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);
134 atomic_set(&mnt->__mnt_writers, 0);
141 kmem_cache_free(mnt_cache, mnt);
146 * Most r/o checks on a fs are for operations that take
147 * discrete amounts of time, like a write() or unlink().
148 * We must keep track of when those operations start
149 * (for permission checks) and when they end, so that
150 * we can determine when writes are able to occur to
154 * __mnt_is_readonly: check whether a mount is read-only
155 * @mnt: the mount to check for its write status
157 * This shouldn't be used directly ouside of the VFS.
158 * It does not guarantee that the filesystem will stay
159 * r/w, just that it is right *now*. This can not and
160 * should not be used in place of IS_RDONLY(inode).
161 * mnt_want/drop_write() will _keep_ the filesystem
164 int __mnt_is_readonly(struct vfsmount *mnt)
166 if (mnt->mnt_flags & MNT_READONLY)
168 if (mnt->mnt_sb->s_flags & MS_RDONLY)
172 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
176 * If holding multiple instances of this lock, they
177 * must be ordered by cpu number.
180 struct lock_class_key lock_class; /* compiles out with !lockdep */
182 struct vfsmount *mnt;
183 } ____cacheline_aligned_in_smp;
184 static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
186 static int __init init_mnt_writers(void)
189 for_each_possible_cpu(cpu) {
190 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
191 spin_lock_init(&writer->lock);
192 lockdep_set_class(&writer->lock, &writer->lock_class);
197 fs_initcall(init_mnt_writers);
199 static void unlock_mnt_writers(void)
202 struct mnt_writer *cpu_writer;
204 for_each_possible_cpu(cpu) {
205 cpu_writer = &per_cpu(mnt_writers, cpu);
206 spin_unlock(&cpu_writer->lock);
210 static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
212 if (!cpu_writer->mnt)
215 * This is in case anyone ever leaves an invalid,
216 * old ->mnt and a count of 0.
218 if (!cpu_writer->count)
220 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
221 cpu_writer->count = 0;
224 * must hold cpu_writer->lock
226 static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
227 struct vfsmount *mnt)
229 if (cpu_writer->mnt == mnt)
231 __clear_mnt_count(cpu_writer);
232 cpu_writer->mnt = mnt;
236 * Most r/o checks on a fs are for operations that take
237 * discrete amounts of time, like a write() or unlink().
238 * We must keep track of when those operations start
239 * (for permission checks) and when they end, so that
240 * we can determine when writes are able to occur to
244 * mnt_want_write - get write access to a mount
245 * @mnt: the mount on which to take a write
247 * This tells the low-level filesystem that a write is
248 * about to be performed to it, and makes sure that
249 * writes are allowed before returning success. When
250 * the write operation is finished, mnt_drop_write()
251 * must be called. This is effectively a refcount.
253 int mnt_want_write(struct vfsmount *mnt)
256 struct mnt_writer *cpu_writer;
258 cpu_writer = &get_cpu_var(mnt_writers);
259 spin_lock(&cpu_writer->lock);
260 if (__mnt_is_readonly(mnt)) {
264 use_cpu_writer_for_mount(cpu_writer, mnt);
267 spin_unlock(&cpu_writer->lock);
268 put_cpu_var(mnt_writers);
271 EXPORT_SYMBOL_GPL(mnt_want_write);
273 static void lock_mnt_writers(void)
276 struct mnt_writer *cpu_writer;
278 for_each_possible_cpu(cpu) {
279 cpu_writer = &per_cpu(mnt_writers, cpu);
280 spin_lock(&cpu_writer->lock);
281 __clear_mnt_count(cpu_writer);
282 cpu_writer->mnt = NULL;
287 * These per-cpu write counts are not guaranteed to have
288 * matched increments and decrements on any given cpu.
289 * A file open()ed for write on one cpu and close()d on
290 * another cpu will imbalance this count. Make sure it
291 * does not get too far out of whack.
293 static void handle_write_count_underflow(struct vfsmount *mnt)
295 if (atomic_read(&mnt->__mnt_writers) >=
296 MNT_WRITER_UNDERFLOW_LIMIT)
299 * It isn't necessary to hold all of the locks
300 * at the same time, but doing it this way makes
301 * us share a lot more code.
305 * vfsmount_lock is for mnt_flags.
307 spin_lock(&vfsmount_lock);
309 * If coalescing the per-cpu writer counts did not
310 * get us back to a positive writer count, we have
313 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
314 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
315 WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
317 mnt, atomic_read(&mnt->__mnt_writers));
318 /* use the flag to keep the dmesg spam down */
319 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
321 spin_unlock(&vfsmount_lock);
322 unlock_mnt_writers();
326 * mnt_drop_write - give up write access to a mount
327 * @mnt: the mount on which to give up write access
329 * Tells the low-level filesystem that we are done
330 * performing writes to it. Must be matched with
331 * mnt_want_write() call above.
333 void mnt_drop_write(struct vfsmount *mnt)
335 int must_check_underflow = 0;
336 struct mnt_writer *cpu_writer;
338 cpu_writer = &get_cpu_var(mnt_writers);
339 spin_lock(&cpu_writer->lock);
341 use_cpu_writer_for_mount(cpu_writer, mnt);
342 if (cpu_writer->count > 0) {
345 must_check_underflow = 1;
346 atomic_dec(&mnt->__mnt_writers);
349 spin_unlock(&cpu_writer->lock);
351 * Logically, we could call this each time,
352 * but the __mnt_writers cacheline tends to
353 * be cold, and makes this expensive.
355 if (must_check_underflow)
356 handle_write_count_underflow(mnt);
358 * This could be done right after the spinlock
359 * is taken because the spinlock keeps us on
360 * the cpu, and disables preemption. However,
361 * putting it here bounds the amount that
362 * __mnt_writers can underflow. Without it,
363 * we could theoretically wrap __mnt_writers.
365 put_cpu_var(mnt_writers);
367 EXPORT_SYMBOL_GPL(mnt_drop_write);
369 static int mnt_make_readonly(struct vfsmount *mnt)
375 * With all the locks held, this value is stable
377 if (atomic_read(&mnt->__mnt_writers) > 0) {
382 * nobody can do a successful mnt_want_write() with all
383 * of the counts in MNT_DENIED_WRITE and the locks held.
