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 vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
180 err = mnt_alloc_id(mnt);
185 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
186 if (!mnt->mnt_devname)
191 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 goto out_free_devname;
195 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
198 mnt->mnt_writers = 0;
201 INIT_LIST_HEAD(&mnt->mnt_hash);
202 INIT_LIST_HEAD(&mnt->mnt_child);
203 INIT_LIST_HEAD(&mnt->mnt_mounts);
204 INIT_LIST_HEAD(&mnt->mnt_list);
205 INIT_LIST_HEAD(&mnt->mnt_expire);
206 INIT_LIST_HEAD(&mnt->mnt_share);
207 INIT_LIST_HEAD(&mnt->mnt_slave_list);
208 INIT_LIST_HEAD(&mnt->mnt_slave);
209 #ifdef CONFIG_FSNOTIFY
210 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
217 kfree(mnt->mnt_devname);
222 kmem_cache_free(mnt_cache, mnt);
227 * Most r/o checks on a fs are for operations that take
228 * discrete amounts of time, like a write() or unlink().
229 * We must keep track of when those operations start
230 * (for permission checks) and when they end, so that
231 * we can determine when writes are able to occur to
235 * __mnt_is_readonly: check whether a mount is read-only
236 * @mnt: the mount to check for its write status
238 * This shouldn't be used directly ouside of the VFS.
239 * It does not guarantee that the filesystem will stay
240 * r/w, just that it is right *now*. This can not and
241 * should not be used in place of IS_RDONLY(inode).
242 * mnt_want/drop_write() will _keep_ the filesystem
245 int __mnt_is_readonly(struct vfsmount *mnt)
247 if (mnt->mnt_flags & MNT_READONLY)
249 if (mnt->mnt_sb->s_flags & MS_RDONLY)
253 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
255 static inline void mnt_inc_writers(struct vfsmount *mnt)
258 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
264 static inline void mnt_dec_writers(struct vfsmount *mnt)
267 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
273 static unsigned int mnt_get_writers(struct vfsmount *mnt)
276 unsigned int count = 0;
279 for_each_possible_cpu(cpu) {
280 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
285 return mnt->mnt_writers;
290 * Most r/o checks on a fs are for operations that take
291 * discrete amounts of time, like a write() or unlink().
292 * We must keep track of when those operations start
293 * (for permission checks) and when they end, so that
294 * we can determine when writes are able to occur to
298 * mnt_want_write - get write access to a mount
299 * @mnt: the mount on which to take a write
301 * This tells the low-level filesystem that a write is
302 * about to be performed to it, and makes sure that
303 * writes are allowed before returning success. When
304 * the write operation is finished, mnt_drop_write()
305 * must be called. This is effectively a refcount.
307 int mnt_want_write(struct vfsmount *mnt)
312 mnt_inc_writers(mnt);
314 * The store to mnt_inc_writers must be visible before we pass
315 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
316 * incremented count after it has set MNT_WRITE_HOLD.
319 while (mnt->mnt_flags & MNT_WRITE_HOLD)
322 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
323 * be set to match its requirements. So we must not load that until
324 * MNT_WRITE_HOLD is cleared.
327 if (__mnt_is_readonly(mnt)) {
328 mnt_dec_writers(mnt);
336 EXPORT_SYMBOL_GPL(mnt_want_write);
339 * mnt_clone_write - get write access to a mount
340 * @mnt: the mount on which to take a write
342 * This is effectively like mnt_want_write, except
343 * it must only be used to take an extra write reference
344 * on a mountpoint that we already know has a write reference
345 * on it. This allows some optimisation.
347 * After finished, mnt_drop_write must be called as usual to
348 * drop the reference.
350 int mnt_clone_write(struct vfsmount *mnt)
352 /* superblock may be r/o */
353 if (__mnt_is_readonly(mnt))
356 mnt_inc_writers(mnt);
360 EXPORT_SYMBOL_GPL(mnt_clone_write);
363 * mnt_want_write_file - get write access to a file's mount
364 * @file: the file who's mount on which to take a write
366 * This is like mnt_want_write, but it takes a file and can
367 * do some optimisations if the file is open for write already
369 int mnt_want_write_file(struct file *file)
371 struct inode *inode = file->f_dentry->d_inode;
372 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
373 return mnt_want_write(file->f_path.mnt);
375 return mnt_clone_write(file->f_path.mnt);
377 EXPORT_SYMBOL_GPL(mnt_want_write_file);
380 * mnt_drop_write - give up write access to a mount
381 * @mnt: the mount on which to give up write access
383 * Tells the low-level filesystem that we are done
384 * performing writes to it. Must be matched with
385 * mnt_want_write() call above.
387 void mnt_drop_write(struct vfsmount *mnt)
390 mnt_dec_writers(mnt);
393 EXPORT_SYMBOL_GPL(mnt_drop_write);
395 void mnt_drop_write_file(struct file *file)
397 mnt_drop_write(file->f_path.mnt);
399 EXPORT_SYMBOL(mnt_drop_write_file);
401 static int mnt_make_readonly(struct vfsmount *mnt)
405 br_write_lock(vfsmount_lock);
406 mnt->mnt_flags |= MNT_WRITE_HOLD;
408 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
409 * should be visible before we do.
414 * With writers on hold, if this value is zero, then there are
415 * definitely no active writers (although held writers may subsequently
416 * increment the count, they'll have to wait, and decrement it after
417 * seeing MNT_READONLY).
419 * It is OK to have counter incremented on one CPU and decremented on
420 * another: the sum will add up correctly. The danger would be when we
421 * sum up each counter, if we read a counter before it is incremented,
422 * but then read another CPU's count which it has been subsequently
423 * decremented from -- we would see more decrements than we should.
424 * MNT_WRITE_HOLD protects against this scenario, because
425 * mnt_want_write first increments count, then smp_mb, then spins on
426 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
427 * we're counting up here.
429 if (mnt_get_writers(mnt) > 0)
432 mnt->mnt_flags |= MNT_READONLY;
434 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
435 * that become unheld will see MNT_READONLY.
438 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
439 br_write_unlock(vfsmount_lock);
443 static void __mnt_unmake_readonly(struct vfsmount *mnt)
445 br_write_lock(vfsmount_lock);
446 mnt->mnt_flags &= ~MNT_READONLY;
447 br_write_unlock(vfsmount_lock);
450 static void free_vfsmnt(struct vfsmount *mnt)
452 kfree(mnt->mnt_devname);
455 free_percpu(mnt->mnt_pcp);
457 kmem_cache_free(mnt_cache, mnt);
461 * find the first or last mount at @dentry on vfsmount @mnt depending on
462 * @dir. If @dir is set return the first mount else return the last mount.
463 * vfsmount_lock must be held for read or write.
465 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
468 struct list_head *head = mount_hashtable + hash(mnt, dentry);
469 struct list_head *tmp = head;
470 struct vfsmount *p, *found = NULL;
473 tmp = dir ? tmp->next : tmp->prev;
477 p = list_entry(tmp, struct vfsmount, mnt_hash);
478 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
487 * lookup_mnt increments the ref count before returning
488 * the vfsmount struct.
