4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/namei.h>
16 #include <linux/security.h>
17 #include <linux/idr.h>
18 #include <linux/acct.h> /* acct_auto_close_mnt */
19 #include <linux/ramfs.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
26 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
27 #define HASH_SIZE (1UL << HASH_SHIFT)
30 static DEFINE_IDA(mnt_id_ida);
31 static DEFINE_IDA(mnt_group_ida);
32 static DEFINE_SPINLOCK(mnt_id_lock);
33 static int mnt_id_start = 0;
34 static int mnt_group_start = 1;
36 static struct list_head *mount_hashtable __read_mostly;
37 static struct kmem_cache *mnt_cache __read_mostly;
38 static struct rw_semaphore namespace_sem;
41 struct kobject *fs_kobj;
42 EXPORT_SYMBOL_GPL(fs_kobj);
45 * vfsmount lock may be taken for read to prevent changes to the
46 * vfsmount hash, ie. during mountpoint lookups or walking back
49 * It should be taken for write in all cases where the vfsmount
50 * tree or hash is modified or when a vfsmount structure is modified.
52 DEFINE_BRLOCK(vfsmount_lock);
54 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
58 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
62 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
65 * allocation is serialized by namespace_sem, but we need the spinlock to
66 * serialize with freeing.
68 static int mnt_alloc_id(struct mount *mnt)
73 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
74 spin_lock(&mnt_id_lock);
75 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
77 mnt_id_start = mnt->mnt_id + 1;
78 spin_unlock(&mnt_id_lock);
85 static void mnt_free_id(struct mount *mnt)
88 spin_lock(&mnt_id_lock);
89 ida_remove(&mnt_id_ida, id);
90 if (mnt_id_start > id)
92 spin_unlock(&mnt_id_lock);
96 * Allocate a new peer group ID
98 * mnt_group_ida is protected by namespace_sem
100 static int mnt_alloc_group_id(struct mount *mnt)
104 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107 res = ida_get_new_above(&mnt_group_ida,
111 mnt_group_start = mnt->mnt_group_id + 1;
117 * Release a peer group ID
119 void mnt_release_group_id(struct mount *mnt)
121 int id = mnt->mnt_group_id;
122 ida_remove(&mnt_group_ida, id);
123 if (mnt_group_start > id)
124 mnt_group_start = id;
125 mnt->mnt_group_id = 0;
129 * vfsmount lock must be held for read
131 static inline void mnt_add_count(struct mount *mnt, int n)
134 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
143 * vfsmount lock must be held for write
145 unsigned int mnt_get_count(struct mount *mnt)
148 unsigned int count = 0;
151 for_each_possible_cpu(cpu) {
152 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
157 return mnt->mnt_count;
161 static struct mount *alloc_vfsmnt(const char *name)
163 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
167 err = mnt_alloc_id(mnt);
172 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
173 if (!mnt->mnt_devname)
178 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
180 goto out_free_devname;
182 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
185 mnt->mnt_writers = 0;
188 INIT_LIST_HEAD(&mnt->mnt_hash);
189 INIT_LIST_HEAD(&mnt->mnt_child);
190 INIT_LIST_HEAD(&mnt->mnt_mounts);
191 INIT_LIST_HEAD(&mnt->mnt_list);
192 INIT_LIST_HEAD(&mnt->mnt_expire);
193 INIT_LIST_HEAD(&mnt->mnt_share);
194 INIT_LIST_HEAD(&mnt->mnt_slave_list);
195 INIT_LIST_HEAD(&mnt->mnt_slave);
196 #ifdef CONFIG_FSNOTIFY
197 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
204 kfree(mnt->mnt_devname);
209 kmem_cache_free(mnt_cache, mnt);
214 * Most r/o checks on a fs are for operations that take
215 * discrete amounts of time, like a write() or unlink().
216 * We must keep track of when those operations start
217 * (for permission checks) and when they end, so that
218 * we can determine when writes are able to occur to
222 * __mnt_is_readonly: check whether a mount is read-only
223 * @mnt: the mount to check for its write status
225 * This shouldn't be used directly ouside of the VFS.
226 * It does not guarantee that the filesystem will stay
227 * r/w, just that it is right *now*. This can not and
228 * should not be used in place of IS_RDONLY(inode).
229 * mnt_want/drop_write() will _keep_ the filesystem
232 int __mnt_is_readonly(struct vfsmount *mnt)
234 if (mnt->mnt_flags & MNT_READONLY)
236 if (mnt->mnt_sb->s_flags & MS_RDONLY)
240 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
242 static inline void mnt_inc_writers(struct mount *mnt)
245 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
251 static inline void mnt_dec_writers(struct mount *mnt)
254 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
260 static unsigned int mnt_get_writers(struct mount *mnt)
263 unsigned int count = 0;
266 for_each_possible_cpu(cpu) {
267 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
272 return mnt->mnt_writers;
276 static int mnt_is_readonly(struct vfsmount *mnt)
278 if (mnt->mnt_sb->s_readonly_remount)
280 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
282 return __mnt_is_readonly(mnt);
286 * Most r/o checks on a fs are for operations that take
287 * discrete amounts of time, like a write() or unlink().
288 * We must keep track of when those operations start
289 * (for permission checks) and when they end, so that
290 * we can determine when writes are able to occur to
294 * mnt_want_write - get write access to a mount
295 * @m: the mount on which to take a write
297 * This tells the low-level filesystem that a write is
298 * about to be performed to it, and makes sure that
299 * writes are allowed before returning success. When
300 * the write operation is finished, mnt_drop_write()
301 * must be called. This is effectively a refcount.
303 int mnt_want_write(struct vfsmount *m)
305 struct mount *mnt = real_mount(m);
309 mnt_inc_writers(mnt);
311 * The store to mnt_inc_writers must be visible before we pass
312 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
313 * incremented count after it has set MNT_WRITE_HOLD.
316 while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
319 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
320 * be set to match its requirements. So we must not load that until
321 * MNT_WRITE_HOLD is cleared.
324 if (mnt_is_readonly(m)) {
325 mnt_dec_writers(mnt);
331 EXPORT_SYMBOL_GPL(mnt_want_write);
334 * mnt_clone_write - get write access to a mount
335 * @mnt: the mount on which to take a write
337 * This is effectively like mnt_want_write, except
338 * it must only be used to take an extra write reference
339 * on a mountpoint that we already know has a write reference
340 * on it. This allows some optimisation.
342 * After finished, mnt_drop_write must be called as usual to
343 * drop the reference.
345 int mnt_clone_write(struct vfsmount *mnt)
347 /* superblock may be r/o */
348 if (__mnt_is_readonly(mnt))
351 mnt_inc_writers(real_mount(mnt));
355 EXPORT_SYMBOL_GPL(mnt_clone_write);
358 * mnt_want_write_file - get write access to a file's mount
359 * @file: the file who's mount on which to take a write
361 * This is like mnt_want_write, but it takes a file and can
362 * do some optimisations if the file is open for write already
364 int mnt_want_write_file(struct file *file)
366 struct inode *inode = file->f_dentry->d_inode;
367 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
368 return mnt_want_write(file->f_path.mnt);
370 return mnt_clone_write(file->f_path.mnt);
372 EXPORT_SYMBOL_GPL(mnt_want_write_file);
375 * mnt_drop_write - give up write access to a mount
376 * @mnt: the mount on which to give up write access
378 * Tells the low-level filesystem that we are done
379 * performing writes to it. Must be matched with
380 * mnt_want_write() call above.
