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/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/acct.h> /* acct_auto_close_mnt */
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
29 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
30 #define HASH_SIZE (1UL << HASH_SHIFT)
33 static DEFINE_IDA(mnt_id_ida);
34 static DEFINE_IDA(mnt_group_ida);
35 static DEFINE_SPINLOCK(mnt_id_lock);
36 static int mnt_id_start = 0;
37 static int mnt_group_start = 1;
39 static struct list_head *mount_hashtable __read_mostly;
40 static struct list_head *mountpoint_hashtable __read_mostly;
41 static struct kmem_cache *mnt_cache __read_mostly;
42 static DECLARE_RWSEM(namespace_sem);
45 struct kobject *fs_kobj;
46 EXPORT_SYMBOL_GPL(fs_kobj);
49 * vfsmount lock may be taken for read to prevent changes to the
50 * vfsmount hash, ie. during mountpoint lookups or walking back
53 * It should be taken for write in all cases where the vfsmount
54 * tree or hash is modified or when a vfsmount structure is modified.
56 DEFINE_BRLOCK(vfsmount_lock);
58 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
60 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
62 tmp = tmp + (tmp >> HASH_SHIFT);
63 return tmp & (HASH_SIZE - 1);
66 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
69 * allocation is serialized by namespace_sem, but we need the spinlock to
70 * serialize with freeing.
72 static int mnt_alloc_id(struct mount *mnt)
77 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
78 spin_lock(&mnt_id_lock);
79 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
81 mnt_id_start = mnt->mnt_id + 1;
82 spin_unlock(&mnt_id_lock);
89 static void mnt_free_id(struct mount *mnt)
92 spin_lock(&mnt_id_lock);
93 ida_remove(&mnt_id_ida, id);
94 if (mnt_id_start > id)
96 spin_unlock(&mnt_id_lock);
100 * Allocate a new peer group ID
102 * mnt_group_ida is protected by namespace_sem
104 static int mnt_alloc_group_id(struct mount *mnt)
108 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
111 res = ida_get_new_above(&mnt_group_ida,
115 mnt_group_start = mnt->mnt_group_id + 1;
121 * Release a peer group ID
123 void mnt_release_group_id(struct mount *mnt)
125 int id = mnt->mnt_group_id;
126 ida_remove(&mnt_group_ida, id);
127 if (mnt_group_start > id)
128 mnt_group_start = id;
129 mnt->mnt_group_id = 0;
133 * vfsmount lock must be held for read
135 static inline void mnt_add_count(struct mount *mnt, int n)
138 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
147 * vfsmount lock must be held for write
149 unsigned int mnt_get_count(struct mount *mnt)
152 unsigned int count = 0;
155 for_each_possible_cpu(cpu) {
156 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
161 return mnt->mnt_count;
165 static struct mount *alloc_vfsmnt(const char *name)
167 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
171 err = mnt_alloc_id(mnt);
176 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
177 if (!mnt->mnt_devname)
182 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
184 goto out_free_devname;
186 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
189 mnt->mnt_writers = 0;
192 INIT_LIST_HEAD(&mnt->mnt_hash);
193 INIT_LIST_HEAD(&mnt->mnt_child);
194 INIT_LIST_HEAD(&mnt->mnt_mounts);
195 INIT_LIST_HEAD(&mnt->mnt_list);
196 INIT_LIST_HEAD(&mnt->mnt_expire);
197 INIT_LIST_HEAD(&mnt->mnt_share);
198 INIT_LIST_HEAD(&mnt->mnt_slave_list);
199 INIT_LIST_HEAD(&mnt->mnt_slave);
200 #ifdef CONFIG_FSNOTIFY
201 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
208 kfree(mnt->mnt_devname);
213 kmem_cache_free(mnt_cache, mnt);
218 * Most r/o checks on a fs are for operations that take
219 * discrete amounts of time, like a write() or unlink().
