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;
676 EXPORT_SYMBOL_GPL(vfs_kern_mount);
678 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
681 struct super_block *sb = old->mnt.mnt_sb;
682 struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
685 if (flag & (CL_SLAVE | CL_PRIVATE))
686 mnt->mnt_group_id = 0; /* not a peer of original */
688 mnt->mnt_group_id = old->mnt_group_id;
690 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
691 int err = mnt_alloc_group_id(mnt);
696 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
697 atomic_inc(&sb->s_active);
698 mnt->mnt.mnt_sb = sb;
699 mnt->mnt.mnt_root = dget(root);
700 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
701 mnt->mnt_parent = mnt;
703 if (flag & CL_SLAVE) {
704 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
705 mnt->mnt_master = old;
706 CLEAR_MNT_SHARED(mnt);
707 } else if (!(flag & CL_PRIVATE)) {
708 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
709 list_add(&mnt->mnt_share, &old->mnt_share);
710 if (IS_MNT_SLAVE(old))
711 list_add(&mnt->mnt_slave, &old->mnt_slave);
712 mnt->mnt_master = old->mnt_master;
714 if (flag & CL_MAKE_SHARED)
717 /* stick the duplicate mount on the same expiry list
718 * as the original if that was on one */
719 if (flag & CL_EXPIRE) {
720 if (!list_empty(&old->mnt_expire))
721 list_add(&mnt->mnt_expire, &old->mnt_expire);
731 static inline void mntfree(struct mount *mnt)
733 struct vfsmount *m = &mnt->mnt;
734 struct super_block *sb = m->mnt_sb;
737 * This probably indicates that somebody messed
738 * up a mnt_want/drop_write() pair. If this
739 * happens, the filesystem was probably unable
740 * to make r/w->r/o transitions.
743 * The locking used to deal with mnt_count decrement provides barriers,
744 * so mnt_get_writers() below is safe.
746 WARN_ON(mnt_get_writers(mnt));
747 fsnotify_vfsmount_delete(m);
750 deactivate_super(sb);
753 static void mntput_no_expire(struct mount *mnt)
757 br_read_lock(vfsmount_lock);
758 if (likely(atomic_read(&mnt->mnt_longterm))) {
759 mnt_add_count(mnt, -1);
760 br_read_unlock(vfsmount_lock);
763 br_read_unlock(vfsmount_lock);
765 br_write_lock(vfsmount_lock);
766 mnt_add_count(mnt, -1);
767 if (mnt_get_count(mnt)) {
768 br_write_unlock(vfsmount_lock);
772 mnt_add_count(mnt, -1);
773 if (likely(mnt_get_count(mnt)))
775 br_write_lock(vfsmount_lock);
777 if (unlikely(mnt->mnt_pinned)) {
778 mnt_add_count(mnt, mnt->mnt_pinned + 1);
780 br_write_unlock(vfsmount_lock);
781 acct_auto_close_mnt(&mnt->mnt);
784 br_write_unlock(vfsmount_lock);
788 void mntput(struct vfsmount *mnt)
791 struct mount *m = real_mount(mnt);
792 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
793 if (unlikely(m->mnt_expiry_mark))
794 m->mnt_expiry_mark = 0;
798 EXPORT_SYMBOL(mntput);
800 struct vfsmount *mntget(struct vfsmount *mnt)
803 mnt_add_count(real_mount(mnt), 1);
806 EXPORT_SYMBOL(mntget);
808 void mnt_pin(struct vfsmount *mnt)
810 br_write_lock(vfsmount_lock);
811 real_mount(mnt)->mnt_pinned++;
812 br_write_unlock(vfsmount_lock);
814 EXPORT_SYMBOL(mnt_pin);
816 void mnt_unpin(struct vfsmount *m)
818 struct mount *mnt = real_mount(m);
819 br_write_lock(vfsmount_lock);
820 if (mnt->mnt_pinned) {
821 mnt_add_count(mnt, 1);
824 br_write_unlock(vfsmount_lock);
826 EXPORT_SYMBOL(mnt_unpin);
828 static inline void mangle(struct seq_file *m, const char *s)
830 seq_escape(m, s, " \t\n\\");
834 * Simple .show_options callback for filesystems which don't want to
835 * implement more complex mount option showing.
837 * See also save_mount_options().
839 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
844 options = rcu_dereference(mnt->mnt_sb->s_options);
846 if (options != NULL && options[0]) {
854 EXPORT_SYMBOL(generic_show_options);
857 * If filesystem uses generic_show_options(), this function should be
858 * called from the fill_super() callback.
