4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/spinlock.h>
15 #include <linux/percpu.h>
16 #include <linux/init.h>
17 #include <linux/kernel.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/cpumask.h>
21 #include <linux/module.h>
22 #include <linux/sysfs.h>
23 #include <linux/seq_file.h>
24 #include <linux/mnt_namespace.h>
25 #include <linux/namei.h>
26 #include <linux/nsproxy.h>
27 #include <linux/security.h>
28 #include <linux/mount.h>
29 #include <linux/ramfs.h>
30 #include <linux/log2.h>
31 #include <linux/idr.h>
32 #include <linux/fs_struct.h>
33 #include <linux/fsnotify.h>
34 #include <asm/uaccess.h>
35 #include <asm/unistd.h>
39 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
40 #define HASH_SIZE (1UL << HASH_SHIFT)
43 static DEFINE_IDA(mnt_id_ida);
44 static DEFINE_IDA(mnt_group_ida);
45 static DEFINE_SPINLOCK(mnt_id_lock);
46 static int mnt_id_start = 0;
47 static int mnt_group_start = 1;
49 static struct list_head *mount_hashtable __read_mostly;
50 static struct kmem_cache *mnt_cache __read_mostly;
51 static struct rw_semaphore namespace_sem;
54 struct kobject *fs_kobj;
55 EXPORT_SYMBOL_GPL(fs_kobj);
58 * vfsmount lock may be taken for read to prevent changes to the
59 * vfsmount hash, ie. during mountpoint lookups or walking back
62 * It should be taken for write in all cases where the vfsmount
63 * tree or hash is modified or when a vfsmount structure is modified.
65 DEFINE_BRLOCK(vfsmount_lock);
67 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
69 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
71 tmp = tmp + (tmp >> HASH_SHIFT);
72 return tmp & (HASH_SIZE - 1);
75 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
78 * allocation is serialized by namespace_sem, but we need the spinlock to
79 * serialize with freeing.
81 static int mnt_alloc_id(struct vfsmount *mnt)
86 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
87 spin_lock(&mnt_id_lock);
88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
90 mnt_id_start = mnt->mnt_id + 1;
91 spin_unlock(&mnt_id_lock);
98 static void mnt_free_id(struct vfsmount *mnt)
100 int id = mnt->mnt_id;
101 spin_lock(&mnt_id_lock);
102 ida_remove(&mnt_id_ida, id);
103 if (mnt_id_start > id)
105 spin_unlock(&mnt_id_lock);
109 * Allocate a new peer group ID
111 * mnt_group_ida is protected by namespace_sem
113 static int mnt_alloc_group_id(struct vfsmount *mnt)
117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
120 res = ida_get_new_above(&mnt_group_ida,
124 mnt_group_start = mnt->mnt_group_id + 1;
130 * Release a peer group ID
132 void mnt_release_group_id(struct vfsmount *mnt)
134 int id = mnt->mnt_group_id;
135 ida_remove(&mnt_group_ida, id);
136 if (mnt_group_start > id)
137 mnt_group_start = id;
138 mnt->mnt_group_id = 0;
142 * vfsmount lock must be held for read
144 static inline void mnt_add_count(struct vfsmount *mnt, int n)
147 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
156 * vfsmount lock must be held for write
158 unsigned int mnt_get_count(struct vfsmount *mnt)
161 unsigned int count = 0;
164 for_each_possible_cpu(cpu) {
165 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
170 return mnt->mnt_count;
174 static struct vfsmount *alloc_vfsmnt(const char *name)
176 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
180 err = mnt_alloc_id(mnt);
185 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
186 if (!mnt->mnt_devname)
191 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 goto out_free_devname;
195 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
198 mnt->mnt_writers = 0;
201 INIT_LIST_HEAD(&mnt->mnt_hash);
202 INIT_LIST_HEAD(&mnt->mnt_child);
203 INIT_LIST_HEAD(&mnt->mnt_mounts);
204 INIT_LIST_HEAD(&mnt->mnt_list);
205 INIT_LIST_HEAD(&mnt->mnt_expire);
206 INIT_LIST_HEAD(&mnt->mnt_share);
207 INIT_LIST_HEAD(&mnt->mnt_slave_list);
208 INIT_LIST_HEAD(&mnt->mnt_slave);
209 #ifdef CONFIG_FSNOTIFY
210 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
217 kfree(mnt->mnt_devname);
222 kmem_cache_free(mnt_cache, mnt);
227 * Most r/o checks on a fs are for operations that take
228 * discrete amounts of time, like a write() or unlink().
229 * We must keep track of when those operations start
230 * (for permission checks) and when they end, so that
231 * we can determine when writes are able to occur to
235 * __mnt_is_readonly: check whether a mount is read-only
236 * @mnt: the mount to check for its write status
238 * This shouldn't be used directly ouside of the VFS.
239 * It does not guarantee that the filesystem will stay
240 * r/w, just that it is right *now*. This can not and
241 * should not be used in place of IS_RDONLY(inode).
242 * mnt_want/drop_write() will _keep_ the filesystem
245 int __mnt_is_readonly(struct vfsmount *mnt)
247 if (mnt->mnt_flags & MNT_READONLY)
249 if (mnt->mnt_sb->s_flags & MS_RDONLY)
253 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
255 static inline void mnt_inc_writers(struct vfsmount *mnt)
258 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
264 static inline void mnt_dec_writers(struct vfsmount *mnt)
267 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
273 static unsigned int mnt_get_writers(struct vfsmount *mnt)
276 unsigned int count = 0;
279 for_each_possible_cpu(cpu) {
280 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
285 return mnt->mnt_writers;
290 * Most r/o checks on a fs are for operations that take
291 * discrete amounts of time, like a write() or unlink().
292 * We must keep track of when those operations start
293 * (for permission checks) and when they end, so that
294 * we can determine when writes are able to occur to
298 * mnt_want_write - get write access to a mount
299 * @mnt: the mount on which to take a write
301 * This tells the low-level filesystem that a write is
302 * about to be performed to it, and makes sure that
303 * writes are allowed before returning success. When
304 * the write operation is finished, mnt_drop_write()
305 * must be called. This is effectively a refcount.
307 int mnt_want_write(struct vfsmount *mnt)
312 mnt_inc_writers(mnt);
314 * The store to mnt_inc_writers must be visible before we pass
315 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
316 * incremented count after it has set MNT_WRITE_HOLD.
319 while (mnt->mnt_flags & MNT_WRITE_HOLD)
322 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
323 * be set to match its requirements. So we must not load that until
324 * MNT_WRITE_HOLD is cleared.
327 if (__mnt_is_readonly(mnt)) {
328 mnt_dec_writers(mnt);
336 EXPORT_SYMBOL_GPL(mnt_want_write);
339 * mnt_clone_write - get write access to a mount
340 * @mnt: the mount on which to take a write
342 * This is effectively like mnt_want_write, except
343 * it must only be used to take an extra write reference
344 * on a mountpoint that we already know has a write reference
345 * on it. This allows some optimisation.
347 * After finished, mnt_drop_write must be called as usual to
348 * drop the reference.
