fs(type, flags, name, data);
667 return ERR_CAST(root);
670 mnt->mnt.mnt_root = root;
671 mnt->mnt.mnt_sb = root->d_sb;
672 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
673 mnt->mnt_parent = mnt;
674 br_write_lock(vfsmount_lock);
675 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
676 br_write_unlock(vfsmount_lock);
679 EXPORT_SYMBOL_GPL(vfs_kern_mount);
681 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
684 struct super_block *sb = old->mnt.mnt_sb;
685 struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
688 if (flag & (CL_SLAVE | CL_PRIVATE))
689 mnt->mnt_group_id = 0; /* not a peer of original */
691 mnt->mnt_group_id = old->mnt_group_id;
693 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
694 int err = mnt_alloc_group_id(mnt);
699 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
700 atomic_inc(&sb->s_active);
701 mnt->mnt.mnt_sb = sb;
702 mnt->mnt.mnt_root = dget(root);
703 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
704 mnt->mnt_parent = mnt;
705 br_write_lock(vfsmount_lock);
706 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
707 br_write_unlock(vfsmount_lock);
709 if (flag & CL_SLAVE) {
710 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
711 mnt->mnt_master = old;
712 CLEAR_MNT_SHARED(mnt);
713 } else if (!(flag & CL_PRIVATE)) {
714 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
715 list_add(&mnt->mnt_share, &old->mnt_share);
716 if (IS_MNT_SLAVE(old))
717 list_add(&mnt->mnt_slave, &old->mnt_slave);
718 mnt->mnt_master = old->mnt_master;
720 if (flag & CL_MAKE_SHARED)
723 /* stick the duplicate mount on the same expiry list
724 * as the original if that was on one */
725 if (flag & CL_EXPIRE) {
726 if (!list_empty(&old->mnt_expire))
727 list_add(&mnt->mnt_expire, &old->mnt_expire);
737 static inline void mntfree(struct mount *mnt)
739 struct vfsmount *m = &mnt->mnt;
740 struct super_block *sb = m->mnt_sb;
743 * This probably indicates that somebody messed
744 * up a mnt_want/drop_write() pair. If this
745 * happens, the filesystem was probably unable
746 * to make r/w->r/o transitions.
749 * The locking used to deal with mnt_count decrement provides barriers,
750 * so mnt_get_writers() below is safe.
752 WARN_ON(mnt_get_writers(mnt));
753 fsnotify_vfsmount_delete(m);
756 deactivate_super(sb);
759 static void mntput_no_expire(struct mount *mnt)
763 br_read_lock(vfsmount_lock);
764 if (likely(atomic_read(&mnt->mnt_longterm))) {
765 mnt_add_count(mnt, -1);
766 br_read_unlock(vfsmount_lock);
769 br_read_unlock(vfsmount_lock);
771 br_write_lock(vfsmount_lock);
772 mnt_add_count(mnt, -1);
773 if (mnt_get_count(mnt)) {
774 br_write_unlock(vfsmount_lock);
778 mnt_add_count(mnt, -1);
779 if (likely(mnt_get_count(mnt)))
781 br_write_lock(vfsmount_lock);
783 if (unlikely(mnt->mnt_pinned)) {
784 mnt_add_count(mnt, mnt->mnt_pinned + 1);
786 br_write_unlock(vfsmount_lock);
787 acct_auto_close_mnt(&mnt->mnt);
790 list_del(&mnt->mnt_instance);
791 br_write_unlock(vfsmount_lock);
795 void mntput(struct vfsmount *mnt)
798 struct mount *m = real_mount(mnt);
799 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
800 if (unlikely(m->mnt_expiry_mark))
801 m->mnt_expiry_mark = 0;
805 EXPORT_SYMBOL(mntput);
807 struct vfsmount *mntget(struct vfsmount *mnt)
810 mnt_add_count(real_mount(mnt), 1);
813 EXPORT_SYMBOL(mntget);
815 void mnt_pin(struct vfsmount *mnt)
817 br_write_lock(vfsmount_lock);
818 real_mount(mnt)->mnt_pinned++;
819 br_write_unlock(vfsmount_lock);
821 EXPORT_SYMBOL(mnt_pin);
823 void mnt_unpin(struct vfsmount *m)
825 struct mount *mnt = real_mount(m);
826 br_write_lock(vfsmount_lock);
827 if (mnt->mnt_pinned) {
828 mnt_add_count(mnt, 1);
831 br_write_unlock(vfsmount_lock);
833 EXPORT_SYMBOL(mnt_unpin);
835 static inline void mangle(struct seq_file *m, const char *s)
837 seq_escape(m, s, " \t\n\\");
841 * Simple .show_options callback for filesystems which don't want to
842 * implement more complex mount option showing.
844 * See also save_mount_options().
846 int generic_show_options(struct seq_file *m, struct dentry *root)
851 options = rcu_dereference(root->d_sb->s_options);
853 if (options != NULL && options[0]) {
861 EXPORT_SYMBOL(generic_show_options);
864 * If filesystem uses generic_show_options(), this function should be
865 * called from the fill_super() callback.
