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 static 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 static void free_vfsmnt(struct vfsmount *mnt)
471 kfree(mnt->mnt_devname);
474 free_percpu(mnt->mnt_pcp);
476 kmem_cache_free(mnt_cache, mnt);
480 * find the first or last mount at @dentry on vfsmount @mnt depending on
481 * @dir. If @dir is set return the first mount else return the last mount.
482 * vfsmount_lock must be held for read or write.
484 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
487 struct list_head *head = mount_hashtable + hash(mnt, dentry);
488 struct list_head *tmp = head;
489 struct vfsmount *p, *found = NULL;
492 tmp = dir ? tmp->next : tmp->prev;
496 p = list_entry(tmp, struct vfsmount, mnt_hash);
497 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
506 * lookup_mnt increments the ref count before returning
507 * the vfsmount struct.
509 struct vfsmount *lookup_mnt(struct path *path)
511 struct vfsmount *child_mnt;
513 br_read_lock(vfsmount_lock);
514 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
516 br_read_unlock(vfsmount_lock);
520 static inline int check_mnt(struct vfsmount *mnt)
522 return mnt->mnt_ns == current->nsproxy->mnt_ns;
526 * vfsmount lock must be held for write
528 static void touch_mnt_namespace(struct mnt_namespace *ns)
532 wake_up_interruptible(&ns->poll);
537 * vfsmount lock must be held for write
539 static void __touch_mnt_namespace(struct mnt_namespace *ns)
541 if (ns && ns->event != event) {
543 wake_up_interruptible(&ns->poll);
548 * Clear dentry's mounted state if it has no remaining mounts.
549 * vfsmount_lock must be held for write.
551 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
555 for (u = 0; u < HASH_SIZE; u++) {
558 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
559 if (p->mnt_mountpoint == dentry)
563 spin_lock(&dentry->d_lock);
564 dentry->d_flags &= ~DCACHE_MOUNTED;
565 spin_unlock(&dentry->d_lock);
569 * vfsmount lock must be held for write
571 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
573 old_path->dentry = mnt->mnt_mountpoint;
574 old_path->mnt = mnt->mnt_parent;
575 mnt->mnt_parent = mnt;
576 mnt->mnt_mountpoint = mnt->mnt_root;
577 list_del_init(&mnt->mnt_child);
578 list_del_init(&mnt->mnt_hash);
579 dentry_reset_mounted(old_path->mnt, old_path->dentry);
583 * vfsmount lock must be held for write
585 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
586 struct vfsmount *child_mnt)
588 child_mnt->mnt_parent = mntget(mnt);
589 child_mnt->mnt_mountpoint = dget(dentry);
590 spin_lock(&dentry->d_lock);
591 dentry->d_flags |= DCACHE_MOUNTED;
592 spin_unlock(&dentry->d_lock);
596 * vfsmount lock must be held for write
598 static void attach_mnt(struct vfsmount *mnt, struct path *path)
600 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
601 list_add_tail(&mnt->mnt_hash, mount_hashtable +
602 hash(path->mnt, path->dentry));
603 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
606 static inline void __mnt_make_longterm(struct vfsmount *mnt)
609 atomic_inc(&mnt->mnt_longterm);
613 /* needs vfsmount lock for write */
614 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
617 atomic_dec(&mnt->mnt_longterm);
622 * vfsmount lock must be held for write
624 static void commit_tree(struct vfsmount *mnt)
626 struct vfsmount *parent = mnt->mnt_parent;
629 struct mnt_namespace *n = parent->mnt_ns;
631 BUG_ON(parent == mnt);
633 list_add_tail(&head, &mnt->mnt_list);
634 list_for_each_entry(m, &head, mnt_list) {
636 __mnt_make_longterm(m);
639 list_splice(&head, n->list.prev);
641 list_add_tail(&mnt->mnt_hash, mount_hashtable +
642 hash(parent, mnt->mnt_mountpoint));
643 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
644 touch_mnt_namespace(n);
647 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
649 struct list_head *next = p->mnt_mounts.next;
650 if (next == &p->mnt_mounts) {
654 next = p->mnt_child.next;
655 if (next != &p->mnt_parent->mnt_mounts)
660 return list_entry(next, struct vfsmount, mnt_child);
663 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
665 struct list_head *prev = p->mnt_mounts.prev;
666 while (prev != &p->mnt_mounts) {
667 p = list_entry(prev, struct vfsmount, mnt_child);
668 prev = p->mnt_mounts.prev;
674 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
676 struct vfsmount *mnt;
680 return ERR_PTR(-ENODEV);
682 mnt = alloc_vfsmnt(name);
684 return ERR_PTR(-ENOMEM);
686 if (flags & MS_KERNMOUNT)
687 mnt->mnt_flags = MNT_INTERNAL;
689 root = mount_fs(type, flags, name, data);
692 return ERR_CAST(root);
695 mnt->mnt_root = root;
696 mnt->mnt_sb = root->d_sb;
697 mnt->mnt_mountpoint = mnt->mnt_root;
698 mnt->mnt_parent = mnt;
701 EXPORT_SYMBOL_GPL(vfs_kern_mount);
703 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
706 struct super_block *sb = old->mnt_sb;
707 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
710 if (flag & (CL_SLAVE | CL_PRIVATE))
711 mnt->mnt_group_id = 0; /* not a peer of original */
713 mnt->mnt_group_id = old->mnt_group_id;
715 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
716 int err = mnt_alloc_group_id(mnt);
721 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
722 atomic_inc(&sb->s_active);
724 mnt->mnt_root = dget(root);
725 mnt->mnt_mountpoint = mnt->mnt_root;
726 mnt->mnt_parent = mnt;
728 if (flag & CL_SLAVE) {
729 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
730 mnt->mnt_master = old;
731 CLEAR_MNT_SHARED(mnt);
732 } else if (!(flag & CL_PRIVATE)) {
733 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
734 list_add(&mnt->mnt_share, &old->mnt_share);
735 if (IS_MNT_SLAVE(old))
736 list_add(&mnt->mnt_slave, &old->mnt_slave);
737 mnt->mnt_master = old->mnt_master;
739 if (flag & CL_MAKE_SHARED)
742 /* stick the duplicate mount on the same expiry list
743 * as the original if that was on one */
744 if (flag & CL_EXPIRE) {
745 if (!list_empty(&old->mnt_expire))
746 list_add(&mnt->mnt_expire, &old->mnt_expire);
756 static inline void mntfree(struct vfsmount *mnt)
758 struct super_block *sb = mnt->mnt_sb;
761 * This probably indicates that somebody messed
762 * up a mnt_want/drop_write() pair. If this
763 * happens, the filesystem was probably unable
764 * to make r/w->r/o transitions.
