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 <asm/uaccess.h>
31 #include <asm/unistd.h>
35 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
36 #define HASH_SIZE (1UL << HASH_SHIFT)
38 /* spinlock for vfsmount related operations, inplace of dcache_lock */
39 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
42 static DEFINE_IDA(mnt_id_ida);
43 static DEFINE_IDA(mnt_group_ida);
45 static struct list_head *mount_hashtable __read_mostly;
46 static struct kmem_cache *mnt_cache __read_mostly;
47 static struct rw_semaphore namespace_sem;
50 struct kobject *fs_kobj;
51 EXPORT_SYMBOL_GPL(fs_kobj);
53 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
55 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
56 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
57 tmp = tmp + (tmp >> HASH_SHIFT);
58 return tmp & (HASH_SIZE - 1);
61 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
63 /* allocation is serialized by namespace_sem */
64 static int mnt_alloc_id(struct vfsmount *mnt)
69 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
70 spin_lock(&vfsmount_lock);
71 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
72 spin_unlock(&vfsmount_lock);
79 static void mnt_free_id(struct vfsmount *mnt)
81 spin_lock(&vfsmount_lock);
82 ida_remove(&mnt_id_ida, mnt->mnt_id);
83 spin_unlock(&vfsmount_lock);
87 * Allocate a new peer group ID
89 * mnt_group_ida is protected by namespace_sem
91 static int mnt_alloc_group_id(struct vfsmount *mnt)
93 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
96 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
100 * Release a peer group ID
102 void mnt_release_group_id(struct vfsmount *mnt)
104 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
105 mnt->mnt_group_id = 0;
108 struct vfsmount *alloc_vfsmnt(const char *name)
110 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
114 err = mnt_alloc_id(mnt);
119 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
120 if (!mnt->mnt_devname)
124 atomic_set(&mnt->mnt_count, 1);
125 INIT_LIST_HEAD(&mnt->mnt_hash);
126 INIT_LIST_HEAD(&mnt->mnt_child);
127 INIT_LIST_HEAD(&mnt->mnt_mounts);
128 INIT_LIST_HEAD(&mnt->mnt_list);
129 INIT_LIST_HEAD(&mnt->mnt_expire);
130 INIT_LIST_HEAD(&mnt->mnt_share);
131 INIT_LIST_HEAD(&mnt->mnt_slave_list);
132 INIT_LIST_HEAD(&mnt->mnt_slave);
133 atomic_set(&mnt->__mnt_writers, 0);
140 kmem_cache_free(mnt_cache, mnt);
145 * Most r/o checks on a fs are for operations that take
146 * discrete amounts of time, like a write() or unlink().
147 * We must keep track of when those operations start
148 * (for permission checks) and when they end, so that
149 * we can determine when writes are able to occur to
153 * __mnt_is_readonly: check whether a mount is read-only
154 * @mnt: the mount to check for its write status
156 * This shouldn't be used directly ouside of the VFS.
157 * It does not guarantee that the filesystem will stay
158 * r/w, just that it is right *now*. This can not and
159 * should not be used in place of IS_RDONLY(inode).
160 * mnt_want/drop_write() will _keep_ the filesystem
163 int __mnt_is_readonly(struct vfsmount *mnt)
165 if (mnt->mnt_flags & MNT_READONLY)
167 if (mnt->mnt_sb->s_flags & MS_RDONLY)
171 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
175 * If holding multiple instances of this lock, they
176 * must be ordered by cpu number.
179 struct lock_class_key lock_class; /* compiles out with !lockdep */
181 struct vfsmount *mnt;
182 } ____cacheline_aligned_in_smp;
183 static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
185 static int __init init_mnt_writers(void)
188 for_each_possible_cpu(cpu) {
189 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
190 spin_lock_init(&writer->lock);
191 lockdep_set_class(&writer->lock, &writer->lock_class);
196 fs_initcall(init_mnt_writers);
198 static void unlock_mnt_writers(void)
201 struct mnt_writer *cpu_writer;
203 for_each_possible_cpu(cpu) {
204 cpu_writer = &per_cpu(mnt_writers, cpu);
205 spin_unlock(&cpu_writer->lock);
209 static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
211 if (!cpu_writer->mnt)
214 * This is in case anyone ever leaves an invalid,
215 * old ->mnt and a count of 0.
217 if (!cpu_writer->count)
219 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
220 cpu_writer->count = 0;
223 * must hold cpu_writer->lock
225 static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
226 struct vfsmount *mnt)
228 if (cpu_writer->mnt == mnt)
230 __clear_mnt_count(cpu_writer);
231 cpu_writer->mnt = mnt;
235 * Most r/o checks on a fs are for operations that take
236 * discrete amounts of time, like a write() or unlink().
237 * We must keep track of when those operations start
238 * (for permission checks) and when they end, so that
239 * we can determine when writes are able to occur to
243 * mnt_want_write - get write access to a mount
244 * @mnt: the mount on which to take a write
246 * This tells the low-level filesystem that a write is
247 * about to be performed to it, and makes sure that
248 * writes are allowed before returning success. When
249 * the write operation is finished, mnt_drop_write()
250 * must be called. This is effectively a refcount.
252 int mnt_want_write(struct vfsmount *mnt)
255 struct mnt_writer *cpu_writer;
257 cpu_writer = &get_cpu_var(mnt_writers);
258 spin_lock(&cpu_writer->lock);
259 if (__mnt_is_readonly(mnt)) {
263 use_cpu_writer_for_mount(cpu_writer, mnt);
266 spin_unlock(&cpu_writer->lock);
267 put_cpu_var(mnt_writers);
270 EXPORT_SYMBOL_GPL(mnt_want_write);
272 static void lock_mnt_writers(void)
275 struct mnt_writer *cpu_writer;
277 for_each_possible_cpu(cpu) {
278 cpu_writer = &per_cpu(mnt_writers, cpu);
279 spin_lock(&cpu_writer->lock);
280 __clear_mnt_count(cpu_writer);
281 cpu_writer->mnt = NULL;
286 * These per-cpu write counts are not guaranteed to have
287 * matched increments and decrements on any given cpu.