860 * The .remount_fs callback usually needs to be handled in a special
861 * way, to make sure, that previous options are not overwritten if the
864 * Also note, that if the filesystem's .remount_fs function doesn't
865 * reset all options to their default value, but changes only newly
866 * given options, then the displayed options will not reflect reality
869 void save_mount_options(struct super_block *sb, char *options)
871 BUG_ON(sb->s_options);
872 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
874 EXPORT_SYMBOL(save_mount_options);
876 void replace_mount_options(struct super_block *sb, char *options)
878 char *old = sb->s_options;
879 rcu_assign_pointer(sb->s_options, options);
885 EXPORT_SYMBOL(replace_mount_options);
887 #ifdef CONFIG_PROC_FS
888 /* iterator; we want it to have access to namespace_sem, thus here... */
889 static void *m_start(struct seq_file *m, loff_t *pos)
891 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
893 down_read(&namespace_sem);
894 return seq_list_start(&p->ns->list, *pos);
897 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
899 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
901 return seq_list_next(v, &p->ns->list, pos);
904 static void m_stop(struct seq_file *m, void *v)
906 up_read(&namespace_sem);
909 static int m_show(struct seq_file *m, void *v)
911 struct proc_mounts *p = container_of(m, struct proc_mounts, m);
912 struct mount *r = list_entry(v, struct mount, mnt_list);
913 return p->show(m, &r->mnt);
916 const struct seq_operations mounts_op = {
922 #endif /* CONFIG_PROC_FS */
925 * may_umount_tree - check if a mount tree is busy
926 * @mnt: root of mount tree
928 * This is called to check if a tree of mounts has any
929 * open files, pwds, chroots or sub mounts that are
932 int may_umount_tree(struct vfsmount *m)
934 struct mount *mnt = real_mount(m);
936 int minimum_refs = 0;
940 /* write lock needed for mnt_get_count */
941 br_write_lock(vfsmount_lock);
942 for (p = mnt; p; p = next_mnt(p, mnt)) {
943 actual_refs += mnt_get_count(p);
946 br_write_unlock(vfsmount_lock);
948 if (actual_refs > minimum_refs)
954 EXPORT_SYMBOL(may_umount_tree);
957 * may_umount - check if a mount point is busy
958 * @mnt: root of mount
960 * This is called to check if a mount point has any
961 * open files, pwds, chroots or sub mounts. If the
962 * mount has sub mounts this will return busy
963 * regardless of whether the sub mounts are busy.
965 * Doesn't take quota and stuff into account. IOW, in some cases it will
966 * give false negatives. The main reason why it's here is that we need
967 * a non-destructive way to look for easily umountable filesystems.
969 int may_umount(struct vfsmount *mnt)
972 down_read(&namespace_sem);
973 br_write_lock(vfsmount_lock);
974 if (propagate_mount_busy(real_mount(mnt), 2))
976 br_write_unlock(vfsmount_lock);
977 up_read(&namespace_sem);
981 EXPORT_SYMBOL(may_umount);
983 void release_mounts(struct list_head *head)
986 while (!list_empty(head)) {
987 mnt = list_first_entry(head, struct mount, mnt_hash);
988 list_del_init(&mnt->mnt_hash);
989 if (mnt_has_parent(mnt)) {
990 struct dentry *dentry;
993 br_write_lock(vfsmount_lock);
994 dentry = mnt->mnt_mountpoint;
996 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
997 mnt->mnt_parent = mnt;
999 br_write_unlock(vfsmount_lock);
1008 * vfsmount lock must be held for write
1009 * namespace_sem must be held for write
1011 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1013 LIST_HEAD(tmp_list);
1016 for (p = mnt; p; p = next_mnt(p, mnt))
1017 list_move(&p->mnt_hash, &tmp_list);
1020 propagate_umount(&tmp_list);
1022 list_for_each_entry(p, &tmp_list, mnt_hash) {
1023 list_del_init(&p->mnt_expire);
1024 list_del_init(&p->mnt_list);
1025 __touch_mnt_namespace(p->mnt_ns);
1027 __mnt_make_shortterm(p);
1028 list_del_init(&p->mnt_child);
1029 if (mnt_has_parent(p)) {
1030 p->mnt_parent->mnt_ghosts++;
1031 dentry_reset_mounted(p->mnt_mountpoint);
1033 change_mnt_propagation(p, MS_PRIVATE);
1035 list_splice(&tmp_list, kill);
1038 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1040 static int do_umount(struct mount *mnt, int flags)
1042 struct super_block *sb = mnt->mnt.mnt_sb;
1044 LIST_HEAD(umount_list);
1046 retval = security_sb_umount(&mnt->mnt, flags);
1051 * Allow userspace to request a mountpoint be expired rather than
1052 * unmounting unconditionally. Unmount only happens if:
1053 * (1) the mark is already set (the mark is cleared by mntput())
1054 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1056 if (flags & MNT_EXPIRE) {
1057 if (&mnt->mnt == current->fs->root.mnt ||
1058 flags & (MNT_FORCE | MNT_DETACH))
1062 * probably don't strictly need the lock here if we examined
1063 * all race cases, but it's a slowpath.
1065 br_write_lock(vfsmount_lock);
1066 if (mnt_get_count(mnt) != 2) {
1067 br_write_unlock(vfsmount_lock);
1070 br_write_unlock(vfsmount_lock);
1072 if (!xchg(&mnt->mnt_expiry_mark, 1))
1077 * If we may have to abort operations to get out of this
1078 * mount, and they will themselves hold resources we must
1079 * allow the fs to do things. In the Unix tradition of
1080 * 'Gee thats tricky lets do it in userspace' the umount_begin
1081 * might fail to complete on the first run through as other tasks
1082 * must return, and the like. Thats for the mount program to worry
1083 * about for the moment.
1086 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1087 sb->s_op->umount_begin(sb);
1091 * No sense to grab the lock for this test, but test itself looks
1092 * somewhat bogus. Suggestions for better replacement?
