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/slab.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/percpu.h>
16 #include <linux/init.h>
17 #include <linux/kernel.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/cpumask.h>
21 #include <linux/module.h>
22 #include <linux/sysfs.h>
23 #include <linux/seq_file.h>
24 #include <linux/mnt_namespace.h>
25 #include <linux/namei.h>
26 #include <linux/nsproxy.h>
27 #include <linux/security.h>
28 #include <linux/mount.h>
29 #include <linux/ramfs.h>
30 #include <linux/log2.h>
31 #include <linux/idr.h>
32 #include <linux/fs_struct.h>
33 #include <linux/fsnotify.h>
34 #include <asm/uaccess.h>
35 #include <asm/unistd.h>
39 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40 #define HASH_SIZE (1UL << HASH_SHIFT)
43 static DEFINE_IDA(mnt_id_ida);
44 static DEFINE_IDA(mnt_group_ida);
45 static DEFINE_SPINLOCK(mnt_id_lock);
46 static int mnt_id_start = 0;
47 static int mnt_group_start = 1;
49 static struct list_head *mount_hashtable __read_mostly;
50 static struct kmem_cache *mnt_cache __read_mostly;
51 static struct rw_semaphore namespace_sem;
54 struct kobject *fs_kobj;
55 EXPORT_SYMBOL_GPL(fs_kobj);
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
65 DEFINE_BRLOCK(vfsmount_lock);
67 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
69 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
71 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
75 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
81 static int mnt_alloc_id(struct mount *mnt)
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
87 spin_lock(&mnt_id_lock);
88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
90 mnt_id_start = mnt->mnt_id + 1;
91 spin_unlock(&mnt_id_lock);
98 static void mnt_free_id(struct mount *mnt)
100 int id = mnt->mnt_id;
101 spin_lock(&mnt_id_lock);
102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
105 spin_unlock(&mnt_id_lock);
109 * Allocate a new peer group ID
111 * mnt_group_ida is protected by namespace_sem
113 static int mnt_alloc_group_id(struct mount *mnt)
117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
120 res = ida_get_new_above(&mnt_group_ida,
124 mnt_group_start = mnt->mnt_group_id + 1;
130 * Release a peer group ID
132 void mnt_release_group_id(struct mount *mnt)
134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
138 mnt->mnt_group_id = 0;
142 * vfsmount lock must be held for read
144 static inline void mnt_add_count(struct mount *mnt, int n)
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
156 * vfsmount lock must be held for write
158 unsigned int mnt_get_count(struct mount *mnt)
161 unsigned int count = 0;
164 for_each_possible_cpu(cpu) {
165 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
170 return mnt->mnt_count;
174 static struct mount *alloc_vfsmnt(const char *name)
176 struct mount *p = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
178 struct vfsmount *mnt = &p->mnt;
181 err = mnt_alloc_id(p);
186 p->mnt_devname = kstrdup(name, GFP_KERNEL);
192 p->mnt_pcp = alloc_percpu(struct mnt_pcp);
194 goto out_free_devname;
196 this_cpu_add(p->mnt_pcp->mnt_count, 1);
202 INIT_LIST_HEAD(&p->mnt_hash);
203 INIT_LIST_HEAD(&p->mnt_child);
204 INIT_LIST_HEAD(&p->mnt_mounts);
205 INIT_LIST_HEAD(&p->mnt_list);
206 INIT_LIST_HEAD(&p->mnt_expire);
207 INIT_LIST_HEAD(&p->mnt_share);
208 INIT_LIST_HEAD(&p->mnt_slave_list);
209 INIT_LIST_HEAD(&p->mnt_slave);
210 #ifdef CONFIG_FSNOTIFY
211 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
218 kfree(p->mnt_devname);
223 kmem_cache_free(mnt_cache, p);
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
246 int __mnt_is_readonly(struct vfsmount *mnt)
248 if (mnt->mnt_flags & MNT_READONLY)
250 if (mnt->mnt_sb->s_flags & MS_RDONLY)
254 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
256 static inline void mnt_inc_writers(struct mount *mnt)
259 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
265 static inline void mnt_dec_writers(struct mount *mnt)
268 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
274 static unsigned int mnt_get_writers(struct mount *mnt)
277 unsigned int count = 0;
280 for_each_possible_cpu(cpu) {
281 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
286 return mnt->mnt_writers;
291 * Most r/o checks on a fs are for operations that take
292 * discrete amounts of time, like a write() or unlink().
293 * We must keep track of when those operations start
294 * (for permission checks) and when they end, so that
295 * we can determine when writes are able to occur to
299 * mnt_want_write - get write access to a mount
300 * @m: the mount on which to take a write
302 * This tells the low-level filesystem that a write is
303 * about to be performed to it, and makes sure that
304 * writes are allowed before returning success. When
305 * the write operation is finished, mnt_drop_write()
306 * must be called. This is effectively a refcount.
308 int mnt_want_write(struct vfsmount *m)
310 struct mount *mnt = real_mount(m);
314 mnt_inc_writers(mnt);
316 * The store to mnt_inc_writers must be visible before we pass
317 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
318 * incremented count after it has set MNT_WRITE_HOLD.
321 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
324 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
325 * be set to match its requirements. So we must not load that until
326 * MNT_WRITE_HOLD is cleared.
329 if (__mnt_is_readonly(m)) {
330 mnt_dec_writers(mnt);
338 EXPORT_SYMBOL_GPL(mnt_want_write);
341 * mnt_clone_write - get write access to a mount
342 * @mnt: the mount on which to take a write
344 * This is effectively like mnt_want_write, except
345 * it must only be used to take an extra write reference
346 * on a mountpoint that we already know has a write reference
347 * on it. This allows some optimisation.
349 * After finished, mnt_drop_write must be called as usual to
350 * drop the reference.
352 int mnt_clone_write(struct vfsmount *mnt)
354 /* superblock may be r/o */
355 if (__mnt_is_readonly(mnt))
358 mnt_inc_writers(real_mount(mnt));
362 EXPORT_SYMBOL_GPL(mnt_clone_write);
365 * mnt_want_write_file - get write access to a file's mount
366 * @file: the file who's mount on which to take a write
368 * This is like mnt_want_write, but it takes a file and can
369 * do some optimisations if the file is open for write already
371 int mnt_want_write_file(struct file *file)
373 struct inode *inode = file->f_dentry->d_inode;
374 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
375 return mnt_want_write(file->f_path.mnt);
377 return mnt_clone_write(file->f_path.mnt);
379 EXPORT_SYMBOL_GPL(mnt_want_write_file);
382 * mnt_drop_write - give up write access to a mount
383 * @mnt: the mount on which to give up write access
385 * Tells the low-level filesystem that we are done
386 * performing writes to it. Must be matched with
387 * mnt_want_write() call above.
389 void mnt_drop_write(struct vfsmount *mnt)
392 mnt_dec_writers(real_mount(mnt));
395 EXPORT_SYMBOL_GPL(mnt_drop_write);
397 void mnt_drop_write_file(struct file *file)
399 mnt_drop_write(file->f_path.mnt);
401 EXPORT_SYMBOL(mnt_drop_write_file);
403 static int mnt_make_readonly(struct mount *mnt)
407 br_write_lock(vfsmount_lock);
408 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
410 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
411 * should be visible before we do.
416 * With writers on hold, if this value is zero, then there are
417 * definitely no active writers (although held writers may subsequently
418 * increment the count, they'll have to wait, and decrement it after
419 * seeing MNT_READONLY).
