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);
155 static inline void mnt_set_count(struct vfsmount *mnt, int n)
158 this_cpu_write(mnt->mnt_pcp->mnt_count, n);
165 * vfsmount lock must be held for read
167 static inline void mnt_inc_count(struct vfsmount *mnt)
169 mnt_add_count(mnt, 1);
173 * vfsmount lock must be held for read
175 static inline void mnt_dec_count(struct vfsmount *mnt)
177 mnt_add_count(mnt, -1);
181 * vfsmount lock must be held for write
183 unsigned int mnt_get_count(struct vfsmount *mnt)
186 unsigned int count = 0;
189 for_each_possible_cpu(cpu) {
190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
195 return mnt->mnt_count;
199 struct vfsmount *alloc_vfsmnt(const char *name)
201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
205 err = mnt_alloc_id(mnt);
210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
211 if (!mnt->mnt_devname)
216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
218 goto out_free_devname;
220 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
223 mnt->mnt_writers = 0;
226 INIT_LIST_HEAD(&mnt->mnt_hash);
227 INIT_LIST_HEAD(&mnt->mnt_child);
228 INIT_LIST_HEAD(&mnt->mnt_mounts);
229 INIT_LIST_HEAD(&mnt->mnt_list);
230 INIT_LIST_HEAD(&mnt->mnt_expire);
231 INIT_LIST_HEAD(&mnt->mnt_share);
232 INIT_LIST_HEAD(&mnt->mnt_slave_list);
233 INIT_LIST_HEAD(&mnt->mnt_slave);
234 #ifdef CONFIG_FSNOTIFY
235 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
242 kfree(mnt->mnt_devname);
247 kmem_cache_free(mnt_cache, mnt);
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
270 int __mnt_is_readonly(struct vfsmount *mnt)
272 if (mnt->mnt_flags & MNT_READONLY)
274 if (mnt->mnt_sb->s_flags & MS_RDONLY)
278 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
280 static inline void mnt_inc_writers(struct vfsmount *mnt)
283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
289 static inline void mnt_dec_writers(struct vfsmount *mnt)
292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
298 static unsigned int mnt_get_writers(struct vfsmount *mnt)
301 unsigned int count = 0;
304 for_each_possible_cpu(cpu) {
305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 return mnt->mnt_writers;
315 * Most r/o checks on a fs are for operations that take
316 * discrete amounts of time, like a write() or unlink().
317 * We must keep track of when those operations start
318 * (for permission checks) and when they end, so that
319 * we can determine when writes are able to occur to
323 * mnt_want_write - get write access to a mount
324 * @mnt: the mount on which to take a write
326 * This tells the low-level filesystem that a write is
327 * about to be performed to it, and makes sure that
328 * writes are allowed before returning success. When
329 * the write operation is finished, mnt_drop_write()
330 * must be called. This is effectively a refcount.
332 int mnt_want_write(struct vfsmount *mnt)
337 mnt_inc_writers(mnt);
339 * The store to mnt_inc_writers must be visible before we pass
340 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
341 * incremented count after it has set MNT_WRITE_HOLD.
344 while (mnt->mnt_flags & MNT_WRITE_HOLD)
347 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
348 * be set to match its requirements. So we must not load that until
349 * MNT_WRITE_HOLD is cleared.
352 if (__mnt_is_readonly(mnt)) {
353 mnt_dec_writers(mnt);
361 EXPORT_SYMBOL_GPL(mnt_want_write);
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
375 int mnt_clone_write(struct vfsmount *mnt)
377 /* superblock may be r/o */
378 if (__mnt_is_readonly(mnt))
381 mnt_inc_writers(mnt);
385 EXPORT_SYMBOL_GPL(mnt_clone_write);
388 * mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
391 * This is like mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
394 int mnt_want_write_file(struct file *file)
396 struct inode *inode = file->f_dentry->d_inode;
397 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
398 return mnt_want_write(file->f_path.mnt);
400 return mnt_clone_write(file->f_path.mnt);
402 EXPORT_SYMBOL_GPL(mnt_want_write_file);
405 * mnt_drop_write - give up write access to a mount
406 * @mnt: the mount on which to give up write access
408 * Tells the low-level filesystem that we are done
409 * performing writes to it. Must be matched with
410 * mnt_want_write() call above.
412 void mnt_drop_write(struct vfsmount *mnt)
415 mnt_dec_writers(mnt);
418 EXPORT_SYMBOL_GPL(mnt_drop_write);
420 static int mnt_make_readonly(struct vfsmount *mnt)
424 br_write_lock(vfsmount_lock);
425 mnt->mnt_flags |= MNT_WRITE_HOLD;
427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
428 * should be visible before we do.
433 * With writers on hold, if this value is zero, then there are
434 * definitely no active writers (although held writers may subsequently
435 * increment the count, they'll have to wait, and decrement it after
436 * seeing MNT_READONLY).
438 * It is OK to have counter incremented on one CPU and decremented on
439 * another: the sum will add up correctly. The danger would be when we
440 * sum up each counter, if we read a counter before it is incremented,
441 * but then read another CPU's count which it has been subsequently
442 * decremented from -- we would see more decrements than we should.
443 * MNT_WRITE_HOLD protects against this scenario, because
444 * mnt_want_write first increments count, then smp_mb, then spins on
445 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
446 * we're counting up here.
448 if (mnt_get_writers(mnt) > 0)
451 mnt->mnt_flags |= MNT_READONLY;
453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
454 * that become unheld will see MNT_READONLY.
457 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
458 br_write_unlock(vfsmount_lock);
462 static void __mnt_unmake_readonly(struct vfsmount *mnt)
464 br_write_lock(vfsmount_lock);
465 mnt->mnt_flags &= ~MNT_READONLY;
466 br_write_unlock(vfsmount_lock);
469 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
472 mnt->mnt_root = dget(sb->s_root);
475 EXPORT_SYMBOL(simple_set_mnt);
477 void free_vfsmnt(struct vfsmount *mnt)
479 kfree(mnt->mnt_devname);
482 free_percpu(mnt->mnt_pcp);
484 kmem_cache_free(mnt_cache, mnt);
488 * find the first or last mount at @dentry on vfsmount @mnt depending on
489 * @dir. If @dir is set return the first mount else return the last mount.
490 * vfsmount_lock must be held for read or write.
492 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
495 struct list_head *head = mount_hashtable + hash(mnt, dentry);
496 struct list_head *tmp = head;
497 struct vfsmount *p, *found = NULL;
500 tmp = dir ? tmp->next : tmp->prev;
504 p = list_entry(tmp, struct vfsmount, mnt_hash);
505 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
514 * lookup_mnt increments the ref count before returning
515 * the vfsmount struct.