288 * A file open()ed for write on one cpu and close()d on
289 * another cpu will imbalance this count. Make sure it
290 * does not get too far out of whack.
292 static void handle_write_count_underflow(struct vfsmount *mnt)
294 if (atomic_read(&mnt->__mnt_writers) >=
295 MNT_WRITER_UNDERFLOW_LIMIT)
298 * It isn't necessary to hold all of the locks
299 * at the same time, but doing it this way makes
300 * us share a lot more code.
304 * vfsmount_lock is for mnt_flags.
306 spin_lock(&vfsmount_lock);
308 * If coalescing the per-cpu writer counts did not
309 * get us back to a positive writer count, we have
312 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
313 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
314 WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
316 mnt, atomic_read(&mnt->__mnt_writers));
317 /* use the flag to keep the dmesg spam down */
318 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
320 spin_unlock(&vfsmount_lock);
321 unlock_mnt_writers();
325 * mnt_drop_write - give up write access to a mount
326 * @mnt: the mount on which to give up write access
328 * Tells the low-level filesystem that we are done
329 * performing writes to it. Must be matched with
330 * mnt_want_write() call above.
332 void mnt_drop_write(struct vfsmount *mnt)
334 int must_check_underflow = 0;
335 struct mnt_writer *cpu_writer;
337 cpu_writer = &get_cpu_var(mnt_writers);
338 spin_lock(&cpu_writer->lock);
340 use_cpu_writer_for_mount(cpu_writer, mnt);
341 if (cpu_writer->count > 0) {
344 must_check_underflow = 1;
345 atomic_dec(&mnt->__mnt_writers);
348 spin_unlock(&cpu_writer->lock);
350 * Logically, we could call this each time,
351 * but the __mnt_writers cacheline tends to
352 * be cold, and makes this expensive.
354 if (must_check_underflow)
355 handle_write_count_underflow(mnt);
357 * This could be done right after the spinlock
358 * is taken because the spinlock keeps us on
359 * the cpu, and disables preemption. However,
360 * putting it here bounds the amount that
361 * __mnt_writers can underflow. Without it,
362 * we could theoretically wrap __mnt_writers.
364 put_cpu_var(mnt_writers);
366 EXPORT_SYMBOL_GPL(mnt_drop_write);
368 static int mnt_make_readonly(struct vfsmount *mnt)
374 * With all the locks held, this value is stable
376 if (atomic_read(&mnt->__mnt_writers) > 0) {
381 * nobody can do a successful mnt_want_write() with all
382 * of the counts in MNT_DENIED_WRITE and the locks held.
384 spin_lock(&vfsmount_lock);
386 mnt->mnt_flags |= MNT_READONLY;
387 spin_unlock(&vfsmount_lock);
389 unlock_mnt_writers();
393 static void __mnt_unmake_readonly(struct vfsmount *mnt)
395 spin_lock(&vfsmount_lock);
396 mnt->mnt_flags &= ~MNT_READONLY;
397 spin_unlock(&vfsmount_lock);
400 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
403 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 if (cpu_writer->mnt != mnt)
619 spin_lock(&cpu_writer->lock);
620 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
621 cpu_writer->count = 0;
623 * Might as well do this so that no one
624 * ever sees the pointer and expects
627 cpu_writer->mnt = NULL;
628 spin_unlock(&cpu_writer->lock);
631 * This probably indicates that somebody messed
632 * up a mnt_want/drop_write() pair. If this
633 * happens, the filesystem was probably unable
634 * to make r/w->r/o transitions.
636 WARN_ON(atomic_read(&mnt->__mnt_writers));
639 deactivate_super(sb);
642 void mntput_no_expire(struct vfsmount *mnt)
645 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
646 if (likely(!mnt->mnt_pinned)) {
647 spin_unlock(&vfsmount_lock);
651 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
653 spin_unlock(&vfsmount_lock);
654 acct_auto_close_mnt(mnt);
655 security_sb_umount_close(mnt);
660 EXPORT_SYMBOL(mntput_no_expire);
662 void mnt_pin(struct vfsmount *mnt)
664 spin_lock(&vfsmount_lock);
666 spin_unlock(&vfsmount_lock);
669 EXPORT_SYMBOL(mnt_pin);
671 void mnt_unpin(struct vfsmount *mnt)
673 spin_lock(&vfsmount_lock);
674 if (mnt->mnt_pinned) {
675 atomic_inc(&mnt->mnt_count);
678 spin_unlock(&vfsmount_lock);
681 EXPORT_SYMBOL(mnt_unpin);
683 static inline void mangle(struct seq_file *m, const char *s)
685 seq_escape(m, s, " \t\n\\");
689 * Simple .show_options callback for filesystems which don't want to
690 * implement more complex mount option showing.
692 * See also save_mount_options().
694 int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
696 const char *options = mnt->mnt_sb->s_options;
698 if (options != NULL && options[0]) {
705 EXPORT_SYMBOL(generic_show_options);
708 * If filesystem uses generic_show_options(), this function should be
709 * called from the fill_super() callback.
