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/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <linux/log2.h>
29 #include <linux/idr.h>
30 #include <linux/fs_struct.h>
31 #include <asm/uaccess.h>
32 #include <asm/unistd.h>
36 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37 #define HASH_SIZE (1UL << HASH_SHIFT)
39 /* spinlock for vfsmount related operations, inplace of dcache_lock */
40 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43 static DEFINE_IDA(mnt_id_ida);
44 static DEFINE_IDA(mnt_group_ida);
46 static struct list_head *mount_hashtable __read_mostly;
47 static struct kmem_cache *mnt_cache __read_mostly;
48 static struct rw_semaphore namespace_sem;
51 struct kobject *fs_kobj;
52 EXPORT_SYMBOL_GPL(fs_kobj);
54 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
56 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
58 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
62 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
64 /* allocation is serialized by namespace_sem */
65 static int mnt_alloc_id(struct vfsmount *mnt)
70 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
71 spin_lock(&vfsmount_lock);
72 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
73 spin_unlock(&vfsmount_lock);
80 static void mnt_free_id(struct vfsmount *mnt)
82 spin_lock(&vfsmount_lock);
83 ida_remove(&mnt_id_ida, mnt->mnt_id);
84 spin_unlock(&vfsmount_lock);
88 * Allocate a new peer group ID
90 * mnt_group_ida is protected by namespace_sem
92 static int mnt_alloc_group_id(struct vfsmount *mnt)
94 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
97 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
101 * Release a peer group ID
103 void mnt_release_group_id(struct vfsmount *mnt)
105 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
106 mnt->mnt_group_id = 0;
109 struct vfsmount *alloc_vfsmnt(const char *name)
111 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
115 err = mnt_alloc_id(mnt);
120 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
121 if (!mnt->mnt_devname)
125 atomic_set(&mnt->mnt_count, 1);
126 INIT_LIST_HEAD(&mnt->mnt_hash);
127 INIT_LIST_HEAD(&mnt->mnt_child);
128 INIT_LIST_HEAD(&mnt->mnt_mounts);
129 INIT_LIST_HEAD(&mnt->mnt_list);
130 INIT_LIST_HEAD(&mnt->mnt_expire);
131 INIT_LIST_HEAD(&mnt->mnt_share);
132 INIT_LIST_HEAD(&mnt->mnt_slave_list);
133 INIT_LIST_HEAD(&mnt->mnt_slave);
135 mnt->mnt_writers = alloc_percpu(int);
136 if (!mnt->mnt_writers)
137 goto out_free_devname;
139 mnt->mnt_writers = 0;
146 kfree(mnt->mnt_devname);
151 kmem_cache_free(mnt_cache, mnt);
156 * Most r/o checks on a fs are for operations that take
157 * discrete amounts of time, like a write() or unlink().
158 * We must keep track of when those operations start
159 * (for permission checks) and when they end, so that
160 * we can determine when writes are able to occur to
164 * __mnt_is_readonly: check whether a mount is read-only
165 * @mnt: the mount to check for its write status
167 * This shouldn't be used directly ouside of the VFS.
168 * It does not guarantee that the filesystem will stay
169 * r/w, just that it is right *now*. This can not and
170 * should not be used in place of IS_RDONLY(inode).
171 * mnt_want/drop_write() will _keep_ the filesystem
174 int __mnt_is_readonly(struct vfsmount *mnt)
176 if (mnt->mnt_flags & MNT_READONLY)
178 if (mnt->mnt_sb->s_flags & MS_RDONLY)
182 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
184 static inline void inc_mnt_writers(struct vfsmount *mnt)
187 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
193 static inline void dec_mnt_writers(struct vfsmount *mnt)
196 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
202 static unsigned int count_mnt_writers(struct vfsmount *mnt)
205 unsigned int count = 0;
208 for_each_possible_cpu(cpu) {
209 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
214 return mnt->mnt_writers;
219 * Most r/o checks on a fs are for operations that take
220 * discrete amounts of time, like a write() or unlink().
221 * We must keep track of when those operations start
222 * (for permission checks) and when they end, so that
223 * we can determine when writes are able to occur to
227 * mnt_want_write - get write access to a mount
228 * @mnt: the mount on which to take a write
230 * This tells the low-level filesystem that a write is
231 * about to be performed to it, and makes sure that
232 * writes are allowed before returning success. When
233 * the write operation is finished, mnt_drop_write()
234 * must be called. This is effectively a refcount.
236 int mnt_want_write(struct vfsmount *mnt)
241 inc_mnt_writers(mnt);
243 * The store to inc_mnt_writers must be visible before we pass
244 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
245 * incremented count after it has set MNT_WRITE_HOLD.
248 while (mnt->mnt_flags & MNT_WRITE_HOLD)
251 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
252 * be set to match its requirements. So we must not load that until
253 * MNT_WRITE_HOLD is cleared.
256 if (__mnt_is_readonly(mnt)) {
257 dec_mnt_writers(mnt);
265 EXPORT_SYMBOL_GPL(mnt_want_write);
268 * mnt_drop_write - give up write access to a mount
269 * @mnt: the mount on which to give up write access
271 * Tells the low-level filesystem that we are done
272 * performing writes to it. Must be matched with
273 * mnt_want_write() call above.
275 void mnt_drop_write(struct vfsmount *mnt)
278 dec_mnt_writers(mnt);
281 EXPORT_SYMBOL_GPL(mnt_drop_write);
283 static int mnt_make_readonly(struct vfsmount *mnt)
287 spin_lock(&vfsmount_lock);
288 mnt->mnt_flags |= MNT_WRITE_HOLD;
290 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
291 * should be visible before we do.
296 * With writers on hold, if this value is zero, then there are
297 * definitely no active writers (although held writers may subsequently
298 * increment the count, they'll have to wait, and decrement it after
299 * seeing MNT_READONLY).
301 * It is OK to have counter incremented on one CPU and decremented on
302 * another: the sum will add up correctly. The danger would be when we
303 * sum up each counter, if we read a counter before it is incremented,
304 * but then read another CPU's count which it has been subsequently
305 * decremented from -- we would see more decrements than we should.