385 spin_lock(&vfsmount_lock);
387 mnt->mnt_flags |= MNT_READONLY;
388 spin_unlock(&vfsmount_lock);
390 unlock_mnt_writers();
394 static void __mnt_unmake_readonly(struct vfsmount *mnt)
396 spin_lock(&vfsmount_lock);
397 mnt->mnt_flags &= ~MNT_READONLY;
398 spin_unlock(&vfsmount_lock);
401 void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
404 mnt->mnt_root = dget(sb->s_root);
407 EXPORT_SYMBOL(simple_set_mnt);
409 void free_vfsmnt(struct vfsmount *mnt)
411 kfree(mnt->mnt_devname);
413 kmem_cache_free(mnt_cache, mnt);
417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
420 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
423 struct list_head *head = mount_hashtable + hash(mnt, dentry);
424 struct list_head *tmp = head;
425 struct vfsmount *p, *found = NULL;
428 tmp = dir ? tmp->next : tmp->prev;
432 p = list_entry(tmp, struct vfsmount, mnt_hash);
433 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
445 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
447 struct vfsmount *child_mnt;
448 spin_lock(&vfsmount_lock);
449 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
451 spin_unlock(&vfsmount_lock);
455 static inline int check_mnt(struct vfsmount *mnt)
457 return mnt->mnt_ns == current->nsproxy->mnt_ns;
460 static void touch_mnt_namespace(struct mnt_namespace *ns)
464 wake_up_interruptible(&ns->poll);
468 static void __touch_mnt_namespace(struct mnt_namespace *ns)
470 if (ns && ns->event != event) {
472 wake_up_interruptible(&ns->poll);
476 static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
478 old_path->dentry = mnt->mnt_mountpoint;
479 old_path->mnt = mnt->mnt_parent;
480 mnt->mnt_parent = mnt;
481 mnt->mnt_mountpoint = mnt->mnt_root;
482 list_del_init(&mnt->mnt_child);
483 list_del_init(&mnt->mnt_hash);
484 old_path->dentry->d_mounted--;
487 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
488 struct vfsmount *child_mnt)
490 child_mnt->mnt_parent = mntget(mnt);
491 child_mnt->mnt_mountpoint = dget(dentry);
495 static void attach_mnt(struct vfsmount *mnt, struct path *path)
497 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
498 list_add_tail(&mnt->mnt_hash, mount_hashtable +
499 hash(path->mnt, path->dentry));
500 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
504 * the caller must hold vfsmount_lock
506 static void commit_tree(struct vfsmount *mnt)
508 struct vfsmount *parent = mnt->mnt_parent;
511 struct mnt_namespace *n = parent->mnt_ns;
513 BUG_ON(parent == mnt);
515 list_add_tail(&head, &mnt->mnt_list);
516 list_for_each_entry(m, &head, mnt_list)
518 list_splice(&head, n->list.prev);
520 list_add_tail(&mnt->mnt_hash, mount_hashtable +
521 hash(parent, mnt->mnt_mountpoint));
522 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
523 touch_mnt_namespace(n);
526 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
528 struct list_head *next = p->mnt_mounts.next;
529 if (next == &p->mnt_mounts) {
533 next = p->mnt_child.next;
534 if (next != &p->mnt_parent->mnt_mounts)
539 return list_entry(next, struct vfsmount, mnt_child);
542 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
544 struct list_head *prev = p->mnt_mounts.prev;
545 while (prev != &p->mnt_mounts) {
546 p = list_entry(prev, struct vfsmount, mnt_child);
547 prev = p->mnt_mounts.prev;
552 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
555 struct super_block *sb = old->mnt_sb;
556 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
559 if (flag & (CL_SLAVE | CL_PRIVATE))
560 mnt->mnt_group_id = 0; /* not a peer of original */
562 mnt->mnt_group_id = old->mnt_group_id;
564 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
565 int err = mnt_alloc_group_id(mnt);
570 mnt->mnt_flags = old->mnt_flags;
571 atomic_inc(&sb->s_active);
573 mnt->mnt_root = dget(root);
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 mnt->mnt_parent = mnt;
577 if (flag & CL_SLAVE) {
578 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
579 mnt->mnt_master = old;
580 CLEAR_MNT_SHARED(mnt);
581 } else if (!(flag & CL_PRIVATE)) {
582 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
583 list_add(&mnt->mnt_share, &old->mnt_share);
584 if (IS_MNT_SLAVE(old))
585 list_add(&mnt->mnt_slave, &old->mnt_slave);
586 mnt->mnt_master = old->mnt_master;
588 if (flag & CL_MAKE_SHARED)
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
593 if (flag & CL_EXPIRE) {
594 if (!list_empty(&old->mnt_expire))
595 list_add(&mnt->mnt_expire, &old->mnt_expire);
605 static inline void __mntput(struct vfsmount *mnt)
608 struct super_block *sb = mnt->mnt_sb;
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
615 for_each_possible_cpu(cpu) {
616 struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
617 spin_lock(&cpu_writer->lock);
618 if (cpu_writer->mnt != mnt) {
619 spin_unlock(&cpu_writer->lock);
622 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
623 cpu_writer->count = 0;
625 * Might as well do this so that no one
626 * ever sees the pointer and expects
629 cpu_writer->mnt = NULL;
630 spin_unlock(&cpu_writer->lock);
633 * This probably indicates that somebody messed
634 * up a mnt_want/drop_write() pair. If this
635 * happens, the filesystem was probably unable
636 * to make r/w->r/o transitions.