767 * The locking used to deal with mnt_count decrement provides barriers,
768 * so mnt_get_writers() below is safe.
770 WARN_ON(mnt_get_writers(mnt));
771 fsnotify_vfsmount_delete(mnt);
774 deactivate_super(sb);
777 static void mntput_no_expire(struct vfsmount *mnt)
781 br_read_lock(vfsmount_lock);
782 if (likely(atomic_read(&mnt->mnt_longterm))) {
784 br_read_unlock(vfsmount_lock);
787 br_read_unlock(vfsmount_lock);
789 br_write_lock(vfsmount_lock);
791 if (mnt_get_count(mnt)) {
792 br_write_unlock(vfsmount_lock);
797 if (likely(mnt_get_count(mnt)))
799 br_write_lock(vfsmount_lock);
801 if (unlikely(mnt->mnt_pinned)) {
802 mnt_add_count(mnt, mnt->mnt_pinned + 1);
804 br_write_unlock(vfsmount_lock);
805 acct_auto_close_mnt(mnt);
808 br_write_unlock(vfsmount_lock);
812 void mntput(struct vfsmount *mnt)
815 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
816 if (unlikely(mnt->mnt_expiry_mark))
817 mnt->mnt_expiry_mark = 0;
818 mntput_no_expire(mnt);
821 EXPORT_SYMBOL(mntput);
823 struct vfsmount *mntget(struct vfsmount *mnt)
829 EXPORT_SYMBOL(mntget);
831 void mnt_pin(struct vfsmount *mnt)
833 br_write_lock(vfsmount_lock);
835 br_write_unlock(vfsmount_lock);
837 EXPORT_SYMBOL(mnt_pin);
839 void mnt_unpin(struct vfsmount *mnt)
841 br_write_lock(vfsmount_lock);
842 if (mnt->mnt_pinned) {
846 br_write_unlock(vfsmount_lock);
848 EXPORT_SYMBOL(mnt_unpin);
850 static inline void mangle(struct seq_file *m, const char *s)
852 seq_escape(m, s, " \t\n\\");
856 * Simple .show_options callback for filesystems which don't want to
857 * implement more complex mount option showing.
859 * See also save_mount_options().
861 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
866 options = rcu_dereference(mnt->mnt_sb->s_options);
868 if (options != NULL && options[0]) {
876 EXPORT_SYMBOL(generic_show_options);
879 * If filesystem uses generic_show_options(), this function should be
880 * called from the fill_super() callback.
882 * The .remount_fs callback usually needs to be handled in a special
883 * way, to make sure, that previous options are not overwritten if the
886 * Also note, that if the filesystem's .remount_fs function doesn't
887 * reset all options to their default value, but changes only newly
888 * given options, then the displayed options will not reflect reality
891 void save_mount_options(struct super_block *sb, char *options)
893 BUG_ON(sb->s_options);
894 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
896 EXPORT_SYMBOL(save_mount_options);
898 void replace_mount_options(struct super_block *sb, char *options)
900 char *old = sb->s_options;
901 rcu_assign_pointer(sb->s_options, options);
907 EXPORT_SYMBOL(replace_mount_options);
909 #ifdef CONFIG_PROC_FS
911 static void *m_start(struct seq_file *m, loff_t *pos)
913 struct proc_mounts *p = m->private;
915 down_read(&namespace_sem);
916 return seq_list_start(&p->ns->list, *pos);
919 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
921 struct proc_mounts *p = m->private;
923 return seq_list_next(v, &p->ns->list, pos);
926 static void m_stop(struct seq_file *m, void *v)
928 up_read(&namespace_sem);
931 int mnt_had_events(struct proc_mounts *p)
933 struct mnt_namespace *ns = p->ns;
936 br_read_lock(vfsmount_lock);
937 if (p->event != ns->event) {
938 p->event = ns->event;
941 br_read_unlock(vfsmount_lock);
946 struct proc_fs_info {
951 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
953 static const struct proc_fs_info fs_info[] = {
954 { MS_SYNCHRONOUS, ",sync" },
955 { MS_DIRSYNC, ",dirsync" },
956 { MS_MANDLOCK, ",mand" },
959 const struct proc_fs_info *fs_infop;
961 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
962 if (sb->s_flags & fs_infop->flag)
963 seq_puts(m, fs_infop->str);
966 return security_sb_show_options(m, sb);
969 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
971 static const struct proc_fs_info mnt_info[] = {
972 { MNT_NOSUID, ",nosuid" },
973 { MNT_NODEV, ",nodev" },
974 { MNT_NOEXEC, ",noexec" },
975 { MNT_NOATIME, ",noatime" },
976 { MNT_NODIRATIME, ",nodiratime" },
977 { MNT_RELATIME, ",relatime" },
980 const struct proc_fs_info *fs_infop;
982 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
983 if (mnt->mnt_flags & fs_infop->flag)
984 seq_puts(m, fs_infop->str);
988 static void show_type(struct seq_file *m, struct super_block *sb)
990 mangle(m, sb->s_type->name);
991 if (sb->s_subtype && sb->s_subtype[0]) {
993 mangle(m, sb->s_subtype);
997 static int show_vfsmnt(struct seq_file *m, void *v)
999 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1001 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1003 if (mnt->mnt_sb->s_op->show_devname) {
1004 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1008 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1011 seq_path(m, &mnt_path, " \t\n\\");
1013 show_type(m, mnt->mnt_sb);
1014 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1015 err = show_sb_opts(m, mnt->mnt_sb);
1018 show_mnt_opts(m, mnt);
1019 if (mnt->mnt_sb->s_op->show_options)
1020 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1021 seq_puts(m, " 0 0\n");
1026 const struct seq_operations mounts_op = {
1033 static int uuid_is_nil(u8 *uuid)
1036 u8 *cp = (u8 *)uuid;
1038 for (i = 0; i < 16; i++) {
1045 static int show_mountinfo(struct seq_file *m, void *v)
1047 struct proc_mounts *p = m->private;
1048 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1049 struct super_block *sb = mnt->mnt_sb;
1050 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1051 struct path root = p->root;
1054 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1055 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1056 if (sb->s_op->show_path)
1057 err = sb->s_op->show_path(m, mnt);
1059 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1063 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1064 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1066 * Mountpoint is outside root, discard that one. Ugly,
1067 * but less so than trying to do that in iterator in a
1068 * race-free way (due to renames).