306 * MNT_WRITE_HOLD protects against this scenario, because
307 * mnt_want_write first increments count, then smp_mb, then spins on
308 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
309 * we're counting up here.
311 if (count_mnt_writers(mnt) > 0)
314 mnt->mnt_flags |= MNT_READONLY;
316 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
317 * that become unheld will see MNT_READONLY.
320 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
321 spin_unlock(&vfsmount_lock);
325 static void __mnt_unmake_readonly(struct vfsmount *mnt)
327 spin_lock(&vfsmount_lock);
328 mnt->mnt_flags &= ~MNT_READONLY;
329 spin_unlock(&vfsmount_lock);
332 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
335 mnt->mnt_root = dget(sb->s_root);
338 EXPORT_SYMBOL(simple_set_mnt);
340 void free_vfsmnt(struct vfsmount *mnt)
342 kfree(mnt->mnt_devname);
345 free_percpu(mnt->mnt_writers);
347 kmem_cache_free(mnt_cache, mnt);
351 * find the first or last mount at @dentry on vfsmount @mnt depending on
352 * @dir. If @dir is set return the first mount else return the last mount.
354 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
357 struct list_head *head = mount_hashtable + hash(mnt, dentry);
358 struct list_head *tmp = head;
359 struct vfsmount *p, *found = NULL;
362 tmp = dir ? tmp->next : tmp->prev;
366 p = list_entry(tmp, struct vfsmount, mnt_hash);
367 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
376 * lookup_mnt increments the ref count before returning
377 * the vfsmount struct.
379 struct vfsmount *lookup_mnt(struct path *path)
381 struct vfsmount *child_mnt;
382 spin_lock(&vfsmount_lock);
383 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
385 spin_unlock(&vfsmount_lock);
389 static inline int check_mnt(struct vfsmount *mnt)
391 return mnt->mnt_ns == current->nsproxy->mnt_ns;
394 static void touch_mnt_namespace(struct mnt_namespace *ns)
398 wake_up_interruptible(&ns->poll);
402 static void __touch_mnt_namespace(struct mnt_namespace *ns)
404 if (ns && ns->event != event) {
406 wake_up_interruptible(&ns->poll);
410 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
412 old_path->dentry = mnt->mnt_mountpoint;
413 old_path->mnt = mnt->mnt_parent;
414 mnt->mnt_parent = mnt;
415 mnt->mnt_mountpoint = mnt->mnt_root;
416 list_del_init(&mnt->mnt_child);
417 list_del_init(&mnt->mnt_hash);
418 old_path->dentry->d_mounted--;
421 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
422 struct vfsmount *child_mnt)
424 child_mnt->mnt_parent = mntget(mnt);
425 child_mnt->mnt_mountpoint = dget(dentry);
429 static void attach_mnt(struct vfsmount *mnt, struct path *path)
431 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
432 list_add_tail(&mnt->mnt_hash, mount_hashtable +
433 hash(path->mnt, path->dentry));
434 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
438 * the caller must hold vfsmount_lock
440 static void commit_tree(struct vfsmount *mnt)
442 struct vfsmount *parent = mnt->mnt_parent;
445 struct mnt_namespace *n = parent->mnt_ns;
447 BUG_ON(parent == mnt);
449 list_add_tail(&head, &mnt->mnt_list);
450 list_for_each_entry(m, &head, mnt_list)
452 list_splice(&head, n->list.prev);
454 list_add_tail(&mnt->mnt_hash, mount_hashtable +
455 hash(parent, mnt->mnt_mountpoint));
456 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
457 touch_mnt_namespace(n);
460 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
462 struct list_head *next = p->mnt_mounts.next;
463 if (next == &p->mnt_mounts) {
467 next = p->mnt_child.next;
468 if (next != &p->mnt_parent->mnt_mounts)
473 return list_entry(next, struct vfsmount, mnt_child);
476 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
478 struct list_head *prev = p->mnt_mounts.prev;
479 while (prev != &p->mnt_mounts) {
480 p = list_entry(prev, struct vfsmount, mnt_child);
481 prev = p->mnt_mounts.prev;
486 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
489 struct super_block *sb = old->mnt_sb;
490 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
493 if (flag & (CL_SLAVE | CL_PRIVATE))
494 mnt->mnt_group_id = 0; /* not a peer of original */
496 mnt->mnt_group_id = old->mnt_group_id;
498 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
499 int err = mnt_alloc_group_id(mnt);
504 mnt->mnt_flags = old->mnt_flags;
505 atomic_inc(&sb->s_active);
507 mnt->mnt_root = dget(root);
508 mnt->mnt_mountpoint = mnt->mnt_root;
509 mnt->mnt_parent = mnt;
511 if (flag & CL_SLAVE) {
512 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
513 mnt->mnt_master = old;
514 CLEAR_MNT_SHARED(mnt);
515 } else if (!(flag & CL_PRIVATE)) {
516 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
517 list_add(&mnt->mnt_share, &old->mnt_share);
518 if (IS_MNT_SLAVE(old))
519 list_add(&mnt->mnt_slave, &old->mnt_slave);
520 mnt->mnt_master = old->mnt_master;
522 if (flag & CL_MAKE_SHARED)
525 /* stick the duplicate mount on the same expiry list
526 * as the original if that was on one */
527 if (flag & CL_EXPIRE) {
528 if (!list_empty(&old->mnt_expire))
529 list_add(&mnt->mnt_expire, &old->mnt_expire);
539 static inline void __mntput(struct vfsmount *mnt)
541 struct super_block *sb = mnt->mnt_sb;
543 * This probably indicates that somebody messed
544 * up a mnt_want/drop_write() pair. If this
545 * happens, the filesystem was probably unable
546 * to make r/w->r/o transitions.