867 * The .remount_fs callback usually needs to be handled in a special
868 * way, to make sure, that previous options are not overwritten if the
871 * Also note, that if the filesystem's .remount_fs function doesn't
872 * reset all options to their default value, but changes only newly
873 * given options, then the displayed options will not reflect reality
876 void save_mount_options(struct super_block *sb, char *options)
878 BUG_ON(sb->s_options);
879 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
881 EXPORT_SYMBOL(save_mount_options);
883 void replace_mount_options(struct super_block *sb, char *options)
885 char *old = sb->s_options;
886 rcu_assign_pointer(sb->s_options, options);
892 EXPORT_SYMBOL(replace_mount_options);
894 #ifdef CONFIG_PROC_FS
895 /* iterator; we want it to have access to namespace_sem, thus here... */
896 static void *m_start(struct seq_file *m, loff_t *pos)
898 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
900 down_read(&namespace_sem);
901 return seq_list_start(&p->ns->list, *pos);
904 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
906 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
908 return seq_list_next(v, &p->ns->list, pos);
911 static void m_stop(struct seq_file *m, void *v)
913 up_read(&namespace_sem);
916 static int m_show(struct seq_file *m, void *v)
918 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
919 struct mount *r = list_entry(v, struct mount, mnt_list);
920 return p->show(m, &r->mnt);
923 const struct seq_operations mounts_op = {
929 #endif /* CONFIG_PROC_FS */
932 * may_umount_tree - check if a mount tree is busy
933 * @mnt: root of mount tree
935 * This is called to check if a tree of mounts has any
936 * open files, pwds, chroots or sub mounts that are
939 int may_umount_tree(struct vfsmount *m)
941 struct mount *mnt = real_mount(m);
943 int minimum_refs = 0;
947 /* write lock needed for mnt_get_count */
948 br_write_lock(vfsmount_lock);
949 for (p = mnt; p; p = next_mnt(p, mnt)) {
950 actual_refs += mnt_get_count(p);
953 br_write_unlock(vfsmount_lock);
955 if (actual_refs > minimum_refs)
961 EXPORT_SYMBOL(may_umount_tree);
964 * may_umount - check if a mount point is busy
965 * @mnt: root of mount
967 * This is called to check if a mount point has any
968 * open files, pwds, chroots or sub mounts. If the
969 * mount has sub mounts this will return busy
970 * regardless of whether the sub mounts are busy.
972 * Doesn't take quota and stuff into account. IOW, in some cases it will
973 * give false negatives. The main reason why it's here is that we need
974 * a non-destructive way to look for easily umountable filesystems.
976 int may_umount(struct vfsmount *mnt)
979 down_read(&namespace_sem);
980 br_write_lock(vfsmount_lock);
981 if (propagate_mount_busy(real_mount(mnt), 2))
983 br_write_unlock(vfsmount_lock);
984 up_read(&namespace_sem);
988 EXPORT_SYMBOL(may_umount);
990 void release_mounts(struct list_head *head)
993 while (!list_empty(head)) {
994 mnt = list_first_entry(head, struct mount, mnt_hash);
995 list_del_init(&mnt->mnt_hash);
996 if (mnt_has_parent(mnt)) {
997 struct dentry *dentry;
1000 br_write_lock(vfsmount_lock);
1001 dentry = mnt->mnt_mountpoint;
1002 m = mnt->mnt_parent;
1003 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1004 mnt->mnt_parent = mnt;
1006 br_write_unlock(vfsmount_lock);
1015 * vfsmount lock must be held for write
1016 * namespace_sem must be held for write
1018 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1020 LIST_HEAD(tmp_list);
1023 for (p = mnt; p; p = next_mnt(p, mnt))
1024 list_move(&p->mnt_hash, &tmp_list);
1027 propagate_umount(&tmp_list);
1029 list_for_each_entry(p, &tmp_list, mnt_hash) {
1030 list_del_init(&p->mnt_expire);
1031 list_del_init(&p->mnt_list);
1032 __touch_mnt_namespace(p->mnt_ns);
1034 __mnt_make_shortterm(p);
1035 list_del_init(&p->mnt_child);
1036 if (mnt_has_parent(p)) {
1037 p->mnt_parent->mnt_ghosts++;
1038 dentry_reset_mounted(p->mnt_mountpoint);
1040 change_mnt_propagation(p, MS_PRIVATE);
1042 list_splice(&tmp_list, kill);
1045 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1047 static int do_umount(struct mount *mnt, int flags)
1049 struct super_block *sb = mnt->mnt.mnt_sb;
1051 LIST_HEAD(umount_list);
1053 retval = security_sb_umount(&mnt->mnt, flags);
1058 * Allow userspace to request a mountpoint be expired rather than
1059 * unmounting unconditionally. Unmount only happens if:
1060 * (1) the mark is already set (the mark is cleared by mntput())
1061 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1063 if (flags & MNT_EXPIRE) {
1064 if (&mnt->mnt == current->fs->root.mnt ||
1065 flags & (MNT_FORCE | MNT_DETACH))
1069 * probably don't strictly need the lock here if we examined
1070 * all race cases, but it's a slowpath.
1072 br_write_lock(vfsmount_lock);
1073 if (mnt_get_count(mnt) != 2) {
1074 br_write_unlock(vfsmount_lock);
1077 br_write_unlock(vfsmount_lock);
1079 if (!xchg(&mnt->mnt_expiry_mark, 1))
1084 * If we may have to abort operations to get out of this
1085 * mount, and they will themselves hold resources we must
1086 * allow the fs to do things. In the Unix tradition of
1087 * 'Gee thats tricky lets do it in userspace' the umount_begin
1088 * might fail to complete on the first run through as other tasks
1089 * must return, and the like. Thats for the mount program to worry
1090 * about for the moment.