1093 * Ho-hum... In principle, we might treat that as umount + switch
1094 * to rootfs. GC would eventually take care of the old vfsmount.
1095 * Actually it makes sense, especially if rootfs would contain a
1096 * /reboot - static binary that would close all descriptors and
1097 * call reboot(9). Then init(8) could umount root and exec /reboot.
1099 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1101 * Special case for "unmounting" root ...
1102 * we just try to remount it readonly.
1104 down_write(&sb->s_umount);
1105 if (!(sb->s_flags & MS_RDONLY))
1106 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1107 up_write(&sb->s_umount);
1111 down_write(&namespace_sem);
1112 br_write_lock(vfsmount_lock);
1115 if (!(flags & MNT_DETACH))
1116 shrink_submounts(mnt, &umount_list);
1119 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1120 if (!list_empty(&mnt->mnt_list))
1121 umount_tree(mnt, 1, &umount_list);
1124 br_write_unlock(vfsmount_lock);
1125 up_write(&namespace_sem);
1126 release_mounts(&umount_list);
1131 * Now umount can handle mount points as well as block devices.
1132 * This is important for filesystems which use unnamed block devices.
1134 * We now support a flag for forced unmount like the other 'big iron'
1135 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1138 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1143 int lookup_flags = 0;
1145 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1148 if (!(flags & UMOUNT_NOFOLLOW))
1149 lookup_flags |= LOOKUP_FOLLOW;
1151 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1154 mnt = real_mount(path.mnt);
1156 if (path.dentry != path.mnt->mnt_root)
1158 if (!check_mnt(mnt))
1162 if (!capable(CAP_SYS_ADMIN))
1165 retval = do_umount(mnt, flags);
1167 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1169 mntput_no_expire(mnt);
1174 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1177 * The 2.0 compatible umount. No flags.
1179 SYSCALL_DEFINE1(oldumount, char __user *, name)
1181 return sys_umount(name, 0);
1186 static int mount_is_safe(struct path *path)
1188 if (capable(CAP_SYS_ADMIN))
1192 if (S_ISLNK(path->dentry->d_inode->i_mode))
1194 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1195 if (current_uid() != path->dentry->d_inode->i_uid)
1198 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1204 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1207 struct mount *res, *p, *q, *r;
1210 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1213 res = q = clone_mnt(mnt, dentry, flag);
1216 q->mnt_mountpoint = mnt->mnt_mountpoint;
1219 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1221 if (!is_subdir(r->mnt_mountpoint, dentry))
1224 for (s = r; s; s = next_mnt(s, r)) {
1225 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1226 s = skip_mnt_tree(s);
1229 while (p != s->mnt_parent) {
1235 path.dentry = p->mnt_mountpoint;
1236 q = clone_mnt(p, p->mnt.mnt_root, flag);
1239 br_write_lock(vfsmount_lock);
1240 list_add_tail(&q->mnt_list, &res->mnt_list);
1241 attach_mnt(q, &path);
1242 br_write_unlock(vfsmount_lock);
1248 LIST_HEAD(umount_list);
1249 br_write_lock(vfsmount_lock);
1250 umount_tree(res, 0, &umount_list);
1251 br_write_unlock(vfsmount_lock);
1252 release_mounts(&umount_list);
1257 struct vfsmount *collect_mounts(struct path *path)
1260 down_write(&namespace_sem);
1261 tree = copy_tree(real_mount(path->mnt), path->dentry,
1262 CL_COPY_ALL | CL_PRIVATE);
1263 up_write(&namespace_sem);
1264 return tree ? &tree->mnt : NULL;
1267 void drop_collected_mounts(struct vfsmount *mnt)
1269 LIST_HEAD(umount_list);
1270 down_write(&namespace_sem);
1271 br_write_lock(vfsmount_lock);
1272 umount_tree(real_mount(mnt), 0, &umount_list);
1273 br_write_unlock(vfsmount_lock);
1274 up_write(&namespace_sem);
1275 release_mounts(&umount_list);
1278 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1279 struct vfsmount *root)
1282 int res = f(root, arg);
1285 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1286 res = f(&mnt->mnt, arg);
1293 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1297 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1298 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1299 mnt_release_group_id(p);
1303 static int invent_group_ids(struct mount *mnt, bool recurse)
1307 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1308 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1309 int err = mnt_alloc_group_id(p);
1311 cleanup_group_ids(mnt, p);
1321 * @source_mnt : mount tree to be attached
1322 * @nd : place the mount tree @source_mnt is attached
1323 * @parent_nd : if non-null, detach the source_mnt from its parent and
1324 * store the parent mount and mountpoint dentry.
1325 * (done when source_mnt is moved)
1327 * NOTE: in the table below explains the semantics when a source mount
1328 * of a given type is attached to a destination mount of a given type.
1329 * ---------------------------------------------------------------------------
1330 * | BIND MOUNT OPERATION |
1331 * |**************************************************************************
1332 * | source-->| shared | private | slave | unbindable |
1336 * |**************************************************************************
1337 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1339 * |non-shared| shared (+) | private | slave (*) | invalid |
1340 * ***************************************************************************
1341 * A bind operation clones the source mount and mounts the clone on the
1342 * destination mount.
