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 vfsmount *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 vfsmount *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 vfsmount *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 vfsmount *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 vfsmount *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 vfsmount *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 vfsmount *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(mnt);
186 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
192 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
194 goto out_free_devname;
196 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
199 mnt->mnt_writers = 0;
202 INIT_LIST_HEAD(&mnt->mnt_hash);
203 INIT_LIST_HEAD(&mnt->mnt_child);
204 INIT_LIST_HEAD(&mnt->mnt_mounts);
205 INIT_LIST_HEAD(&mnt->mnt_list);
206 INIT_LIST_HEAD(&mnt->mnt_expire);
207 INIT_LIST_HEAD(&mnt->mnt_share);
208 INIT_LIST_HEAD(&mnt->mnt_slave_list);
209 INIT_LIST_HEAD(&mnt->mnt_slave);
210 #ifdef CONFIG_FSNOTIFY
211 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
218 kfree(p->mnt.mnt_devname);
221 mnt_free_id(&p->mnt);
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 vfsmount *mnt)
259 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
265 static inline void mnt_dec_writers(struct vfsmount *mnt)
268 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
274 static unsigned int mnt_get_writers(struct vfsmount *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 * @mnt: 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 *mnt)
313 mnt_inc_writers(mnt);
315 * The store to mnt_inc_writers must be visible before we pass
316 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
317 * incremented count after it has set MNT_WRITE_HOLD.
320 while (mnt->mnt_flags & MNT_WRITE_HOLD)
323 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
324 * be set to match its requirements. So we must not load that until
325 * MNT_WRITE_HOLD is cleared.
328 if (__mnt_is_readonly(mnt)) {
329 mnt_dec_writers(mnt);
337 EXPORT_SYMBOL_GPL(mnt_want_write);
340 * mnt_clone_write - get write access to a mount
341 * @mnt: the mount on which to take a write
343 * This is effectively like mnt_want_write, except
344 * it must only be used to take an extra write reference
345 * on a mountpoint that we already know has a write reference
346 * on it. This allows some optimisation.
348 * After finished, mnt_drop_write must be called as usual to
349 * drop the reference.
351 int mnt_clone_write(struct vfsmount *mnt)
353 /* superblock may be r/o */
354 if (__mnt_is_readonly(mnt))
357 mnt_inc_writers(mnt);
361 EXPORT_SYMBOL_GPL(mnt_clone_write);
364 * mnt_want_write_file - get write access to a file's mount
365 * @file: the file who's mount on which to take a write
367 * This is like mnt_want_write, but it takes a file and can
368 * do some optimisations if the file is open for write already
370 int mnt_want_write_file(struct file *file)
372 struct inode *inode = file->f_dentry->d_inode;
373 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
374 return mnt_want_write(file->f_path.mnt);
376 return mnt_clone_write(file->f_path.mnt);
378 EXPORT_SYMBOL_GPL(mnt_want_write_file);
381 * mnt_drop_write - give up write access to a mount
382 * @mnt: the mount on which to give up write access
384 * Tells the low-level filesystem that we are done
385 * performing writes to it. Must be matched with
386 * mnt_want_write() call above.
388 void mnt_drop_write(struct vfsmount *mnt)
391 mnt_dec_writers(mnt);
394 EXPORT_SYMBOL_GPL(mnt_drop_write);
396 void mnt_drop_write_file(struct file *file)
398 mnt_drop_write(file->f_path.mnt);
400 EXPORT_SYMBOL(mnt_drop_write_file);
402 static int mnt_make_readonly(struct vfsmount *mnt)
406 br_write_lock(vfsmount_lock);
407 mnt->mnt_flags |= MNT_WRITE_HOLD;
409 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
410 * should be visible before we do.
415 * With writers on hold, if this value is zero, then there are
416 * definitely no active writers (although held writers may subsequently
417 * increment the count, they'll have to wait, and decrement it after
418 * seeing MNT_READONLY).
420 * It is OK to have counter incremented on one CPU and decremented on
421 * another: the sum will add up correctly. The danger would be when we
422 * sum up each counter, if we read a counter before it is incremented,
423 * but then read another CPU's count which it has been subsequently
424 * decremented from -- we would see more decrements than we should.
425 * MNT_WRITE_HOLD protects against this scenario, because
426 * mnt_want_write first increments count, then smp_mb, then spins on
427 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
428 * we're counting up here.
430 if (mnt_get_writers(mnt) > 0)
433 mnt->mnt_flags |= MNT_READONLY;
435 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
436 * that become unheld will see MNT_READONLY.
439 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
440 br_write_unlock(vfsmount_lock);
444 static void __mnt_unmake_readonly(struct vfsmount *mnt)
446 br_write_lock(vfsmount_lock);
447 mnt->mnt_flags &= ~MNT_READONLY;
448 br_write_unlock(vfsmount_lock);
451 static void free_vfsmnt(struct vfsmount *mnt)
453 struct mount *p = real_mount(mnt);
454 kfree(mnt->mnt_devname);
457 free_percpu(mnt->mnt_pcp);
459 kmem_cache_free(mnt_cache, p);
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 vfsmount *__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 vfsmount *p, *found = NULL;
475 tmp = dir ? tmp->next : tmp->prev;
479 p = list_entry(tmp, struct vfsmount, mnt_hash);
480 if (p->mnt_parent == 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 vfsmount *child_mnt;
496 br_read_lock(vfsmount_lock);
497 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
499 br_read_unlock(vfsmount_lock);
503 static inline int check_mnt(struct vfsmount *mnt)
505 return mnt->mnt_ns == current->nsproxy->mnt_ns;
509 * vfsmount lock must be held for write
511 static void touch_mnt_namespace(struct mnt_namespace *ns)
515 wake_up_interruptible(&ns->poll);
520 * vfsmount lock must be held for write
522 static void __touch_mnt_namespace(struct mnt_namespace *ns)
524 if (ns && ns->event != event) {
526 wake_up_interruptible(&ns->poll);
531 * Clear dentry's mounted state if it has no remaining mounts.
532 * vfsmount_lock must be held for write.