517 struct vfsmount *lookup_mnt(struct path *path)
519 struct vfsmount *child_mnt;
521 br_read_lock(vfsmount_lock);
522 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
524 br_read_unlock(vfsmount_lock);
528 static inline int check_mnt(struct vfsmount *mnt)
530 return mnt->mnt_ns == current->nsproxy->mnt_ns;
534 * vfsmount lock must be held for write
536 static void touch_mnt_namespace(struct mnt_namespace *ns)
540 wake_up_interruptible(&ns->poll);
545 * vfsmount lock must be held for write
547 static void __touch_mnt_namespace(struct mnt_namespace *ns)
549 if (ns && ns->event != event) {
551 wake_up_interruptible(&ns->poll);
556 * Clear dentry's mounted state if it has no remaining mounts.
557 * vfsmount_lock must be held for write.
559 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
563 for (u = 0; u < HASH_SIZE; u++) {
566 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
567 if (p->mnt_mountpoint == 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 detach_mnt(struct vfsmount *mnt, struct path *old_path)
581 old_path->dentry = mnt->mnt_mountpoint;
582 old_path->mnt = mnt->mnt_parent;
583 mnt->mnt_parent = mnt;
584 mnt->mnt_mountpoint = mnt->mnt_root;
585 list_del_init(&mnt->mnt_child);
586 list_del_init(&mnt->mnt_hash);
587 dentry_reset_mounted(old_path->mnt, old_path->dentry);
591 * vfsmount lock must be held for write
593 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
594 struct vfsmount *child_mnt)
596 child_mnt->mnt_parent = mntget(mnt);
597 child_mnt->mnt_mountpoint = dget(dentry);
598 spin_lock(&dentry->d_lock);
599 dentry->d_flags |= DCACHE_MOUNTED;
600 spin_unlock(&dentry->d_lock);
604 * vfsmount lock must be held for write
606 static void attach_mnt(struct vfsmount *mnt, struct path *path)
608 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
609 list_add_tail(&mnt->mnt_hash, mount_hashtable +
610 hash(path->mnt, path->dentry));
611 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
614 static inline void __mnt_make_longterm(struct vfsmount *mnt)
617 atomic_inc(&mnt->mnt_longterm);
621 /* needs vfsmount lock for write */
622 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
625 atomic_dec(&mnt->mnt_longterm);
630 * vfsmount lock must be held for write
632 static void commit_tree(struct vfsmount *mnt)
634 struct vfsmount *parent = mnt->mnt_parent;
637 struct mnt_namespace *n = parent->mnt_ns;
639 BUG_ON(parent == mnt);
641 list_add_tail(&head, &mnt->mnt_list);
642 list_for_each_entry(m, &head, mnt_list) {
644 __mnt_make_longterm(m);
647 list_splice(&head, n->list.prev);
649 list_add_tail(&mnt->mnt_hash, mount_hashtable +
650 hash(parent, mnt->mnt_mountpoint));
651 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
652 touch_mnt_namespace(n);
655 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
657 struct list_head *next = p->mnt_mounts.next;
658 if (next == &p->mnt_mounts) {
662 next = p->mnt_child.next;
663 if (next != &p->mnt_parent->mnt_mounts)
668 return list_entry(next, struct vfsmount, mnt_child);
671 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
673 struct list_head *prev = p->mnt_mounts.prev;
674 while (prev != &p->mnt_mounts) {
675 p = list_entry(prev, struct vfsmount, mnt_child);
676 prev = p->mnt_mounts.prev;
681 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
684 struct super_block *sb = old->mnt_sb;
685 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
688 if (flag & (CL_SLAVE | CL_PRIVATE))
689 mnt->mnt_group_id = 0; /* not a peer of original */
691 mnt->mnt_group_id = old->mnt_group_id;
693 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
694 int err = mnt_alloc_group_id(mnt);
699 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
700 atomic_inc(&sb->s_active);
702 mnt->mnt_root = dget(root);
703 mnt->mnt_mountpoint = mnt->mnt_root;
704 mnt->mnt_parent = mnt;
706 if (flag & CL_SLAVE) {
707 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
708 mnt->mnt_master = old;
709 CLEAR_MNT_SHARED(mnt);
710 } else if (!(flag & CL_PRIVATE)) {
711 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
712 list_add(&mnt->mnt_share, &old->mnt_share);
713 if (IS_MNT_SLAVE(old))
714 list_add(&mnt->mnt_slave, &old->mnt_slave);
715 mnt->mnt_master = old->mnt_master;
717 if (flag & CL_MAKE_SHARED)
720 /* stick the duplicate mount on the same expiry list
721 * as the original if that was on one */
722 if (flag & CL_EXPIRE) {
723 if (!list_empty(&old->mnt_expire))
724 list_add(&mnt->mnt_expire, &old->mnt_expire);
734 static inline void mntfree(struct vfsmount *mnt)
736 struct super_block *sb = mnt->mnt_sb;
739 * This probably indicates that somebody messed
740 * up a mnt_want/drop_write() pair. If this
741 * happens, the filesystem was probably unable
742 * to make r/w->r/o transitions.
745 * The locking used to deal with mnt_count decrement provides barriers,
746 * so mnt_get_writers() below is safe.
748 WARN_ON(mnt_get_writers(mnt));
749 fsnotify_vfsmount_delete(mnt);
752 deactivate_super(sb);
755 static void mntput_no_expire(struct vfsmount *mnt)
759 br_read_lock(vfsmount_lock);
760 if (likely(atomic_read(&mnt->mnt_longterm))) {
762 br_read_unlock(vfsmount_lock);
765 br_read_unlock(vfsmount_lock);
767 br_write_lock(vfsmount_lock);
769 if (mnt_get_count(mnt)) {
770 br_write_unlock(vfsmount_lock);
775 if (likely(mnt_get_count(mnt)))
777 br_write_lock(vfsmount_lock);
779 if (unlikely(mnt->mnt_pinned)) {
780 mnt_add_count(mnt, mnt->mnt_pinned + 1);
782 br_write_unlock(vfsmount_lock);
783 acct_auto_close_mnt(mnt);
786 br_write_unlock(vfsmount_lock);
790 void mntput(struct vfsmount *mnt)
793 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
794 if (unlikely(mnt->mnt_expiry_mark))
795 mnt->mnt_expiry_mark = 0;
796 mntput_no_expire(mnt);
799 EXPORT_SYMBOL(mntput);
801 struct vfsmount *mntget(struct vfsmount *mnt)
807 EXPORT_SYMBOL(mntget);
809 void mnt_pin(struct vfsmount *mnt)
811 br_write_lock(vfsmount_lock);
813 br_write_unlock(vfsmount_lock);
815 EXPORT_SYMBOL(mnt_pin);
817 void mnt_unpin(struct vfsmount *mnt)
819 br_write_lock(vfsmount_lock);
820 if (mnt->mnt_pinned) {
824 br_write_unlock(vfsmount_lock);
826 EXPORT_SYMBOL(mnt_unpin);
828 static inline void mangle(struct seq_file *m, const char *s)
830 seq_escape(m, s, " \t\n\\");
834 * Simple .show_options callback for filesystems which don't want to
835 * implement more complex mount option showing.