549 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
550 * provides barriers, so count_mnt_writers() below is safe. AV
552 WARN_ON(count_mnt_writers(mnt));
555 deactivate_super(sb);
558 void mntput_no_expire(struct vfsmount *mnt)
561 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
562 if (likely(!mnt->mnt_pinned)) {
563 spin_unlock(&vfsmount_lock);
567 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
569 spin_unlock(&vfsmount_lock);
570 acct_auto_close_mnt(mnt);
571 security_sb_umount_close(mnt);
576 EXPORT_SYMBOL(mntput_no_expire);
578 void mnt_pin(struct vfsmount *mnt)
580 spin_lock(&vfsmount_lock);
582 spin_unlock(&vfsmount_lock);
585 EXPORT_SYMBOL(mnt_pin);
587 void mnt_unpin(struct vfsmount *mnt)
589 spin_lock(&vfsmount_lock);
590 if (mnt->mnt_pinned) {
591 atomic_inc(&mnt->mnt_count);
594 spin_unlock(&vfsmount_lock);
597 EXPORT_SYMBOL(mnt_unpin);
599 static inline void mangle(struct seq_file *m, const char *s)
601 seq_escape(m, s, " \t\n\\");
605 * Simple .show_options callback for filesystems which don't want to
606 * implement more complex mount option showing.
608 * See also save_mount_options().
610 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
615 options = rcu_dereference(mnt->mnt_sb->s_options);
617 if (options != NULL && options[0]) {
625 EXPORT_SYMBOL(generic_show_options);
628 * If filesystem uses generic_show_options(), this function should be
629 * called from the fill_super() callback.
631 * The .remount_fs callback usually needs to be handled in a special
632 * way, to make sure, that previous options are not overwritten if the
635 * Also note, that if the filesystem's .remount_fs function doesn't
636 * reset all options to their default value, but changes only newly
637 * given options, then the displayed options will not reflect reality
640 void save_mount_options(struct super_block *sb, char *options)
642 BUG_ON(sb->s_options);
643 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
645 EXPORT_SYMBOL(save_mount_options);
647 void replace_mount_options(struct super_block *sb, char *options)
649 char *old = sb->s_options;
650 rcu_assign_pointer(sb->s_options, options);
656 EXPORT_SYMBOL(replace_mount_options);
658 #ifdef CONFIG_PROC_FS
660 static void *m_start(struct seq_file *m, loff_t *pos)
662 struct proc_mounts *p = m->private;
664 down_read(&namespace_sem);
665 return seq_list_start(&p->ns->list, *pos);
668 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
670 struct proc_mounts *p = m->private;
672 return seq_list_next(v, &p->ns->list, pos);
675 static void m_stop(struct seq_file *m, void *v)
677 up_read(&namespace_sem);
680 struct proc_fs_info {
685 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
687 static const struct proc_fs_info fs_info[] = {
688 { MS_SYNCHRONOUS, ",sync" },
689 { MS_DIRSYNC, ",dirsync" },
690 { MS_MANDLOCK, ",mand" },
693 const struct proc_fs_info *fs_infop;
695 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
696 if (sb->s_flags & fs_infop->flag)
697 seq_puts(m, fs_infop->str);
700 return security_sb_show_options(m, sb);
703 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
705 static const struct proc_fs_info mnt_info[] = {
706 { MNT_NOSUID, ",nosuid" },
707 { MNT_NODEV, ",nodev" },
708 { MNT_NOEXEC, ",noexec" },
709 { MNT_NOATIME, ",noatime" },
710 { MNT_NODIRATIME, ",nodiratime" },
711 { MNT_RELATIME, ",relatime" },
712 { MNT_STRICTATIME, ",strictatime" },
715 const struct proc_fs_info *fs_infop;
717 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
718 if (mnt->mnt_flags & fs_infop->flag)
719 seq_puts(m, fs_infop->str);
723 static void show_type(struct seq_file *m, struct super_block *sb)
725 mangle(m, sb->s_type->name);
726 if (sb->s_subtype && sb->s_subtype[0]) {
728 mangle(m, sb->s_subtype);
732 static int show_vfsmnt(struct seq_file *m, void *v)
734 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
736 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
738 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
740 seq_path(m, &mnt_path, " \t\n\\");
742 show_type(m, mnt->mnt_sb);
743 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
744 err = show_sb_opts(m, mnt->mnt_sb);
747 show_mnt_opts(m, mnt);
748 if (mnt->mnt_sb->s_op->show_options)
749 err = mnt->mnt_sb->s_op->show_options(m, mnt);
750 seq_puts(m, " 0 0\n");
755 const struct seq_operations mounts_op = {
762 static int show_mountinfo(struct seq_file *m, void *v)
764 struct proc_mounts *p = m->private;
765 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
766 struct super_block *sb = mnt->mnt_sb;
767 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
768 struct path root = p->root;
771 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
772 MAJOR(sb->s_dev), MINOR(sb->s_dev));
773 seq_dentry(m, mnt->mnt_root, " \t\n\\");
775 seq_path_root(m, &mnt_path, &root, " \t\n\\");
776 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
778 * Mountpoint is outside root, discard that one. Ugly,
779 * but less so than trying to do that in iterator in a
780 * race-free way (due to renames).
