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/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/acct.h>
18 #include <linux/capability.h>
19 #include <linux/cpumask.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/mnt_namespace.h>
24 #include <linux/namei.h>
25 #include <linux/nsproxy.h>
26 #include <linux/security.h>
27 #include <linux/mount.h>
28 #include <linux/ramfs.h>
29 #include <linux/log2.h>
30 #include <linux/idr.h>
31 #include <linux/fs_struct.h>
32 #include <asm/uaccess.h>
33 #include <asm/unistd.h>
37 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
38 #define HASH_SIZE (1UL << HASH_SHIFT)
40 /* spinlock for vfsmount related operations, inplace of dcache_lock */
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
44 static DEFINE_IDA(mnt_id_ida);
45 static DEFINE_IDA(mnt_group_ida);
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);
57 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
60 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
61 tmp = tmp + (tmp >> HASH_SHIFT);
62 return tmp & (HASH_SIZE - 1);
65 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67 /* allocation is serialized by namespace_sem */
68 static int mnt_alloc_id(struct vfsmount *mnt)
73 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
74 spin_lock(&vfsmount_lock);
75 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
77 mnt_id_start = mnt->mnt_id + 1;
78 spin_unlock(&vfsmount_lock);
85 static void mnt_free_id(struct vfsmount *mnt)
88 spin_lock(&vfsmount_lock);
89 ida_remove(&mnt_id_ida, id);
90 if (mnt_id_start > id)
92 spin_unlock(&vfsmount_lock);
96 * Allocate a new peer group ID
98 * mnt_group_ida is protected by namespace_sem
100 static int mnt_alloc_group_id(struct vfsmount *mnt)
104 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107 res = ida_get_new_above(&mnt_group_ida,
111 mnt_group_start = mnt->mnt_group_id + 1;
117 * Release a peer group ID
119 void mnt_release_group_id(struct vfsmount *mnt)
121 int id = mnt->mnt_group_id;
122 ida_remove(&mnt_group_ida, id);
123 if (mnt_group_start > id)
124 mnt_group_start = id;
125 mnt->mnt_group_id = 0;
128 struct vfsmount *alloc_vfsmnt(const char *name)
130 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
134 err = mnt_alloc_id(mnt);
139 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
140 if (!mnt->mnt_devname)
144 atomic_set(&mnt->mnt_count, 1);
145 INIT_LIST_HEAD(&mnt->mnt_hash);
146 INIT_LIST_HEAD(&mnt->mnt_child);
147 INIT_LIST_HEAD(&mnt->mnt_mounts);
148 INIT_LIST_HEAD(&mnt->mnt_list);
149 INIT_LIST_HEAD(&mnt->mnt_expire);
150 INIT_LIST_HEAD(&mnt->mnt_share);
151 INIT_LIST_HEAD(&mnt->mnt_slave_list);
152 INIT_LIST_HEAD(&mnt->mnt_slave);
154 mnt->mnt_writers = alloc_percpu(int);
155 if (!mnt->mnt_writers)
156 goto out_free_devname;
158 mnt->mnt_writers = 0;
165 kfree(mnt->mnt_devname);
170 kmem_cache_free(mnt_cache, mnt);
175 * Most r/o checks on a fs are for operations that take
176 * discrete amounts of time, like a write() or unlink().
177 * We must keep track of when those operations start
178 * (for permission checks) and when they end, so that
179 * we can determine when writes are able to occur to
183 * __mnt_is_readonly: check whether a mount is read-only
184 * @mnt: the mount to check for its write status
186 * This shouldn't be used directly ouside of the VFS.
187 * It does not guarantee that the filesystem will stay
188 * r/w, just that it is right *now*. This can not and
189 * should not be used in place of IS_RDONLY(inode).
190 * mnt_want/drop_write() will _keep_ the filesystem
193 int __mnt_is_readonly(struct vfsmount *mnt)
195 if (mnt->mnt_flags & MNT_READONLY)
197 if (mnt->mnt_sb->s_flags & MS_RDONLY)
201 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
203 static inline void inc_mnt_writers(struct vfsmount *mnt)
206 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
212 static inline void dec_mnt_writers(struct vfsmount *mnt)
215 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
221 static unsigned int count_mnt_writers(struct vfsmount *mnt)
224 unsigned int count = 0;
227 for_each_possible_cpu(cpu) {
228 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
233 return mnt->mnt_writers;
238 * Most r/o checks on a fs are for operations that take
239 * discrete amounts of time, like a write() or unlink().
240 * We must keep track of when those operations start
241 * (for permission checks) and when they end, so that
242 * we can determine when writes are able to occur to
246 * mnt_want_write - get write access to a mount
247 * @mnt: the mount on which to take a write
249 * This tells the low-level filesystem that a write is
250 * about to be performed to it, and makes sure that
251 * writes are allowed before returning success. When
252 * the write operation is finished, mnt_drop_write()
253 * must be called. This is effectively a refcount.
255 int mnt_want_write(struct vfsmount *mnt)
260 inc_mnt_writers(mnt);
262 * The store to inc_mnt_writers must be visible before we pass
263 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
264 * incremented count after it has set MNT_WRITE_HOLD.
267 while (mnt->mnt_flags & MNT_WRITE_HOLD)
270 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
271 * be set to match its requirements. So we must not load that until
272 * MNT_WRITE_HOLD is cleared.
275 if (__mnt_is_readonly(mnt)) {
276 dec_mnt_writers(mnt);
284 EXPORT_SYMBOL_GPL(mnt_want_write);
287 * mnt_clone_write - get write access to a mount
288 * @mnt: the mount on which to take a write
290 * This is effectively like mnt_want_write, except
291 * it must only be used to take an extra write reference
292 * on a mountpoint that we already know has a write reference
293 * on it. This allows some optimisation.
295 * After finished, mnt_drop_write must be called as usual to
296 * drop the reference.
298 int mnt_clone_write(struct vfsmount *mnt)
300 /* superblock may be r/o */
301 if (__mnt_is_readonly(mnt))
304 inc_mnt_writers(mnt);
308 EXPORT_SYMBOL_GPL(mnt_clone_write);
311 * mnt_want_write_file - get write access to a file's mount
312 * @file: the file who's mount on which to take a write
314 * This is like mnt_want_write, but it takes a file and can
315 * do some optimisations if the file is open for write already
317 int mnt_want_write_file(struct file *file)
319 if (!(file->f_mode & FMODE_WRITE))
320 return mnt_want_write(file->f_path.mnt);
322 return mnt_clone_write(file->f_path.mnt);
324 EXPORT_SYMBOL_GPL(mnt_want_write_file);
327 * mnt_drop_write - give up write access to a mount
328 * @mnt: the mount on which to give up write access
330 * Tells the low-level filesystem that we are done
331 * performing writes to it. Must be matched with
332 * mnt_want_write() call above.
