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);
359 EXPORT_SYMBOL_GPL(mnt_want_write);
362 * mnt_clone_write - get write access to a mount
363 * @mnt: the mount on which to take a write
365 * This is effectively like mnt_want_write, except
366 * it must only be used to take an extra write reference
367 * on a mountpoint that we already know has a write reference
368 * on it. This allows some optimisation.
370 * After finished, mnt_drop_write must be called as usual to
371 * drop the reference.
373 int mnt_clone_write(struct vfsmount *mnt)
375 /* superblock may be r/o */
376 if (__mnt_is_readonly(mnt))
379 mnt_inc_writers(mnt);
383 EXPORT_SYMBOL_GPL(mnt_clone_write);
386 * mnt_want_write_file - get write access to a file's mount
387 * @file: the file who's mount on which to take a write
389 * This is like mnt_want_write, but it takes a file and can
390 * do some optimisations if the file is open for write already
392 int mnt_want_write_file(struct file *file)
394 struct inode *inode = file->f_dentry->d_inode;
395 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
396 return mnt_want_write(file->f_path.mnt);
398 return mnt_clone_write(file->f_path.mnt);
400 EXPORT_SYMBOL_GPL(mnt_want_write_file);
403 * mnt_drop_write - give up write access to a mount
404 * @mnt: the mount on which to give up write access
406 * Tells the low-level filesystem that we are done
407 * performing writes to it. Must be matched with
408 * mnt_want_write() call above.
410 void mnt_drop_write(struct vfsmount *mnt)
413 mnt_dec_writers(mnt);
416 EXPORT_SYMBOL_GPL(mnt_drop_write);
418 static int mnt_make_readonly(struct vfsmount *mnt)
422 br_write_lock(vfsmount_lock);
423 mnt->mnt_flags |= MNT_WRITE_HOLD;
425 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
426 * should be visible before we do.
431 * With writers on hold, if this value is zero, then there are
432 * definitely no active writers (although held writers may subsequently
433 * increment the count, they'll have to wait, and decrement it after
434 * seeing MNT_READONLY).
436 * It is OK to have counter incremented on one CPU and decremented on
437 * another: the sum will add up correctly. The danger would be when we
438 * sum up each counter, if we read a counter before it is incremented,
439 * but then read another CPU's count which it has been subsequently
440 * decremented from -- we would see more decrements than we should.
441 * MNT_WRITE_HOLD protects against this scenario, because
442 * mnt_want_write first increments count, then smp_mb, then spins on
443 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
444 * we're counting up here.
446 if (mnt_get_writers(mnt) > 0)
449 mnt->mnt_flags |= MNT_READONLY;
451 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
452 * that become unheld will see MNT_READONLY.
455 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
456 br_write_unlock(vfsmount_lock);
460 static void __mnt_unmake_readonly(struct vfsmount *mnt)
462 br_write_lock(vfsmount_lock);
463 mnt->mnt_flags &= ~MNT_READONLY;
464 br_write_unlock(vfsmount_lock);
467 static void free_vfsmnt(struct vfsmount *mnt)
469 kfree(mnt->mnt_devname);
472 free_percpu(mnt->mnt_pcp);
474 kmem_cache_free(mnt_cache, mnt);
478 * find the first or last mount at @dentry on vfsmount @mnt depending on
479 * @dir. If @dir is set return the first mount else return the last mount.
480 * vfsmount_lock must be held for read or write.
482 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
485 struct list_head *head = mount_hashtable + hash(mnt, dentry);
486 struct list_head *tmp = head;
487 struct vfsmount *p, *found = NULL;
490 tmp = dir ? tmp->next : tmp->prev;
494 p = list_entry(tmp, struct vfsmount, mnt_hash);
495 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
504 * lookup_mnt increments the ref count before returning
505 * the vfsmount struct.
507 struct vfsmount *lookup_mnt(struct path *path)
509 struct vfsmount *child_mnt;
511 br_read_lock(vfsmount_lock);
512 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
514 br_read_unlock(vfsmount_lock);
518 static inline int check_mnt(struct vfsmount *mnt)
520 return mnt->mnt_ns == current->nsproxy->mnt_ns;
524 * vfsmount lock must be held for write
526 static void touch_mnt_namespace(struct mnt_namespace *ns)
530 wake_up_interruptible(&ns->poll);
535 * vfsmount lock must be held for write
537 static void __touch_mnt_namespace(struct mnt_namespace *ns)
539 if (ns && ns->event != event) {
541 wake_up_interruptible(&ns->poll);
546 * Clear dentry's mounted state if it has no remaining mounts.
547 * vfsmount_lock must be held for write.