1344 * (++) the cloned mount is propagated to all the mounts in the propagation
1345 * tree of the destination mount and the cloned mount is added to
1346 * the peer group of the source mount.
1347 * (+) the cloned mount is created under the destination mount and is marked
1348 * as shared. The cloned mount is added to the peer group of the source
1350 * (+++) the mount is propagated to all the mounts in the propagation tree
1351 * of the destination mount and the cloned mount is made slave
1352 * of the same master as that of the source mount. The cloned mount
1353 * is marked as 'shared and slave'.
1354 * (*) the cloned mount is made a slave of the same master as that of the
1357 * ---------------------------------------------------------------------------
1358 * | MOVE MOUNT OPERATION |
1359 * |**************************************************************************
1360 * | source-->| shared | private | slave | unbindable |
1364 * |**************************************************************************
1365 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1367 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1368 * ***************************************************************************
1370 * (+) the mount is moved to the destination. And is then propagated to
1371 * all the mounts in the propagation tree of the destination mount.
1372 * (+*) the mount is moved to the destination.
1373 * (+++) the mount is moved to the destination and is then propagated to
1374 * all the mounts belonging to the destination mount's propagation tree.
1375 * the mount is marked as 'shared and slave'.
1376 * (*) the mount continues to be a slave at the new location.
1378 * if the source mount is a tree, the operations explained above is
1379 * applied to each mount in the tree.
1380 * Must be called without spinlocks held, since this function can sleep
1383 static int attach_recursive_mnt(struct mount *source_mnt,
1384 struct path *path, struct path *parent_path)
1386 LIST_HEAD(tree_list);
1387 struct mount *dest_mnt = real_mount(path->mnt);
1388 struct dentry *dest_dentry = path->dentry;
1389 struct mount *child, *p;
1392 if (IS_MNT_SHARED(dest_mnt)) {
1393 err = invent_group_ids(source_mnt, true);
1397 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1399 goto out_cleanup_ids;
1401 br_write_lock(vfsmount_lock);
1403 if (IS_MNT_SHARED(dest_mnt)) {
1404 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1408 detach_mnt(source_mnt, parent_path);
1409 attach_mnt(source_mnt, path);
1410 touch_mnt_namespace(source_mnt->mnt_ns);
1412 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1413 commit_tree(source_mnt);
1416 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1417 list_del_init(&child->mnt_hash);
1420 br_write_unlock(vfsmount_lock);
1425 if (IS_MNT_SHARED(dest_mnt))
1426 cleanup_group_ids(source_mnt, NULL);
1431 static int lock_mount(struct path *path)
1433 struct vfsmount *mnt;
1435 mutex_lock(&path->dentry->d_inode->i_mutex);
1436 if (unlikely(cant_mount(path->dentry))) {
1437 mutex_unlock(&path->dentry->d_inode->i_mutex);
1440 down_write(&namespace_sem);
1441 mnt = lookup_mnt(path);
1444 up_write(&namespace_sem);
1445 mutex_unlock(&path->dentry->d_inode->i_mutex);
1448 path->dentry = dget(mnt->mnt_root);
1452 static void unlock_mount(struct path *path)
1454 up_write(&namespace_sem);
1455 mutex_unlock(&path->dentry->d_inode->i_mutex);
1458 static int graft_tree(struct mount *mnt, struct path *path)
1460 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1463 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1464 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1467 if (d_unlinked(path->dentry))
1470 return attach_recursive_mnt(mnt, path, NULL);
1474 * Sanity check the flags to change_mnt_propagation.
1477 static int flags_to_propagation_type(int flags)
1479 int type = flags & ~(MS_REC | MS_SILENT);
1481 /* Fail if any non-propagation flags are set */
1482 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1484 /* Only one propagation flag should be set */
1485 if (!is_power_of_2(type))
1491 * recursively change the type of the mountpoint.
1493 static int do_change_type(struct path *path, int flag)
1496 struct mount *mnt = real_mount(path->mnt);
1497 int recurse = flag & MS_REC;
1501 if (!capable(CAP_SYS_ADMIN))
1504 if (path->dentry != path->mnt->mnt_root)
1507 type = flags_to_propagation_type(flag);
1511 down_write(&namespace_sem);
1512 if (type == MS_SHARED) {
1513 err = invent_group_ids(mnt, recurse);
1518 br_write_lock(vfsmount_lock);
1519 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1520 change_mnt_propagation(m, type);
1521 br_write_unlock(vfsmount_lock);
1524 up_write(&namespace_sem);
1529 * do loopback mount.
1531 static int do_loopback(struct path *path, char *old_name,
1534 LIST_HEAD(umount_list);
1535 struct path old_path;
1536 struct mount *mnt = NULL, *old;
1537 int err = mount_is_safe(path);
1540 if (!old_name || !*old_name)
1542 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1546 err = lock_mount(path);
1550 old = real_mount(old_path.mnt);
1553 if (IS_MNT_UNBINDABLE(old))
1556 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1561 mnt = copy_tree(old, old_path.dentry, 0);
1563 mnt = clone_mnt(old, old_path.dentry, 0);
1568 err = graft_tree(mnt, path);
1570 br_write_lock(vfsmount_lock);
1571 umount_tree(mnt, 0, &umount_list);
1572 br_write_unlock(vfsmount_lock);
1576 release_mounts(&umount_list);
1578 path_put(&old_path);
1582 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1585 int readonly_request = 0;
1587 if (ms_flags & MS_RDONLY)
1588 readonly_request = 1;
1589 if (readonly_request == __mnt_is_readonly(mnt))
1592 if (readonly_request)
1593 error = mnt_make_readonly(real_mount(mnt));
1595 __mnt_unmake_readonly(real_mount(mnt));
1600 * change filesystem flags. dir should be a physical root of filesystem.