334 void mnt_drop_write(struct vfsmount *mnt)
337 dec_mnt_writers(mnt);
340 EXPORT_SYMBOL_GPL(mnt_drop_write);
342 static int mnt_make_readonly(struct vfsmount *mnt)
346 spin_lock(&vfsmount_lock);
347 mnt->mnt_flags |= MNT_WRITE_HOLD;
349 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
350 * should be visible before we do.
355 * With writers on hold, if this value is zero, then there are
356 * definitely no active writers (although held writers may subsequently
357 * increment the count, they'll have to wait, and decrement it after
358 * seeing MNT_READONLY).
360 * It is OK to have counter incremented on one CPU and decremented on
361 * another: the sum will add up correctly. The danger would be when we
362 * sum up each counter, if we read a counter before it is incremented,
363 * but then read another CPU's count which it has been subsequently
364 * decremented from -- we would see more decrements than we should.
365 * MNT_WRITE_HOLD protects against this scenario, because
366 * mnt_want_write first increments count, then smp_mb, then spins on
367 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
368 * we're counting up here.
370 if (count_mnt_writers(mnt) > 0)
373 mnt->mnt_flags |= MNT_READONLY;
375 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
376 * that become unheld will see MNT_READONLY.
379 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
380 spin_unlock(&vfsmount_lock);
384 static void __mnt_unmake_readonly(struct vfsmount *mnt)
386 spin_lock(&vfsmount_lock);
387 mnt->mnt_flags &= ~MNT_READONLY;
388 spin_unlock(&vfsmount_lock);
391 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
394 mnt->mnt_root = dget(sb->s_root);
397 EXPORT_SYMBOL(simple_set_mnt);
399 void free_vfsmnt(struct vfsmount *mnt)
401 kfree(mnt->mnt_devname);
404 free_percpu(mnt->mnt_writers);
406 kmem_cache_free(mnt_cache, mnt);
410 * find the first or last mount at @dentry on vfsmount @mnt depending on
411 * @dir. If @dir is set return the first mount else return the last mount.
413 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
416 struct list_head *head = mount_hashtable + hash(mnt, dentry);
417 struct list_head *tmp = head;
418 struct vfsmount *p, *found = NULL;
421 tmp = dir ? tmp->next : tmp->prev;
425 p = list_entry(tmp, struct vfsmount, mnt_hash);
426 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
435 * lookup_mnt increments the ref count before returning
436 * the vfsmount struct.
438 struct vfsmount *lookup_mnt(struct path *path)
440 struct vfsmount *child_mnt;
441 spin_lock(&vfsmount_lock);
442 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
444 spin_unlock(&vfsmount_lock);
448 static inline int check_mnt(struct vfsmount *mnt)
450 return mnt->mnt_ns == current->nsproxy->mnt_ns;
453 static void touch_mnt_namespace(struct mnt_namespace *ns)
457 wake_up_interruptible(&ns->poll);
461 static void __touch_mnt_namespace(struct mnt_namespace *ns)
463 if (ns && ns->event != event) {
465 wake_up_interruptible(&ns->poll);
469 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
471 old_path->dentry = mnt->mnt_mountpoint;
472 old_path->mnt = mnt->mnt_parent;
473 mnt->mnt_parent = mnt;
474 mnt->mnt_mountpoint = mnt->mnt_root;
475 list_del_init(&mnt->mnt_child);
476 list_del_init(&mnt->mnt_hash);
477 old_path->dentry->d_mounted--;
480 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
481 struct vfsmount *child_mnt)
483 child_mnt->mnt_parent = mntget(mnt);
484 child_mnt->mnt_mountpoint = dget(dentry);
488 static void attach_mnt(struct vfsmount *mnt, struct path *path)
490 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
491 list_add_tail(&mnt->mnt_hash, mount_hashtable +
492 hash(path->mnt, path->dentry));
493 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
497 * the caller must hold vfsmount_lock
499 static void commit_tree(struct vfsmount *mnt)
501 struct vfsmount *parent = mnt->mnt_parent;
504 struct mnt_namespace *n = parent->mnt_ns;
506 BUG_ON(parent == mnt);
508 list_add_tail(&head, &mnt->mnt_list);
509 list_for_each_entry(m, &head, mnt_list)
511 list_splice(&head, n->list.prev);
513 list_add_tail(&mnt->mnt_hash, mount_hashtable +
514 hash(parent, mnt->mnt_mountpoint));
515 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
516 touch_mnt_namespace(n);
519 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
521 struct list_head *next = p->mnt_mounts.next;
522 if (next == &p->mnt_mounts) {
526 next = p->mnt_child.next;
527 if (next != &p->mnt_parent->mnt_mounts)
532 return list_entry(next, struct vfsmount, mnt_child);
535 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
537 struct list_head *prev = p->mnt_mounts.prev;
538 while (prev != &p->mnt_mounts) {
539 p = list_entry(prev, struct vfsmount, mnt_child);
540 prev = p->mnt_mounts.prev;
545 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
548 struct super_block *sb = old->mnt_sb;
549 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
552 if (flag & (CL_SLAVE | CL_PRIVATE))
553 mnt->mnt_group_id = 0; /* not a peer of original */
555 mnt->mnt_group_id = old->mnt_group_id;
557 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
558 int err = mnt_alloc_group_id(mnt);
563 mnt->mnt_flags = old->mnt_flags;
564 atomic_inc(&sb->s_active);
566 mnt->mnt_root = dget(root);
567 mnt->mnt_mountpoint = mnt->mnt_root;
568 mnt->mnt_parent = mnt;
570 if (flag & CL_SLAVE) {
571 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
572 mnt->mnt_master = old;
573 CLEAR_MNT_SHARED(mnt);
574 } else if (!(flag & CL_PRIVATE)) {
575 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
576 list_add(&mnt->mnt_share, &old->mnt_share);
577 if (IS_MNT_SLAVE(old))
578 list_add(&mnt->mnt_slave, &old->mnt_slave);
579 mnt->mnt_master = old->mnt_master;
581 if (flag & CL_MAKE_SHARED)
584 /* stick the duplicate mount on the same expiry list
585 * as the original if that was on one */
586 if (flag & CL_EXPIRE) {
587 if (!list_empty(&old->mnt_expire))
588 list_add(&mnt->mnt_expire, &old->mnt_expire);
598 static inline void __mntput(struct vfsmount *mnt)
600 struct super_block *sb = mnt->mnt_sb;
602 * This probably indicates that somebody messed
603 * up a mnt_want/drop_write() pair. If this
604 * happens, the filesystem was probably unable
605 * to make r/w->r/o transitions.