549 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry)
553 for (u = 0; u < HASH_SIZE; u++) {
556 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
557 if (p->mnt_mountpoint == dentry)
561 spin_lock(&dentry->d_lock);
562 dentry->d_flags &= ~DCACHE_MOUNTED;
563 spin_unlock(&dentry->d_lock);
567 * vfsmount lock must be held for write
569 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
571 old_path->dentry = mnt->mnt_mountpoint;
572 old_path->mnt = mnt->mnt_parent;
573 mnt->mnt_parent = mnt;
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 list_del_init(&mnt->mnt_child);
576 list_del_init(&mnt->mnt_hash);
577 dentry_reset_mounted(old_path->mnt, old_path->dentry);
581 * vfsmount lock must be held for write
583 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
584 struct vfsmount *child_mnt)
586 child_mnt->mnt_parent = mntget(mnt);
587 child_mnt->mnt_mountpoint = dget(dentry);
588 spin_lock(&dentry->d_lock);
589 dentry->d_flags |= DCACHE_MOUNTED;
590 spin_unlock(&dentry->d_lock);
594 * vfsmount lock must be held for write
596 static void attach_mnt(struct vfsmount *mnt, struct path *path)
598 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
599 list_add_tail(&mnt->mnt_hash, mount_hashtable +
600 hash(path->mnt, path->dentry));
601 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
604 static inline void __mnt_make_longterm(struct vfsmount *mnt)
607 atomic_inc(&mnt->mnt_longterm);
611 /* needs vfsmount lock for write */
612 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
615 atomic_dec(&mnt->mnt_longterm);
620 * vfsmount lock must be held for write
622 static void commit_tree(struct vfsmount *mnt)
624 struct vfsmount *parent = mnt->mnt_parent;
627 struct mnt_namespace *n = parent->mnt_ns;
629 BUG_ON(parent == mnt);
631 list_add_tail(&head, &mnt->mnt_list);
632 list_for_each_entry(m, &head, mnt_list) {
634 __mnt_make_longterm(m);
637 list_splice(&head, n->list.prev);
639 list_add_tail(&mnt->mnt_hash, mount_hashtable +
640 hash(parent, mnt->mnt_mountpoint));
641 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
642 touch_mnt_namespace(n);
645 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
647 struct list_head *next = p->mnt_mounts.next;
648 if (next == &p->mnt_mounts) {
652 next = p->mnt_child.next;
653 if (next != &p->mnt_parent->mnt_mounts)
658 return list_entry(next, struct vfsmount, mnt_child);
661 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
663 struct list_head *prev = p->mnt_mounts.prev;
664 while (prev != &p->mnt_mounts) {
665 p = list_entry(prev, struct vfsmount, mnt_child);
666 prev = p->mnt_mounts.prev;
672 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
674 struct vfsmount *mnt;
678 return ERR_PTR(-ENODEV);
680 mnt = alloc_vfsmnt(name);
682 return ERR_PTR(-ENOMEM);
684 if (flags & MS_KERNMOUNT)
685 mnt->mnt_flags = MNT_INTERNAL;
687 root = mount_fs(type, flags, name, data);
690 return ERR_CAST(root);
693 mnt->mnt_root = root;
694 mnt->mnt_sb = root->d_sb;
695 mnt->mnt_mountpoint = mnt->mnt_root;
696 mnt->mnt_parent = mnt;
699 EXPORT_SYMBOL_GPL(vfs_kern_mount);
701 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
704 struct super_block *sb = old->mnt_sb;
705 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
708 if (flag & (CL_SLAVE | CL_PRIVATE))
709 mnt->mnt_group_id = 0; /* not a peer of original */
711 mnt->mnt_group_id = old->mnt_group_id;
713 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
714 int err = mnt_alloc_group_id(mnt);
719 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
720 atomic_inc(&sb->s_active);
722 mnt->mnt_root = dget(root);
723 mnt->mnt_mountpoint = mnt->mnt_root;
724 mnt->mnt_parent = mnt;
726 if (flag & CL_SLAVE) {
727 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
728 mnt->mnt_master = old;
729 CLEAR_MNT_SHARED(mnt);
730 } else if (!(flag & CL_PRIVATE)) {
731 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
732 list_add(&mnt->mnt_share, &old->mnt_share);
733 if (IS_MNT_SLAVE(old))
734 list_add(&mnt->mnt_slave, &old->mnt_slave);
735 mnt->mnt_master = old->mnt_master;
737 if (flag & CL_MAKE_SHARED)
740 /* stick the duplicate mount on the same expiry list
741 * as the original if that was on one */
742 if (flag & CL_EXPIRE) {
743 if (!list_empty(&old->mnt_expire))
744 list_add(&mnt->mnt_expire, &old->mnt_expire);
754 static inline void mntfree(struct vfsmount *mnt)
756 struct super_block *sb = mnt->mnt_sb;
759 * This probably indicates that somebody messed
760 * up a mnt_want/drop_write() pair. If this
761 * happens, the filesystem was probably unable
762 * to make r/w->r/o transitions.
765 * The locking used to deal with mnt_count decrement provides barriers,
766 * so mnt_get_writers() below is safe.
768 WARN_ON(mnt_get_writers(mnt));
769 fsnotify_vfsmount_delete(mnt);
772 deactivate_super(sb);
775 static void mntput_no_expire(struct vfsmount *mnt)
779 br_read_lock(vfsmount_lock);
780 if (likely(atomic_read(&mnt->mnt_longterm))) {
782 br_read_unlock(vfsmount_lock);
785 br_read_unlock(vfsmount_lock);
787 br_write_lock(vfsmount_lock);
789 if (mnt_get_count(mnt)) {
790 br_write_unlock(vfsmount_lock);
795 if (likely(mnt_get_count(mnt)))
797 br_write_lock(vfsmount_lock);
799 if (unlikely(mnt->mnt_pinned)) {
800 mnt_add_count(mnt, mnt->mnt_pinned + 1);
802 br_write_unlock(vfsmount_lock);
803 acct_auto_close_mnt(mnt);
806 br_write_unlock(vfsmount_lock);
810 void mntput(struct vfsmount *mnt)
813 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
814 if (unlikely(mnt->mnt_expiry_mark))
815 mnt->mnt_expiry_mark = 0;
816 mntput_no_expire(mnt);
819 EXPORT_SYMBOL(mntput);
821 struct vfsmount *mntget(struct vfsmount *mnt)
827 EXPORT_SYMBOL(mntget);
829 void mnt_pin(struct vfsmount *mnt)
831 br_write_lock(vfsmount_lock);
833 br_write_unlock(vfsmount_lock);
835 EXPORT_SYMBOL(mnt_pin);
837 void mnt_unpin(struct vfsmount *mnt)
839 br_write_lock(vfsmount_lock);
840 if (mnt->mnt_pinned) {
844 br_write_unlock(vfsmount_lock);
846 EXPORT_SYMBOL(mnt_unpin);
848 static inline void mangle(struct seq_file *m, const char *s)
850 seq_escape(m, s, " \t\n\\");
854 * Simple .show_options callback for filesystems which don't want to
855 * implement more complex mount option showing.
857 * See also save_mount_options().
859 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
864 options = rcu_dereference(mnt->mnt_sb->s_options);
866 if (options != NULL && options[0]) {
874 EXPORT_SYMBOL(generic_show_options);
877 * If filesystem uses generic_show_options(), this function should be
878 * called from the fill_super() callback.