490 struct vfsmount *lookup_mnt(struct path *path)
492 struct vfsmount *child_mnt;
494 br_read_lock(vfsmount_lock);
495 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
497 br_read_unlock(vfsmount_lock);
501 static inline int check_mnt(struct vfsmount *mnt)
503 return mnt->mnt_ns == current->nsproxy->mnt_ns;
507 * vfsmount lock must be held for write
509 static void touch_mnt_namespace(struct mnt_namespace *ns)
513 wake_up_interruptible(&ns->poll);
518 * vfsmount lock must be held for write
520 static void __touch_mnt_namespace(struct mnt_namespace *ns)
522 if (ns && ns->event != event) {
524 wake_up_interruptible(&ns->poll);
529 * Clear dentry's mounted state if it has no remaining mounts.
530 * vfsmount_lock must be held for write.
532 static void dentry_reset_mounted(struct dentry *dentry)
536 for (u = 0; u < HASH_SIZE; u++) {
539 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
540 if (p->mnt_mountpoint == dentry)
544 spin_lock(&dentry->d_lock);
545 dentry->d_flags &= ~DCACHE_MOUNTED;
546 spin_unlock(&dentry->d_lock);
550 * vfsmount lock must be held for write
552 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
554 old_path->dentry = mnt->mnt_mountpoint;
555 old_path->mnt = mnt->mnt_parent;
556 mnt->mnt_parent = mnt;
557 mnt->mnt_mountpoint = mnt->mnt_root;
558 list_del_init(&mnt->mnt_child);
559 list_del_init(&mnt->mnt_hash);
560 dentry_reset_mounted(old_path->dentry);
564 * vfsmount lock must be held for write
566 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
567 struct vfsmount *child_mnt)
569 child_mnt->mnt_parent = mntget(mnt);
570 child_mnt->mnt_mountpoint = dget(dentry);
571 spin_lock(&dentry->d_lock);
572 dentry->d_flags |= DCACHE_MOUNTED;
573 spin_unlock(&dentry->d_lock);
577 * vfsmount lock must be held for write
579 static void attach_mnt(struct vfsmount *mnt, struct path *path)
581 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
582 list_add_tail(&mnt->mnt_hash, mount_hashtable +
583 hash(path->mnt, path->dentry));
584 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
587 static inline void __mnt_make_longterm(struct vfsmount *mnt)
590 atomic_inc(&mnt->mnt_longterm);
594 /* needs vfsmount lock for write */
595 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
598 atomic_dec(&mnt->mnt_longterm);
603 * vfsmount lock must be held for write
605 static void commit_tree(struct vfsmount *mnt)
607 struct vfsmount *parent = mnt->mnt_parent;
610 struct mnt_namespace *n = parent->mnt_ns;
612 BUG_ON(parent == mnt);
614 list_add_tail(&head, &mnt->mnt_list);
615 list_for_each_entry(m, &head, mnt_list) {
617 __mnt_make_longterm(m);
620 list_splice(&head, n->list.prev);
622 list_add_tail(&mnt->mnt_hash, mount_hashtable +
623 hash(parent, mnt->mnt_mountpoint));
624 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
625 touch_mnt_namespace(n);
628 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
630 struct list_head *next = p->mnt_mounts.next;
631 if (next == &p->mnt_mounts) {
635 next = p->mnt_child.next;
636 if (next != &p->mnt_parent->mnt_mounts)
641 return list_entry(next, struct vfsmount, mnt_child);
644 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
646 struct list_head *prev = p->mnt_mounts.prev;
647 while (prev != &p->mnt_mounts) {
648 p = list_entry(prev, struct vfsmount, mnt_child);
649 prev = p->mnt_mounts.prev;
655 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
657 struct vfsmount *mnt;
661 return ERR_PTR(-ENODEV);
663 mnt = alloc_vfsmnt(name);
665 return ERR_PTR(-ENOMEM);
667 if (flags & MS_KERNMOUNT)
668 mnt->mnt_flags = MNT_INTERNAL;
670 root = mount_fs(type, flags, name, data);
673 return ERR_CAST(root);
676 mnt->mnt_root = root;
677 mnt->mnt_sb = root->d_sb;
678 mnt->mnt_mountpoint = mnt->mnt_root;
679 mnt->mnt_parent = mnt;
682 EXPORT_SYMBOL_GPL(vfs_kern_mount);
684 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
687 struct super_block *sb = old->mnt_sb;
688 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
691 if (flag & (CL_SLAVE | CL_PRIVATE))
692 mnt->mnt_group_id = 0; /* not a peer of original */
694 mnt->mnt_group_id = old->mnt_group_id;
696 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
697 int err = mnt_alloc_group_id(mnt);
702 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
703 atomic_inc(&sb->s_active);
705 mnt->mnt_root = dget(root);
706 mnt->mnt_mountpoint = mnt->mnt_root;
707 mnt->mnt_parent = mnt;
709 if (flag & CL_SLAVE) {
710 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
711 mnt->mnt_master = old;
712 CLEAR_MNT_SHARED(mnt);
713 } else if (!(flag & CL_PRIVATE)) {
714 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
715 list_add(&mnt->mnt_share, &old->mnt_share);
716 if (IS_MNT_SLAVE(old))
717 list_add(&mnt->mnt_slave, &old->mnt_slave);
718 mnt->mnt_master = old->mnt_master;
720 if (flag & CL_MAKE_SHARED)
723 /* stick the duplicate mount on the same expiry list
724 * as the original if that was on one */
725 if (flag & CL_EXPIRE) {
726 if (!list_empty(&old->mnt_expire))
727 list_add(&mnt->mnt_expire, &old->mnt_expire);
737 static inline void mntfree(struct vfsmount *mnt)
739 struct super_block *sb = mnt->mnt_sb;
742 * This probably indicates that somebody messed
743 * up a mnt_want/drop_write() pair. If this
744 * happens, the filesystem was probably unable
745 * to make r/w->r/o transitions.
748 * The locking used to deal with mnt_count decrement provides barriers,
749 * so mnt_get_writers() below is safe.