350 int mnt_clone_write(struct vfsmount *mnt)
352 /* superblock may be r/o */
353 if (__mnt_is_readonly(mnt))
356 mnt_inc_writers(mnt);
360 EXPORT_SYMBOL_GPL(mnt_clone_write);
363 * mnt_want_write_file - get write access to a file's mount
364 * @file: the file who's mount on which to take a write
366 * This is like mnt_want_write, but it takes a file and can
367 * do some optimisations if the file is open for write already
369 int mnt_want_write_file(struct file *file)
371 struct inode *inode = file->f_dentry->d_inode;
372 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
373 return mnt_want_write(file->f_path.mnt);
375 return mnt_clone_write(file->f_path.mnt);
377 EXPORT_SYMBOL_GPL(mnt_want_write_file);
380 * mnt_drop_write - give up write access to a mount
381 * @mnt: the mount on which to give up write access
383 * Tells the low-level filesystem that we are done
384 * performing writes to it. Must be matched with
385 * mnt_want_write() call above.
387 void mnt_drop_write(struct vfsmount *mnt)
390 mnt_dec_writers(mnt);
393 EXPORT_SYMBOL_GPL(mnt_drop_write);
395 static int mnt_make_readonly(struct vfsmount *mnt)
399 br_write_lock(vfsmount_lock);
400 mnt->mnt_flags |= MNT_WRITE_HOLD;
402 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
403 * should be visible before we do.
408 * With writers on hold, if this value is zero, then there are
409 * definitely no active writers (although held writers may subsequently
410 * increment the count, they'll have to wait, and decrement it after
411 * seeing MNT_READONLY).
413 * It is OK to have counter incremented on one CPU and decremented on
414 * another: the sum will add up correctly. The danger would be when we
415 * sum up each counter, if we read a counter before it is incremented,
416 * but then read another CPU's count which it has been subsequently
417 * decremented from -- we would see more decrements than we should.
418 * MNT_WRITE_HOLD protects against this scenario, because
419 * mnt_want_write first increments count, then smp_mb, then spins on
420 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
421 * we're counting up here.
423 if (mnt_get_writers(mnt) > 0)
426 mnt->mnt_flags |= MNT_READONLY;
428 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
429 * that become unheld will see MNT_READONLY.
432 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
433 br_write_unlock(vfsmount_lock);
437 static void __mnt_unmake_readonly(struct vfsmount *mnt)
439 br_write_lock(vfsmount_lock);
440 mnt->mnt_flags &= ~MNT_READONLY;
441 br_write_unlock(vfsmount_lock);
444 static void free_vfsmnt(struct vfsmount *mnt)
446 kfree(mnt->mnt_devname);
449 free_percpu(mnt->mnt_pcp);
451 kmem_cache_free(mnt_cache, mnt);
455 * find the first or last mount at @dentry on vfsmount @mnt depending on
456 * @dir. If @dir is set return the first mount else return the last mount.
457 * vfsmount_lock must be held for read or write.
459 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
462 struct list_head *head = mount_hashtable + hash(mnt, dentry);
463 struct list_head *tmp = head;
464 struct vfsmount *p, *found = NULL;
467 tmp = dir ? tmp->next : tmp->prev;
471 p = list_entry(tmp, struct vfsmount, mnt_hash);
472 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
481 * lookup_mnt increments the ref count before returning
482 * the vfsmount struct.
484 struct vfsmount *lookup_mnt(struct path *path)
486 struct vfsmount *child_mnt;
488 br_read_lock(vfsmount_lock);
489 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
491 br_read_unlock(vfsmount_lock);
495 static inline int check_mnt(struct vfsmount *mnt)
497 return mnt->mnt_ns == current->nsproxy->mnt_ns;
501 * vfsmount lock must be held for write
503 static void touch_mnt_namespace(struct mnt_namespace *ns)
507 wake_up_interruptible(&ns->poll);
512 * vfsmount lock must be held for write
514 static void __touch_mnt_namespace(struct mnt_namespace *ns)
516 if (ns && ns->event != event) {
518 wake_up_interruptible(&ns->poll);
523 * Clear dentry's mounted state if it has no remaining mounts.
524 * vfsmount_lock must be held for write.
526 static void dentry_reset_mounted(struct dentry *dentry)
530 for (u = 0; u < HASH_SIZE; u++) {
533 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
534 if (p->mnt_mountpoint == dentry)
538 spin_lock(&dentry->d_lock);
539 dentry->d_flags &= ~DCACHE_MOUNTED;
540 spin_unlock(&dentry->d_lock);
544 * vfsmount lock must be held for write
546 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
548 old_path->dentry = mnt->mnt_mountpoint;
549 old_path->mnt = mnt->mnt_parent;
550 mnt->mnt_parent = mnt;
551 mnt->mnt_mountpoint = mnt->mnt_root;
552 list_del_init(&mnt->mnt_child);
553 list_del_init(&mnt->mnt_hash);
554 dentry_reset_mounted(old_path->dentry);
558 * vfsmount lock must be held for write
560 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
561 struct vfsmount *child_mnt)
563 child_mnt->mnt_parent = mntget(mnt);
564 child_mnt->mnt_mountpoint = dget(dentry);
565 spin_lock(&dentry->d_lock);
566 dentry->d_flags |= DCACHE_MOUNTED;
567 spin_unlock(&dentry->d_lock);
571 * vfsmount lock must be held for write
573 static void attach_mnt(struct vfsmount *mnt, struct path *path)
575 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
576 list_add_tail(&mnt->mnt_hash, mount_hashtable +
577 hash(path->mnt, path->dentry));
578 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
581 static inline void __mnt_make_longterm(struct vfsmount *mnt)
584 atomic_inc(&mnt->mnt_longterm);
588 /* needs vfsmount lock for write */
589 static inline void __mnt_make_shortterm(struct vfsmount *mnt)
592 atomic_dec(&mnt->mnt_longterm);
597 * vfsmount lock must be held for write
599 static void commit_tree(struct vfsmount *mnt)
601 struct vfsmount *parent = mnt->mnt_parent;
604 struct mnt_namespace *n = parent->mnt_ns;
606 BUG_ON(parent == mnt);
608 list_add_tail(&head, &mnt->mnt_list);
609 list_for_each_entry(m, &head, mnt_list) {
611 __mnt_make_longterm(m);
614 list_splice(&head, n->list.prev);
616 list_add_tail(&mnt->mnt_hash, mount_hashtable +
617 hash(parent, mnt->mnt_mountpoint));
618 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
619 touch_mnt_namespace(n);
622 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
624 struct list_head *next = p->mnt_mounts.next;
625 if (next == &p->mnt_mounts) {
629 next = p->mnt_child.next;
630 if (next != &p->mnt_parent->mnt_mounts)
635 return list_entry(next, struct vfsmount, mnt_child);
638 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
640 struct list_head *prev = p->mnt_mounts.prev;
641 while (prev != &p->mnt_mounts) {
642 p = list_entry(prev, struct vfsmount, mnt_child);
643 prev = p->mnt_mounts.prev;
649 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
651 struct vfsmount *mnt;
655 return ERR_PTR(-ENODEV);
657 mnt = alloc_vfsmnt(name);
659 return ERR_PTR(-ENOMEM);
661 if (flags & MS_KERNMOUNT)
662 mnt->mnt_flags = MNT_INTERNAL;
664 root = mount_fs(type, flags, name, data);
667 return ERR_CAST(root);
670 mnt->mnt_root = root;
671 mnt->mnt_sb = root->d_sb;
672 mnt->mnt_mountpoint = mnt->mnt_root;
673 mnt->mnt_parent = mnt;
676 EXPORT_SYMBOL_GPL(vfs_kern_mount);
678 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
681 struct super_block *sb = old->mnt_sb;
682 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
685 if (flag & (CL_SLAVE | CL_PRIVATE))
686 mnt->mnt_group_id = 0; /* not a peer of original */
688 mnt->mnt_group_id = old->mnt_group_id;
690 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
691 int err = mnt_alloc_group_id(mnt);
696 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD;
697 atomic_inc(&sb->s_active);
699 mnt->mnt_root = dget(root);
700 mnt->mnt_mountpoint = mnt->mnt_root;
701 mnt->mnt_parent = mnt;
703 if (flag & CL_SLAVE) {
704 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
705 mnt->mnt_master = old;
706 CLEAR_MNT_SHARED(mnt);
707 } else if (!(flag & CL_PRIVATE)) {
708 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
709 list_add(&mnt->mnt_share, &old->mnt_share);
710 if (IS_MNT_SLAVE(old))
711 list_add(&mnt->mnt_slave, &old->mnt_slave);
712 mnt->mnt_master = old->mnt_master;
714 if (flag & CL_MAKE_SHARED)
717 /* stick the duplicate mount on the same expiry list
718 * as the original if that was on one */
719 if (flag & CL_EXPIRE) {
720 if (!list_empty(&old->mnt_expire))
721 list_add(&mnt->mnt_expire, &old->mnt_expire);
731 static inline void mntfree(struct vfsmount *mnt)
733 struct super_block *sb = mnt->mnt_sb;
736 * This probably indicates that somebody messed
737 * up a mnt_want/drop_write() pair. If this
738 * happens, the filesystem was probably unable
739 * to make r/w->r/o transitions.