880 * The .remount_fs callback usually needs to be handled in a special
881 * way, to make sure, that previous options are not overwritten if the
884 * Also note, that if the filesystem's .remount_fs function doesn't
885 * reset all options to their default value, but changes only newly
886 * given options, then the displayed options will not reflect reality
889 void save_mount_options(struct super_block *sb, char *options)
891 BUG_ON(sb->s_options);
892 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
894 EXPORT_SYMBOL(save_mount_options);
896 void replace_mount_options(struct super_block *sb, char *options)
898 char *old = sb->s_options;
899 rcu_assign_pointer(sb->s_options, options);
905 EXPORT_SYMBOL(replace_mount_options);
907 #ifdef CONFIG_PROC_FS
909 static void *m_start(struct seq_file *m, loff_t *pos)
911 struct proc_mounts *p = m->private;
913 down_read(&namespace_sem);
914 return seq_list_start(&p->ns->list, *pos);
917 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
919 struct proc_mounts *p = m->private;
921 return seq_list_next(v, &p->ns->list, pos);
924 static void m_stop(struct seq_file *m, void *v)
926 up_read(&namespace_sem);
929 int mnt_had_events(struct proc_mounts *p)
931 struct mnt_namespace *ns = p->ns;
934 br_read_lock(vfsmount_lock);
935 if (p->m.poll_event != ns->event) {
936 p->m.poll_event = ns->event;
939 br_read_unlock(vfsmount_lock);
944 struct proc_fs_info {
949 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
951 static const struct proc_fs_info fs_info[] = {
952 { MS_SYNCHRONOUS, ",sync" },
953 { MS_DIRSYNC, ",dirsync" },
954 { MS_MANDLOCK, ",mand" },
957 const struct proc_fs_info *fs_infop;
959 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
960 if (sb->s_flags & fs_infop->flag)
961 seq_puts(m, fs_infop->str);
964 return security_sb_show_options(m, sb);
967 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
969 static const struct proc_fs_info mnt_info[] = {
970 { MNT_NOSUID, ",nosuid" },
971 { MNT_NODEV, ",nodev" },
972 { MNT_NOEXEC, ",noexec" },
973 { MNT_NOATIME, ",noatime" },
974 { MNT_NODIRATIME, ",nodiratime" },
975 { MNT_RELATIME, ",relatime" },
978 const struct proc_fs_info *fs_infop;
980 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
981 if (mnt->mnt_flags & fs_infop->flag)
982 seq_puts(m, fs_infop->str);
986 static void show_type(struct seq_file *m, struct super_block *sb)
988 mangle(m, sb->s_type->name);
989 if (sb->s_subtype && sb->s_subtype[0]) {
991 mangle(m, sb->s_subtype);
995 static int show_vfsmnt(struct seq_file *m, void *v)
997 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
999 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1001 if (mnt->mnt_sb->s_op->show_devname) {
1002 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1006 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1009 seq_path(m, &mnt_path, " \t\n\\");
1011 show_type(m, mnt->mnt_sb);
1012 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
1013 err = show_sb_opts(m, mnt->mnt_sb);
1016 show_mnt_opts(m, mnt);
1017 if (mnt->mnt_sb->s_op->show_options)
1018 err = mnt->mnt_sb->s_op->show_options(m, mnt);
1019 seq_puts(m, " 0 0\n");
1024 const struct seq_operations mounts_op = {
1031 static int show_mountinfo(struct seq_file *m, void *v)
1033 struct proc_mounts *p = m->private;
1034 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1035 struct super_block *sb = mnt->mnt_sb;
1036 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1037 struct path root = p->root;
1040 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1041 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1042 if (sb->s_op->show_path)
1043 err = sb->s_op->show_path(m, mnt);
1045 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1049 seq_path_root(m, &mnt_path, &root, " \t\n\\");
1050 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
1052 * Mountpoint is outside root, discard that one. Ugly,
1053 * but less so than trying to do that in iterator in a
1054 * race-free way (due to renames).
1058 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1059 show_mnt_opts(m, mnt);
1061 /* Tagged fields ("foo:X" or "bar") */
1062 if (IS_MNT_SHARED(mnt))
1063 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1064 if (IS_MNT_SLAVE(mnt)) {
1065 int master = mnt->mnt_master->mnt_group_id;
1066 int dom = get_dominating_id(mnt, &p->root);
1067 seq_printf(m, " master:%i", master);
1068 if (dom && dom != master)
1069 seq_printf(m, " propagate_from:%i", dom);
1071 if (IS_MNT_UNBINDABLE(mnt))
1072 seq_puts(m, " unbindable");
1074 /* Filesystem specific data */
1078 if (sb->s_op->show_devname)
1079 err = sb->s_op->show_devname(m, mnt);
1081 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1084 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1085 err = show_sb_opts(m, sb);
1088 if (sb->s_op->show_options)
1089 err = sb->s_op->show_options(m, mnt);
1095 const struct seq_operations mountinfo_op = {
1099 .show = show_mountinfo,
1102 static int show_vfsstat(struct seq_file *m, void *v)
1104 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1105 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1109 if (mnt->mnt_sb->s_op->show_devname) {
1110 seq_puts(m, "device ");
1111 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1113 if (mnt->mnt_devname) {
1114 seq_puts(m, "device ");
1115 mangle(m, mnt->mnt_devname);
1117 seq_puts(m, "no device");
1121 seq_puts(m, " mounted on ");
1122 seq_path(m, &mnt_path, " \t\n\\");
1125 /* file system type */
1126 seq_puts(m, "with fstype ");
1127 show_type(m, mnt->mnt_sb);
1129 /* optional statistics */
1130 if (mnt->mnt_sb->s_op->show_stats) {
1133 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1140 const struct seq_operations mountstats_op = {
1144 .show = show_vfsstat,
1146 #endif /* CONFIG_PROC_FS */
1149 * may_umount_tree - check if a mount tree is busy
1150 * @mnt: root of mount tree
1152 * This is called to check if a tree of mounts has any
1153 * open files, pwds, chroots or sub mounts that are
1156 int may_umount_tree(struct vfsmount *mnt)
1158 int actual_refs = 0;
1159 int minimum_refs = 0;
1162 /* write lock needed for mnt_get_count */
1163 br_write_lock(vfsmount_lock);
1164 for (p = mnt; p; p = next_mnt(p, mnt)) {
1165 actual_refs += mnt_get_count(p);
1168 br_write_unlock(vfsmount_lock);
1170 if (actual_refs > minimum_refs)
1176 EXPORT_SYMBOL(may_umount_tree);
1179 * may_umount - check if a mount point is busy
1180 * @mnt: root of mount
1182 * This is called to check if a mount point has any
1183 * open files, pwds, chroots or sub mounts. If the
1184 * mount has sub mounts this will return busy
1185 * regardless of whether the sub mounts are busy.
1187 * Doesn't take quota and stuff into account. IOW, in some cases it will
1188 * give false negatives. The main reason why it's here is that we need
1189 * a non-destructive way to look for easily umountable filesystems.