382 void mnt_drop_write(struct vfsmount *mnt)
385 mnt_dec_writers(real_mount(mnt));
388 EXPORT_SYMBOL_GPL(mnt_drop_write);
390 void mnt_drop_write_file(struct file *file)
392 mnt_drop_write(file->f_path.mnt);
394 EXPORT_SYMBOL(mnt_drop_write_file);
396 static int mnt_make_readonly(struct mount *mnt)
400 br_write_lock(&vfsmount_lock);
401 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
403 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
404 * should be visible before we do.
409 * With writers on hold, if this value is zero, then there are
410 * definitely no active writers (although held writers may subsequently
411 * increment the count, they'll have to wait, and decrement it after
412 * seeing MNT_READONLY).
414 * It is OK to have counter incremented on one CPU and decremented on
415 * another: the sum will add up correctly. The danger would be when we
416 * sum up each counter, if we read a counter before it is incremented,
417 * but then read another CPU's count which it has been subsequently
418 * decremented from -- we would see more decrements than we should.
419 * MNT_WRITE_HOLD protects against this scenario, because
420 * mnt_want_write first increments count, then smp_mb, then spins on
421 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
422 * we're counting up here.
424 if (mnt_get_writers(mnt) > 0)
427 mnt->mnt.mnt_flags |= MNT_READONLY;
429 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
430 * that become unheld will see MNT_READONLY.
433 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
434 br_write_unlock(&vfsmount_lock);
438 static void __mnt_unmake_readonly(struct mount *mnt)
440 br_write_lock(&vfsmount_lock);
441 mnt->mnt.mnt_flags &= ~MNT_READONLY;
442 br_write_unlock(&vfsmount_lock);
445 int sb_prepare_remount_readonly(struct super_block *sb)
450 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
451 if (atomic_long_read(&sb->s_remove_count))
454 br_write_lock(&vfsmount_lock);
455 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
456 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
457 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
459 if (mnt_get_writers(mnt) > 0) {
465 if (!err && atomic_long_read(&sb->s_remove_count))
469 sb->s_readonly_remount = 1;
472 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
473 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
474 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
476 br_write_unlock(&vfsmount_lock);
481 static void free_vfsmnt(struct mount *mnt)
483 kfree(mnt->mnt_devname);
486 free_percpu(mnt->mnt_pcp);
488 kmem_cache_free(mnt_cache, mnt);
492 * find the first or last mount at @dentry on vfsmount @mnt depending on
493 * @dir. If @dir is set return the first mount else return the last mount.
494 * vfsmount_lock must be held for read or write.
496 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
499 struct list_head *head = mount_hashtable + hash(mnt, dentry);
500 struct list_head *tmp = head;
501 struct mount *p, *found = NULL;
504 tmp = dir ? tmp->next : tmp->prev;
508 p = list_entry(tmp, struct mount, mnt_hash);
509 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
518 * lookup_mnt increments the ref count before returning
519 * the vfsmount struct.
521 struct vfsmount *lookup_mnt(struct path *path)
523 struct mount *child_mnt;
525 br_read_lock(&vfsmount_lock);
526 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
528 mnt_add_count(child_mnt, 1);
529 br_read_unlock(&vfsmount_lock);
530 return &child_mnt->mnt;
532 br_read_unlock(&vfsmount_lock);
537 static inline int check_mnt(struct mount *mnt)
539 return mnt->mnt_ns == current->nsproxy->mnt_ns;
543 * vfsmount lock must be held for write
545 static void touch_mnt_namespace(struct mnt_namespace *ns)
549 wake_up_interruptible(&ns->poll);
554 * vfsmount lock must be held for write
556 static void __touch_mnt_namespace(struct mnt_namespace *ns)
558 if (ns && ns->event != event) {
560 wake_up_interruptible(&ns->poll);
565 * Clear dentry's mounted state if it has no remaining mounts.
566 * vfsmount_lock must be held for write.
568 static void dentry_reset_mounted(struct dentry *dentry)
572 for (u = 0; u < HASH_SIZE; u++) {
575 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
576 if (p->mnt_mountpoint == dentry)
580 spin_lock(&dentry->d_lock);
581 dentry->d_flags &= ~DCACHE_MOUNTED;
582 spin_unlock(&dentry->d_lock);
586 * vfsmount lock must be held for write
588 static void detach_mnt(struct mount *mnt, struct path *old_path)
590 old_path->dentry = mnt->mnt_mountpoint;
591 old_path->mnt = &mnt->mnt_parent->mnt;
592 mnt->mnt_parent = mnt;
593 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
594 list_del_init(&mnt->mnt_child);
595 list_del_init(&mnt->mnt_hash);
596 dentry_reset_mounted(old_path->dentry);
600 * vfsmount lock must be held for write
602 void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
603 struct mount *child_mnt)
605 mnt_add_count(mnt, 1); /* essentially, that's mntget */
606 child_mnt->mnt_mountpoint = dget(dentry);
607 child_mnt->mnt_parent = mnt;
608 spin_lock(&dentry->d_lock);
609 dentry->d_flags |= DCACHE_MOUNTED;
610 spin_unlock(&dentry->d_lock);
614 * vfsmount lock must be held for write
616 static void attach_mnt(struct mount *mnt, struct path *path)
618 mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
619 list_add_tail(&mnt->mnt_hash, mount_hashtable +
620 hash(path->mnt, path->dentry));
621 list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
625 * vfsmount lock must be held for write
627 static void commit_tree(struct mount *mnt)
629 struct mount *parent = mnt->mnt_parent;
632 struct mnt_namespace *n = parent->mnt_ns;
634 BUG_ON(parent == mnt);
636 list_add_tail(&head, &mnt->mnt_list);
637 list_for_each_entry(m, &head, mnt_list)
640 list_splice(&head, n->list.prev);
642 list_add_tail(&mnt->mnt_hash, mount_hashtable +
643 hash(&parent->mnt, mnt->mnt_mountpoint));
644 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
645 touch_mnt_namespace(n);
648 static struct mount *next_mnt(struct mount *p, struct mount *root)
650 struct list_head *next = p->mnt_mounts.next;
651 if (next == &p->mnt_mounts) {
655 next = p->mnt_child.next;
656 if (next != &p->mnt_parent->mnt_mounts)
661 return list_entry(next, struct mount, mnt_child);
664 static struct mount *skip_mnt_tree(struct mount *p)
666 struct list_head *prev = p->mnt_mounts.prev;
667 while (prev != &p->mnt_mounts) {
668 p = list_entry(prev, struct mount, mnt_child);
669 prev = p->mnt_mounts.prev;
675 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
681 return ERR_PTR(-ENODEV);
683 mnt = alloc_vfsmnt(name);
685 return ERR_PTR(-ENOMEM);
687 if (flags & MS_KERNMOUNT)
688 mnt->mnt.mnt_flags = MNT_INTERNAL;