837 * See also save_mount_options().
839 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
844 options = rcu_dereference(mnt->mnt_sb->s_options);
846 if (options != NULL && options[0]) {
854 EXPORT_SYMBOL(generic_show_options);
857 * If filesystem uses generic_show_options(), this function should be
858 * called from the fill_super() callback.
860 * The .remount_fs callback usually needs to be handled in a special
861 * way, to make sure, that previous options are not overwritten if the
864 * Also note, that if the filesystem's .remount_fs function doesn't
865 * reset all options to their default value, but changes only newly
866 * given options, then the displayed options will not reflect reality
869 void save_mount_options(struct super_block *sb, char *options)
871 BUG_ON(sb->s_options);
872 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
874 EXPORT_SYMBOL(save_mount_options);
876 void replace_mount_options(struct super_block *sb, char *options)
878 char *old = sb->s_options;
879 rcu_assign_pointer(sb->s_options, options);
885 EXPORT_SYMBOL(replace_mount_options);
887 #ifdef CONFIG_PROC_FS
889 static void *m_start(struct seq_file *m, loff_t *pos)
891 struct proc_mounts *p = m->private;
893 down_read(&namespace_sem);
894 return seq_list_start(&p->ns->list, *pos);
897 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
899 struct proc_mounts *p = m->private;
901 return seq_list_next(v, &p->ns->list, pos);
904 static void m_stop(struct seq_file *m, void *v)
906 up_read(&namespace_sem);
909 int mnt_had_events(struct proc_mounts *p)
911 struct mnt_namespace *ns = p->ns;
914 br_read_lock(vfsmount_lock);
915 if (p->event != ns->event) {
916 p->event = ns->event;
919 br_read_unlock(vfsmount_lock);
924 struct proc_fs_info {
929 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
931 static const struct proc_fs_info fs_info[] = {
932 { MS_SYNCHRONOUS, ",sync" },
933 { MS_DIRSYNC, ",dirsync" },
934 { MS_MANDLOCK, ",mand" },
937 const struct proc_fs_info *fs_infop;
939 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
940 if (sb->s_flags & fs_infop->flag)
941 seq_puts(m, fs_infop->str);
944 return security_sb_show_options(m, sb);
947 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
949 static const struct proc_fs_info mnt_info[] = {
950 { MNT_NOSUID, ",nosuid" },
951 { MNT_NODEV, ",nodev" },
952 { MNT_NOEXEC, ",noexec" },
953 { MNT_NOATIME, ",noatime" },
954 { MNT_NODIRATIME, ",nodiratime" },
955 { MNT_RELATIME, ",relatime" },
958 const struct proc_fs_info *fs_infop;
960 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
961 if (mnt->mnt_flags & fs_infop->flag)
962 seq_puts(m, fs_infop->str);
966 static void show_type(struct seq_file *m, struct super_block *sb)
968 mangle(m, sb->s_type->name);
969 if (sb->s_subtype && sb->s_subtype[0]) {
971 mangle(m, sb->s_subtype);
975 static int show_vfsmnt(struct seq_file *m, void *v)
977 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
979 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
981 if (mnt->mnt_sb->s_op->show_devname) {
982 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
986 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
989 seq_path(m, &mnt_path, " \t\n\\");
991 show_type(m, mnt->mnt_sb);
992 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
993 err = show_sb_opts(m, mnt->mnt_sb);
996 show_mnt_opts(m, mnt);
997 if (mnt->mnt_sb->s_op->show_options)
998 err = mnt->mnt_sb->s_op->show_options(m, mnt);
999 seq_puts(m, " 0 0\n");
1004 const struct seq_operations mounts_op = {
1011 static int uuid_is_nil(u8 *uuid)
1014 u8 *cp = (u8 *)uuid;
1016 for (i = 0; i < 16; i++) {
1023 static int show_mountinfo(struct seq_file *m, void *v)
1025 struct proc_mounts *p = m->private;
1026 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1027 struct super_block *sb = mnt->mnt_sb;
1028 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1029 struct path root = p->root;
1032 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1033 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1034 if (sb->s_op->show_path)
1035 err = sb->s_op->show_path(m, mnt);
1037 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1041 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1042 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1044 * Mountpoint is outside root, discard that one. Ugly,
1045 * but less so than trying to do that in iterator in a
1046 * race-free way (due to renames).
1050 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1051 show_mnt_opts(m, mnt);
1053 /* Tagged fields ("foo:X" or "bar") */
1054 if (IS_MNT_SHARED(mnt))
1055 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1056 if (IS_MNT_SLAVE(mnt)) {
1057 int master = mnt->mnt_master->mnt_group_id;
1058 int dom = get_dominating_id(mnt, &p->root);
1059 seq_printf(m, " master:%i", master);
1060 if (dom && dom != master)
1061 seq_printf(m, " propagate_from:%i", dom);
1063 if (IS_MNT_UNBINDABLE(mnt))
1064 seq_puts(m, " unbindable");
1066 if (!uuid_is_nil(mnt->mnt_sb->s_uuid))
1067 /* print the uuid */
1068 seq_printf(m, " uuid:%pU", mnt->mnt_sb->s_uuid);
1070 /* Filesystem specific data */
1074 if (sb->s_op->show_devname)
1075 err = sb->s_op->show_devname(m, mnt);
1077 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1080 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1081 err = show_sb_opts(m, sb);
1084 if (sb->s_op->show_options)
1085 err = sb->s_op->show_options(m, mnt);
1091 const struct seq_operations mountinfo_op = {
1095 .show = show_mountinfo,
1098 static int show_vfsstat(struct seq_file *m, void *v)
1100 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1101 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1105 if (mnt->mnt_sb->s_op->show_devname) {
1106 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1108 if (mnt->mnt_devname) {
1109 seq_puts(m, "device ");
1110 mangle(m, mnt->mnt_devname);
1112 seq_puts(m, "no device");
1116 seq_puts(m, " mounted on ");
1117 seq_path(m, &mnt_path, " \t\n\\");
1120 /* file system type */
1121 seq_puts(m, "with fstype ");
1122 show_type(m, mnt->mnt_sb);
1124 /* optional statistics */
1125 if (mnt->mnt_sb->s_op->show_stats) {
1128 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1135 const struct seq_operations mountstats_op = {
1139 .show = show_vfsstat,
1141 #endif /* CONFIG_PROC_FS */
1144 * may_umount_tree - check if a mount tree is busy
1145 * @mnt: root of mount tree
1147 * This is called to check if a tree of mounts has any
1148 * open files, pwds, chroots or sub mounts that are
1151 int may_umount_tree(struct vfsmount *mnt)
1153 int actual_refs = 0;
1154 int minimum_refs = 0;
1157 /* write lock needed for mnt_get_count */
1158 br_write_lock(vfsmount_lock);
1159 for (p = mnt; p; p = next_mnt(p, mnt)) {
1160 actual_refs += mnt_get_count(p);
1163 br_write_unlock(vfsmount_lock);
1165 if (actual_refs > minimum_refs)
1171 EXPORT_SYMBOL(may_umount_tree);
1174 * may_umount - check if a mount point is busy
1175 * @mnt: root of mount
1177 * This is called to check if a mount point has any
1178 * open files, pwds, chroots or sub mounts. If the
1179 * mount has sub mounts this will return busy
1180 * regardless of whether the sub mounts are busy.