608 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
609 * provides barriers, so count_mnt_writers() below is safe. AV
611 WARN_ON(count_mnt_writers(mnt));
614 deactivate_super(sb);
617 void mntput_no_expire(struct vfsmount *mnt)
620 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
621 if (likely(!mnt->mnt_pinned)) {
622 spin_unlock(&vfsmount_lock);
626 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
628 spin_unlock(&vfsmount_lock);
629 acct_auto_close_mnt(mnt);
630 security_sb_umount_close(mnt);
635 EXPORT_SYMBOL(mntput_no_expire);
637 void mnt_pin(struct vfsmount *mnt)
639 spin_lock(&vfsmount_lock);
641 spin_unlock(&vfsmount_lock);
644 EXPORT_SYMBOL(mnt_pin);
646 void mnt_unpin(struct vfsmount *mnt)
648 spin_lock(&vfsmount_lock);
649 if (mnt->mnt_pinned) {
650 atomic_inc(&mnt->mnt_count);
653 spin_unlock(&vfsmount_lock);
656 EXPORT_SYMBOL(mnt_unpin);
658 static inline void mangle(struct seq_file *m, const char *s)
660 seq_escape(m, s, " \t\n\\");
664 * Simple .show_options callback for filesystems which don't want to
665 * implement more complex mount option showing.
667 * See also save_mount_options().
669 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
674 options = rcu_dereference(mnt->mnt_sb->s_options);
676 if (options != NULL && options[0]) {
684 EXPORT_SYMBOL(generic_show_options);
687 * If filesystem uses generic_show_options(), this function should be
688 * called from the fill_super() callback.
690 * The .remount_fs callback usually needs to be handled in a special
691 * way, to make sure, that previous options are not overwritten if the
694 * Also note, that if the filesystem's .remount_fs function doesn't
695 * reset all options to their default value, but changes only newly
696 * given options, then the displayed options will not reflect reality
699 void save_mount_options(struct super_block *sb, char *options)
701 BUG_ON(sb->s_options);
702 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
704 EXPORT_SYMBOL(save_mount_options);
706 void replace_mount_options(struct super_block *sb, char *options)
708 char *old = sb->s_options;
709 rcu_assign_pointer(sb->s_options, options);
715 EXPORT_SYMBOL(replace_mount_options);
717 #ifdef CONFIG_PROC_FS
719 static void *m_start(struct seq_file *m, loff_t *pos)
721 struct proc_mounts *p = m->private;
723 down_read(&namespace_sem);
724 return seq_list_start(&p->ns->list, *pos);
727 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
729 struct proc_mounts *p = m->private;
731 return seq_list_next(v, &p->ns->list, pos);
734 static void m_stop(struct seq_file *m, void *v)
736 up_read(&namespace_sem);
739 struct proc_fs_info {
744 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
746 static const struct proc_fs_info fs_info[] = {
747 { MS_SYNCHRONOUS, ",sync" },
748 { MS_DIRSYNC, ",dirsync" },
749 { MS_MANDLOCK, ",mand" },
752 const struct proc_fs_info *fs_infop;
754 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
755 if (sb->s_flags & fs_infop->flag)
756 seq_puts(m, fs_infop->str);
759 return security_sb_show_options(m, sb);
762 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
764 static const struct proc_fs_info mnt_info[] = {
765 { MNT_NOSUID, ",nosuid" },
766 { MNT_NODEV, ",nodev" },
767 { MNT_NOEXEC, ",noexec" },
768 { MNT_NOATIME, ",noatime" },
769 { MNT_NODIRATIME, ",nodiratime" },
770 { MNT_RELATIME, ",relatime" },
771 { MNT_STRICTATIME, ",strictatime" },
774 const struct proc_fs_info *fs_infop;
776 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
777 if (mnt->mnt_flags & fs_infop->flag)
778 seq_puts(m, fs_infop->str);
782 static void show_type(struct seq_file *m, struct super_block *sb)
784 mangle(m, sb->s_type->name);
785 if (sb->s_subtype && sb->s_subtype[0]) {
787 mangle(m, sb->s_subtype);
791 static int show_vfsmnt(struct seq_file *m, void *v)
793 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
795 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
797 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
799 seq_path(m, &mnt_path, " \t\n\\");
801 show_type(m, mnt->mnt_sb);
802 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
803 err = show_sb_opts(m, mnt->mnt_sb);
806 show_mnt_opts(m, mnt);
807 if (mnt->mnt_sb->s_op->show_options)
808 err = mnt->mnt_sb->s_op->show_options(m, mnt);
809 seq_puts(m, " 0 0\n");
814 const struct seq_operations mounts_op = {
821 static int show_mountinfo(struct seq_file *m, void *v)
823 struct proc_mounts *p = m->private;
824 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
825 struct super_block *sb = mnt->mnt_sb;
826 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
827 struct path root = p->root;
830 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
831 MAJOR(sb->s_dev), MINOR(sb->s_dev));
832 seq_dentry(m, mnt->mnt_root, " \t\n\\");
834 seq_path_root(m, &mnt_path, &root, " \t\n\\");
835 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
837 * Mountpoint is outside root, discard that one. Ugly,
838 * but less so than trying to do that in iterator in a
839 * race-free way (due to renames).