638 WARN_ON(atomic_read(&mnt->__mnt_writers));
641 deactivate_super(sb);
644 void mntput_no_expire(struct vfsmount *mnt)
647 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
648 if (likely(!mnt->mnt_pinned)) {
649 spin_unlock(&vfsmount_lock);
653 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
655 spin_unlock(&vfsmount_lock);
656 acct_auto_close_mnt(mnt);
657 security_sb_umount_close(mnt);
662 EXPORT_SYMBOL(mntput_no_expire);
664 void mnt_pin(struct vfsmount *mnt)
666 spin_lock(&vfsmount_lock);
668 spin_unlock(&vfsmount_lock);
671 EXPORT_SYMBOL(mnt_pin);
673 void mnt_unpin(struct vfsmount *mnt)
675 spin_lock(&vfsmount_lock);
676 if (mnt->mnt_pinned) {
677 atomic_inc(&mnt->mnt_count);
680 spin_unlock(&vfsmount_lock);
683 EXPORT_SYMBOL(mnt_unpin);
685 static inline void mangle(struct seq_file *m, const char *s)
687 seq_escape(m, s, " \t\n\\");
691 * Simple .show_options callback for filesystems which don't want to
692 * implement more complex mount option showing.
694 * See also save_mount_options().
696 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
698 const char *options = mnt->mnt_sb->s_options;
700 if (options != NULL && options[0]) {
707 EXPORT_SYMBOL(generic_show_options);
710 * If filesystem uses generic_show_options(), this function should be
711 * called from the fill_super() callback.
713 * The .remount_fs callback usually needs to be handled in a special
714 * way, to make sure, that previous options are not overwritten if the
717 * Also note, that if the filesystem's .remount_fs function doesn't
718 * reset all options to their default value, but changes only newly
719 * given options, then the displayed options will not reflect reality
722 void save_mount_options(struct super_block *sb, char *options)
724 kfree(sb->s_options);
725 sb->s_options = kstrdup(options, GFP_KERNEL);
727 EXPORT_SYMBOL(save_mount_options);
729 #ifdef CONFIG_PROC_FS
731 static void *m_start(struct seq_file *m, loff_t *pos)
733 struct proc_mounts *p = m->private;
735 down_read(&namespace_sem);
736 return seq_list_start(&p->ns->list, *pos);
739 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
741 struct proc_mounts *p = m->private;
743 return seq_list_next(v, &p->ns->list, pos);
746 static void m_stop(struct seq_file *m, void *v)
748 up_read(&namespace_sem);
751 struct proc_fs_info {
756 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
758 static const struct proc_fs_info fs_info[] = {
759 { MS_SYNCHRONOUS, ",sync" },
760 { MS_DIRSYNC, ",dirsync" },
761 { MS_MANDLOCK, ",mand" },
764 const struct proc_fs_info *fs_infop;
766 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
767 if (sb->s_flags & fs_infop->flag)
768 seq_puts(m, fs_infop->str);
771 return security_sb_show_options(m, sb);
774 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
776 static const struct proc_fs_info mnt_info[] = {
777 { MNT_NOSUID, ",nosuid" },
778 { MNT_NODEV, ",nodev" },
779 { MNT_NOEXEC, ",noexec" },
780 { MNT_NOATIME, ",noatime" },
781 { MNT_NODIRATIME, ",nodiratime" },
782 { MNT_RELATIME, ",relatime" },
783 { MNT_STRICTATIME, ",strictatime" },
786 const struct proc_fs_info *fs_infop;
788 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
789 if (mnt->mnt_flags & fs_infop->flag)
790 seq_puts(m, fs_infop->str);
794 static void show_type(struct seq_file *m, struct super_block *sb)
796 mangle(m, sb->s_type->name);
797 if (sb->s_subtype && sb->s_subtype[0]) {
799 mangle(m, sb->s_subtype);
803 static int show_vfsmnt(struct seq_file *m, void *v)
805 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
807 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
809 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
811 seq_path(m, &mnt_path, " \t\n\\");
813 show_type(m, mnt->mnt_sb);
814 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
815 err = show_sb_opts(m, mnt->mnt_sb);
818 show_mnt_opts(m, mnt);
819 if (mnt->mnt_sb->s_op->show_options)
820 err = mnt->mnt_sb->s_op->show_options(m, mnt);
821 seq_puts(m, " 0 0\n");
826 const struct seq_operations mounts_op = {
833 static int show_mountinfo(struct seq_file *m, void *v)
835 struct proc_mounts *p = m->private;
836 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
837 struct super_block *sb = mnt->mnt_sb;
838 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
839 struct path root = p->root;
842 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
843 MAJOR(sb->s_dev), MINOR(sb->s_dev));
844 seq_dentry(m, mnt->mnt_root, " \t\n\\");
846 seq_path_root(m, &mnt_path, &root, " \t\n\\");
847 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
849 * Mountpoint is outside root, discard that one. Ugly,
850 * but less so than trying to do that in iterator in a
851 * race-free way (due to renames).
855 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
856 show_mnt_opts(m, mnt);
858 /* Tagged fields ("foo:X" or "bar") */
859 if (IS_MNT_SHARED(mnt))
860 seq_printf(m, " shared:%i", mnt->mnt_group_id);
861 if (IS_MNT_SLAVE(mnt)) {
862 int master = mnt->mnt_master->mnt_group_id;
863 int dom = get_dominating_id(mnt, &p->root);
864 seq_printf(m, " master:%i", master);
865 if (dom && dom != master)
866 seq_printf(m, " propagate_from:%i", dom);
868 if (IS_MNT_UNBINDABLE(mnt))
869 seq_puts(m, " unbindable");
871 /* Filesystem specific data */
875 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
876 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
877 err = show_sb_opts(m, sb);
880 if (sb->s_op->show_options)
881 err = sb->s_op->show_options(m, mnt);
887 const struct seq_operations mountinfo_op = {
891 .show = show_mountinfo,
894 static int show_vfsstat(struct seq_file *m, void *v)
896 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
897 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
901 if (mnt->mnt_devname) {
902 seq_puts(m, "device ");
903 mangle(m, mnt->mnt_devname);
905 seq_puts(m, "no device");
908 seq_puts(m, " mounted on ");
909 seq_path(m, &mnt_path, " \t\n\\");
912 /* file system type */
913 seq_puts(m, "with fstype ");
914 show_type(m, mnt->mnt_sb);
916 /* optional statistics */
917 if (mnt->mnt_sb->s_op->show_stats) {
919 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
926 const struct seq_operations mountstats_op = {
930 .show = show_vfsstat,
932 #endif /* CONFIG_PROC_FS */
935 * may_umount_tree - check if a mount tree is busy
936 * @mnt: root of mount tree
938 * This is called to check if a tree of mounts has any
939 * open files, pwds, chroots or sub mounts that are
942 int may_umount_tree(struct vfsmount *mnt)
945 int minimum_refs = 0;
948 spin_lock(&vfsmount_lock);
949 for (p = mnt; p; p = next_mnt(p, mnt)) {
950 actual_refs += atomic_read(&p->mnt_count);
953 spin_unlock(&vfsmount_lock);
955 if (actual_refs > minimum_refs)
961 EXPORT_SYMBOL(may_umount_tree);
964 * may_umount - check if a mount point is busy
965 * @mnt: root of mount
967 * This is called to check if a mount point has any
968 * open files, pwds, chroots or sub mounts. If the
969 * mount has sub mounts this will return busy
970 * regardless of whether the sub mounts are busy.