534 static void dentry_reset_mounted(struct dentry *dentry)
538 for (u = 0; u < HASH_SIZE; u++) {
541 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
542 if (p->mnt_mountpoint == dentry)
546 spin_lock(&dentry->d_lock);
547 dentry->d_flags &= ~DCACHE_MOUNTED;
548 spin_unlock(&dentry->d_lock);
552 * vfsmount lock must be held for write
554 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
556 old_path->dentry = mnt->mnt_mountpoint;
557 old_path->mnt = mnt->mnt_parent;
558 mnt->mnt_parent = mnt;
559 mnt->mnt_mountpoint = mnt->mnt_root;
560 list_del_init(&mnt->mnt_child);
561 list_del_init(&mnt->mnt_hash);
562 dentry_reset_mounted(old_path->dentry);
566 * vfsmount lock must be held for write
568 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
569 struct vfsmount *child_mnt)
571 child_mnt->mnt_parent = mntget(mnt);
572 child_mnt->mnt_mountpoint = dget(dentry);
573 spin_lock(&dentry->d_lock);
574 dentry->d_flags |= DCACHE_MOUNTED;
575 spin_unlock(&dentry->d_lock);
579 * vfsmount lock must be held for write
581 static void attach_mnt(struct vfsmount *mnt, struct path *path)
583 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
584 list_add_tail(&mnt->mnt_hash, mount_hashtable +
585 hash(path->mnt, path->dentry));
586 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
589 static inline void __mnt_make_longterm(struct vfsmount *mnt)
592 atomic_inc(&mnt->mnt_longterm);
596 /* needs vfsmount lock for write */
597 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
600 atomic_dec(&mnt->mnt_longterm);
605 * vfsmount lock must be held for write
607 static void commit_tree(struct vfsmount *mnt)
609 struct vfsmount *parent = mnt->mnt_parent;
612 struct mnt_namespace *n = parent->mnt_ns;
614 BUG_ON(parent == mnt);
616 list_add_tail(&head, &mnt->mnt_list);
617 list_for_each_entry(m, &head, mnt_list) {
619 __mnt_make_longterm(m);
622 list_splice(&head, n->list.prev);
624 list_add_tail(&mnt->mnt_hash, mount_hashtable +
625 hash(parent, mnt->mnt_mountpoint));
626 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
627 touch_mnt_namespace(n);
630 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
632 struct list_head *next = p->mnt_mounts.next;
633 if (next == &p->mnt_mounts) {
637 next = p->mnt_child.next;
638 if (next != &p->mnt_parent->mnt_mounts)
643 return list_entry(next, struct vfsmount, mnt_child);
646 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
648 struct list_head *prev = p->mnt_mounts.prev;
649 while (prev != &p->mnt_mounts) {
650 p = list_entry(prev, struct vfsmount, mnt_child);
651 prev = p->mnt_mounts.prev;
657 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
659 struct vfsmount *mnt;
663 return ERR_PTR(-ENODEV);
665 mnt = alloc_vfsmnt(name);
667 return ERR_PTR(-ENOMEM);
669 if (flags & MS_KERNMOUNT)
670 mnt->mnt_flags = MNT_INTERNAL;
672 root = mount_fs(type, flags, name, data);
675 return ERR_CAST(root);
678 mnt->mnt_root = root;
679 mnt->mnt_sb = root->d_sb;
680 mnt->mnt_mountpoint = mnt->mnt_root;
681 mnt->mnt_parent = mnt;
684 EXPORT_SYMBOL_GPL(vfs_kern_mount);
686 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
689 struct super_block *sb = old->mnt_sb;
690 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
693 if (flag & (CL_SLAVE | CL_PRIVATE))
694 mnt->mnt_group_id = 0; /* not a peer of original */
696 mnt->mnt_group_id = old->mnt_group_id;
698 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
699 int err = mnt_alloc_group_id(mnt);
704 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
705 atomic_inc(&sb->s_active);
707 mnt->mnt_root = dget(root);
708 mnt->mnt_mountpoint = mnt->mnt_root;
709 mnt->mnt_parent = mnt;
711 if (flag & CL_SLAVE) {
712 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
713 mnt->mnt_master = old;
714 CLEAR_MNT_SHARED(mnt);
715 } else if (!(flag & CL_PRIVATE)) {
716 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
717 list_add(&mnt->mnt_share, &old->mnt_share);
718 if (IS_MNT_SLAVE(old))
719 list_add(&mnt->mnt_slave, &old->mnt_slave);
720 mnt->mnt_master = old->mnt_master;
722 if (flag & CL_MAKE_SHARED)
725 /* stick the duplicate mount on the same expiry list
726 * as the original if that was on one */
727 if (flag & CL_EXPIRE) {
728 if (!list_empty(&old->mnt_expire))
729 list_add(&mnt->mnt_expire, &old->mnt_expire);
739 static inline void mntfree(struct vfsmount *mnt)
741 struct super_block *sb = mnt->mnt_sb;
744 * This probably indicates that somebody messed
745 * up a mnt_want/drop_write() pair. If this
746 * happens, the filesystem was probably unable
747 * to make r/w->r/o transitions.
750 * The locking used to deal with mnt_count decrement provides barriers,
751 * so mnt_get_writers() below is safe.
753 WARN_ON(mnt_get_writers(mnt));
754 fsnotify_vfsmount_delete(mnt);
757 deactivate_super(sb);
760 static void mntput_no_expire(struct vfsmount *mnt)
764 br_read_lock(vfsmount_lock);
765 if (likely(atomic_read(&mnt->mnt_longterm))) {
766 mnt_add_count(mnt, -1);
767 br_read_unlock(vfsmount_lock);
770 br_read_unlock(vfsmount_lock);
772 br_write_lock(vfsmount_lock);
773 mnt_add_count(mnt, -1);
774 if (mnt_get_count(mnt)) {
775 br_write_unlock(vfsmount_lock);
779 mnt_add_count(mnt, -1);
780 if (likely(mnt_get_count(mnt)))
782 br_write_lock(vfsmount_lock);
784 if (unlikely(mnt->mnt_pinned)) {
785 mnt_add_count(mnt, mnt->mnt_pinned + 1);
787 br_write_unlock(vfsmount_lock);
788 acct_auto_close_mnt(mnt);
791 br_write_unlock(vfsmount_lock);
795 void mntput(struct vfsmount *mnt)
798 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
799 if (unlikely(mnt->mnt_expiry_mark))
800 mnt->mnt_expiry_mark = 0;
801 mntput_no_expire(mnt);
804 EXPORT_SYMBOL(mntput);
806 struct vfsmount *mntget(struct vfsmount *mnt)
809 mnt_add_count(mnt, 1);
812 EXPORT_SYMBOL(mntget);
814 void mnt_pin(struct vfsmount *mnt)
816 br_write_lock(vfsmount_lock);
818 br_write_unlock(vfsmount_lock);
820 EXPORT_SYMBOL(mnt_pin);
822 void mnt_unpin(struct vfsmount *mnt)
824 br_write_lock(vfsmount_lock);
825 if (mnt->mnt_pinned) {
826 mnt_add_count(mnt, 1);
829 br_write_unlock(vfsmount_lock);
831 EXPORT_SYMBOL(mnt_unpin);
833 static inline void mangle(struct seq_file *m, const char *s)
835 seq_escape(m, s, " \t\n\\");
839 * Simple .show_options callback for filesystems which don't want to
840 * implement more complex mount option showing.
842 * See also save_mount_options().