1093 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1094 sb->s_op->umount_begin(sb);
1098 * No sense to grab the lock for this test, but test itself looks
1099 * somewhat bogus. Suggestions for better replacement?
1100 * Ho-hum... In principle, we might treat that as umount + switch
1101 * to rootfs. GC would eventually take care of the old vfsmount.
1102 * Actually it makes sense, especially if rootfs would contain a
1103 * /reboot - static binary that would close all descriptors and
1104 * call reboot(9). Then init(8) could umount root and exec /reboot.
1106 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1108 * Special case for "unmounting" root ...
1109 * we just try to remount it readonly.
1111 down_write(&sb->s_umount);
1112 if (!(sb->s_flags & MS_RDONLY))
1113 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1114 up_write(&sb->s_umount);
1118 down_write(&namespace_sem);
1119 br_write_lock(vfsmount_lock);
1122 if (!(flags & MNT_DETACH))
1123 shrink_submounts(mnt, &umount_list);
1126 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1127 if (!list_empty(&mnt->mnt_list))
1128 umount_tree(mnt, 1, &umount_list);
1131 br_write_unlock(vfsmount_lock);
1132 up_write(&namespace_sem);
1133 release_mounts(&umount_list);
1138 * Now umount can handle mount points as well as block devices.
1139 * This is important for filesystems which use unnamed block devices.
1141 * We now support a flag for forced unmount like the other 'big iron'
1142 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1145 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1150 int lookup_flags = 0;
1152 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1155 if (!(flags & UMOUNT_NOFOLLOW))
1156 lookup_flags |= LOOKUP_FOLLOW;
1158 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1161 mnt = real_mount(path.mnt);
1163 if (path.dentry != path.mnt->mnt_root)
1165 if (!check_mnt(mnt))
1169 if (!capable(CAP_SYS_ADMIN))
1172 retval = do_umount(mnt, flags);
1174 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1176 mntput_no_expire(mnt);
1181 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1184 * The 2.0 compatible umount. No flags.
1186 SYSCALL_DEFINE1(oldumount, char __user *, name)
1188 return sys_umount(name, 0);
1193 static int mount_is_safe(struct path *path)
1195 if (capable(CAP_SYS_ADMIN))
1199 if (S_ISLNK(path->dentry->d_inode->i_mode))
1201 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1202 if (current_uid() != path->dentry->d_inode->i_uid)
1205 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1211 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1214 struct mount *res, *p, *q, *r;
1217 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1220 res = q = clone_mnt(mnt, dentry, flag);
1223 q->mnt_mountpoint = mnt->mnt_mountpoint;
1226 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1228 if (!is_subdir(r->mnt_mountpoint, dentry))
1231 for (s = r; s; s = next_mnt(s, r)) {
1232 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1233 s = skip_mnt_tree(s);
1236 while (p != s->mnt_parent) {
1242 path.dentry = p->mnt_mountpoint;
1243 q = clone_mnt(p, p->mnt.mnt_root, flag);
1246 br_write_lock(vfsmount_lock);
1247 list_add_tail(&q->mnt_list, &res->mnt_list);
1248 attach_mnt(q, &path);
1249 br_write_unlock(vfsmount_lock);
1255 LIST_HEAD(umount_list);
1256 br_write_lock(vfsmount_lock);
1257 umount_tree(res, 0, &umount_list);
1258 br_write_unlock(vfsmount_lock);
1259 release_mounts(&umount_list);
1264 struct vfsmount *collect_mounts(struct path *path)
1267 down_write(&namespace_sem);
1268 tree = copy_tree(real_mount(path->mnt), path->dentry,
1269 CL_COPY_ALL | CL_PRIVATE);
1270 up_write(&namespace_sem);
1271 return tree ? &tree->mnt : NULL;
1274 void drop_collected_mounts(struct vfsmount *mnt)
1276 LIST_HEAD(umount_list);
1277 down_write(&namespace_sem);
1278 br_write_lock(vfsmount_lock);
1279 umount_tree(real_mount(mnt), 0, &umount_list);
1280 br_write_unlock(vfsmount_lock);
1281 up_write(&namespace_sem);
1282 release_mounts(&umount_list);
1285 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1286 struct vfsmount *root)
1289 int res = f(root, arg);
1292 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1293 res = f(&mnt->mnt, arg);
1300 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1304 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1305 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1306 mnt_release_group_id(p);
1310 static int invent_group_ids(struct mount *mnt, bool recurse)
1314 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1315 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1316 int err = mnt_alloc_group_id(p);
1318 cleanup_group_ids(mnt, p);
1328 * @source_mnt : mount tree to be attached
1329 * @nd : place the mount tree @source_mnt is attached
1330 * @parent_nd : if non-null, detach the source_mnt from its parent and
1331 * store the parent mount and mountpoint dentry.
1332 * (done when source_mnt is moved)
1334 * NOTE: in the table below explains the semantics when a source mount
1335 * of a given type is attached to a destination mount of a given type.