972 * Doesn't take quota and stuff into account. IOW, in some cases it will
973 * give false negatives. The main reason why it's here is that we need
974 * a non-destructive way to look for easily umountable filesystems.
976 int may_umount(struct vfsmount *mnt)
979 spin_lock(&vfsmount_lock);
980 if (propagate_mount_busy(mnt, 2))
982 spin_unlock(&vfsmount_lock);
986 EXPORT_SYMBOL(may_umount);
988 void release_mounts(struct list_head *head)
990 struct vfsmount *mnt;
991 while (!list_empty(head)) {
992 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
993 list_del_init(&mnt->mnt_hash);
994 if (mnt->mnt_parent != mnt) {
995 struct dentry *dentry;
997 spin_lock(&vfsmount_lock);
998 dentry = mnt->mnt_mountpoint;
1000 mnt->mnt_mountpoint = mnt->mnt_root;
1001 mnt->mnt_parent = mnt;
1003 spin_unlock(&vfsmount_lock);
1011 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1015 for (p = mnt; p; p = next_mnt(p, mnt))
1016 list_move(&p->mnt_hash, kill);
1019 propagate_umount(kill);
1021 list_for_each_entry(p, kill, mnt_hash) {
1022 list_del_init(&p->mnt_expire);
1023 list_del_init(&p->mnt_list);
1024 __touch_mnt_namespace(p->mnt_ns);
1026 list_del_init(&p->mnt_child);
1027 if (p->mnt_parent != p) {
1028 p->mnt_parent->mnt_ghosts++;
1029 p->mnt_mountpoint->d_mounted--;
1031 change_mnt_propagation(p, MS_PRIVATE);
1035 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1037 static int do_umount(struct vfsmount *mnt, int flags)
1039 struct super_block *sb = mnt->mnt_sb;
1041 LIST_HEAD(umount_list);
1043 retval = security_sb_umount(mnt, flags);
1048 * Allow userspace to request a mountpoint be expired rather than
1049 * unmounting unconditionally. Unmount only happens if:
1050 * (1) the mark is already set (the mark is cleared by mntput())
1051 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1053 if (flags & MNT_EXPIRE) {
1054 if (mnt == current->fs->root.mnt ||
1055 flags & (MNT_FORCE | MNT_DETACH))
1058 if (atomic_read(&mnt->mnt_count) != 2)
1061 if (!xchg(&mnt->mnt_expiry_mark, 1))
1066 * If we may have to abort operations to get out of this
1067 * mount, and they will themselves hold resources we must
1068 * allow the fs to do things. In the Unix tradition of
1069 * 'Gee thats tricky lets do it in userspace' the umount_begin
1070 * might fail to complete on the first run through as other tasks
1071 * must return, and the like. Thats for the mount program to worry
1072 * about for the moment.
1075 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1076 sb->s_op->umount_begin(sb);
1080 * No sense to grab the lock for this test, but test itself looks
1081 * somewhat bogus. Suggestions for better replacement?
1082 * Ho-hum... In principle, we might treat that as umount + switch
1083 * to rootfs. GC would eventually take care of the old vfsmount.
1084 * Actually it makes sense, especially if rootfs would contain a
1085 * /reboot - static binary that would close all descriptors and
1086 * call reboot(9). Then init(8) could umount root and exec /reboot.
1088 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1090 * Special case for "unmounting" root ...
1091 * we just try to remount it readonly.
1093 down_write(&sb->s_umount);
1094 if (!(sb->s_flags & MS_RDONLY)) {
1096 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1099 up_write(&sb->s_umount);
1103 down_write(&namespace_sem);
1104 spin_lock(&vfsmount_lock);
1107 if (!(flags & MNT_DETACH))
1108 shrink_submounts(mnt, &umount_list);
1111 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1112 if (!list_empty(&mnt->mnt_list))
1113 umount_tree(mnt, 1, &umount_list);
1116 spin_unlock(&vfsmount_lock);
1118 security_sb_umount_busy(mnt);
1119 up_write(&namespace_sem);
1120 release_mounts(&umount_list);
1125 * Now umount can handle mount points as well as block devices.
1126 * This is important for filesystems which use unnamed block devices.
1128 * We now support a flag for forced unmount like the other 'big iron'
1129 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1132 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1137 retval = user_path(name, &path);
1141 if (path.dentry != path.mnt->mnt_root)
1143 if (!check_mnt(path.mnt))
1147 if (!capable(CAP_SYS_ADMIN))
1150 retval = do_umount(path.mnt, flags);
1152 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1154 mntput_no_expire(path.mnt);
1159 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1162 * The 2.0 compatible umount. No flags.