751 WARN_ON(mnt_get_writers(mnt));
752 fsnotify_vfsmount_delete(mnt);
755 deactivate_super(sb);
758 static void mntput_no_expire(struct vfsmount *mnt)
762 br_read_lock(vfsmount_lock);
763 if (likely(atomic_read(&mnt->mnt_longterm))) {
764 mnt_add_count(mnt, -1);
765 br_read_unlock(vfsmount_lock);
768 br_read_unlock(vfsmount_lock);
770 br_write_lock(vfsmount_lock);
771 mnt_add_count(mnt, -1);
772 if (mnt_get_count(mnt)) {
773 br_write_unlock(vfsmount_lock);
777 mnt_add_count(mnt, -1);
778 if (likely(mnt_get_count(mnt)))
780 br_write_lock(vfsmount_lock);
782 if (unlikely(mnt->mnt_pinned)) {
783 mnt_add_count(mnt, mnt->mnt_pinned + 1);
785 br_write_unlock(vfsmount_lock);
786 acct_auto_close_mnt(mnt);
789 br_write_unlock(vfsmount_lock);
793 void mntput(struct vfsmount *mnt)
796 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
797 if (unlikely(mnt->mnt_expiry_mark))
798 mnt->mnt_expiry_mark = 0;
799 mntput_no_expire(mnt);
802 EXPORT_SYMBOL(mntput);
804 struct vfsmount *mntget(struct vfsmount *mnt)
807 mnt_add_count(mnt, 1);
810 EXPORT_SYMBOL(mntget);
812 void mnt_pin(struct vfsmount *mnt)
814 br_write_lock(vfsmount_lock);
816 br_write_unlock(vfsmount_lock);
818 EXPORT_SYMBOL(mnt_pin);
820 void mnt_unpin(struct vfsmount *mnt)
822 br_write_lock(vfsmount_lock);
823 if (mnt->mnt_pinned) {
824 mnt_add_count(mnt, 1);
827 br_write_unlock(vfsmount_lock);
829 EXPORT_SYMBOL(mnt_unpin);
831 static inline void mangle(struct seq_file *m, const char *s)
833 seq_escape(m, s, " \t\n\\");
837 * Simple .show_options callback for filesystems which don't want to
838 * implement more complex mount option showing.
840 * See also save_mount_options().
842 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
847 options = rcu_dereference(mnt->mnt_sb->s_options);
849 if (options != NULL && options[0]) {
857 EXPORT_SYMBOL(generic_show_options);
860 * If filesystem uses generic_show_options(), this function should be
861 * called from the fill_super() callback.
863 * The .remount_fs callback usually needs to be handled in a special
864 * way, to make sure, that previous options are not overwritten if the
867 * Also note, that if the filesystem's .remount_fs function doesn't
868 * reset all options to their default value, but changes only newly
869 * given options, then the displayed options will not reflect reality
872 void save_mount_options(struct super_block *sb, char *options)
874 BUG_ON(sb->s_options);
875 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
877 EXPORT_SYMBOL(save_mount_options);
879 void replace_mount_options(struct super_block *sb, char *options)
881 char *old = sb->s_options;
882 rcu_assign_pointer(sb->s_options, options);
888 EXPORT_SYMBOL(replace_mount_options);
890 #ifdef CONFIG_PROC_FS
892 static void *m_start(struct seq_file *m, loff_t *pos)
894 struct proc_mounts *p = m->private;
896 down_read(&namespace_sem);
897 return seq_list_start(&p->ns->list, *pos);
900 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
902 struct proc_mounts *p = m->private;
904 return seq_list_next(v, &p->ns->list, pos);
907 static void m_stop(struct seq_file *m, void *v)
909 up_read(&namespace_sem);
912 int mnt_had_events(struct proc_mounts *p)
914 struct mnt_namespace *ns = p->ns;
917 br_read_lock(vfsmount_lock);
918 if (p->m.poll_event != ns->event) {
919 p->m.poll_event = ns->event;
922 br_read_unlock(vfsmount_lock);
927 struct proc_fs_info {
932 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
934 static const struct proc_fs_info fs_info[] = {
935 { MS_SYNCHRONOUS, ",sync" },
936 { MS_DIRSYNC, ",dirsync" },
937 { MS_MANDLOCK, ",mand" },
940 const struct proc_fs_info *fs_infop;
942 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
943 if (sb->s_flags & fs_infop->flag)
944 seq_puts(m, fs_infop->str);
947 return security_sb_show_options(m, sb);
950 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
952 static const struct proc_fs_info mnt_info[] = {
953 { MNT_NOSUID, ",nosuid" },
954 { MNT_NODEV, ",nodev" },
955 { MNT_NOEXEC, ",noexec" },
956 { MNT_NOATIME, ",noatime" },
957 { MNT_NODIRATIME, ",nodiratime" },
958 { MNT_RELATIME, ",relatime" },
961 const struct proc_fs_info *fs_infop;
963 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
964 if (mnt->mnt_flags & fs_infop->flag)
965 seq_puts(m, fs_infop->str);
969 static void show_type(struct seq_file *m, struct super_block *sb)
971 mangle(m, sb->s_type->name);
972 if (sb->s_subtype && sb->s_subtype[0]) {
974 mangle(m, sb->s_subtype);
978 static int show_vfsmnt(struct seq_file *m, void *v)
980 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
982 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
984 if (mnt->mnt_sb->s_op->show_devname) {
985 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
989 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
992 seq_path(m, &mnt_path, " \t\n\\");
994 show_type(m, mnt->mnt_sb);
995 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
996 err = show_sb_opts(m, mnt->mnt_sb);
999 show_mnt_opts(m, mnt);
1000 if (mnt->mnt_sb->s_op->show_options)
1001 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1002 seq_puts(m, " 0 0\n");
1007 const struct seq_operations mounts_op = {
1014 static int show_mountinfo(struct seq_file *m, void *v)
1016 struct proc_mounts *p = m->private;
1017 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1018 struct super_block *sb = mnt->mnt_sb;
1019 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1020 struct path root = p->root;
1023 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1024 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1025 if (sb->s_op->show_path)
1026 err = sb->s_op->show_path(m, mnt);
1028 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1033 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1034 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1038 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1039 show_mnt_opts(m, mnt);
1041 /* Tagged fields ("foo:X" or "bar") */
1042 if (IS_MNT_SHARED(mnt))
1043 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1044 if (IS_MNT_SLAVE(mnt)) {
1045 int master = mnt->mnt_master->mnt_group_id;
1046 int dom = get_dominating_id(mnt, &p->root);
1047 seq_printf(m, " master:%i", master);
1048 if (dom && dom != master)
1049 seq_printf(m, " propagate_from:%i", dom);
1051 if (IS_MNT_UNBINDABLE(mnt))
1052 seq_puts(m, " unbindable");
1054 /* Filesystem specific data */
1058 if (sb->s_op->show_devname)
1059 err = sb->s_op->show_devname(m, mnt);
1061 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1064 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1065 err = show_sb_opts(m, sb);
1068 if (sb->s_op->show_options)
1069 err = sb->s_op->show_options(m, mnt);
1075 const struct seq_operations mountinfo_op = {
1079 .show = show_mountinfo,
1082 static int show_vfsstat(struct seq_file *m, void *v)
1084 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1085 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1089 if (mnt->mnt_sb->s_op->show_devname) {
1090 seq_puts(m, "device ");
1091 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1093 if (mnt->mnt_devname) {
1094 seq_puts(m, "device ");
1095 mangle(m, mnt->mnt_devname);
1097 seq_puts(m, "no device");
1101 seq_puts(m, " mounted on ");
1102 seq_path(m, &mnt_path, " \t\n\\");
1105 /* file system type */
1106 seq_puts(m, "with fstype ");
1107 show_type(m, mnt->mnt_sb);
1109 /* optional statistics */
1110 if (mnt->mnt_sb->s_op->show_stats) {
1113 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1120 const struct seq_operations mountstats_op = {
1124 .show = show_vfsstat,
1126 #endif /* CONFIG_PROC_FS */
1129 * may_umount_tree - check if a mount tree is busy
1130 * @mnt: root of mount tree
1132 * This is called to check if a tree of mounts has any
1133 * open files, pwds, chroots or sub mounts that are
1136 int may_umount_tree(struct vfsmount *mnt)
1138 int actual_refs = 0;
1139 int minimum_refs = 0;
1142 /* write lock needed for mnt_get_count */
1143 br_write_lock(vfsmount_lock);
1144 for (p = mnt; p; p = next_mnt(p, mnt)) {
1145 actual_refs += mnt_get_count(p);
1148 br_write_unlock(vfsmount_lock);
1150 if (actual_refs > minimum_refs)
1156 EXPORT_SYMBOL(may_umount_tree);
1159 * may_umount - check if a mount point is busy
1160 * @mnt: root of mount
1162 * This is called to check if a mount point has any
1163 * open files, pwds, chroots or sub mounts. If the
1164 * mount has sub mounts this will return busy
1165 * regardless of whether the sub mounts are busy.