1191 int may_umount(struct vfsmount *mnt)
1194 down_read(&namespace_sem);
1195 br_write_lock(vfsmount_lock);
1196 if (propagate_mount_busy(mnt, 2))
1198 br_write_unlock(vfsmount_lock);
1199 up_read(&namespace_sem);
1203 EXPORT_SYMBOL(may_umount);
1205 void release_mounts(struct list_head *head)
1207 struct vfsmount *mnt;
1208 while (!list_empty(head)) {
1209 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1210 list_del_init(&mnt->mnt_hash);
1211 if (mnt->mnt_parent != mnt) {
1212 struct dentry *dentry;
1215 br_write_lock(vfsmount_lock);
1216 dentry = mnt->mnt_mountpoint;
1217 m = mnt->mnt_parent;
1218 mnt->mnt_mountpoint = mnt->mnt_root;
1219 mnt->mnt_parent = mnt;
1221 br_write_unlock(vfsmount_lock);
1230 * vfsmount lock must be held for write
1231 * namespace_sem must be held for write
1233 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1235 LIST_HEAD(tmp_list);
1238 for (p = mnt; p; p = next_mnt(p, mnt))
1239 list_move(&p->mnt_hash, &tmp_list);
1242 propagate_umount(&tmp_list);
1244 list_for_each_entry(p, &tmp_list, mnt_hash) {
1245 list_del_init(&p->mnt_expire);
1246 list_del_init(&p->mnt_list);
1247 __touch_mnt_namespace(p->mnt_ns);
1249 __mnt_make_shortterm(p);
1250 list_del_init(&p->mnt_child);
1251 if (p->mnt_parent != p) {
1252 p->mnt_parent->mnt_ghosts++;
1253 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint);
1255 change_mnt_propagation(p, MS_PRIVATE);
1257 list_splice(&tmp_list, kill);
1260 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1262 static int do_umount(struct vfsmount *mnt, int flags)
1264 struct super_block *sb = mnt->mnt_sb;
1266 LIST_HEAD(umount_list);
1268 retval = security_sb_umount(mnt, flags);
1273 * Allow userspace to request a mountpoint be expired rather than
1274 * unmounting unconditionally. Unmount only happens if:
1275 * (1) the mark is already set (the mark is cleared by mntput())
1276 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1278 if (flags & MNT_EXPIRE) {
1279 if (mnt == current->fs->root.mnt ||
1280 flags & (MNT_FORCE | MNT_DETACH))
1284 * probably don't strictly need the lock here if we examined
1285 * all race cases, but it's a slowpath.
1287 br_write_lock(vfsmount_lock);
1288 if (mnt_get_count(mnt) != 2) {
1289 br_write_unlock(vfsmount_lock);
1292 br_write_unlock(vfsmount_lock);
1294 if (!xchg(&mnt->mnt_expiry_mark, 1))
1299 * If we may have to abort operations to get out of this
1300 * mount, and they will themselves hold resources we must
1301 * allow the fs to do things. In the Unix tradition of
1302 * 'Gee thats tricky lets do it in userspace' the umount_begin
1303 * might fail to complete on the first run through as other tasks
1304 * must return, and the like. Thats for the mount program to worry
1305 * about for the moment.
1308 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1309 sb->s_op->umount_begin(sb);
1313 * No sense to grab the lock for this test, but test itself looks
1314 * somewhat bogus. Suggestions for better replacement?
1315 * Ho-hum... In principle, we might treat that as umount + switch
1316 * to rootfs. GC would eventually take care of the old vfsmount.
1317 * Actually it makes sense, especially if rootfs would contain a
1318 * /reboot - static binary that would close all descriptors and
1319 * call reboot(9). Then init(8) could umount root and exec /reboot.
1321 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1323 * Special case for "unmounting" root ...
1324 * we just try to remount it readonly.
1326 down_write(&sb->s_umount);
1327 if (!(sb->s_flags & MS_RDONLY))
1328 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1329 up_write(&sb->s_umount);
1333 down_write(&namespace_sem);
1334 br_write_lock(vfsmount_lock);
1337 if (!(flags & MNT_DETACH))
1338 shrink_submounts(mnt, &umount_list);
1341 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1342 if (!list_empty(&mnt->mnt_list))
1343 umount_tree(mnt, 1, &umount_list);
1346 br_write_unlock(vfsmount_lock);
1347 up_write(&namespace_sem);
1348 release_mounts(&umount_list);
1353 * Now umount can handle mount points as well as block devices.
1354 * This is important for filesystems which use unnamed block devices.
1356 * We now support a flag for forced unmount like the other 'big iron'
1357 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1360 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1364 int lookup_flags = 0;
1366 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1369 if (!(flags & UMOUNT_NOFOLLOW))
1370 lookup_flags |= LOOKUP_FOLLOW;
1372 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1376 if (path.dentry != path.mnt->mnt_root)
1378 if (!check_mnt(path.mnt))
1382 if (!capable(CAP_SYS_ADMIN))
1385 retval = do_umount(path.mnt, flags);
1387 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1389 mntput_no_expire(path.mnt);
1394 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1397 * The 2.0 compatible umount. No flags.
1399 SYSCALL_DEFINE1(oldumount, char __user *, name)
1401 return sys_umount(name, 0);
1406 static int mount_is_safe(struct path *path)
1408 if (capable(CAP_SYS_ADMIN))
1412 if (S_ISLNK(path->dentry->d_inode->i_mode))
1414 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1415 if (current_uid() != path->dentry->d_inode->i_uid)
1418 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1424 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1427 struct vfsmount *res, *p, *q, *r, *s;
1430 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1433 res = q = clone_mnt(mnt, dentry, flag);
1436 q->mnt_mountpoint = mnt->mnt_mountpoint;
1439 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1440 if (!is_subdir(r->mnt_mountpoint, dentry))
1443 for (s = r; s; s = next_mnt(s, r)) {
1444 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1445 s = skip_mnt_tree(s);
1448 while (p != s->mnt_parent) {
1454 path.dentry = p->mnt_mountpoint;
1455 q = clone_mnt(p, p->mnt_root, flag);
1458 br_write_lock(vfsmount_lock);
1459 list_add_tail(&q->mnt_list, &res->mnt_list);
1460 attach_mnt(q, &path);
1461 br_write_unlock(vfsmount_lock);
1467 LIST_HEAD(umount_list);
1468 br_write_lock(vfsmount_lock);
1469 umount_tree(res, 0, &umount_list);
1470 br_write_unlock(vfsmount_lock);
1471 release_mounts(&umount_list);
1476 struct vfsmount *collect_mounts(struct path *path)
1478 struct vfsmount *tree;
1479 down_write(&namespace_sem);
1480 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1481 up_write(&namespace_sem);
1485 void drop_collected_mounts(struct vfsmount *mnt)
1487 LIST_HEAD(umount_list);
1488 down_write(&namespace_sem);
1489 br_write_lock(vfsmount_lock);
1490 umount_tree(mnt, 0, &umount_list);
1491 br_write_unlock(vfsmount_lock);
1492 up_write(&namespace_sem);
1493 release_mounts(&umount_list);
1496 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1497 struct vfsmount *root)
1499 struct vfsmount *mnt;
1500 int res = f(root, arg);
1503 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1511 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1515 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1516 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1517 mnt_release_group_id(p);
1521 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1525 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1526 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1527 int err = mnt_alloc_group_id(p);
1529 cleanup_group_ids(mnt, p);
1539 * @source_mnt : mount tree to be attached
1540 * @nd : place the mount tree @source_mnt is attached
1541 * @parent_nd : if non-null, detach the source_mnt from its parent and
1542 * store the parent mount and mountpoint dentry.