1164 SYSCALL_DEFINE1(oldumount, char __user *, name)
1166 return sys_umount(name, 0);
1171 static int mount_is_safe(struct path *path)
1173 if (capable(CAP_SYS_ADMIN))
1177 if (S_ISLNK(path->dentry->d_inode->i_mode))
1179 if (path->dentry->d_inode->i_mode & S_ISVTX) {
1180 if (current_uid() != path->dentry->d_inode->i_uid)
1183 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1189 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1192 struct vfsmount *res, *p, *q, *r, *s;
1195 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1198 res = q = clone_mnt(mnt, dentry, flag);
1201 q->mnt_mountpoint = mnt->mnt_mountpoint;
1204 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1205 if (!is_subdir(r->mnt_mountpoint, dentry))
1208 for (s = r; s; s = next_mnt(s, r)) {
1209 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1210 s = skip_mnt_tree(s);
1213 while (p != s->mnt_parent) {
1219 path.dentry = p->mnt_mountpoint;
1220 q = clone_mnt(p, p->mnt_root, flag);
1223 spin_lock(&vfsmount_lock);
1224 list_add_tail(&q->mnt_list, &res->mnt_list);
1225 attach_mnt(q, &path);
1226 spin_unlock(&vfsmount_lock);
1232 LIST_HEAD(umount_list);
1233 spin_lock(&vfsmount_lock);
1234 umount_tree(res, 0, &umount_list);
1235 spin_unlock(&vfsmount_lock);
1236 release_mounts(&umount_list);
1241 struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1243 struct vfsmount *tree;
1244 down_write(&namespace_sem);
1245 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
1246 up_write(&namespace_sem);
1250 void drop_collected_mounts(struct vfsmount *mnt)
1252 LIST_HEAD(umount_list);
1253 down_write(&namespace_sem);
1254 spin_lock(&vfsmount_lock);
1255 umount_tree(mnt, 0, &umount_list);
1256 spin_unlock(&vfsmount_lock);
1257 up_write(&namespace_sem);
1258 release_mounts(&umount_list);
1261 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1265 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1266 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1267 mnt_release_group_id(p);
1271 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1275 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1276 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1277 int err = mnt_alloc_group_id(p);
1279 cleanup_group_ids(mnt, p);
1289 * @source_mnt : mount tree to be attached
1290 * @nd : place the mount tree @source_mnt is attached
1291 * @parent_nd : if non-null, detach the source_mnt from its parent and
1292 * store the parent mount and mountpoint dentry.
1293 * (done when source_mnt is moved)
1295 * NOTE: in the table below explains the semantics when a source mount
1296 * of a given type is attached to a destination mount of a given type.
1297 * ---------------------------------------------------------------------------
1298 * | BIND MOUNT OPERATION |
1299 * |**************************************************************************
1300 * | source-->| shared | private | slave | unbindable |
1304 * |**************************************************************************
1305 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1307 * |non-shared| shared (+) | private | slave (*) | invalid |
1308 * ***************************************************************************
1309 * A bind operation clones the source mount and mounts the clone on the
1310 * destination mount.
1312 * (++) the cloned mount is propagated to all the mounts in the propagation
1313 * tree of the destination mount and the cloned mount is added to
1314 * the peer group of the source mount.
1315 * (+) the cloned mount is created under the destination mount and is marked
1316 * as shared. The cloned mount is added to the peer group of the source
1318 * (+++) the mount is propagated to all the mounts in the propagation tree
1319 * of the destination mount and the cloned mount is made slave
1320 * of the same master as that of the source mount. The cloned mount
1321 * is marked as 'shared and slave'.
1322 * (*) the cloned mount is made a slave of the same master as that of the
1325 * ---------------------------------------------------------------------------
1326 * | MOVE MOUNT OPERATION |
1327 * |**************************************************************************
1328 * | source-->| shared | private | slave | unbindable |
1332 * |**************************************************************************
1333 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1335 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1336 * ***************************************************************************
1338 * (+) the mount is moved to the destination. And is then propagated to
1339 * all the mounts in the propagation tree of the destination mount.
1340 * (+*) the mount is moved to the destination.
1341 * (+++) the mount is moved to the destination and is then propagated to
1342 * all the mounts belonging to the destination mount's propagation tree.
1343 * the mount is marked as 'shared and slave'.
1344 * (*) the mount continues to be a slave at the new location.
1346 * if the source mount is a tree, the operations explained above is
1347 * applied to each mount in the tree.
1348 * Must be called without spinlocks held, since this function can sleep
1351 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1352 struct path *path, struct path *parent_path)
1354 LIST_HEAD(tree_list);
1355 struct vfsmount *dest_mnt = path->mnt;
1356 struct dentry *dest_dentry = path->dentry;
1357 struct vfsmount *child, *p;
1360 if (IS_MNT_SHARED(dest_mnt)) {
1361 err = invent_group_ids(source_mnt, true);
1365 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1367 goto out_cleanup_ids;
1369 if (IS_MNT_SHARED(dest_mnt)) {
1370 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1374 spin_lock(&vfsmount_lock);
1376 detach_mnt(source_mnt, parent_path);
1377 attach_mnt(source_mnt, path);
1378 touch_mnt_namespace(parent_path->mnt->mnt_ns);
1380 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1381 commit_tree(source_mnt);
1384 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1385 list_del_init(&child->mnt_hash);
1388 spin_unlock(&vfsmount_lock);
1392 if (IS_MNT_SHARED(dest_mnt))
1393 cleanup_group_ids(source_mnt, NULL);
1398 static int graft_tree(struct vfsmount *mnt, struct path *path)
1401 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1404 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1405 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1409 mutex_lock(&path->dentry->d_inode->i_mutex);
1410 if (IS_DEADDIR(path->dentry->d_inode))
1413 err = security_sb_check_sb(mnt, path);
1418 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1419 err = attach_recursive_mnt(mnt, path, NULL);
1421 mutex_unlock(&path->dentry->d_inode->i_mutex);
1423 security_sb_post_addmount(mnt, path);
1428 * recursively change the type of the mountpoint.
1430 static int do_change_type(struct path *path, int flag)
1432 struct vfsmount *m, *mnt = path->mnt;
1433 int recurse = flag & MS_REC;
1434 int type = flag & ~MS_REC;
1437 if (!capable(CAP_SYS_ADMIN))
1440 if (path->dentry != path->mnt->mnt_root)
1443 down_write(&namespace_sem);
1444 if (type == MS_SHARED) {
1445 err = invent_group_ids(mnt, recurse);
1450 spin_lock(&vfsmount_lock);
1451 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1452 change_mnt_propagation(m, type);
1453 spin_unlock(&vfsmount_lock);
1456 up_write(&namespace_sem);
1461 * do loopback mount.