844 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
849 options = rcu_dereference(mnt->mnt_sb->s_options);
851 if (options != NULL && options[0]) {
859 EXPORT_SYMBOL(generic_show_options);
862 * If filesystem uses generic_show_options(), this function should be
863 * called from the fill_super() callback.
865 * The .remount_fs callback usually needs to be handled in a special
866 * way, to make sure, that previous options are not overwritten if the
869 * Also note, that if the filesystem's .remount_fs function doesn't
870 * reset all options to their default value, but changes only newly
871 * given options, then the displayed options will not reflect reality
874 void save_mount_options(struct super_block *sb, char *options)
876 BUG_ON(sb->s_options);
877 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
879 EXPORT_SYMBOL(save_mount_options);
881 void replace_mount_options(struct super_block *sb, char *options)
883 char *old = sb->s_options;
884 rcu_assign_pointer(sb->s_options, options);
890 EXPORT_SYMBOL(replace_mount_options);
892 #ifdef CONFIG_PROC_FS
894 static void *m_start(struct seq_file *m, loff_t *pos)
896 struct proc_mounts *p = m->private;
898 down_read(&namespace_sem);
899 return seq_list_start(&p->ns->list, *pos);
902 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
904 struct proc_mounts *p = m->private;
906 return seq_list_next(v, &p->ns->list, pos);
909 static void m_stop(struct seq_file *m, void *v)
911 up_read(&namespace_sem);
914 int mnt_had_events(struct proc_mounts *p)
916 struct mnt_namespace *ns = p->ns;
919 br_read_lock(vfsmount_lock);
920 if (p->m.poll_event != ns->event) {
921 p->m.poll_event = ns->event;
924 br_read_unlock(vfsmount_lock);
929 struct proc_fs_info {
934 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
936 static const struct proc_fs_info fs_info[] = {
937 { MS_SYNCHRONOUS, ",sync" },
938 { MS_DIRSYNC, ",dirsync" },
939 { MS_MANDLOCK, ",mand" },
942 const struct proc_fs_info *fs_infop;
944 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
945 if (sb->s_flags & fs_infop->flag)
946 seq_puts(m, fs_infop->str);
949 return security_sb_show_options(m, sb);
952 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
954 static const struct proc_fs_info mnt_info[] = {
955 { MNT_NOSUID, ",nosuid" },
956 { MNT_NODEV, ",nodev" },
957 { MNT_NOEXEC, ",noexec" },
958 { MNT_NOATIME, ",noatime" },
959 { MNT_NODIRATIME, ",nodiratime" },
960 { MNT_RELATIME, ",relatime" },
963 const struct proc_fs_info *fs_infop;
965 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
966 if (mnt->mnt_flags & fs_infop->flag)
967 seq_puts(m, fs_infop->str);
971 static void show_type(struct seq_file *m, struct super_block *sb)
973 mangle(m, sb->s_type->name);
974 if (sb->s_subtype && sb->s_subtype[0]) {
976 mangle(m, sb->s_subtype);
980 static int show_vfsmnt(struct seq_file *m, void *v)
982 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
984 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
986 if (mnt->mnt_sb->s_op->show_devname) {
987 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
991 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
994 seq_path(m, &mnt_path, " \t\n\\");
996 show_type(m, mnt->mnt_sb);
997 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
998 err = show_sb_opts(m, mnt->mnt_sb);
1001 show_mnt_opts(m, mnt);
1002 if (mnt->mnt_sb->s_op->show_options)
1003 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1004 seq_puts(m, " 0 0\n");
1009 const struct seq_operations mounts_op = {
1016 static int show_mountinfo(struct seq_file *m, void *v)
1018 struct proc_mounts *p = m->private;
1019 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1020 struct super_block *sb = mnt->mnt_sb;
1021 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1022 struct path root = p->root;
1025 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1026 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1027 if (sb->s_op->show_path)
1028 err = sb->s_op->show_path(m, mnt);
1030 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1035 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1036 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1040 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1041 show_mnt_opts(m, mnt);
1043 /* Tagged fields ("foo:X" or "bar") */
1044 if (IS_MNT_SHARED(mnt))
1045 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1046 if (IS_MNT_SLAVE(mnt)) {
1047 int master = mnt->mnt_master->mnt_group_id;
1048 int dom = get_dominating_id(mnt, &p->root);
1049 seq_printf(m, " master:%i", master);
1050 if (dom && dom != master)
1051 seq_printf(m, " propagate_from:%i", dom);
1053 if (IS_MNT_UNBINDABLE(mnt))
1054 seq_puts(m, " unbindable");
1056 /* Filesystem specific data */
1060 if (sb->s_op->show_devname)
1061 err = sb->s_op->show_devname(m, mnt);
1063 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1066 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1067 err = show_sb_opts(m, sb);
1070 if (sb->s_op->show_options)
1071 err = sb->s_op->show_options(m, mnt);
1077 const struct seq_operations mountinfo_op = {
1081 .show = show_mountinfo,
1084 static int show_vfsstat(struct seq_file *m, void *v)
1086 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1087 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1091 if (mnt->mnt_sb->s_op->show_devname) {
1092 seq_puts(m, "device ");
1093 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1095 if (mnt->mnt_devname) {
1096 seq_puts(m, "device ");
1097 mangle(m, mnt->mnt_devname);
1099 seq_puts(m, "no device");
1103 seq_puts(m, " mounted on ");
1104 seq_path(m, &mnt_path, " \t\n\\");
1107 /* file system type */
1108 seq_puts(m, "with fstype ");
1109 show_type(m, mnt->mnt_sb);
1111 /* optional statistics */
1112 if (mnt->mnt_sb->s_op->show_stats) {
1115 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1122 const struct seq_operations mountstats_op = {
1126 .show = show_vfsstat,
1128 #endif /* CONFIG_PROC_FS */
1131 * may_umount_tree - check if a mount tree is busy
1132 * @mnt: root of mount tree
1134 * This is called to check if a tree of mounts has any
1135 * open files, pwds, chroots or sub mounts that are
1138 int may_umount_tree(struct vfsmount *mnt)
1140 int actual_refs = 0;
1141 int minimum_refs = 0;
1144 /* write lock needed for mnt_get_count */
1145 br_write_lock(vfsmount_lock);
1146 for (p = mnt; p; p = next_mnt(p, mnt)) {
1147 actual_refs += mnt_get_count(p);
1150 br_write_unlock(vfsmount_lock);
1152 if (actual_refs > minimum_refs)
1158 EXPORT_SYMBOL(may_umount_tree);
1161 * may_umount - check if a mount point is busy
1162 * @mnt: root of mount
1164 * This is called to check if a mount point has any
1165 * open files, pwds, chroots or sub mounts. If the
1166 * mount has sub mounts this will return busy
1167 * regardless of whether the sub mounts are busy.