784 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
785 show_mnt_opts(m, mnt);
787 /* Tagged fields ("foo:X" or "bar") */
788 if (IS_MNT_SHARED(mnt))
789 seq_printf(m, " shared:%i", mnt->mnt_group_id);
790 if (IS_MNT_SLAVE(mnt)) {
791 int master = mnt->mnt_master->mnt_group_id;
792 int dom = get_dominating_id(mnt, &p->root);
793 seq_printf(m, " master:%i", master);
794 if (dom && dom != master)
795 seq_printf(m, " propagate_from:%i", dom);
797 if (IS_MNT_UNBINDABLE(mnt))
798 seq_puts(m, " unbindable");
800 /* Filesystem specific data */
804 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
805 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
806 err = show_sb_opts(m, sb);
809 if (sb->s_op->show_options)
810 err = sb->s_op->show_options(m, mnt);
816 const struct seq_operations mountinfo_op = {
820 .show = show_mountinfo,
823 static int show_vfsstat(struct seq_file *m, void *v)
825 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
826 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
830 if (mnt->mnt_devname) {
831 seq_puts(m, "device ");
832 mangle(m, mnt->mnt_devname);
834 seq_puts(m, "no device");
837 seq_puts(m, " mounted on ");
838 seq_path(m, &mnt_path, " \t\n\\");
841 /* file system type */
842 seq_puts(m, "with fstype ");
843 show_type(m, mnt->mnt_sb);
845 /* optional statistics */
846 if (mnt->mnt_sb->s_op->show_stats) {
848 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
855 const struct seq_operations mountstats_op = {
859 .show = show_vfsstat,
861 #endif /* CONFIG_PROC_FS */
864 * may_umount_tree - check if a mount tree is busy
865 * @mnt: root of mount tree
867 * This is called to check if a tree of mounts has any
868 * open files, pwds, chroots or sub mounts that are
871 int may_umount_tree(struct vfsmount *mnt)
874 int minimum_refs = 0;
877 spin_lock(&vfsmount_lock);
878 for (p = mnt; p; p = next_mnt(p, mnt)) {
879 actual_refs += atomic_read(&p->mnt_count);
882 spin_unlock(&vfsmount_lock);
884 if (actual_refs > minimum_refs)
890 EXPORT_SYMBOL(may_umount_tree);
893 * may_umount - check if a mount point is busy
894 * @mnt: root of mount
896 * This is called to check if a mount point has any
897 * open files, pwds, chroots or sub mounts. If the
898 * mount has sub mounts this will return busy
899 * regardless of whether the sub mounts are busy.
901 * Doesn't take quota and stuff into account. IOW, in some cases it will
902 * give false negatives. The main reason why it's here is that we need
903 * a non-destructive way to look for easily umountable filesystems.
905 int may_umount(struct vfsmount *mnt)
908 spin_lock(&vfsmount_lock);
909 if (propagate_mount_busy(mnt, 2))
911 spin_unlock(&vfsmount_lock);
915 EXPORT_SYMBOL(may_umount);
917 void release_mounts(struct list_head *head)
919 struct vfsmount *mnt;
920 while (!list_empty(head)) {
921 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
922 list_del_init(&mnt->mnt_hash);
923 if (mnt->mnt_parent != mnt) {
924 struct dentry *dentry;
926 spin_lock(&vfsmount_lock);
927 dentry = mnt->mnt_mountpoint;
929 mnt->mnt_mountpoint = mnt->mnt_root;
930 mnt->mnt_parent = mnt;
932 spin_unlock(&vfsmount_lock);
940 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
944 for (p = mnt; p; p = next_mnt(p, mnt))
945 list_move(&p->mnt_hash, kill);
948 propagate_umount(kill);
950 list_for_each_entry(p, kill, mnt_hash) {
951 list_del_init(&p->mnt_expire);
952 list_del_init(&p->mnt_list);
953 __touch_mnt_namespace(p->mnt_ns);
955 list_del_init(&p->mnt_child);
956 if (p->mnt_parent != p) {
957 p->mnt_parent->mnt_ghosts++;
958 p->mnt_mountpoint->d_mounted--;
960 change_mnt_propagation(p, MS_PRIVATE);
964 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
966 static int do_umount(struct vfsmount *mnt, int flags)
968 struct super_block *sb = mnt->mnt_sb;
970 LIST_HEAD(umount_list);
972 retval = security_sb_umount(mnt, flags);
977 * Allow userspace to request a mountpoint be expired rather than
978 * unmounting unconditionally. Unmount only happens if:
979 * (1) the mark is already set (the mark is cleared by mntput())
980 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
982 if (flags & MNT_EXPIRE) {
983 if (mnt == current->fs->root.mnt ||
984 flags & (MNT_FORCE | MNT_DETACH))
987 if (atomic_read(&mnt->mnt_count) != 2)
990 if (!xchg(&mnt->mnt_expiry_mark, 1))
995 * If we may have to abort operations to get out of this
996 * mount, and they will themselves hold resources we must
997 * allow the fs to do things. In the Unix tradition of
998 * 'Gee thats tricky lets do it in userspace' the umount_begin
999 * might fail to complete on the first run through as other tasks
1000 * must return, and the like. Thats for the mount program to worry
1001 * about for the moment.
1004 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1005 sb->s_op->umount_begin(sb);
1009 * No sense to grab the lock for this test, but test itself looks
1010 * somewhat bogus. Suggestions for better replacement?
1011 * Ho-hum... In principle, we might treat that as umount + switch
1012 * to rootfs. GC would eventually take care of the old vfsmount.
1013 * Actually it makes sense, especially if rootfs would contain a
1014 * /reboot - static binary that would close all descriptors and
1015 * call reboot(9). Then init(8) could umount root and exec /reboot.
1017 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1019 * Special case for "unmounting" root ...
1020 * we just try to remount it readonly.
1022 down_write(&sb->s_umount);
1023 if (!(sb->s_flags & MS_RDONLY)) {
1025 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1028 up_write(&sb->s_umount);
1032 down_write(&namespace_sem);
1033 spin_lock(&vfsmount_lock);
1036 if (!(flags & MNT_DETACH))
1037 shrink_submounts(mnt, &umount_list);
1040 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1041 if (!list_empty(&mnt->mnt_list))
1042 umount_tree(mnt, 1, &umount_list);
1045 spin_unlock(&vfsmount_lock);
1047 security_sb_umount_busy(mnt);
1048 up_write(&namespace_sem);
1049 release_mounts(&umount_list);
1054 * Now umount can handle mount points as well as block devices.
1055 * This is important for filesystems which use unnamed block devices.
1057 * We now support a flag for forced unmount like the other 'big iron'
1058 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1061 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1066 retval = user_path(name, &path);
1070 if (path.dentry != path.mnt->mnt_root)
1072 if (!check_mnt(path.mnt))
1076 if (!capable(CAP_SYS_ADMIN))
1079 retval = do_umount(path.mnt, flags);
1081 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1083 mntput_no_expire(path.mnt);
1088 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1091 * The 2.0 compatible umount. No flags.