843 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
844 show_mnt_opts(m, mnt);
846 /* Tagged fields ("foo:X" or "bar") */
847 if (IS_MNT_SHARED(mnt))
848 seq_printf(m, " shared:%i", mnt->mnt_group_id);
849 if (IS_MNT_SLAVE(mnt)) {
850 int master = mnt->mnt_master->mnt_group_id;
851 int dom = get_dominating_id(mnt, &p->root);
852 seq_printf(m, " master:%i", master);
853 if (dom && dom != master)
854 seq_printf(m, " propagate_from:%i", dom);
856 if (IS_MNT_UNBINDABLE(mnt))
857 seq_puts(m, " unbindable");
859 /* Filesystem specific data */
863 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
864 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
865 err = show_sb_opts(m, sb);
868 if (sb->s_op->show_options)
869 err = sb->s_op->show_options(m, mnt);
875 const struct seq_operations mountinfo_op = {
879 .show = show_mountinfo,
882 static int show_vfsstat(struct seq_file *m, void *v)
884 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
885 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
889 if (mnt->mnt_devname) {
890 seq_puts(m, "device ");
891 mangle(m, mnt->mnt_devname);
893 seq_puts(m, "no device");
896 seq_puts(m, " mounted on ");
897 seq_path(m, &mnt_path, " \t\n\\");
900 /* file system type */
901 seq_puts(m, "with fstype ");
902 show_type(m, mnt->mnt_sb);
904 /* optional statistics */
905 if (mnt->mnt_sb->s_op->show_stats) {
907 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
914 const struct seq_operations mountstats_op = {
918 .show = show_vfsstat,
920 #endif /* CONFIG_PROC_FS */
923 * may_umount_tree - check if a mount tree is busy
924 * @mnt: root of mount tree
926 * This is called to check if a tree of mounts has any
927 * open files, pwds, chroots or sub mounts that are
930 int may_umount_tree(struct vfsmount *mnt)
933 int minimum_refs = 0;
936 spin_lock(&vfsmount_lock);
937 for (p = mnt; p; p = next_mnt(p, mnt)) {
938 actual_refs += atomic_read(&p->mnt_count);
941 spin_unlock(&vfsmount_lock);
943 if (actual_refs > minimum_refs)
949 EXPORT_SYMBOL(may_umount_tree);
952 * may_umount - check if a mount point is busy
953 * @mnt: root of mount
955 * This is called to check if a mount point has any
956 * open files, pwds, chroots or sub mounts. If the
957 * mount has sub mounts this will return busy
958 * regardless of whether the sub mounts are busy.
960 * Doesn't take quota and stuff into account. IOW, in some cases it will
961 * give false negatives. The main reason why it's here is that we need
962 * a non-destructive way to look for easily umountable filesystems.
964 int may_umount(struct vfsmount *mnt)
967 spin_lock(&vfsmount_lock);
968 if (propagate_mount_busy(mnt, 2))
970 spin_unlock(&vfsmount_lock);
974 EXPORT_SYMBOL(may_umount);
976 void release_mounts(struct list_head *head)
978 struct vfsmount *mnt;
979 while (!list_empty(head)) {
980 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
981 list_del_init(&mnt->mnt_hash);
982 if (mnt->mnt_parent != mnt) {
983 struct dentry *dentry;
985 spin_lock(&vfsmount_lock);
986 dentry = mnt->mnt_mountpoint;
988 mnt->mnt_mountpoint = mnt->mnt_root;
989 mnt->mnt_parent = mnt;
991 spin_unlock(&vfsmount_lock);
999 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1003 for (p = mnt; p; p = next_mnt(p, mnt))
1004 list_move(&p->mnt_hash, kill);
1007 propagate_umount(kill);
1009 list_for_each_entry(p, kill, mnt_hash) {
1010 list_del_init(&p->mnt_expire);
1011 list_del_init(&p->mnt_list);
1012 __touch_mnt_namespace(p->mnt_ns);
1014 list_del_init(&p->mnt_child);
1015 if (p->mnt_parent != p) {
1016 p->mnt_parent->mnt_ghosts++;
1017 p->mnt_mountpoint->d_mounted--;
1019 change_mnt_propagation(p, MS_PRIVATE);
1023 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1025 static int do_umount(struct vfsmount *mnt, int flags)
1027 struct super_block *sb = mnt->mnt_sb;
1029 LIST_HEAD(umount_list);
1031 retval = security_sb_umount(mnt, flags);
1036 * Allow userspace to request a mountpoint be expired rather than
1037 * unmounting unconditionally. Unmount only happens if:
1038 * (1) the mark is already set (the mark is cleared by mntput())
1039 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1041 if (flags & MNT_EXPIRE) {
1042 if (mnt == current->fs->root.mnt ||
1043 flags & (MNT_FORCE | MNT_DETACH))
1046 if (atomic_read(&mnt->mnt_count) != 2)
1049 if (!xchg(&mnt->mnt_expiry_mark, 1))
1054 * If we may have to abort operations to get out of this
1055 * mount, and they will themselves hold resources we must
1056 * allow the fs to do things. In the Unix tradition of
1057 * 'Gee thats tricky lets do it in userspace' the umount_begin
1058 * might fail to complete on the first run through as other tasks
1059 * must return, and the like. Thats for the mount program to worry
1060 * about for the moment.
1063 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1064 sb->s_op->umount_begin(sb);
1068 * No sense to grab the lock for this test, but test itself looks
1069 * somewhat bogus. Suggestions for better replacement?
1070 * Ho-hum... In principle, we might treat that as umount + switch
1071 * to rootfs. GC would eventually take care of the old vfsmount.
1072 * Actually it makes sense, especially if rootfs would contain a
1073 * /reboot - static binary that would close all descriptors and
1074 * call reboot(9). Then init(8) could umount root and exec /reboot.
1076 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1078 * Special case for "unmounting" root ...
1079 * we just try to remount it readonly.
1081 down_write(&sb->s_umount);
1082 if (!(sb->s_flags & MS_RDONLY))
1083 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1084 up_write(&sb->s_umount);
1088 down_write(&namespace_sem);
1089 spin_lock(&vfsmount_lock);
1092 if (!(flags & MNT_DETACH))
1093 shrink_submounts(mnt, &umount_list);
1096 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1097 if (!list_empty(&mnt->mnt_list))
1098 umount_tree(mnt, 1, &umount_list);
1101 spin_unlock(&vfsmount_lock);
1103 security_sb_umount_busy(mnt);
1104 up_write(&namespace_sem);
1105 release_mounts(&umount_list);
1110 * Now umount can handle mount points as well as block devices.
1111 * This is important for filesystems which use unnamed block devices.