1543 * (done when source_mnt is moved)
1545 * NOTE: in the table below explains the semantics when a source mount
1546 * of a given type is attached to a destination mount of a given type.
1547 * ---------------------------------------------------------------------------
1548 * | BIND MOUNT OPERATION |
1549 * |**************************************************************************
1550 * | source-->| shared | private | slave | unbindable |
1554 * |**************************************************************************
1555 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1557 * |non-shared| shared (+) | private | slave (*) | invalid |
1558 * ***************************************************************************
1559 * A bind operation clones the source mount and mounts the clone on the
1560 * destination mount.
1562 * (++) the cloned mount is propagated to all the mounts in the propagation
1563 * tree of the destination mount and the cloned mount is added to
1564 * the peer group of the source mount.
1565 * (+) the cloned mount is created under the destination mount and is marked
1566 * as shared. The cloned mount is added to the peer group of the source
1568 * (+++) the mount is propagated to all the mounts in the propagation tree
1569 * of the destination mount and the cloned mount is made slave
1570 * of the same master as that of the source mount. The cloned mount
1571 * is marked as 'shared and slave'.
1572 * (*) the cloned mount is made a slave of the same master as that of the
1575 * ---------------------------------------------------------------------------
1576 * | MOVE MOUNT OPERATION |
1577 * |**************************************************************************
1578 * | source-->| shared | private | slave | unbindable |
1582 * |**************************************************************************
1583 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1585 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1586 * ***************************************************************************
1588 * (+) the mount is moved to the destination. And is then propagated to
1589 * all the mounts in the propagation tree of the destination mount.
1590 * (+*) the mount is moved to the destination.
1591 * (+++) the mount is moved to the destination and is then propagated to
1592 * all the mounts belonging to the destination mount's propagation tree.
1593 * the mount is marked as 'shared and slave'.
1594 * (*) the mount continues to be a slave at the new location.
1596 * if the source mount is a tree, the operations explained above is
1597 * applied to each mount in the tree.
1598 * Must be called without spinlocks held, since this function can sleep
1601 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1602 struct path *path, struct path *parent_path)
1604 LIST_HEAD(tree_list);
1605 struct vfsmount *dest_mnt = path->mnt;
1606 struct dentry *dest_dentry = path->dentry;
1607 struct vfsmount *child, *p;
1610 if (IS_MNT_SHARED(dest_mnt)) {
1611 err = invent_group_ids(source_mnt, true);
1615 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1617 goto out_cleanup_ids;
1619 br_write_lock(vfsmount_lock);
1621 if (IS_MNT_SHARED(dest_mnt)) {
1622 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1626 detach_mnt(source_mnt, parent_path);
1627 attach_mnt(source_mnt, path);
1628 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1630 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1631 commit_tree(source_mnt);
1634 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1635 list_del_init(&child->mnt_hash);
1638 br_write_unlock(vfsmount_lock);
1643 if (IS_MNT_SHARED(dest_mnt))
1644 cleanup_group_ids(source_mnt, NULL);
1649 static int lock_mount(struct path *path)
1651 struct vfsmount *mnt;
1653 mutex_lock(&path->dentry->d_inode->i_mutex);
1654 if (unlikely(cant_mount(path->dentry))) {
1655 mutex_unlock(&path->dentry->d_inode->i_mutex);
1658 down_write(&namespace_sem);
1659 mnt = lookup_mnt(path);
1662 up_write(&namespace_sem);
1663 mutex_unlock(&path->dentry->d_inode->i_mutex);
1666 path->dentry = dget(mnt->mnt_root);
1670 static void unlock_mount(struct path *path)
1672 up_write(&namespace_sem);
1673 mutex_unlock(&path->dentry->d_inode->i_mutex);
1676 static int graft_tree(struct vfsmount *mnt, struct path *path)
1678 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1681 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1682 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1685 if (d_unlinked(path->dentry))
1688 return attach_recursive_mnt(mnt, path, NULL);
1692 * Sanity check the flags to change_mnt_propagation.
1695 static int flags_to_propagation_type(int flags)
1697 int type = flags & ~(MS_REC | MS_SILENT);
1699 /* Fail if any non-propagation flags are set */
1700 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1702 /* Only one propagation flag should be set */
1703 if (!is_power_of_2(type))
1709 * recursively change the type of the mountpoint.
1711 static int do_change_type(struct path *path, int flag)
1713 struct vfsmount *m, *mnt = path->mnt;
1714 int recurse = flag & MS_REC;
1718 if (!capable(CAP_SYS_ADMIN))
1721 if (path->dentry != path->mnt->mnt_root)
1724 type = flags_to_propagation_type(flag);
1728 down_write(&namespace_sem);
1729 if (type == MS_SHARED) {
1730 err = invent_group_ids(mnt, recurse);
1735 br_write_lock(vfsmount_lock);
1736 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1737 change_mnt_propagation(m, type);
1738 br_write_unlock(vfsmount_lock);
1741 up_write(&namespace_sem);
1746 * do loopback mount.
1748 static int do_loopback(struct path *path, char *old_name,
1751 LIST_HEAD(umount_list);
1752 struct path old_path;
1753 struct vfsmount *mnt = NULL;
1754 int err = mount_is_safe(path);
1757 if (!old_name || !*old_name)
1759 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1763 err = lock_mount(path);
1768 if (IS_MNT_UNBINDABLE(old_path.mnt))
1771 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1776 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1778 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1783 err = graft_tree(mnt, path);
1785 br_write_lock(vfsmount_lock);
1786 umount_tree(mnt, 0, &umount_list);
1787 br_write_unlock(vfsmount_lock);
1791 release_mounts(&umount_list);
1793 path_put(&old_path);
1797 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1800 int readonly_request = 0;
1802 if (ms_flags & MS_RDONLY)
1803 readonly_request = 1;
1804 if (readonly_request == __mnt_is_readonly(mnt))
1807 if (readonly_request)
1808 error = mnt_make_readonly(mnt);
1810 __mnt_unmake_readonly(mnt);
1815 * change filesystem flags. dir should be a physical root of filesystem.