1182 * Doesn't take quota and stuff into account. IOW, in some cases it will
1183 * give false negatives. The main reason why it's here is that we need
1184 * a non-destructive way to look for easily umountable filesystems.
1186 int may_umount(struct vfsmount *mnt)
1189 down_read(&namespace_sem);
1190 br_write_lock(vfsmount_lock);
1191 if (propagate_mount_busy(mnt, 2))
1193 br_write_unlock(vfsmount_lock);
1194 up_read(&namespace_sem);
1198 EXPORT_SYMBOL(may_umount);
1200 void release_mounts(struct list_head *head)
1202 struct vfsmount *mnt;
1203 while (!list_empty(head)) {
1204 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1205 list_del_init(&mnt->mnt_hash);
1206 if (mnt->mnt_parent != mnt) {
1207 struct dentry *dentry;
1210 br_write_lock(vfsmount_lock);
1211 dentry = mnt->mnt_mountpoint;
1212 m = mnt->mnt_parent;
1213 mnt->mnt_mountpoint = mnt->mnt_root;
1214 mnt->mnt_parent = mnt;
1216 br_write_unlock(vfsmount_lock);
1225 * vfsmount lock must be held for write
1226 * namespace_sem must be held for write
1228 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1230 LIST_HEAD(tmp_list);
1233 for (p = mnt; p; p = next_mnt(p, mnt))
1234 list_move(&p->mnt_hash, &tmp_list);
1237 propagate_umount(&tmp_list);
1239 list_for_each_entry(p, &tmp_list, mnt_hash) {
1240 list_del_init(&p->mnt_expire);
1241 list_del_init(&p->mnt_list);
1242 __touch_mnt_namespace(p->mnt_ns);
1244 __mnt_make_shortterm(p);
1245 list_del_init(&p->mnt_child);
1246 if (p->mnt_parent != p) {
1247 p->mnt_parent->mnt_ghosts++;
1248 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1250 change_mnt_propagation(p, MS_PRIVATE);
1252 list_splice(&tmp_list, kill);
1255 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1257 static int do_umount(struct vfsmount *mnt, int flags)
1259 struct super_block *sb = mnt->mnt_sb;
1261 LIST_HEAD(umount_list);
1263 retval = security_sb_umount(mnt, flags);
1268 * Allow userspace to request a mountpoint be expired rather than
1269 * unmounting unconditionally. Unmount only happens if:
1270 * (1) the mark is already set (the mark is cleared by mntput())
1271 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1273 if (flags & MNT_EXPIRE) {
1274 if (mnt == current->fs->root.mnt ||
1275 flags & (MNT_FORCE | MNT_DETACH))
1279 * probably don't strictly need the lock here if we examined
1280 * all race cases, but it's a slowpath.
1282 br_write_lock(vfsmount_lock);
1283 if (mnt_get_count(mnt) != 2) {
1284 br_write_unlock(vfsmount_lock);
1287 br_write_unlock(vfsmount_lock);
1289 if (!xchg(&mnt->mnt_expiry_mark, 1))
1294 * If we may have to abort operations to get out of this
1295 * mount, and they will themselves hold resources we must
1296 * allow the fs to do things. In the Unix tradition of
1297 * 'Gee thats tricky lets do it in userspace' the umount_begin
1298 * might fail to complete on the first run through as other tasks
1299 * must return, and the like. Thats for the mount program to worry
1300 * about for the moment.
1303 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1304 sb->s_op->umount_begin(sb);
1308 * No sense to grab the lock for this test, but test itself looks
1309 * somewhat bogus. Suggestions for better replacement?
1310 * Ho-hum... In principle, we might treat that as umount + switch
1311 * to rootfs. GC would eventually take care of the old vfsmount.
1312 * Actually it makes sense, especially if rootfs would contain a
1313 * /reboot - static binary that would close all descriptors and
1314 * call reboot(9). Then init(8) could umount root and exec /reboot.
1316 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1318 * Special case for "unmounting" root ...
1319 * we just try to remount it readonly.
1321 down_write(&sb->s_umount);
1322 if (!(sb->s_flags & MS_RDONLY))
1323 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1324 up_write(&sb->s_umount);
1328 down_write(&namespace_sem);
1329 br_write_lock(vfsmount_lock);
1332 if (!(flags & MNT_DETACH))
1333 shrink_submounts(mnt, &umount_list);
1336 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1337 if (!list_empty(&mnt->mnt_list))
1338 umount_tree(mnt, 1, &umount_list);
1341 br_write_unlock(vfsmount_lock);
1342 up_write(&namespace_sem);
1343 release_mounts(&umount_list);
1348 * Now umount can handle mount points as well as block devices.
1349 * This is important for filesystems which use unnamed block devices.
1351 * We now support a flag for forced unmount like the other 'big iron'
1352 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1355 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1359 int lookup_flags = 0;
1361 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1364 if (!(flags & UMOUNT_NOFOLLOW))
1365 lookup_flags |= LOOKUP_FOLLOW;
1367 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1371 if (path.dentry != path.mnt->mnt_root)
1373 if (!check_mnt(path.mnt))
1377 if (!capable(CAP_SYS_ADMIN))
1380 retval = do_umount(path.mnt, flags);
1382 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1384 mntput_no_expire(path.mnt);
1389 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1392 * The 2.0 compatible umount. No flags.