421 * It is OK to have counter incremented on one CPU and decremented on
422 * another: the sum will add up correctly. The danger would be when we
423 * sum up each counter, if we read a counter before it is incremented,
424 * but then read another CPU's count which it has been subsequently
425 * decremented from -- we would see more decrements than we should.
426 * MNT_WRITE_HOLD protects against this scenario, because
427 * mnt_want_write first increments count, then smp_mb, then spins on
428 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
429 * we're counting up here.
431 if (mnt_get_writers(mnt) > 0)
434 mnt->mnt.mnt_flags |= MNT_READONLY;
436 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
437 * that become unheld will see MNT_READONLY.
440 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
441 br_write_unlock(vfsmount_lock);
445 static void __mnt_unmake_readonly(struct mount *mnt)
447 br_write_lock(vfsmount_lock);
448 mnt->mnt.mnt_flags &= ~MNT_READONLY;
449 br_write_unlock(vfsmount_lock);
452 static void free_vfsmnt(struct mount *mnt)
454 kfree(mnt->mnt_devname);
457 free_percpu(mnt->mnt_pcp);
459 kmem_cache_free(mnt_cache, mnt);
463 * find the first or last mount at @dentry on vfsmount @mnt depending on
464 * @dir. If @dir is set return the first mount else return the last mount.
465 * vfsmount_lock must be held for read or write.
467 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
470 struct list_head *head = mount_hashtable + hash(mnt, dentry);
471 struct list_head *tmp = head;
472 struct mount *p, *found = NULL;
475 tmp = dir ? tmp->next : tmp->prev;
479 p = list_entry(tmp, struct mount, mnt_hash);
480 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
489 * lookup_mnt increments the ref count before returning
490 * the vfsmount struct.
492 struct vfsmount *lookup_mnt(struct path *path)
494 struct mount *child_mnt;
496 br_read_lock(vfsmount_lock);
497 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
499 mnt_add_count(child_mnt, 1);
500 br_read_unlock(vfsmount_lock);
501 return &child_mnt->mnt;
503 br_read_unlock(vfsmount_lock);
508 static inline int check_mnt(struct mount *mnt)
510 return mnt->mnt_ns == current->nsproxy->mnt_ns;
514 * vfsmount lock must be held for write
516 static void touch_mnt_namespace(struct mnt_namespace *ns)
520 wake_up_interruptible(&ns->poll);
525 * vfsmount lock must be held for write
527 static void __touch_mnt_namespace(struct mnt_namespace *ns)
529 if (ns && ns->event != event) {
531 wake_up_interruptible(&ns->poll);
536 * Clear dentry's mounted state if it has no remaining mounts.
537 * vfsmount_lock must be held for write.
539 static void dentry_reset_mounted(struct dentry *dentry)
543 for (u = 0; u < HASH_SIZE; u++) {
546 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
547 if (p->mnt_mountpoint == dentry)
551 spin_lock(&dentry->d_lock);
552 dentry->d_flags &= ~DCACHE_MOUNTED;
553 spin_unlock(&dentry->d_lock);
557 * vfsmount lock must be held for write
559 static void detach_mnt(struct mount *mnt, struct path *old_path)
561 old_path->dentry = mnt->mnt_mountpoint;
562 old_path->mnt = &mnt->mnt_parent->mnt;
563 mnt->mnt_parent = mnt;
564 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
565 list_del_init(&mnt->mnt_child);
566 list_del_init(&mnt->mnt_hash);
567 dentry_reset_mounted(old_path->dentry);
571 * vfsmount lock must be held for write
573 void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
574 struct mount *child_mnt)
576 child_mnt->mnt_parent = real_mount(mntget(&mnt->mnt));
577 child_mnt->mnt_mountpoint = dget(dentry);
578 spin_lock(&dentry->d_lock);
579 dentry->d_flags |= DCACHE_MOUNTED;
580 spin_unlock(&dentry->d_lock);
584 * vfsmount lock must be held for write
586 static void attach_mnt(struct mount *mnt, struct path *path)
588 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
589 list_add_tail(&mnt->mnt_hash, mount_hashtable +
590 hash(path->mnt, path->dentry));
591 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
594 static inline void __mnt_make_longterm(struct mount *mnt)
597 atomic_inc(&mnt->mnt_longterm);
601 /* needs vfsmount lock for write */
602 static inline void __mnt_make_shortterm(struct mount *mnt)
605 atomic_dec(&mnt->mnt_longterm);
610 * vfsmount lock must be held for write
612 static void commit_tree(struct mount *mnt)
614 struct mount *parent = mnt->mnt_parent;
617 struct mnt_namespace *n = parent->mnt_ns;
619 BUG_ON(parent == mnt);
621 list_add_tail(&head, &mnt->mnt_list);
622 list_for_each_entry(m, &head, mnt_list) {
624 __mnt_make_longterm(m);
627 list_splice(&head, n->list.prev);
629 list_add_tail(&mnt->mnt_hash, mount_hashtable +
630 hash(&parent->mnt, mnt->mnt_mountpoint));
631 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
632 touch_mnt_namespace(n);
635 static struct mount *next_mnt(struct mount *p, struct vfsmount *root)
637 struct list_head *next = p->mnt_mounts.next;
638 if (next == &p->mnt_mounts) {
642 next = p->mnt_child.next;
643 if (next != &p->mnt_parent->mnt_mounts)
648 return list_entry(next, struct mount, mnt_child);
651 static struct mount *skip_mnt_tree(struct mount *p)
653 struct list_head *prev = p->mnt_mounts.prev;
654 while (prev != &p->mnt_mounts) {
655 p = list_entry(prev, struct mount, mnt_child);
656 prev = p->mnt_mounts.prev;
662 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
668 return ERR_PTR(-ENODEV);
670 mnt = alloc_vfsmnt(name);
672 return ERR_PTR(-ENOMEM);
674 if (flags & MS_KERNMOUNT)
675 mnt->mnt.mnt_flags = MNT_INTERNAL;
677 root = mount_fs(type, flags, name, data);
680 return ERR_CAST(root);
683 mnt->mnt.mnt_root = root;
684 mnt->mnt.mnt_sb = root->d_sb;
685 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
686 mnt->mnt_parent = mnt;
689 EXPORT_SYMBOL_GPL(vfs_kern_mount);
691 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
694 struct super_block *sb = old->mnt.mnt_sb;
695 struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
698 if (flag & (CL_SLAVE | CL_PRIVATE))
699 mnt->mnt_group_id = 0; /* not a peer of original */
701 mnt->mnt_group_id = old->mnt_group_id;
703 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
704 int err = mnt_alloc_group_id(mnt);
709 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
710 atomic_inc(&sb->s_active);
711 mnt->mnt.mnt_sb = sb;
712 mnt->mnt.mnt_root = dget(root);
713 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
714 mnt->mnt_parent = mnt;
716 if (flag & CL_SLAVE) {
717 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
718 mnt->mnt_master = old;
719 CLEAR_MNT_SHARED(mnt);
720 } else if (!(flag & CL_PRIVATE)) {
721 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
722 list_add(&mnt->mnt_share, &old->mnt_share);
723 if (IS_MNT_SLAVE(old))
724 list_add(&mnt->mnt_slave, &old->mnt_slave);
725 mnt->mnt_master = old->mnt_master;
727 if (flag & CL_MAKE_SHARED)
730 /* stick the duplicate mount on the same expiry list
731 * as the original if that was on one */
732 if (flag & CL_EXPIRE) {
733 if (!list_empty(&old->mnt_expire))
734 list_add(&mnt->mnt_expire, &old->mnt_expire);
744 static inline void mntfree(struct mount *mnt)
746 struct vfsmount *m = &mnt->mnt;
747 struct super_block *sb = m->mnt_sb;
750 * This probably indicates that somebody messed
751 * up a mnt_want/drop_write() pair. If this
752 * happens, the filesystem was probably unable
753 * to make r/w->r/o transitions.