1113 * We now support a flag for forced unmount like the other 'big iron'
1114 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1117 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1122 retval = user_path(name, &path);
1126 if (path.dentry != path.mnt->mnt_root)
1128 if (!check_mnt(path.mnt))
1132 if (!capable(CAP_SYS_ADMIN))
1135 retval = do_umount(path.mnt, flags);
1137 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1139 mntput_no_expire(path.mnt);
1144 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1147 * The 2.0 compatible umount. No flags.
1149 SYSCALL_DEFINE1(oldumount, char __user *, name)
1151 return sys_umount(name, 0);
1156 static int mount_is_safe(struct path *path)
1158 if (capable(CAP_SYS_ADMIN))
1162 if (S_ISLNK(path->dentry->d_inode->i_mode))
1164 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1165 if (current_uid() != path->dentry->d_inode->i_uid)
1168 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1174 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1177 struct vfsmount *res, *p, *q, *r, *s;
1180 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1183 res = q = clone_mnt(mnt, dentry, flag);
1186 q->mnt_mountpoint = mnt->mnt_mountpoint;
1189 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1190 if (!is_subdir(r->mnt_mountpoint, dentry))
1193 for (s = r; s; s = next_mnt(s, r)) {
1194 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1195 s = skip_mnt_tree(s);
1198 while (p != s->mnt_parent) {
1204 path.dentry = p->mnt_mountpoint;
1205 q = clone_mnt(p, p->mnt_root, flag);
1208 spin_lock(&vfsmount_lock);
1209 list_add_tail(&q->mnt_list, &res->mnt_list);
1210 attach_mnt(q, &path);
1211 spin_unlock(&vfsmount_lock);
1217 LIST_HEAD(umount_list);
1218 spin_lock(&vfsmount_lock);
1219 umount_tree(res, 0, &umount_list);
1220 spin_unlock(&vfsmount_lock);
1221 release_mounts(&umount_list);
1226 struct vfsmount *collect_mounts(struct path *path)
1228 struct vfsmount *tree;
1229 down_write(&namespace_sem);
1230 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1231 up_write(&namespace_sem);
1235 void drop_collected_mounts(struct vfsmount *mnt)
1237 LIST_HEAD(umount_list);
1238 down_write(&namespace_sem);
1239 spin_lock(&vfsmount_lock);
1240 umount_tree(mnt, 0, &umount_list);
1241 spin_unlock(&vfsmount_lock);
1242 up_write(&namespace_sem);
1243 release_mounts(&umount_list);
1246 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1250 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1251 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1252 mnt_release_group_id(p);
1256 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1260 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1261 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1262 int err = mnt_alloc_group_id(p);
1264 cleanup_group_ids(mnt, p);
1274 * @source_mnt : mount tree to be attached
1275 * @nd : place the mount tree @source_mnt is attached
1276 * @parent_nd : if non-null, detach the source_mnt from its parent and
1277 * store the parent mount and mountpoint dentry.
1278 * (done when source_mnt is moved)
1280 * NOTE: in the table below explains the semantics when a source mount
1281 * of a given type is attached to a destination mount of a given type.
1282 * ---------------------------------------------------------------------------
1283 * | BIND MOUNT OPERATION |
1284 * |**************************************************************************
1285 * | source-->| shared | private | slave | unbindable |
1289 * |**************************************************************************
1290 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1292 * |non-shared| shared (+) | private | slave (*) | invalid |
1293 * ***************************************************************************
1294 * A bind operation clones the source mount and mounts the clone on the
1295 * destination mount.
1297 * (++) the cloned mount is propagated to all the mounts in the propagation
1298 * tree of the destination mount and the cloned mount is added to
1299 * the peer group of the source mount.
1300 * (+) the cloned mount is created under the destination mount and is marked
1301 * as shared. The cloned mount is added to the peer group of the source
1303 * (+++) the mount is propagated to all the mounts in the propagation tree
1304 * of the destination mount and the cloned mount is made slave
1305 * of the same master as that of the source mount. The cloned mount
1306 * is marked as 'shared and slave'.
1307 * (*) the cloned mount is made a slave of the same master as that of the
1310 * ---------------------------------------------------------------------------
1311 * | MOVE MOUNT OPERATION |
1312 * |**************************************************************************
1313 * | source-->| shared | private | slave | unbindable |
1317 * |**************************************************************************
1318 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1320 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1321 * ***************************************************************************
1323 * (+) the mount is moved to the destination. And is then propagated to
1324 * all the mounts in the propagation tree of the destination mount.
1325 * (+*) the mount is moved to the destination.
1326 * (+++) the mount is moved to the destination and is then propagated to
1327 * all the mounts belonging to the destination mount's propagation tree.
1328 * the mount is marked as 'shared and slave'.
1329 * (*) the mount continues to be a slave at the new location.
1331 * if the source mount is a tree, the operations explained above is
1332 * applied to each mount in the tree.
1333 * Must be called without spinlocks held, since this function can sleep
1336 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1337 struct path *path, struct path *parent_path)
1339 LIST_HEAD(tree_list);
1340 struct vfsmount *dest_mnt = path->mnt;
1341 struct dentry *dest_dentry = path->dentry;
1342 struct vfsmount *child, *p;
1345 if (IS_MNT_SHARED(dest_mnt)) {
1346 err = invent_group_ids(source_mnt, true);
1350 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1352 goto out_cleanup_ids;
1354 if (IS_MNT_SHARED(dest_mnt)) {
1355 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1359 spin_lock(&vfsmount_lock);
1361 detach_mnt(source_mnt, parent_path);
1362 attach_mnt(source_mnt, path);
1363 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1365 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1366 commit_tree(source_mnt);
1369 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1370 list_del_init(&child->mnt_hash);
1373 spin_unlock(&vfsmount_lock);
1377 if (IS_MNT_SHARED(dest_mnt))
1378 cleanup_group_ids(source_mnt, NULL);
1383 static int graft_tree(struct vfsmount *mnt, struct path *path)
1386 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1389 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1390 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1394 mutex_lock(&path->dentry->d_inode->i_mutex);
1395 if (IS_DEADDIR(path->dentry->d_inode))
1398 err = security_sb_check_sb(mnt, path);
1403 if (!d_unlinked(path->dentry))
1404 err = attach_recursive_mnt(mnt, path, NULL);
1406 mutex_unlock(&path->dentry->d_inode->i_mutex);
1408 security_sb_post_addmount(mnt, path);
1413 * recursively change the type of the mountpoint.