711 * The .remount_fs callback usually needs to be handled in a special
712 * way, to make sure, that previous options are not overwritten if the
715 * Also note, that if the filesystem's .remount_fs function doesn't
716 * reset all options to their default value, but changes only newly
717 * given options, then the displayed options will not reflect reality
720 void save_mount_options(struct super_block *sb, char *options)
722 kfree(sb->s_options);
723 sb->s_options = kstrdup(options, GFP_KERNEL);
725 EXPORT_SYMBOL(save_mount_options);
727 #ifdef CONFIG_PROC_FS
729 static void *m_start(struct seq_file *m, loff_t *pos)
731 struct proc_mounts *p = m->private;
733 down_read(&namespace_sem);
734 return seq_list_start(&p->ns->list, *pos);
737 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
739 struct proc_mounts *p = m->private;
741 return seq_list_next(v, &p->ns->list, pos);
744 static void m_stop(struct seq_file *m, void *v)
746 up_read(&namespace_sem);
749 struct proc_fs_info {
754 static int show_sb_opts(struct seq_file *m, struct super_block *sb)
756 static const struct proc_fs_info fs_info[] = {
757 { MS_SYNCHRONOUS, ",sync" },
758 { MS_DIRSYNC, ",dirsync" },
759 { MS_MANDLOCK, ",mand" },
762 const struct proc_fs_info *fs_infop;
764 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
765 if (sb->s_flags & fs_infop->flag)
766 seq_puts(m, fs_infop->str);
769 return security_sb_show_options(m, sb);
772 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
774 static const struct proc_fs_info mnt_info[] = {
775 { MNT_NOSUID, ",nosuid" },
776 { MNT_NODEV, ",nodev" },
777 { MNT_NOEXEC, ",noexec" },
778 { MNT_NOATIME, ",noatime" },
779 { MNT_NODIRATIME, ",nodiratime" },
780 { MNT_RELATIME, ",relatime" },
783 const struct proc_fs_info *fs_infop;
785 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
786 if (mnt->mnt_flags & fs_infop->flag)
787 seq_puts(m, fs_infop->str);
791 static void show_type(struct seq_file *m, struct super_block *sb)
793 mangle(m, sb->s_type->name);
794 if (sb->s_subtype && sb->s_subtype[0]) {
796 mangle(m, sb->s_subtype);
800 static int show_vfsmnt(struct seq_file *m, void *v)
802 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
804 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
806 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
808 seq_path(m, &mnt_path, " \t\n\\");
810 show_type(m, mnt->mnt_sb);
811 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
812 err = show_sb_opts(m, mnt->mnt_sb);
815 show_mnt_opts(m, mnt);
816 if (mnt->mnt_sb->s_op->show_options)
817 err = mnt->mnt_sb->s_op->show_options(m, mnt);
818 seq_puts(m, " 0 0\n");
823 const struct seq_operations mounts_op = {
830 static int show_mountinfo(struct seq_file *m, void *v)
832 struct proc_mounts *p = m->private;
833 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
834 struct super_block *sb = mnt->mnt_sb;
835 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
836 struct path root = p->root;
839 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
840 MAJOR(sb->s_dev), MINOR(sb->s_dev));
841 seq_dentry(m, mnt->mnt_root, " \t\n\\");
843 seq_path_root(m, &mnt_path, &root, " \t\n\\");
844 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
846 * Mountpoint is outside root, discard that one. Ugly,
847 * but less so than trying to do that in iterator in a
848 * race-free way (due to renames).
852 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
853 show_mnt_opts(m, mnt);
855 /* Tagged fields ("foo:X" or "bar") */
856 if (IS_MNT_SHARED(mnt))
857 seq_printf(m, " shared:%i", mnt->mnt_group_id);
858 if (IS_MNT_SLAVE(mnt)) {
859 int master = mnt->mnt_master->mnt_group_id;
860 int dom = get_dominating_id(mnt, &p->root);
861 seq_printf(m, " master:%i", master);
862 if (dom && dom != master)
863 seq_printf(m, " propagate_from:%i", dom);
865 if (IS_MNT_UNBINDABLE(mnt))
866 seq_puts(m, " unbindable");
868 /* Filesystem specific data */
872 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
873 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
874 err = show_sb_opts(m, sb);
877 if (sb->s_op->show_options)
878 err = sb->s_op->show_options(m, mnt);
884 const struct seq_operations mountinfo_op = {
888 .show = show_mountinfo,
891 static int show_vfsstat(struct seq_file *m, void *v)
893 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
894 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
898 if (mnt->mnt_devname) {
899 seq_puts(m, "device ");
900 mangle(m, mnt->mnt_devname);
902 seq_puts(m, "no device");
905 seq_puts(m, " mounted on ");
906 seq_path(m, &mnt_path, " \t\n\\");
909 /* file system type */
910 seq_puts(m, "with fstype ");
911 show_type(m, mnt->mnt_sb);
913 /* optional statistics */
914 if (mnt->mnt_sb->s_op->show_stats) {
916 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
923 const struct seq_operations mountstats_op = {
927 .show = show_vfsstat,
929 #endif /* CONFIG_PROC_FS */
932 * may_umount_tree - check if a mount tree is busy
933 * @mnt: root of mount tree
935 * This is called to check if a tree of mounts has any
936 * open files, pwds, chroots or sub mounts that are
939 int may_umount_tree(struct vfsmount *mnt)
942 int minimum_refs = 0;
945 spin_lock(&vfsmount_lock);
946 for (p = mnt; p; p = next_mnt(p, mnt)) {
947 actual_refs += atomic_read(&p->mnt_count);
950 spin_unlock(&vfsmount_lock);
952 if (actual_refs > minimum_refs)
958 EXPORT_SYMBOL(may_umount_tree);
961 * may_umount - check if a mount point is busy
962 * @mnt: root of mount
964 * This is called to check if a mount point has any
965 * open files, pwds, chroots or sub mounts. If the
966 * mount has sub mounts this will return busy
967 * regardless of whether the sub mounts are busy.
969 * Doesn't take quota and stuff into account. IOW, in some cases it will
970 * give false negatives. The main reason why it's here is that we need
971 * a non-destructive way to look for easily umountable filesystems.