1072 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1073 show_mnt_opts(m, mnt);
1075 /* Tagged fields ("foo:X" or "bar") */
1076 if (IS_MNT_SHARED(mnt))
1077 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1078 if (IS_MNT_SLAVE(mnt)) {
1079 int master = mnt->mnt_master->mnt_group_id;
1080 int dom = get_dominating_id(mnt, &p->root);
1081 seq_printf(m, " master:%i", master);
1082 if (dom && dom != master)
1083 seq_printf(m, " propagate_from:%i", dom);
1085 if (IS_MNT_UNBINDABLE(mnt))
1086 seq_puts(m, " unbindable");
1088 if (!uuid_is_nil(mnt->mnt_sb->s_uuid))
1089 /* print the uuid */
1090 seq_printf(m, " uuid:%pU", mnt->mnt_sb->s_uuid);
1092 /* Filesystem specific data */
1096 if (sb->s_op->show_devname)
1097 err = sb->s_op->show_devname(m, mnt);
1099 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1102 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1103 err = show_sb_opts(m, sb);
1106 if (sb->s_op->show_options)
1107 err = sb->s_op->show_options(m, mnt);
1113 const struct seq_operations mountinfo_op = {
1117 .show = show_mountinfo,
1120 static int show_vfsstat(struct seq_file *m, void *v)
1122 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1123 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1127 if (mnt->mnt_sb->s_op->show_devname) {
1128 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1130 if (mnt->mnt_devname) {
1131 seq_puts(m, "device ");
1132 mangle(m, mnt->mnt_devname);
1134 seq_puts(m, "no device");
1138 seq_puts(m, " mounted on ");
1139 seq_path(m, &mnt_path, " \t\n\\");
1142 /* file system type */
1143 seq_puts(m, "with fstype ");
1144 show_type(m, mnt->mnt_sb);
1146 /* optional statistics */
1147 if (mnt->mnt_sb->s_op->show_stats) {
1150 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1157 const struct seq_operations mountstats_op = {
1161 .show = show_vfsstat,
1163 #endif /* CONFIG_PROC_FS */
1166 * may_umount_tree - check if a mount tree is busy
1167 * @mnt: root of mount tree
1169 * This is called to check if a tree of mounts has any
1170 * open files, pwds, chroots or sub mounts that are
1173 int may_umount_tree(struct vfsmount *mnt)
1175 int actual_refs = 0;
1176 int minimum_refs = 0;
1179 /* write lock needed for mnt_get_count */
1180 br_write_lock(vfsmount_lock);
1181 for (p = mnt; p; p = next_mnt(p, mnt)) {
1182 actual_refs += mnt_get_count(p);
1185 br_write_unlock(vfsmount_lock);
1187 if (actual_refs > minimum_refs)
1193 EXPORT_SYMBOL(may_umount_tree);
1196 * may_umount - check if a mount point is busy
1197 * @mnt: root of mount
1199 * This is called to check if a mount point has any
1200 * open files, pwds, chroots or sub mounts. If the
1201 * mount has sub mounts this will return busy
1202 * regardless of whether the sub mounts are busy.
1204 * Doesn't take quota and stuff into account. IOW, in some cases it will
1205 * give false negatives. The main reason why it's here is that we need
1206 * a non-destructive way to look for easily umountable filesystems.
1208 int may_umount(struct vfsmount *mnt)
1211 down_read(&namespace_sem);
1212 br_write_lock(vfsmount_lock);
1213 if (propagate_mount_busy(mnt, 2))
1215 br_write_unlock(vfsmount_lock);
1216 up_read(&namespace_sem);
1220 EXPORT_SYMBOL(may_umount);
1222 void release_mounts(struct list_head *head)
1224 struct vfsmount *mnt;
1225 while (!list_empty(head)) {
1226 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1227 list_del_init(&mnt->mnt_hash);
1228 if (mnt->mnt_parent != mnt) {
1229 struct dentry *dentry;
1232 br_write_lock(vfsmount_lock);
1233 dentry = mnt->mnt_mountpoint;
1234 m = mnt->mnt_parent;
1235 mnt->mnt_mountpoint = mnt->mnt_root;
1236 mnt->mnt_parent = mnt;
1238 br_write_unlock(vfsmount_lock);
1247 * vfsmount lock must be held for write
1248 * namespace_sem must be held for write
1250 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1252 LIST_HEAD(tmp_list);
1255 for (p = mnt; p; p = next_mnt(p, mnt))
1256 list_move(&p->mnt_hash, &tmp_list);
1259 propagate_umount(&tmp_list);
1261 list_for_each_entry(p, &tmp_list, mnt_hash) {
1262 list_del_init(&p->mnt_expire);
1263 list_del_init(&p->mnt_list);
1264 __touch_mnt_namespace(p->mnt_ns);
1266 __mnt_make_shortterm(p);
1267 list_del_init(&p->mnt_child);
1268 if (p->mnt_parent != p) {
1269 p->mnt_parent->mnt_ghosts++;
1270 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1272 change_mnt_propagation(p, MS_PRIVATE);
1274 list_splice(&tmp_list, kill);
1277 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1279 static int do_umount(struct vfsmount *mnt, int flags)
1281 struct super_block *sb = mnt->mnt_sb;
1283 LIST_HEAD(umount_list);
1285 retval = security_sb_umount(mnt, flags);
1290 * Allow userspace to request a mountpoint be expired rather than
1291 * unmounting unconditionally. Unmount only happens if:
1292 * (1) the mark is already set (the mark is cleared by mntput())
1293 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1295 if (flags & MNT_EXPIRE) {
1296 if (mnt == current->fs->root.mnt ||
1297 flags & (MNT_FORCE | MNT_DETACH))
1301 * probably don't strictly need the lock here if we examined
1302 * all race cases, but it's a slowpath.