1169 * Doesn't take quota and stuff into account. IOW, in some cases it will
1170 * give false negatives. The main reason why it's here is that we need
1171 * a non-destructive way to look for easily umountable filesystems.
1173 int may_umount(struct vfsmount *mnt)
1176 down_read(&namespace_sem);
1177 br_write_lock(vfsmount_lock);
1178 if (propagate_mount_busy(mnt, 2))
1180 br_write_unlock(vfsmount_lock);
1181 up_read(&namespace_sem);
1185 EXPORT_SYMBOL(may_umount);
1187 void release_mounts(struct list_head *head)
1189 struct vfsmount *mnt;
1190 while (!list_empty(head)) {
1191 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1192 list_del_init(&mnt->mnt_hash);
1193 if (mnt_has_parent(mnt)) {
1194 struct dentry *dentry;
1197 br_write_lock(vfsmount_lock);
1198 dentry = mnt->mnt_mountpoint;
1199 m = mnt->mnt_parent;
1200 mnt->mnt_mountpoint = mnt->mnt_root;
1201 mnt->mnt_parent = mnt;
1203 br_write_unlock(vfsmount_lock);
1212 * vfsmount lock must be held for write
1213 * namespace_sem must be held for write
1215 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1217 LIST_HEAD(tmp_list);
1220 for (p = mnt; p; p = next_mnt(p, mnt))
1221 list_move(&p->mnt_hash, &tmp_list);
1224 propagate_umount(&tmp_list);
1226 list_for_each_entry(p, &tmp_list, mnt_hash) {
1227 list_del_init(&p->mnt_expire);
1228 list_del_init(&p->mnt_list);
1229 __touch_mnt_namespace(p->mnt_ns);
1231 __mnt_make_shortterm(p);
1232 list_del_init(&p->mnt_child);
1233 if (mnt_has_parent(p)) {
1234 p->mnt_parent->mnt_ghosts++;
1235 dentry_reset_mounted(p->mnt_mountpoint);
1237 change_mnt_propagation(p, MS_PRIVATE);
1239 list_splice(&tmp_list, kill);
1242 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1244 static int do_umount(struct vfsmount *mnt, int flags)
1246 struct super_block *sb = mnt->mnt_sb;
1248 LIST_HEAD(umount_list);
1250 retval = security_sb_umount(mnt, flags);
1255 * Allow userspace to request a mountpoint be expired rather than
1256 * unmounting unconditionally. Unmount only happens if:
1257 * (1) the mark is already set (the mark is cleared by mntput())
1258 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1260 if (flags & MNT_EXPIRE) {
1261 if (mnt == current->fs->root.mnt ||
1262 flags & (MNT_FORCE | MNT_DETACH))
1266 * probably don't strictly need the lock here if we examined
1267 * all race cases, but it's a slowpath.
1269 br_write_lock(vfsmount_lock);
1270 if (mnt_get_count(mnt) != 2) {
1271 br_write_unlock(vfsmount_lock);
1274 br_write_unlock(vfsmount_lock);
1276 if (!xchg(&mnt->mnt_expiry_mark, 1))
1281 * If we may have to abort operations to get out of this
1282 * mount, and they will themselves hold resources we must
1283 * allow the fs to do things. In the Unix tradition of
1284 * 'Gee thats tricky lets do it in userspace' the umount_begin
1285 * might fail to complete on the first run through as other tasks
1286 * must return, and the like. Thats for the mount program to worry
1287 * about for the moment.
1290 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1291 sb->s_op->umount_begin(sb);
1295 * No sense to grab the lock for this test, but test itself looks
1296 * somewhat bogus. Suggestions for better replacement?
1297 * Ho-hum... In principle, we might treat that as umount + switch
1298 * to rootfs. GC would eventually take care of the old vfsmount.
1299 * Actually it makes sense, especially if rootfs would contain a
1300 * /reboot - static binary that would close all descriptors and
1301 * call reboot(9). Then init(8) could umount root and exec /reboot.
1303 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1305 * Special case for "unmounting" root ...
1306 * we just try to remount it readonly.
1308 down_write(&sb->s_umount);
1309 if (!(sb->s_flags & MS_RDONLY))
1310 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1311 up_write(&sb->s_umount);
1315 down_write(&namespace_sem);
1316 br_write_lock(vfsmount_lock);
1319 if (!(flags & MNT_DETACH))
1320 shrink_submounts(mnt, &umount_list);
1323 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1324 if (!list_empty(&mnt->mnt_list))
1325 umount_tree(mnt, 1, &umount_list);
1328 br_write_unlock(vfsmount_lock);
1329 up_write(&namespace_sem);
1330 release_mounts(&umount_list);
1335 * Now umount can handle mount points as well as block devices.
1336 * This is important for filesystems which use unnamed block devices.
1338 * We now support a flag for forced unmount like the other 'big iron'
1339 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1342 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1346 int lookup_flags = 0;
1348 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1351 if (!(flags & UMOUNT_NOFOLLOW))
1352 lookup_flags |= LOOKUP_FOLLOW;
1354 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1358 if (path.dentry != path.mnt->mnt_root)
1360 if (!check_mnt(path.mnt))
1364 if (!capable(CAP_SYS_ADMIN))
1367 retval = do_umount(path.mnt, flags);
1369 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1371 mntput_no_expire(path.mnt);
1376 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1379 * The 2.0 compatible umount. No flags.
1381 SYSCALL_DEFINE1(oldumount, char __user *, name)
1383 return sys_umount(name, 0);
1388 static int mount_is_safe(struct path *path)
1390 if (capable(CAP_SYS_ADMIN))
1394 if (S_ISLNK(path->dentry->d_inode->i_mode))
1396 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1397 if (current_uid() != path->dentry->d_inode->i_uid)
1400 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1406 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1409 struct vfsmount *res, *p, *q, *r, *s;
1412 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1415 res = q = clone_mnt(mnt, dentry, flag);
1418 q->mnt_mountpoint = mnt->mnt_mountpoint;
1421 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1422 if (!is_subdir(r->mnt_mountpoint, dentry))
1425 for (s = r; s; s = next_mnt(s, r)) {
1426 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1427 s = skip_mnt_tree(s);
1430 while (p != s->mnt_parent) {
1436 path.dentry = p->mnt_mountpoint;
1437 q = clone_mnt(p, p->mnt_root, flag);
1440 br_write_lock(vfsmount_lock);
1441 list_add_tail(&q->mnt_list, &res->mnt_list);
1442 attach_mnt(q, &path);
1443 br_write_unlock(vfsmount_lock);
1449 LIST_HEAD(umount_list);
1450 br_write_lock(vfsmount_lock);
1451 umount_tree(res, 0, &umount_list);
1452 br_write_unlock(vfsmount_lock);
1453 release_mounts(&umount_list);
1458 struct vfsmount *collect_mounts(struct path *path)
1460 struct vfsmount *tree;
1461 down_write(&namespace_sem);
1462 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1463 up_write(&namespace_sem);
1467 void drop_collected_mounts(struct vfsmount *mnt)
1469 LIST_HEAD(umount_list);
1470 down_write(&namespace_sem);
1471 br_write_lock(vfsmount_lock);
1472 umount_tree(mnt, 0, &umount_list);
1473 br_write_unlock(vfsmount_lock);
1474 up_write(&namespace_sem);
1475 release_mounts(&umount_list);
1478 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1479 struct vfsmount *root)
1481 struct vfsmount *mnt;
1482 int res = f(root, arg);
1485 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1493 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1497 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1498 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1499 mnt_release_group_id(p);
1503 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1507 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1508 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1509 int err = mnt_alloc_group_id(p);
1511 cleanup_group_ids(mnt, p);
1521 * @source_mnt : mount tree to be attached
1522 * @nd : place the mount tree @source_mnt is attached
1523 * @parent_nd : if non-null, detach the source_mnt from its parent and
1524 * store the parent mount and mountpoint dentry.