1336 * ---------------------------------------------------------------------------
1337 * | BIND MOUNT OPERATION |
1338 * |**************************************************************************
1339 * | source-->| shared | private | slave | unbindable |
1343 * |**************************************************************************
1344 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1346 * |non-shared| shared (+) | private | slave (*) | invalid |
1347 * ***************************************************************************
1348 * A bind operation clones the source mount and mounts the clone on the
1349 * destination mount.
1351 * (++) the cloned mount is propagated to all the mounts in the propagation
1352 * tree of the destination mount and the cloned mount is added to
1353 * the peer group of the source mount.
1354 * (+) the cloned mount is created under the destination mount and is marked
1355 * as shared. The cloned mount is added to the peer group of the source
1357 * (+++) the mount is propagated to all the mounts in the propagation tree
1358 * of the destination mount and the cloned mount is made slave
1359 * of the same master as that of the source mount. The cloned mount
1360 * is marked as 'shared and slave'.
1361 * (*) the cloned mount is made a slave of the same master as that of the
1364 * ---------------------------------------------------------------------------
1365 * | MOVE MOUNT OPERATION |
1366 * |**************************************************************************
1367 * | source-->| shared | private | slave | unbindable |
1371 * |**************************************************************************
1372 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1374 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1375 * ***************************************************************************
1377 * (+) the mount is moved to the destination. And is then propagated to
1378 * all the mounts in the propagation tree of the destination mount.
1379 * (+*) the mount is moved to the destination.
1380 * (+++) the mount is moved to the destination and is then propagated to
1381 * all the mounts belonging to the destination mount's propagation tree.
1382 * the mount is marked as 'shared and slave'.
1383 * (*) the mount continues to be a slave at the new location.
1385 * if the source mount is a tree, the operations explained above is
1386 * applied to each mount in the tree.
1387 * Must be called without spinlocks held, since this function can sleep
1390 static int attach_recursive_mnt(struct mount *source_mnt,
1391 struct path *path, struct path *parent_path)
1393 LIST_HEAD(tree_list);
1394 struct mount *dest_mnt = real_mount(path->mnt);
1395 struct dentry *dest_dentry = path->dentry;
1396 struct mount *child, *p;
1399 if (IS_MNT_SHARED(dest_mnt)) {
1400 err = invent_group_ids(source_mnt, true);
1404 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1406 goto out_cleanup_ids;
1408 br_write_lock(vfsmount_lock);
1410 if (IS_MNT_SHARED(dest_mnt)) {
1411 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1415 detach_mnt(source_mnt, parent_path);
1416 attach_mnt(source_mnt, path);
1417 touch_mnt_namespace(source_mnt->mnt_ns);
1419 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1420 commit_tree(source_mnt);
1423 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1424 list_del_init(&child->mnt_hash);
1427 br_write_unlock(vfsmount_lock);
1432 if (IS_MNT_SHARED(dest_mnt))
1433 cleanup_group_ids(source_mnt, NULL);
1438 static int lock_mount(struct path *path)
1440 struct vfsmount *mnt;
1442 mutex_lock(&path->dentry->d_inode->i_mutex);
1443 if (unlikely(cant_mount(path->dentry))) {
1444 mutex_unlock(&path->dentry->d_inode->i_mutex);
1447 down_write(&namespace_sem);
1448 mnt = lookup_mnt(path);
1451 up_write(&namespace_sem);
1452 mutex_unlock(&path->dentry->d_inode->i_mutex);
1455 path->dentry = dget(mnt->mnt_root);
1459 static void unlock_mount(struct path *path)
1461 up_write(&namespace_sem);
1462 mutex_unlock(&path->dentry->d_inode->i_mutex);
1465 static int graft_tree(struct mount *mnt, struct path *path)
1467 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1470 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1471 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1474 if (d_unlinked(path->dentry))
1477 return attach_recursive_mnt(mnt, path, NULL);
1481 * Sanity check the flags to change_mnt_propagation.
1484 static int flags_to_propagation_type(int flags)
1486 int type = flags & ~(MS_REC | MS_SILENT);
1488 /* Fail if any non-propagation flags are set */
1489 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1491 /* Only one propagation flag should be set */
1492 if (!is_power_of_2(type))
1498 * recursively change the type of the mountpoint.
1500 static int do_change_type(struct path *path, int flag)
1503 struct mount *mnt = real_mount(path->mnt);
1504 int recurse = flag & MS_REC;
1508 if (!capable(CAP_SYS_ADMIN))
1511 if (path->dentry != path->mnt->mnt_root)
1514 type = flags_to_propagation_type(flag);
1518 down_write(&namespace_sem);
1519 if (type == MS_SHARED) {
1520 err = invent_group_ids(mnt, recurse);
1525 br_write_lock(vfsmount_lock);
1526 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1527 change_mnt_propagation(m, type);
1528 br_write_unlock(vfsmount_lock);
1531 up_write(&namespace_sem);
1536 * do loopback mount.