1463 static int do_loopback(struct path *path, char *old_name,
1466 struct path old_path;
1467 struct vfsmount *mnt = NULL;
1468 int err = mount_is_safe(path);
1471 if (!old_name || !*old_name)
1473 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1477 down_write(&namespace_sem);
1479 if (IS_MNT_UNBINDABLE(old_path.mnt))
1482 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1487 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1489 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1494 err = graft_tree(mnt, path);
1496 LIST_HEAD(umount_list);
1497 spin_lock(&vfsmount_lock);
1498 umount_tree(mnt, 0, &umount_list);
1499 spin_unlock(&vfsmount_lock);
1500 release_mounts(&umount_list);
1504 up_write(&namespace_sem);
1505 path_put(&old_path);
1509 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1512 int readonly_request = 0;
1514 if (ms_flags & MS_RDONLY)
1515 readonly_request = 1;
1516 if (readonly_request == __mnt_is_readonly(mnt))
1519 if (readonly_request)
1520 error = mnt_make_readonly(mnt);
1522 __mnt_unmake_readonly(mnt);
1527 * change filesystem flags. dir should be a physical root of filesystem.
1528 * If you've mounted a non-root directory somewhere and want to do remount
1529 * on it - tough luck.
1531 static int do_remount(struct path *path, int flags, int mnt_flags,
1535 struct super_block *sb = path->mnt->mnt_sb;
1537 if (!capable(CAP_SYS_ADMIN))
1540 if (!check_mnt(path->mnt))
1543 if (path->dentry != path->mnt->mnt_root)
1546 down_write(&sb->s_umount);
1547 if (flags & MS_BIND)
1548 err = change_mount_flags(path->mnt, flags);
1550 err = do_remount_sb(sb, flags, data, 0);
1552 path->mnt->mnt_flags = mnt_flags;
1553 up_write(&sb->s_umount);
1555 security_sb_post_remount(path->mnt, flags, data);
1557 spin_lock(&vfsmount_lock);
1558 touch_mnt_namespace(path->mnt->mnt_ns);
1559 spin_unlock(&vfsmount_lock);
1564 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1567 for (p = mnt; p; p = next_mnt(p, mnt)) {
1568 if (IS_MNT_UNBINDABLE(p))
1574 static int do_move_mount(struct path *path, char *old_name)
1576 struct path old_path, parent_path;
1579 if (!capable(CAP_SYS_ADMIN))
1581 if (!old_name || !*old_name)
1583 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1587 down_write(&namespace_sem);
1588 while (d_mountpoint(path->dentry) &&
1589 follow_down(&path->mnt, &path->dentry))
1592 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1596 mutex_lock(&path->dentry->d_inode->i_mutex);
1597 if (IS_DEADDIR(path->dentry->d_inode))
1600 if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
1604 if (old_path.dentry != old_path.mnt->mnt_root)
1607 if (old_path.mnt == old_path.mnt->mnt_parent)
1610 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1611 S_ISDIR(old_path.dentry->d_inode->i_mode))
1614 * Don't move a mount residing in a shared parent.
1616 if (old_path.mnt->mnt_parent &&
1617 IS_MNT_SHARED(old_path.mnt->mnt_parent))
1620 * Don't move a mount tree containing unbindable mounts to a destination
1621 * mount which is shared.
1623 if (IS_MNT_SHARED(path->mnt) &&
1624 tree_contains_unbindable(old_path.mnt))
1627 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1628 if (p == old_path.mnt)
1631 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1635 /* if the mount is moved, it should no longer be expire
1637 list_del_init(&old_path.mnt->mnt_expire);
1639 mutex_unlock(&path->dentry->d_inode->i_mutex);
1641 up_write(&namespace_sem);
1643 path_put(&parent_path);
1644 path_put(&old_path);
1649 * create a new mount for userspace and request it to be added into the
1652 static int do_new_mount(struct path *path, char *type, int flags,
1653 int mnt_flags, char *name, void *data)
1655 struct vfsmount *mnt;
1657 if (!type || !memchr(type, 0, PAGE_SIZE))
1660 /* we need capabilities... */
1661 if (!capable(CAP_SYS_ADMIN))
1664 mnt = do_kern_mount(type, flags, name, data);
1666 return PTR_ERR(mnt);
1668 return do_add_mount(mnt, path, mnt_flags, NULL);
1672 * add a mount into a namespace's mount tree
1673 * - provide the option of adding the new mount to an expiration list
1675 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1676 int mnt_flags, struct list_head *fslist)
1680 down_write(&namespace_sem);
1681 /* Something was mounted here while we slept */
1682 while (d_mountpoint(path->dentry) &&
1683 follow_down(&path->mnt, &path->dentry))
1686 if (!check_mnt(path->mnt))
1689 /* Refuse the same filesystem on the same mount point */
1691 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1692 path->mnt->mnt_root == path->dentry)
1696 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1699 newmnt->mnt_flags = mnt_flags;
1700 if ((err = graft_tree(newmnt, path)))
1703 if (fslist) /* add to the specified expiration list */
1704 list_add_tail(&newmnt->mnt_expire, fslist);
1706 up_write(&namespace_sem);
1710 up_write(&namespace_sem);
1715 EXPORT_SYMBOL_GPL(do_add_mount);
1718 * process a list of expirable mountpoints with the intent of discarding any
1719 * mountpoints that aren't in use and haven't been touched since last we came
1722 void mark_mounts_for_expiry(struct list_head *mounts)
1724 struct vfsmount *mnt, *next;
1725 LIST_HEAD(graveyard);
1728 if (list_empty(mounts))
1731 down_write(&namespace_sem);
1732 spin_lock(&vfsmount_lock);
1734 /* extract from the expiration list every vfsmount that matches the
1735 * following criteria:
1736 * - only referenced by its parent vfsmount
1737 * - still marked for expiry (marked on the last call here; marks are
1738 * cleared by mntput())
1740 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1741 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1742 propagate_mount_busy(mnt, 1))
1744 list_move(&mnt->mnt_expire, &graveyard);
1746 while (!list_empty(&graveyard)) {
1747 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1748 touch_mnt_namespace(mnt->mnt_ns);
1749 umount_tree(mnt, 1, &umounts);
1751 spin_unlock(&vfsmount_lock);
1752 up_write(&namespace_sem);
1754 release_mounts(&umounts);
1757 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1760 * Ripoff of 'select_parent()'
1762 * search the list of submounts for a given mountpoint, and move any
1763 * shrinkable submounts to the 'graveyard' list.