1525 * (done when source_mnt is moved)
1527 * NOTE: in the table below explains the semantics when a source mount
1528 * of a given type is attached to a destination mount of a given type.
1529 * ---------------------------------------------------------------------------
1530 * | BIND MOUNT OPERATION |
1531 * |**************************************************************************
1532 * | source-->| shared | private | slave | unbindable |
1536 * |**************************************************************************
1537 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1539 * |non-shared| shared (+) | private | slave (*) | invalid |
1540 * ***************************************************************************
1541 * A bind operation clones the source mount and mounts the clone on the
1542 * destination mount.
1544 * (++) the cloned mount is propagated to all the mounts in the propagation
1545 * tree of the destination mount and the cloned mount is added to
1546 * the peer group of the source mount.
1547 * (+) the cloned mount is created under the destination mount and is marked
1548 * as shared. The cloned mount is added to the peer group of the source
1550 * (+++) the mount is propagated to all the mounts in the propagation tree
1551 * of the destination mount and the cloned mount is made slave
1552 * of the same master as that of the source mount. The cloned mount
1553 * is marked as 'shared and slave'.
1554 * (*) the cloned mount is made a slave of the same master as that of the
1557 * ---------------------------------------------------------------------------
1558 * | MOVE MOUNT OPERATION |
1559 * |**************************************************************************
1560 * | source-->| shared | private | slave | unbindable |
1564 * |**************************************************************************
1565 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1567 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1568 * ***************************************************************************
1570 * (+) the mount is moved to the destination. And is then propagated to
1571 * all the mounts in the propagation tree of the destination mount.
1572 * (+*) the mount is moved to the destination.
1573 * (+++) the mount is moved to the destination and is then propagated to
1574 * all the mounts belonging to the destination mount's propagation tree.
1575 * the mount is marked as 'shared and slave'.
1576 * (*) the mount continues to be a slave at the new location.
1578 * if the source mount is a tree, the operations explained above is
1579 * applied to each mount in the tree.
1580 * Must be called without spinlocks held, since this function can sleep
1583 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1584 struct path *path, struct path *parent_path)
1586 LIST_HEAD(tree_list);
1587 struct vfsmount *dest_mnt = path->mnt;
1588 struct dentry *dest_dentry = path->dentry;
1589 struct vfsmount *child, *p;
1592 if (IS_MNT_SHARED(dest_mnt)) {
1593 err = invent_group_ids(source_mnt, true);
1597 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1599 goto out_cleanup_ids;
1601 br_write_lock(vfsmount_lock);
1603 if (IS_MNT_SHARED(dest_mnt)) {
1604 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1608 detach_mnt(source_mnt, parent_path);
1609 attach_mnt(source_mnt, path);
1610 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1612 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1613 commit_tree(source_mnt);
1616 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1617 list_del_init(&child->mnt_hash);
1620 br_write_unlock(vfsmount_lock);
1625 if (IS_MNT_SHARED(dest_mnt))
1626 cleanup_group_ids(source_mnt, NULL);
1631 static int lock_mount(struct path *path)
1633 struct vfsmount *mnt;
1635 mutex_lock(&path->dentry->d_inode->i_mutex);
1636 if (unlikely(cant_mount(path->dentry))) {
1637 mutex_unlock(&path->dentry->d_inode->i_mutex);
1640 down_write(&namespace_sem);
1641 mnt = lookup_mnt(path);
1644 up_write(&namespace_sem);
1645 mutex_unlock(&path->dentry->d_inode->i_mutex);
1648 path->dentry = dget(mnt->mnt_root);
1652 static void unlock_mount(struct path *path)
1654 up_write(&namespace_sem);
1655 mutex_unlock(&path->dentry->d_inode->i_mutex);
1658 static int graft_tree(struct vfsmount *mnt, struct path *path)
1660 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1663 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1664 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1667 if (d_unlinked(path->dentry))
1670 return attach_recursive_mnt(mnt, path, NULL);
1674 * Sanity check the flags to change_mnt_propagation.
1677 static int flags_to_propagation_type(int flags)
1679 int type = flags & ~(MS_REC | MS_SILENT);
1681 /* Fail if any non-propagation flags are set */
1682 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1684 /* Only one propagation flag should be set */
1685 if (!is_power_of_2(type))
1691 * recursively change the type of the mountpoint.
1693 static int do_change_type(struct path *path, int flag)
1695 struct vfsmount *m, *mnt = path->mnt;
1696 int recurse = flag & MS_REC;
1700 if (!capable(CAP_SYS_ADMIN))
1703 if (path->dentry != path->mnt->mnt_root)
1706 type = flags_to_propagation_type(flag);
1710 down_write(&namespace_sem);
1711 if (type == MS_SHARED) {
1712 err = invent_group_ids(mnt, recurse);
1717 br_write_lock(vfsmount_lock);
1718 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1719 change_mnt_propagation(m, type);
1720 br_write_unlock(vfsmount_lock);
1723 up_write(&namespace_sem);
1728 * do loopback mount.
1730 static int do_loopback(struct path *path, char *old_name,
1733 LIST_HEAD(umount_list);
1734 struct path old_path;
1735 struct vfsmount *mnt = NULL;
1736 int err = mount_is_safe(path);
1739 if (!old_name || !*old_name)
1741 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1745 err = lock_mount(path);
1750 if (IS_MNT_UNBINDABLE(old_path.mnt))
1753 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1758 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1760 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1765 err = graft_tree(mnt, path);
1767 br_write_lock(vfsmount_lock);
1768 umount_tree(mnt, 0, &umount_list);
1769 br_write_unlock(vfsmount_lock);
1773 release_mounts(&umount_list);
1775 path_put(&old_path);
1779 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1782 int readonly_request = 0;
1784 if (ms_flags & MS_RDONLY)
1785 readonly_request = 1;
1786 if (readonly_request == __mnt_is_readonly(mnt))
1789 if (readonly_request)
1790 error = mnt_make_readonly(mnt);
1792 __mnt_unmake_readonly(mnt);
1797 * change filesystem flags. dir should be a physical root of filesystem.