973 int may_umount(struct vfsmount *mnt)
976 spin_lock(&vfsmount_lock);
977 if (propagate_mount_busy(mnt, 2))
979 spin_unlock(&vfsmount_lock);
983 EXPORT_SYMBOL(may_umount);
985 void release_mounts(struct list_head *head)
987 struct vfsmount *mnt;
988 while (!list_empty(head)) {
989 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
990 list_del_init(&mnt->mnt_hash);
991 if (mnt->mnt_parent != mnt) {
992 struct dentry *dentry;
994 spin_lock(&vfsmount_lock);
995 dentry = mnt->mnt_mountpoint;
997 mnt->mnt_mountpoint = mnt->mnt_root;
998 mnt->mnt_parent = mnt;
1000 spin_unlock(&vfsmount_lock);
1008 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1012 for (p = mnt; p; p = next_mnt(p, mnt))
1013 list_move(&p->mnt_hash, kill);
1016 propagate_umount(kill);
1018 list_for_each_entry(p, kill, mnt_hash) {
1019 list_del_init(&p->mnt_expire);
1020 list_del_init(&p->mnt_list);
1021 __touch_mnt_namespace(p->mnt_ns);
1023 list_del_init(&p->mnt_child);
1024 if (p->mnt_parent != p) {
1025 p->mnt_parent->mnt_ghosts++;
1026 p->mnt_mountpoint->d_mounted--;
1028 change_mnt_propagation(p, MS_PRIVATE);
1032 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1034 static int do_umount(struct vfsmount *mnt, int flags)
1036 struct super_block *sb = mnt->mnt_sb;
1038 LIST_HEAD(umount_list);
1040 retval = security_sb_umount(mnt, flags);
1045 * Allow userspace to request a mountpoint be expired rather than
1046 * unmounting unconditionally. Unmount only happens if:
1047 * (1) the mark is already set (the mark is cleared by mntput())
1048 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1050 if (flags & MNT_EXPIRE) {
1051 if (mnt == current->fs->root.mnt ||
1052 flags & (MNT_FORCE | MNT_DETACH))
1055 if (atomic_read(&mnt->mnt_count) != 2)
1058 if (!xchg(&mnt->mnt_expiry_mark, 1))
1063 * If we may have to abort operations to get out of this
1064 * mount, and they will themselves hold resources we must
1065 * allow the fs to do things. In the Unix tradition of
1066 * 'Gee thats tricky lets do it in userspace' the umount_begin
1067 * might fail to complete on the first run through as other tasks
1068 * must return, and the like. Thats for the mount program to worry
1069 * about for the moment.
1072 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1074 sb->s_op->umount_begin(sb);
1079 * No sense to grab the lock for this test, but test itself looks
1080 * somewhat bogus. Suggestions for better replacement?
1081 * Ho-hum... In principle, we might treat that as umount + switch
1082 * to rootfs. GC would eventually take care of the old vfsmount.
1083 * Actually it makes sense, especially if rootfs would contain a
1084 * /reboot - static binary that would close all descriptors and
1085 * call reboot(9). Then init(8) could umount root and exec /reboot.
1087 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1089 * Special case for "unmounting" root ...
1090 * we just try to remount it readonly.
1092 down_write(&sb->s_umount);
1093 if (!(sb->s_flags & MS_RDONLY)) {
1095 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1098 up_write(&sb->s_umount);
1102 down_write(&namespace_sem);
1103 spin_lock(&vfsmount_lock);
1106 if (!(flags & MNT_DETACH))
1107 shrink_submounts(mnt, &umount_list);
1110 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1111 if (!list_empty(&mnt->mnt_list))
1112 umount_tree(mnt, 1, &umount_list);
1115 spin_unlock(&vfsmount_lock);
1117 security_sb_umount_busy(mnt);
1118 up_write(&namespace_sem);
1119 release_mounts(&umount_list);
1124 * Now umount can handle mount points as well as block devices.
1125 * This is important for filesystems which use unnamed block devices.
1127 * We now support a flag for forced unmount like the other 'big iron'
1128 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1131 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1136 retval = user_path(name, &path);
1140 if (path.dentry != path.mnt->mnt_root)
1142 if (!check_mnt(path.mnt))
1146 if (!capable(CAP_SYS_ADMIN))
1149 retval = do_umount(path.mnt, flags);
1151 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1153 mntput_no_expire(path.mnt);
1158 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1161 * The 2.0 compatible umount. No flags.
1163 SYSCALL_DEFINE1(oldumount, char __user *, name)
1165 return sys_umount(name, 0);
1170 static int mount_is_safe(struct nameidata *nd)
1172 if (capable(CAP_SYS_ADMIN))
1176 if (S_ISLNK(nd->path.dentry->d_inode->i_mode))
1178 if (nd->path.dentry->d_inode->i_mode & S_ISVTX) {
1179 if (current->uid != nd->path.dentry->d_inode->i_uid)
1182 if (vfs_permission(nd, MAY_WRITE))
1188 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1191 struct vfsmount *res, *p, *q, *r, *s;
1194 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1197 res = q = clone_mnt(mnt, dentry, flag);
1200 q->mnt_mountpoint = mnt->mnt_mountpoint;
1203 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1204 if (!is_subdir(r->mnt_mountpoint, dentry))
1207 for (s = r; s; s = next_mnt(s, r)) {
1208 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1209 s = skip_mnt_tree(s);
1212 while (p != s->mnt_parent) {
1218 path.dentry = p->mnt_mountpoint;
1219 q = clone_mnt(p, p->mnt_root, flag);
1222 spin_lock(&vfsmount_lock);
1223 list_add_tail(&q->mnt_list, &res->mnt_list);
1224 attach_mnt(q, &path);
1225 spin_unlock(&vfsmount_lock);
1231 LIST_HEAD(umount_list);
1232 spin_lock(&vfsmount_lock);
1233 umount_tree(res, 0, &umount_list);
1234 spin_unlock(&vfsmount_lock);
1235 release_mounts(&umount_list);
1240 struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1242 struct vfsmount *tree;
1243 down_write(&namespace_sem);
1244 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
1245 up_write(&namespace_sem);
1249 void drop_collected_mounts(struct vfsmount *mnt)
1251 LIST_HEAD(umount_list);
1252 down_write(&namespace_sem);
1253 spin_lock(&vfsmount_lock);
1254 umount_tree(mnt, 0, &umount_list);
1255 spin_unlock(&vfsmount_lock);
1256 up_write(&namespace_sem);
1257 release_mounts(&umount_list);
1260 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1264 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1265 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1266 mnt_release_group_id(p);
1270 static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1274 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1275 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1276 int err = mnt_alloc_group_id(p);
1278 cleanup_group_ids(mnt, p);
1288 * @source_mnt : mount tree to be attached
1289 * @nd : place the mount tree @source_mnt is attached
1290 * @parent_nd : if non-null, detach the source_mnt from its parent and
1291 * store the parent mount and mountpoint dentry.