1816 * If you've mounted a non-root directory somewhere and want to do remount
1817 * on it - tough luck.
1819 static int do_remount(struct path *path, int flags, int mnt_flags,
1823 struct super_block *sb = path->mnt->mnt_sb;
1825 if (!capable(CAP_SYS_ADMIN))
1828 if (!check_mnt(path->mnt))
1831 if (path->dentry != path->mnt->mnt_root)
1834 err = security_sb_remount(sb, data);
1838 down_write(&sb->s_umount);
1839 if (flags & MS_BIND)
1840 err = change_mount_flags(path->mnt, flags);
1842 err = do_remount_sb(sb, flags, data, 0);
1844 br_write_lock(vfsmount_lock);
1845 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1846 path->mnt->mnt_flags = mnt_flags;
1847 br_write_unlock(vfsmount_lock);
1849 up_write(&sb->s_umount);
1851 br_write_lock(vfsmount_lock);
1852 touch_mnt_namespace(path->mnt->mnt_ns);
1853 br_write_unlock(vfsmount_lock);
1858 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1861 for (p = mnt; p; p = next_mnt(p, mnt)) {
1862 if (IS_MNT_UNBINDABLE(p))
1868 static int do_move_mount(struct path *path, char *old_name)
1870 struct path old_path, parent_path;
1873 if (!capable(CAP_SYS_ADMIN))
1875 if (!old_name || !*old_name)
1877 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1881 err = lock_mount(path);
1886 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1889 if (d_unlinked(path->dentry))
1893 if (old_path.dentry != old_path.mnt->mnt_root)
1896 if (old_path.mnt == old_path.mnt->mnt_parent)
1899 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1900 S_ISDIR(old_path.dentry->d_inode->i_mode))
1903 * Don't move a mount residing in a shared parent.
1905 if (old_path.mnt->mnt_parent &&
1906 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1909 * Don't move a mount tree containing unbindable mounts to a destination
1910 * mount which is shared.
1912 if (IS_MNT_SHARED(path->mnt) &&
1913 tree_contains_unbindable(old_path.mnt))
1916 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1917 if (p == old_path.mnt)
1920 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1924 /* if the mount is moved, it should no longer be expire
1926 list_del_init(&old_path.mnt->mnt_expire);
1931 path_put(&parent_path);
1932 path_put(&old_path);
1936 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1939 const char *subtype = strchr(fstype, '.');
1948 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1950 if (!mnt->mnt_sb->s_subtype)
1956 return ERR_PTR(err);
1960 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1962 struct file_system_type *type = get_fs_type(fstype);
1963 struct vfsmount *mnt;
1965 return ERR_PTR(-ENODEV);
1966 mnt = vfs_kern_mount(type, flags, name, data);
1967 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1968 !mnt->mnt_sb->s_subtype)
1969 mnt = fs_set_subtype(mnt, fstype);
1970 put_filesystem(type);
1973 EXPORT_SYMBOL_GPL(do_kern_mount);
1976 * add a mount into a namespace's mount tree
1978 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1982 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1984 err = lock_mount(path);
1989 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1992 /* Refuse the same filesystem on the same mount point */
1994 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1995 path->mnt->mnt_root == path->dentry)
1999 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
2002 newmnt->mnt_flags = mnt_flags;
2003 err = graft_tree(newmnt, path);
2011 * create a new mount for userspace and request it to be added into the
2014 static int do_new_mount(struct path *path, char *type, int flags,
2015 int mnt_flags, char *name, void *data)
2017 struct vfsmount *mnt;
2023 /* we need capabilities... */
2024 if (!capable(CAP_SYS_ADMIN))
2027 mnt = do_kern_mount(type, flags, name, data);
2029 return PTR_ERR(mnt);
2031 err = do_add_mount(mnt, path, mnt_flags);
2037 int finish_automount(struct vfsmount *m, struct path *path)
2040 /* The new mount record should have at least 2 refs to prevent it being
2041 * expired before we get a chance to add it
2043 BUG_ON(mnt_get_count(m) < 2);
2045 if (m->mnt_sb == path->mnt->mnt_sb &&
2046 m->mnt_root == path->dentry) {
2051 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2055 /* remove m from any expiration list it may be on */
2056 if (!list_empty(&m->mnt_expire)) {
2057 down_write(&namespace_sem);
2058 br_write_lock(vfsmount_lock);
2059 list_del_init(&m->mnt_expire);
2060 br_write_unlock(vfsmount_lock);
2061 up_write(&namespace_sem);
2069 * mnt_set_expiry - Put a mount on an expiration list
2070 * @mnt: The mount to list.
2071 * @expiry_list: The list to add the mount to.
2073 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2075 down_write(&namespace_sem);
2076 br_write_lock(vfsmount_lock);
2078 list_add_tail(&mnt->mnt_expire, expiry_list);
2080 br_write_unlock(vfsmount_lock);
2081 up_write(&namespace_sem);
2083 EXPORT_SYMBOL(mnt_set_expiry);
2086 * process a list of expirable mountpoints with the intent of discarding any
2087 * mountpoints that aren't in use and haven't been touched since last we came
2090 void mark_mounts_for_expiry(struct list_head *mounts)
2092 struct vfsmount *mnt, *next;
2093 LIST_HEAD(graveyard);
2096 if (list_empty(mounts))
2099 down_write(&namespace_sem);
2100 br_write_lock(vfsmount_lock);
2102 /* extract from the expiration list every vfsmount that matches the
2103 * following criteria:
2104 * - only referenced by its parent vfsmount
2105 * - still marked for expiry (marked on the last call here; marks are
2106 * cleared by mntput())
2108 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2109 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2110 propagate_mount_busy(mnt, 1))
2112 list_move(&mnt->mnt_expire, &graveyard);
2114 while (!list_empty(&graveyard)) {
2115 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2116 touch_mnt_namespace(mnt->mnt_ns);
2117 umount_tree(mnt, 1, &umounts);
2119 br_write_unlock(vfsmount_lock);
2120 up_write(&namespace_sem);
2122 release_mounts(&umounts);
2125 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2128 * Ripoff of 'select_parent()'
2130 * search the list of submounts for a given mountpoint, and move any
2131 * shrinkable submounts to the 'graveyard' list.