1304 br_write_lock(vfsmount_lock);
1305 if (mnt_get_count(mnt) != 2) {
1306 br_write_unlock(vfsmount_lock);
1309 br_write_unlock(vfsmount_lock);
1311 if (!xchg(&mnt->mnt_expiry_mark, 1))
1316 * If we may have to abort operations to get out of this
1317 * mount, and they will themselves hold resources we must
1318 * allow the fs to do things. In the Unix tradition of
1319 * 'Gee thats tricky lets do it in userspace' the umount_begin
1320 * might fail to complete on the first run through as other tasks
1321 * must return, and the like. Thats for the mount program to worry
1322 * about for the moment.
1325 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1326 sb->s_op->umount_begin(sb);
1330 * No sense to grab the lock for this test, but test itself looks
1331 * somewhat bogus. Suggestions for better replacement?
1332 * Ho-hum... In principle, we might treat that as umount + switch
1333 * to rootfs. GC would eventually take care of the old vfsmount.
1334 * Actually it makes sense, especially if rootfs would contain a
1335 * /reboot - static binary that would close all descriptors and
1336 * call reboot(9). Then init(8) could umount root and exec /reboot.
1338 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1340 * Special case for "unmounting" root ...
1341 * we just try to remount it readonly.
1343 down_write(&sb->s_umount);
1344 if (!(sb->s_flags & MS_RDONLY))
1345 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1346 up_write(&sb->s_umount);
1350 down_write(&namespace_sem);
1351 br_write_lock(vfsmount_lock);
1354 if (!(flags & MNT_DETACH))
1355 shrink_submounts(mnt, &umount_list);
1358 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1359 if (!list_empty(&mnt->mnt_list))
1360 umount_tree(mnt, 1, &umount_list);
1363 br_write_unlock(vfsmount_lock);
1364 up_write(&namespace_sem);
1365 release_mounts(&umount_list);
1370 * Now umount can handle mount points as well as block devices.
1371 * This is important for filesystems which use unnamed block devices.
1373 * We now support a flag for forced unmount like the other 'big iron'
1374 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1377 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1381 int lookup_flags = 0;
1383 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1386 if (!(flags & UMOUNT_NOFOLLOW))
1387 lookup_flags |= LOOKUP_FOLLOW;
1389 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1393 if (path.dentry != path.mnt->mnt_root)
1395 if (!check_mnt(path.mnt))
1399 if (!capable(CAP_SYS_ADMIN))
1402 retval = do_umount(path.mnt, flags);
1404 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1406 mntput_no_expire(path.mnt);
1411 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1414 * The 2.0 compatible umount. No flags.
1416 SYSCALL_DEFINE1(oldumount, char __user *, name)
1418 return sys_umount(name, 0);
1423 static int mount_is_safe(struct path *path)
1425 if (capable(CAP_SYS_ADMIN))
1429 if (S_ISLNK(path->dentry->d_inode->i_mode))
1431 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1432 if (current_uid() != path->dentry->d_inode->i_uid)
1435 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1441 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1444 struct vfsmount *res, *p, *q, *r, *s;
1447 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1450 res = q = clone_mnt(mnt, dentry, flag);
1453 q->mnt_mountpoint = mnt->mnt_mountpoint;
1456 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1457 if (!is_subdir(r->mnt_mountpoint, dentry))
1460 for (s = r; s; s = next_mnt(s, r)) {
1461 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1462 s = skip_mnt_tree(s);
1465 while (p != s->mnt_parent) {
1471 path.dentry = p->mnt_mountpoint;
1472 q = clone_mnt(p, p->mnt_root, flag);
1475 br_write_lock(vfsmount_lock);
1476 list_add_tail(&q->mnt_list, &res->mnt_list);
1477 attach_mnt(q, &path);
1478 br_write_unlock(vfsmount_lock);
1484 LIST_HEAD(umount_list);
1485 br_write_lock(vfsmount_lock);
1486 umount_tree(res, 0, &umount_list);
1487 br_write_unlock(vfsmount_lock);
1488 release_mounts(&umount_list);
1493 struct vfsmount *collect_mounts(struct path *path)
1495 struct vfsmount *tree;
1496 down_write(&namespace_sem);
1497 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1498 up_write(&namespace_sem);
1502 void drop_collected_mounts(struct vfsmount *mnt)
1504 LIST_HEAD(umount_list);
1505 down_write(&namespace_sem);
1506 br_write_lock(vfsmount_lock);
1507 umount_tree(mnt, 0, &umount_list);
1508 br_write_unlock(vfsmount_lock);
1509 up_write(&namespace_sem);
1510 release_mounts(&umount_list);
1513 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1514 struct vfsmount *root)
1516 struct vfsmount *mnt;
1517 int res = f(root, arg);
1520 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1528 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1532 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1533 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1534 mnt_release_group_id(p);
1538 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1542 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1543 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1544 int err = mnt_alloc_group_id(p);
1546 cleanup_group_ids(mnt, p);
1556 * @source_mnt : mount tree to be attached
1557 * @nd : place the mount tree @source_mnt is attached
1558 * @parent_nd : if non-null, detach the source_mnt from its parent and
1559 * store the parent mount and mountpoint dentry.
1560 * (done when source_mnt is moved)
1562 * NOTE: in the table below explains the semantics when a source mount
1563 * of a given type is attached to a destination mount of a given type.
1564 * ---------------------------------------------------------------------------
1565 * | BIND MOUNT OPERATION |
1566 * |**************************************************************************
1567 * | source-->| shared | private | slave | unbindable |
1571 * |**************************************************************************
1572 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1574 * |non-shared| shared (+) | private | slave (*) | invalid |
1575 * ***************************************************************************
1576 * A bind operation clones the source mount and mounts the clone on the
1577 * destination mount.