742 * The locking used to deal with mnt_count decrement provides barriers,
743 * so mnt_get_writers() below is safe.
745 WARN_ON(mnt_get_writers(mnt));
746 fsnotify_vfsmount_delete(mnt);
749 deactivate_super(sb);
752 static void mntput_no_expire(struct vfsmount *mnt)
756 br_read_lock(vfsmount_lock);
757 if (likely(atomic_read(&mnt->mnt_longterm))) {
758 mnt_add_count(mnt, -1);
759 br_read_unlock(vfsmount_lock);
762 br_read_unlock(vfsmount_lock);
764 br_write_lock(vfsmount_lock);
765 mnt_add_count(mnt, -1);
766 if (mnt_get_count(mnt)) {
767 br_write_unlock(vfsmount_lock);
771 mnt_add_count(mnt, -1);
772 if (likely(mnt_get_count(mnt)))
774 br_write_lock(vfsmount_lock);
776 if (unlikely(mnt->mnt_pinned)) {
777 mnt_add_count(mnt, mnt->mnt_pinned + 1);
779 br_write_unlock(vfsmount_lock);
780 acct_auto_close_mnt(mnt);
783 br_write_unlock(vfsmount_lock);
787 void mntput(struct vfsmount *mnt)
790 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
791 if (unlikely(mnt->mnt_expiry_mark))
792 mnt->mnt_expiry_mark = 0;
793 mntput_no_expire(mnt);
796 EXPORT_SYMBOL(mntput);
798 struct vfsmount *mntget(struct vfsmount *mnt)
801 mnt_add_count(mnt, 1);
804 EXPORT_SYMBOL(mntget);
806 void mnt_pin(struct vfsmount *mnt)
808 br_write_lock(vfsmount_lock);
810 br_write_unlock(vfsmount_lock);
812 EXPORT_SYMBOL(mnt_pin);
814 void mnt_unpin(struct vfsmount *mnt)
816 br_write_lock(vfsmount_lock);
817 if (mnt->mnt_pinned) {
818 mnt_add_count(mnt, 1);
821 br_write_unlock(vfsmount_lock);
823 EXPORT_SYMBOL(mnt_unpin);
825 static inline void mangle(struct seq_file *m, const char *s)
827 seq_escape(m, s, " \t\n\\");
831 * Simple .show_options callback for filesystems which don't want to
832 * implement more complex mount option showing.
834 * See also save_mount_options().
836 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
841 options = rcu_dereference(mnt->mnt_sb->s_options);
843 if (options != NULL && options[0]) {
851 EXPORT_SYMBOL(generic_show_options);
854 * If filesystem uses generic_show_options(), this function should be
855 * called from the fill_super() callback.
857 * The .remount_fs callback usually needs to be handled in a special
858 * way, to make sure, that previous options are not overwritten if the
861 * Also note, that if the filesystem's .remount_fs function doesn't
862 * reset all options to their default value, but changes only newly
863 * given options, then the displayed options will not reflect reality
866 void save_mount_options(struct super_block *sb, char *options)
868 BUG_ON(sb->s_options);
869 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
871 EXPORT_SYMBOL(save_mount_options);
873 void replace_mount_options(struct super_block *sb, char *options)
875 char *old = sb->s_options;
876 rcu_assign_pointer(sb->s_options, options);
882 EXPORT_SYMBOL(replace_mount_options);
884 #ifdef CONFIG_PROC_FS
886 static void *m_start(struct seq_file *m, loff_t *pos)
888 struct proc_mounts *p = m->private;
890 down_read(&namespace_sem);
891 return seq_list_start(&p->ns->list, *pos);
894 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
896 struct proc_mounts *p = m->private;
898 return seq_list_next(v, &p->ns->list, pos);
901 static void m_stop(struct seq_file *m, void *v)
903 up_read(&namespace_sem);
906 int mnt_had_events(struct proc_mounts *p)
908 struct mnt_namespace *ns = p->ns;
911 br_read_lock(vfsmount_lock);
912 if (p->m.poll_event != ns->event) {
913 p->m.poll_event = ns->event;
916 br_read_unlock(vfsmount_lock);
921 struct proc_fs_info {
926 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
928 static const struct proc_fs_info fs_info[] = {
929 { MS_SYNCHRONOUS, ",sync" },
930 { MS_DIRSYNC, ",dirsync" },
931 { MS_MANDLOCK, ",mand" },
934 const struct proc_fs_info *fs_infop;
936 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
937 if (sb->s_flags & fs_infop->flag)
938 seq_puts(m, fs_infop->str);
941 return security_sb_show_options(m, sb);
944 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
946 static const struct proc_fs_info mnt_info[] = {
947 { MNT_NOSUID, ",nosuid" },
948 { MNT_NODEV, ",nodev" },
949 { MNT_NOEXEC, ",noexec" },
950 { MNT_NOATIME, ",noatime" },
951 { MNT_NODIRATIME, ",nodiratime" },
952 { MNT_RELATIME, ",relatime" },
955 const struct proc_fs_info *fs_infop;
957 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
958 if (mnt->mnt_flags & fs_infop->flag)
959 seq_puts(m, fs_infop->str);
963 static void show_type(struct seq_file *m, struct super_block *sb)
965 mangle(m, sb->s_type->name);
966 if (sb->s_subtype && sb->s_subtype[0]) {
968 mangle(m, sb->s_subtype);
972 static int show_vfsmnt(struct seq_file *m, void *v)
974 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
976 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
978 if (mnt->mnt_sb->s_op->show_devname) {
979 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
983 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
986 seq_path(m, &mnt_path, " \t\n\\");
988 show_type(m, mnt->mnt_sb);
989 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
990 err = show_sb_opts(m, mnt->mnt_sb);
993 show_mnt_opts(m, mnt);
994 if (mnt->mnt_sb->s_op->show_options)
995 err = mnt->mnt_sb->s_op->show_options(m, mnt);
996 seq_puts(m, " 0 0\n");
1001 const struct seq_operations mounts_op = {
1008 static int show_mountinfo(struct seq_file *m, void *v)
1010 struct proc_mounts *p = m->private;
1011 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1012 struct super_block *sb = mnt->mnt_sb;
1013 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1014 struct path root = p->root;
1017 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
1018 MAJOR(sb->s_dev), MINOR(sb->s_dev));
1019 if (sb->s_op->show_path)
1020 err = sb->s_op->show_path(m, mnt);
1022 seq_dentry(m, mnt->mnt_root, " \t\n\\");
1027 /* mountpoints outside of chroot jail will give SEQ_SKIP on this */
1028 err = seq_path_root(m, &mnt_path, &root, " \t\n\\");