690 root = mount_fs(type, flags, name, data);
693 return ERR_CAST(root);
696 mnt->mnt.mnt_root = root;
697 mnt->mnt.mnt_sb = root->d_sb;
698 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
699 mnt->mnt_parent = mnt;
700 br_write_lock(&vfsmount_lock);
701 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
702 br_write_unlock(&vfsmount_lock);
705 EXPORT_SYMBOL_GPL(vfs_kern_mount);
707 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
710 struct super_block *sb = old->mnt.mnt_sb;
711 struct mount *mnt = alloc_vfsmnt(old->mnt_devname);
714 if (flag & (CL_SLAVE | CL_PRIVATE))
715 mnt->mnt_group_id = 0; /* not a peer of original */
717 mnt->mnt_group_id = old->mnt_group_id;
719 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
720 int err = mnt_alloc_group_id(mnt);
725 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
726 atomic_inc(&sb->s_active);
727 mnt->mnt.mnt_sb = sb;
728 mnt->mnt.mnt_root = dget(root);
729 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
730 mnt->mnt_parent = mnt;
731 br_write_lock(&vfsmount_lock);
732 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
733 br_write_unlock(&vfsmount_lock);
735 if (flag & CL_SLAVE) {
736 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
737 mnt->mnt_master = old;
738 CLEAR_MNT_SHARED(mnt);
739 } else if (!(flag & CL_PRIVATE)) {
740 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
741 list_add(&mnt->mnt_share, &old->mnt_share);
742 if (IS_MNT_SLAVE(old))
743 list_add(&mnt->mnt_slave, &old->mnt_slave);
744 mnt->mnt_master = old->mnt_master;
746 if (flag & CL_MAKE_SHARED)
749 /* stick the duplicate mount on the same expiry list
750 * as the original if that was on one */
751 if (flag & CL_EXPIRE) {
752 if (!list_empty(&old->mnt_expire))
753 list_add(&mnt->mnt_expire, &old->mnt_expire);
763 static inline void mntfree(struct mount *mnt)
765 struct vfsmount *m = &mnt->mnt;
766 struct super_block *sb = m->mnt_sb;
769 * This probably indicates that somebody messed
770 * up a mnt_want/drop_write() pair. If this
771 * happens, the filesystem was probably unable
772 * to make r/w->r/o transitions.
775 * The locking used to deal with mnt_count decrement provides barriers,
776 * so mnt_get_writers() below is safe.
778 WARN_ON(mnt_get_writers(mnt));
779 fsnotify_vfsmount_delete(m);
782 deactivate_super(sb);
785 static void mntput_no_expire(struct mount *mnt)
789 br_read_lock(&vfsmount_lock);
790 if (likely(mnt->mnt_ns)) {
791 /* shouldn't be the last one */
792 mnt_add_count(mnt, -1);
793 br_read_unlock(&vfsmount_lock);
796 br_read_unlock(&vfsmount_lock);
798 br_write_lock(&vfsmount_lock);
799 mnt_add_count(mnt, -1);
800 if (mnt_get_count(mnt)) {
801 br_write_unlock(&vfsmount_lock);
805 mnt_add_count(mnt, -1);
806 if (likely(mnt_get_count(mnt)))
808 br_write_lock(&vfsmount_lock);
810 if (unlikely(mnt->mnt_pinned)) {
811 mnt_add_count(mnt, mnt->mnt_pinned + 1);
813 br_write_unlock(&vfsmount_lock);
814 acct_auto_close_mnt(&mnt->mnt);
818 list_del(&mnt->mnt_instance);
819 br_write_unlock(&vfsmount_lock);
823 void mntput(struct vfsmount *mnt)
826 struct mount *m = real_mount(mnt);
827 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
828 if (unlikely(m->mnt_expiry_mark))
829 m->mnt_expiry_mark = 0;
833 EXPORT_SYMBOL(mntput);
835 struct vfsmount *mntget(struct vfsmount *mnt)
838 mnt_add_count(real_mount(mnt), 1);
841 EXPORT_SYMBOL(mntget);
843 void mnt_pin(struct vfsmount *mnt)
845 br_write_lock(&vfsmount_lock);
846 real_mount(mnt)->mnt_pinned++;
847 br_write_unlock(&vfsmount_lock);
849 EXPORT_SYMBOL(mnt_pin);
851 void mnt_unpin(struct vfsmount *m)
853 struct mount *mnt = real_mount(m);
854 br_write_lock(&vfsmount_lock);
855 if (mnt->mnt_pinned) {
856 mnt_add_count(mnt, 1);
859 br_write_unlock(&vfsmount_lock);
861 EXPORT_SYMBOL(mnt_unpin);
863 static inline void mangle(struct seq_file *m, const char *s)
865 seq_escape(m, s, " \t\n\\");
869 * Simple .show_options callback for filesystems which don't want to
870 * implement more complex mount option showing.
872 * See also save_mount_options().
874 int generic_show_options(struct seq_file *m, struct dentry *root)
879 options = rcu_dereference(root->d_sb->s_options);
881 if (options != NULL && options[0]) {
889 EXPORT_SYMBOL(generic_show_options);
892 * If filesystem uses generic_show_options(), this function should be
893 * called from the fill_super() callback.
895 * The .remount_fs callback usually needs to be handled in a special
896 * way, to make sure, that previous options are not overwritten if the
899 * Also note, that if the filesystem's .remount_fs function doesn't
900 * reset all options to their default value, but changes only newly
901 * given options, then the displayed options will not reflect reality
904 void save_mount_options(struct super_block *sb, char *options)
906 BUG_ON(sb->s_options);
907 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
909 EXPORT_SYMBOL(save_mount_options);
911 void replace_mount_options(struct super_block *sb, char *options)
913 char *old = sb->s_options;
914 rcu_assign_pointer(sb->s_options, options);
920 EXPORT_SYMBOL(replace_mount_options);
922 #ifdef CONFIG_PROC_FS
923 /* iterator; we want it to have access to namespace_sem, thus here... */
924 static void *m_start(struct seq_file *m, loff_t *pos)
926 struct proc_mounts *p = proc_mounts(m);
928 down_read(&namespace_sem);
929 return seq_list_start(&p->ns->list, *pos);
932 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
934 struct proc_mounts *p = proc_mounts(m);
936 return seq_list_next(v, &p->ns->list, pos);
939 static void m_stop(struct seq_file *m, void *v)
941 up_read(&namespace_sem);
944 static int m_show(struct seq_file *m, void *v)
946 struct proc_mounts *p = proc_mounts(m);
947 struct mount *r = list_entry(v, struct mount, mnt_list);
948 return p->show(m, &r->mnt);
951 const struct seq_operations mounts_op = {
957 #endif /* CONFIG_PROC_FS */
960 * may_umount_tree - check if a mount tree is busy
961 * @mnt: root of mount tree
963 * This is called to check if a tree of mounts has any
964 * open files, pwds, chroots or sub mounts that are
967 int may_umount_tree(struct vfsmount *m)
969 struct mount *mnt = real_mount(m);
971 int minimum_refs = 0;
975 /* write lock needed for mnt_get_count */
976 br_write_lock(&vfsmount_lock);
977 for (p = mnt; p; p = next_mnt(p, mnt)) {
978 actual_refs += mnt_get_count(p);
981 br_write_unlock(&vfsmount_lock);
983 if (actual_refs > minimum_refs)
989 EXPORT_SYMBOL(may_umount_tree);
992 * may_umount - check if a mount point is busy
993 * @mnt: root of mount
995 * This is called to check if a mount point has any
996 * open files, pwds, chroots or sub mounts. If the
997 * mount has sub mounts this will return busy
998 * regardless of whether the sub mounts are busy.