1579 * (++) the cloned mount is propagated to all the mounts in the propagation
1580 * tree of the destination mount and the cloned mount is added to
1581 * the peer group of the source mount.
1582 * (+) the cloned mount is created under the destination mount and is marked
1583 * as shared. The cloned mount is added to the peer group of the source
1585 * (+++) the mount is propagated to all the mounts in the propagation tree
1586 * of the destination mount and the cloned mount is made slave
1587 * of the same master as that of the source mount. The cloned mount
1588 * is marked as 'shared and slave'.
1589 * (*) the cloned mount is made a slave of the same master as that of the
1592 * ---------------------------------------------------------------------------
1593 * | MOVE MOUNT OPERATION |
1594 * |**************************************************************************
1595 * | source-->| shared | private | slave | unbindable |
1599 * |**************************************************************************
1600 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1602 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1603 * ***************************************************************************
1605 * (+) the mount is moved to the destination. And is then propagated to
1606 * all the mounts in the propagation tree of the destination mount.
1607 * (+*) the mount is moved to the destination.
1608 * (+++) the mount is moved to the destination and is then propagated to
1609 * all the mounts belonging to the destination mount's propagation tree.
1610 * the mount is marked as 'shared and slave'.
1611 * (*) the mount continues to be a slave at the new location.
1613 * if the source mount is a tree, the operations explained above is
1614 * applied to each mount in the tree.
1615 * Must be called without spinlocks held, since this function can sleep
1618 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1619 struct path *path, struct path *parent_path)
1621 LIST_HEAD(tree_list);
1622 struct vfsmount *dest_mnt = path->mnt;
1623 struct dentry *dest_dentry = path->dentry;
1624 struct vfsmount *child, *p;
1627 if (IS_MNT_SHARED(dest_mnt)) {
1628 err = invent_group_ids(source_mnt, true);
1632 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1634 goto out_cleanup_ids;
1636 br_write_lock(vfsmount_lock);
1638 if (IS_MNT_SHARED(dest_mnt)) {
1639 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1643 detach_mnt(source_mnt, parent_path);
1644 attach_mnt(source_mnt, path);
1645 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1647 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1648 commit_tree(source_mnt);
1651 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1652 list_del_init(&child->mnt_hash);
1655 br_write_unlock(vfsmount_lock);
1660 if (IS_MNT_SHARED(dest_mnt))
1661 cleanup_group_ids(source_mnt, NULL);
1666 static int graft_tree(struct vfsmount *mnt, struct path *path)
1669 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1672 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1673 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1677 mutex_lock(&path->dentry->d_inode->i_mutex);
1678 if (cant_mount(path->dentry))
1681 if (!d_unlinked(path->dentry))
1682 err = attach_recursive_mnt(mnt, path, NULL);
1684 mutex_unlock(&path->dentry->d_inode->i_mutex);
1689 * Sanity check the flags to change_mnt_propagation.
1692 static int flags_to_propagation_type(int flags)
1694 int type = flags & ~MS_REC;
1696 /* Fail if any non-propagation flags are set */
1697 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1699 /* Only one propagation flag should be set */
1700 if (!is_power_of_2(type))
1706 * recursively change the type of the mountpoint.
1708 static int do_change_type(struct path *path, int flag)
1710 struct vfsmount *m, *mnt = path->mnt;
1711 int recurse = flag & MS_REC;
1715 if (!capable(CAP_SYS_ADMIN))
1718 if (path->dentry != path->mnt->mnt_root)
1721 type = flags_to_propagation_type(flag);
1725 down_write(&namespace_sem);
1726 if (type == MS_SHARED) {
1727 err = invent_group_ids(mnt, recurse);
1732 br_write_lock(vfsmount_lock);
1733 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1734 change_mnt_propagation(m, type);
1735 br_write_unlock(vfsmount_lock);
1738 up_write(&namespace_sem);
1743 * do loopback mount.
1745 static int do_loopback(struct path *path, char *old_name,
1748 struct path old_path;
1749 struct vfsmount *mnt = NULL;
1750 int err = mount_is_safe(path);
1753 if (!old_name || !*old_name)
1755 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1759 down_write(&namespace_sem);
1761 if (IS_MNT_UNBINDABLE(old_path.mnt))
1764 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1769 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1771 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1776 err = graft_tree(mnt, path);
1778 LIST_HEAD(umount_list);
1780 br_write_lock(vfsmount_lock);
1781 umount_tree(mnt, 0, &umount_list);
1782 br_write_unlock(vfsmount_lock);
1783 release_mounts(&umount_list);
1787 up_write(&namespace_sem);
1788 path_put(&old_path);
1792 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1795 int readonly_request = 0;
1797 if (ms_flags & MS_RDONLY)
1798 readonly_request = 1;
1799 if (readonly_request == __mnt_is_readonly(mnt))
1802 if (readonly_request)
1803 error = mnt_make_readonly(mnt);
1805 __mnt_unmake_readonly(mnt);
1810 * change filesystem flags. dir should be a physical root of filesystem.
1811 * If you've mounted a non-root directory somewhere and want to do remount
1812 * on it - tough luck.