1032 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
1033 show_mnt_opts(m, mnt);
1035 /* Tagged fields ("foo:X" or "bar") */
1036 if (IS_MNT_SHARED(mnt))
1037 seq_printf(m, " shared:%i", mnt->mnt_group_id);
1038 if (IS_MNT_SLAVE(mnt)) {
1039 int master = mnt->mnt_master->mnt_group_id;
1040 int dom = get_dominating_id(mnt, &p->root);
1041 seq_printf(m, " master:%i", master);
1042 if (dom && dom != master)
1043 seq_printf(m, " propagate_from:%i", dom);
1045 if (IS_MNT_UNBINDABLE(mnt))
1046 seq_puts(m, " unbindable");
1048 /* Filesystem specific data */
1052 if (sb->s_op->show_devname)
1053 err = sb->s_op->show_devname(m, mnt);
1055 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
1058 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
1059 err = show_sb_opts(m, sb);
1062 if (sb->s_op->show_options)
1063 err = sb->s_op->show_options(m, mnt);
1069 const struct seq_operations mountinfo_op = {
1073 .show = show_mountinfo,
1076 static int show_vfsstat(struct seq_file *m, void *v)
1078 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
1079 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
1083 if (mnt->mnt_sb->s_op->show_devname) {
1084 seq_puts(m, "device ");
1085 err = mnt->mnt_sb->s_op->show_devname(m, mnt);
1087 if (mnt->mnt_devname) {
1088 seq_puts(m, "device ");
1089 mangle(m, mnt->mnt_devname);
1091 seq_puts(m, "no device");
1095 seq_puts(m, " mounted on ");
1096 seq_path(m, &mnt_path, " \t\n\\");
1099 /* file system type */
1100 seq_puts(m, "with fstype ");
1101 show_type(m, mnt->mnt_sb);
1103 /* optional statistics */
1104 if (mnt->mnt_sb->s_op->show_stats) {
1107 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
1114 const struct seq_operations mountstats_op = {
1118 .show = show_vfsstat,
1120 #endif /* CONFIG_PROC_FS */
1123 * may_umount_tree - check if a mount tree is busy
1124 * @mnt: root of mount tree
1126 * This is called to check if a tree of mounts has any
1127 * open files, pwds, chroots or sub mounts that are
1130 int may_umount_tree(struct vfsmount *mnt)
1132 int actual_refs = 0;
1133 int minimum_refs = 0;
1136 /* write lock needed for mnt_get_count */
1137 br_write_lock(vfsmount_lock);
1138 for (p = mnt; p; p = next_mnt(p, mnt)) {
1139 actual_refs += mnt_get_count(p);
1142 br_write_unlock(vfsmount_lock);
1144 if (actual_refs > minimum_refs)
1150 EXPORT_SYMBOL(may_umount_tree);
1153 * may_umount - check if a mount point is busy
1154 * @mnt: root of mount
1156 * This is called to check if a mount point has any
1157 * open files, pwds, chroots or sub mounts. If the
1158 * mount has sub mounts this will return busy
1159 * regardless of whether the sub mounts are busy.
1161 * Doesn't take quota and stuff into account. IOW, in some cases it will
1162 * give false negatives. The main reason why it's here is that we need
1163 * a non-destructive way to look for easily umountable filesystems.
1165 int may_umount(struct vfsmount *mnt)
1168 down_read(&namespace_sem);
1169 br_write_lock(vfsmount_lock);
1170 if (propagate_mount_busy(mnt, 2))
1172 br_write_unlock(vfsmount_lock);
1173 up_read(&namespace_sem);
1177 EXPORT_SYMBOL(may_umount);
1179 void release_mounts(struct list_head *head)
1181 struct vfsmount *mnt;
1182 while (!list_empty(head)) {
1183 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1184 list_del_init(&mnt->mnt_hash);
1185 if (mnt_has_parent(mnt)) {
1186 struct dentry *dentry;
1189 br_write_lock(vfsmount_lock);
1190 dentry = mnt->mnt_mountpoint;
1191 m = mnt->mnt_parent;
1192 mnt->mnt_mountpoint = mnt->mnt_root;
1193 mnt->mnt_parent = mnt;
1195 br_write_unlock(vfsmount_lock);
1204 * vfsmount lock must be held for write
1205 * namespace_sem must be held for write
1207 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1209 LIST_HEAD(tmp_list);
1212 for (p = mnt; p; p = next_mnt(p, mnt))
1213 list_move(&p->mnt_hash, &tmp_list);
1216 propagate_umount(&tmp_list);
1218 list_for_each_entry(p, &tmp_list, mnt_hash) {
1219 list_del_init(&p->mnt_expire);
1220 list_del_init(&p->mnt_list);
1221 __touch_mnt_namespace(p->mnt_ns);
1223 __mnt_make_shortterm(p);
1224 list_del_init(&p->mnt_child);
1225 if (mnt_has_parent(p)) {
1226 p->mnt_parent->mnt_ghosts++;
1227 dentry_reset_mounted(p->mnt_mountpoint);
1229 change_mnt_propagation(p, MS_PRIVATE);
1231 list_splice(&tmp_list, kill);
1234 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1236 static int do_umount(struct vfsmount *mnt, int flags)
1238 struct super_block *sb = mnt->mnt_sb;
1240 LIST_HEAD(umount_list);
1242 retval = security_sb_umount(mnt, flags);
1247 * Allow userspace to request a mountpoint be expired rather than
1248 * unmounting unconditionally. Unmount only happens if:
1249 * (1) the mark is already set (the mark is cleared by mntput())
1250 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1252 if (flags & MNT_EXPIRE) {
1253 if (mnt == current->fs->root.mnt ||
1254 flags & (MNT_FORCE | MNT_DETACH))
1258 * probably don't strictly need the lock here if we examined
1259 * all race cases, but it's a slowpath.
1261 br_write_lock(vfsmount_lock);
1262 if (mnt_get_count(mnt) != 2) {
1263 br_write_unlock(vfsmount_lock);
1266 br_write_unlock(vfsmount_lock);
1268 if (!xchg(&mnt->mnt_expiry_mark, 1))
1273 * If we may have to abort operations to get out of this
1274 * mount, and they will themselves hold resources we must
1275 * allow the fs to do things. In the Unix tradition of
1276 * 'Gee thats tricky lets do it in userspace' the umount_begin
1277 * might fail to complete on the first run through as other tasks
1278 * must return, and the like. Thats for the mount program to worry
1279 * about for the moment.
1282 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1283 sb->s_op->umount_begin(sb);
1287 * No sense to grab the lock for this test, but test itself looks
1288 * somewhat bogus. Suggestions for better replacement?
1289 * Ho-hum... In principle, we might treat that as umount + switch
1290 * to rootfs. GC would eventually take care of the old vfsmount.