1093 SYSCALL_DEFINE1(oldumount, char __user *, name)
1095 return sys_umount(name, 0);
1100 static int mount_is_safe(struct path *path)
1102 if (capable(CAP_SYS_ADMIN))
1106 if (S_ISLNK(path->dentry->d_inode->i_mode))
1108 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1109 if (current_uid() != path->dentry->d_inode->i_uid)
1112 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1118 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1121 struct vfsmount *res, *p, *q, *r, *s;
1124 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1127 res = q = clone_mnt(mnt, dentry, flag);
1130 q->mnt_mountpoint = mnt->mnt_mountpoint;
1133 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1134 if (!is_subdir(r->mnt_mountpoint, dentry))
1137 for (s = r; s; s = next_mnt(s, r)) {
1138 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1139 s = skip_mnt_tree(s);
1142 while (p != s->mnt_parent) {
1148 path.dentry = p->mnt_mountpoint;
1149 q = clone_mnt(p, p->mnt_root, flag);
1152 spin_lock(&vfsmount_lock);
1153 list_add_tail(&q->mnt_list, &res->mnt_list);
1154 attach_mnt(q, &path);
1155 spin_unlock(&vfsmount_lock);
1161 LIST_HEAD(umount_list);
1162 spin_lock(&vfsmount_lock);
1163 umount_tree(res, 0, &umount_list);
1164 spin_unlock(&vfsmount_lock);
1165 release_mounts(&umount_list);
1170 struct vfsmount *collect_mounts(struct path *path)
1172 struct vfsmount *tree;
1173 down_write(&namespace_sem);
1174 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
1175 up_write(&namespace_sem);
1179 void drop_collected_mounts(struct vfsmount *mnt)
1181 LIST_HEAD(umount_list);
1182 down_write(&namespace_sem);
1183 spin_lock(&vfsmount_lock);
1184 umount_tree(mnt, 0, &umount_list);
1185 spin_unlock(&vfsmount_lock);
1186 up_write(&namespace_sem);
1187 release_mounts(&umount_list);
1190 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1194 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1195 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1196 mnt_release_group_id(p);
1200 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1204 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1205 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1206 int err = mnt_alloc_group_id(p);
1208 cleanup_group_ids(mnt, p);
1218 * @source_mnt : mount tree to be attached
1219 * @nd : place the mount tree @source_mnt is attached
1220 * @parent_nd : if non-null, detach the source_mnt from its parent and
1221 * store the parent mount and mountpoint dentry.
1222 * (done when source_mnt is moved)
1224 * NOTE: in the table below explains the semantics when a source mount
1225 * of a given type is attached to a destination mount of a given type.
1226 * ---------------------------------------------------------------------------
1227 * | BIND MOUNT OPERATION |
1228 * |**************************************************************************
1229 * | source-->| shared | private | slave | unbindable |
1233 * |**************************************************************************
1234 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1236 * |non-shared| shared (+) | private | slave (*) | invalid |
1237 * ***************************************************************************
1238 * A bind operation clones the source mount and mounts the clone on the
1239 * destination mount.
1241 * (++) the cloned mount is propagated to all the mounts in the propagation
1242 * tree of the destination mount and the cloned mount is added to
1243 * the peer group of the source mount.
1244 * (+) the cloned mount is created under the destination mount and is marked
1245 * as shared. The cloned mount is added to the peer group of the source
1247 * (+++) the mount is propagated to all the mounts in the propagation tree
1248 * of the destination mount and the cloned mount is made slave
1249 * of the same master as that of the source mount. The cloned mount
1250 * is marked as 'shared and slave'.
1251 * (*) the cloned mount is made a slave of the same master as that of the
1254 * ---------------------------------------------------------------------------
1255 * | MOVE MOUNT OPERATION |
1256 * |**************************************************************************
1257 * | source-->| shared | private | slave | unbindable |
1261 * |**************************************************************************
1262 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1264 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1265 * ***************************************************************************
1267 * (+) the mount is moved to the destination. And is then propagated to
1268 * all the mounts in the propagation tree of the destination mount.
1269 * (+*) the mount is moved to the destination.
1270 * (+++) the mount is moved to the destination and is then propagated to
1271 * all the mounts belonging to the destination mount's propagation tree.
1272 * the mount is marked as 'shared and slave'.
1273 * (*) the mount continues to be a slave at the new location.
1275 * if the source mount is a tree, the operations explained above is
1276 * applied to each mount in the tree.
1277 * Must be called without spinlocks held, since this function can sleep
1280 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1281 struct path *path, struct path *parent_path)
1283 LIST_HEAD(tree_list);
1284 struct vfsmount *dest_mnt = path->mnt;
1285 struct dentry *dest_dentry = path->dentry;
1286 struct vfsmount *child, *p;
1289 if (IS_MNT_SHARED(dest_mnt)) {
1290 err = invent_group_ids(source_mnt, true);
1294 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1296 goto out_cleanup_ids;
1298 if (IS_MNT_SHARED(dest_mnt)) {
1299 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1303 spin_lock(&vfsmount_lock);
1305 detach_mnt(source_mnt, parent_path);
1306 attach_mnt(source_mnt, path);
1307 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1309 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1310 commit_tree(source_mnt);
1313 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1314 list_del_init(&child->mnt_hash);
1317 spin_unlock(&vfsmount_lock);
1321 if (IS_MNT_SHARED(dest_mnt))
1322 cleanup_group_ids(source_mnt, NULL);
1327 static int graft_tree(struct vfsmount *mnt, struct path *path)
1330 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1333 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1334 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1338 mutex_lock(&path->dentry->d_inode->i_mutex);
1339 if (IS_DEADDIR(path->dentry->d_inode))
1342 err = security_sb_check_sb(mnt, path);
1347 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1348 err = attach_recursive_mnt(mnt, path, NULL);
1350 mutex_unlock(&path->dentry->d_inode->i_mutex);
1352 security_sb_post_addmount(mnt, path);
1357 * recursively change the type of the mountpoint.