1798 * If you've mounted a non-root directory somewhere and want to do remount
1799 * on it - tough luck.
1801 static int do_remount(struct path *path, int flags, int mnt_flags,
1805 struct super_block *sb = path->mnt->mnt_sb;
1807 if (!capable(CAP_SYS_ADMIN))
1810 if (!check_mnt(path->mnt))
1813 if (path->dentry != path->mnt->mnt_root)
1816 err = security_sb_remount(sb, data);
1820 down_write(&sb->s_umount);
1821 if (flags & MS_BIND)
1822 err = change_mount_flags(path->mnt, flags);
1824 err = do_remount_sb(sb, flags, data, 0);
1826 br_write_lock(vfsmount_lock);
1827 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1828 path->mnt->mnt_flags = mnt_flags;
1829 br_write_unlock(vfsmount_lock);
1831 up_write(&sb->s_umount);
1833 br_write_lock(vfsmount_lock);
1834 touch_mnt_namespace(path->mnt->mnt_ns);
1835 br_write_unlock(vfsmount_lock);
1840 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1843 for (p = mnt; p; p = next_mnt(p, mnt)) {
1844 if (IS_MNT_UNBINDABLE(p))
1850 static int do_move_mount(struct path *path, char *old_name)
1852 struct path old_path, parent_path;
1855 if (!capable(CAP_SYS_ADMIN))
1857 if (!old_name || !*old_name)
1859 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1863 err = lock_mount(path);
1868 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1871 if (d_unlinked(path->dentry))
1875 if (old_path.dentry != old_path.mnt->mnt_root)
1878 if (!mnt_has_parent(old_path.mnt))
1881 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1882 S_ISDIR(old_path.dentry->d_inode->i_mode))
1885 * Don't move a mount residing in a shared parent.
1887 if (IS_MNT_SHARED(old_path.mnt->mnt_parent))
1890 * Don't move a mount tree containing unbindable mounts to a destination
1891 * mount which is shared.
1893 if (IS_MNT_SHARED(path->mnt) &&
1894 tree_contains_unbindable(old_path.mnt))
1897 for (p = path->mnt; mnt_has_parent(p); p = p->mnt_parent)
1898 if (p == old_path.mnt)
1901 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1905 /* if the mount is moved, it should no longer be expire
1907 list_del_init(&old_path.mnt->mnt_expire);
1912 path_put(&parent_path);
1913 path_put(&old_path);
1917 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1920 const char *subtype = strchr(fstype, '.');
1929 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1931 if (!mnt->mnt_sb->s_subtype)
1937 return ERR_PTR(err);
1940 static struct vfsmount *
1941 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1943 struct file_system_type *type = get_fs_type(fstype);
1944 struct vfsmount *mnt;
1946 return ERR_PTR(-ENODEV);
1947 mnt = vfs_kern_mount(type, flags, name, data);
1948 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1949 !mnt->mnt_sb->s_subtype)
1950 mnt = fs_set_subtype(mnt, fstype);
1951 put_filesystem(type);
1956 * add a mount into a namespace's mount tree
1958 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1962 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1964 err = lock_mount(path);
1969 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1972 /* Refuse the same filesystem on the same mount point */
1974 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1975 path->mnt->mnt_root == path->dentry)
1979 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1982 newmnt->mnt_flags = mnt_flags;
1983 err = graft_tree(newmnt, path);
1991 * create a new mount for userspace and request it to be added into the
1994 static int do_new_mount(struct path *path, char *type, int flags,
1995 int mnt_flags, char *name, void *data)
1997 struct vfsmount *mnt;
2003 /* we need capabilities... */
2004 if (!capable(CAP_SYS_ADMIN))
2007 mnt = do_kern_mount(type, flags, name, data);
2009 return PTR_ERR(mnt);
2011 err = do_add_mount(mnt, path, mnt_flags);
2017 int finish_automount(struct vfsmount *m, struct path *path)
2020 /* The new mount record should have at least 2 refs to prevent it being
2021 * expired before we get a chance to add it
2023 BUG_ON(mnt_get_count(m) < 2);
2025 if (m->mnt_sb == path->mnt->mnt_sb &&
2026 m->mnt_root == path->dentry) {
2031 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2035 /* remove m from any expiration list it may be on */
2036 if (!list_empty(&m->mnt_expire)) {
2037 down_write(&namespace_sem);
2038 br_write_lock(vfsmount_lock);
2039 list_del_init(&m->mnt_expire);
2040 br_write_unlock(vfsmount_lock);
2041 up_write(&namespace_sem);
2049 * mnt_set_expiry - Put a mount on an expiration list
2050 * @mnt: The mount to list.
2051 * @expiry_list: The list to add the mount to.
2053 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2055 down_write(&namespace_sem);
2056 br_write_lock(vfsmount_lock);
2058 list_add_tail(&mnt->mnt_expire, expiry_list);
2060 br_write_unlock(vfsmount_lock);
2061 up_write(&namespace_sem);
2063 EXPORT_SYMBOL(mnt_set_expiry);
2066 * process a list of expirable mountpoints with the intent of discarding any
2067 * mountpoints that aren't in use and haven't been touched since last we came
2070 void mark_mounts_for_expiry(struct list_head *mounts)
2072 struct vfsmount *mnt, *next;
2073 LIST_HEAD(graveyard);
2076 if (list_empty(mounts))
2079 down_write(&namespace_sem);
2080 br_write_lock(vfsmount_lock);
2082 /* extract from the expiration list every vfsmount that matches the
2083 * following criteria:
2084 * - only referenced by its parent vfsmount
2085 * - still marked for expiry (marked on the last call here; marks are
2086 * cleared by mntput())
2088 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2089 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2090 propagate_mount_busy(mnt, 1))
2092 list_move(&mnt->mnt_expire, &graveyard);
2094 while (!list_empty(&graveyard)) {
2095 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2096 touch_mnt_namespace(mnt->mnt_ns);
2097 umount_tree(mnt, 1, &umounts);
2099 br_write_unlock(vfsmount_lock);
2100 up_write(&namespace_sem);
2102 release_mounts(&umounts);
2105 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2108 * Ripoff of 'select_parent()'
2110 * search the list of submounts for a given mountpoint, and move any
2111 * shrinkable submounts to the 'graveyard' list.
2113 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2115 struct vfsmount *this_parent = parent;
2116 struct list_head *next;
2120 next = this_parent->mnt_mounts.next;
2122 while (next != &this_parent->mnt_mounts) {
2123 struct list_head *tmp = next;
2124 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2127 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2130 * Descend a level if the d_mounts list is non-empty.