1415 static int do_change_type(struct path *path, int flag)
1417 struct vfsmount *m, *mnt = path->mnt;
1418 int recurse = flag & MS_REC;
1419 int type = flag & ~MS_REC;
1422 if (!capable(CAP_SYS_ADMIN))
1425 if (path->dentry != path->mnt->mnt_root)
1428 down_write(&namespace_sem);
1429 if (type == MS_SHARED) {
1430 err = invent_group_ids(mnt, recurse);
1435 spin_lock(&vfsmount_lock);
1436 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1437 change_mnt_propagation(m, type);
1438 spin_unlock(&vfsmount_lock);
1441 up_write(&namespace_sem);
1446 * do loopback mount.
1448 static int do_loopback(struct path *path, char *old_name,
1451 struct path old_path;
1452 struct vfsmount *mnt = NULL;
1453 int err = mount_is_safe(path);
1456 if (!old_name || !*old_name)
1458 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1462 down_write(&namespace_sem);
1464 if (IS_MNT_UNBINDABLE(old_path.mnt))
1467 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1472 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1474 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1479 err = graft_tree(mnt, path);
1481 LIST_HEAD(umount_list);
1482 spin_lock(&vfsmount_lock);
1483 umount_tree(mnt, 0, &umount_list);
1484 spin_unlock(&vfsmount_lock);
1485 release_mounts(&umount_list);
1489 up_write(&namespace_sem);
1490 path_put(&old_path);
1494 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1497 int readonly_request = 0;
1499 if (ms_flags & MS_RDONLY)
1500 readonly_request = 1;
1501 if (readonly_request == __mnt_is_readonly(mnt))
1504 if (readonly_request)
1505 error = mnt_make_readonly(mnt);
1507 __mnt_unmake_readonly(mnt);
1512 * change filesystem flags. dir should be a physical root of filesystem.
1513 * If you've mounted a non-root directory somewhere and want to do remount
1514 * on it - tough luck.
1516 static int do_remount(struct path *path, int flags, int mnt_flags,
1520 struct super_block *sb = path->mnt->mnt_sb;
1522 if (!capable(CAP_SYS_ADMIN))
1525 if (!check_mnt(path->mnt))
1528 if (path->dentry != path->mnt->mnt_root)
1531 down_write(&sb->s_umount);
1532 if (flags & MS_BIND)
1533 err = change_mount_flags(path->mnt, flags);
1535 err = do_remount_sb(sb, flags, data, 0);
1537 path->mnt->mnt_flags = mnt_flags;
1538 up_write(&sb->s_umount);
1540 security_sb_post_remount(path->mnt, flags, data);
1542 spin_lock(&vfsmount_lock);
1543 touch_mnt_namespace(path->mnt->mnt_ns);
1544 spin_unlock(&vfsmount_lock);
1549 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1552 for (p = mnt; p; p = next_mnt(p, mnt)) {
1553 if (IS_MNT_UNBINDABLE(p))
1559 static int do_move_mount(struct path *path, char *old_name)
1561 struct path old_path, parent_path;
1564 if (!capable(CAP_SYS_ADMIN))
1566 if (!old_name || !*old_name)
1568 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1572 down_write(&namespace_sem);
1573 while (d_mountpoint(path->dentry) &&
1577 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1581 mutex_lock(&path->dentry->d_inode->i_mutex);
1582 if (IS_DEADDIR(path->dentry->d_inode))
1585 if (d_unlinked(path->dentry))
1589 if (old_path.dentry != old_path.mnt->mnt_root)
1592 if (old_path.mnt == old_path.mnt->mnt_parent)
1595 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1596 S_ISDIR(old_path.dentry->d_inode->i_mode))
1599 * Don't move a mount residing in a shared parent.
1601 if (old_path.mnt->mnt_parent &&
1602 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1605 * Don't move a mount tree containing unbindable mounts to a destination
1606 * mount which is shared.
1608 if (IS_MNT_SHARED(path->mnt) &&
1609 tree_contains_unbindable(old_path.mnt))
1612 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1613 if (p == old_path.mnt)
1616 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1620 /* if the mount is moved, it should no longer be expire
1622 list_del_init(&old_path.mnt->mnt_expire);
1624 mutex_unlock(&path->dentry->d_inode->i_mutex);
1626 up_write(&namespace_sem);
1628 path_put(&parent_path);
1629 path_put(&old_path);
1634 * create a new mount for userspace and request it to be added into the
1637 static int do_new_mount(struct path *path, char *type, int flags,
1638 int mnt_flags, char *name, void *data)
1640 struct vfsmount *mnt;
1642 if (!type || !memchr(type, 0, PAGE_SIZE))
1645 /* we need capabilities... */
1646 if (!capable(CAP_SYS_ADMIN))
1650 mnt = do_kern_mount(type, flags, name, data);
1653 return PTR_ERR(mnt);
1655 return do_add_mount(mnt, path, mnt_flags, NULL);
1659 * add a mount into a namespace's mount tree
1660 * - provide the option of adding the new mount to an expiration list
1662 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1663 int mnt_flags, struct list_head *fslist)
1667 down_write(&namespace_sem);
1668 /* Something was mounted here while we slept */
1669 while (d_mountpoint(path->dentry) &&
1673 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1676 /* Refuse the same filesystem on the same mount point */
1678 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1679 path->mnt->mnt_root == path->dentry)
1683 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1686 newmnt->mnt_flags = mnt_flags;
1687 if ((err = graft_tree(newmnt, path)))
1690 if (fslist) /* add to the specified expiration list */
1691 list_add_tail(&newmnt->mnt_expire, fslist);
1693 up_write(&namespace_sem);
1697 up_write(&namespace_sem);
1702 EXPORT_SYMBOL_GPL(do_add_mount);
1705 * process a list of expirable mountpoints with the intent of discarding any
1706 * mountpoints that aren't in use and haven't been touched since last we came
1709 void mark_mounts_for_expiry(struct list_head *mounts)
1711 struct vfsmount *mnt, *next;
1712 LIST_HEAD(graveyard);
1715 if (list_empty(mounts))
1718 down_write(&namespace_sem);
1719 spin_lock(&vfsmount_lock);
1721 /* extract from the expiration list every vfsmount that matches the
1722 * following criteria:
1723 * - only referenced by its parent vfsmount
1724 * - still marked for expiry (marked on the last call here; marks are
1725 * cleared by mntput())
1727 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1728 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1729 propagate_mount_busy(mnt, 1))
1731 list_move(&mnt->mnt_expire, &graveyard);
1733 while (!list_empty(&graveyard)) {
1734 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1735 touch_mnt_namespace(mnt->mnt_ns);
1736 umount_tree(mnt, 1, &umounts);
1738 spin_unlock(&vfsmount_lock);
1739 up_write(&namespace_sem);
1741 release_mounts(&umounts);
1744 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1747 * Ripoff of 'select_parent()'
1749 * search the list of submounts for a given mountpoint, and move any
1750 * shrinkable submounts to the 'graveyard' list.