1292 * (done when source_mnt is moved)
1294 * NOTE: in the table below explains the semantics when a source mount
1295 * of a given type is attached to a destination mount of a given type.
1296 * ---------------------------------------------------------------------------
1297 * | BIND MOUNT OPERATION |
1298 * |**************************************************************************
1299 * | source-->| shared | private | slave | unbindable |
1303 * |**************************************************************************
1304 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1306 * |non-shared| shared (+) | private | slave (*) | invalid |
1307 * ***************************************************************************
1308 * A bind operation clones the source mount and mounts the clone on the
1309 * destination mount.
1311 * (++) the cloned mount is propagated to all the mounts in the propagation
1312 * tree of the destination mount and the cloned mount is added to
1313 * the peer group of the source mount.
1314 * (+) the cloned mount is created under the destination mount and is marked
1315 * as shared. The cloned mount is added to the peer group of the source
1317 * (+++) the mount is propagated to all the mounts in the propagation tree
1318 * of the destination mount and the cloned mount is made slave
1319 * of the same master as that of the source mount. The cloned mount
1320 * is marked as 'shared and slave'.
1321 * (*) the cloned mount is made a slave of the same master as that of the
1324 * ---------------------------------------------------------------------------
1325 * | MOVE MOUNT OPERATION |
1326 * |**************************************************************************
1327 * | source-->| shared | private | slave | unbindable |
1331 * |**************************************************************************
1332 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1334 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1335 * ***************************************************************************
1337 * (+) the mount is moved to the destination. And is then propagated to
1338 * all the mounts in the propagation tree of the destination mount.
1339 * (+*) the mount is moved to the destination.
1340 * (+++) the mount is moved to the destination and is then propagated to
1341 * all the mounts belonging to the destination mount's propagation tree.
1342 * the mount is marked as 'shared and slave'.
1343 * (*) the mount continues to be a slave at the new location.
1345 * if the source mount is a tree, the operations explained above is
1346 * applied to each mount in the tree.
1347 * Must be called without spinlocks held, since this function can sleep
1350 static int attach_recursive_mnt(struct vfsmount *source_mnt,
1351 struct path *path, struct path *parent_path)
1353 LIST_HEAD(tree_list);
1354 struct vfsmount *dest_mnt = path->mnt;
1355 struct dentry *dest_dentry = path->dentry;
1356 struct vfsmount *child, *p;
1359 if (IS_MNT_SHARED(dest_mnt)) {
1360 err = invent_group_ids(source_mnt, true);
1364 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1366 goto out_cleanup_ids;
1368 if (IS_MNT_SHARED(dest_mnt)) {
1369 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1373 spin_lock(&vfsmount_lock);
1375 detach_mnt(source_mnt, parent_path);
1376 attach_mnt(source_mnt, path);
1377 touch_mnt_namespace(current->nsproxy->mnt_ns);
1379 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1380 commit_tree(source_mnt);
1383 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1384 list_del_init(&child->mnt_hash);
1387 spin_unlock(&vfsmount_lock);
1391 if (IS_MNT_SHARED(dest_mnt))
1392 cleanup_group_ids(source_mnt, NULL);
1397 static int graft_tree(struct vfsmount *mnt, struct path *path)
1400 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1403 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1404 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1408 mutex_lock(&path->dentry->d_inode->i_mutex);
1409 if (IS_DEADDIR(path->dentry->d_inode))
1412 err = security_sb_check_sb(mnt, path);
1417 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1418 err = attach_recursive_mnt(mnt, path, NULL);
1420 mutex_unlock(&path->dentry->d_inode->i_mutex);
1422 security_sb_post_addmount(mnt, path);
1427 * recursively change the type of the mountpoint.
1428 * noinline this do_mount helper to save do_mount stack space.
1430 static noinline int do_change_type(struct nameidata *nd, int flag)
1432 struct vfsmount *m, *mnt = nd->path.mnt;
1433 int recurse = flag & MS_REC;
1434 int type = flag & ~MS_REC;
1437 if (!capable(CAP_SYS_ADMIN))
1440 if (nd->path.dentry != nd->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.
1462 * noinline this do_mount helper to save do_mount stack space.
1464 static noinline int do_loopback(struct nameidata *nd, char *old_name,
1467 struct nameidata old_nd;
1468 struct vfsmount *mnt = NULL;
1469 int err = mount_is_safe(nd);
1472 if (!old_name || !*old_name)
1474 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1478 down_write(&namespace_sem);
1480 if (IS_MNT_UNBINDABLE(old_nd.path.mnt))
1483 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
1488 mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0);
1490 mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0);
1495 err = graft_tree(mnt, &nd->path);
1497 LIST_HEAD(umount_list);
1498 spin_lock(&vfsmount_lock);
1499 umount_tree(mnt, 0, &umount_list);
1500 spin_unlock(&vfsmount_lock);
1501 release_mounts(&umount_list);
1505 up_write(&namespace_sem);
1506 path_put(&old_nd.path);
1510 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1513 int readonly_request = 0;
1515 if (ms_flags & MS_RDONLY)
1516 readonly_request = 1;
1517 if (readonly_request == __mnt_is_readonly(mnt))
1520 if (readonly_request)
1521 error = mnt_make_readonly(mnt);
1523 __mnt_unmake_readonly(mnt);
1528 * change filesystem flags. dir should be a physical root of filesystem.
1529 * If you've mounted a non-root directory somewhere and want to do remount
1530 * on it - tough luck.
1531 * noinline this do_mount helper to save do_mount stack space.