1394 SYSCALL_DEFINE1(oldumount, char __user *, name)
1396 return sys_umount(name, 0);
1401 static int mount_is_safe(struct path *path)
1403 if (capable(CAP_SYS_ADMIN))
1407 if (S_ISLNK(path->dentry->d_inode->i_mode))
1409 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1410 if (current_uid() != path->dentry->d_inode->i_uid)
1413 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1419 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1422 struct vfsmount *res, *p, *q, *r, *s;
1425 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1428 res = q = clone_mnt(mnt, dentry, flag);
1431 q->mnt_mountpoint = mnt->mnt_mountpoint;
1434 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1435 if (!is_subdir(r->mnt_mountpoint, dentry))
1438 for (s = r; s; s = next_mnt(s, r)) {
1439 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1440 s = skip_mnt_tree(s);
1443 while (p != s->mnt_parent) {
1449 path.dentry = p->mnt_mountpoint;
1450 q = clone_mnt(p, p->mnt_root, flag);
1453 br_write_lock(vfsmount_lock);
1454 list_add_tail(&q->mnt_list, &res->mnt_list);
1455 attach_mnt(q, &path);
1456 br_write_unlock(vfsmount_lock);
1462 LIST_HEAD(umount_list);
1463 br_write_lock(vfsmount_lock);
1464 umount_tree(res, 0, &umount_list);
1465 br_write_unlock(vfsmount_lock);
1466 release_mounts(&umount_list);
1471 struct vfsmount *collect_mounts(struct path *path)
1473 struct vfsmount *tree;
1474 down_write(&namespace_sem);
1475 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1476 up_write(&namespace_sem);
1480 void drop_collected_mounts(struct vfsmount *mnt)
1482 LIST_HEAD(umount_list);
1483 down_write(&namespace_sem);
1484 br_write_lock(vfsmount_lock);
1485 umount_tree(mnt, 0, &umount_list);
1486 br_write_unlock(vfsmount_lock);
1487 up_write(&namespace_sem);
1488 release_mounts(&umount_list);
1491 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1492 struct vfsmount *root)
1494 struct vfsmount *mnt;
1495 int res = f(root, arg);
1498 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1506 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1510 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1511 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1512 mnt_release_group_id(p);
1516 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1520 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1521 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1522 int err = mnt_alloc_group_id(p);
1524 cleanup_group_ids(mnt, p);
1534 * @source_mnt : mount tree to be attached
1535 * @nd : place the mount tree @source_mnt is attached
1536 * @parent_nd : if non-null, detach the source_mnt from its parent and
1537 * store the parent mount and mountpoint dentry.
1538 * (done when source_mnt is moved)
1540 * NOTE: in the table below explains the semantics when a source mount
1541 * of a given type is attached to a destination mount of a given type.
1542 * ---------------------------------------------------------------------------
1543 * | BIND MOUNT OPERATION |
1544 * |**************************************************************************
1545 * | source-->| shared | private | slave | unbindable |
1549 * |**************************************************************************
1550 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1552 * |non-shared| shared (+) | private | slave (*) | invalid |
1553 * ***************************************************************************
1554 * A bind operation clones the source mount and mounts the clone on the
1555 * destination mount.
1557 * (++) the cloned mount is propagated to all the mounts in the propagation
1558 * tree of the destination mount and the cloned mount is added to
1559 * the peer group of the source mount.
1560 * (+) the cloned mount is created under the destination mount and is marked
1561 * as shared. The cloned mount is added to the peer group of the source
1563 * (+++) the mount is propagated to all the mounts in the propagation tree
1564 * of the destination mount and the cloned mount is made slave
1565 * of the same master as that of the source mount. The cloned mount
1566 * is marked as 'shared and slave'.
1567 * (*) the cloned mount is made a slave of the same master as that of the
1570 * ---------------------------------------------------------------------------
1571 * | MOVE MOUNT OPERATION |
1572 * |**************************************************************************
1573 * | source-->| shared | private | slave | unbindable |
1577 * |**************************************************************************
1578 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1580 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1581 * ***************************************************************************
1583 * (+) the mount is moved to the destination. And is then propagated to
1584 * all the mounts in the propagation tree of the destination mount.
1585 * (+*) the mount is moved to the destination.
1586 * (+++) the mount is moved to the destination and is then propagated to
1587 * all the mounts belonging to the destination mount's propagation tree.
1588 * the mount is marked as 'shared and slave'.
1589 * (*) the mount continues to be a slave at the new location.
1591 * if the source mount is a tree, the operations explained above is
1592 * applied to each mount in the tree.
1593 * Must be called without spinlocks held, since this function can sleep
1596 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1597 struct path *path, struct path *parent_path)
1599 LIST_HEAD(tree_list);
1600 struct vfsmount *dest_mnt = path->mnt;
1601 struct dentry *dest_dentry = path->dentry;
1602 struct vfsmount *child, *p;
1605 if (IS_MNT_SHARED(dest_mnt)) {
1606 err = invent_group_ids(source_mnt, true);
1610 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1612 goto out_cleanup_ids;
1614 br_write_lock(vfsmount_lock);
1616 if (IS_MNT_SHARED(dest_mnt)) {
1617 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1621 detach_mnt(source_mnt, parent_path);
1622 attach_mnt(source_mnt, path);
1623 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1625 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1626 commit_tree(source_mnt);
1629 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1630 list_del_init(&child->mnt_hash);
1633 br_write_unlock(vfsmount_lock);
1638 if (IS_MNT_SHARED(dest_mnt))
1639 cleanup_group_ids(source_mnt, NULL);
1644 static int graft_tree(struct vfsmount *mnt, struct path *path)
1647 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1650 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1651 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1655 mutex_lock(&path->dentry->d_inode->i_mutex);
1656 if (cant_mount(path->dentry))
1659 if (!d_unlinked(path->dentry))
1660 err = attach_recursive_mnt(mnt, path, NULL);
1662 mutex_unlock(&path->dentry->d_inode->i_mutex);
1667 * Sanity check the flags to change_mnt_propagation.
1670 static int flags_to_propagation_type(int flags)
1672 int type = flags & ~MS_REC;
1674 /* Fail if any non-propagation flags are set */
1675 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1677 /* Only one propagation flag should be set */
1678 if (!is_power_of_2(type))
1684 * recursively change the type of the mountpoint.