1752 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1754 struct vfsmount *this_parent = parent;
1755 struct list_head *next;
1759 next = this_parent->mnt_mounts.next;
1761 while (next != &this_parent->mnt_mounts) {
1762 struct list_head *tmp = next;
1763 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1766 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1769 * Descend a level if the d_mounts list is non-empty.
1771 if (!list_empty(&mnt->mnt_mounts)) {
1776 if (!propagate_mount_busy(mnt, 1)) {
1777 list_move_tail(&mnt->mnt_expire, graveyard);
1782 * All done at this level ... ascend and resume the search
1784 if (this_parent != parent) {
1785 next = this_parent->mnt_child.next;
1786 this_parent = this_parent->mnt_parent;
1793 * process a list of expirable mountpoints with the intent of discarding any
1794 * submounts of a specific parent mountpoint
1796 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1798 LIST_HEAD(graveyard);
1801 /* extract submounts of 'mountpoint' from the expiration list */
1802 while (select_submounts(mnt, &graveyard)) {
1803 while (!list_empty(&graveyard)) {
1804 m = list_first_entry(&graveyard, struct vfsmount,
1806 touch_mnt_namespace(m->mnt_ns);
1807 umount_tree(m, 1, umounts);
1813 * Some copy_from_user() implementations do not return the exact number of
1814 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1815 * Note that this function differs from copy_from_user() in that it will oops
1816 * on bad values of `to', rather than returning a short copy.
1818 static long exact_copy_from_user(void *to, const void __user * from,
1822 const char __user *f = from;
1825 if (!access_ok(VERIFY_READ, from, n))
1829 if (__get_user(c, f)) {
1840 int copy_mount_options(const void __user * data, unsigned long *where)
1850 if (!(page = __get_free_page(GFP_KERNEL)))
1853 /* We only care that *some* data at the address the user
1854 * gave us is valid. Just in case, we'll zero
1855 * the remainder of the page.
1857 /* copy_from_user cannot cross TASK_SIZE ! */
1858 size = TASK_SIZE - (unsigned long)data;
1859 if (size > PAGE_SIZE)
1862 i = size - exact_copy_from_user((void *)page, data, size);
1868 memset((char *)page + i, 0, PAGE_SIZE - i);
1874 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1875 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1877 * data is a (void *) that can point to any structure up to
1878 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1879 * information (or be NULL).
1881 * Pre-0.97 versions of mount() didn't have a flags word.
1882 * When the flags word was introduced its top half was required
1883 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1884 * Therefore, if this magic number is present, it carries no information
1885 * and must be discarded.
1887 long do_mount(char *dev_name, char *dir_name, char *type_page,
1888 unsigned long flags, void *data_page)
1895 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1896 flags &= ~MS_MGC_MSK;
1898 /* Basic sanity checks */
1900 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1902 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1906 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1908 /* Default to relatime unless overriden */
1909 if (!(flags & MS_NOATIME))
1910 mnt_flags |= MNT_RELATIME;
1912 /* Separate the per-mountpoint flags */
1913 if (flags & MS_NOSUID)
1914 mnt_flags |= MNT_NOSUID;
1915 if (flags & MS_NODEV)
1916 mnt_flags |= MNT_NODEV;
1917 if (flags & MS_NOEXEC)
1918 mnt_flags |= MNT_NOEXEC;
1919 if (flags & MS_NOATIME)
1920 mnt_flags |= MNT_NOATIME;
1921 if (flags & MS_NODIRATIME)
1922 mnt_flags |= MNT_NODIRATIME;
1923 if (flags & MS_STRICTATIME)
1924 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1925 if (flags & MS_RDONLY)
1926 mnt_flags |= MNT_READONLY;
1928 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1929 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1932 /* ... and get the mountpoint */
1933 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1937 retval = security_sb_mount(dev_name, &path,
1938 type_page, flags, data_page);
1942 if (flags & MS_REMOUNT)
1943 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1945 else if (flags & MS_BIND)
1946 retval = do_loopback(&path, dev_name, flags & MS_REC);
1947 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1948 retval = do_change_type(&path, flags);
1949 else if (flags & MS_MOVE)
1950 retval = do_move_mount(&path, dev_name);
1952 retval = do_new_mount(&path, type_page, flags, mnt_flags,
1953 dev_name, data_page);
1959 static struct mnt_namespace *alloc_mnt_ns(void)
1961 struct mnt_namespace *new_ns;
1963 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1965 return ERR_PTR(-ENOMEM);
1966 atomic_set(&new_ns->count, 1);
1967 new_ns->root = NULL;
1968 INIT_LIST_HEAD(&new_ns->list);
1969 init_waitqueue_head(&new_ns->poll);
1975 * Allocate a new namespace structure and populate it with contents
1976 * copied from the namespace of the passed in task structure.
1978 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1979 struct fs_struct *fs)
1981 struct mnt_namespace *new_ns;
1982 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1983 struct vfsmount *p, *q;
1985 new_ns = alloc_mnt_ns();
1989 down_write(&namespace_sem);
1990 /* First pass: copy the tree topology */
1991 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1992 CL_COPY_ALL | CL_EXPIRE);
1993 if (!new_ns->root) {
1994 up_write(&namespace_sem);
1996 return ERR_PTR(-ENOMEM);
1998 spin_lock(&vfsmount_lock);
1999 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2000 spin_unlock(&vfsmount_lock);
2003 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2004 * as belonging to new namespace. We have already acquired a private
2005 * fs_struct, so tsk->fs->lock is not needed.