1814 static int do_remount(struct path *path, int flags, int mnt_flags,
1818 struct super_block *sb = path->mnt->mnt_sb;
1820 if (!capable(CAP_SYS_ADMIN))
1823 if (!check_mnt(path->mnt))
1826 if (path->dentry != path->mnt->mnt_root)
1829 err = security_sb_remount(sb, data);
1833 down_write(&sb->s_umount);
1834 if (flags & MS_BIND)
1835 err = change_mount_flags(path->mnt, flags);
1837 err = do_remount_sb(sb, flags, data, 0);
1839 br_write_lock(vfsmount_lock);
1840 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1841 path->mnt->mnt_flags = mnt_flags;
1842 br_write_unlock(vfsmount_lock);
1844 up_write(&sb->s_umount);
1846 br_write_lock(vfsmount_lock);
1847 touch_mnt_namespace(path->mnt->mnt_ns);
1848 br_write_unlock(vfsmount_lock);
1853 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1856 for (p = mnt; p; p = next_mnt(p, mnt)) {
1857 if (IS_MNT_UNBINDABLE(p))
1863 static int do_move_mount(struct path *path, char *old_name)
1865 struct path old_path, parent_path;
1868 if (!capable(CAP_SYS_ADMIN))
1870 if (!old_name || !*old_name)
1872 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1876 down_write(&namespace_sem);
1877 err = follow_down(path, true);
1882 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1886 mutex_lock(&path->dentry->d_inode->i_mutex);
1887 if (cant_mount(path->dentry))
1890 if (d_unlinked(path->dentry))
1894 if (old_path.dentry != old_path.mnt->mnt_root)
1897 if (old_path.mnt == old_path.mnt->mnt_parent)
1900 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1901 S_ISDIR(old_path.dentry->d_inode->i_mode))
1904 * Don't move a mount residing in a shared parent.
1906 if (old_path.mnt->mnt_parent &&
1907 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1910 * Don't move a mount tree containing unbindable mounts to a destination
1911 * mount which is shared.
1913 if (IS_MNT_SHARED(path->mnt) &&
1914 tree_contains_unbindable(old_path.mnt))
1917 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1918 if (p == old_path.mnt)
1921 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1925 /* if the mount is moved, it should no longer be expire
1927 list_del_init(&old_path.mnt->mnt_expire);
1929 mutex_unlock(&path->dentry->d_inode->i_mutex);
1931 up_write(&namespace_sem);
1933 path_put(&parent_path);
1934 path_put(&old_path);
1938 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1941 const char *subtype = strchr(fstype, '.');
1950 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1952 if (!mnt->mnt_sb->s_subtype)
1958 return ERR_PTR(err);
1962 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1964 struct file_system_type *type = get_fs_type(fstype);
1965 struct vfsmount *mnt;
1967 return ERR_PTR(-ENODEV);
1968 mnt = vfs_kern_mount(type, flags, name, data);
1969 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1970 !mnt->mnt_sb->s_subtype)
1971 mnt = fs_set_subtype(mnt, fstype);
1972 put_filesystem(type);
1975 EXPORT_SYMBOL_GPL(do_kern_mount);
1978 * add a mount into a namespace's mount tree
1980 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1984 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1986 down_write(&namespace_sem);
1987 /* Something was mounted here while we slept */
1988 err = follow_down(path, true);
1993 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1996 /* Refuse the same filesystem on the same mount point */
1998 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1999 path->mnt->mnt_root == path->dentry)
2003 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2006 newmnt->mnt_flags = mnt_flags;
2007 err = graft_tree(newmnt, path);
2010 up_write(&namespace_sem);
2015 * create a new mount for userspace and request it to be added into the
2018 static int do_new_mount(struct path *path, char *type, int flags,
2019 int mnt_flags, char *name, void *data)
2021 struct vfsmount *mnt;
2027 /* we need capabilities... */
2028 if (!capable(CAP_SYS_ADMIN))
2031 mnt = do_kern_mount(type, flags, name, data);
2033 return PTR_ERR(mnt);
2035 err = do_add_mount(mnt, path, mnt_flags);
2041 int finish_automount(struct vfsmount *m, struct path *path)
2044 /* The new mount record should have at least 2 refs to prevent it being
2045 * expired before we get a chance to add it
2047 BUG_ON(mnt_get_count(m) < 2);
2049 if (m->mnt_sb == path->mnt->mnt_sb &&
2050 m->mnt_root == path->dentry) {
2055 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2059 /* remove m from any expiration list it may be on */
2060 if (!list_empty(&m->mnt_expire)) {
2061 down_write(&namespace_sem);
2062 br_write_lock(vfsmount_lock);
2063 list_del_init(&m->mnt_expire);
2064 br_write_unlock(vfsmount_lock);
2065 up_write(&namespace_sem);
2073 * mnt_set_expiry - Put a mount on an expiration list
2074 * @mnt: The mount to list.
2075 * @expiry_list: The list to add the mount to.
2077 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2079 down_write(&namespace_sem);
2080 br_write_lock(vfsmount_lock);
2082 list_add_tail(&mnt->mnt_expire, expiry_list);
2084 br_write_unlock(vfsmount_lock);
2085 up_write(&namespace_sem);
2087 EXPORT_SYMBOL(mnt_set_expiry);
2090 * process a list of expirable mountpoints with the intent of discarding any
2091 * mountpoints that aren't in use and haven't been touched since last we came
2094 void mark_mounts_for_expiry(struct list_head *mounts)
2096 struct vfsmount *mnt, *next;
2097 LIST_HEAD(graveyard);
2100 if (list_empty(mounts))
2103 down_write(&namespace_sem);
2104 br_write_lock(vfsmount_lock);
2106 /* extract from the expiration list every vfsmount that matches the
2107 * following criteria:
2108 * - only referenced by its parent vfsmount
2109 * - still marked for expiry (marked on the last call here; marks are
2110 * cleared by mntput())
2112 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2113 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2114 propagate_mount_busy(mnt, 1))
2116 list_move(&mnt->mnt_expire, &graveyard);
2118 while (!list_empty(&graveyard)) {
2119 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2120 touch_mnt_namespace(mnt->mnt_ns);
2121 umount_tree(mnt, 1, &umounts);
2123 br_write_unlock(vfsmount_lock);
2124 up_write(&namespace_sem);
2126 release_mounts(&umounts);
2129 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2132 * Ripoff of 'select_parent()'
2134 * search the list of submounts for a given mountpoint, and move any
2135 * shrinkable submounts to the 'graveyard' list.
2137 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2139 struct vfsmount *this_parent = parent;
2140 struct list_head *next;
2144 next = this_parent->mnt_mounts.next;
2146 while (next != &this_parent->mnt_mounts) {
2147 struct list_head *tmp = next;
2148 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2151 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2154 * Descend a level if the d_mounts list is non-empty.