1291 * Actually it makes sense, especially if rootfs would contain a
1292 * /reboot - static binary that would close all descriptors and
1293 * call reboot(9). Then init(8) could umount root and exec /reboot.
1295 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1297 * Special case for "unmounting" root ...
1298 * we just try to remount it readonly.
1300 down_write(&sb->s_umount);
1301 if (!(sb->s_flags & MS_RDONLY))
1302 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1303 up_write(&sb->s_umount);
1307 down_write(&namespace_sem);
1308 br_write_lock(vfsmount_lock);
1311 if (!(flags & MNT_DETACH))
1312 shrink_submounts(mnt, &umount_list);
1315 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1316 if (!list_empty(&mnt->mnt_list))
1317 umount_tree(mnt, 1, &umount_list);
1320 br_write_unlock(vfsmount_lock);
1321 up_write(&namespace_sem);
1322 release_mounts(&umount_list);
1327 * Now umount can handle mount points as well as block devices.
1328 * This is important for filesystems which use unnamed block devices.
1330 * We now support a flag for forced unmount like the other 'big iron'
1331 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1334 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1338 int lookup_flags = 0;
1340 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1343 if (!(flags & UMOUNT_NOFOLLOW))
1344 lookup_flags |= LOOKUP_FOLLOW;
1346 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1350 if (path.dentry != path.mnt->mnt_root)
1352 if (!check_mnt(path.mnt))
1356 if (!capable(CAP_SYS_ADMIN))
1359 retval = do_umount(path.mnt, flags);
1361 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1363 mntput_no_expire(path.mnt);
1368 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1371 * The 2.0 compatible umount. No flags.
1373 SYSCALL_DEFINE1(oldumount, char __user *, name)
1375 return sys_umount(name, 0);
1380 static int mount_is_safe(struct path *path)
1382 if (capable(CAP_SYS_ADMIN))
1386 if (S_ISLNK(path->dentry->d_inode->i_mode))
1388 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1389 if (current_uid() != path->dentry->d_inode->i_uid)
1392 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1398 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1401 struct vfsmount *res, *p, *q, *r, *s;
1404 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1407 res = q = clone_mnt(mnt, dentry, flag);
1410 q->mnt_mountpoint = mnt->mnt_mountpoint;
1413 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1414 if (!is_subdir(r->mnt_mountpoint, dentry))
1417 for (s = r; s; s = next_mnt(s, r)) {
1418 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1419 s = skip_mnt_tree(s);
1422 while (p != s->mnt_parent) {
1428 path.dentry = p->mnt_mountpoint;
1429 q = clone_mnt(p, p->mnt_root, flag);
1432 br_write_lock(vfsmount_lock);
1433 list_add_tail(&q->mnt_list, &res->mnt_list);
1434 attach_mnt(q, &path);
1435 br_write_unlock(vfsmount_lock);
1441 LIST_HEAD(umount_list);
1442 br_write_lock(vfsmount_lock);
1443 umount_tree(res, 0, &umount_list);
1444 br_write_unlock(vfsmount_lock);
1445 release_mounts(&umount_list);
1450 struct vfsmount *collect_mounts(struct path *path)
1452 struct vfsmount *tree;
1453 down_write(&namespace_sem);
1454 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1455 up_write(&namespace_sem);
1459 void drop_collected_mounts(struct vfsmount *mnt)
1461 LIST_HEAD(umount_list);
1462 down_write(&namespace_sem);
1463 br_write_lock(vfsmount_lock);
1464 umount_tree(mnt, 0, &umount_list);
1465 br_write_unlock(vfsmount_lock);
1466 up_write(&namespace_sem);
1467 release_mounts(&umount_list);
1470 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1471 struct vfsmount *root)
1473 struct vfsmount *mnt;
1474 int res = f(root, arg);
1477 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1485 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1489 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1490 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1491 mnt_release_group_id(p);
1495 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1499 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1500 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1501 int err = mnt_alloc_group_id(p);
1503 cleanup_group_ids(mnt, p);
1513 * @source_mnt : mount tree to be attached
1514 * @nd : place the mount tree @source_mnt is attached
1515 * @parent_nd : if non-null, detach the source_mnt from its parent and
1516 * store the parent mount and mountpoint dentry.
1517 * (done when source_mnt is moved)
1519 * NOTE: in the table below explains the semantics when a source mount
1520 * of a given type is attached to a destination mount of a given type.
1521 * ---------------------------------------------------------------------------
1522 * | BIND MOUNT OPERATION |
1523 * |**************************************************************************
1524 * | source-->| shared | private | slave | unbindable |
1528 * |**************************************************************************
1529 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1531 * |non-shared| shared (+) | private | slave (*) | invalid |
1532 * ***************************************************************************
1533 * A bind operation clones the source mount and mounts the clone on the
1534 * destination mount.
1536 * (++) the cloned mount is propagated to all the mounts in the propagation
1537 * tree of the destination mount and the cloned mount is added to
1538 * the peer group of the source mount.
1539 * (+) the cloned mount is created under the destination mount and is marked
1540 * as shared. The cloned mount is added to the peer group of the source
1542 * (+++) the mount is propagated to all the mounts in the propagation tree
1543 * of the destination mount and the cloned mount is made slave
1544 * of the same master as that of the source mount. The cloned mount
1545 * is marked as 'shared and slave'.
1546 * (*) the cloned mount is made a slave of the same master as that of the
1549 * ---------------------------------------------------------------------------
1550 * | MOVE MOUNT OPERATION |
1551 * |**************************************************************************
1552 * | source-->| shared | private | slave | unbindable |
1556 * |**************************************************************************
1557 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1559 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1560 * ***************************************************************************
1562 * (+) the mount is moved to the destination. And is then propagated to
1563 * all the mounts in the propagation tree of the destination mount.
1564 * (+*) the mount is moved to the destination.
1565 * (+++) the mount is moved to the destination and is then propagated to
1566 * all the mounts belonging to the destination mount's propagation tree.
1567 * the mount is marked as 'shared and slave'.
1568 * (*) the mount continues to be a slave at the new location.
1570 * if the source mount is a tree, the operations explained above is
1571 * applied to each mount in the tree.
1572 * Must be called without spinlocks held, since this function can sleep
1575 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1576 struct path *path, struct path *parent_path)
1578 LIST_HEAD(tree_list);
1579 struct vfsmount *dest_mnt = path->mnt;
1580 struct dentry *dest_dentry = path->dentry;
1581 struct vfsmount *child, *p;
1584 if (IS_MNT_SHARED(dest_mnt)) {
1585 err = invent_group_ids(source_mnt, true);
1589 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1591 goto out_cleanup_ids;
1593 br_write_lock(vfsmount_lock);
1595 if (IS_MNT_SHARED(dest_mnt)) {
1596 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1600 detach_mnt(source_mnt, parent_path);
1601 attach_mnt(source_mnt, path);
1602 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1604 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1605 commit_tree(source_mnt);
1608 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1609 list_del_init(&child->mnt_hash);
1612 br_write_unlock(vfsmount_lock);
1617 if (IS_MNT_SHARED(dest_mnt))
1618 cleanup_group_ids(source_mnt, NULL);
1623 static int lock_mount(struct path *path)
1625 struct vfsmount *mnt;
1627 mutex_lock(&path->dentry->d_inode->i_mutex);
1628 if (unlikely(cant_mount(path->dentry))) {
1629 mutex_unlock(&path->dentry->d_inode->i_mutex);
1632 down_write(&namespace_sem);
1633 mnt = lookup_mnt(path);
1636 up_write(&namespace_sem);
1637 mutex_unlock(&path->dentry->d_inode->i_mutex);
1640 path->dentry = dget(mnt->mnt_root);
1644 static void unlock_mount(struct path *path)
1646 up_write(&namespace_sem);
1647 mutex_unlock(&path->dentry->d_inode->i_mutex);
1650 static int graft_tree(struct vfsmount *mnt, struct path *path)
1652 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1655 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1656 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1659 if (d_unlinked(path->dentry))
1662 return attach_recursive_mnt(mnt, path, NULL);
1666 * Sanity check the flags to change_mnt_propagation.