756 * The locking used to deal with mnt_count decrement provides barriers,
757 * so mnt_get_writers() below is safe.
759 WARN_ON(mnt_get_writers(mnt));
760 fsnotify_vfsmount_delete(m);
763 deactivate_super(sb);
766 static void mntput_no_expire(struct mount *mnt)
770 br_read_lock(vfsmount_lock);
771 if (likely(atomic_read(&mnt->mnt_longterm))) {
772 mnt_add_count(mnt, -1);
773 br_read_unlock(vfsmount_lock);
776 br_read_unlock(vfsmount_lock);
778 br_write_lock(vfsmount_lock);
779 mnt_add_count(mnt, -1);
780 if (mnt_get_count(mnt)) {
781 br_write_unlock(vfsmount_lock);
785 mnt_add_count(mnt, -1);
786 if (likely(mnt_get_count(mnt)))
788 br_write_lock(vfsmount_lock);
790 if (unlikely(mnt->mnt_pinned)) {
791 mnt_add_count(mnt, mnt->mnt_pinned + 1);
793 br_write_unlock(vfsmount_lock);
794 acct_auto_close_mnt(&mnt->mnt);
797 br_write_unlock(vfsmount_lock);
801 void mntput(struct vfsmount *mnt)
804 struct mount *m = real_mount(mnt);
805 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
806 if (unlikely(m->mnt_expiry_mark))
807 m->mnt_expiry_mark = 0;
811 EXPORT_SYMBOL(mntput);
813 struct vfsmount *mntget(struct vfsmount *mnt)
816 mnt_add_count(real_mount(mnt), 1);
819 EXPORT_SYMBOL(mntget);
821 void mnt_pin(struct vfsmount *mnt)
823 br_write_lock(vfsmount_lock);
824 real_mount(mnt)->mnt_pinned++;
825 br_write_unlock(vfsmount_lock);
827 EXPORT_SYMBOL(mnt_pin);
829 void mnt_unpin(struct vfsmount *m)
831 struct mount *mnt = real_mount(m);
832 br_write_lock(vfsmount_lock);
833 if (mnt->mnt_pinned) {
834 mnt_add_count(mnt, 1);
837 br_write_unlock(vfsmount_lock);
839 EXPORT_SYMBOL(mnt_unpin);
841 static inline void mangle(struct seq_file *m, const char *s)
843 seq_escape(m, s, " \t\n\\");
847 * Simple .show_options callback for filesystems which don't want to
848 * implement more complex mount option showing.
850 * See also save_mount_options().
852 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
857 options = rcu_dereference(mnt->mnt_sb->s_options);
859 if (options != NULL && options[0]) {
867 EXPORT_SYMBOL(generic_show_options);
870 * If filesystem uses generic_show_options(), this function should be
871 * called from the fill_super() callback.
873 * The .remount_fs callback usually needs to be handled in a special
874 * way, to make sure, that previous options are not overwritten if the
877 * Also note, that if the filesystem's .remount_fs function doesn't
878 * reset all options to their default value, but changes only newly
879 * given options, then the displayed options will not reflect reality
882 void save_mount_options(struct super_block *sb, char *options)
884 BUG_ON(sb->s_options);
885 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
887 EXPORT_SYMBOL(save_mount_options);
889 void replace_mount_options(struct super_block *sb, char *options)
891 char *old = sb->s_options;
892 rcu_assign_pointer(sb->s_options, options);
898 EXPORT_SYMBOL(replace_mount_options);
900 #ifdef CONFIG_PROC_FS
902 static void *m_start(struct seq_file *m, loff_t *pos)
904 struct proc_mounts *p = m->private;
906 down_read(&namespace_sem);
907 return seq_list_start(&p->ns->list, *pos);
910 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
912 struct proc_mounts *p = m->private;
914 return seq_list_next(v, &p->ns->list, pos);
917 static void m_stop(struct seq_file *m, void *v)
919 up_read(&namespace_sem);
922 int mnt_had_events(struct proc_mounts *p)
924 struct mnt_namespace *ns = p->ns;
927 br_read_lock(vfsmount_lock);
928 if (p->m.poll_event != ns->event) {
929 p->m.poll_event = ns->event;
932 br_read_unlock(vfsmount_lock);
937 struct proc_fs_info {
942 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
944 static const struct proc_fs_info fs_info[] = {
945 { MS_SYNCHRONOUS, ",sync" },
946 { MS_DIRSYNC, ",dirsync" },
947 { MS_MANDLOCK, ",mand" },
950 const struct proc_fs_info *fs_infop;
952 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
953 if (sb->s_flags & fs_infop->flag)
954 seq_puts(m, fs_infop->str);
957 return security_sb_show_options(m, sb);
960 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
962 static const struct proc_fs_info mnt_info[] = {
963 { MNT_NOSUID, ",nosuid" },
964 { MNT_NODEV, ",nodev" },
965 { MNT_NOEXEC, ",noexec" },
966 { MNT_NOATIME, ",noatime" },
967 { MNT_NODIRATIME, ",nodiratime" },
968 { MNT_RELATIME, ",relatime" },
971 const struct proc_fs_info *fs_infop;
973 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
974 if (mnt->mnt_flags & fs_infop->flag)
975 seq_puts(m, fs_infop->str);
979 static void show_type(struct seq_file *m, struct super_block *sb)
981 mangle(m, sb->s_type->name);
982 if (sb->s_subtype && sb->s_subtype[0]) {
984 mangle(m, sb->s_subtype);
988 static int show_vfsmnt(struct seq_file *m, void *v)
990 struct mount *r = list_entry(v, struct mount, mnt_list);
991 struct vfsmount *mnt = &r->mnt;
993 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
995 if (mnt->mnt_sb->s_op->show_devname) {
996 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1000 mangle(m, r->mnt_devname ? r->mnt_devname : "none");
1003 seq_path(m, &mnt_path, " \t\n\\");
1005 show_type(m, mnt->mnt_sb);
1006 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1007 err = show_sb_opts(m, mnt->mnt_sb);
1010 show_mnt_opts(m, mnt);
1011 if (mnt->mnt_sb->s_op->show_options)
1012 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1013 seq_puts(m, " 0 0\n");
1018 const struct seq_operations mounts_op = {
1025 static int show_mountinfo(struct seq_file *m, void *v)
1027 struct proc_mounts *p = m->private;
1028 struct mount *r = list_entry(v, struct mount, mnt_list);
1029 struct vfsmount *mnt = &r->mnt;
1030 struct super_block *sb = mnt->mnt_sb;
1031 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1032 struct path root = p->root;
1035 seq_printf(m, "%i %i %u:%u ", r->mnt_id, r->mnt_parent->mnt_id,
1036 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1037 if (sb->s_op->show_path)
1038 err = sb->s_op->show_path(m, mnt);
1040 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1045 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1046 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1050 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1051 show_mnt_opts(m, mnt);
1053 /* Tagged fields ("foo:X" or "bar") */
1054 if (IS_MNT_SHARED(r))
1055 seq_printf(m, " shared:%i", r->mnt_group_id);
1056 if (IS_MNT_SLAVE(r)) {
1057 int master = r->mnt_master->mnt_group_id;
1058 int dom = get_dominating_id(r, &p->root);
1059 seq_printf(m, " master:%i", master);
1060 if (dom && dom != master)
1061 seq_printf(m, " propagate_from:%i", dom);
1063 if (IS_MNT_UNBINDABLE(r))