1000 * Doesn't take quota and stuff into account. IOW, in some cases it will
1001 * give false negatives. The main reason why it's here is that we need
1002 * a non-destructive way to look for easily umountable filesystems.
1004 int may_umount(struct vfsmount *mnt)
1007 down_read(&namespace_sem);
1008 br_write_lock(&vfsmount_lock);
1009 if (propagate_mount_busy(real_mount(mnt), 2))
1011 br_write_unlock(&vfsmount_lock);
1012 up_read(&namespace_sem);
1016 EXPORT_SYMBOL(may_umount);
1018 void release_mounts(struct list_head *head)
1021 while (!list_empty(head)) {
1022 mnt = list_first_entry(head, struct mount, mnt_hash);
1023 list_del_init(&mnt->mnt_hash);
1024 if (mnt_has_parent(mnt)) {
1025 struct dentry *dentry;
1028 br_write_lock(&vfsmount_lock);
1029 dentry = mnt->mnt_mountpoint;
1030 m = mnt->mnt_parent;
1031 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1032 mnt->mnt_parent = mnt;
1034 br_write_unlock(&vfsmount_lock);
1043 * vfsmount lock must be held for write
1044 * namespace_sem must be held for write
1046 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1048 LIST_HEAD(tmp_list);
1051 for (p = mnt; p; p = next_mnt(p, mnt))
1052 list_move(&p->mnt_hash, &tmp_list);
1055 propagate_umount(&tmp_list);
1057 list_for_each_entry(p, &tmp_list, mnt_hash) {
1058 list_del_init(&p->mnt_expire);
1059 list_del_init(&p->mnt_list);
1060 __touch_mnt_namespace(p->mnt_ns);
1062 list_del_init(&p->mnt_child);
1063 if (mnt_has_parent(p)) {
1064 p->mnt_parent->mnt_ghosts++;
1065 dentry_reset_mounted(p->mnt_mountpoint);
1067 change_mnt_propagation(p, MS_PRIVATE);
1069 list_splice(&tmp_list, kill);
1072 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1074 static int do_umount(struct mount *mnt, int flags)
1076 struct super_block *sb = mnt->mnt.mnt_sb;
1078 LIST_HEAD(umount_list);
1080 retval = security_sb_umount(&mnt->mnt, flags);
1085 * Allow userspace to request a mountpoint be expired rather than
1086 * unmounting unconditionally. Unmount only happens if:
1087 * (1) the mark is already set (the mark is cleared by mntput())
1088 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1090 if (flags & MNT_EXPIRE) {
1091 if (&mnt->mnt == current->fs->root.mnt ||
1092 flags & (MNT_FORCE | MNT_DETACH))
1096 * probably don't strictly need the lock here if we examined
1097 * all race cases, but it's a slowpath.
1099 br_write_lock(&vfsmount_lock);
1100 if (mnt_get_count(mnt) != 2) {
1101 br_write_unlock(&vfsmount_lock);
1104 br_write_unlock(&vfsmount_lock);
1106 if (!xchg(&mnt->mnt_expiry_mark, 1))
1111 * If we may have to abort operations to get out of this
1112 * mount, and they will themselves hold resources we must
1113 * allow the fs to do things. In the Unix tradition of
1114 * 'Gee thats tricky lets do it in userspace' the umount_begin
1115 * might fail to complete on the first run through as other tasks
1116 * must return, and the like. Thats for the mount program to worry
1117 * about for the moment.
1120 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1121 sb->s_op->umount_begin(sb);
1125 * No sense to grab the lock for this test, but test itself looks
1126 * somewhat bogus. Suggestions for better replacement?
1127 * Ho-hum... In principle, we might treat that as umount + switch
1128 * to rootfs. GC would eventually take care of the old vfsmount.
1129 * Actually it makes sense, especially if rootfs would contain a
1130 * /reboot - static binary that would close all descriptors and
1131 * call reboot(9). Then init(8) could umount root and exec /reboot.
1133 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1135 * Special case for "unmounting" root ...
1136 * we just try to remount it readonly.
1138 down_write(&sb->s_umount);
1139 if (!(sb->s_flags & MS_RDONLY))
1140 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1141 up_write(&sb->s_umount);
1145 down_write(&namespace_sem);
1146 br_write_lock(&vfsmount_lock);
1149 if (!(flags & MNT_DETACH))
1150 shrink_submounts(mnt, &umount_list);
1153 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1154 if (!list_empty(&mnt->mnt_list))
1155 umount_tree(mnt, 1, &umount_list);
1158 br_write_unlock(&vfsmount_lock);
1159 up_write(&namespace_sem);
1160 release_mounts(&umount_list);
1165 * Now umount can handle mount points as well as block devices.
1166 * This is important for filesystems which use unnamed block devices.
1168 * We now support a flag for forced unmount like the other 'big iron'
1169 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1172 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1177 int lookup_flags = 0;
1179 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1182 if (!(flags & UMOUNT_NOFOLLOW))
1183 lookup_flags |= LOOKUP_FOLLOW;
1185 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1188 mnt = real_mount(path.mnt);
1190 if (path.dentry != path.mnt->mnt_root)
1192 if (!check_mnt(mnt))
1196 if (!capable(CAP_SYS_ADMIN))
1199 retval = do_umount(mnt, flags);
1201 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1203 mntput_no_expire(mnt);
1208 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1211 * The 2.0 compatible umount. No flags.