1669 static int flags_to_propagation_type(int flags)
1671 int type = flags & ~(MS_REC | MS_SILENT);
1673 /* Fail if any non-propagation flags are set */
1674 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1676 /* Only one propagation flag should be set */
1677 if (!is_power_of_2(type))
1683 * recursively change the type of the mountpoint.
1685 static int do_change_type(struct path *path, int flag)
1687 struct vfsmount *m, *mnt = path->mnt;
1688 int recurse = flag & MS_REC;
1692 if (!capable(CAP_SYS_ADMIN))
1695 if (path->dentry != path->mnt->mnt_root)
1698 type = flags_to_propagation_type(flag);
1702 down_write(&namespace_sem);
1703 if (type == MS_SHARED) {
1704 err = invent_group_ids(mnt, recurse);
1709 br_write_lock(vfsmount_lock);
1710 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1711 change_mnt_propagation(m, type);
1712 br_write_unlock(vfsmount_lock);
1715 up_write(&namespace_sem);
1720 * do loopback mount.
1722 static int do_loopback(struct path *path, char *old_name,
1725 LIST_HEAD(umount_list);
1726 struct path old_path;
1727 struct vfsmount *mnt = NULL;
1728 int err = mount_is_safe(path);
1731 if (!old_name || !*old_name)
1733 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1737 err = lock_mount(path);
1742 if (IS_MNT_UNBINDABLE(old_path.mnt))
1745 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1750 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1752 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1757 err = graft_tree(mnt, path);
1759 br_write_lock(vfsmount_lock);
1760 umount_tree(mnt, 0, &umount_list);
1761 br_write_unlock(vfsmount_lock);
1765 release_mounts(&umount_list);
1767 path_put(&old_path);
1771 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1774 int readonly_request = 0;
1776 if (ms_flags & MS_RDONLY)
1777 readonly_request = 1;
1778 if (readonly_request == __mnt_is_readonly(mnt))
1781 if (readonly_request)
1782 error = mnt_make_readonly(mnt);
1784 __mnt_unmake_readonly(mnt);
1789 * change filesystem flags. dir should be a physical root of filesystem.
1790 * If you've mounted a non-root directory somewhere and want to do remount
1791 * on it - tough luck.
1793 static int do_remount(struct path *path, int flags, int mnt_flags,
1797 struct super_block *sb = path->mnt->mnt_sb;
1799 if (!capable(CAP_SYS_ADMIN))
1802 if (!check_mnt(path->mnt))
1805 if (path->dentry != path->mnt->mnt_root)
1808 err = security_sb_remount(sb, data);
1812 down_write(&sb->s_umount);
1813 if (flags & MS_BIND)
1814 err = change_mount_flags(path->mnt, flags);
1816 err = do_remount_sb(sb, flags, data, 0);
1818 br_write_lock(vfsmount_lock);
1819 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
1820 path->mnt->mnt_flags = mnt_flags;
1821 br_write_unlock(vfsmount_lock);
1823 up_write(&sb->s_umount);
1825 br_write_lock(vfsmount_lock);
1826 touch_mnt_namespace(path->mnt->mnt_ns);
1827 br_write_unlock(vfsmount_lock);
1832 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1835 for (p = mnt; p; p = next_mnt(p, mnt)) {
1836 if (IS_MNT_UNBINDABLE(p))
1842 static int do_move_mount(struct path *path, char *old_name)
1844 struct path old_path, parent_path;
1847 if (!capable(CAP_SYS_ADMIN))
1849 if (!old_name || !*old_name)
1851 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1855 err = lock_mount(path);
1860 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1863 if (d_unlinked(path->dentry))
1867 if (old_path.dentry != old_path.mnt->mnt_root)
1870 if (!mnt_has_parent(old_path.mnt))
1873 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1874 S_ISDIR(old_path.dentry->d_inode->i_mode))
1877 * Don't move a mount residing in a shared parent.
1879 if (IS_MNT_SHARED(old_path.mnt->mnt_parent))
1882 * Don't move a mount tree containing unbindable mounts to a destination
1883 * mount which is shared.
1885 if (IS_MNT_SHARED(path->mnt) &&
1886 tree_contains_unbindable(old_path.mnt))
1889 for (p = path->mnt; mnt_has_parent(p); p = p->mnt_parent)
1890 if (p == old_path.mnt)
1893 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1897 /* if the mount is moved, it should no longer be expire
1899 list_del_init(&old_path.mnt->mnt_expire);
1904 path_put(&parent_path);
1905 path_put(&old_path);
1909 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1912 const char *subtype = strchr(fstype, '.');
1921 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1923 if (!mnt->mnt_sb->s_subtype)
1929 return ERR_PTR(err);
1933 do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1935 struct file_system_type *type = get_fs_type(fstype);
1936 struct vfsmount *mnt;
1938 return ERR_PTR(-ENODEV);
1939 mnt = vfs_kern_mount(type, flags, name, data);
1940 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1941 !mnt->mnt_sb->s_subtype)
1942 mnt = fs_set_subtype(mnt, fstype);
1943 put_filesystem(type);
1946 EXPORT_SYMBOL_GPL(do_kern_mount);
1949 * add a mount into a namespace's mount tree
1951 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags)
1955 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1957 err = lock_mount(path);
1962 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1965 /* Refuse the same filesystem on the same mount point */
1967 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1968 path->mnt->mnt_root == path->dentry)
1972 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1975 newmnt->mnt_flags = mnt_flags;
1976 err = graft_tree(newmnt, path);
1984 * create a new mount for userspace and request it to be added into the
1987 static int do_new_mount(struct path *path, char *type, int flags,
1988 int mnt_flags, char *name, void *data)
1990 struct vfsmount *mnt;
1996 /* we need capabilities... */
1997 if (!capable(CAP_SYS_ADMIN))
2000 mnt = do_kern_mount(type, flags, name, data);
2002 return PTR_ERR(mnt);
2004 err = do_add_mount(mnt, path, mnt_flags);
2010 int finish_automount(struct vfsmount *m, struct path *path)
2013 /* The new mount record should have at least 2 refs to prevent it being
2014 * expired before we get a chance to add it
2016 BUG_ON(mnt_get_count(m) < 2);
2018 if (m->mnt_sb == path->mnt->mnt_sb &&
2019 m->mnt_root == path->dentry) {
2024 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2028 /* remove m from any expiration list it may be on */
2029 if (!list_empty(&m->mnt_expire)) {
2030 down_write(&namespace_sem);
2031 br_write_lock(vfsmount_lock);
2032 list_del_init(&m->mnt_expire);
2033 br_write_unlock(vfsmount_lock);
2034 up_write(&namespace_sem);
2042 * mnt_set_expiry - Put a mount on an expiration list
2043 * @mnt: The mount to list.
2044 * @expiry_list: The list to add the mount to.