1064 seq_puts(m, " unbindable");
1066 /* Filesystem specific data */
1070 if (sb->s_op->show_devname)
1071 err = sb->s_op->show_devname(m, mnt);
1073 mangle(m, r->mnt_devname ? r->mnt_devname : "none");
1076 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1077 err = show_sb_opts(m, sb);
1080 if (sb->s_op->show_options)
1081 err = sb->s_op->show_options(m, mnt);
1087 const struct seq_operations mountinfo_op = {
1091 .show = show_mountinfo,
1094 static int show_vfsstat(struct seq_file *m, void *v)
1096 struct mount *r = list_entry(v, struct mount, mnt_list);
1097 struct vfsmount *mnt = &r->mnt;
1098 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1102 if (mnt->mnt_sb->s_op->show_devname) {
1103 seq_puts(m, "device ");
1104 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1106 if (r->mnt_devname) {
1107 seq_puts(m, "device ");
1108 mangle(m, r->mnt_devname);
1110 seq_puts(m, "no device");
1114 seq_puts(m, " mounted on ");
1115 seq_path(m, &mnt_path, " \t\n\\");
1118 /* file system type */
1119 seq_puts(m, "with fstype ");
1120 show_type(m, mnt->mnt_sb);
1122 /* optional statistics */
1123 if (mnt->mnt_sb->s_op->show_stats) {
1126 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1133 const struct seq_operations mountstats_op = {
1137 .show = show_vfsstat,
1139 #endif /* CONFIG_PROC_FS */
1142 * may_umount_tree - check if a mount tree is busy
1143 * @mnt: root of mount tree
1145 * This is called to check if a tree of mounts has any
1146 * open files, pwds, chroots or sub mounts that are
1149 int may_umount_tree(struct vfsmount *mnt)
1151 int actual_refs = 0;
1152 int minimum_refs = 0;
1156 /* write lock needed for mnt_get_count */
1157 br_write_lock(vfsmount_lock);
1158 for (p = real_mount(mnt); p; p = next_mnt(p, mnt)) {
1159 actual_refs += mnt_get_count(p);
1162 br_write_unlock(vfsmount_lock);
1164 if (actual_refs > minimum_refs)
1170 EXPORT_SYMBOL(may_umount_tree);
1173 * may_umount - check if a mount point is busy
1174 * @mnt: root of mount
1176 * This is called to check if a mount point has any
1177 * open files, pwds, chroots or sub mounts. If the
1178 * mount has sub mounts this will return busy
1179 * regardless of whether the sub mounts are busy.
1181 * Doesn't take quota and stuff into account. IOW, in some cases it will
1182 * give false negatives. The main reason why it's here is that we need
1183 * a non-destructive way to look for easily umountable filesystems.
1185 int may_umount(struct vfsmount *mnt)
1188 down_read(&namespace_sem);
1189 br_write_lock(vfsmount_lock);
1190 if (propagate_mount_busy(real_mount(mnt), 2))
1192 br_write_unlock(vfsmount_lock);
1193 up_read(&namespace_sem);
1197 EXPORT_SYMBOL(may_umount);
1199 void release_mounts(struct list_head *head)
1202 while (!list_empty(head)) {
1203 mnt = list_first_entry(head, struct mount, mnt_hash);
1204 list_del_init(&mnt->mnt_hash);
1205 if (mnt_has_parent(mnt)) {
1206 struct dentry *dentry;
1209 br_write_lock(vfsmount_lock);
1210 dentry = mnt->mnt_mountpoint;
1211 m = mnt->mnt_parent;
1212 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1213 mnt->mnt_parent = mnt;
1215 br_write_unlock(vfsmount_lock);
1224 * vfsmount lock must be held for write
1225 * namespace_sem must be held for write
1227 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1229 LIST_HEAD(tmp_list);
1232 for (p = mnt; p; p = next_mnt(p, &mnt->mnt))
1233 list_move(&p->mnt_hash, &tmp_list);
1236 propagate_umount(&tmp_list);
1238 list_for_each_entry(p, &tmp_list, mnt_hash) {
1239 list_del_init(&p->mnt_expire);
1240 list_del_init(&p->mnt_list);
1241 __touch_mnt_namespace(p->mnt_ns);
1243 __mnt_make_shortterm(p);
1244 list_del_init(&p->mnt_child);
1245 if (mnt_has_parent(p)) {
1246 p->mnt_parent->mnt_ghosts++;
1247 dentry_reset_mounted(p->mnt_mountpoint);
1249 change_mnt_propagation(p, MS_PRIVATE);
1251 list_splice(&tmp_list, kill);
1254 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1256 static int do_umount(struct mount *mnt, int flags)
1258 struct super_block *sb = mnt->mnt.mnt_sb;
1260 LIST_HEAD(umount_list);
1262 retval = security_sb_umount(&mnt->mnt, flags);
1267 * Allow userspace to request a mountpoint be expired rather than
1268 * unmounting unconditionally. Unmount only happens if:
1269 * (1) the mark is already set (the mark is cleared by mntput())
1270 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1272 if (flags & MNT_EXPIRE) {
1273 if (&mnt->mnt == current->fs->root.mnt ||
1274 flags & (MNT_FORCE | MNT_DETACH))
1278 * probably don't strictly need the lock here if we examined
1279 * all race cases, but it's a slowpath.
1281 br_write_lock(vfsmount_lock);
1282 if (mnt_get_count(mnt) != 2) {
1283 br_write_unlock(vfsmount_lock);
1286 br_write_unlock(vfsmount_lock);
1288 if (!xchg(&mnt->mnt_expiry_mark, 1))
1293 * If we may have to abort operations to get out of this
1294 * mount, and they will themselves hold resources we must
1295 * allow the fs to do things. In the Unix tradition of
1296 * 'Gee thats tricky lets do it in userspace' the umount_begin
1297 * might fail to complete on the first run through as other tasks
1298 * must return, and the like. Thats for the mount program to worry
1299 * about for the moment.
1302 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1303 sb->s_op->umount_begin(sb);
1307 * No sense to grab the lock for this test, but test itself looks
1308 * somewhat bogus. Suggestions for better replacement?
1309 * Ho-hum... In principle, we might treat that as umount + switch
1310 * to rootfs. GC would eventually take care of the old vfsmount.
1311 * Actually it makes sense, especially if rootfs would contain a
1312 * /reboot - static binary that would close all descriptors and
1313 * call reboot(9). Then init(8) could umount root and exec /reboot.
1315 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1317 * Special case for "unmounting" root ...
1318 * we just try to remount it readonly.
1320 down_write(&sb->s_umount);
1321 if (!(sb->s_flags & MS_RDONLY))
1322 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1323 up_write(&sb->s_umount);
1327 down_write(&namespace_sem);
1328 br_write_lock(vfsmount_lock);
1331 if (!(flags & MNT_DETACH))
1332 shrink_submounts(mnt, &umount_list);
1335 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1336 if (!list_empty(&mnt->mnt_list))
1337 umount_tree(mnt, 1, &umount_list);
1340 br_write_unlock(vfsmount_lock);
1341 up_write(&namespace_sem);
1342 release_mounts(&umount_list);
1347 * Now umount can handle mount points as well as block devices.
1348 * This is important for filesystems which use unnamed block devices.