1167 * Doesn't take quota and stuff into account. IOW, in some cases it will
1168 * give false negatives. The main reason why it's here is that we need
1169 * a non-destructive way to look for easily umountable filesystems.
1171 int may_umount(struct vfsmount *mnt)
1174 down_read(&namespace_sem);
1175 br_write_lock(vfsmount_lock);
1176 if (propagate_mount_busy(mnt, 2))
1178 br_write_unlock(vfsmount_lock);
1179 up_read(&namespace_sem);
1183 EXPORT_SYMBOL(may_umount);
1185 void release_mounts(struct list_head *head)
1187 struct vfsmount *mnt;
1188 while (!list_empty(head)) {
1189 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1190 list_del_init(&mnt->mnt_hash);
1191 if (mnt_has_parent(mnt)) {
1192 struct dentry *dentry;
1195 br_write_lock(vfsmount_lock);
1196 dentry = mnt->mnt_mountpoint;
1197 m = mnt->mnt_parent;
1198 mnt->mnt_mountpoint = mnt->mnt_root;
1199 mnt->mnt_parent = mnt;
1201 br_write_unlock(vfsmount_lock);
1210 * vfsmount lock must be held for write
1211 * namespace_sem must be held for write
1213 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1215 LIST_HEAD(tmp_list);
1218 for (p = mnt; p; p = next_mnt(p, mnt))
1219 list_move(&p->mnt_hash, &tmp_list);
1222 propagate_umount(&tmp_list);
1224 list_for_each_entry(p, &tmp_list, mnt_hash) {
1225 list_del_init(&p->mnt_expire);
1226 list_del_init(&p->mnt_list);
1227 __touch_mnt_namespace(p->mnt_ns);
1229 __mnt_make_shortterm(p);
1230 list_del_init(&p->mnt_child);
1231 if (mnt_has_parent(p)) {
1232 p->mnt_parent->mnt_ghosts++;
1233 dentry_reset_mounted(p->mnt_mountpoint);
1235 change_mnt_propagation(p, MS_PRIVATE);
1237 list_splice(&tmp_list, kill);
1240 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1242 static int do_umount(struct vfsmount *mnt, int flags)
1244 struct super_block *sb = mnt->mnt_sb;
1246 LIST_HEAD(umount_list);
1248 retval = security_sb_umount(mnt, flags);
1253 * Allow userspace to request a mountpoint be expired rather than
1254 * unmounting unconditionally. Unmount only happens if:
1255 * (1) the mark is already set (the mark is cleared by mntput())
1256 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1258 if (flags & MNT_EXPIRE) {
1259 if (mnt == current->fs->root.mnt ||
1260 flags & (MNT_FORCE | MNT_DETACH))
1264 * probably don't strictly need the lock here if we examined
1265 * all race cases, but it's a slowpath.
1267 br_write_lock(vfsmount_lock);
1268 if (mnt_get_count(mnt) != 2) {
1269 br_write_unlock(vfsmount_lock);
1272 br_write_unlock(vfsmount_lock);
1274 if (!xchg(&mnt->mnt_expiry_mark, 1))
1279 * If we may have to abort operations to get out of this
1280 * mount, and they will themselves hold resources we must
1281 * allow the fs to do things. In the Unix tradition of
1282 * 'Gee thats tricky lets do it in userspace' the umount_begin
1283 * might fail to complete on the first run through as other tasks
1284 * must return, and the like. Thats for the mount program to worry
1285 * about for the moment.
1288 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1289 sb->s_op->umount_begin(sb);
1293 * No sense to grab the lock for this test, but test itself looks
1294 * somewhat bogus. Suggestions for better replacement?
1295 * Ho-hum... In principle, we might treat that as umount + switch
1296 * to rootfs. GC would eventually take care of the old vfsmount.
1297 * Actually it makes sense, especially if rootfs would contain a
1298 * /reboot - static binary that would close all descriptors and
1299 * call reboot(9). Then init(8) could umount root and exec /reboot.
1301 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1303 * Special case for "unmounting" root ...
1304 * we just try to remount it readonly.
1306 down_write(&sb->s_umount);
1307 if (!(sb->s_flags & MS_RDONLY))
1308 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1309 up_write(&sb->s_umount);
1313 down_write(&namespace_sem);
1314 br_write_lock(vfsmount_lock);
1317 if (!(flags & MNT_DETACH))
1318 shrink_submounts(mnt, &umount_list);
1321 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1322 if (!list_empty(&mnt->mnt_list))
1323 umount_tree(mnt, 1, &umount_list);
1326 br_write_unlock(vfsmount_lock);
1327 up_write(&namespace_sem);
1328 release_mounts(&umount_list);
1333 * Now umount can handle mount points as well as block devices.
1334 * This is important for filesystems which use unnamed block devices.
1336 * We now support a flag for forced unmount like the other 'big iron'
1337 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1340 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1344 int lookup_flags = 0;
1346 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1349 if (!(flags & UMOUNT_NOFOLLOW))
1350 lookup_flags |= LOOKUP_FOLLOW;
1352 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1356 if (path.dentry != path.mnt->mnt_root)
1358 if (!check_mnt(path.mnt))
1362 if (!capable(CAP_SYS_ADMIN))
1365 retval = do_umount(path.mnt, flags);
1367 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1369 mntput_no_expire(path.mnt);
1374 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1377 * The 2.0 compatible umount. No flags.