1765 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1767 struct vfsmount *this_parent = parent;
1768 struct list_head *next;
1772 next = this_parent->mnt_mounts.next;
1774 while (next != &this_parent->mnt_mounts) {
1775 struct list_head *tmp = next;
1776 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1779 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1782 * Descend a level if the d_mounts list is non-empty.
1784 if (!list_empty(&mnt->mnt_mounts)) {
1789 if (!propagate_mount_busy(mnt, 1)) {
1790 list_move_tail(&mnt->mnt_expire, graveyard);
1795 * All done at this level ... ascend and resume the search
1797 if (this_parent != parent) {
1798 next = this_parent->mnt_child.next;
1799 this_parent = this_parent->mnt_parent;
1806 * process a list of expirable mountpoints with the intent of discarding any
1807 * submounts of a specific parent mountpoint
1809 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1811 LIST_HEAD(graveyard);
1814 /* extract submounts of 'mountpoint' from the expiration list */
1815 while (select_submounts(mnt, &graveyard)) {
1816 while (!list_empty(&graveyard)) {
1817 m = list_first_entry(&graveyard, struct vfsmount,
1819 touch_mnt_namespace(m->mnt_ns);
1820 umount_tree(m, 1, umounts);
1826 * Some copy_from_user() implementations do not return the exact number of
1827 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1828 * Note that this function differs from copy_from_user() in that it will oops
1829 * on bad values of `to', rather than returning a short copy.
1831 static long exact_copy_from_user(void *to, const void __user * from,
1835 const char __user *f = from;
1838 if (!access_ok(VERIFY_READ, from, n))
1842 if (__get_user(c, f)) {
1853 int copy_mount_options(const void __user * data, unsigned long *where)
1863 if (!(page = __get_free_page(GFP_KERNEL)))
1866 /* We only care that *some* data at the address the user
1867 * gave us is valid. Just in case, we'll zero
1868 * the remainder of the page.
1870 /* copy_from_user cannot cross TASK_SIZE ! */
1871 size = TASK_SIZE - (unsigned long)data;
1872 if (size > PAGE_SIZE)
1875 i = size - exact_copy_from_user((void *)page, data, size);
1881 memset((char *)page + i, 0, PAGE_SIZE - i);
1887 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1888 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1890 * data is a (void *) that can point to any structure up to
1891 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1892 * information (or be NULL).
1894 * Pre-0.97 versions of mount() didn't have a flags word.
1895 * When the flags word was introduced its top half was required
1896 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1897 * Therefore, if this magic number is present, it carries no information
1898 * and must be discarded.
1900 long do_mount(char *dev_name, char *dir_name, char *type_page,
1901 unsigned long flags, void *data_page)
1908 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1909 flags &= ~MS_MGC_MSK;
1911 /* Basic sanity checks */
1913 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1915 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1919 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1921 /* Default to relatime unless overriden */
1922 if (!(flags & MS_NOATIME))
1923 mnt_flags |= MNT_RELATIME;
1925 /* Separate the per-mountpoint flags */
1926 if (flags & MS_NOSUID)
1927 mnt_flags |= MNT_NOSUID;
1928 if (flags & MS_NODEV)
1929 mnt_flags |= MNT_NODEV;
1930 if (flags & MS_NOEXEC)
1931 mnt_flags |= MNT_NOEXEC;
1932 if (flags & MS_NOATIME)
1933 mnt_flags |= MNT_NOATIME;
1934 if (flags & MS_NODIRATIME)
1935 mnt_flags |= MNT_NODIRATIME;
1936 if (flags & MS_STRICTATIME)
1937 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1938 if (flags & MS_RDONLY)
1939 mnt_flags |= MNT_READONLY;
1941 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1942 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1945 /* ... and get the mountpoint */
1946 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1950 retval = security_sb_mount(dev_name, &path,
1951 type_page, flags, data_page);
1955 if (flags & MS_REMOUNT)
1956 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1958 else if (flags & MS_BIND)
1959 retval = do_loopback(&path, dev_name, flags & MS_REC);
1960 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1961 retval = do_change_type(&path, flags);
1962 else if (flags & MS_MOVE)
1963 retval = do_move_mount(&path, dev_name);
1965 retval = do_new_mount(&path, type_page, flags, mnt_flags,
1966 dev_name, data_page);
1973 * Allocate a new namespace structure and populate it with contents
1974 * copied from the namespace of the passed in task structure.
1976 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1977 struct fs_struct *fs)
1979 struct mnt_namespace *new_ns;
1980 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1981 struct vfsmount *p, *q;
1983 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1985 return ERR_PTR(-ENOMEM);
1987 atomic_set(&new_ns->count, 1);
1988 INIT_LIST_HEAD(&new_ns->list);
1989 init_waitqueue_head(&new_ns->poll);
1992 down_write(&namespace_sem);
1993 /* First pass: copy the tree topology */
1994 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1995 CL_COPY_ALL | CL_EXPIRE);
1996 if (!new_ns->root) {
1997 up_write(&namespace_sem);
1999 return ERR_PTR(-ENOMEM);
2001 spin_lock(&vfsmount_lock);
2002 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2003 spin_unlock(&vfsmount_lock);
2006 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2007 * as belonging to new namespace. We have already acquired a private
2008 * fs_struct, so tsk->fs->lock is not needed.