2132 if (!list_empty(&mnt->mnt_mounts)) {
2137 if (!propagate_mount_busy(mnt, 1)) {
2138 list_move_tail(&mnt->mnt_expire, graveyard);
2143 * All done at this level ... ascend and resume the search
2145 if (this_parent != parent) {
2146 next = this_parent->mnt_child.next;
2147 this_parent = this_parent->mnt_parent;
2154 * process a list of expirable mountpoints with the intent of discarding any
2155 * submounts of a specific parent mountpoint
2157 * vfsmount_lock must be held for write
2159 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2161 LIST_HEAD(graveyard);
2164 /* extract submounts of 'mountpoint' from the expiration list */
2165 while (select_submounts(mnt, &graveyard)) {
2166 while (!list_empty(&graveyard)) {
2167 m = list_first_entry(&graveyard, struct vfsmount,
2169 touch_mnt_namespace(m->mnt_ns);
2170 umount_tree(m, 1, umounts);
2176 * Some copy_from_user() implementations do not return the exact number of
2177 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2178 * Note that this function differs from copy_from_user() in that it will oops
2179 * on bad values of `to', rather than returning a short copy.
2181 static long exact_copy_from_user(void *to, const void __user * from,
2185 const char __user *f = from;
2188 if (!access_ok(VERIFY_READ, from, n))
2192 if (__get_user(c, f)) {
2203 int copy_mount_options(const void __user * data, unsigned long *where)
2213 if (!(page = __get_free_page(GFP_KERNEL)))
2216 /* We only care that *some* data at the address the user
2217 * gave us is valid. Just in case, we'll zero
2218 * the remainder of the page.
2220 /* copy_from_user cannot cross TASK_SIZE ! */
2221 size = TASK_SIZE - (unsigned long)data;
2222 if (size > PAGE_SIZE)
2225 i = size - exact_copy_from_user((void *)page, data, size);
2231 memset((char *)page + i, 0, PAGE_SIZE - i);
2236 int copy_mount_string(const void __user *data, char **where)
2245 tmp = strndup_user(data, PAGE_SIZE);
2247 return PTR_ERR(tmp);
2254 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2255 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2257 * data is a (void *) that can point to any structure up to
2258 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2259 * information (or be NULL).
2261 * Pre-0.97 versions of mount() didn't have a flags word.
2262 * When the flags word was introduced its top half was required
2263 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2264 * Therefore, if this magic number is present, it carries no information
2265 * and must be discarded.
2267 long do_mount(char *dev_name, char *dir_name, char *type_page,
2268 unsigned long flags, void *data_page)
2275 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2276 flags &= ~MS_MGC_MSK;
2278 /* Basic sanity checks */
2280 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2284 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2286 /* ... and get the mountpoint */
2287 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2291 retval = security_sb_mount(dev_name, &path,
2292 type_page, flags, data_page);
2296 /* Default to relatime unless overriden */
2297 if (!(flags & MS_NOATIME))
2298 mnt_flags |= MNT_RELATIME;
2300 /* Separate the per-mountpoint flags */
2301 if (flags & MS_NOSUID)
2302 mnt_flags |= MNT_NOSUID;
2303 if (flags & MS_NODEV)
2304 mnt_flags |= MNT_NODEV;
2305 if (flags & MS_NOEXEC)
2306 mnt_flags |= MNT_NOEXEC;
2307 if (flags & MS_NOATIME)
2308 mnt_flags |= MNT_NOATIME;
2309 if (flags & MS_NODIRATIME)
2310 mnt_flags |= MNT_NODIRATIME;
2311 if (flags & MS_STRICTATIME)
2312 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2313 if (flags & MS_RDONLY)
2314 mnt_flags |= MNT_READONLY;
2316 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2317 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2320 if (flags & MS_REMOUNT)
2321 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2323 else if (flags & MS_BIND)
2324 retval = do_loopback(&path, dev_name, flags & MS_REC);
2325 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2326 retval = do_change_type(&path, flags);
2327 else if (flags & MS_MOVE)
2328 retval = do_move_mount(&path, dev_name);
2330 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2331 dev_name, data_page);
2337 static struct mnt_namespace *alloc_mnt_ns(void)
2339 struct mnt_namespace *new_ns;
2341 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2343 return ERR_PTR(-ENOMEM);
2344 atomic_set(&new_ns->count, 1);
2345 new_ns->root = NULL;
2346 INIT_LIST_HEAD(&new_ns->list);
2347 init_waitqueue_head(&new_ns->poll);
2352 void mnt_make_longterm(struct vfsmount *mnt)
2354 __mnt_make_longterm(mnt);
2357 void mnt_make_shortterm(struct vfsmount *mnt)
2360 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2362 br_write_lock(vfsmount_lock);
2363 atomic_dec(&mnt->mnt_longterm);
2364 br_write_unlock(vfsmount_lock);
2369 * Allocate a new namespace structure and populate it with contents
2370 * copied from the namespace of the passed in task structure.
2372 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2373 struct fs_struct *fs)
2375 struct mnt_namespace *new_ns;
2376 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2377 struct vfsmount *p, *q;
2379 new_ns = alloc_mnt_ns();
2383 down_write(&namespace_sem);
2384 /* First pass: copy the tree topology */
2385 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2386 CL_COPY_ALL | CL_EXPIRE);
2387 if (!new_ns->root) {
2388 up_write(&namespace_sem);
2390 return ERR_PTR(-ENOMEM);
2392 br_write_lock(vfsmount_lock);
2393 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2394 br_write_unlock(vfsmount_lock);
2397 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2398 * as belonging to new namespace. We have already acquired a private
2399 * fs_struct, so tsk->fs->lock is not needed.