1350 * We now support a flag for forced unmount like the other 'big iron'
1351 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1354 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1359 int lookup_flags = 0;
1361 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1364 if (!(flags & UMOUNT_NOFOLLOW))
1365 lookup_flags |= LOOKUP_FOLLOW;
1367 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1370 mnt = real_mount(path.mnt);
1372 if (path.dentry != path.mnt->mnt_root)
1374 if (!check_mnt(mnt))
1378 if (!capable(CAP_SYS_ADMIN))
1381 retval = do_umount(mnt, flags);
1383 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1385 mntput_no_expire(mnt);
1390 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1393 * The 2.0 compatible umount. No flags.
1395 SYSCALL_DEFINE1(oldumount, char __user *, name)
1397 return sys_umount(name, 0);
1402 static int mount_is_safe(struct path *path)
1404 if (capable(CAP_SYS_ADMIN))
1408 if (S_ISLNK(path->dentry->d_inode->i_mode))
1410 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1411 if (current_uid() != path->dentry->d_inode->i_uid)
1414 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1420 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1423 struct mount *res, *p, *q, *r;
1426 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1429 res = q = clone_mnt(mnt, dentry, flag);
1432 q->mnt_mountpoint = mnt->mnt_mountpoint;
1435 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1437 if (!is_subdir(r->mnt_mountpoint, dentry))
1440 for (s = r; s; s = next_mnt(s, &r->mnt)) {
1441 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1442 s = skip_mnt_tree(s);
1445 while (p != s->mnt_parent) {
1451 path.dentry = p->mnt_mountpoint;
1452 q = clone_mnt(p, p->mnt.mnt_root, flag);
1455 br_write_lock(vfsmount_lock);
1456 list_add_tail(&q->mnt_list, &res->mnt_list);
1457 attach_mnt(q, &path);
1458 br_write_unlock(vfsmount_lock);
1464 LIST_HEAD(umount_list);
1465 br_write_lock(vfsmount_lock);
1466 umount_tree(res, 0, &umount_list);
1467 br_write_unlock(vfsmount_lock);
1468 release_mounts(&umount_list);
1473 struct vfsmount *collect_mounts(struct path *path)
1476 down_write(&namespace_sem);
1477 tree = copy_tree(real_mount(path->mnt), path->dentry,
1478 CL_COPY_ALL | CL_PRIVATE);
1479 up_write(&namespace_sem);
1480 return tree ? &tree->mnt : NULL;
1483 void drop_collected_mounts(struct vfsmount *mnt)
1485 LIST_HEAD(umount_list);
1486 down_write(&namespace_sem);
1487 br_write_lock(vfsmount_lock);
1488 umount_tree(real_mount(mnt), 0, &umount_list);
1489 br_write_unlock(vfsmount_lock);
1490 up_write(&namespace_sem);
1491 release_mounts(&umount_list);
1494 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1495 struct vfsmount *root)
1498 int res = f(root, arg);
1501 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1502 res = f(&mnt->mnt, arg);
1509 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1513 for (p = mnt; p != end; p = next_mnt(p, &mnt->mnt)) {
1514 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1515 mnt_release_group_id(p);
1519 static int invent_group_ids(struct mount *mnt, bool recurse)
1523 for (p = mnt; p; p = recurse ? next_mnt(p, &mnt->mnt) : NULL) {
1524 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1525 int err = mnt_alloc_group_id(p);
1527 cleanup_group_ids(mnt, p);
1537 * @source_mnt : mount tree to be attached
1538 * @nd : place the mount tree @source_mnt is attached
1539 * @parent_nd : if non-null, detach the source_mnt from its parent and
1540 * store the parent mount and mountpoint dentry.
1541 * (done when source_mnt is moved)
1543 * NOTE: in the table below explains the semantics when a source mount
1544 * of a given type is attached to a destination mount of a given type.
1545 * ---------------------------------------------------------------------------
1546 * | BIND MOUNT OPERATION |
1547 * |**************************************************************************
1548 * | source-->| shared | private | slave | unbindable |
1552 * |**************************************************************************
1553 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1555 * |non-shared| shared (+) | private | slave (*) | invalid |
1556 * ***************************************************************************
1557 * A bind operation clones the source mount and mounts the clone on the
1558 * destination mount.
1560 * (++) the cloned mount is propagated to all the mounts in the propagation
1561 * tree of the destination mount and the cloned mount is added to
1562 * the peer group of the source mount.
1563 * (+) the cloned mount is created under the destination mount and is marked
1564 * as shared. The cloned mount is added to the peer group of the source
1566 * (+++) the mount is propagated to all the mounts in the propagation tree
1567 * of the destination mount and the cloned mount is made slave
1568 * of the same master as that of the source mount. The cloned mount
1569 * is marked as 'shared and slave'.
1570 * (*) the cloned mount is made a slave of the same master as that of the
1573 * ---------------------------------------------------------------------------
1574 * | MOVE MOUNT OPERATION |
1575 * |**************************************************************************
1576 * | source-->| shared | private | slave | unbindable |
1580 * |**************************************************************************
1581 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1583 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1584 * ***************************************************************************
1586 * (+) the mount is moved to the destination. And is then propagated to
1587 * all the mounts in the propagation tree of the destination mount.
1588 * (+*) the mount is moved to the destination.
1589 * (+++) the mount is moved to the destination and is then propagated to
1590 * all the mounts belonging to the destination mount's propagation tree.
1591 * the mount is marked as 'shared and slave'.
1592 * (*) the mount continues to be a slave at the new location.
1594 * if the source mount is a tree, the operations explained above is
1595 * applied to each mount in the tree.
1596 * Must be called without spinlocks held, since this function can sleep
1599 static int attach_recursive_mnt(struct mount *source_mnt,
1600 struct path *path, struct path *parent_path)
1602 LIST_HEAD(tree_list);
1603 struct mount *dest_mnt = real_mount(path->mnt);
1604 struct dentry *dest_dentry = path->dentry;
1605 struct mount *child, *p;
1608 if (IS_MNT_SHARED(dest_mnt)) {
1609 err = invent_group_ids(source_mnt, true);
1613 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1615 goto out_cleanup_ids;
1617 br_write_lock(vfsmount_lock);
1619 if (IS_MNT_SHARED(dest_mnt)) {
1620 for (p = source_mnt; p; p = next_mnt(p, &source_mnt->mnt))
1624 detach_mnt(source_mnt, parent_path);
1625 attach_mnt(source_mnt, path);
1626 touch_mnt_namespace(source_mnt->mnt_ns);
1628 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1629 commit_tree(source_mnt);
1632 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1633 list_del_init(&child->mnt_hash);
1636 br_write_unlock(vfsmount_lock);
1641 if (IS_MNT_SHARED(dest_mnt))
1642 cleanup_group_ids(source_mnt, NULL);
1647 static int lock_mount(struct path *path)
1649 struct vfsmount *mnt;
1651 mutex_lock(&path->dentry->d_inode->i_mutex);
1652 if (unlikely(cant_mount(path->dentry))) {
1653 mutex_unlock(&path->dentry->d_inode->i_mutex);
1656 down_write(&namespace_sem);
1657 mnt = lookup_mnt(path);
1660 up_write(&namespace_sem);
1661 mutex_unlock(&path->dentry->d_inode->i_mutex);
1664 path->dentry = dget(mnt->mnt_root);
1668 static void unlock_mount(struct path *path)
1670 up_write(&namespace_sem);
1671 mutex_unlock(&path->dentry->d_inode->i_mutex);
1674 static int graft_tree(struct mount *mnt, struct path *path)
1676 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1679 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1680 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1683 if (d_unlinked(path->dentry))
1686 return attach_recursive_mnt(mnt, path, NULL);
1690 * Sanity check the flags to change_mnt_propagation.