2133 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2135 struct vfsmount *this_parent = parent;
2136 struct list_head *next;
2140 next = this_parent->mnt_mounts.next;
2142 while (next != &this_parent->mnt_mounts) {
2143 struct list_head *tmp = next;
2144 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2147 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2150 * Descend a level if the d_mounts list is non-empty.
2152 if (!list_empty(&mnt->mnt_mounts)) {
2157 if (!propagate_mount_busy(mnt, 1)) {
2158 list_move_tail(&mnt->mnt_expire, graveyard);
2163 * All done at this level ... ascend and resume the search
2165 if (this_parent != parent) {
2166 next = this_parent->mnt_child.next;
2167 this_parent = this_parent->mnt_parent;
2174 * process a list of expirable mountpoints with the intent of discarding any
2175 * submounts of a specific parent mountpoint
2177 * vfsmount_lock must be held for write
2179 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2181 LIST_HEAD(graveyard);
2184 /* extract submounts of 'mountpoint' from the expiration list */
2185 while (select_submounts(mnt, &graveyard)) {
2186 while (!list_empty(&graveyard)) {
2187 m = list_first_entry(&graveyard, struct vfsmount,
2189 touch_mnt_namespace(m->mnt_ns);
2190 umount_tree(m, 1, umounts);
2196 * Some copy_from_user() implementations do not return the exact number of
2197 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2198 * Note that this function differs from copy_from_user() in that it will oops
2199 * on bad values of `to', rather than returning a short copy.
2201 static long exact_copy_from_user(void *to, const void __user * from,
2205 const char __user *f = from;
2208 if (!access_ok(VERIFY_READ, from, n))
2212 if (__get_user(c, f)) {
2223 int copy_mount_options(const void __user * data, unsigned long *where)
2233 if (!(page = __get_free_page(GFP_KERNEL)))
2236 /* We only care that *some* data at the address the user
2237 * gave us is valid. Just in case, we'll zero
2238 * the remainder of the page.
2240 /* copy_from_user cannot cross TASK_SIZE ! */
2241 size = TASK_SIZE - (unsigned long)data;
2242 if (size > PAGE_SIZE)
2245 i = size - exact_copy_from_user((void *)page, data, size);
2251 memset((char *)page + i, 0, PAGE_SIZE - i);
2256 int copy_mount_string(const void __user *data, char **where)
2265 tmp = strndup_user(data, PAGE_SIZE);
2267 return PTR_ERR(tmp);
2274 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2275 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2277 * data is a (void *) that can point to any structure up to
2278 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2279 * information (or be NULL).
2281 * Pre-0.97 versions of mount() didn't have a flags word.
2282 * When the flags word was introduced its top half was required
2283 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2284 * Therefore, if this magic number is present, it carries no information
2285 * and must be discarded.
2287 long do_mount(char *dev_name, char *dir_name, char *type_page,
2288 unsigned long flags, void *data_page)
2295 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2296 flags &= ~MS_MGC_MSK;
2298 /* Basic sanity checks */
2300 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2304 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2306 /* ... and get the mountpoint */
2307 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2311 retval = security_sb_mount(dev_name, &path,
2312 type_page, flags, data_page);
2316 /* Default to relatime unless overriden */
2317 if (!(flags & MS_NOATIME))
2318 mnt_flags |= MNT_RELATIME;
2320 /* Separate the per-mountpoint flags */
2321 if (flags & MS_NOSUID)
2322 mnt_flags |= MNT_NOSUID;
2323 if (flags & MS_NODEV)
2324 mnt_flags |= MNT_NODEV;
2325 if (flags & MS_NOEXEC)
2326 mnt_flags |= MNT_NOEXEC;
2327 if (flags & MS_NOATIME)
2328 mnt_flags |= MNT_NOATIME;
2329 if (flags & MS_NODIRATIME)
2330 mnt_flags |= MNT_NODIRATIME;
2331 if (flags & MS_STRICTATIME)
2332 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2333 if (flags & MS_RDONLY)
2334 mnt_flags |= MNT_READONLY;
2336 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2337 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2340 if (flags & MS_REMOUNT)
2341 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2343 else if (flags & MS_BIND)
2344 retval = do_loopback(&path, dev_name, flags & MS_REC);
2345 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2346 retval = do_change_type(&path, flags);
2347 else if (flags & MS_MOVE)
2348 retval = do_move_mount(&path, dev_name);
2350 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2351 dev_name, data_page);
2357 static struct mnt_namespace *alloc_mnt_ns(void)
2359 struct mnt_namespace *new_ns;
2361 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2363 return ERR_PTR(-ENOMEM);
2364 atomic_set(&new_ns->count, 1);
2365 new_ns->root = NULL;
2366 INIT_LIST_HEAD(&new_ns->list);
2367 init_waitqueue_head(&new_ns->poll);
2372 void mnt_make_longterm(struct vfsmount *mnt)
2374 __mnt_make_longterm(mnt);
2377 void mnt_make_shortterm(struct vfsmount *mnt)
2380 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2382 br_write_lock(vfsmount_lock);
2383 atomic_dec(&mnt->mnt_longterm);
2384 br_write_unlock(vfsmount_lock);
2389 * Allocate a new namespace structure and populate it with contents
2390 * copied from the namespace of the passed in task structure.
2392 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2393 struct fs_struct *fs)
2395 struct mnt_namespace *new_ns;
2396 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2397 struct vfsmount *p, *q;
2399 new_ns = alloc_mnt_ns();
2403 down_write(&namespace_sem);
2404 /* First pass: copy the tree topology */
2405 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2406 CL_COPY_ALL | CL_EXPIRE);
2407 if (!new_ns->root) {
2408 up_write(&namespace_sem);
2410 return ERR_PTR(-ENOMEM);
2412 br_write_lock(vfsmount_lock);
2413 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2414 br_write_unlock(vfsmount_lock);
2417 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2418 * as belonging to new namespace. We have already acquired a private
2419 * fs_struct, so tsk->fs->lock is not needed.