1686 static int do_change_type(struct path *path, int flag)
1688 struct vfsmount *m, *mnt = path->mnt;
1689 int recurse = flag & MS_REC;
1693 if (!capable(CAP_SYS_ADMIN))
1696 if (path->dentry != path->mnt->mnt_root)
1699 type = flags_to_propagation_type(flag);
1703 down_write(&namespace_sem);
1704 if (type == MS_SHARED) {
1705 err = invent_group_ids(mnt, recurse);
1710 br_write_lock(vfsmount_lock);
1711 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1712 change_mnt_propagation(m, type);
1713 br_write_unlock(vfsmount_lock);
1716 up_write(&namespace_sem);
1721 * do loopback mount.
1723 static int do_loopback(struct path *path, char *old_name,
1726 struct path old_path;
1727 struct vfsmount *mnt = NULL;
1728 int err = mount_is_safe(path);
1731 if (!old_name || !*old_name)
1733 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1737 down_write(&namespace_sem);
1739 if (IS_MNT_UNBINDABLE(old_path.mnt))
1742 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1747 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1749 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1754 err = graft_tree(mnt, path);
1756 LIST_HEAD(umount_list);
1758 br_write_lock(vfsmount_lock);
1759 umount_tree(mnt, 0, &umount_list);
1760 br_write_unlock(vfsmount_lock);
1761 release_mounts(&umount_list);
1765 up_write(&namespace_sem);
1766 path_put(&old_path);
1770 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1773 int readonly_request = 0;
1775 if (ms_flags & MS_RDONLY)
1776 readonly_request = 1;
1777 if (readonly_request == __mnt_is_readonly(mnt))
1780 if (readonly_request)
1781 error = mnt_make_readonly(mnt);
1783 __mnt_unmake_readonly(mnt);
1788 * change filesystem flags. dir should be a physical root of filesystem.
1789 * If you've mounted a non-root directory somewhere and want to do remount
1790 * on it - tough luck.
1792 static int do_remount(struct path *path, int flags, int mnt_flags,
1796 struct super_block *sb = path->mnt->mnt_sb;
1798 if (!capable(CAP_SYS_ADMIN))
1801 if (!check_mnt(path->mnt))
1804 if (path->dentry != path->mnt->mnt_root)
1807 err = security_sb_remount(sb, data);
1811 down_write(&sb->s_umount);
1812 if (flags & MS_BIND)
1813 err = change_mount_flags(path->mnt, flags);
1815 err = do_remount_sb(sb, flags, data, 0);
1817 br_write_lock(vfsmount_lock);
1818 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1819 path->mnt->mnt_flags = mnt_flags;
1820 br_write_unlock(vfsmount_lock);
1822 up_write(&sb->s_umount);
1824 br_write_lock(vfsmount_lock);
1825 touch_mnt_namespace(path->mnt->mnt_ns);
1826 br_write_unlock(vfsmount_lock);
1831 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1834 for (p = mnt; p; p = next_mnt(p, mnt)) {
1835 if (IS_MNT_UNBINDABLE(p))
1841 static int do_move_mount(struct path *path, char *old_name)
1843 struct path old_path, parent_path;
1846 if (!capable(CAP_SYS_ADMIN))
1848 if (!old_name || !*old_name)
1850 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1854 down_write(&namespace_sem);
1855 err = follow_down(path, true);
1860 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1864 mutex_lock(&path->dentry->d_inode->i_mutex);
1865 if (cant_mount(path->dentry))
1868 if (d_unlinked(path->dentry))
1872 if (old_path.dentry != old_path.mnt->mnt_root)
1875 if (old_path.mnt == old_path.mnt->mnt_parent)
1878 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1879 S_ISDIR(old_path.dentry->d_inode->i_mode))
1882 * Don't move a mount residing in a shared parent.
1884 if (old_path.mnt->mnt_parent &&
1885 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; p->mnt_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);
1907 mutex_unlock(&path->dentry->d_inode->i_mutex);
1909 up_write(&namespace_sem);
1911 path_put(&parent_path);
1912 path_put(&old_path);
1916 static int do_add_mount(struct vfsmount *, struct path *, int);
1919 * create a new mount for userspace and request it to be added into the
1922 static int do_new_mount(struct path *path, char *type, int flags,
1923 int mnt_flags, char *name, void *data)
1925 struct vfsmount *mnt;
1931 /* we need capabilities... */
1932 if (!capable(CAP_SYS_ADMIN))
1935 mnt = do_kern_mount(type, flags, name, data);
1937 return PTR_ERR(mnt);
1939 err = do_add_mount(mnt, path, mnt_flags);
1945 int finish_automount(struct vfsmount *m, struct path *path)
1948 /* The new mount record should have at least 2 refs to prevent it being
1949 * expired before we get a chance to add it
1951 BUG_ON(mnt_get_count(m) < 2);
1953 if (m->mnt_sb == path->mnt->mnt_sb &&
1954 m->mnt_root == path->dentry) {
1959 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1963 /* remove m from any expiration list it may be on */
1964 if (!list_empty(&m->mnt_expire)) {
1965 down_write(&namespace_sem);
1966 br_write_lock(vfsmount_lock);
1967 list_del_init(&m->mnt_expire);
1968 br_write_unlock(vfsmount_lock);
1969 up_write(&namespace_sem);
1977 * add a mount into a namespace's mount tree
1979 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1983 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1985 down_write(&namespace_sem);
1986 /* Something was mounted here while we slept */
1987 err = follow_down(path, true);
1992 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1995 /* Refuse the same filesystem on the same mount point */
1997 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1998 path->mnt->mnt_root == path->dentry)
2002 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2005 newmnt->mnt_flags = mnt_flags;
2006 err = graft_tree(newmnt, path);
2009 up_write(&namespace_sem);
2014 * mnt_set_expiry - Put a mount on an expiration list
2015 * @mnt: The mount to list.
2016 * @expiry_list: The list to add the mount to.
2018 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2020 down_write(&namespace_sem);
2021 br_write_lock(vfsmount_lock);
2023 list_add_tail(&mnt->mnt_expire, expiry_list);
2025 br_write_unlock(vfsmount_lock);
2026 up_write(&namespace_sem);
2028 EXPORT_SYMBOL(mnt_set_expiry);
2031 * process a list of expirable mountpoints with the intent of discarding any
2032 * mountpoints that aren't in use and haven't been touched since last we came
2035 void mark_mounts_for_expiry(struct list_head *mounts)
2037 struct vfsmount *mnt, *next;
2038 LIST_HEAD(graveyard);
2041 if (list_empty(mounts))
2044 down_write(&namespace_sem);
2045 br_write_lock(vfsmount_lock);
2047 /* extract from the expiration list every vfsmount that matches the
2048 * following criteria:
2049 * - only referenced by its parent vfsmount
2050 * - still marked for expiry (marked on the last call here; marks are
2051 * cleared by mntput())
2053 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2054 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2055 propagate_mount_busy(mnt, 1))
2057 list_move(&mnt->mnt_expire, &graveyard);
2059 while (!list_empty(&graveyard)) {
2060 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2061 touch_mnt_namespace(mnt->mnt_ns);
2062 umount_tree(mnt, 1, &umounts);
2064 br_write_unlock(vfsmount_lock);
2065 up_write(&namespace_sem);
2067 release_mounts(&umounts);
2070 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2073 * Ripoff of 'select_parent()'
2075 * search the list of submounts for a given mountpoint, and move any
2076 * shrinkable submounts to the 'graveyard' list.