1601 * If you've mounted a non-root directory somewhere and want to do remount
1602 * on it - tough luck.
1604 static int do_remount(struct path *path, int flags, int mnt_flags,
1608 struct super_block *sb = path->mnt->mnt_sb;
1609 struct mount *mnt = real_mount(path->mnt);
1611 if (!capable(CAP_SYS_ADMIN))
1614 if (!check_mnt(mnt))
1617 if (path->dentry != path->mnt->mnt_root)
1620 err = security_sb_remount(sb, data);
1624 down_write(&sb->s_umount);
1625 if (flags & MS_BIND)
1626 err = change_mount_flags(path->mnt, flags);
1628 err = do_remount_sb(sb, flags, data, 0);
1630 br_write_lock(vfsmount_lock);
1631 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1632 mnt->mnt.mnt_flags = mnt_flags;
1633 br_write_unlock(vfsmount_lock);
1635 up_write(&sb->s_umount);
1637 br_write_lock(vfsmount_lock);
1638 touch_mnt_namespace(mnt->mnt_ns);
1639 br_write_unlock(vfsmount_lock);
1644 static inline int tree_contains_unbindable(struct mount *mnt)
1647 for (p = mnt; p; p = next_mnt(p, mnt)) {
1648 if (IS_MNT_UNBINDABLE(p))
1654 static int do_move_mount(struct path *path, char *old_name)
1656 struct path old_path, parent_path;
1660 if (!capable(CAP_SYS_ADMIN))
1662 if (!old_name || !*old_name)
1664 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1668 err = lock_mount(path);
1672 old = real_mount(old_path.mnt);
1673 p = real_mount(path->mnt);
1676 if (!check_mnt(p) || !check_mnt(old))
1679 if (d_unlinked(path->dentry))
1683 if (old_path.dentry != old_path.mnt->mnt_root)
1686 if (!mnt_has_parent(old))
1689 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1690 S_ISDIR(old_path.dentry->d_inode->i_mode))
1693 * Don't move a mount residing in a shared parent.
1695 if (IS_MNT_SHARED(old->mnt_parent))
1698 * Don't move a mount tree containing unbindable mounts to a destination
1699 * mount which is shared.
1701 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1704 for (; mnt_has_parent(p); p = p->mnt_parent)
1708 err = attach_recursive_mnt(old, path, &parent_path);
1712 /* if the mount is moved, it should no longer be expire
1714 list_del_init(&old->mnt_expire);
1719 path_put(&parent_path);
1720 path_put(&old_path);
1724 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1727 const char *subtype = strchr(fstype, '.');
1736 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1738 if (!mnt->mnt_sb->s_subtype)
1744 return ERR_PTR(err);
1747 static struct vfsmount *
1748 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1750 struct file_system_type *type = get_fs_type(fstype);
1751 struct vfsmount *mnt;
1753 return ERR_PTR(-ENODEV);
1754 mnt = vfs_kern_mount(type, flags, name, data);
1755 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1756 !mnt->mnt_sb->s_subtype)
1757 mnt = fs_set_subtype(mnt, fstype);
1758 put_filesystem(type);
1763 * add a mount into a namespace's mount tree
1765 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1769 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1771 err = lock_mount(path);
1776 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1779 /* Refuse the same filesystem on the same mount point */
1781 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1782 path->mnt->mnt_root == path->dentry)
1786 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1789 newmnt->mnt.mnt_flags = mnt_flags;
1790 err = graft_tree(newmnt, path);
1798 * create a new mount for userspace and request it to be added into the
1801 static int do_new_mount(struct path *path, char *type, int flags,
1802 int mnt_flags, char *name, void *data)
1804 struct vfsmount *mnt;
1810 /* we need capabilities... */
1811 if (!capable(CAP_SYS_ADMIN))
1814 mnt = do_kern_mount(type, flags, name, data);
1816 return PTR_ERR(mnt);
1818 err = do_add_mount(real_mount(mnt), path, mnt_flags);
1824 int finish_automount(struct vfsmount *m, struct path *path)
1826 struct mount *mnt = real_mount(m);
1828 /* The new mount record should have at least 2 refs to prevent it being
1829 * expired before we get a chance to add it
1831 BUG_ON(mnt_get_count(mnt) < 2);
1833 if (m->mnt_sb == path->mnt->mnt_sb &&
1834 m->mnt_root == path->dentry) {
1839 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1843 /* remove m from any expiration list it may be on */
1844 if (!list_empty(&mnt->mnt_expire)) {
1845 down_write(&namespace_sem);
1846 br_write_lock(vfsmount_lock);
1847 list_del_init(&mnt->mnt_expire);
1848 br_write_unlock(vfsmount_lock);
1849 up_write(&namespace_sem);
1857 * mnt_set_expiry - Put a mount on an expiration list
1858 * @mnt: The mount to list.