1379 SYSCALL_DEFINE1(oldumount, char __user *, name)
1381 return sys_umount(name, 0);
1386 static int mount_is_safe(struct path *path)
1388 if (capable(CAP_SYS_ADMIN))
1392 if (S_ISLNK(path->dentry->d_inode->i_mode))
1394 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1395 if (current_uid() != path->dentry->d_inode->i_uid)
1398 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1404 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1407 struct vfsmount *res, *p, *q, *r, *s;
1410 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1413 res = q = clone_mnt(mnt, dentry, flag);
1416 q->mnt_mountpoint = mnt->mnt_mountpoint;
1419 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1420 if (!is_subdir(r->mnt_mountpoint, dentry))
1423 for (s = r; s; s = next_mnt(s, r)) {
1424 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1425 s = skip_mnt_tree(s);
1428 while (p != s->mnt_parent) {
1434 path.dentry = p->mnt_mountpoint;
1435 q = clone_mnt(p, p->mnt_root, flag);
1438 br_write_lock(vfsmount_lock);
1439 list_add_tail(&q->mnt_list, &res->mnt_list);
1440 attach_mnt(q, &path);
1441 br_write_unlock(vfsmount_lock);
1447 LIST_HEAD(umount_list);
1448 br_write_lock(vfsmount_lock);
1449 umount_tree(res, 0, &umount_list);
1450 br_write_unlock(vfsmount_lock);
1451 release_mounts(&umount_list);
1456 struct vfsmount *collect_mounts(struct path *path)
1458 struct vfsmount *tree;
1459 down_write(&namespace_sem);
1460 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1461 up_write(&namespace_sem);
1465 void drop_collected_mounts(struct vfsmount *mnt)
1467 LIST_HEAD(umount_list);
1468 down_write(&namespace_sem);
1469 br_write_lock(vfsmount_lock);
1470 umount_tree(mnt, 0, &umount_list);
1471 br_write_unlock(vfsmount_lock);
1472 up_write(&namespace_sem);
1473 release_mounts(&umount_list);
1476 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1477 struct vfsmount *root)
1479 struct vfsmount *mnt;
1480 int res = f(root, arg);
1483 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1491 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1495 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1496 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1497 mnt_release_group_id(p);
1501 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1505 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1506 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1507 int err = mnt_alloc_group_id(p);
1509 cleanup_group_ids(mnt, p);
1519 * @source_mnt : mount tree to be attached
1520 * @nd : place the mount tree @source_mnt is attached
1521 * @parent_nd : if non-null, detach the source_mnt from its parent and
1522 * store the parent mount and mountpoint dentry.
1523 * (done when source_mnt is moved)
1525 * NOTE: in the table below explains the semantics when a source mount
1526 * of a given type is attached to a destination mount of a given type.
1527 * ---------------------------------------------------------------------------
1528 * | BIND MOUNT OPERATION |
1529 * |**************************************************************************
1530 * | source-->| shared | private | slave | unbindable |
1534 * |**************************************************************************
1535 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1537 * |non-shared| shared (+) | private | slave (*) | invalid |
1538 * ***************************************************************************
1539 * A bind operation clones the source mount and mounts the clone on the
1540 * destination mount.
1542 * (++) the cloned mount is propagated to all the mounts in the propagation
1543 * tree of the destination mount and the cloned mount is added to
1544 * the peer group of the source mount.
1545 * (+) the cloned mount is created under the destination mount and is marked
1546 * as shared. The cloned mount is added to the peer group of the source
1548 * (+++) the mount is propagated to all the mounts in the propagation tree
1549 * of the destination mount and the cloned mount is made slave
1550 * of the same master as that of the source mount. The cloned mount
1551 * is marked as 'shared and slave'.
1552 * (*) the cloned mount is made a slave of the same master as that of the
1555 * ---------------------------------------------------------------------------
1556 * | MOVE MOUNT OPERATION |
1557 * |**************************************************************************
1558 * | source-->| shared | private | slave | unbindable |
1562 * |**************************************************************************
1563 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1565 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1566 * ***************************************************************************
1568 * (+) the mount is moved to the destination. And is then propagated to
1569 * all the mounts in the propagation tree of the destination mount.
1570 * (+*) the mount is moved to the destination.
1571 * (+++) the mount is moved to the destination and is then propagated to
1572 * all the mounts belonging to the destination mount's propagation tree.
1573 * the mount is marked as 'shared and slave'.
1574 * (*) the mount continues to be a slave at the new location.
1576 * if the source mount is a tree, the operations explained above is
1577 * applied to each mount in the tree.
1578 * Must be called without spinlocks held, since this function can sleep
1581 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1582 struct path *path, struct path *parent_path)
1584 LIST_HEAD(tree_list);
1585 struct vfsmount *dest_mnt = path->mnt;
1586 struct dentry *dest_dentry = path->dentry;
1587 struct vfsmount *child, *p;
1590 if (IS_MNT_SHARED(dest_mnt)) {
1591 err = invent_group_ids(source_mnt, true);
1595 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1597 goto out_cleanup_ids;
1599 br_write_lock(vfsmount_lock);
1601 if (IS_MNT_SHARED(dest_mnt)) {
1602 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1606 detach_mnt(source_mnt, parent_path);
1607 attach_mnt(source_mnt, path);
1608 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1610 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1611 commit_tree(source_mnt);
1614 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1615 list_del_init(&child->mnt_hash);
1618 br_write_unlock(vfsmount_lock);
1623 if (IS_MNT_SHARED(dest_mnt))
1624 cleanup_group_ids(source_mnt, NULL);
1629 static int lock_mount(struct path *path)
1631 struct vfsmount *mnt;
1633 mutex_lock(&path->dentry->d_inode->i_mutex);
1634 if (unlikely(cant_mount(path->dentry))) {
1635 mutex_unlock(&path->dentry->d_inode->i_mutex);
1638 down_write(&namespace_sem);
1639 mnt = lookup_mnt(path);
1642 up_write(&namespace_sem);
1643 mutex_unlock(&path->dentry->d_inode->i_mutex);
1646 path->dentry = dget(mnt->mnt_root);
1650 static void unlock_mount(struct path *path)
1652 up_write(&namespace_sem);
1653 mutex_unlock(&path->dentry->d_inode->i_mutex);
1656 static int graft_tree(struct vfsmount *mnt, struct path *path)
1658 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1661 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1662 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1665 if (d_unlinked(path->dentry))
1668 return attach_recursive_mnt(mnt, path, NULL);
1672 * Sanity check the flags to change_mnt_propagation.
1675 static int flags_to_propagation_type(int flags)
1677 int type = flags & ~(MS_REC | MS_SILENT);
1679 /* Fail if any non-propagation flags are set */
1680 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1682 /* Only one propagation flag should be set */
1683 if (!is_power_of_2(type))
1689 * recursively change the type of the mountpoint.
1691 static int do_change_type(struct path *path, int flag)
1693 struct vfsmount *m, *mnt = path->mnt;
1694 int recurse = flag & MS_REC;
1698 if (!capable(CAP_SYS_ADMIN))
1701 if (path->dentry != path->mnt->mnt_root)
1704 type = flags_to_propagation_type(flag);
1708 down_write(&namespace_sem);
1709 if (type == MS_SHARED) {
1710 err = invent_group_ids(mnt, recurse);
1715 br_write_lock(vfsmount_lock);
1716 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1717 change_mnt_propagation(m, type);
1718 br_write_unlock(vfsmount_lock);
1721 up_write(&namespace_sem);
1726 * do loopback mount.