1538 static int do_loopback(struct path *path, char *old_name,
1541 LIST_HEAD(umount_list);
1542 struct path old_path;
1543 struct mount *mnt = NULL, *old;
1544 int err = mount_is_safe(path);
1547 if (!old_name || !*old_name)
1549 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1553 err = lock_mount(path);
1557 old = real_mount(old_path.mnt);
1560 if (IS_MNT_UNBINDABLE(old))
1563 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1568 mnt = copy_tree(old, old_path.dentry, 0);
1570 mnt = clone_mnt(old, old_path.dentry, 0);
1575 err = graft_tree(mnt, path);
1577 br_write_lock(vfsmount_lock);
1578 umount_tree(mnt, 0, &umount_list);
1579 br_write_unlock(vfsmount_lock);
1583 release_mounts(&umount_list);
1585 path_put(&old_path);
1589 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1592 int readonly_request = 0;
1594 if (ms_flags & MS_RDONLY)
1595 readonly_request = 1;
1596 if (readonly_request == __mnt_is_readonly(mnt))
1599 if (readonly_request)
1600 error = mnt_make_readonly(real_mount(mnt));
1602 __mnt_unmake_readonly(real_mount(mnt));
1607 * change filesystem flags. dir should be a physical root of filesystem.
1608 * If you've mounted a non-root directory somewhere and want to do remount
1609 * on it - tough luck.
1611 static int do_remount(struct path *path, int flags, int mnt_flags,
1615 struct super_block *sb = path->mnt->mnt_sb;
1616 struct mount *mnt = real_mount(path->mnt);
1618 if (!capable(CAP_SYS_ADMIN))
1621 if (!check_mnt(mnt))
1624 if (path->dentry != path->mnt->mnt_root)
1627 err = security_sb_remount(sb, data);
1631 down_write(&sb->s_umount);
1632 if (flags & MS_BIND)
1633 err = change_mount_flags(path->mnt, flags);
1635 err = do_remount_sb(sb, flags, data, 0);
1637 br_write_lock(vfsmount_lock);
1638 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1639 mnt->mnt.mnt_flags = mnt_flags;
1640 br_write_unlock(vfsmount_lock);
1642 up_write(&sb->s_umount);
1644 br_write_lock(vfsmount_lock);
1645 touch_mnt_namespace(mnt->mnt_ns);
1646 br_write_unlock(vfsmount_lock);
1651 static inline int tree_contains_unbindable(struct mount *mnt)
1654 for (p = mnt; p; p = next_mnt(p, mnt)) {
1655 if (IS_MNT_UNBINDABLE(p))
1661 static int do_move_mount(struct path *path, char *old_name)
1663 struct path old_path, parent_path;
1667 if (!capable(CAP_SYS_ADMIN))
1669 if (!old_name || !*old_name)
1671 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1675 err = lock_mount(path);
1679 old = real_mount(old_path.mnt);
1680 p = real_mount(path->mnt);
1683 if (!check_mnt(p) || !check_mnt(old))
1686 if (d_unlinked(path->dentry))
1690 if (old_path.dentry != old_path.mnt->mnt_root)
1693 if (!mnt_has_parent(old))
1696 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1697 S_ISDIR(old_path.dentry->d_inode->i_mode))
1700 * Don't move a mount residing in a shared parent.
1702 if (IS_MNT_SHARED(old->mnt_parent))
1705 * Don't move a mount tree containing unbindable mounts to a destination
1706 * mount which is shared.
1708 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1711 for (; mnt_has_parent(p); p = p->mnt_parent)
1715 err = attach_recursive_mnt(old, path, &parent_path);
1719 /* if the mount is moved, it should no longer be expire
1721 list_del_init(&old->mnt_expire);
1726 path_put(&parent_path);
1727 path_put(&old_path);
1731 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1734 const char *subtype = strchr(fstype, '.');
1743 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1745 if (!mnt->mnt_sb->s_subtype)
1751 return ERR_PTR(err);
1754 static struct vfsmount *
1755 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1757 struct file_system_type *type = get_fs_type(fstype);
1758 struct vfsmount *mnt;
1760 return ERR_PTR(-ENODEV);
1761 mnt = vfs_kern_mount(type, flags, name, data);
1762 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1763 !mnt->mnt_sb->s_subtype)
1764 mnt = fs_set_subtype(mnt, fstype);
1765 put_filesystem(type);
1770 * add a mount into a namespace's mount tree
1772 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1776 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1778 err = lock_mount(path);
1783 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1786 /* Refuse the same filesystem on the same mount point */
1788 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1789 path->mnt->mnt_root == path->dentry)
1793 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1796 newmnt->mnt.mnt_flags = mnt_flags;
1797 err = graft_tree(newmnt, path);
1805 * create a new mount for userspace and request it to be added into the
1808 static int do_new_mount(struct path *path, char *type, int flags,
1809 int mnt_flags, char *name, void *data)
1811 struct vfsmount *mnt;
1817 /* we need capabilities... */
1818 if (!capable(CAP_SYS_ADMIN))
1821 mnt = do_kern_mount(type, flags, name, data);
1823 return PTR_ERR(mnt);
1825 err = do_add_mount(real_mount(mnt), path, mnt_flags);
1831 int finish_automount(struct vfsmount *m, struct path *path)
1833 struct mount *mnt = real_mount(m);
1835 /* The new mount record should have at least 2 refs to prevent it being
1836 * expired before we get a chance to add it
1838 BUG_ON(mnt_get_count(mnt) < 2);
1840 if (m->mnt_sb == path->mnt->mnt_sb &&
1841 m->mnt_root == path->dentry) {
1846 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1850 /* remove m from any expiration list it may be on */
1851 if (!list_empty(&mnt->mnt_expire)) {
1852 down_write(&namespace_sem);
1853 br_write_lock(vfsmount_lock);
1854 list_del_init(&mnt->mnt_expire);
1855 br_write_unlock(vfsmount_lock);
1856 up_write(&namespace_sem);
1864 * mnt_set_expiry - Put a mount on an expiration list
1865 * @mnt: The mount to list.