1859 * @expiry_list: The list to add the mount to.
1861 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1863 down_write(&namespace_sem);
1864 br_write_lock(vfsmount_lock);
1866 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1868 br_write_unlock(vfsmount_lock);
1869 up_write(&namespace_sem);
1871 EXPORT_SYMBOL(mnt_set_expiry);
1874 * process a list of expirable mountpoints with the intent of discarding any
1875 * mountpoints that aren't in use and haven't been touched since last we came
1878 void mark_mounts_for_expiry(struct list_head *mounts)
1880 struct mount *mnt, *next;
1881 LIST_HEAD(graveyard);
1884 if (list_empty(mounts))
1887 down_write(&namespace_sem);
1888 br_write_lock(vfsmount_lock);
1890 /* extract from the expiration list every vfsmount that matches the
1891 * following criteria:
1892 * - only referenced by its parent vfsmount
1893 * - still marked for expiry (marked on the last call here; marks are
1894 * cleared by mntput())
1896 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1897 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1898 propagate_mount_busy(mnt, 1))
1900 list_move(&mnt->mnt_expire, &graveyard);
1902 while (!list_empty(&graveyard)) {
1903 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
1904 touch_mnt_namespace(mnt->mnt_ns);
1905 umount_tree(mnt, 1, &umounts);
1907 br_write_unlock(vfsmount_lock);
1908 up_write(&namespace_sem);
1910 release_mounts(&umounts);
1913 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1916 * Ripoff of 'select_parent()'
1918 * search the list of submounts for a given mountpoint, and move any
1919 * shrinkable submounts to the 'graveyard' list.
1921 static int select_submounts(struct mount *parent, struct list_head *graveyard)
1923 struct mount *this_parent = parent;
1924 struct list_head *next;
1928 next = this_parent->mnt_mounts.next;
1930 while (next != &this_parent->mnt_mounts) {
1931 struct list_head *tmp = next;
1932 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
1935 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
1938 * Descend a level if the d_mounts list is non-empty.
1940 if (!list_empty(&mnt->mnt_mounts)) {
1945 if (!propagate_mount_busy(mnt, 1)) {
1946 list_move_tail(&mnt->mnt_expire, graveyard);
1951 * All done at this level ... ascend and resume the search
1953 if (this_parent != parent) {
1954 next = this_parent->mnt_child.next;
1955 this_parent = this_parent->mnt_parent;
1962 * process a list of expirable mountpoints with the intent of discarding any
1963 * submounts of a specific parent mountpoint
1965 * vfsmount_lock must be held for write
1967 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
1969 LIST_HEAD(graveyard);
1972 /* extract submounts of 'mountpoint' from the expiration list */
1973 while (select_submounts(mnt, &graveyard)) {
1974 while (!list_empty(&graveyard)) {
1975 m = list_first_entry(&graveyard, struct mount,
1977 touch_mnt_namespace(m->mnt_ns);
1978 umount_tree(m, 1, umounts);
1984 * Some copy_from_user() implementations do not return the exact number of
1985 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1986 * Note that this function differs from copy_from_user() in that it will oops
1987 * on bad values of `to', rather than returning a short copy.
1989 static long exact_copy_from_user(void *to, const void __user * from,
1993 const char __user *f = from;
1996 if (!access_ok(VERIFY_READ, from, n))
2000 if (__get_user(c, f)) {
2011 int copy_mount_options(const void __user * data, unsigned long *where)
2021 if (!(page = __get_free_page(GFP_KERNEL)))
2024 /* We only care that *some* data at the address the user
2025 * gave us is valid. Just in case, we'll zero
2026 * the remainder of the page.
2028 /* copy_from_user cannot cross TASK_SIZE ! */
2029 size = TASK_SIZE - (unsigned long)data;
2030 if (size > PAGE_SIZE)
2033 i = size - exact_copy_from_user((void *)page, data, size);
2039 memset((char *)page + i, 0, PAGE_SIZE - i);
2044 int copy_mount_string(const void __user *data, char **where)
2053 tmp = strndup_user(data, PAGE_SIZE);
2055 return PTR_ERR(tmp);
2062 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2063 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2065 * data is a (void *) that can point to any structure up to
2066 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2067 * information (or be NULL).
2069 * Pre-0.97 versions of mount() didn't have a flags word.
2070 * When the flags word was introduced its top half was required
2071 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2072 * Therefore, if this magic number is present, it carries no information
2073 * and must be discarded.