1213 SYSCALL_DEFINE1(oldumount, char __user *, name)
1215 return sys_umount(name, 0);
1220 static int mount_is_safe(struct path *path)
1222 if (capable(CAP_SYS_ADMIN))
1226 if (S_ISLNK(path->dentry->d_inode->i_mode))
1228 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1229 if (current_uid() != path->dentry->d_inode->i_uid)
1232 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1238 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1241 struct mount *res, *p, *q, *r;
1244 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1247 res = q = clone_mnt(mnt, dentry, flag);
1250 q->mnt_mountpoint = mnt->mnt_mountpoint;
1253 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1255 if (!is_subdir(r->mnt_mountpoint, dentry))
1258 for (s = r; s; s = next_mnt(s, r)) {
1259 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1260 s = skip_mnt_tree(s);
1263 while (p != s->mnt_parent) {
1269 path.dentry = p->mnt_mountpoint;
1270 q = clone_mnt(p, p->mnt.mnt_root, flag);
1273 br_write_lock(&vfsmount_lock);
1274 list_add_tail(&q->mnt_list, &res->mnt_list);
1275 attach_mnt(q, &path);
1276 br_write_unlock(&vfsmount_lock);
1282 LIST_HEAD(umount_list);
1283 br_write_lock(&vfsmount_lock);
1284 umount_tree(res, 0, &umount_list);
1285 br_write_unlock(&vfsmount_lock);
1286 release_mounts(&umount_list);
1291 struct vfsmount *collect_mounts(struct path *path)
1294 down_write(&namespace_sem);
1295 tree = copy_tree(real_mount(path->mnt), path->dentry,
1296 CL_COPY_ALL | CL_PRIVATE);
1297 up_write(&namespace_sem);
1298 return tree ? &tree->mnt : NULL;
1301 void drop_collected_mounts(struct vfsmount *mnt)
1303 LIST_HEAD(umount_list);
1304 down_write(&namespace_sem);
1305 br_write_lock(&vfsmount_lock);
1306 umount_tree(real_mount(mnt), 0, &umount_list);
1307 br_write_unlock(&vfsmount_lock);
1308 up_write(&namespace_sem);
1309 release_mounts(&umount_list);
1312 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1313 struct vfsmount *root)
1316 int res = f(root, arg);
1319 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1320 res = f(&mnt->mnt, arg);
1327 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1331 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1332 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1333 mnt_release_group_id(p);
1337 static int invent_group_ids(struct mount *mnt, bool recurse)
1341 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1342 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1343 int err = mnt_alloc_group_id(p);
1345 cleanup_group_ids(mnt, p);
1355 * @source_mnt : mount tree to be attached
1356 * @nd : place the mount tree @source_mnt is attached
1357 * @parent_nd : if non-null, detach the source_mnt from its parent and
1358 * store the parent mount and mountpoint dentry.
1359 * (done when source_mnt is moved)
1361 * NOTE: in the table below explains the semantics when a source mount
1362 * of a given type is attached to a destination mount of a given type.
1363 * ---------------------------------------------------------------------------
1364 * | BIND MOUNT OPERATION |
1365 * |**************************************************************************
1366 * | source-->| shared | private | slave | unbindable |
1370 * |**************************************************************************
1371 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1373 * |non-shared| shared (+) | private | slave (*) | invalid |
1374 * ***************************************************************************
1375 * A bind operation clones the source mount and mounts the clone on the
1376 * destination mount.
1378 * (++) the cloned mount is propagated to all the mounts in the propagation
1379 * tree of the destination mount and the cloned mount is added to
1380 * the peer group of the source mount.
1381 * (+) the cloned mount is created under the destination mount and is marked
1382 * as shared. The cloned mount is added to the peer group of the source
1384 * (+++) the mount is propagated to all the mounts in the propagation tree
1385 * of the destination mount and the cloned mount is made slave
1386 * of the same master as that of the source mount. The cloned mount
1387 * is marked as 'shared and slave'.
1388 * (*) the cloned mount is made a slave of the same master as that of the
1391 * ---------------------------------------------------------------------------
1392 * | MOVE MOUNT OPERATION |
1393 * |**************************************************************************
1394 * | source-->| shared | private | slave | unbindable |
1398 * |**************************************************************************
1399 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1401 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1402 * ***************************************************************************
1404 * (+) the mount is moved to the destination. And is then propagated to
1405 * all the mounts in the propagation tree of the destination mount.
1406 * (+*) the mount is moved to the destination.
1407 * (+++) the mount is moved to the destination and is then propagated to
1408 * all the mounts belonging to the destination mount's propagation tree.
1409 * the mount is marked as 'shared and slave'.
1410 * (*) the mount continues to be a slave at the new location.
1412 * if the source mount is a tree, the operations explained above is
1413 * applied to each mount in the tree.
1414 * Must be called without spinlocks held, since this function can sleep
1417 static int attach_recursive_mnt(struct mount *source_mnt,
1418 struct path *path, struct path *parent_path)
1420 LIST_HEAD(tree_list);
1421 struct mount *dest_mnt = real_mount(path->mnt);
1422 struct dentry *dest_dentry = path->dentry;
1423 struct mount *child, *p;
1426 if (IS_MNT_SHARED(dest_mnt)) {
1427 err = invent_group_ids(source_mnt, true);
1431 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1433 goto out_cleanup_ids;
1435 br_write_lock(&vfsmount_lock);
1437 if (IS_MNT_SHARED(dest_mnt)) {
1438 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1442 detach_mnt(source_mnt, parent_path);
1443 attach_mnt(source_mnt, path);
1444 touch_mnt_namespace(source_mnt->mnt_ns);
1446 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1447 commit_tree(source_mnt);
1450 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1451 list_del_init(&child->mnt_hash);
1454 br_write_unlock(&vfsmount_lock);
1459 if (IS_MNT_SHARED(dest_mnt))
1460 cleanup_group_ids(source_mnt, NULL);
1465 static int lock_mount(struct path *path)
1467 struct vfsmount *mnt;
1469 mutex_lock(&path->dentry->d_inode->i_mutex);
1470 if (unlikely(cant_mount(path->dentry))) {
1471 mutex_unlock(&path->dentry->d_inode->i_mutex);
1474 down_write(&namespace_sem);
1475 mnt = lookup_mnt(path);
1478 up_write(&namespace_sem);
1479 mutex_unlock(&path->dentry->d_inode->i_mutex);
1482 path->dentry = dget(mnt->mnt_root);
1486 static void unlock_mount(struct path *path)
1488 up_write(&namespace_sem);
1489 mutex_unlock(&path->dentry->d_inode->i_mutex);
1492 static int graft_tree(struct mount *mnt, struct path *path)
1494 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1497 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1498 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1501 if (d_unlinked(path->dentry))
1504 return attach_recursive_mnt(mnt, path, NULL);
1508 * Sanity check the flags to change_mnt_propagation.
1511 static int flags_to_propagation_type(int flags)
1513 int type = flags & ~(MS_REC | MS_SILENT);
1515 /* Fail if any non-propagation flags are set */
1516 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1518 /* Only one propagation flag should be set */
1519 if (!is_power_of_2(type))
1525 * recursively change the type of the mountpoint.
1527 static int do_change_type(struct path *path, int flag)
1530 struct mount *mnt = real_mount(path->mnt);
1531 int recurse = flag & MS_REC;
1535 if (!capable(CAP_SYS_ADMIN))
1538 if (path->dentry != path->mnt->mnt_root)
1541 type = flags_to_propagation_type(flag);
1545 down_write(&namespace_sem);
1546 if (type == MS_SHARED) {
1547 err = invent_group_ids(mnt, recurse);
1552 br_write_lock(&vfsmount_lock);
1553 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1554 change_mnt_propagation(m, type);
1555 br_write_unlock(&vfsmount_lock);
1558 up_write(&namespace_sem);
1563 * do loopback mount.