1359 static int do_change_type(struct path *path, int flag)
1361 struct vfsmount *m, *mnt = path->mnt;
1362 int recurse = flag & MS_REC;
1363 int type = flag & ~MS_REC;
1366 if (!capable(CAP_SYS_ADMIN))
1369 if (path->dentry != path->mnt->mnt_root)
1372 down_write(&namespace_sem);
1373 if (type == MS_SHARED) {
1374 err = invent_group_ids(mnt, recurse);
1379 spin_lock(&vfsmount_lock);
1380 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1381 change_mnt_propagation(m, type);
1382 spin_unlock(&vfsmount_lock);
1385 up_write(&namespace_sem);
1390 * do loopback mount.
1392 static int do_loopback(struct path *path, char *old_name,
1395 struct path old_path;
1396 struct vfsmount *mnt = NULL;
1397 int err = mount_is_safe(path);
1400 if (!old_name || !*old_name)
1402 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1406 down_write(&namespace_sem);
1408 if (IS_MNT_UNBINDABLE(old_path.mnt))
1411 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1416 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1418 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1423 err = graft_tree(mnt, path);
1425 LIST_HEAD(umount_list);
1426 spin_lock(&vfsmount_lock);
1427 umount_tree(mnt, 0, &umount_list);
1428 spin_unlock(&vfsmount_lock);
1429 release_mounts(&umount_list);
1433 up_write(&namespace_sem);
1434 path_put(&old_path);
1438 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1441 int readonly_request = 0;
1443 if (ms_flags & MS_RDONLY)
1444 readonly_request = 1;
1445 if (readonly_request == __mnt_is_readonly(mnt))
1448 if (readonly_request)
1449 error = mnt_make_readonly(mnt);
1451 __mnt_unmake_readonly(mnt);
1456 * change filesystem flags. dir should be a physical root of filesystem.
1457 * If you've mounted a non-root directory somewhere and want to do remount
1458 * on it - tough luck.
1460 static int do_remount(struct path *path, int flags, int mnt_flags,
1464 struct super_block *sb = path->mnt->mnt_sb;
1466 if (!capable(CAP_SYS_ADMIN))
1469 if (!check_mnt(path->mnt))
1472 if (path->dentry != path->mnt->mnt_root)
1475 down_write(&sb->s_umount);
1476 if (flags & MS_BIND)
1477 err = change_mount_flags(path->mnt, flags);
1479 err = do_remount_sb(sb, flags, data, 0);
1481 path->mnt->mnt_flags = mnt_flags;
1482 up_write(&sb->s_umount);
1484 security_sb_post_remount(path->mnt, flags, data);
1486 spin_lock(&vfsmount_lock);
1487 touch_mnt_namespace(path->mnt->mnt_ns);
1488 spin_unlock(&vfsmount_lock);
1493 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1496 for (p = mnt; p; p = next_mnt(p, mnt)) {
1497 if (IS_MNT_UNBINDABLE(p))
1503 static int do_move_mount(struct path *path, char *old_name)
1505 struct path old_path, parent_path;
1508 if (!capable(CAP_SYS_ADMIN))
1510 if (!old_name || !*old_name)
1512 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1516 down_write(&namespace_sem);
1517 while (d_mountpoint(path->dentry) &&
1521 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1525 mutex_lock(&path->dentry->d_inode->i_mutex);
1526 if (IS_DEADDIR(path->dentry->d_inode))
1529 if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
1533 if (old_path.dentry != old_path.mnt->mnt_root)
1536 if (old_path.mnt == old_path.mnt->mnt_parent)
1539 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1540 S_ISDIR(old_path.dentry->d_inode->i_mode))
1543 * Don't move a mount residing in a shared parent.
1545 if (old_path.mnt->mnt_parent &&
1546 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1549 * Don't move a mount tree containing unbindable mounts to a destination
1550 * mount which is shared.
1552 if (IS_MNT_SHARED(path->mnt) &&
1553 tree_contains_unbindable(old_path.mnt))
1556 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1557 if (p == old_path.mnt)
1560 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1564 /* if the mount is moved, it should no longer be expire
1566 list_del_init(&old_path.mnt->mnt_expire);
1568 mutex_unlock(&path->dentry->d_inode->i_mutex);
1570 up_write(&namespace_sem);
1572 path_put(&parent_path);
1573 path_put(&old_path);
1578 * create a new mount for userspace and request it to be added into the
1581 static int do_new_mount(struct path *path, char *type, int flags,
1582 int mnt_flags, char *name, void *data)
1584 struct vfsmount *mnt;
1586 if (!type || !memchr(type, 0, PAGE_SIZE))
1589 /* we need capabilities... */
1590 if (!capable(CAP_SYS_ADMIN))
1593 mnt = do_kern_mount(type, flags, name, data);
1595 return PTR_ERR(mnt);
1597 return do_add_mount(mnt, path, mnt_flags, NULL);
1601 * add a mount into a namespace's mount tree
1602 * - provide the option of adding the new mount to an expiration list
1604 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1605 int mnt_flags, struct list_head *fslist)
1609 down_write(&namespace_sem);
1610 /* Something was mounted here while we slept */
1611 while (d_mountpoint(path->dentry) &&
1615 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
1618 /* Refuse the same filesystem on the same mount point */
1620 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1621 path->mnt->mnt_root == path->dentry)
1625 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1628 newmnt->mnt_flags = mnt_flags;
1629 if ((err = graft_tree(newmnt, path)))
1632 if (fslist) /* add to the specified expiration list */
1633 list_add_tail(&newmnt->mnt_expire, fslist);
1635 up_write(&namespace_sem);
1639 up_write(&namespace_sem);
1644 EXPORT_SYMBOL_GPL(do_add_mount);
1647 * process a list of expirable mountpoints with the intent of discarding any
1648 * mountpoints that aren't in use and haven't been touched since last we came
1651 void mark_mounts_for_expiry(struct list_head *mounts)
1653 struct vfsmount *mnt, *next;
1654 LIST_HEAD(graveyard);
1657 if (list_empty(mounts))
1660 down_write(&namespace_sem);
1661 spin_lock(&vfsmount_lock);
1663 /* extract from the expiration list every vfsmount that matches the
1664 * following criteria:
1665 * - only referenced by its parent vfsmount
1666 * - still marked for expiry (marked on the last call here; marks are
1667 * cleared by mntput())
1669 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1670 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1671 propagate_mount_busy(mnt, 1))
1673 list_move(&mnt->mnt_expire, &graveyard);
1675 while (!list_empty(&graveyard)) {
1676 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1677 touch_mnt_namespace(mnt->mnt_ns);
1678 umount_tree(mnt, 1, &umounts);
1680 spin_unlock(&vfsmount_lock);
1681 up_write(&namespace_sem);
1683 release_mounts(&umounts);
1686 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1689 * Ripoff of 'select_parent()'
1691 * search the list of submounts for a given mountpoint, and move any
1692 * shrinkable submounts to the 'graveyard' list.