2078 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2080 struct vfsmount *this_parent = parent;
2081 struct list_head *next;
2085 next = this_parent->mnt_mounts.next;
2087 while (next != &this_parent->mnt_mounts) {
2088 struct list_head *tmp = next;
2089 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2092 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2095 * Descend a level if the d_mounts list is non-empty.
2097 if (!list_empty(&mnt->mnt_mounts)) {
2102 if (!propagate_mount_busy(mnt, 1)) {
2103 list_move_tail(&mnt->mnt_expire, graveyard);
2108 * All done at this level ... ascend and resume the search
2110 if (this_parent != parent) {
2111 next = this_parent->mnt_child.next;
2112 this_parent = this_parent->mnt_parent;
2119 * process a list of expirable mountpoints with the intent of discarding any
2120 * submounts of a specific parent mountpoint
2122 * vfsmount_lock must be held for write
2124 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2126 LIST_HEAD(graveyard);
2129 /* extract submounts of 'mountpoint' from the expiration list */
2130 while (select_submounts(mnt, &graveyard)) {
2131 while (!list_empty(&graveyard)) {
2132 m = list_first_entry(&graveyard, struct vfsmount,
2134 touch_mnt_namespace(m->mnt_ns);
2135 umount_tree(m, 1, umounts);
2141 * Some copy_from_user() implementations do not return the exact number of
2142 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2143 * Note that this function differs from copy_from_user() in that it will oops
2144 * on bad values of `to', rather than returning a short copy.
2146 static long exact_copy_from_user(void *to, const void __user * from,
2150 const char __user *f = from;
2153 if (!access_ok(VERIFY_READ, from, n))
2157 if (__get_user(c, f)) {
2168 int copy_mount_options(const void __user * data, unsigned long *where)
2178 if (!(page = __get_free_page(GFP_KERNEL)))
2181 /* We only care that *some* data at the address the user
2182 * gave us is valid. Just in case, we'll zero
2183 * the remainder of the page.
2185 /* copy_from_user cannot cross TASK_SIZE ! */
2186 size = TASK_SIZE - (unsigned long)data;
2187 if (size > PAGE_SIZE)
2190 i = size - exact_copy_from_user((void *)page, data, size);
2196 memset((char *)page + i, 0, PAGE_SIZE - i);
2201 int copy_mount_string(const void __user *data, char **where)
2210 tmp = strndup_user(data, PAGE_SIZE);
2212 return PTR_ERR(tmp);
2219 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2220 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2222 * data is a (void *) that can point to any structure up to
2223 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2224 * information (or be NULL).
2226 * Pre-0.97 versions of mount() didn't have a flags word.
2227 * When the flags word was introduced its top half was required
2228 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2229 * Therefore, if this magic number is present, it carries no information
2230 * and must be discarded.
2232 long do_mount(char *dev_name, char *dir_name, char *type_page,
2233 unsigned long flags, void *data_page)
2240 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2241 flags &= ~MS_MGC_MSK;
2243 /* Basic sanity checks */
2245 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2249 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2251 /* ... and get the mountpoint */
2252 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2256 retval = security_sb_mount(dev_name, &path,
2257 type_page, flags, data_page);
2261 /* Default to relatime unless overriden */
2262 if (!(flags & MS_NOATIME))
2263 mnt_flags |= MNT_RELATIME;
2265 /* Separate the per-mountpoint flags */
2266 if (flags & MS_NOSUID)
2267 mnt_flags |= MNT_NOSUID;
2268 if (flags & MS_NODEV)
2269 mnt_flags |= MNT_NODEV;
2270 if (flags & MS_NOEXEC)
2271 mnt_flags |= MNT_NOEXEC;
2272 if (flags & MS_NOATIME)
2273 mnt_flags |= MNT_NOATIME;
2274 if (flags & MS_NODIRATIME)
2275 mnt_flags |= MNT_NODIRATIME;
2276 if (flags & MS_STRICTATIME)
2277 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2278 if (flags & MS_RDONLY)
2279 mnt_flags |= MNT_READONLY;
2281 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2282 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2285 if (flags & MS_REMOUNT)
2286 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2288 else if (flags & MS_BIND)
2289 retval = do_loopback(&path, dev_name, flags & MS_REC);
2290 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2291 retval = do_change_type(&path, flags);
2292 else if (flags & MS_MOVE)
2293 retval = do_move_mount(&path, dev_name);
2295 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2296 dev_name, data_page);
2302 static struct mnt_namespace *alloc_mnt_ns(void)
2304 struct mnt_namespace *new_ns;
2306 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2308 return ERR_PTR(-ENOMEM);
2309 atomic_set(&new_ns->count, 1);
2310 new_ns->root = NULL;
2311 INIT_LIST_HEAD(&new_ns->list);
2312 init_waitqueue_head(&new_ns->poll);
2317 void mnt_make_longterm(struct vfsmount *mnt)
2319 __mnt_make_longterm(mnt);
2322 void mnt_make_shortterm(struct vfsmount *mnt)
2325 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2327 br_write_lock(vfsmount_lock);
2328 atomic_dec(&mnt->mnt_longterm);
2329 br_write_unlock(vfsmount_lock);
2334 * Allocate a new namespace structure and populate it with contents
2335 * copied from the namespace of the passed in task structure.
2337 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2338 struct fs_struct *fs)
2340 struct mnt_namespace *new_ns;
2341 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2342 struct vfsmount *p, *q;
2344 new_ns = alloc_mnt_ns();
2348 down_write(&namespace_sem);
2349 /* First pass: copy the tree topology */
2350 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2351 CL_COPY_ALL | CL_EXPIRE);
2352 if (!new_ns->root) {
2353 up_write(&namespace_sem);
2355 return ERR_PTR(-ENOMEM);
2357 br_write_lock(vfsmount_lock);
2358 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2359 br_write_unlock(vfsmount_lock);
2362 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2363 * as belonging to new namespace. We have already acquired a private
2364 * fs_struct, so tsk->fs->lock is not needed.