2156 if (!list_empty(&mnt->mnt_mounts)) {
2161 if (!propagate_mount_busy(mnt, 1)) {
2162 list_move_tail(&mnt->mnt_expire, graveyard);
2167 * All done at this level ... ascend and resume the search
2169 if (this_parent != parent) {
2170 next = this_parent->mnt_child.next;
2171 this_parent = this_parent->mnt_parent;
2178 * process a list of expirable mountpoints with the intent of discarding any
2179 * submounts of a specific parent mountpoint
2181 * vfsmount_lock must be held for write
2183 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2185 LIST_HEAD(graveyard);
2188 /* extract submounts of 'mountpoint' from the expiration list */
2189 while (select_submounts(mnt, &graveyard)) {
2190 while (!list_empty(&graveyard)) {
2191 m = list_first_entry(&graveyard, struct vfsmount,
2193 touch_mnt_namespace(m->mnt_ns);
2194 umount_tree(m, 1, umounts);
2200 * Some copy_from_user() implementations do not return the exact number of
2201 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2202 * Note that this function differs from copy_from_user() in that it will oops
2203 * on bad values of `to', rather than returning a short copy.
2205 static long exact_copy_from_user(void *to, const void __user * from,
2209 const char __user *f = from;
2212 if (!access_ok(VERIFY_READ, from, n))
2216 if (__get_user(c, f)) {
2227 int copy_mount_options(const void __user * data, unsigned long *where)
2237 if (!(page = __get_free_page(GFP_KERNEL)))
2240 /* We only care that *some* data at the address the user
2241 * gave us is valid. Just in case, we'll zero
2242 * the remainder of the page.
2244 /* copy_from_user cannot cross TASK_SIZE ! */
2245 size = TASK_SIZE - (unsigned long)data;
2246 if (size > PAGE_SIZE)
2249 i = size - exact_copy_from_user((void *)page, data, size);
2255 memset((char *)page + i, 0, PAGE_SIZE - i);
2260 int copy_mount_string(const void __user *data, char **where)
2269 tmp = strndup_user(data, PAGE_SIZE);
2271 return PTR_ERR(tmp);
2278 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2279 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2281 * data is a (void *) that can point to any structure up to
2282 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2283 * information (or be NULL).
2285 * Pre-0.97 versions of mount() didn't have a flags word.
2286 * When the flags word was introduced its top half was required
2287 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2288 * Therefore, if this magic number is present, it carries no information
2289 * and must be discarded.
2291 long do_mount(char *dev_name, char *dir_name, char *type_page,
2292 unsigned long flags, void *data_page)
2299 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2300 flags &= ~MS_MGC_MSK;
2302 /* Basic sanity checks */
2304 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2308 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2310 /* ... and get the mountpoint */
2311 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2315 retval = security_sb_mount(dev_name, &path,
2316 type_page, flags, data_page);
2320 /* Default to relatime unless overriden */
2321 if (!(flags & MS_NOATIME))
2322 mnt_flags |= MNT_RELATIME;
2324 /* Separate the per-mountpoint flags */
2325 if (flags & MS_NOSUID)
2326 mnt_flags |= MNT_NOSUID;
2327 if (flags & MS_NODEV)
2328 mnt_flags |= MNT_NODEV;
2329 if (flags & MS_NOEXEC)
2330 mnt_flags |= MNT_NOEXEC;
2331 if (flags & MS_NOATIME)
2332 mnt_flags |= MNT_NOATIME;
2333 if (flags & MS_NODIRATIME)
2334 mnt_flags |= MNT_NODIRATIME;
2335 if (flags & MS_STRICTATIME)
2336 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2337 if (flags & MS_RDONLY)
2338 mnt_flags |= MNT_READONLY;
2340 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2341 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2344 if (flags & MS_REMOUNT)
2345 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2347 else if (flags & MS_BIND)
2348 retval = do_loopback(&path, dev_name, flags & MS_REC);
2349 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2350 retval = do_change_type(&path, flags);
2351 else if (flags & MS_MOVE)
2352 retval = do_move_mount(&path, dev_name);
2354 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2355 dev_name, data_page);
2361 static struct mnt_namespace *alloc_mnt_ns(void)
2363 struct mnt_namespace *new_ns;
2365 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2367 return ERR_PTR(-ENOMEM);
2368 atomic_set(&new_ns->count, 1);
2369 new_ns->root = NULL;
2370 INIT_LIST_HEAD(&new_ns->list);
2371 init_waitqueue_head(&new_ns->poll);
2376 void mnt_make_longterm(struct vfsmount *mnt)
2378 __mnt_make_longterm(mnt);
2381 void mnt_make_shortterm(struct vfsmount *mnt)
2384 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2386 br_write_lock(vfsmount_lock);
2387 atomic_dec(&mnt->mnt_longterm);
2388 br_write_unlock(vfsmount_lock);
2393 * Allocate a new namespace structure and populate it with contents
2394 * copied from the namespace of the passed in task structure.
2396 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2397 struct fs_struct *fs)
2399 struct mnt_namespace *new_ns;
2400 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2401 struct vfsmount *p, *q;
2403 new_ns = alloc_mnt_ns();
2407 down_write(&namespace_sem);
2408 /* First pass: copy the tree topology */
2409 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2410 CL_COPY_ALL | CL_EXPIRE);
2411 if (!new_ns->root) {
2412 up_write(&namespace_sem);
2414 return ERR_PTR(-ENOMEM);
2416 br_write_lock(vfsmount_lock);
2417 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2418 br_write_unlock(vfsmount_lock);
2421 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2422 * as belonging to new namespace. We have already acquired a private
2423 * fs_struct, so tsk->fs->lock is not needed.