1866 * @expiry_list: The list to add the mount to.
1868 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1870 down_write(&namespace_sem);
1871 br_write_lock(vfsmount_lock);
1873 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1875 br_write_unlock(vfsmount_lock);
1876 up_write(&namespace_sem);
1878 EXPORT_SYMBOL(mnt_set_expiry);
1881 * process a list of expirable mountpoints with the intent of discarding any
1882 * mountpoints that aren't in use and haven't been touched since last we came
1885 void mark_mounts_for_expiry(struct list_head *mounts)
1887 struct mount *mnt, *next;
1888 LIST_HEAD(graveyard);
1891 if (list_empty(mounts))
1894 down_write(&namespace_sem);
1895 br_write_lock(vfsmount_lock);
1897 /* extract from the expiration list every vfsmount that matches the
1898 * following criteria:
1899 * - only referenced by its parent vfsmount
1900 * - still marked for expiry (marked on the last call here; marks are
1901 * cleared by mntput())
1903 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1904 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1905 propagate_mount_busy(mnt, 1))
1907 list_move(&mnt->mnt_expire, &graveyard);
1909 while (!list_empty(&graveyard)) {
1910 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
1911 touch_mnt_namespace(mnt->mnt_ns);
1912 umount_tree(mnt, 1, &umounts);
1914 br_write_unlock(vfsmount_lock);
1915 up_write(&namespace_sem);
1917 release_mounts(&umounts);
1920 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1923 * Ripoff of 'select_parent()'
1925 * search the list of submounts for a given mountpoint, and move any
1926 * shrinkable submounts to the 'graveyard' list.
1928 static int select_submounts(struct mount *parent, struct list_head *graveyard)
1930 struct mount *this_parent = parent;
1931 struct list_head *next;
1935 next = this_parent->mnt_mounts.next;
1937 while (next != &this_parent->mnt_mounts) {
1938 struct list_head *tmp = next;
1939 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
1942 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
1945 * Descend a level if the d_mounts list is non-empty.
1947 if (!list_empty(&mnt->mnt_mounts)) {
1952 if (!propagate_mount_busy(mnt, 1)) {
1953 list_move_tail(&mnt->mnt_expire, graveyard);
1958 * All done at this level ... ascend and resume the search
1960 if (this_parent != parent) {
1961 next = this_parent->mnt_child.next;
1962 this_parent = this_parent->mnt_parent;
1969 * process a list of expirable mountpoints with the intent of discarding any
1970 * submounts of a specific parent mountpoint
1972 * vfsmount_lock must be held for write
1974 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
1976 LIST_HEAD(graveyard);
1979 /* extract submounts of 'mountpoint' from the expiration list */
1980 while (select_submounts(mnt, &graveyard)) {
1981 while (!list_empty(&graveyard)) {
1982 m = list_first_entry(&graveyard, struct mount,
1984 touch_mnt_namespace(m->mnt_ns);
1985 umount_tree(m, 1, umounts);
1991 * Some copy_from_user() implementations do not return the exact number of
1992 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1993 * Note that this function differs from copy_from_user() in that it will oops
1994 * on bad values of `to', rather than returning a short copy.
1996 static long exact_copy_from_user(void *to, const void __user * from,
2000 const char __user *f = from;
2003 if (!access_ok(VERIFY_READ, from, n))
2007 if (__get_user(c, f)) {
2018 int copy_mount_options(const void __user * data, unsigned long *where)
2028 if (!(page = __get_free_page(GFP_KERNEL)))
2031 /* We only care that *some* data at the address the user
2032 * gave us is valid. Just in case, we'll zero
2033 * the remainder of the page.
2035 /* copy_from_user cannot cross TASK_SIZE ! */
2036 size = TASK_SIZE - (unsigned long)data;
2037 if (size > PAGE_SIZE)
2040 i = size - exact_copy_from_user((void *)page, data, size);
2046 memset((char *)page + i, 0, PAGE_SIZE - i);
2051 int copy_mount_string(const void __user *data, char **where)
2060 tmp = strndup_user(data, PAGE_SIZE);
2062 return PTR_ERR(tmp);
2069 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2070 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2072 * data is a (void *) that can point to any structure up to
2073 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2074 * information (or be NULL).
2076 * Pre-0.97 versions of mount() didn't have a flags word.
2077 * When the flags word was introduced its top half was required
2078 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2079 * Therefore, if this magic number is present, it carries no information
2080 * and must be discarded.