1728 static int do_loopback(struct path *path, char *old_name,
1731 LIST_HEAD(umount_list);
1732 struct path old_path;
1733 struct vfsmount *mnt = NULL;
1734 int err = mount_is_safe(path);
1737 if (!old_name || !*old_name)
1739 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1743 err = lock_mount(path);
1748 if (IS_MNT_UNBINDABLE(old_path.mnt))
1751 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1756 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1758 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1763 err = graft_tree(mnt, path);
1765 br_write_lock(vfsmount_lock);
1766 umount_tree(mnt, 0, &umount_list);
1767 br_write_unlock(vfsmount_lock);
1771 release_mounts(&umount_list);
1773 path_put(&old_path);
1777 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1780 int readonly_request = 0;
1782 if (ms_flags & MS_RDONLY)
1783 readonly_request = 1;
1784 if (readonly_request == __mnt_is_readonly(mnt))
1787 if (readonly_request)
1788 error = mnt_make_readonly(mnt);
1790 __mnt_unmake_readonly(mnt);
1795 * change filesystem flags. dir should be a physical root of filesystem.
1796 * If you've mounted a non-root directory somewhere and want to do remount
1797 * on it - tough luck.
1799 static int do_remount(struct path *path, int flags, int mnt_flags,
1803 struct super_block *sb = path->mnt->mnt_sb;
1805 if (!capable(CAP_SYS_ADMIN))
1808 if (!check_mnt(path->mnt))
1811 if (path->dentry != path->mnt->mnt_root)
1814 err = security_sb_remount(sb, data);
1818 down_write(&sb->s_umount);
1819 if (flags & MS_BIND)
1820 err = change_mount_flags(path->mnt, flags);
1822 err = do_remount_sb(sb, flags, data, 0);
1824 br_write_lock(vfsmount_lock);
1825 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1826 path->mnt->mnt_flags = mnt_flags;
1827 br_write_unlock(vfsmount_lock);
1829 up_write(&sb->s_umount);
1831 br_write_lock(vfsmount_lock);
1832 touch_mnt_namespace(path->mnt->mnt_ns);
1833 br_write_unlock(vfsmount_lock);
1838 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1841 for (p = mnt; p; p = next_mnt(p, mnt)) {
1842 if (IS_MNT_UNBINDABLE(p))
1848 static int do_move_mount(struct path *path, char *old_name)
1850 struct path old_path, parent_path;
1853 if (!capable(CAP_SYS_ADMIN))
1855 if (!old_name || !*old_name)
1857 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1861 err = lock_mount(path);
1866 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1869 if (d_unlinked(path->dentry))
1873 if (old_path.dentry != old_path.mnt->mnt_root)
1876 if (!mnt_has_parent(old_path.mnt))
1879 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1880 S_ISDIR(old_path.dentry->d_inode->i_mode))
1883 * Don't move a mount residing in a shared parent.
1885 if (IS_MNT_SHARED(old_path.mnt->mnt_parent))
1888 * Don't move a mount tree containing unbindable mounts to a destination
1889 * mount which is shared.
1891 if (IS_MNT_SHARED(path->mnt) &&
1892 tree_contains_unbindable(old_path.mnt))
1895 for (p = path->mnt; mnt_has_parent(p); p = p->mnt_parent)
1896 if (p == old_path.mnt)
1899 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1903 /* if the mount is moved, it should no longer be expire
1905 list_del_init(&old_path.mnt->mnt_expire);
1910 path_put(&parent_path);
1911 path_put(&old_path);
1915 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1918 const char *subtype = strchr(fstype, '.');
1927 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1929 if (!mnt->mnt_sb->s_subtype)
1935 return ERR_PTR(err);
1938 static struct vfsmount *
1939 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1941 struct file_system_type *type = get_fs_type(fstype);
1942 struct vfsmount *mnt;
1944 return ERR_PTR(-ENODEV);
1945 mnt = vfs_kern_mount(type, flags, name, data);
1946 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1947 !mnt->mnt_sb->s_subtype)
1948 mnt = fs_set_subtype(mnt, fstype);
1949 put_filesystem(type);
1954 * add a mount into a namespace's mount tree
1956 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1960 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1962 err = lock_mount(path);
1967 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1970 /* Refuse the same filesystem on the same mount point */
1972 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1973 path->mnt->mnt_root == path->dentry)
1977 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1980 newmnt->mnt_flags = mnt_flags;
1981 err = graft_tree(newmnt, path);
1989 * create a new mount for userspace and request it to be added into the
1992 static int do_new_mount(struct path *path, char *type, int flags,
1993 int mnt_flags, char *name, void *data)
1995 struct vfsmount *mnt;
2001 /* we need capabilities... */
2002 if (!capable(CAP_SYS_ADMIN))
2005 mnt = do_kern_mount(type, flags, name, data);
2007 return PTR_ERR(mnt);
2009 err = do_add_mount(mnt, path, mnt_flags);
2015 int finish_automount(struct vfsmount *m, struct path *path)
2018 /* The new mount record should have at least 2 refs to prevent it being
2019 * expired before we get a chance to add it
2021 BUG_ON(mnt_get_count(m) < 2);
2023 if (m->mnt_sb == path->mnt->mnt_sb &&
2024 m->mnt_root == path->dentry) {
2029 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2033 /* remove m from any expiration list it may be on */
2034 if (!list_empty(&m->mnt_expire)) {
2035 down_write(&namespace_sem);
2036 br_write_lock(vfsmount_lock);
2037 list_del_init(&m->mnt_expire);
2038 br_write_unlock(vfsmount_lock);
2039 up_write(&namespace_sem);
2047 * mnt_set_expiry - Put a mount on an expiration list
2048 * @mnt: The mount to list.
2049 * @expiry_list: The list to add the mount to.
2051 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2053 down_write(&namespace_sem);
2054 br_write_lock(vfsmount_lock);
2056 list_add_tail(&mnt->mnt_expire, expiry_list);
2058 br_write_unlock(vfsmount_lock);
2059 up_write(&namespace_sem);
2061 EXPORT_SYMBOL(mnt_set_expiry);
2064 * process a list of expirable mountpoints with the intent of discarding any
2065 * mountpoints that aren't in use and haven't been touched since last we came
2068 void mark_mounts_for_expiry(struct list_head *mounts)
2070 struct vfsmount *mnt, *next;
2071 LIST_HEAD(graveyard);
2074 if (list_empty(mounts))
2077 down_write(&namespace_sem);
2078 br_write_lock(vfsmount_lock);
2080 /* extract from the expiration list every vfsmount that matches the
2081 * following criteria:
2082 * - only referenced by its parent vfsmount
2083 * - still marked for expiry (marked on the last call here; marks are
2084 * cleared by mntput())
2086 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2087 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2088 propagate_mount_busy(mnt, 1))
2090 list_move(&mnt->mnt_expire, &graveyard);
2092 while (!list_empty(&graveyard)) {
2093 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2094 touch_mnt_namespace(mnt->mnt_ns);
2095 umount_tree(mnt, 1, &umounts);
2097 br_write_unlock(vfsmount_lock);
2098 up_write(&namespace_sem);
2100 release_mounts(&umounts);
2103 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2106 * Ripoff of 'select_parent()'
2108 * search the list of submounts for a given mountpoint, and move any
2109 * shrinkable submounts to the 'graveyard' list.