1533 static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags,
1537 struct super_block *sb = nd->path.mnt->mnt_sb;
1539 if (!capable(CAP_SYS_ADMIN))
1542 if (!check_mnt(nd->path.mnt))
1545 if (nd->path.dentry != nd->path.mnt->mnt_root)
1548 down_write(&sb->s_umount);
1549 if (flags & MS_BIND)
1550 err = change_mount_flags(nd->path.mnt, flags);
1552 err = do_remount_sb(sb, flags, data, 0);
1554 nd->path.mnt->mnt_flags = mnt_flags;
1555 up_write(&sb->s_umount);
1557 security_sb_post_remount(nd->path.mnt, flags, data);
1559 spin_lock(&vfsmount_lock);
1560 touch_mnt_namespace(nd->path.mnt->mnt_ns);
1561 spin_unlock(&vfsmount_lock);
1566 static inline int tree_contains_unbindable(struct vfsmount *mnt)
1569 for (p = mnt; p; p = next_mnt(p, mnt)) {
1570 if (IS_MNT_UNBINDABLE(p))
1577 * noinline this do_mount helper to save do_mount stack space.
1579 static noinline int do_move_mount(struct nameidata *nd, char *old_name)
1581 struct nameidata old_nd;
1582 struct path parent_path;
1585 if (!capable(CAP_SYS_ADMIN))
1587 if (!old_name || !*old_name)
1589 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
1593 down_write(&namespace_sem);
1594 while (d_mountpoint(nd->path.dentry) &&
1595 follow_down(&nd->path.mnt, &nd->path.dentry))
1598 if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt))
1602 mutex_lock(&nd->path.dentry->d_inode->i_mutex);
1603 if (IS_DEADDIR(nd->path.dentry->d_inode))
1606 if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry))
1610 if (old_nd.path.dentry != old_nd.path.mnt->mnt_root)
1613 if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent)
1616 if (S_ISDIR(nd->path.dentry->d_inode->i_mode) !=
1617 S_ISDIR(old_nd.path.dentry->d_inode->i_mode))
1620 * Don't move a mount residing in a shared parent.
1622 if (old_nd.path.mnt->mnt_parent &&
1623 IS_MNT_SHARED(old_nd.path.mnt->mnt_parent))
1626 * Don't move a mount tree containing unbindable mounts to a destination
1627 * mount which is shared.
1629 if (IS_MNT_SHARED(nd->path.mnt) &&
1630 tree_contains_unbindable(old_nd.path.mnt))
1633 for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent)
1634 if (p == old_nd.path.mnt)
1637 err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path);
1641 /* if the mount is moved, it should no longer be expire
1643 list_del_init(&old_nd.path.mnt->mnt_expire);
1645 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
1647 up_write(&namespace_sem);
1649 path_put(&parent_path);
1650 path_put(&old_nd.path);
1655 * create a new mount for userspace and request it to be added into the
1657 * noinline this do_mount helper to save do_mount stack space.
1659 static noinline int do_new_mount(struct nameidata *nd, char *type, int flags,
1660 int mnt_flags, char *name, void *data)
1662 struct vfsmount *mnt;
1664 if (!type || !memchr(type, 0, PAGE_SIZE))
1667 /* we need capabilities... */
1668 if (!capable(CAP_SYS_ADMIN))
1671 mnt = do_kern_mount(type, flags, name, data);
1673 return PTR_ERR(mnt);
1675 return do_add_mount(mnt, &nd->path, mnt_flags, NULL);
1679 * add a mount into a namespace's mount tree
1680 * - provide the option of adding the new mount to an expiration list
1682 int do_add_mount(struct vfsmount *newmnt, struct path *path,
1683 int mnt_flags, struct list_head *fslist)
1687 down_write(&namespace_sem);
1688 /* Something was mounted here while we slept */
1689 while (d_mountpoint(path->dentry) &&
1690 follow_down(&path->mnt, &path->dentry))
1693 if (!check_mnt(path->mnt))
1696 /* Refuse the same filesystem on the same mount point */
1698 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1699 path->mnt->mnt_root == path->dentry)
1703 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1706 newmnt->mnt_flags = mnt_flags;
1707 if ((err = graft_tree(newmnt, path)))
1710 if (fslist) /* add to the specified expiration list */
1711 list_add_tail(&newmnt->mnt_expire, fslist);
1713 up_write(&namespace_sem);
1717 up_write(&namespace_sem);
1722 EXPORT_SYMBOL_GPL(do_add_mount);
1725 * process a list of expirable mountpoints with the intent of discarding any
1726 * mountpoints that aren't in use and haven't been touched since last we came
1729 void mark_mounts_for_expiry(struct list_head *mounts)
1731 struct vfsmount *mnt, *next;
1732 LIST_HEAD(graveyard);
1735 if (list_empty(mounts))
1738 down_write(&namespace_sem);
1739 spin_lock(&vfsmount_lock);
1741 /* extract from the expiration list every vfsmount that matches the
1742 * following criteria:
1743 * - only referenced by its parent vfsmount
1744 * - still marked for expiry (marked on the last call here; marks are
1745 * cleared by mntput())
1747 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1748 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1749 propagate_mount_busy(mnt, 1))
1751 list_move(&mnt->mnt_expire, &graveyard);
1753 while (!list_empty(&graveyard)) {
1754 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1755 touch_mnt_namespace(mnt->mnt_ns);
1756 umount_tree(mnt, 1, &umounts);
1758 spin_unlock(&vfsmount_lock);
1759 up_write(&namespace_sem);
1761 release_mounts(&umounts);
1764 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1767 * Ripoff of 'select_parent()'
1769 * search the list of submounts for a given mountpoint, and move any
1770 * shrinkable submounts to the 'graveyard' list.
1772 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1774 struct vfsmount *this_parent = parent;
1775 struct list_head *next;
1779 next = this_parent->mnt_mounts.next;
1781 while (next != &this_parent->mnt_mounts) {
1782 struct list_head *tmp = next;
1783 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1786 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1789 * Descend a level if the d_mounts list is non-empty.