2082 long do_mount(char *dev_name, char *dir_name, char *type_page,
2083 unsigned long flags, void *data_page)
2090 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2091 flags &= ~MS_MGC_MSK;
2093 /* Basic sanity checks */
2095 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2099 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2101 /* ... and get the mountpoint */
2102 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2106 retval = security_sb_mount(dev_name, &path,
2107 type_page, flags, data_page);
2111 /* Default to relatime unless overriden */
2112 if (!(flags & MS_NOATIME))
2113 mnt_flags |= MNT_RELATIME;
2115 /* Separate the per-mountpoint flags */
2116 if (flags & MS_NOSUID)
2117 mnt_flags |= MNT_NOSUID;
2118 if (flags & MS_NODEV)
2119 mnt_flags |= MNT_NODEV;
2120 if (flags & MS_NOEXEC)
2121 mnt_flags |= MNT_NOEXEC;
2122 if (flags & MS_NOATIME)
2123 mnt_flags |= MNT_NOATIME;
2124 if (flags & MS_NODIRATIME)
2125 mnt_flags |= MNT_NODIRATIME;
2126 if (flags & MS_STRICTATIME)
2127 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2128 if (flags & MS_RDONLY)
2129 mnt_flags |= MNT_READONLY;
2131 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2132 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2135 if (flags & MS_REMOUNT)
2136 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2138 else if (flags & MS_BIND)
2139 retval = do_loopback(&path, dev_name, flags & MS_REC);
2140 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2141 retval = do_change_type(&path, flags);
2142 else if (flags & MS_MOVE)
2143 retval = do_move_mount(&path, dev_name);
2145 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2146 dev_name, data_page);
2152 static struct mnt_namespace *alloc_mnt_ns(void)
2154 struct mnt_namespace *new_ns;
2156 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2158 return ERR_PTR(-ENOMEM);
2159 atomic_set(&new_ns->count, 1);
2160 new_ns->root = NULL;
2161 INIT_LIST_HEAD(&new_ns->list);
2162 init_waitqueue_head(&new_ns->poll);
2167 void mnt_make_longterm(struct vfsmount *mnt)
2169 __mnt_make_longterm(real_mount(mnt));
2172 void mnt_make_shortterm(struct vfsmount *m)
2175 struct mount *mnt = real_mount(m);
2176 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2178 br_write_lock(vfsmount_lock);
2179 atomic_dec(&mnt->mnt_longterm);
2180 br_write_unlock(vfsmount_lock);
2185 * Allocate a new namespace structure and populate it with contents
2186 * copied from the namespace of the passed in task structure.
2188 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2189 struct fs_struct *fs)
2191 struct mnt_namespace *new_ns;
2192 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2193 struct mount *p, *q;
2194 struct mount *old = mnt_ns->root;
2197 new_ns = alloc_mnt_ns();
2201 down_write(&namespace_sem);
2202 /* First pass: copy the tree topology */
2203 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2205 up_write(&namespace_sem);
2207 return ERR_PTR(-ENOMEM);
2210 br_write_lock(vfsmount_lock);
2211 list_add_tail(&new_ns->list, &new->mnt_list);
2212 br_write_unlock(vfsmount_lock);
2215 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2216 * as belonging to new namespace. We have already acquired a private
2217 * fs_struct, so tsk->fs->lock is not needed.
2223 __mnt_make_longterm(q);
2225 if (&p->mnt == fs->root.mnt) {
2226 fs->root.mnt = mntget(&q->mnt);
2227 __mnt_make_longterm(q);
2228 mnt_make_shortterm(&p->mnt);
2231 if (&p->mnt == fs->pwd.mnt) {
2232 fs->pwd.mnt = mntget(&q->mnt);
2233 __mnt_make_longterm(q);
2234 mnt_make_shortterm(&p->mnt);
2238 p = next_mnt(p, old);
2239 q = next_mnt(q, new);
2241 up_write(&namespace_sem);
2251 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2252 struct fs_struct *new_fs)
2254 struct mnt_namespace *new_ns;
2259 if (!(flags & CLONE_NEWNS))
2262 new_ns = dup_mnt_ns(ns, new_fs);
2269 * create_mnt_ns - creates a private namespace and adds a root filesystem
2270 * @mnt: pointer to the new root filesystem mountpoint
2272 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2274 struct mnt_namespace *new_ns = alloc_mnt_ns();
2275 if (!IS_ERR(new_ns)) {
2276 struct mount *mnt = real_mount(m);
2277 mnt->mnt_ns = new_ns;
2278 __mnt_make_longterm(mnt);
2280 list_add(&new_ns->list, &mnt->mnt_list);
2287 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2289 struct mnt_namespace *ns;
2290 struct super_block *s;
2294 ns = create_mnt_ns(mnt);
2296 return ERR_CAST(ns);
2298 err = vfs_path_lookup(mnt->mnt_root, mnt,
2299 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2304 return ERR_PTR(err);
2306 /* trade a vfsmount reference for active sb one */
2307 s = path.mnt->mnt_sb;
2308 atomic_inc(&s->s_active);
2310 /* lock the sucker */
2311 down_write(&s->s_umount);
2312 /* ... and return the root of (sub)tree on it */
2315 EXPORT_SYMBOL(mount_subtree);
2317 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2318 char __user *, type, unsigned long, flags, void __user *, data)
2324 unsigned long data_page;
2326 ret = copy_mount_string(type, &kernel_type);
2330 kernel_dir = getname(dir_name);
2331 if (IS_ERR(kernel_dir)) {
2332 ret = PTR_ERR(kernel_dir);
2336 ret = copy_mount_string(dev_name, &kernel_dev);
2340 ret = copy_mount_options(data, &data_page);
2344 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2345 (void *) data_page);
2347 free_page(data_page);
2351 putname(kernel_dir);
2359 * Return true if path is reachable from root
2361 * namespace_sem or vfsmount_lock is held
2363 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2364 const struct path *root)
2366 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2367 dentry = mnt->mnt_mountpoint;
2368 mnt = mnt->mnt_parent;
2370 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2373 int path_is_under(struct path *path1, struct path *path2)
2376 br_read_lock(vfsmount_lock);
2377 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2378 br_read_unlock(vfsmount_lock);
2381 EXPORT_SYMBOL(path_is_under);
2384 * pivot_root Semantics:
2385 * Moves the root file system of the current process to the directory put_old,
2386 * makes new_root as the new root file system of the current process, and sets
2387 * root/cwd of all processes which had them on the current root to new_root.