1565 static int do_loopback(struct path *path, char *old_name,
1568 LIST_HEAD(umount_list);
1569 struct path old_path;
1570 struct mount *mnt = NULL, *old;
1571 int err = mount_is_safe(path);
1574 if (!old_name || !*old_name)
1576 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1580 err = lock_mount(path);
1584 old = real_mount(old_path.mnt);
1587 if (IS_MNT_UNBINDABLE(old))
1590 if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1595 mnt = copy_tree(old, old_path.dentry, 0);
1597 mnt = clone_mnt(old, old_path.dentry, 0);
1602 err = graft_tree(mnt, path);
1604 br_write_lock(&vfsmount_lock);
1605 umount_tree(mnt, 0, &umount_list);
1606 br_write_unlock(&vfsmount_lock);
1610 release_mounts(&umount_list);
1612 path_put(&old_path);
1616 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1619 int readonly_request = 0;
1621 if (ms_flags & MS_RDONLY)
1622 readonly_request = 1;
1623 if (readonly_request == __mnt_is_readonly(mnt))
1626 if (readonly_request)
1627 error = mnt_make_readonly(real_mount(mnt));
1629 __mnt_unmake_readonly(real_mount(mnt));
1634 * change filesystem flags. dir should be a physical root of filesystem.
1635 * If you've mounted a non-root directory somewhere and want to do remount
1636 * on it - tough luck.
1638 static int do_remount(struct path *path, int flags, int mnt_flags,
1642 struct super_block *sb = path->mnt->mnt_sb;
1643 struct mount *mnt = real_mount(path->mnt);
1645 if (!capable(CAP_SYS_ADMIN))
1648 if (!check_mnt(mnt))
1651 if (path->dentry != path->mnt->mnt_root)
1654 err = security_sb_remount(sb, data);
1658 down_write(&sb->s_umount);
1659 if (flags & MS_BIND)
1660 err = change_mount_flags(path->mnt, flags);
1662 err = do_remount_sb(sb, flags, data, 0);
1664 br_write_lock(&vfsmount_lock);
1665 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1666 mnt->mnt.mnt_flags = mnt_flags;
1667 br_write_unlock(&vfsmount_lock);
1669 up_write(&sb->s_umount);
1671 br_write_lock(&vfsmount_lock);
1672 touch_mnt_namespace(mnt->mnt_ns);
1673 br_write_unlock(&vfsmount_lock);
1678 static inline int tree_contains_unbindable(struct mount *mnt)
1681 for (p = mnt; p; p = next_mnt(p, mnt)) {
1682 if (IS_MNT_UNBINDABLE(p))
1688 static int do_move_mount(struct path *path, char *old_name)
1690 struct path old_path, parent_path;
1694 if (!capable(CAP_SYS_ADMIN))
1696 if (!old_name || !*old_name)
1698 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1702 err = lock_mount(path);
1706 old = real_mount(old_path.mnt);
1707 p = real_mount(path->mnt);
1710 if (!check_mnt(p) || !check_mnt(old))
1713 if (d_unlinked(path->dentry))
1717 if (old_path.dentry != old_path.mnt->mnt_root)
1720 if (!mnt_has_parent(old))
1723 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1724 S_ISDIR(old_path.dentry->d_inode->i_mode))
1727 * Don't move a mount residing in a shared parent.
1729 if (IS_MNT_SHARED(old->mnt_parent))
1732 * Don't move a mount tree containing unbindable mounts to a destination
1733 * mount which is shared.
1735 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1738 for (; mnt_has_parent(p); p = p->mnt_parent)
1742 err = attach_recursive_mnt(old, path, &parent_path);
1746 /* if the mount is moved, it should no longer be expire
1748 list_del_init(&old->mnt_expire);
1753 path_put(&parent_path);
1754 path_put(&old_path);
1758 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1761 const char *subtype = strchr(fstype, '.');
1770 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1772 if (!mnt->mnt_sb->s_subtype)
1778 return ERR_PTR(err);
1781 static struct vfsmount *
1782 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1784 struct file_system_type *type = get_fs_type(fstype);
1785 struct vfsmount *mnt;
1787 return ERR_PTR(-ENODEV);
1788 mnt = vfs_kern_mount(type, flags, name, data);
1789 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1790 !mnt->mnt_sb->s_subtype)
1791 mnt = fs_set_subtype(mnt, fstype);
1792 put_filesystem(type);
1797 * add a mount into a namespace's mount tree
1799 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1803 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1805 err = lock_mount(path);
1810 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1813 /* Refuse the same filesystem on the same mount point */
1815 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1816 path->mnt->mnt_root == path->dentry)
1820 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1823 newmnt->mnt.mnt_flags = mnt_flags;
1824 err = graft_tree(newmnt, path);
1832 * create a new mount for userspace and request it to be added into the
1835 static int do_new_mount(struct path *path, char *type, int flags,
1836 int mnt_flags, char *name, void *data)
1838 struct vfsmount *mnt;
1844 /* we need capabilities... */
1845 if (!capable(CAP_SYS_ADMIN))
1848 mnt = do_kern_mount(type, flags, name, data);
1850 return PTR_ERR(mnt);
1852 err = do_add_mount(real_mount(mnt), path, mnt_flags);
1858 int finish_automount(struct vfsmount *m, struct path *path)
1860 struct mount *mnt = real_mount(m);
1862 /* The new mount record should have at least 2 refs to prevent it being
1863 * expired before we get a chance to add it
1865 BUG_ON(mnt_get_count(mnt) < 2);
1867 if (m->mnt_sb == path->mnt->mnt_sb &&
1868 m->mnt_root == path->dentry) {
1873 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1877 /* remove m from any expiration list it may be on */
1878 if (!list_empty(&mnt->mnt_expire)) {
1879 down_write(&namespace_sem);
1880 br_write_lock(&vfsmount_lock);
1881 list_del_init(&mnt->mnt_expire);
1882 br_write_unlock(&vfsmount_lock);
1883 up_write(&namespace_sem);
1891 * mnt_set_expiry - Put a mount on an expiration list
1892 * @mnt: The mount to list.
1893 * @expiry_list: The list to add the mount to.