2075 long do_mount(char *dev_name, char *dir_name, char *type_page,
2076 unsigned long flags, void *data_page)
2083 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2084 flags &= ~MS_MGC_MSK;
2086 /* Basic sanity checks */
2088 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2092 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2094 /* ... and get the mountpoint */
2095 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2099 retval = security_sb_mount(dev_name, &path,
2100 type_page, flags, data_page);
2104 /* Default to relatime unless overriden */
2105 if (!(flags & MS_NOATIME))
2106 mnt_flags |= MNT_RELATIME;
2108 /* Separate the per-mountpoint flags */
2109 if (flags & MS_NOSUID)
2110 mnt_flags |= MNT_NOSUID;
2111 if (flags & MS_NODEV)
2112 mnt_flags |= MNT_NODEV;
2113 if (flags & MS_NOEXEC)
2114 mnt_flags |= MNT_NOEXEC;
2115 if (flags & MS_NOATIME)
2116 mnt_flags |= MNT_NOATIME;
2117 if (flags & MS_NODIRATIME)
2118 mnt_flags |= MNT_NODIRATIME;
2119 if (flags & MS_STRICTATIME)
2120 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2121 if (flags & MS_RDONLY)
2122 mnt_flags |= MNT_READONLY;
2124 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2125 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2128 if (flags & MS_REMOUNT)
2129 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2131 else if (flags & MS_BIND)
2132 retval = do_loopback(&path, dev_name, flags & MS_REC);
2133 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2134 retval = do_change_type(&path, flags);
2135 else if (flags & MS_MOVE)
2136 retval = do_move_mount(&path, dev_name);
2138 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2139 dev_name, data_page);
2145 static struct mnt_namespace *alloc_mnt_ns(void)
2147 struct mnt_namespace *new_ns;
2149 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2151 return ERR_PTR(-ENOMEM);
2152 atomic_set(&new_ns->count, 1);
2153 new_ns->root = NULL;
2154 INIT_LIST_HEAD(&new_ns->list);
2155 init_waitqueue_head(&new_ns->poll);
2160 void mnt_make_longterm(struct vfsmount *mnt)
2162 __mnt_make_longterm(real_mount(mnt));
2165 void mnt_make_shortterm(struct vfsmount *m)
2168 struct mount *mnt = real_mount(m);
2169 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2171 br_write_lock(vfsmount_lock);
2172 atomic_dec(&mnt->mnt_longterm);
2173 br_write_unlock(vfsmount_lock);
2178 * Allocate a new namespace structure and populate it with contents
2179 * copied from the namespace of the passed in task structure.
2181 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2182 struct fs_struct *fs)
2184 struct mnt_namespace *new_ns;
2185 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2186 struct mount *p, *q;
2187 struct mount *old = mnt_ns->root;
2190 new_ns = alloc_mnt_ns();
2194 down_write(&namespace_sem);
2195 /* First pass: copy the tree topology */
2196 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2198 up_write(&namespace_sem);
2200 return ERR_PTR(-ENOMEM);
2203 br_write_lock(vfsmount_lock);
2204 list_add_tail(&new_ns->list, &new->mnt_list);
2205 br_write_unlock(vfsmount_lock);
2208 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2209 * as belonging to new namespace. We have already acquired a private
2210 * fs_struct, so tsk->fs->lock is not needed.
2216 __mnt_make_longterm(q);
2218 if (&p->mnt == fs->root.mnt) {
2219 fs->root.mnt = mntget(&q->mnt);
2220 __mnt_make_longterm(q);
2221 mnt_make_shortterm(&p->mnt);
2224 if (&p->mnt == fs->pwd.mnt) {
2225 fs->pwd.mnt = mntget(&q->mnt);
2226 __mnt_make_longterm(q);
2227 mnt_make_shortterm(&p->mnt);
2231 p = next_mnt(p, old);
2232 q = next_mnt(q, new);
2234 up_write(&namespace_sem);
2244 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2245 struct fs_struct *new_fs)
2247 struct mnt_namespace *new_ns;
2252 if (!(flags & CLONE_NEWNS))
2255 new_ns = dup_mnt_ns(ns, new_fs);
2262 * create_mnt_ns - creates a private namespace and adds a root filesystem
2263 * @mnt: pointer to the new root filesystem mountpoint
2265 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2267 struct mnt_namespace *new_ns = alloc_mnt_ns();
2268 if (!IS_ERR(new_ns)) {
2269 struct mount *mnt = real_mount(m);
2270 mnt->mnt_ns = new_ns;
2271 __mnt_make_longterm(mnt);
2273 list_add(&new_ns->list, &mnt->mnt_list);
2280 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2282 struct mnt_namespace *ns;
2283 struct super_block *s;
2287 ns = create_mnt_ns(mnt);
2289 return ERR_CAST(ns);
2291 err = vfs_path_lookup(mnt->mnt_root, mnt,
2292 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2297 return ERR_PTR(err);
2299 /* trade a vfsmount reference for active sb one */
2300 s = path.mnt->mnt_sb;
2301 atomic_inc(&s->s_active);
2303 /* lock the sucker */
2304 down_write(&s->s_umount);
2305 /* ... and return the root of (sub)tree on it */
2308 EXPORT_SYMBOL(mount_subtree);
2310 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2311 char __user *, type, unsigned long, flags, void __user *, data)
2317 unsigned long data_page;
2319 ret = copy_mount_string(type, &kernel_type);
2323 kernel_dir = getname(dir_name);
2324 if (IS_ERR(kernel_dir)) {
2325 ret = PTR_ERR(kernel_dir);
2329 ret = copy_mount_string(dev_name, &kernel_dev);
2333 ret = copy_mount_options(data, &data_page);
2337 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2338 (void *) data_page);
2340 free_page(data_page);
2344 putname(kernel_dir);
2352 * Return true if path is reachable from root
2354 * namespace_sem or vfsmount_lock is held
2356 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2357 const struct path *root)
2359 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2360 dentry = mnt->mnt_mountpoint;
2361 mnt = mnt->mnt_parent;
2363 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2366 int path_is_under(struct path *path1, struct path *path2)
2369 br_read_lock(vfsmount_lock);
2370 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2371 br_read_unlock(vfsmount_lock);
2374 EXPORT_SYMBOL(path_is_under);
2377 * pivot_root Semantics:
2378 * Moves the root file system of the current process to the directory put_old,
2379 * makes new_root as the new root file system of the current process, and sets
2380 * root/cwd of all processes which had them on the current root to new_root.