2015 if (p == fs->root.mnt) {
2017 fs->root.mnt = mntget(q);
2019 if (p == fs->pwd.mnt) {
2021 fs->pwd.mnt = mntget(q);
2024 p = next_mnt(p, mnt_ns->root);
2025 q = next_mnt(q, new_ns->root);
2027 up_write(&namespace_sem);
2037 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2038 struct fs_struct *new_fs)
2040 struct mnt_namespace *new_ns;
2045 if (!(flags & CLONE_NEWNS))
2048 new_ns = dup_mnt_ns(ns, new_fs);
2054 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2055 char __user *, type, unsigned long, flags, void __user *, data)
2058 unsigned long data_page;
2059 unsigned long type_page;
2060 unsigned long dev_page;
2063 retval = copy_mount_options(type, &type_page);
2067 dir_page = getname(dir_name);
2068 retval = PTR_ERR(dir_page);
2069 if (IS_ERR(dir_page))
2072 retval = copy_mount_options(dev_name, &dev_page);
2076 retval = copy_mount_options(data, &data_page);
2081 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2082 flags, (void *)data_page);
2084 free_page(data_page);
2087 free_page(dev_page);
2091 free_page(type_page);
2096 * pivot_root Semantics:
2097 * Moves the root file system of the current process to the directory put_old,
2098 * makes new_root as the new root file system of the current process, and sets
2099 * root/cwd of all processes which had them on the current root to new_root.
2102 * The new_root and put_old must be directories, and must not be on the
2103 * same file system as the current process root. The put_old must be
2104 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2105 * pointed to by put_old must yield the same directory as new_root. No other
2106 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2108 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2109 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2110 * in this situation.
2113 * - we don't move root/cwd if they are not at the root (reason: if something
2114 * cared enough to change them, it's probably wrong to force them elsewhere)
2115 * - it's okay to pick a root that isn't the root of a file system, e.g.
2116 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2117 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2120 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2121 const char __user *, put_old)
2123 struct vfsmount *tmp;
2124 struct path new, old, parent_path, root_parent, root;
2127 if (!capable(CAP_SYS_ADMIN))
2130 error = user_path_dir(new_root, &new);
2134 if (!check_mnt(new.mnt))
2137 error = user_path_dir(put_old, &old);
2141 error = security_sb_pivotroot(&old, &new);
2147 read_lock(¤t->fs->lock);
2148 root = current->fs->root;
2149 path_get(¤t->fs->root);
2150 read_unlock(¤t->fs->lock);
2151 down_write(&namespace_sem);
2152 mutex_lock(&old.dentry->d_inode->i_mutex);
2154 if (IS_MNT_SHARED(old.mnt) ||
2155 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2156 IS_MNT_SHARED(root.mnt->mnt_parent))
2158 if (!check_mnt(root.mnt))
2161 if (IS_DEADDIR(new.dentry->d_inode))
2163 if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
2165 if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
2168 if (new.mnt == root.mnt ||
2169 old.mnt == root.mnt)
2170 goto out2; /* loop, on the same file system */
2172 if (root.mnt->mnt_root != root.dentry)
2173 goto out2; /* not a mountpoint */
2174 if (root.mnt->mnt_parent == root.mnt)
2175 goto out2; /* not attached */
2176 if (new.mnt->mnt_root != new.dentry)
2177 goto out2; /* not a mountpoint */
2178 if (new.mnt->mnt_parent == new.mnt)
2179 goto out2; /* not attached */
2180 /* make sure we can reach put_old from new_root */
2182 spin_lock(&vfsmount_lock);
2183 if (tmp != new.mnt) {
2185 if (tmp->mnt_parent == tmp)
2186 goto out3; /* already mounted on put_old */
2187 if (tmp->mnt_parent == new.mnt)
2189 tmp = tmp->mnt_parent;
2191 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2193 } else if (!is_subdir(old.dentry, new.dentry))
2195 detach_mnt(new.mnt, &parent_path);
2196 detach_mnt(root.mnt, &root_parent);
2197 /* mount old root on put_old */
2198 attach_mnt(root.mnt, &old);
2199 /* mount new_root on / */
2200 attach_mnt(new.mnt, &root_parent);
2201 touch_mnt_namespace(current->nsproxy->mnt_ns);
2202 spin_unlock(&vfsmount_lock);
2203 chroot_fs_refs(&root, &new);
2204 security_sb_post_pivotroot(&root, &new);
2206 path_put(&root_parent);
2207 path_put(&parent_path);
2209 mutex_unlock(&old.dentry->d_inode->i_mutex);
2210 up_write(&namespace_sem);
2218 spin_unlock(&vfsmount_lock);
2222 static void __init init_mount_tree(void)
2224 struct vfsmount *mnt;
2225 struct mnt_namespace *ns;
2228 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2230 panic("Can't create rootfs");
2231 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2233 panic("Can't allocate initial namespace");
2234 atomic_set(&ns->count, 1);
2235 INIT_LIST_HEAD(&ns->list);
2236 init_waitqueue_head(&ns->poll);
2238 list_add(&mnt->mnt_list, &ns->list);
2242 init_task.nsproxy->mnt_ns = ns;
2245 root.mnt = ns->root;
2246 root.dentry = ns->root->mnt_root;
2248 set_fs_pwd(current->fs, &root);
2249 set_fs_root(current->fs, &root);
2252 void __init mnt_init(void)
2257 init_rwsem(&namespace_sem);
2259 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2260 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2262 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2264 if (!mount_hashtable)
2265 panic("Failed to allocate mount hash table\n");
2267 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2269 for (u = 0; u < HASH_SIZE; u++)
2270 INIT_LIST_HEAD(&mount_hashtable[u]);
2274 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2276 fs_kobj = kobject_create_and_add("fs", NULL);
2278 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2283 void __put_mnt_ns(struct mnt_namespace *ns)
2285 struct vfsmount *root = ns->root;
2286 LIST_HEAD(umount_list);
2288 spin_unlock(&vfsmount_lock);
2289 down_write(&namespace_sem);
2290 spin_lock(&vfsmount_lock);
2291 umount_tree(root, 0, &umount_list);
2292 spin_unlock(&vfsmount_lock);
2293 up_write(&namespace_sem);
2294 release_mounts(&umount_list);