2429 __mnt_make_longterm(q);
2431 if (p == fs->root.mnt) {
2432 fs->root.mnt = mntget(q);
2433 __mnt_make_longterm(q);
2434 mnt_make_shortterm(p);
2437 if (p == fs->pwd.mnt) {
2438 fs->pwd.mnt = mntget(q);
2439 __mnt_make_longterm(q);
2440 mnt_make_shortterm(p);
2444 p = next_mnt(p, mnt_ns->root);
2445 q = next_mnt(q, new_ns->root);
2447 up_write(&namespace_sem);
2457 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2458 struct fs_struct *new_fs)
2460 struct mnt_namespace *new_ns;
2465 if (!(flags & CLONE_NEWNS))
2468 new_ns = dup_mnt_ns(ns, new_fs);
2475 * create_mnt_ns - creates a private namespace and adds a root filesystem
2476 * @mnt: pointer to the new root filesystem mountpoint
2478 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2480 struct mnt_namespace *new_ns;
2482 new_ns = alloc_mnt_ns();
2483 if (!IS_ERR(new_ns)) {
2484 mnt->mnt_ns = new_ns;
2485 __mnt_make_longterm(mnt);
2487 list_add(&new_ns->list, &new_ns->root->mnt_list);
2491 EXPORT_SYMBOL(create_mnt_ns);
2493 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2494 char __user *, type, unsigned long, flags, void __user *, data)
2500 unsigned long data_page;
2502 ret = copy_mount_string(type, &kernel_type);
2506 kernel_dir = getname(dir_name);
2507 if (IS_ERR(kernel_dir)) {
2508 ret = PTR_ERR(kernel_dir);
2512 ret = copy_mount_string(dev_name, &kernel_dev);
2516 ret = copy_mount_options(data, &data_page);
2520 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2521 (void *) data_page);
2523 free_page(data_page);
2527 putname(kernel_dir);
2535 * pivot_root Semantics:
2536 * Moves the root file system of the current process to the directory put_old,
2537 * makes new_root as the new root file system of the current process, and sets
2538 * root/cwd of all processes which had them on the current root to new_root.
2541 * The new_root and put_old must be directories, and must not be on the
2542 * same file system as the current process root. The put_old must be
2543 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2544 * pointed to by put_old must yield the same directory as new_root. No other
2545 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2547 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2548 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2549 * in this situation.
2552 * - we don't move root/cwd if they are not at the root (reason: if something
2553 * cared enough to change them, it's probably wrong to force them elsewhere)
2554 * - it's okay to pick a root that isn't the root of a file system, e.g.
2555 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2556 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2559 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2560 const char __user *, put_old)
2562 struct vfsmount *tmp;
2563 struct path new, old, parent_path, root_parent, root;
2566 if (!capable(CAP_SYS_ADMIN))
2569 error = user_path_dir(new_root, &new);
2573 error = user_path_dir(put_old, &old);
2577 error = security_sb_pivotroot(&old, &new);
2583 get_fs_root(current->fs, &root);
2584 down_write(&namespace_sem);
2585 mutex_lock(&old.dentry->d_inode->i_mutex);
2587 if (IS_MNT_SHARED(old.mnt) ||
2588 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2589 IS_MNT_SHARED(root.mnt->mnt_parent))
2591 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2594 if (cant_mount(old.dentry))
2596 if (d_unlinked(new.dentry))
2598 if (d_unlinked(old.dentry))
2601 if (new.mnt == root.mnt ||
2602 old.mnt == root.mnt)
2603 goto out2; /* loop, on the same file system */
2605 if (root.mnt->mnt_root != root.dentry)
2606 goto out2; /* not a mountpoint */
2607 if (root.mnt->mnt_parent == root.mnt)
2608 goto out2; /* not attached */
2609 if (new.mnt->mnt_root != new.dentry)
2610 goto out2; /* not a mountpoint */
2611 if (new.mnt->mnt_parent == new.mnt)
2612 goto out2; /* not attached */
2613 /* make sure we can reach put_old from new_root */
2615 if (tmp != new.mnt) {
2617 if (tmp->mnt_parent == tmp)
2618 goto out2; /* already mounted on put_old */
2619 if (tmp->mnt_parent == new.mnt)
2621 tmp = tmp->mnt_parent;
2623 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2625 } else if (!is_subdir(old.dentry, new.dentry))
2627 br_write_lock(vfsmount_lock);
2628 detach_mnt(new.mnt, &parent_path);
2629 detach_mnt(root.mnt, &root_parent);
2630 /* mount old root on put_old */
2631 attach_mnt(root.mnt, &old);
2632 /* mount new_root on / */
2633 attach_mnt(new.mnt, &root_parent);
2634 touch_mnt_namespace(current->nsproxy->mnt_ns);
2635 br_write_unlock(vfsmount_lock);
2636 chroot_fs_refs(&root, &new);
2639 path_put(&root_parent);
2640 path_put(&parent_path);
2642 mutex_unlock(&old.dentry->d_inode->i_mutex);
2643 up_write(&namespace_sem);
2652 static void __init init_mount_tree(void)
2654 struct vfsmount *mnt;
2655 struct mnt_namespace *ns;
2658 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2660 panic("Can't create rootfs");
2662 ns = create_mnt_ns(mnt);
2664 panic("Can't allocate initial namespace");
2666 init_task.nsproxy->mnt_ns = ns;
2669 root.mnt = ns->root;
2670 root.dentry = ns->root->mnt_root;
2672 set_fs_pwd(current->fs, &root);
2673 set_fs_root(current->fs, &root);
2676 void __init mnt_init(void)
2681 init_rwsem(&namespace_sem);
2683 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2684 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2686 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2688 if (!mount_hashtable)
2689 panic("Failed to allocate mount hash table\n");
2691 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2693 for (u = 0; u < HASH_SIZE; u++)
2694 INIT_LIST_HEAD(&mount_hashtable[u]);
2696 br_lock_init(vfsmount_lock);
2700 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2702 fs_kobj = kobject_create_and_add("fs", NULL);
2704 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2709 void put_mnt_ns(struct mnt_namespace *ns)
2711 LIST_HEAD(umount_list);
2713 if (!atomic_dec_and_test(&ns->count))
2715 down_write(&namespace_sem);
2716 br_write_lock(vfsmount_lock);
2717 umount_tree(ns->root, 0, &umount_list);
2718 br_write_unlock(vfsmount_lock);
2719 up_write(&namespace_sem);
2720 release_mounts(&umount_list);
2723 EXPORT_SYMBOL(put_mnt_ns);
2725 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2727 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2729 EXPORT_SYMBOL_GPL(kern_mount_data);