2046 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2048 down_write(&namespace_sem);
2049 br_write_lock(vfsmount_lock);
2051 list_add_tail(&mnt->mnt_expire, expiry_list);
2053 br_write_unlock(vfsmount_lock);
2054 up_write(&namespace_sem);
2056 EXPORT_SYMBOL(mnt_set_expiry);
2059 * process a list of expirable mountpoints with the intent of discarding any
2060 * mountpoints that aren't in use and haven't been touched since last we came
2063 void mark_mounts_for_expiry(struct list_head *mounts)
2065 struct vfsmount *mnt, *next;
2066 LIST_HEAD(graveyard);
2069 if (list_empty(mounts))
2072 down_write(&namespace_sem);
2073 br_write_lock(vfsmount_lock);
2075 /* extract from the expiration list every vfsmount that matches the
2076 * following criteria:
2077 * - only referenced by its parent vfsmount
2078 * - still marked for expiry (marked on the last call here; marks are
2079 * cleared by mntput())
2081 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2082 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2083 propagate_mount_busy(mnt, 1))
2085 list_move(&mnt->mnt_expire, &graveyard);
2087 while (!list_empty(&graveyard)) {
2088 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
2089 touch_mnt_namespace(mnt->mnt_ns);
2090 umount_tree(mnt, 1, &umounts);
2092 br_write_unlock(vfsmount_lock);
2093 up_write(&namespace_sem);
2095 release_mounts(&umounts);
2098 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2101 * Ripoff of 'select_parent()'
2103 * search the list of submounts for a given mountpoint, and move any
2104 * shrinkable submounts to the 'graveyard' list.
2106 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
2108 struct vfsmount *this_parent = parent;
2109 struct list_head *next;
2113 next = this_parent->mnt_mounts.next;
2115 while (next != &this_parent->mnt_mounts) {
2116 struct list_head *tmp = next;
2117 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
2120 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
2123 * Descend a level if the d_mounts list is non-empty.
2125 if (!list_empty(&mnt->mnt_mounts)) {
2130 if (!propagate_mount_busy(mnt, 1)) {
2131 list_move_tail(&mnt->mnt_expire, graveyard);
2136 * All done at this level ... ascend and resume the search
2138 if (this_parent != parent) {
2139 next = this_parent->mnt_child.next;
2140 this_parent = this_parent->mnt_parent;
2147 * process a list of expirable mountpoints with the intent of discarding any
2148 * submounts of a specific parent mountpoint
2150 * vfsmount_lock must be held for write
2152 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
2154 LIST_HEAD(graveyard);
2157 /* extract submounts of 'mountpoint' from the expiration list */
2158 while (select_submounts(mnt, &graveyard)) {
2159 while (!list_empty(&graveyard)) {
2160 m = list_first_entry(&graveyard, struct vfsmount,
2162 touch_mnt_namespace(m->mnt_ns);
2163 umount_tree(m, 1, umounts);
2169 * Some copy_from_user() implementations do not return the exact number of
2170 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2171 * Note that this function differs from copy_from_user() in that it will oops
2172 * on bad values of `to', rather than returning a short copy.
2174 static long exact_copy_from_user(void *to, const void __user * from,
2178 const char __user *f = from;
2181 if (!access_ok(VERIFY_READ, from, n))
2185 if (__get_user(c, f)) {
2196 int copy_mount_options(const void __user * data, unsigned long *where)
2206 if (!(page = __get_free_page(GFP_KERNEL)))
2209 /* We only care that *some* data at the address the user
2210 * gave us is valid. Just in case, we'll zero
2211 * the remainder of the page.
2213 /* copy_from_user cannot cross TASK_SIZE ! */
2214 size = TASK_SIZE - (unsigned long)data;
2215 if (size > PAGE_SIZE)
2218 i = size - exact_copy_from_user((void *)page, data, size);
2224 memset((char *)page + i, 0, PAGE_SIZE - i);
2229 int copy_mount_string(const void __user *data, char **where)
2238 tmp = strndup_user(data, PAGE_SIZE);
2240 return PTR_ERR(tmp);
2247 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2248 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2250 * data is a (void *) that can point to any structure up to
2251 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2252 * information (or be NULL).
2254 * Pre-0.97 versions of mount() didn't have a flags word.
2255 * When the flags word was introduced its top half was required
2256 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2257 * Therefore, if this magic number is present, it carries no information
2258 * and must be discarded.
2260 long do_mount(char *dev_name, char *dir_name, char *type_page,
2261 unsigned long flags, void *data_page)
2268 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2269 flags &= ~MS_MGC_MSK;
2271 /* Basic sanity checks */
2273 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2277 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2279 /* ... and get the mountpoint */
2280 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2284 retval = security_sb_mount(dev_name, &path,
2285 type_page, flags, data_page);
2289 /* Default to relatime unless overriden */
2290 if (!(flags & MS_NOATIME))
2291 mnt_flags |= MNT_RELATIME;
2293 /* Separate the per-mountpoint flags */
2294 if (flags & MS_NOSUID)
2295 mnt_flags |= MNT_NOSUID;
2296 if (flags & MS_NODEV)
2297 mnt_flags |= MNT_NODEV;
2298 if (flags & MS_NOEXEC)
2299 mnt_flags |= MNT_NOEXEC;
2300 if (flags & MS_NOATIME)
2301 mnt_flags |= MNT_NOATIME;
2302 if (flags & MS_NODIRATIME)
2303 mnt_flags |= MNT_NODIRATIME;
2304 if (flags & MS_STRICTATIME)
2305 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2306 if (flags & MS_RDONLY)
2307 mnt_flags |= MNT_READONLY;
2309 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2310 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2313 if (flags & MS_REMOUNT)
2314 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2316 else if (flags & MS_BIND)
2317 retval = do_loopback(&path, dev_name, flags & MS_REC);
2318 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2319 retval = do_change_type(&path, flags);
2320 else if (flags & MS_MOVE)
2321 retval = do_move_mount(&path, dev_name);
2323 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2324 dev_name, data_page);
2330 static struct mnt_namespace *alloc_mnt_ns(void)
2332 struct mnt_namespace *new_ns;
2334 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2336 return ERR_PTR(-ENOMEM);
2337 atomic_set(&new_ns->count, 1);
2338 new_ns->root = NULL;
2339 INIT_LIST_HEAD(&new_ns->list);
2340 init_waitqueue_head(&new_ns->poll);
2345 void mnt_make_longterm(struct vfsmount *mnt)
2347 __mnt_make_longterm(mnt);
2350 void mnt_make_shortterm(struct vfsmount *mnt)
2353 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2355 br_write_lock(vfsmount_lock);
2356 atomic_dec(&mnt->mnt_longterm);
2357 br_write_unlock(vfsmount_lock);
2362 * Allocate a new namespace structure and populate it with contents
2363 * copied from the namespace of the passed in task structure.
2365 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2366 struct fs_struct *fs)
2368 struct mnt_namespace *new_ns;
2369 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2370 struct vfsmount *p, *q;
2372 new_ns = alloc_mnt_ns();
2376 down_write(&namespace_sem);
2377 /* First pass: copy the tree topology */
2378 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2379 CL_COPY_ALL | CL_EXPIRE);
2380 if (!new_ns->root) {
2381 up_write(&namespace_sem);
2383 return ERR_PTR(-ENOMEM);
2385 br_write_lock(vfsmount_lock);
2386 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2387 br_write_unlock(vfsmount_lock);
2390 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2391 * as belonging to new namespace. We have already acquired a private
2392 * fs_struct, so tsk->fs->lock is not needed.