2370 __mnt_make_longterm(q);
2372 if (p == fs->root.mnt) {
2373 fs->root.mnt = mntget(q);
2374 __mnt_make_longterm(q);
2375 mnt_make_shortterm(p);
2378 if (p == fs->pwd.mnt) {
2379 fs->pwd.mnt = mntget(q);
2380 __mnt_make_longterm(q);
2381 mnt_make_shortterm(p);
2385 p = next_mnt(p, mnt_ns->root);
2386 q = next_mnt(q, new_ns->root);
2388 up_write(&namespace_sem);
2398 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2399 struct fs_struct *new_fs)
2401 struct mnt_namespace *new_ns;
2406 if (!(flags & CLONE_NEWNS))
2409 new_ns = dup_mnt_ns(ns, new_fs);
2416 * create_mnt_ns - creates a private namespace and adds a root filesystem
2417 * @mnt: pointer to the new root filesystem mountpoint
2419 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2421 struct mnt_namespace *new_ns;
2423 new_ns = alloc_mnt_ns();
2424 if (!IS_ERR(new_ns)) {
2425 mnt->mnt_ns = new_ns;
2426 __mnt_make_longterm(mnt);
2428 list_add(&new_ns->list, &new_ns->root->mnt_list);
2432 EXPORT_SYMBOL(create_mnt_ns);
2434 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2435 char __user *, type, unsigned long, flags, void __user *, data)
2441 unsigned long data_page;
2443 ret = copy_mount_string(type, &kernel_type);
2447 kernel_dir = getname(dir_name);
2448 if (IS_ERR(kernel_dir)) {
2449 ret = PTR_ERR(kernel_dir);
2453 ret = copy_mount_string(dev_name, &kernel_dev);
2457 ret = copy_mount_options(data, &data_page);
2461 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2462 (void *) data_page);
2464 free_page(data_page);
2468 putname(kernel_dir);
2476 * pivot_root Semantics:
2477 * Moves the root file system of the current process to the directory put_old,
2478 * makes new_root as the new root file system of the current process, and sets
2479 * root/cwd of all processes which had them on the current root to new_root.
2482 * The new_root and put_old must be directories, and must not be on the
2483 * same file system as the current process root. The put_old must be
2484 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2485 * pointed to by put_old must yield the same directory as new_root. No other
2486 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2488 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2489 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2490 * in this situation.
2493 * - we don't move root/cwd if they are not at the root (reason: if something
2494 * cared enough to change them, it's probably wrong to force them elsewhere)
2495 * - it's okay to pick a root that isn't the root of a file system, e.g.
2496 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2497 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2500 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2501 const char __user *, put_old)
2503 struct vfsmount *tmp;
2504 struct path new, old, parent_path, root_parent, root;
2507 if (!capable(CAP_SYS_ADMIN))
2510 error = user_path_dir(new_root, &new);
2514 if (!check_mnt(new.mnt))
2517 error = user_path_dir(put_old, &old);
2521 error = security_sb_pivotroot(&old, &new);
2527 get_fs_root(current->fs, &root);
2528 down_write(&namespace_sem);
2529 mutex_lock(&old.dentry->d_inode->i_mutex);
2531 if (IS_MNT_SHARED(old.mnt) ||
2532 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2533 IS_MNT_SHARED(root.mnt->mnt_parent))
2535 if (!check_mnt(root.mnt))
2538 if (cant_mount(old.dentry))
2540 if (d_unlinked(new.dentry))
2542 if (d_unlinked(old.dentry))
2545 if (new.mnt == root.mnt ||
2546 old.mnt == root.mnt)
2547 goto out2; /* loop, on the same file system */
2549 if (root.mnt->mnt_root != root.dentry)
2550 goto out2; /* not a mountpoint */
2551 if (root.mnt->mnt_parent == root.mnt)
2552 goto out2; /* not attached */
2553 if (new.mnt->mnt_root != new.dentry)
2554 goto out2; /* not a mountpoint */
2555 if (new.mnt->mnt_parent == new.mnt)
2556 goto out2; /* not attached */
2557 /* make sure we can reach put_old from new_root */
2559 br_write_lock(vfsmount_lock);
2560 if (tmp != new.mnt) {
2562 if (tmp->mnt_parent == tmp)
2563 goto out3; /* already mounted on put_old */
2564 if (tmp->mnt_parent == new.mnt)
2566 tmp = tmp->mnt_parent;
2568 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2570 } else if (!is_subdir(old.dentry, new.dentry))
2572 detach_mnt(new.mnt, &parent_path);
2573 detach_mnt(root.mnt, &root_parent);
2574 /* mount old root on put_old */
2575 attach_mnt(root.mnt, &old);
2576 /* mount new_root on / */
2577 attach_mnt(new.mnt, &root_parent);
2578 touch_mnt_namespace(current->nsproxy->mnt_ns);
2579 br_write_unlock(vfsmount_lock);
2580 chroot_fs_refs(&root, &new);
2583 path_put(&root_parent);
2584 path_put(&parent_path);
2586 mutex_unlock(&old.dentry->d_inode->i_mutex);
2587 up_write(&namespace_sem);
2595 br_write_unlock(vfsmount_lock);
2599 static void __init init_mount_tree(void)
2601 struct vfsmount *mnt;
2602 struct mnt_namespace *ns;
2605 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2607 panic("Can't create rootfs");
2609 ns = create_mnt_ns(mnt);
2611 panic("Can't allocate initial namespace");
2613 init_task.nsproxy->mnt_ns = ns;
2616 root.mnt = ns->root;
2617 root.dentry = ns->root->mnt_root;
2619 set_fs_pwd(current->fs, &root);
2620 set_fs_root(current->fs, &root);
2623 void __init mnt_init(void)
2628 init_rwsem(&namespace_sem);
2630 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2631 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2633 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2635 if (!mount_hashtable)
2636 panic("Failed to allocate mount hash table\n");
2638 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2640 for (u = 0; u < HASH_SIZE; u++)
2641 INIT_LIST_HEAD(&mount_hashtable[u]);
2643 br_lock_init(vfsmount_lock);
2647 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2649 fs_kobj = kobject_create_and_add("fs", NULL);
2651 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2656 void put_mnt_ns(struct mnt_namespace *ns)
2658 LIST_HEAD(umount_list);
2660 if (!atomic_dec_and_test(&ns->count))
2662 down_write(&namespace_sem);
2663 br_write_lock(vfsmount_lock);
2664 umount_tree(ns->root, 0, &umount_list);
2665 br_write_unlock(vfsmount_lock);
2666 up_write(&namespace_sem);
2667 release_mounts(&umount_list);
2670 EXPORT_SYMBOL(put_mnt_ns);