1693 static int flags_to_propagation_type(int flags)
1695 int type = flags & ~(MS_REC | MS_SILENT);
1697 /* Fail if any non-propagation flags are set */
1698 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1700 /* Only one propagation flag should be set */
1701 if (!is_power_of_2(type))
1707 * recursively change the type of the mountpoint.
1709 static int do_change_type(struct path *path, int flag)
1712 struct mount *mnt = real_mount(path->mnt);
1713 int recurse = flag & MS_REC;
1717 if (!capable(CAP_SYS_ADMIN))
1720 if (path->dentry != path->mnt->mnt_root)
1723 type = flags_to_propagation_type(flag);
1727 down_write(&namespace_sem);
1728 if (type == MS_SHARED) {
1729 err = invent_group_ids(mnt, recurse);
1734 br_write_lock(vfsmount_lock);
1735 for (m = mnt; m; m = (recurse ? next_mnt(m, &mnt->mnt) : NULL))
1736 change_mnt_propagation(m, type);
1737 br_write_unlock(vfsmount_lock);
1740 up_write(&namespace_sem);
1745 * do loopback mount.
1747 static int do_loopback(struct path *path, char *old_name,
1750 LIST_HEAD(umount_list);
1751 struct path old_path;
1752 struct mount *mnt = NULL, *old;
1753 int err = mount_is_safe(path);
1756 if (!old_name || !*old_name)
1758 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1762 err = lock_mount(path);
1766 old = real_mount(old_path.mnt);
1769 if (IS_MNT_UNBINDABLE(old))
1772 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1777 mnt = copy_tree(old, old_path.dentry, 0);
1779 mnt = clone_mnt(old, old_path.dentry, 0);
1784 err = graft_tree(mnt, path);
1786 br_write_lock(vfsmount_lock);
1787 umount_tree(mnt, 0, &umount_list);
1788 br_write_unlock(vfsmount_lock);
1792 release_mounts(&umount_list);
1794 path_put(&old_path);
1798 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1801 int readonly_request = 0;
1803 if (ms_flags & MS_RDONLY)
1804 readonly_request = 1;
1805 if (readonly_request == __mnt_is_readonly(mnt))
1808 if (readonly_request)
1809 error = mnt_make_readonly(real_mount(mnt));
1811 __mnt_unmake_readonly(real_mount(mnt));
1816 * change filesystem flags. dir should be a physical root of filesystem.
1817 * If you've mounted a non-root directory somewhere and want to do remount
1818 * on it - tough luck.
1820 static int do_remount(struct path *path, int flags, int mnt_flags,
1824 struct super_block *sb = path->mnt->mnt_sb;
1825 struct mount *mnt = real_mount(path->mnt);
1827 if (!capable(CAP_SYS_ADMIN))
1830 if (!check_mnt(mnt))
1833 if (path->dentry != path->mnt->mnt_root)
1836 err = security_sb_remount(sb, data);
1840 down_write(&sb->s_umount);
1841 if (flags & MS_BIND)
1842 err = change_mount_flags(path->mnt, flags);
1844 err = do_remount_sb(sb, flags, data, 0);
1846 br_write_lock(vfsmount_lock);
1847 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1848 mnt->mnt.mnt_flags = mnt_flags;
1849 br_write_unlock(vfsmount_lock);
1851 up_write(&sb->s_umount);
1853 br_write_lock(vfsmount_lock);
1854 touch_mnt_namespace(mnt->mnt_ns);
1855 br_write_unlock(vfsmount_lock);
1860 static inline int tree_contains_unbindable(struct mount *mnt)
1863 for (p = mnt; p; p = next_mnt(p, &mnt->mnt)) {
1864 if (IS_MNT_UNBINDABLE(p))
1870 static int do_move_mount(struct path *path, char *old_name)
1872 struct path old_path, parent_path;
1876 if (!capable(CAP_SYS_ADMIN))
1878 if (!old_name || !*old_name)
1880 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1884 err = lock_mount(path);
1888 old = real_mount(old_path.mnt);
1889 p = real_mount(path->mnt);
1892 if (!check_mnt(p) || !check_mnt(old))
1895 if (d_unlinked(path->dentry))
1899 if (old_path.dentry != old_path.mnt->mnt_root)
1902 if (!mnt_has_parent(old))
1905 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1906 S_ISDIR(old_path.dentry->d_inode->i_mode))
1909 * Don't move a mount residing in a shared parent.
1911 if (IS_MNT_SHARED(old->mnt_parent))
1914 * Don't move a mount tree containing unbindable mounts to a destination
1915 * mount which is shared.
1917 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1920 for (; mnt_has_parent(p); p = p->mnt_parent)
1924 err = attach_recursive_mnt(old, path, &parent_path);
1928 /* if the mount is moved, it should no longer be expire
1930 list_del_init(&old->mnt_expire);
1935 path_put(&parent_path);
1936 path_put(&old_path);
1940 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1943 const char *subtype = strchr(fstype, '.');
1952 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1954 if (!mnt->mnt_sb->s_subtype)
1960 return ERR_PTR(err);
1963 static struct vfsmount *
1964 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1966 struct file_system_type *type = get_fs_type(fstype);
1967 struct vfsmount *mnt;
1969 return ERR_PTR(-ENODEV);
1970 mnt = vfs_kern_mount(type, flags, name, data);
1971 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1972 !mnt->mnt_sb->s_subtype)
1973 mnt = fs_set_subtype(mnt, fstype);
1974 put_filesystem(type);
1979 * add a mount into a namespace's mount tree
1981 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1985 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1987 err = lock_mount(path);
1992 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1995 /* Refuse the same filesystem on the same mount point */
1997 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1998 path->mnt->mnt_root == path->dentry)
2002 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2005 newmnt->mnt.mnt_flags = mnt_flags;
2006 err = graft_tree(newmnt, path);
2014 * create a new mount for userspace and request it to be added into the
2017 static int do_new_mount(struct path *path, char *type, int flags,
2018 int mnt_flags, char *name, void *data)
2020 struct vfsmount *mnt;
2026 /* we need capabilities... */
2027 if (!capable(CAP_SYS_ADMIN))
2030 mnt = do_kern_mount(type, flags, name, data);
2032 return PTR_ERR(mnt);
2034 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2040 int finish_automount(struct vfsmount *m, struct path *path)
2042 struct mount *mnt = real_mount(m);
2044 /* The new mount record should have at least 2 refs to prevent it being
2045 * expired before we get a chance to add it
2047 BUG_ON(mnt_get_count(mnt) < 2);
2049 if (m->mnt_sb == path->mnt->mnt_sb &&
2050 m->mnt_root == path->dentry) {
2055 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2059 /* remove m from any expiration list it may be on */
2060 if (!list_empty(&mnt->mnt_expire)) {
2061 down_write(&namespace_sem);
2062 br_write_lock(vfsmount_lock);
2063 list_del_init(&mnt->mnt_expire);
2064 br_write_unlock(vfsmount_lock);
2065 up_write(&namespace_sem);
2073 * mnt_set_expiry - Put a mount on an expiration list
2074 * @mnt: The mount to list.
2075 * @expiry_list: The list to add the mount to.