2111 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2113 struct vfsmount *this_parent = parent;
2114 struct list_head *next;
2118 next = this_parent->mnt_mounts.next;
2120 while (next != &this_parent->mnt_mounts) {
2121 struct list_head *tmp = next;
2122 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2125 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2128 * Descend a level if the d_mounts list is non-empty.
2130 if (!list_empty(&mnt->mnt_mounts)) {
2135 if (!propagate_mount_busy(mnt, 1)) {
2136 list_move_tail(&mnt->mnt_expire, graveyard);
2141 * All done at this level ... ascend and resume the search
2143 if (this_parent != parent) {
2144 next = this_parent->mnt_child.next;
2145 this_parent = this_parent->mnt_parent;
2152 * process a list of expirable mountpoints with the intent of discarding any
2153 * submounts of a specific parent mountpoint
2155 * vfsmount_lock must be held for write
2157 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2159 LIST_HEAD(graveyard);
2162 /* extract submounts of 'mountpoint' from the expiration list */
2163 while (select_submounts(mnt, &graveyard)) {
2164 while (!list_empty(&graveyard)) {
2165 m = list_first_entry(&graveyard, struct vfsmount,
2167 touch_mnt_namespace(m->mnt_ns);
2168 umount_tree(m, 1, umounts);
2174 * Some copy_from_user() implementations do not return the exact number of
2175 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2176 * Note that this function differs from copy_from_user() in that it will oops
2177 * on bad values of `to', rather than returning a short copy.
2179 static long exact_copy_from_user(void *to, const void __user * from,
2183 const char __user *f = from;
2186 if (!access_ok(VERIFY_READ, from, n))
2190 if (__get_user(c, f)) {
2201 int copy_mount_options(const void __user * data, unsigned long *where)
2211 if (!(page = __get_free_page(GFP_KERNEL)))
2214 /* We only care that *some* data at the address the user
2215 * gave us is valid. Just in case, we'll zero
2216 * the remainder of the page.
2218 /* copy_from_user cannot cross TASK_SIZE ! */
2219 size = TASK_SIZE - (unsigned long)data;
2220 if (size > PAGE_SIZE)
2223 i = size - exact_copy_from_user((void *)page, data, size);
2229 memset((char *)page + i, 0, PAGE_SIZE - i);
2234 int copy_mount_string(const void __user *data, char **where)
2243 tmp = strndup_user(data, PAGE_SIZE);
2245 return PTR_ERR(tmp);
2252 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2253 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2255 * data is a (void *) that can point to any structure up to
2256 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2257 * information (or be NULL).
2259 * Pre-0.97 versions of mount() didn't have a flags word.
2260 * When the flags word was introduced its top half was required
2261 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2262 * Therefore, if this magic number is present, it carries no information
2263 * and must be discarded.
2265 long do_mount(char *dev_name, char *dir_name, char *type_page,
2266 unsigned long flags, void *data_page)
2273 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2274 flags &= ~MS_MGC_MSK;
2276 /* Basic sanity checks */
2278 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2282 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2284 /* ... and get the mountpoint */
2285 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2289 retval = security_sb_mount(dev_name, &path,
2290 type_page, flags, data_page);
2294 /* Default to relatime unless overriden */
2295 if (!(flags & MS_NOATIME))
2296 mnt_flags |= MNT_RELATIME;
2298 /* Separate the per-mountpoint flags */
2299 if (flags & MS_NOSUID)
2300 mnt_flags |= MNT_NOSUID;
2301 if (flags & MS_NODEV)
2302 mnt_flags |= MNT_NODEV;
2303 if (flags & MS_NOEXEC)
2304 mnt_flags |= MNT_NOEXEC;
2305 if (flags & MS_NOATIME)
2306 mnt_flags |= MNT_NOATIME;
2307 if (flags & MS_NODIRATIME)
2308 mnt_flags |= MNT_NODIRATIME;
2309 if (flags & MS_STRICTATIME)
2310 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2311 if (flags & MS_RDONLY)
2312 mnt_flags |= MNT_READONLY;
2314 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2315 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2318 if (flags & MS_REMOUNT)
2319 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2321 else if (flags & MS_BIND)
2322 retval = do_loopback(&path, dev_name, flags & MS_REC);
2323 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2324 retval = do_change_type(&path, flags);
2325 else if (flags & MS_MOVE)
2326 retval = do_move_mount(&path, dev_name);
2328 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2329 dev_name, data_page);
2335 static struct mnt_namespace *alloc_mnt_ns(void)
2337 struct mnt_namespace *new_ns;
2339 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2341 return ERR_PTR(-ENOMEM);
2342 atomic_set(&new_ns->count, 1);
2343 new_ns->root = NULL;
2344 INIT_LIST_HEAD(&new_ns->list);
2345 init_waitqueue_head(&new_ns->poll);
2350 void mnt_make_longterm(struct vfsmount *mnt)
2352 __mnt_make_longterm(mnt);
2355 void mnt_make_shortterm(struct vfsmount *mnt)
2358 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2360 br_write_lock(vfsmount_lock);
2361 atomic_dec(&mnt->mnt_longterm);
2362 br_write_unlock(vfsmount_lock);
2367 * Allocate a new namespace structure and populate it with contents
2368 * copied from the namespace of the passed in task structure.
2370 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2371 struct fs_struct *fs)
2373 struct mnt_namespace *new_ns;
2374 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2375 struct vfsmount *p, *q;
2377 new_ns = alloc_mnt_ns();
2381 down_write(&namespace_sem);
2382 /* First pass: copy the tree topology */
2383 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2384 CL_COPY_ALL | CL_EXPIRE);
2385 if (!new_ns->root) {
2386 up_write(&namespace_sem);
2388 return ERR_PTR(-ENOMEM);
2390 br_write_lock(vfsmount_lock);
2391 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2392 br_write_unlock(vfsmount_lock);
2395 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2396 * as belonging to new namespace. We have already acquired a private
2397 * fs_struct, so tsk->fs->lock is not needed.