2383 * The new_root and put_old must be directories, and must not be on the
2384 * same file system as the current process root. The put_old must be
2385 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2386 * pointed to by put_old must yield the same directory as new_root. No other
2387 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2389 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2390 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2391 * in this situation.
2394 * - we don't move root/cwd if they are not at the root (reason: if something
2395 * cared enough to change them, it's probably wrong to force them elsewhere)
2396 * - it's okay to pick a root that isn't the root of a file system, e.g.
2397 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2398 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2401 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2402 const char __user *, put_old)
2404 struct path new, old, parent_path, root_parent, root;
2405 struct mount *new_mnt, *root_mnt;
2408 if (!capable(CAP_SYS_ADMIN))
2411 error = user_path_dir(new_root, &new);
2415 error = user_path_dir(put_old, &old);
2419 error = security_sb_pivotroot(&old, &new);
2423 get_fs_root(current->fs, &root);
2424 error = lock_mount(&old);
2429 new_mnt = real_mount(new.mnt);
2430 root_mnt = real_mount(root.mnt);
2431 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2432 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2433 IS_MNT_SHARED(root_mnt->mnt_parent))
2435 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2438 if (d_unlinked(new.dentry))
2440 if (d_unlinked(old.dentry))
2443 if (new.mnt == root.mnt ||
2444 old.mnt == root.mnt)
2445 goto out4; /* loop, on the same file system */
2447 if (root.mnt->mnt_root != root.dentry)
2448 goto out4; /* not a mountpoint */
2449 if (!mnt_has_parent(root_mnt))
2450 goto out4; /* not attached */
2451 if (new.mnt->mnt_root != new.dentry)
2452 goto out4; /* not a mountpoint */
2453 if (!mnt_has_parent(new_mnt))
2454 goto out4; /* not attached */
2455 /* make sure we can reach put_old from new_root */
2456 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2458 br_write_lock(vfsmount_lock);
2459 detach_mnt(new_mnt, &parent_path);
2460 detach_mnt(root_mnt, &root_parent);
2461 /* mount old root on put_old */
2462 attach_mnt(root_mnt, &old);
2463 /* mount new_root on / */
2464 attach_mnt(new_mnt, &root_parent);
2465 touch_mnt_namespace(current->nsproxy->mnt_ns);
2466 br_write_unlock(vfsmount_lock);
2467 chroot_fs_refs(&root, &new);
2472 path_put(&root_parent);
2473 path_put(&parent_path);
2485 static void __init init_mount_tree(void)
2487 struct vfsmount *mnt;
2488 struct mnt_namespace *ns;
2491 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2493 panic("Can't create rootfs");
2495 ns = create_mnt_ns(mnt);
2497 panic("Can't allocate initial namespace");
2499 init_task.nsproxy->mnt_ns = ns;
2503 root.dentry = mnt->mnt_root;
2505 set_fs_pwd(current->fs, &root);
2506 set_fs_root(current->fs, &root);
2509 void __init mnt_init(void)
2514 init_rwsem(&namespace_sem);
2516 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2517 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2519 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2521 if (!mount_hashtable)
2522 panic("Failed to allocate mount hash table\n");
2524 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2526 for (u = 0; u < HASH_SIZE; u++)
2527 INIT_LIST_HEAD(&mount_hashtable[u]);
2529 br_lock_init(vfsmount_lock);
2533 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2535 fs_kobj = kobject_create_and_add("fs", NULL);
2537 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2542 void put_mnt_ns(struct mnt_namespace *ns)
2544 LIST_HEAD(umount_list);
2546 if (!atomic_dec_and_test(&ns->count))
2548 down_write(&namespace_sem);
2549 br_write_lock(vfsmount_lock);
2550 umount_tree(ns->root, 0, &umount_list);
2551 br_write_unlock(vfsmount_lock);
2552 up_write(&namespace_sem);
2553 release_mounts(&umount_list);
2557 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2559 struct vfsmount *mnt;
2560 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2563 * it is a longterm mount, don't release mnt until
2564 * we unmount before file sys is unregistered
2566 mnt_make_longterm(mnt);
2570 EXPORT_SYMBOL_GPL(kern_mount_data);
2572 void kern_unmount(struct vfsmount *mnt)
2574 /* release long term mount so mount point can be released */
2575 if (!IS_ERR_OR_NULL(mnt)) {
2576 mnt_make_shortterm(mnt);
2580 EXPORT_SYMBOL(kern_unmount);
2582 bool our_mnt(struct vfsmount *mnt)
2584 return check_mnt(real_mount(mnt));