2077 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2079 down_write(&namespace_sem);
2080 br_write_lock(vfsmount_lock);
2082 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2084 br_write_unlock(vfsmount_lock);
2085 up_write(&namespace_sem);
2087 EXPORT_SYMBOL(mnt_set_expiry);
2090 * process a list of expirable mountpoints with the intent of discarding any
2091 * mountpoints that aren't in use and haven't been touched since last we came
2094 void mark_mounts_for_expiry(struct list_head *mounts)
2096 struct mount *mnt, *next;
2097 LIST_HEAD(graveyard);
2100 if (list_empty(mounts))
2103 down_write(&namespace_sem);
2104 br_write_lock(vfsmount_lock);
2106 /* extract from the expiration list every vfsmount that matches the
2107 * following criteria:
2108 * - only referenced by its parent vfsmount
2109 * - still marked for expiry (marked on the last call here; marks are
2110 * cleared by mntput())
2112 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2113 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2114 propagate_mount_busy(mnt, 1))
2116 list_move(&mnt->mnt_expire, &graveyard);
2118 while (!list_empty(&graveyard)) {
2119 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2120 touch_mnt_namespace(mnt->mnt_ns);
2121 umount_tree(mnt, 1, &umounts);
2123 br_write_unlock(vfsmount_lock);
2124 up_write(&namespace_sem);
2126 release_mounts(&umounts);
2129 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2132 * Ripoff of 'select_parent()'
2134 * search the list of submounts for a given mountpoint, and move any
2135 * shrinkable submounts to the 'graveyard' list.
2137 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2139 struct mount *this_parent = parent;
2140 struct list_head *next;
2144 next = this_parent->mnt_mounts.next;
2146 while (next != &this_parent->mnt_mounts) {
2147 struct list_head *tmp = next;
2148 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2151 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2154 * Descend a level if the d_mounts list is non-empty.
2156 if (!list_empty(&mnt->mnt_mounts)) {
2161 if (!propagate_mount_busy(mnt, 1)) {
2162 list_move_tail(&mnt->mnt_expire, graveyard);
2167 * All done at this level ... ascend and resume the search
2169 if (this_parent != parent) {
2170 next = this_parent->mnt_child.next;
2171 this_parent = this_parent->mnt_parent;
2178 * process a list of expirable mountpoints with the intent of discarding any
2179 * submounts of a specific parent mountpoint
2181 * vfsmount_lock must be held for write
2183 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2185 LIST_HEAD(graveyard);
2188 /* extract submounts of 'mountpoint' from the expiration list */
2189 while (select_submounts(mnt, &graveyard)) {
2190 while (!list_empty(&graveyard)) {
2191 m = list_first_entry(&graveyard, struct mount,
2193 touch_mnt_namespace(m->mnt_ns);
2194 umount_tree(m, 1, umounts);
2200 * Some copy_from_user() implementations do not return the exact number of
2201 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2202 * Note that this function differs from copy_from_user() in that it will oops
2203 * on bad values of `to', rather than returning a short copy.
2205 static long exact_copy_from_user(void *to, const void __user * from,
2209 const char __user *f = from;
2212 if (!access_ok(VERIFY_READ, from, n))
2216 if (__get_user(c, f)) {
2227 int copy_mount_options(const void __user * data, unsigned long *where)
2237 if (!(page = __get_free_page(GFP_KERNEL)))
2240 /* We only care that *some* data at the address the user
2241 * gave us is valid. Just in case, we'll zero
2242 * the remainder of the page.
2244 /* copy_from_user cannot cross TASK_SIZE ! */
2245 size = TASK_SIZE - (unsigned long)data;
2246 if (size > PAGE_SIZE)
2249 i = size - exact_copy_from_user((void *)page, data, size);
2255 memset((char *)page + i, 0, PAGE_SIZE - i);
2260 int copy_mount_string(const void __user *data, char **where)
2269 tmp = strndup_user(data, PAGE_SIZE);
2271 return PTR_ERR(tmp);
2278 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2279 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2281 * data is a (void *) that can point to any structure up to
2282 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2283 * information (or be NULL).
2285 * Pre-0.97 versions of mount() didn't have a flags word.
2286 * When the flags word was introduced its top half was required
2287 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2288 * Therefore, if this magic number is present, it carries no information
2289 * and must be discarded.
2291 long do_mount(char *dev_name, char *dir_name, char *type_page,
2292 unsigned long flags, void *data_page)
2299 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2300 flags &= ~MS_MGC_MSK;
2302 /* Basic sanity checks */
2304 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2308 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2310 /* ... and get the mountpoint */
2311 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2315 retval = security_sb_mount(dev_name, &path,
2316 type_page, flags, data_page);
2320 /* Default to relatime unless overriden */
2321 if (!(flags & MS_NOATIME))
2322 mnt_flags |= MNT_RELATIME;
2324 /* Separate the per-mountpoint flags */
2325 if (flags & MS_NOSUID)
2326 mnt_flags |= MNT_NOSUID;
2327 if (flags & MS_NODEV)
2328 mnt_flags |= MNT_NODEV;
2329 if (flags & MS_NOEXEC)
2330 mnt_flags |= MNT_NOEXEC;
2331 if (flags & MS_NOATIME)
2332 mnt_flags |= MNT_NOATIME;
2333 if (flags & MS_NODIRATIME)
2334 mnt_flags |= MNT_NODIRATIME;
2335 if (flags & MS_STRICTATIME)
2336 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2337 if (flags & MS_RDONLY)
2338 mnt_flags |= MNT_READONLY;
2340 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2341 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2344 if (flags & MS_REMOUNT)
2345 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2347 else if (flags & MS_BIND)
2348 retval = do_loopback(&path, dev_name, flags & MS_REC);
2349 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2350 retval = do_change_type(&path, flags);
2351 else if (flags & MS_MOVE)
2352 retval = do_move_mount(&path, dev_name);
2354 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2355 dev_name, data_page);
2361 static struct mnt_namespace *alloc_mnt_ns(void)
2363 struct mnt_namespace *new_ns;
2365 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2367 return ERR_PTR(-ENOMEM);
2368 atomic_set(&new_ns->count, 1);
2369 new_ns->root = NULL;
2370 INIT_LIST_HEAD(&new_ns->list);
2371 init_waitqueue_head(&new_ns->poll);
2376 void mnt_make_longterm(struct vfsmount *mnt)
2378 __mnt_make_longterm(real_mount(mnt));
2381 void mnt_make_shortterm(struct vfsmount *m)
2384 struct mount *mnt = real_mount(m);
2385 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2387 br_write_lock(vfsmount_lock);
2388 atomic_dec(&mnt->mnt_longterm);
2389 br_write_unlock(vfsmount_lock);
2394 * Allocate a new namespace structure and populate it with contents
2395 * copied from the namespace of the passed in task structure.
2397 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2398 struct fs_struct *fs)
2400 struct mnt_namespace *new_ns;
2401 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2402 struct mount *p, *q;
2405 new_ns = alloc_mnt_ns();
2409 down_write(&namespace_sem);
2410 /* First pass: copy the tree topology */
2411 new = copy_tree(real_mount(mnt_ns->root), mnt_ns->root->mnt_root,
2412 CL_COPY_ALL | CL_EXPIRE);
2414 up_write(&namespace_sem);
2416 return ERR_PTR(-ENOMEM);
2418 new_ns->root = &new->mnt;
2419 br_write_lock(vfsmount_lock);
2420 list_add_tail(&new_ns->list, &new->mnt_list);
2421 br_write_unlock(vfsmount_lock);
2424 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2425 * as belonging to new namespace. We have already acquired a private
2426 * fs_struct, so tsk->fs->lock is not needed.