1791 if (!list_empty(&mnt->mnt_mounts)) {
1796 if (!propagate_mount_busy(mnt, 1)) {
1797 list_move_tail(&mnt->mnt_expire, graveyard);
1802 * All done at this level ... ascend and resume the search
1804 if (this_parent != parent) {
1805 next = this_parent->mnt_child.next;
1806 this_parent = this_parent->mnt_parent;
1813 * process a list of expirable mountpoints with the intent of discarding any
1814 * submounts of a specific parent mountpoint
1816 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1818 LIST_HEAD(graveyard);
1821 /* extract submounts of 'mountpoint' from the expiration list */
1822 while (select_submounts(mnt, &graveyard)) {
1823 while (!list_empty(&graveyard)) {
1824 m = list_first_entry(&graveyard, struct vfsmount,
1826 touch_mnt_namespace(mnt->mnt_ns);
1827 umount_tree(mnt, 1, umounts);
1833 * Some copy_from_user() implementations do not return the exact number of
1834 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1835 * Note that this function differs from copy_from_user() in that it will oops
1836 * on bad values of `to', rather than returning a short copy.
1838 static long exact_copy_from_user(void *to, const void __user * from,
1842 const char __user *f = from;
1845 if (!access_ok(VERIFY_READ, from, n))
1849 if (__get_user(c, f)) {
1860 int copy_mount_options(const void __user * data, unsigned long *where)
1870 if (!(page = __get_free_page(GFP_KERNEL)))
1873 /* We only care that *some* data at the address the user
1874 * gave us is valid. Just in case, we'll zero
1875 * the remainder of the page.
1877 /* copy_from_user cannot cross TASK_SIZE ! */
1878 size = TASK_SIZE - (unsigned long)data;
1879 if (size > PAGE_SIZE)
1882 i = size - exact_copy_from_user((void *)page, data, size);
1888 memset((char *)page + i, 0, PAGE_SIZE - i);
1894 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1895 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1897 * data is a (void *) that can point to any structure up to
1898 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1899 * information (or be NULL).
1901 * Pre-0.97 versions of mount() didn't have a flags word.
1902 * When the flags word was introduced its top half was required
1903 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1904 * Therefore, if this magic number is present, it carries no information
1905 * and must be discarded.
1907 long do_mount(char *dev_name, char *dir_name, char *type_page,
1908 unsigned long flags, void *data_page)
1910 struct nameidata nd;
1915 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1916 flags &= ~MS_MGC_MSK;
1918 /* Basic sanity checks */
1920 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1922 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1926 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1928 /* Separate the per-mountpoint flags */
1929 if (flags & MS_NOSUID)
1930 mnt_flags |= MNT_NOSUID;
1931 if (flags & MS_NODEV)
1932 mnt_flags |= MNT_NODEV;
1933 if (flags & MS_NOEXEC)
1934 mnt_flags |= MNT_NOEXEC;
1935 if (flags & MS_NOATIME)
1936 mnt_flags |= MNT_NOATIME;
1937 if (flags & MS_NODIRATIME)
1938 mnt_flags |= MNT_NODIRATIME;
1939 if (flags & MS_RELATIME)
1940 mnt_flags |= MNT_RELATIME;
1941 if (flags & MS_RDONLY)
1942 mnt_flags |= MNT_READONLY;
1944 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1945 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
1947 /* ... and get the mountpoint */
1948 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1952 retval = security_sb_mount(dev_name, &nd.path,
1953 type_page, flags, data_page);
1957 if (flags & MS_REMOUNT)
1958 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1960 else if (flags & MS_BIND)
1961 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1962 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1963 retval = do_change_type(&nd, flags);
1964 else if (flags & MS_MOVE)
1965 retval = do_move_mount(&nd, dev_name);
1967 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1968 dev_name, data_page);
1975 * Allocate a new namespace structure and populate it with contents
1976 * copied from the namespace of the passed in task structure.
1978 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1979 struct fs_struct *fs)
1981 struct mnt_namespace *new_ns;
1982 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1983 struct vfsmount *p, *q;
1985 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1987 return ERR_PTR(-ENOMEM);
1989 atomic_set(&new_ns->count, 1);
1990 INIT_LIST_HEAD(&new_ns->list);
1991 init_waitqueue_head(&new_ns->poll);
1994 down_write(&namespace_sem);
1995 /* First pass: copy the tree topology */
1996 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1997 CL_COPY_ALL | CL_EXPIRE);
1998 if (!new_ns->root) {
1999 up_write(&namespace_sem);
2001 return ERR_PTR(-ENOMEM);;
2003 spin_lock(&vfsmount_lock);
2004 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2005 spin_unlock(&vfsmount_lock);
2008 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2009 * as belonging to new namespace. We have already acquired a private
2010 * fs_struct, so tsk->fs->lock is not needed.
2017 if (p == fs->root.mnt) {
2019 fs->root.mnt = mntget(q);
2021 if (p == fs->pwd.mnt) {
2023 fs->pwd.mnt = mntget(q);
2026 p = next_mnt(p, mnt_ns->root);
2027 q = next_mnt(q, new_ns->root);
2029 up_write(&namespace_sem);
2039 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2040 struct fs_struct *new_fs)
2042 struct mnt_namespace *new_ns;
2047 if (!(flags & CLONE_NEWNS))
2050 new_ns = dup_mnt_ns(ns, new_fs);
2056 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2057 char __user *, type, unsigned long, flags, void __user *, data)
2060 unsigned long data_page;
2061 unsigned long type_page;
2062 unsigned long dev_page;
2065 retval = copy_mount_options(type, &type_page);
2069 dir_page = getname(dir_name);
2070 retval = PTR_ERR(dir_page);
2071 if (IS_ERR(dir_page))
2074 retval = copy_mount_options(dev_name, &dev_page);
2078 retval = copy_mount_options(data, &data_page);
2083 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2084 flags, (void *)data_page);
2086 free_page(data_page);
2089 free_page(dev_page);
2093 free_page(type_page);
2098 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2099 * It can block. Requires the big lock held.
2101 void set_fs_root(struct fs_struct *fs, struct path *path)
2103 struct path old_root;
2105 write_lock(&fs->lock);
2106 old_root = fs->root;
2109 write_unlock(&fs->lock);
2110 if (old_root.dentry)
2111 path_put(&old_root);
2115 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2116 * It can block. Requires the big lock held.