2390 * The new_root and put_old must be directories, and must not be on the
2391 * same file system as the current process root. The put_old must be
2392 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2393 * pointed to by put_old must yield the same directory as new_root. No other
2394 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2396 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2397 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2398 * in this situation.
2401 * - we don't move root/cwd if they are not at the root (reason: if something
2402 * cared enough to change them, it's probably wrong to force them elsewhere)
2403 * - it's okay to pick a root that isn't the root of a file system, e.g.
2404 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2405 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2408 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2409 const char __user *, put_old)
2411 struct path new, old, parent_path, root_parent, root;
2412 struct mount *new_mnt, *root_mnt;
2415 if (!capable(CAP_SYS_ADMIN))
2418 error = user_path_dir(new_root, &new);
2422 error = user_path_dir(put_old, &old);
2426 error = security_sb_pivotroot(&old, &new);
2430 get_fs_root(current->fs, &root);
2431 error = lock_mount(&old);
2436 new_mnt = real_mount(new.mnt);
2437 root_mnt = real_mount(root.mnt);
2438 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2439 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2440 IS_MNT_SHARED(root_mnt->mnt_parent))
2442 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2445 if (d_unlinked(new.dentry))
2447 if (d_unlinked(old.dentry))
2450 if (new.mnt == root.mnt ||
2451 old.mnt == root.mnt)
2452 goto out4; /* loop, on the same file system */
2454 if (root.mnt->mnt_root != root.dentry)
2455 goto out4; /* not a mountpoint */
2456 if (!mnt_has_parent(root_mnt))
2457 goto out4; /* not attached */
2458 if (new.mnt->mnt_root != new.dentry)
2459 goto out4; /* not a mountpoint */
2460 if (!mnt_has_parent(new_mnt))
2461 goto out4; /* not attached */
2462 /* make sure we can reach put_old from new_root */
2463 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2465 br_write_lock(vfsmount_lock);
2466 detach_mnt(new_mnt, &parent_path);
2467 detach_mnt(root_mnt, &root_parent);
2468 /* mount old root on put_old */
2469 attach_mnt(root_mnt, &old);
2470 /* mount new_root on / */
2471 attach_mnt(new_mnt, &root_parent);
2472 touch_mnt_namespace(current->nsproxy->mnt_ns);
2473 br_write_unlock(vfsmount_lock);
2474 chroot_fs_refs(&root, &new);
2479 path_put(&root_parent);
2480 path_put(&parent_path);
2492 static void __init init_mount_tree(void)
2494 struct vfsmount *mnt;
2495 struct mnt_namespace *ns;
2498 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2500 panic("Can't create rootfs");
2502 ns = create_mnt_ns(mnt);
2504 panic("Can't allocate initial namespace");
2506 init_task.nsproxy->mnt_ns = ns;
2510 root.dentry = mnt->mnt_root;
2512 set_fs_pwd(current->fs, &root);
2513 set_fs_root(current->fs, &root);
2516 void __init mnt_init(void)
2521 init_rwsem(&namespace_sem);
2523 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2524 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2526 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2528 if (!mount_hashtable)
2529 panic("Failed to allocate mount hash table\n");
2531 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2533 for (u = 0; u < HASH_SIZE; u++)
2534 INIT_LIST_HEAD(&mount_hashtable[u]);
2536 br_lock_init(vfsmount_lock);
2540 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2542 fs_kobj = kobject_create_and_add("fs", NULL);
2544 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2549 void put_mnt_ns(struct mnt_namespace *ns)
2551 LIST_HEAD(umount_list);
2553 if (!atomic_dec_and_test(&ns->count))
2555 down_write(&namespace_sem);
2556 br_write_lock(vfsmount_lock);
2557 umount_tree(ns->root, 0, &umount_list);
2558 br_write_unlock(vfsmount_lock);
2559 up_write(&namespace_sem);
2560 release_mounts(&umount_list);
2564 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2566 struct vfsmount *mnt;
2567 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2570 * it is a longterm mount, don't release mnt until
2571 * we unmount before file sys is unregistered
2573 mnt_make_longterm(mnt);
2577 EXPORT_SYMBOL_GPL(kern_mount_data);
2579 void kern_unmount(struct vfsmount *mnt)
2581 /* release long term mount so mount point can be released */
2582 if (!IS_ERR_OR_NULL(mnt)) {
2583 mnt_make_shortterm(mnt);
2587 EXPORT_SYMBOL(kern_unmount);
2589 bool our_mnt(struct vfsmount *mnt)
2591 return check_mnt(real_mount(mnt));