2428 p = real_mount(mnt_ns->root);
2432 __mnt_make_longterm(q);
2434 if (&p->mnt == fs->root.mnt) {
2435 fs->root.mnt = mntget(&q->mnt);
2436 __mnt_make_longterm(q);
2437 mnt_make_shortterm(&p->mnt);
2440 if (&p->mnt == fs->pwd.mnt) {
2441 fs->pwd.mnt = mntget(&q->mnt);
2442 __mnt_make_longterm(q);
2443 mnt_make_shortterm(&p->mnt);
2447 p = next_mnt(p, mnt_ns->root);
2448 q = next_mnt(q, new_ns->root);
2450 up_write(&namespace_sem);
2460 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2461 struct fs_struct *new_fs)
2463 struct mnt_namespace *new_ns;
2468 if (!(flags & CLONE_NEWNS))
2471 new_ns = dup_mnt_ns(ns, new_fs);
2478 * create_mnt_ns - creates a private namespace and adds a root filesystem
2479 * @mnt: pointer to the new root filesystem mountpoint
2481 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2483 struct mnt_namespace *new_ns = alloc_mnt_ns();
2484 if (!IS_ERR(new_ns)) {
2485 struct mount *mnt = real_mount(m);
2486 mnt->mnt_ns = new_ns;
2487 __mnt_make_longterm(mnt);
2489 list_add(&new_ns->list, &mnt->mnt_list);
2496 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2498 struct mnt_namespace *ns;
2499 struct super_block *s;
2503 ns = create_mnt_ns(mnt);
2505 return ERR_CAST(ns);
2507 err = vfs_path_lookup(mnt->mnt_root, mnt,
2508 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2513 return ERR_PTR(err);
2515 /* trade a vfsmount reference for active sb one */
2516 s = path.mnt->mnt_sb;
2517 atomic_inc(&s->s_active);
2519 /* lock the sucker */
2520 down_write(&s->s_umount);
2521 /* ... and return the root of (sub)tree on it */
2524 EXPORT_SYMBOL(mount_subtree);
2526 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2527 char __user *, type, unsigned long, flags, void __user *, data)
2533 unsigned long data_page;
2535 ret = copy_mount_string(type, &kernel_type);
2539 kernel_dir = getname(dir_name);
2540 if (IS_ERR(kernel_dir)) {
2541 ret = PTR_ERR(kernel_dir);
2545 ret = copy_mount_string(dev_name, &kernel_dev);
2549 ret = copy_mount_options(data, &data_page);
2553 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2554 (void *) data_page);
2556 free_page(data_page);
2560 putname(kernel_dir);
2568 * Return true if path is reachable from root
2570 * namespace_sem or vfsmount_lock is held
2572 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2573 const struct path *root)
2575 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2576 dentry = mnt->mnt_mountpoint;
2577 mnt = mnt->mnt_parent;
2579 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2582 int path_is_under(struct path *path1, struct path *path2)
2585 br_read_lock(vfsmount_lock);
2586 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2587 br_read_unlock(vfsmount_lock);
2590 EXPORT_SYMBOL(path_is_under);
2593 * pivot_root Semantics:
2594 * Moves the root file system of the current process to the directory put_old,
2595 * makes new_root as the new root file system of the current process, and sets
2596 * root/cwd of all processes which had them on the current root to new_root.
2599 * The new_root and put_old must be directories, and must not be on the
2600 * same file system as the current process root. The put_old must be
2601 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2602 * pointed to by put_old must yield the same directory as new_root. No other
2603 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2605 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2606 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2607 * in this situation.
2610 * - we don't move root/cwd if they are not at the root (reason: if something
2611 * cared enough to change them, it's probably wrong to force them elsewhere)
2612 * - it's okay to pick a root that isn't the root of a file system, e.g.
2613 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2614 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2617 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2618 const char __user *, put_old)
2620 struct path new, old, parent_path, root_parent, root;
2621 struct mount *new_mnt, *root_mnt;
2624 if (!capable(CAP_SYS_ADMIN))
2627 error = user_path_dir(new_root, &new);
2631 error = user_path_dir(put_old, &old);
2635 error = security_sb_pivotroot(&old, &new);
2639 get_fs_root(current->fs, &root);
2640 error = lock_mount(&old);
2645 new_mnt = real_mount(new.mnt);
2646 root_mnt = real_mount(root.mnt);
2647 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2648 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2649 IS_MNT_SHARED(root_mnt->mnt_parent))
2651 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2654 if (d_unlinked(new.dentry))
2656 if (d_unlinked(old.dentry))
2659 if (new.mnt == root.mnt ||
2660 old.mnt == root.mnt)
2661 goto out4; /* loop, on the same file system */
2663 if (root.mnt->mnt_root != root.dentry)
2664 goto out4; /* not a mountpoint */
2665 if (!mnt_has_parent(root_mnt))
2666 goto out4; /* not attached */
2667 if (new.mnt->mnt_root != new.dentry)
2668 goto out4; /* not a mountpoint */
2669 if (!mnt_has_parent(new_mnt))
2670 goto out4; /* not attached */
2671 /* make sure we can reach put_old from new_root */
2672 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2674 br_write_lock(vfsmount_lock);
2675 detach_mnt(new_mnt, &parent_path);
2676 detach_mnt(root_mnt, &root_parent);
2677 /* mount old root on put_old */
2678 attach_mnt(root_mnt, &old);
2679 /* mount new_root on / */
2680 attach_mnt(new_mnt, &root_parent);
2681 touch_mnt_namespace(current->nsproxy->mnt_ns);
2682 br_write_unlock(vfsmount_lock);
2683 chroot_fs_refs(&root, &new);
2688 path_put(&root_parent);
2689 path_put(&parent_path);
2701 static void __init init_mount_tree(void)
2703 struct vfsmount *mnt;
2704 struct mnt_namespace *ns;
2707 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2709 panic("Can't create rootfs");
2711 ns = create_mnt_ns(mnt);
2713 panic("Can't allocate initial namespace");
2715 init_task.nsproxy->mnt_ns = ns;
2718 root.mnt = ns->root;
2719 root.dentry = ns->root->mnt_root;
2721 set_fs_pwd(current->fs, &root);
2722 set_fs_root(current->fs, &root);
2725 void __init mnt_init(void)
2730 init_rwsem(&namespace_sem);
2732 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2733 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2735 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2737 if (!mount_hashtable)
2738 panic("Failed to allocate mount hash table\n");
2740 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2742 for (u = 0; u < HASH_SIZE; u++)
2743 INIT_LIST_HEAD(&mount_hashtable[u]);
2745 br_lock_init(vfsmount_lock);
2749 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2751 fs_kobj = kobject_create_and_add("fs", NULL);
2753 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2758 void put_mnt_ns(struct mnt_namespace *ns)
2760 LIST_HEAD(umount_list);
2762 if (!atomic_dec_and_test(&ns->count))
2764 down_write(&namespace_sem);
2765 br_write_lock(vfsmount_lock);
2766 umount_tree(real_mount(ns->root), 0, &umount_list);
2767 br_write_unlock(vfsmount_lock);
2768 up_write(&namespace_sem);
2769 release_mounts(&umount_list);
2773 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2775 struct vfsmount *mnt;
2776 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2779 * it is a longterm mount, don't release mnt until
2780 * we unmount before file sys is unregistered
2782 mnt_make_longterm(mnt);
2786 EXPORT_SYMBOL_GPL(kern_mount_data);
2788 void kern_unmount(struct vfsmount *mnt)
2790 /* release long term mount so mount point can be released */
2791 if (!IS_ERR_OR_NULL(mnt)) {
2792 mnt_make_shortterm(mnt);
2796 EXPORT_SYMBOL(kern_unmount);
2798 bool our_mnt(struct vfsmount *mnt)
2800 return check_mnt(real_mount(mnt));