1694 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1696 struct vfsmount *this_parent = parent;
1697 struct list_head *next;
1701 next = this_parent->mnt_mounts.next;
1703 while (next != &this_parent->mnt_mounts) {
1704 struct list_head *tmp = next;
1705 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1708 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1711 * Descend a level if the d_mounts list is non-empty.
1713 if (!list_empty(&mnt->mnt_mounts)) {
1718 if (!propagate_mount_busy(mnt, 1)) {
1719 list_move_tail(&mnt->mnt_expire, graveyard);
1724 * All done at this level ... ascend and resume the search
1726 if (this_parent != parent) {
1727 next = this_parent->mnt_child.next;
1728 this_parent = this_parent->mnt_parent;
1735 * process a list of expirable mountpoints with the intent of discarding any
1736 * submounts of a specific parent mountpoint
1738 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1740 LIST_HEAD(graveyard);
1743 /* extract submounts of 'mountpoint' from the expiration list */
1744 while (select_submounts(mnt, &graveyard)) {
1745 while (!list_empty(&graveyard)) {
1746 m = list_first_entry(&graveyard, struct vfsmount,
1748 touch_mnt_namespace(m->mnt_ns);
1749 umount_tree(m, 1, umounts);
1755 * Some copy_from_user() implementations do not return the exact number of
1756 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1757 * Note that this function differs from copy_from_user() in that it will oops
1758 * on bad values of `to', rather than returning a short copy.
1760 static long exact_copy_from_user(void *to, const void __user * from,
1764 const char __user *f = from;
1767 if (!access_ok(VERIFY_READ, from, n))
1771 if (__get_user(c, f)) {
1782 int copy_mount_options(const void __user * data, unsigned long *where)
1792 if (!(page = __get_free_page(GFP_KERNEL)))
1795 /* We only care that *some* data at the address the user
1796 * gave us is valid. Just in case, we'll zero
1797 * the remainder of the page.
1799 /* copy_from_user cannot cross TASK_SIZE ! */
1800 size = TASK_SIZE - (unsigned long)data;
1801 if (size > PAGE_SIZE)
1804 i = size - exact_copy_from_user((void *)page, data, size);
1810 memset((char *)page + i, 0, PAGE_SIZE - i);
1816 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1817 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1819 * data is a (void *) that can point to any structure up to
1820 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1821 * information (or be NULL).
1823 * Pre-0.97 versions of mount() didn't have a flags word.
1824 * When the flags word was introduced its top half was required
1825 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1826 * Therefore, if this magic number is present, it carries no information
1827 * and must be discarded.
1829 long do_mount(char *dev_name, char *dir_name, char *type_page,
1830 unsigned long flags, void *data_page)
1837 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1838 flags &= ~MS_MGC_MSK;
1840 /* Basic sanity checks */
1842 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1844 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1848 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1850 /* Default to relatime unless overriden */
1851 if (!(flags & MS_NOATIME))
1852 mnt_flags |= MNT_RELATIME;
1854 /* Separate the per-mountpoint flags */
1855 if (flags & MS_NOSUID)
1856 mnt_flags |= MNT_NOSUID;
1857 if (flags & MS_NODEV)
1858 mnt_flags |= MNT_NODEV;
1859 if (flags & MS_NOEXEC)
1860 mnt_flags |= MNT_NOEXEC;
1861 if (flags & MS_NOATIME)
1862 mnt_flags |= MNT_NOATIME;
1863 if (flags & MS_NODIRATIME)
1864 mnt_flags |= MNT_NODIRATIME;
1865 if (flags & MS_STRICTATIME)
1866 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1867 if (flags & MS_RDONLY)
1868 mnt_flags |= MNT_READONLY;
1870 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1871 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1874 /* ... and get the mountpoint */
1875 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1879 retval = security_sb_mount(dev_name, &path,
1880 type_page, flags, data_page);
1884 if (flags & MS_REMOUNT)
1885 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1887 else if (flags & MS_BIND)
1888 retval = do_loopback(&path, dev_name, flags & MS_REC);
1889 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1890 retval = do_change_type(&path, flags);
1891 else if (flags & MS_MOVE)
1892 retval = do_move_mount(&path, dev_name);
1894 retval = do_new_mount(&path, type_page, flags, mnt_flags,
1895 dev_name, data_page);
1902 * Allocate a new namespace structure and populate it with contents
1903 * copied from the namespace of the passed in task structure.
1905 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1906 struct fs_struct *fs)
1908 struct mnt_namespace *new_ns;
1909 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1910 struct vfsmount *p, *q;
1912 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1914 return ERR_PTR(-ENOMEM);
1916 atomic_set(&new_ns->count, 1);
1917 INIT_LIST_HEAD(&new_ns->list);
1918 init_waitqueue_head(&new_ns->poll);
1921 down_write(&namespace_sem);
1922 /* First pass: copy the tree topology */
1923 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1924 CL_COPY_ALL | CL_EXPIRE);
1925 if (!new_ns->root) {
1926 up_write(&namespace_sem);
1928 return ERR_PTR(-ENOMEM);
1930 spin_lock(&vfsmount_lock);
1931 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1932 spin_unlock(&vfsmount_lock);
1935 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1936 * as belonging to new namespace. We have already acquired a private
1937 * fs_struct, so tsk->fs->lock is not needed.