2425 __mnt_make_longterm(q);
2427 if (p == fs->root.mnt) {
2428 fs->root.mnt = mntget(q);
2429 __mnt_make_longterm(q);
2430 mnt_make_shortterm(p);
2433 if (p == fs->pwd.mnt) {
2434 fs->pwd.mnt = mntget(q);
2435 __mnt_make_longterm(q);
2436 mnt_make_shortterm(p);
2440 p = next_mnt(p, mnt_ns->root);
2441 q = next_mnt(q, new_ns->root);
2443 up_write(&namespace_sem);
2453 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2454 struct fs_struct *new_fs)
2456 struct mnt_namespace *new_ns;
2461 if (!(flags & CLONE_NEWNS))
2464 new_ns = dup_mnt_ns(ns, new_fs);
2471 * create_mnt_ns - creates a private namespace and adds a root filesystem
2472 * @mnt: pointer to the new root filesystem mountpoint
2474 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2476 struct mnt_namespace *new_ns;
2478 new_ns = alloc_mnt_ns();
2479 if (!IS_ERR(new_ns)) {
2480 mnt->mnt_ns = new_ns;
2481 __mnt_make_longterm(mnt);
2483 list_add(&new_ns->list, &new_ns->root->mnt_list);
2487 EXPORT_SYMBOL(create_mnt_ns);
2489 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2490 char __user *, type, unsigned long, flags, void __user *, data)
2496 unsigned long data_page;
2498 ret = copy_mount_string(type, &kernel_type);
2502 kernel_dir = getname(dir_name);
2503 if (IS_ERR(kernel_dir)) {
2504 ret = PTR_ERR(kernel_dir);
2508 ret = copy_mount_string(dev_name, &kernel_dev);
2512 ret = copy_mount_options(data, &data_page);
2516 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2517 (void *) data_page);
2519 free_page(data_page);
2523 putname(kernel_dir);
2531 * pivot_root Semantics:
2532 * Moves the root file system of the current process to the directory put_old,
2533 * makes new_root as the new root file system of the current process, and sets
2534 * root/cwd of all processes which had them on the current root to new_root.
2537 * The new_root and put_old must be directories, and must not be on the
2538 * same file system as the current process root. The put_old must be
2539 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2540 * pointed to by put_old must yield the same directory as new_root. No other
2541 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2543 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2544 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2545 * in this situation.
2548 * - we don't move root/cwd if they are not at the root (reason: if something
2549 * cared enough to change them, it's probably wrong to force them elsewhere)
2550 * - it's okay to pick a root that isn't the root of a file system, e.g.
2551 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2552 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2555 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2556 const char __user *, put_old)
2558 struct vfsmount *tmp;
2559 struct path new, old, parent_path, root_parent, root;
2562 if (!capable(CAP_SYS_ADMIN))
2565 error = user_path_dir(new_root, &new);
2569 error = user_path_dir(put_old, &old);
2573 error = security_sb_pivotroot(&old, &new);
2577 get_fs_root(current->fs, &root);
2578 error = lock_mount(&old);
2583 if (IS_MNT_SHARED(old.mnt) ||
2584 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2585 IS_MNT_SHARED(root.mnt->mnt_parent))
2587 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2590 if (d_unlinked(new.dentry))
2592 if (d_unlinked(old.dentry))
2595 if (new.mnt == root.mnt ||
2596 old.mnt == root.mnt)
2597 goto out4; /* loop, on the same file system */
2599 if (root.mnt->mnt_root != root.dentry)
2600 goto out4; /* not a mountpoint */
2601 if (root.mnt->mnt_parent == root.mnt)
2602 goto out4; /* not attached */
2603 if (new.mnt->mnt_root != new.dentry)
2604 goto out4; /* not a mountpoint */
2605 if (new.mnt->mnt_parent == new.mnt)
2606 goto out4; /* not attached */
2607 /* make sure we can reach put_old from new_root */
2609 if (tmp != new.mnt) {
2611 if (tmp->mnt_parent == tmp)
2612 goto out4; /* already mounted on put_old */
2613 if (tmp->mnt_parent == new.mnt)
2615 tmp = tmp->mnt_parent;
2617 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2619 } else if (!is_subdir(old.dentry, new.dentry))
2621 br_write_lock(vfsmount_lock);
2622 detach_mnt(new.mnt, &parent_path);
2623 detach_mnt(root.mnt, &root_parent);
2624 /* mount old root on put_old */
2625 attach_mnt(root.mnt, &old);
2626 /* mount new_root on / */
2627 attach_mnt(new.mnt, &root_parent);
2628 touch_mnt_namespace(current->nsproxy->mnt_ns);
2629 br_write_unlock(vfsmount_lock);
2630 chroot_fs_refs(&root, &new);
2635 path_put(&root_parent);
2636 path_put(&parent_path);
2648 static void __init init_mount_tree(void)
2650 struct vfsmount *mnt;
2651 struct mnt_namespace *ns;
2654 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2656 panic("Can't create rootfs");
2658 ns = create_mnt_ns(mnt);
2660 panic("Can't allocate initial namespace");
2662 init_task.nsproxy->mnt_ns = ns;
2665 root.mnt = ns->root;
2666 root.dentry = ns->root->mnt_root;
2668 set_fs_pwd(current->fs, &root);
2669 set_fs_root(current->fs, &root);
2672 void __init mnt_init(void)
2677 init_rwsem(&namespace_sem);
2679 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2680 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2682 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2684 if (!mount_hashtable)
2685 panic("Failed to allocate mount hash table\n");
2687 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2689 for (u = 0; u < HASH_SIZE; u++)
2690 INIT_LIST_HEAD(&mount_hashtable[u]);
2692 br_lock_init(vfsmount_lock);
2696 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2698 fs_kobj = kobject_create_and_add("fs", NULL);
2700 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2705 void put_mnt_ns(struct mnt_namespace *ns)
2707 LIST_HEAD(umount_list);
2709 if (!atomic_dec_and_test(&ns->count))
2711 down_write(&namespace_sem);
2712 br_write_lock(vfsmount_lock);
2713 umount_tree(ns->root, 0, &umount_list);
2714 br_write_unlock(vfsmount_lock);
2715 up_write(&namespace_sem);
2716 release_mounts(&umount_list);
2719 EXPORT_SYMBOL(put_mnt_ns);
2721 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2723 struct vfsmount *mnt;
2724 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2727 * it is a longterm mount, don't release mnt until
2728 * we unmount before file sys is unregistered
2730 mnt_make_longterm(mnt);
2734 EXPORT_SYMBOL_GPL(kern_mount_data);
2736 void kern_unmount(struct vfsmount *mnt)
2738 /* release long term mount so mount point can be released */
2739 if (!IS_ERR_OR_NULL(mnt)) {
2740 mnt_make_shortterm(mnt);
2744 EXPORT_SYMBOL(kern_unmount);