2012 if (p == fs->root.mnt) {
2014 fs->root.mnt = mntget(q);
2016 if (p == fs->pwd.mnt) {
2018 fs->pwd.mnt = mntget(q);
2021 p = next_mnt(p, mnt_ns->root);
2022 q = next_mnt(q, new_ns->root);
2024 up_write(&namespace_sem);
2034 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2035 struct fs_struct *new_fs)
2037 struct mnt_namespace *new_ns;
2042 if (!(flags & CLONE_NEWNS))
2045 new_ns = dup_mnt_ns(ns, new_fs);
2052 * create_mnt_ns - creates a private namespace and adds a root filesystem
2053 * @mnt: pointer to the new root filesystem mountpoint
2055 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
2057 struct mnt_namespace *new_ns;
2059 new_ns = alloc_mnt_ns();
2060 if (!IS_ERR(new_ns)) {
2061 mnt->mnt_ns = new_ns;
2063 list_add(&new_ns->list, &new_ns->root->mnt_list);
2067 EXPORT_SYMBOL(create_mnt_ns);
2069 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2070 char __user *, type, unsigned long, flags, void __user *, data)
2073 unsigned long data_page;
2074 unsigned long type_page;
2075 unsigned long dev_page;
2078 retval = copy_mount_options(type, &type_page);
2082 dir_page = getname(dir_name);
2083 retval = PTR_ERR(dir_page);
2084 if (IS_ERR(dir_page))
2087 retval = copy_mount_options(dev_name, &dev_page);
2091 retval = copy_mount_options(data, &data_page);
2095 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2096 flags, (void *)data_page);
2097 free_page(data_page);
2100 free_page(dev_page);
2104 free_page(type_page);
2109 * pivot_root Semantics:
2110 * Moves the root file system of the current process to the directory put_old,
2111 * makes new_root as the new root file system of the current process, and sets
2112 * root/cwd of all processes which had them on the current root to new_root.
2115 * The new_root and put_old must be directories, and must not be on the
2116 * same file system as the current process root. The put_old must be
2117 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2118 * pointed to by put_old must yield the same directory as new_root. No other
2119 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2121 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2122 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2123 * in this situation.
2126 * - we don't move root/cwd if they are not at the root (reason: if something
2127 * cared enough to change them, it's probably wrong to force them elsewhere)
2128 * - it's okay to pick a root that isn't the root of a file system, e.g.
2129 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2130 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2133 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2134 const char __user *, put_old)
2136 struct vfsmount *tmp;
2137 struct path new, old, parent_path, root_parent, root;
2140 if (!capable(CAP_SYS_ADMIN))
2143 error = user_path_dir(new_root, &new);
2147 if (!check_mnt(new.mnt))
2150 error = user_path_dir(put_old, &old);
2154 error = security_sb_pivotroot(&old, &new);
2160 read_lock(¤t->fs->lock);
2161 root = current->fs->root;
2162 path_get(¤t->fs->root);
2163 read_unlock(¤t->fs->lock);
2164 down_write(&namespace_sem);
2165 mutex_lock(&old.dentry->d_inode->i_mutex);
2167 if (IS_MNT_SHARED(old.mnt) ||
2168 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2169 IS_MNT_SHARED(root.mnt->mnt_parent))
2171 if (!check_mnt(root.mnt))
2174 if (IS_DEADDIR(new.dentry->d_inode))
2176 if (d_unlinked(new.dentry))
2178 if (d_unlinked(old.dentry))
2181 if (new.mnt == root.mnt ||
2182 old.mnt == root.mnt)
2183 goto out2; /* loop, on the same file system */
2185 if (root.mnt->mnt_root != root.dentry)
2186 goto out2; /* not a mountpoint */
2187 if (root.mnt->mnt_parent == root.mnt)
2188 goto out2; /* not attached */
2189 if (new.mnt->mnt_root != new.dentry)
2190 goto out2; /* not a mountpoint */
2191 if (new.mnt->mnt_parent == new.mnt)
2192 goto out2; /* not attached */
2193 /* make sure we can reach put_old from new_root */
2195 spin_lock(&vfsmount_lock);
2196 if (tmp != new.mnt) {
2198 if (tmp->mnt_parent == tmp)
2199 goto out3; /* already mounted on put_old */
2200 if (tmp->mnt_parent == new.mnt)
2202 tmp = tmp->mnt_parent;
2204 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2206 } else if (!is_subdir(old.dentry, new.dentry))
2208 detach_mnt(new.mnt, &parent_path);
2209 detach_mnt(root.mnt, &root_parent);
2210 /* mount old root on put_old */
2211 attach_mnt(root.mnt, &old);
2212 /* mount new_root on / */
2213 attach_mnt(new.mnt, &root_parent);
2214 touch_mnt_namespace(current->nsproxy->mnt_ns);
2215 spin_unlock(&vfsmount_lock);
2216 chroot_fs_refs(&root, &new);
2217 security_sb_post_pivotroot(&root, &new);
2219 path_put(&root_parent);
2220 path_put(&parent_path);
2222 mutex_unlock(&old.dentry->d_inode->i_mutex);
2223 up_write(&namespace_sem);
2231 spin_unlock(&vfsmount_lock);
2235 static void __init init_mount_tree(void)
2237 struct vfsmount *mnt;
2238 struct mnt_namespace *ns;
2241 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2243 panic("Can't create rootfs");
2244 ns = create_mnt_ns(mnt);
2246 panic("Can't allocate initial namespace");
2248 init_task.nsproxy->mnt_ns = ns;
2251 root.mnt = ns->root;
2252 root.dentry = ns->root->mnt_root;
2254 set_fs_pwd(current->fs, &root);
2255 set_fs_root(current->fs, &root);
2258 void __init mnt_init(void)
2263 init_rwsem(&namespace_sem);
2265 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2266 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2268 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2270 if (!mount_hashtable)
2271 panic("Failed to allocate mount hash table\n");
2273 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2275 for (u = 0; u < HASH_SIZE; u++)
2276 INIT_LIST_HEAD(&mount_hashtable[u]);
2280 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2282 fs_kobj = kobject_create_and_add("fs", NULL);
2284 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2289 void put_mnt_ns(struct mnt_namespace *ns)
2291 struct vfsmount *root;
2292 LIST_HEAD(umount_list);
2294 if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock))
2298 spin_unlock(&vfsmount_lock);
2299 down_write(&namespace_sem);
2300 spin_lock(&vfsmount_lock);
2301 umount_tree(root, 0, &umount_list);
2302 spin_unlock(&vfsmount_lock);
2303 up_write(&namespace_sem);
2304 release_mounts(&umount_list);
2307 EXPORT_SYMBOL(put_mnt_ns);