1895 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1897 down_write(&namespace_sem);
1898 br_write_lock(&vfsmount_lock);
1900 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1902 br_write_unlock(&vfsmount_lock);
1903 up_write(&namespace_sem);
1905 EXPORT_SYMBOL(mnt_set_expiry);
1908 * process a list of expirable mountpoints with the intent of discarding any
1909 * mountpoints that aren't in use and haven't been touched since last we came
1912 void mark_mounts_for_expiry(struct list_head *mounts)
1914 struct mount *mnt, *next;
1915 LIST_HEAD(graveyard);
1918 if (list_empty(mounts))
1921 down_write(&namespace_sem);
1922 br_write_lock(&vfsmount_lock);
1924 /* extract from the expiration list every vfsmount that matches the
1925 * following criteria:
1926 * - only referenced by its parent vfsmount
1927 * - still marked for expiry (marked on the last call here; marks are
1928 * cleared by mntput())
1930 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1931 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1932 propagate_mount_busy(mnt, 1))
1934 list_move(&mnt->mnt_expire, &graveyard);
1936 while (!list_empty(&graveyard)) {
1937 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
1938 touch_mnt_namespace(mnt->mnt_ns);
1939 umount_tree(mnt, 1, &umounts);
1941 br_write_unlock(&vfsmount_lock);
1942 up_write(&namespace_sem);
1944 release_mounts(&umounts);
1947 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1950 * Ripoff of 'select_parent()'
1952 * search the list of submounts for a given mountpoint, and move any
1953 * shrinkable submounts to the 'graveyard' list.
1955 static int select_submounts(struct mount *parent, struct list_head *graveyard)
1957 struct mount *this_parent = parent;
1958 struct list_head *next;
1962 next = this_parent->mnt_mounts.next;
1964 while (next != &this_parent->mnt_mounts) {
1965 struct list_head *tmp = next;
1966 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
1969 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
1972 * Descend a level if the d_mounts list is non-empty.
1974 if (!list_empty(&mnt->mnt_mounts)) {
1979 if (!propagate_mount_busy(mnt, 1)) {
1980 list_move_tail(&mnt->mnt_expire, graveyard);
1985 * All done at this level ... ascend and resume the search
1987 if (this_parent != parent) {
1988 next = this_parent->mnt_child.next;
1989 this_parent = this_parent->mnt_parent;
1996 * process a list of expirable mountpoints with the intent of discarding any
1997 * submounts of a specific parent mountpoint
1999 * vfsmount_lock must be held for write
2001 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2003 LIST_HEAD(graveyard);
2006 /* extract submounts of 'mountpoint' from the expiration list */
2007 while (select_submounts(mnt, &graveyard)) {
2008 while (!list_empty(&graveyard)) {
2009 m = list_first_entry(&graveyard, struct mount,
2011 touch_mnt_namespace(m->mnt_ns);
2012 umount_tree(m, 1, umounts);
2018 * Some copy_from_user() implementations do not return the exact number of
2019 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2020 * Note that this function differs from copy_from_user() in that it will oops
2021 * on bad values of `to', rather than returning a short copy.
2023 static long exact_copy_from_user(void *to, const void __user * from,
2027 const char __user *f = from;
2030 if (!access_ok(VERIFY_READ, from, n))
2034 if (__get_user(c, f)) {
2045 int copy_mount_options(const void __user * data, unsigned long *where)
2055 if (!(page = __get_free_page(GFP_KERNEL)))
2058 /* We only care that *some* data at the address the user
2059 * gave us is valid. Just in case, we'll zero
2060 * the remainder of the page.
2062 /* copy_from_user cannot cross TASK_SIZE ! */
2063 size = TASK_SIZE - (unsigned long)data;
2064 if (size > PAGE_SIZE)
2067 i = size - exact_copy_from_user((void *)page, data, size);
2073 memset((char *)page + i, 0, PAGE_SIZE - i);
2078 int copy_mount_string(const void __user *data, char **where)
2087 tmp = strndup_user(data, PAGE_SIZE);
2089 return PTR_ERR(tmp);
2096 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2097 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2099 * data is a (void *) that can point to any structure up to
2100 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2101 * information (or be NULL).
2103 * Pre-0.97 versions of mount() didn't have a flags word.
2104 * When the flags word was introduced its top half was required
2105 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2106 * Therefore, if this magic number is present, it carries no information
2107 * and must be discarded.
2109 long do_mount(char *dev_name, char *dir_name, char *type_page,
2110 unsigned long flags, void *data_page)
2117 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2118 flags &= ~MS_MGC_MSK;
2120 /* Basic sanity checks */
2122 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2126 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2128 /* ... and get the mountpoint */
2129 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2133 retval = security_sb_mount(dev_name, &path,
2134 type_page, flags, data_page);
2138 /* Default to relatime unless overriden */
2139 if (!(flags & MS_NOATIME))
2140 mnt_flags |= MNT_RELATIME;
2142 /* Separate the per-mountpoint flags */
2143 if (flags & MS_NOSUID)
2144 mnt_flags |= MNT_NOSUID;
2145 if (flags & MS_NODEV)
2146 mnt_flags |= MNT_NODEV;
2147 if (flags & MS_NOEXEC)
2148 mnt_flags |= MNT_NOEXEC;
2149 if (flags & MS_NOATIME)
2150 mnt_flags |= MNT_NOATIME;
2151 if (flags & MS_NODIRATIME)
2152 mnt_flags |= MNT_NODIRATIME;
2153 if (flags & MS_STRICTATIME)
2154 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2155 if (flags & MS_RDONLY)
2156 mnt_flags |= MNT_READONLY;
2158 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2159 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2162 if (flags & MS_REMOUNT)
2163 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2165 else if (flags & MS_BIND)
2166 retval = do_loopback(&path, dev_name, flags & MS_REC);
2167 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2168 retval = do_change_type(&path, flags);
2169 else if (flags & MS_MOVE)
2170 retval = do_move_mount(&path, dev_name);
2172 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2173 dev_name, data_page);
2179 static struct mnt_namespace *alloc_mnt_ns(void)
2181 struct mnt_namespace *new_ns;
2183 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2185 return ERR_PTR(-ENOMEM);
2186 atomic_set(&new_ns->count, 1);
2187 new_ns->root = NULL;
2188 INIT_LIST_HEAD(&new_ns->list);
2189 init_waitqueue_head(&new_ns->poll);
2195 * Allocate a new namespace structure and populate it with contents
2196 * copied from the namespace of the passed in task structure.
2198 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2199 struct fs_struct *fs)
2201 struct mnt_namespace *new_ns;
2202 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2203 struct mount *p, *q;
2204 struct mount *old = mnt_ns->root;
2207 new_ns = alloc_mnt_ns();
2211 down_write(&namespace_sem);
2212 /* First pass: copy the tree topology */
2213 new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2215 up_write(&namespace_sem);
2217 return ERR_PTR(-ENOMEM);
2220 br_write_lock(&vfsmount_lock);
2221 list_add_tail(&new_ns->list, &new->mnt_list);
2222 br_write_unlock(&vfsmount_lock);
2225 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2226 * as belonging to new namespace. We have already acquired a private
2227 * fs_struct, so tsk->fs->lock is not needed.