1944 if (p == fs->root.mnt) {
1946 fs->root.mnt = mntget(q);
1948 if (p == fs->pwd.mnt) {
1950 fs->pwd.mnt = mntget(q);
1953 p = next_mnt(p, mnt_ns->root);
1954 q = next_mnt(q, new_ns->root);
1956 up_write(&namespace_sem);
1966 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
1967 struct fs_struct *new_fs)
1969 struct mnt_namespace *new_ns;
1974 if (!(flags & CLONE_NEWNS))
1977 new_ns = dup_mnt_ns(ns, new_fs);
1983 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
1984 char __user *, type, unsigned long, flags, void __user *, data)
1987 unsigned long data_page;
1988 unsigned long type_page;
1989 unsigned long dev_page;
1992 retval = copy_mount_options(type, &type_page);
1996 dir_page = getname(dir_name);
1997 retval = PTR_ERR(dir_page);
1998 if (IS_ERR(dir_page))
2001 retval = copy_mount_options(dev_name, &dev_page);
2005 retval = copy_mount_options(data, &data_page);
2010 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2011 flags, (void *)data_page);
2013 free_page(data_page);
2016 free_page(dev_page);
2020 free_page(type_page);
2025 * pivot_root Semantics:
2026 * Moves the root file system of the current process to the directory put_old,
2027 * makes new_root as the new root file system of the current process, and sets
2028 * root/cwd of all processes which had them on the current root to new_root.
2031 * The new_root and put_old must be directories, and must not be on the
2032 * same file system as the current process root. The put_old must be
2033 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2034 * pointed to by put_old must yield the same directory as new_root. No other
2035 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2037 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2038 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2039 * in this situation.
2042 * - we don't move root/cwd if they are not at the root (reason: if something
2043 * cared enough to change them, it's probably wrong to force them elsewhere)
2044 * - it's okay to pick a root that isn't the root of a file system, e.g.
2045 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2046 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2049 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2050 const char __user *, put_old)
2052 struct vfsmount *tmp;
2053 struct path new, old, parent_path, root_parent, root;
2056 if (!capable(CAP_SYS_ADMIN))
2059 error = user_path_dir(new_root, &new);
2063 if (!check_mnt(new.mnt))
2066 error = user_path_dir(put_old, &old);
2070 error = security_sb_pivotroot(&old, &new);
2076 read_lock(¤t->fs->lock);
2077 root = current->fs->root;
2078 path_get(¤t->fs->root);
2079 read_unlock(¤t->fs->lock);
2080 down_write(&namespace_sem);
2081 mutex_lock(&old.dentry->d_inode->i_mutex);
2083 if (IS_MNT_SHARED(old.mnt) ||
2084 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2085 IS_MNT_SHARED(root.mnt->mnt_parent))
2087 if (!check_mnt(root.mnt))
2090 if (IS_DEADDIR(new.dentry->d_inode))
2092 if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
2094 if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
2097 if (new.mnt == root.mnt ||
2098 old.mnt == root.mnt)
2099 goto out2; /* loop, on the same file system */
2101 if (root.mnt->mnt_root != root.dentry)
2102 goto out2; /* not a mountpoint */
2103 if (root.mnt->mnt_parent == root.mnt)
2104 goto out2; /* not attached */
2105 if (new.mnt->mnt_root != new.dentry)
2106 goto out2; /* not a mountpoint */
2107 if (new.mnt->mnt_parent == new.mnt)
2108 goto out2; /* not attached */
2109 /* make sure we can reach put_old from new_root */
2111 spin_lock(&vfsmount_lock);
2112 if (tmp != new.mnt) {
2114 if (tmp->mnt_parent == tmp)
2115 goto out3; /* already mounted on put_old */
2116 if (tmp->mnt_parent == new.mnt)
2118 tmp = tmp->mnt_parent;
2120 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2122 } else if (!is_subdir(old.dentry, new.dentry))
2124 detach_mnt(new.mnt, &parent_path);
2125 detach_mnt(root.mnt, &root_parent);
2126 /* mount old root on put_old */
2127 attach_mnt(root.mnt, &old);
2128 /* mount new_root on / */
2129 attach_mnt(new.mnt, &root_parent);
2130 touch_mnt_namespace(current->nsproxy->mnt_ns);
2131 spin_unlock(&vfsmount_lock);
2132 chroot_fs_refs(&root, &new);
2133 security_sb_post_pivotroot(&root, &new);
2135 path_put(&root_parent);
2136 path_put(&parent_path);
2138 mutex_unlock(&old.dentry->d_inode->i_mutex);
2139 up_write(&namespace_sem);
2147 spin_unlock(&vfsmount_lock);
2151 static void __init init_mount_tree(void)
2153 struct vfsmount *mnt;
2154 struct mnt_namespace *ns;
2157 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2159 panic("Can't create rootfs");
2160 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2162 panic("Can't allocate initial namespace");
2163 atomic_set(&ns->count, 1);
2164 INIT_LIST_HEAD(&ns->list);
2165 init_waitqueue_head(&ns->poll);
2167 list_add(&mnt->mnt_list, &ns->list);
2171 init_task.nsproxy->mnt_ns = ns;
2174 root.mnt = ns->root;
2175 root.dentry = ns->root->mnt_root;
2177 set_fs_pwd(current->fs, &root);
2178 set_fs_root(current->fs, &root);
2181 void __init mnt_init(void)
2186 init_rwsem(&namespace_sem);
2188 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2189 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2191 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2193 if (!mount_hashtable)
2194 panic("Failed to allocate mount hash table\n");
2196 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2198 for (u = 0; u < HASH_SIZE; u++)
2199 INIT_LIST_HEAD(&mount_hashtable[u]);
2203 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2205 fs_kobj = kobject_create_and_add("fs", NULL);
2207 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2212 void __put_mnt_ns(struct mnt_namespace *ns)
2214 struct vfsmount *root = ns->root;
2215 LIST_HEAD(umount_list);
2217 spin_unlock(&vfsmount_lock);
2218 down_write(&namespace_sem);
2219 spin_lock(&vfsmount_lock);
2220 umount_tree(root, 0, &umount_list);
2221 spin_unlock(&vfsmount_lock);
2222 up_write(&namespace_sem);
2223 release_mounts(&umount_list);