2403 __mnt_make_longterm(q);
2405 if (p == fs->root.mnt) {
2406 fs->root.mnt = mntget(q);
2407 __mnt_make_longterm(q);
2408 mnt_make_shortterm(p);
2411 if (p == fs->pwd.mnt) {
2412 fs->pwd.mnt = mntget(q);
2413 __mnt_make_longterm(q);
2414 mnt_make_shortterm(p);
2418 p = next_mnt(p, mnt_ns->root);
2419 q = next_mnt(q, new_ns->root);
2421 up_write(&namespace_sem);
2431 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2432 struct fs_struct *new_fs)
2434 struct mnt_namespace *new_ns;
2439 if (!(flags & CLONE_NEWNS))
2442 new_ns = dup_mnt_ns(ns, new_fs);
2449 * create_mnt_ns - creates a private namespace and adds a root filesystem
2450 * @mnt: pointer to the new root filesystem mountpoint
2452 static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2454 struct mnt_namespace *new_ns;
2456 new_ns = alloc_mnt_ns();
2457 if (!IS_ERR(new_ns)) {
2458 mnt->mnt_ns = new_ns;
2459 __mnt_make_longterm(mnt);
2461 list_add(&new_ns->list, &new_ns->root->mnt_list);
2468 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2470 struct mnt_namespace *ns;
2471 struct super_block *s;
2475 ns = create_mnt_ns(mnt);
2477 return ERR_CAST(ns);
2479 err = vfs_path_lookup(mnt->mnt_root, mnt,
2480 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2485 return ERR_PTR(err);
2487 /* trade a vfsmount reference for active sb one */
2488 s = path.mnt->mnt_sb;
2489 atomic_inc(&s->s_active);
2491 /* lock the sucker */
2492 down_write(&s->s_umount);
2493 /* ... and return the root of (sub)tree on it */
2496 EXPORT_SYMBOL(mount_subtree);
2498 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2499 char __user *, type, unsigned long, flags, void __user *, data)
2505 unsigned long data_page;
2507 ret = copy_mount_string(type, &kernel_type);
2511 kernel_dir = getname(dir_name);
2512 if (IS_ERR(kernel_dir)) {
2513 ret = PTR_ERR(kernel_dir);
2517 ret = copy_mount_string(dev_name, &kernel_dev);
2521 ret = copy_mount_options(data, &data_page);
2525 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2526 (void *) data_page);
2528 free_page(data_page);
2532 putname(kernel_dir);
2540 * Return true if path is reachable from root
2542 * namespace_sem or vfsmount_lock is held
2544 bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
2545 const struct path *root)
2547 while (mnt != root->mnt && mnt_has_parent(mnt)) {
2548 dentry = mnt->mnt_mountpoint;
2549 mnt = mnt->mnt_parent;
2551 return mnt == root->mnt && is_subdir(dentry, root->dentry);
2554 int path_is_under(struct path *path1, struct path *path2)
2557 br_read_lock(vfsmount_lock);
2558 res = is_path_reachable(path1->mnt, path1->dentry, path2);
2559 br_read_unlock(vfsmount_lock);
2562 EXPORT_SYMBOL(path_is_under);
2565 * pivot_root Semantics:
2566 * Moves the root file system of the current process to the directory put_old,
2567 * makes new_root as the new root file system of the current process, and sets
2568 * root/cwd of all processes which had them on the current root to new_root.
2571 * The new_root and put_old must be directories, and must not be on the
2572 * same file system as the current process root. The put_old must be
2573 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2574 * pointed to by put_old must yield the same directory as new_root. No other
2575 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2577 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2578 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2579 * in this situation.
2582 * - we don't move root/cwd if they are not at the root (reason: if something
2583 * cared enough to change them, it's probably wrong to force them elsewhere)
2584 * - it's okay to pick a root that isn't the root of a file system, e.g.
2585 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2586 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2589 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2590 const char __user *, put_old)
2592 struct path new, old, parent_path, root_parent, root;
2595 if (!capable(CAP_SYS_ADMIN))
2598 error = user_path_dir(new_root, &new);
2602 error = user_path_dir(put_old, &old);
2606 error = security_sb_pivotroot(&old, &new);
2610 get_fs_root(current->fs, &root);
2611 error = lock_mount(&old);
2616 if (IS_MNT_SHARED(old.mnt) ||
2617 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2618 IS_MNT_SHARED(root.mnt->mnt_parent))
2620 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2623 if (d_unlinked(new.dentry))
2625 if (d_unlinked(old.dentry))
2628 if (new.mnt == root.mnt ||
2629 old.mnt == root.mnt)
2630 goto out4; /* loop, on the same file system */
2632 if (root.mnt->mnt_root != root.dentry)
2633 goto out4; /* not a mountpoint */
2634 if (!mnt_has_parent(root.mnt))
2635 goto out4; /* not attached */
2636 if (new.mnt->mnt_root != new.dentry)
2637 goto out4; /* not a mountpoint */
2638 if (!mnt_has_parent(new.mnt))
2639 goto out4; /* not attached */
2640 /* make sure we can reach put_old from new_root */
2641 if (!is_path_reachable(old.mnt, old.dentry, &new))
2643 br_write_lock(vfsmount_lock);
2644 detach_mnt(new.mnt, &parent_path);
2645 detach_mnt(root.mnt, &root_parent);
2646 /* mount old root on put_old */
2647 attach_mnt(root.mnt, &old);
2648 /* mount new_root on / */
2649 attach_mnt(new.mnt, &root_parent);
2650 touch_mnt_namespace(current->nsproxy->mnt_ns);
2651 br_write_unlock(vfsmount_lock);
2652 chroot_fs_refs(&root, &new);
2657 path_put(&root_parent);
2658 path_put(&parent_path);
2670 static void __init init_mount_tree(void)
2672 struct vfsmount *mnt;
2673 struct mnt_namespace *ns;
2676 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2678 panic("Can't create rootfs");
2680 ns = create_mnt_ns(mnt);
2682 panic("Can't allocate initial namespace");
2684 init_task.nsproxy->mnt_ns = ns;
2687 root.mnt = ns->root;
2688 root.dentry = ns->root->mnt_root;
2690 set_fs_pwd(current->fs, &root);
2691 set_fs_root(current->fs, &root);
2694 void __init mnt_init(void)
2699 init_rwsem(&namespace_sem);
2701 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2702 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2704 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2706 if (!mount_hashtable)
2707 panic("Failed to allocate mount hash table\n");
2709 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2711 for (u = 0; u < HASH_SIZE; u++)
2712 INIT_LIST_HEAD(&mount_hashtable[u]);
2714 br_lock_init(vfsmount_lock);
2718 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2720 fs_kobj = kobject_create_and_add("fs", NULL);
2722 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2727 void put_mnt_ns(struct mnt_namespace *ns)
2729 LIST_HEAD(umount_list);
2731 if (!atomic_dec_and_test(&ns->count))
2733 down_write(&namespace_sem);
2734 br_write_lock(vfsmount_lock);
2735 umount_tree(ns->root, 0, &umount_list);
2736 br_write_unlock(vfsmount_lock);
2737 up_write(&namespace_sem);
2738 release_mounts(&umount_list);
2742 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2744 struct vfsmount *mnt;
2745 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2748 * it is a longterm mount, don't release mnt until
2749 * we unmount before file sys is unregistered
2751 mnt_make_longterm(mnt);
2755 EXPORT_SYMBOL_GPL(kern_mount_data);
2757 void kern_unmount(struct vfsmount *mnt)
2759 /* release long term mount so mount point can be released */
2760 if (!IS_ERR_OR_NULL(mnt)) {
2761 mnt_make_shortterm(mnt);
2765 EXPORT_SYMBOL(kern_unmount);
2767 bool our_mnt(struct vfsmount *mnt)
2769 return check_mnt(mnt);