2234 if (&p->mnt == fs->root.mnt) {
2235 fs->root.mnt = mntget(&q->mnt);
2238 if (&p->mnt == fs->pwd.mnt) {
2239 fs->pwd.mnt = mntget(&q->mnt);
2243 p = next_mnt(p, old);
2244 q = next_mnt(q, new);
2246 up_write(&namespace_sem);
2256 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2257 struct fs_struct *new_fs)
2259 struct mnt_namespace *new_ns;
2264 if (!(flags & CLONE_NEWNS))
2267 new_ns = dup_mnt_ns(ns, new_fs);
2274 * create_mnt_ns - creates a private namespace and adds a root filesystem
2275 * @mnt: pointer to the new root filesystem mountpoint
2277 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2279 struct mnt_namespace *new_ns = alloc_mnt_ns();
2280 if (!IS_ERR(new_ns)) {
2281 struct mount *mnt = real_mount(m);
2282 mnt->mnt_ns = new_ns;
2284 list_add(&new_ns->list, &mnt->mnt_list);
2291 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2293 struct mnt_namespace *ns;
2294 struct super_block *s;
2298 ns = create_mnt_ns(mnt);
2300 return ERR_CAST(ns);
2302 err = vfs_path_lookup(mnt->mnt_root, mnt,
2303 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2308 return ERR_PTR(err);
2310 /* trade a vfsmount reference for active sb one */
2311 s = path.mnt->mnt_sb;
2312 atomic_inc(&s->s_active);
2314 /* lock the sucker */
2315 down_write(&s->s_umount);
2316 /* ... and return the root of (sub)tree on it */
2319 EXPORT_SYMBOL(mount_subtree);
2321 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2322 char __user *, type, unsigned long, flags, void __user *, data)
2328 unsigned long data_page;
2330 ret = copy_mount_string(type, &kernel_type);
2334 kernel_dir = getname(dir_name);
2335 if (IS_ERR(kernel_dir)) {
2336 ret = PTR_ERR(kernel_dir);
2340 ret = copy_mount_string(dev_name, &kernel_dev);
2344 ret = copy_mount_options(data, &data_page);
2348 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2349 (void *) data_page);
2351 free_page(data_page);
2355 putname(kernel_dir);
2363 * Return true if path is reachable from root
2365 * namespace_sem or vfsmount_lock is held
2367 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2368 const struct path *root)
2370 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2371 dentry = mnt->mnt_mountpoint;
2372 mnt = mnt->mnt_parent;
2374 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2377 int path_is_under(struct path *path1, struct path *path2)
2380 br_read_lock(&vfsmount_lock);
2381 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2382 br_read_unlock(&vfsmount_lock);
2385 EXPORT_SYMBOL(path_is_under);
2388 * pivot_root Semantics:
2389 * Moves the root file system of the current process to the directory put_old,
2390 * makes new_root as the new root file system of the current process, and sets
2391 * root/cwd of all processes which had them on the current root to new_root.
2394 * The new_root and put_old must be directories, and must not be on the
2395 * same file system as the current process root. The put_old must be
2396 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2397 * pointed to by put_old must yield the same directory as new_root. No other
2398 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2400 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2401 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2402 * in this situation.
2405 * - we don't move root/cwd if they are not at the root (reason: if something
2406 * cared enough to change them, it's probably wrong to force them elsewhere)
2407 * - it's okay to pick a root that isn't the root of a file system, e.g.
2408 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2409 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2412 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2413 const char __user *, put_old)
2415 struct path new, old, parent_path, root_parent, root;
2416 struct mount *new_mnt, *root_mnt;
2419 if (!capable(CAP_SYS_ADMIN))
2422 error = user_path_dir(new_root, &new);
2426 error = user_path_dir(put_old, &old);
2430 error = security_sb_pivotroot(&old, &new);
2434 get_fs_root(current->fs, &root);
2435 error = lock_mount(&old);
2440 new_mnt = real_mount(new.mnt);
2441 root_mnt = real_mount(root.mnt);
2442 if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2443 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2444 IS_MNT_SHARED(root_mnt->mnt_parent))
2446 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2449 if (d_unlinked(new.dentry))
2451 if (d_unlinked(old.dentry))
2454 if (new.mnt == root.mnt ||
2455 old.mnt == root.mnt)
2456 goto out4; /* loop, on the same file system */
2458 if (root.mnt->mnt_root != root.dentry)
2459 goto out4; /* not a mountpoint */
2460 if (!mnt_has_parent(root_mnt))
2461 goto out4; /* not attached */
2462 if (new.mnt->mnt_root != new.dentry)
2463 goto out4; /* not a mountpoint */
2464 if (!mnt_has_parent(new_mnt))
2465 goto out4; /* not attached */
2466 /* make sure we can reach put_old from new_root */
2467 if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2469 br_write_lock(&vfsmount_lock);
2470 detach_mnt(new_mnt, &parent_path);
2471 detach_mnt(root_mnt, &root_parent);
2472 /* mount old root on put_old */
2473 attach_mnt(root_mnt, &old);
2474 /* mount new_root on / */
2475 attach_mnt(new_mnt, &root_parent);
2476 touch_mnt_namespace(current->nsproxy->mnt_ns);
2477 br_write_unlock(&vfsmount_lock);
2478 chroot_fs_refs(&root, &new);
2483 path_put(&root_parent);
2484 path_put(&parent_path);
2496 static void __init init_mount_tree(void)
2498 struct vfsmount *mnt;
2499 struct mnt_namespace *ns;
2502 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2504 panic("Can't create rootfs");
2506 ns = create_mnt_ns(mnt);
2508 panic("Can't allocate initial namespace");
2510 init_task.nsproxy->mnt_ns = ns;
2514 root.dentry = mnt->mnt_root;
2516 set_fs_pwd(current->fs, &root);
2517 set_fs_root(current->fs, &root);
2520 void __init mnt_init(void)
2525 init_rwsem(&namespace_sem);
2527 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2528 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2530 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2532 if (!mount_hashtable)
2533 panic("Failed to allocate mount hash table\n");
2535 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2537 for (u = 0; u < HASH_SIZE; u++)
2538 INIT_LIST_HEAD(&mount_hashtable[u]);
2540 br_lock_init(&vfsmount_lock);
2544 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2546 fs_kobj = kobject_create_and_add("fs", NULL);
2548 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2553 void put_mnt_ns(struct mnt_namespace *ns)
2555 LIST_HEAD(umount_list);
2557 if (!atomic_dec_and_test(&ns->count))
2559 down_write(&namespace_sem);
2560 br_write_lock(&vfsmount_lock);
2561 umount_tree(ns->root, 0, &umount_list);
2562 br_write_unlock(&vfsmount_lock);
2563 up_write(&namespace_sem);
2564 release_mounts(&umount_list);
2568 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2570 struct vfsmount *mnt;
2571 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2574 * it is a longterm mount, don't release mnt until
2575 * we unmount before file sys is unregistered
2577 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2581 EXPORT_SYMBOL_GPL(kern_mount_data);
2583 void kern_unmount(struct vfsmount *mnt)
2585 /* release long term mount so mount point can be released */
2586 if (!IS_ERR_OR_NULL(mnt)) {
2587 br_write_lock(&vfsmount_lock);
2588 real_mount(mnt)->mnt_ns = NULL;
2589 br_write_unlock(&vfsmount_lock);
2593 EXPORT_SYMBOL(kern_unmount);
2595 bool our_mnt(struct vfsmount *mnt)
2597 return check_mnt(real_mount(mnt));