2398 __mnt_make_longterm(q);
2400 if (p == fs->root.mnt) {
2401 fs->root.mnt = mntget(q);
2402 __mnt_make_longterm(q);
2403 mnt_make_shortterm(p);
2406 if (p == fs->pwd.mnt) {
2407 fs->pwd.mnt = mntget(q);
2408 __mnt_make_longterm(q);
2409 mnt_make_shortterm(p);
2413 p = next_mnt(p, mnt_ns->root);
2414 q = next_mnt(q, new_ns->root);
2416 up_write(&namespace_sem);
2426 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2427 struct fs_struct *new_fs)
2429 struct mnt_namespace *new_ns;
2434 if (!(flags & CLONE_NEWNS))
2437 new_ns = dup_mnt_ns(ns, new_fs);
2444 * create_mnt_ns - creates a private namespace and adds a root filesystem
2445 * @mnt: pointer to the new root filesystem mountpoint
2447 static struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2449 struct mnt_namespace *new_ns;
2451 new_ns = alloc_mnt_ns();
2452 if (!IS_ERR(new_ns)) {
2453 mnt->mnt_ns = new_ns;
2454 __mnt_make_longterm(mnt);
2456 list_add(&new_ns->list, &new_ns->root->mnt_list);
2463 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2465 struct mnt_namespace *ns;
2466 struct super_block *s;
2470 ns = create_mnt_ns(mnt);
2472 return ERR_CAST(ns);
2474 err = vfs_path_lookup(mnt->mnt_root, mnt,
2475 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2480 return ERR_PTR(err);
2482 /* trade a vfsmount reference for active sb one */
2483 s = path.mnt->mnt_sb;
2484 atomic_inc(&s->s_active);
2486 /* lock the sucker */
2487 down_write(&s->s_umount);
2488 /* ... and return the root of (sub)tree on it */
2491 EXPORT_SYMBOL(mount_subtree);
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 * Return true if path is reachable from root
2537 * namespace_sem or vfsmount_lock is held
2539 bool is_path_reachable(struct vfsmount *mnt, struct dentry *dentry,
2540 const struct path *root)
2542 while (mnt != root->mnt && mnt_has_parent(mnt)) {
2543 dentry = mnt->mnt_mountpoint;
2544 mnt = mnt->mnt_parent;
2546 return mnt == root->mnt && is_subdir(dentry, root->dentry);
2549 int path_is_under(struct path *path1, struct path *path2)
2552 br_read_lock(vfsmount_lock);
2553 res = is_path_reachable(path1->mnt, path1->dentry, path2);
2554 br_read_unlock(vfsmount_lock);
2557 EXPORT_SYMBOL(path_is_under);
2560 * pivot_root Semantics:
2561 * Moves the root file system of the current process to the directory put_old,
2562 * makes new_root as the new root file system of the current process, and sets
2563 * root/cwd of all processes which had them on the current root to new_root.
2566 * The new_root and put_old must be directories, and must not be on the
2567 * same file system as the current process root. The put_old must be
2568 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2569 * pointed to by put_old must yield the same directory as new_root. No other
2570 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2572 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2573 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2574 * in this situation.
2577 * - we don't move root/cwd if they are not at the root (reason: if something
2578 * cared enough to change them, it's probably wrong to force them elsewhere)
2579 * - it's okay to pick a root that isn't the root of a file system, e.g.
2580 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2581 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2584 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2585 const char __user *, put_old)
2587 struct path new, old, parent_path, root_parent, root;
2590 if (!capable(CAP_SYS_ADMIN))
2593 error = user_path_dir(new_root, &new);
2597 error = user_path_dir(put_old, &old);
2601 error = security_sb_pivotroot(&old, &new);
2605 get_fs_root(current->fs, &root);
2606 error = lock_mount(&old);
2611 if (IS_MNT_SHARED(old.mnt) ||
2612 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2613 IS_MNT_SHARED(root.mnt->mnt_parent))
2615 if (!check_mnt(root.mnt) || !check_mnt(new.mnt))
2618 if (d_unlinked(new.dentry))
2620 if (d_unlinked(old.dentry))
2623 if (new.mnt == root.mnt ||
2624 old.mnt == root.mnt)
2625 goto out4; /* loop, on the same file system */
2627 if (root.mnt->mnt_root != root.dentry)
2628 goto out4; /* not a mountpoint */
2629 if (!mnt_has_parent(root.mnt))
2630 goto out4; /* not attached */
2631 if (new.mnt->mnt_root != new.dentry)
2632 goto out4; /* not a mountpoint */
2633 if (!mnt_has_parent(new.mnt))
2634 goto out4; /* not attached */
2635 /* make sure we can reach put_old from new_root */
2636 if (!is_path_reachable(old.mnt, old.dentry, &new))
2638 br_write_lock(vfsmount_lock);
2639 detach_mnt(new.mnt, &parent_path);
2640 detach_mnt(root.mnt, &root_parent);
2641 /* mount old root on put_old */
2642 attach_mnt(root.mnt, &old);
2643 /* mount new_root on / */
2644 attach_mnt(new.mnt, &root_parent);
2645 touch_mnt_namespace(current->nsproxy->mnt_ns);
2646 br_write_unlock(vfsmount_lock);
2647 chroot_fs_refs(&root, &new);
2652 path_put(&root_parent);
2653 path_put(&parent_path);
2665 static void __init init_mount_tree(void)
2667 struct vfsmount *mnt;
2668 struct mnt_namespace *ns;
2671 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2673 panic("Can't create rootfs");
2675 ns = create_mnt_ns(mnt);
2677 panic("Can't allocate initial namespace");
2679 init_task.nsproxy->mnt_ns = ns;
2682 root.mnt = ns->root;
2683 root.dentry = ns->root->mnt_root;
2685 set_fs_pwd(current->fs, &root);
2686 set_fs_root(current->fs, &root);
2689 void __init mnt_init(void)
2694 init_rwsem(&namespace_sem);
2696 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2697 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2699 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2701 if (!mount_hashtable)
2702 panic("Failed to allocate mount hash table\n");
2704 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2706 for (u = 0; u < HASH_SIZE; u++)
2707 INIT_LIST_HEAD(&mount_hashtable[u]);
2709 br_lock_init(vfsmount_lock);
2713 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2715 fs_kobj = kobject_create_and_add("fs", NULL);
2717 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2722 void put_mnt_ns(struct mnt_namespace *ns)
2724 LIST_HEAD(umount_list);
2726 if (!atomic_dec_and_test(&ns->count))
2728 down_write(&namespace_sem);
2729 br_write_lock(vfsmount_lock);
2730 umount_tree(ns->root, 0, &umount_list);
2731 br_write_unlock(vfsmount_lock);
2732 up_write(&namespace_sem);
2733 release_mounts(&umount_list);
2737 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2739 struct vfsmount *mnt;
2740 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2743 * it is a longterm mount, don't release mnt until
2744 * we unmount before file sys is unregistered
2746 mnt_make_longterm(mnt);
2750 EXPORT_SYMBOL_GPL(kern_mount_data);
2752 void kern_unmount(struct vfsmount *mnt)
2754 /* release long term mount so mount point can be released */
2755 if (!IS_ERR_OR_NULL(mnt)) {
2756 mnt_make_shortterm(mnt);
2760 EXPORT_SYMBOL(kern_unmount);
2762 bool our_mnt(struct vfsmount *mnt)
2764 return check_mnt(mnt);