2118 void set_fs_pwd(struct fs_struct *fs, struct path *path)
2120 struct path old_pwd;
2122 write_lock(&fs->lock);
2126 write_unlock(&fs->lock);
2132 static void chroot_fs_refs(struct path *old_root, struct path *new_root)
2134 struct task_struct *g, *p;
2135 struct fs_struct *fs;
2137 read_lock(&tasklist_lock);
2138 do_each_thread(g, p) {
2142 atomic_inc(&fs->count);
2144 if (fs->root.dentry == old_root->dentry
2145 && fs->root.mnt == old_root->mnt)
2146 set_fs_root(fs, new_root);
2147 if (fs->pwd.dentry == old_root->dentry
2148 && fs->pwd.mnt == old_root->mnt)
2149 set_fs_pwd(fs, new_root);
2153 } while_each_thread(g, p);
2154 read_unlock(&tasklist_lock);
2158 * pivot_root Semantics:
2159 * Moves the root file system of the current process to the directory put_old,
2160 * makes new_root as the new root file system of the current process, and sets
2161 * root/cwd of all processes which had them on the current root to new_root.
2164 * The new_root and put_old must be directories, and must not be on the
2165 * same file system as the current process root. The put_old must be
2166 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2167 * pointed to by put_old must yield the same directory as new_root. No other
2168 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2170 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2171 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2172 * in this situation.
2175 * - we don't move root/cwd if they are not at the root (reason: if something
2176 * cared enough to change them, it's probably wrong to force them elsewhere)
2177 * - it's okay to pick a root that isn't the root of a file system, e.g.
2178 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2179 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2182 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2183 const char __user *, put_old)
2185 struct vfsmount *tmp;
2186 struct path new, old, parent_path, root_parent, root;
2189 if (!capable(CAP_SYS_ADMIN))
2192 error = user_path_dir(new_root, &new);
2196 if (!check_mnt(new.mnt))
2199 error = user_path_dir(put_old, &old);
2203 error = security_sb_pivotroot(&old, &new);
2209 read_lock(¤t->fs->lock);
2210 root = current->fs->root;
2211 path_get(¤t->fs->root);
2212 read_unlock(¤t->fs->lock);
2213 down_write(&namespace_sem);
2214 mutex_lock(&old.dentry->d_inode->i_mutex);
2216 if (IS_MNT_SHARED(old.mnt) ||
2217 IS_MNT_SHARED(new.mnt->mnt_parent) ||
2218 IS_MNT_SHARED(root.mnt->mnt_parent))
2220 if (!check_mnt(root.mnt))
2223 if (IS_DEADDIR(new.dentry->d_inode))
2225 if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
2227 if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
2230 if (new.mnt == root.mnt ||
2231 old.mnt == root.mnt)
2232 goto out2; /* loop, on the same file system */
2234 if (root.mnt->mnt_root != root.dentry)
2235 goto out2; /* not a mountpoint */
2236 if (root.mnt->mnt_parent == root.mnt)
2237 goto out2; /* not attached */
2238 if (new.mnt->mnt_root != new.dentry)
2239 goto out2; /* not a mountpoint */
2240 if (new.mnt->mnt_parent == new.mnt)
2241 goto out2; /* not attached */
2242 /* make sure we can reach put_old from new_root */
2244 spin_lock(&vfsmount_lock);
2245 if (tmp != new.mnt) {
2247 if (tmp->mnt_parent == tmp)
2248 goto out3; /* already mounted on put_old */
2249 if (tmp->mnt_parent == new.mnt)
2251 tmp = tmp->mnt_parent;
2253 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2255 } else if (!is_subdir(old.dentry, new.dentry))
2257 detach_mnt(new.mnt, &parent_path);
2258 detach_mnt(root.mnt, &root_parent);
2259 /* mount old root on put_old */
2260 attach_mnt(root.mnt, &old);
2261 /* mount new_root on / */
2262 attach_mnt(new.mnt, &root_parent);
2263 touch_mnt_namespace(current->nsproxy->mnt_ns);
2264 spin_unlock(&vfsmount_lock);
2265 chroot_fs_refs(&root, &new);
2266 security_sb_post_pivotroot(&root, &new);
2268 path_put(&root_parent);
2269 path_put(&parent_path);
2271 mutex_unlock(&old.dentry->d_inode->i_mutex);
2272 up_write(&namespace_sem);
2280 spin_unlock(&vfsmount_lock);
2284 static void __init init_mount_tree(void)
2286 struct vfsmount *mnt;
2287 struct mnt_namespace *ns;
2290 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2292 panic("Can't create rootfs");
2293 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2295 panic("Can't allocate initial namespace");
2296 atomic_set(&ns->count, 1);
2297 INIT_LIST_HEAD(&ns->list);
2298 init_waitqueue_head(&ns->poll);
2300 list_add(&mnt->mnt_list, &ns->list);
2304 init_task.nsproxy->mnt_ns = ns;
2307 root.mnt = ns->root;
2308 root.dentry = ns->root->mnt_root;
2310 set_fs_pwd(current->fs, &root);
2311 set_fs_root(current->fs, &root);
2314 void __init mnt_init(void)
2319 init_rwsem(&namespace_sem);
2321 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2322 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2324 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2326 if (!mount_hashtable)
2327 panic("Failed to allocate mount hash table\n");
2329 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2331 for (u = 0; u < HASH_SIZE; u++)
2332 INIT_LIST_HEAD(&mount_hashtable[u]);
2336 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2338 fs_kobj = kobject_create_and_add("fs", NULL);
2340 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2345 void __put_mnt_ns(struct mnt_namespace *ns)
2347 struct vfsmount *root = ns->root;
2348 LIST_HEAD(umount_list);
2350 spin_unlock(&vfsmount_lock);
2351 down_write(&namespace_sem);
2352 spin_lock(&vfsmount_lock);
2353 umount_tree(root, 0, &umount_list);
2354 spin_unlock(&vfsmount_lock);
2355 up_write(&namespace_sem);
2356 release_mounts(&umount_list);