2405 __mnt_make_longterm(q);
2407 if (p == fs->root.mnt) {
2408 fs->root.mnt = mntget(q);
2409 __mnt_make_longterm(q);
2410 mnt_make_shortterm(p);
2413 if (p == fs->pwd.mnt) {
2414 fs->pwd.mnt = mntget(q);
2415 __mnt_make_longterm(q);
2416 mnt_make_shortterm(p);
2420 p = next_mnt(p, mnt_ns->root);
2421 q = next_mnt(q, new_ns->root);
2423 up_write(&namespace_sem);
2433 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2434 struct fs_struct *new_fs)
2436 struct mnt_namespace *new_ns;
2441 if (!(flags & CLONE_NEWNS))
2444 new_ns = dup_mnt_ns(ns, new_fs);
2451 * create_mnt_ns - creates a private namespace and adds a root filesystem
2452 * @mnt: pointer to the new root filesystem mountpoint
2454 static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2456 struct mnt_namespace *new_ns;
2458 new_ns = alloc_mnt_ns();
2459 if (!IS_ERR(new_ns)) {
2460 mnt->mnt_ns = new_ns;
2461 __mnt_make_longterm(mnt);
2463 list_add(&new_ns->list, &new_ns->root->mnt_list);
2470 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2472 struct mnt_namespace *ns;
2473 struct super_block *s;
2477 ns = create_mnt_ns(mnt);
2479 return ERR_CAST(ns);
2481 err = vfs_path_lookup(mnt->mnt_root, mnt,
2482 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2487 return ERR_PTR(err);
2489 /* trade a vfsmount reference for active sb one */
2490 s = path.mnt->mnt_sb;
2491 atomic_inc(&s->s_active);
2493 /* lock the sucker */
2494 down_write(&s->s_umount);
2495 /* ... and return the root of (sub)tree on it */
2498 EXPORT_SYMBOL(mount_subtree);
2500 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2501 char __user *, type, unsigned long, flags, void __user *, data)
2507 unsigned long data_page;
2509 ret = copy_mount_string(type, &kernel_type);
2513 kernel_dir = getname(dir_name);
2514 if (IS_ERR(kernel_dir)) {
2515 ret = PTR_ERR(kernel_dir);
2519 ret = copy_mount_string(dev_name, &kernel_dev);
2523 ret = copy_mount_options(data, &data_page);
2527 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2528 (void *) data_page);
2530 free_page(data_page);
2534 putname(kernel_dir);
2542 * Return true if path is reachable from root
2544 * namespace_sem or vfsmount_lock is held
2546 bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
2547 const struct path *root)
2549 while (mnt != root->mnt && mnt_has_parent(mnt)) {
2550 dentry = mnt->mnt_mountpoint;
2551 mnt = mnt->mnt_parent;
2553 return mnt == root->mnt && is_subdir(dentry, root->dentry);
2556 int path_is_under(struct path *path1, struct path *path2)
2559 br_read_lock(vfsmount_lock);
2560 res = is_path_reachable(path1->mnt, path1->dentry, path2);
2561 br_read_unlock(vfsmount_lock);
2564 EXPORT_SYMBOL(path_is_under);
2567 * pivot_root Semantics:
2568 * Moves the root file system of the current process to the directory put_old,
2569 * makes new_root as the new root file system of the current process, and sets
2570 * root/cwd of all processes which had them on the current root to new_root.
2573 * The new_root and put_old must be directories, and must not be on the
2574 * same file system as the current process root. The put_old must be
2575 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2576 * pointed to by put_old must yield the same directory as new_root. No other
2577 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2579 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2580 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2581 * in this situation.
2584 * - we don't move root/cwd if they are not at the root (reason: if something
2585 * cared enough to change them, it's probably wrong to force them elsewhere)
2586 * - it's okay to pick a root that isn't the root of a file system, e.g.
2587 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2588 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2591 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2592 const char __user *, put_old)
2594 struct path new, old, parent_path, root_parent, root;
2597 if (!capable(CAP_SYS_ADMIN))
2600 error = user_path_dir(new_root, &new);
2604 error = user_path_dir(put_old, &old);
2608 error = security_sb_pivotroot(&old, &new);
2612 get_fs_root(current->fs, &root);
2613 error = lock_mount(&old);
2618 if (IS_MNT_SHARED(old.mnt) ||
2619 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2620 IS_MNT_SHARED(root.mnt->mnt_parent))
2622 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2625 if (d_unlinked(new.dentry))
2627 if (d_unlinked(old.dentry))
2630 if (new.mnt == root.mnt ||
2631 old.mnt == root.mnt)
2632 goto out4; /* loop, on the same file system */
2634 if (root.mnt->mnt_root != root.dentry)
2635 goto out4; /* not a mountpoint */
2636 if (!mnt_has_parent(root.mnt))
2637 goto out4; /* not attached */
2638 if (new.mnt->mnt_root != new.dentry)
2639 goto out4; /* not a mountpoint */
2640 if (!mnt_has_parent(new.mnt))
2641 goto out4; /* not attached */
2642 /* make sure we can reach put_old from new_root */
2643 if (!is_path_reachable(old.mnt, old.dentry, &new))
2645 br_write_lock(vfsmount_lock);
2646 detach_mnt(new.mnt, &parent_path);
2647 detach_mnt(root.mnt, &root_parent);
2648 /* mount old root on put_old */
2649 attach_mnt(root.mnt, &old);
2650 /* mount new_root on / */
2651 attach_mnt(new.mnt, &root_parent);
2652 touch_mnt_namespace(current->nsproxy->mnt_ns);
2653 br_write_unlock(vfsmount_lock);
2654 chroot_fs_refs(&root, &new);
2659 path_put(&root_parent);
2660 path_put(&parent_path);
2672 static void __init init_mount_tree(void)
2674 struct vfsmount *mnt;
2675 struct mnt_namespace *ns;
2678 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2680 panic("Can't create rootfs");
2682 ns = create_mnt_ns(mnt);
2684 panic("Can't allocate initial namespace");
2686 init_task.nsproxy->mnt_ns = ns;
2689 root.mnt = ns->root;
2690 root.dentry = ns->root->mnt_root;
2692 set_fs_pwd(current->fs, &root);
2693 set_fs_root(current->fs, &root);
2696 void __init mnt_init(void)
2701 init_rwsem(&namespace_sem);
2703 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2704 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2706 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2708 if (!mount_hashtable)
2709 panic("Failed to allocate mount hash table\n");
2711 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2713 for (u = 0; u < HASH_SIZE; u++)
2714 INIT_LIST_HEAD(&mount_hashtable[u]);
2716 br_lock_init(vfsmount_lock);
2720 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2722 fs_kobj = kobject_create_and_add("fs", NULL);
2724 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2729 void put_mnt_ns(struct mnt_namespace *ns)
2731 LIST_HEAD(umount_list);
2733 if (!atomic_dec_and_test(&ns->count))
2735 down_write(&namespace_sem);
2736 br_write_lock(vfsmount_lock);
2737 umount_tree(ns->root, 0, &umount_list);
2738 br_write_unlock(vfsmount_lock);
2739 up_write(&namespace_sem);
2740 release_mounts(&umount_list);
2744 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2746 struct vfsmount *mnt;
2747 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2750 * it is a longterm mount, don't release mnt until
2751 * we unmount before file sys is unregistered
2753 mnt_make_longterm(mnt);
2757 EXPORT_SYMBOL_GPL(kern_mount_data);
2759 void kern_unmount(struct vfsmount *mnt)
2761 /* release long term mount so mount point can be released */
2762 if (!IS_ERR_OR_NULL(mnt)) {
2763 mnt_make_shortterm(mnt);
2767 EXPORT_SYMBOL(kern_unmount);
2769 bool our_mnt(struct vfsmount *mnt)
2771 return check_mnt(mnt);