220 * We must keep track of when those operations start
221 * (for permission checks) and when they end, so that
222 * we can determine when writes are able to occur to
226 * __mnt_is_readonly: check whether a mount is read-only
227 * @mnt: the mount to check for its write status
229 * This shouldn't be used directly ouside of the VFS.
230 * It does not guarantee that the filesystem will stay
231 * r/w, just that it is right *now*. This can not and
232 * should not be used in place of IS_RDONLY(inode).
233 * mnt_want/drop_write() will _keep_ the filesystem
236 int __mnt_is_readonly(struct vfsmount *mnt)
238 if (mnt->mnt_flags & MNT_READONLY)
240 if (mnt->mnt_sb->s_flags & MS_RDONLY)
244 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
246 static inline void mnt_inc_writers(struct mount *mnt)
249 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
255 static inline void mnt_dec_writers(struct mount *mnt)
258 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
264 static unsigned int mnt_get_writers(struct mount *mnt)
267 unsigned int count = 0;
270 for_each_possible_cpu(cpu) {
271 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
276 return mnt->mnt_writers;
280 static int mnt_is_readonly(struct vfsmount *mnt)
282 if (mnt->mnt_sb->s_readonly_remount)
284 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
286 return __mnt_is_readonly(mnt);
290 * Most r/o & frozen checks on a fs are for operations that take discrete
291 * amounts of time, like a write() or unlink(). We must keep track of when
292 * those operations start (for permission checks) and when they end, so that we
293 * can determine when writes are able to occur to a filesystem.
296 * __mnt_want_write - get write access to a mount without freeze protection
297 * @m: the mount on which to take a write
299 * This tells the low-level filesystem that a write is about to be performed to
300 * it, and makes sure that writes are allowed (mnt it read-write) before
301 * returning success. This operation does not protect against filesystem being
302 * frozen. When the write operation is finished, __mnt_drop_write() must be
303 * called. This is effectively a refcount.
305 int __mnt_want_write(struct vfsmount *m)
307 struct mount *mnt = real_mount(m);
311 mnt_inc_writers(mnt);
313 * The store to mnt_inc_writers must be visible before we pass
314 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
315 * incremented count after it has set MNT_WRITE_HOLD.
318 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
321 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
322 * be set to match its requirements. So we must not load that until
323 * MNT_WRITE_HOLD is cleared.
326 if (mnt_is_readonly(m)) {
327 mnt_dec_writers(mnt);
336 * mnt_want_write - get write access to a mount
337 * @m: the mount on which to take a write
339 * This tells the low-level filesystem that a write is about to be performed to
340 * it, and makes sure that writes are allowed (mount is read-write, filesystem
341 * is not frozen) before returning success. When the write operation is
342 * finished, mnt_drop_write() must be called. This is effectively a refcount.
344 int mnt_want_write(struct vfsmount *m)
348 sb_start_write(m->mnt_sb);
349 ret = __mnt_want_write(m);
351 sb_end_write(m->mnt_sb);
354 EXPORT_SYMBOL_GPL(mnt_want_write);
357 * mnt_clone_write - get write access to a mount
358 * @mnt: the mount on which to take a write
360 * This is effectively like mnt_want_write, except
361 * it must only be used to take an extra write reference
362 * on a mountpoint that we already know has a write reference
363 * on it. This allows some optimisation.
365 * After finished, mnt_drop_write must be called as usual to
366 * drop the reference.
368 int mnt_clone_write(struct vfsmount *mnt)
370 /* superblock may be r/o */
371 if (__mnt_is_readonly(mnt))
374 mnt_inc_writers(real_mount(mnt));
378 EXPORT_SYMBOL_GPL(mnt_clone_write);
381 * __mnt_want_write_file - get write access to a file's mount
382 * @file: the file who's mount on which to take a write
384 * This is like __mnt_want_write, but it takes a file and can
385 * do some optimisations if the file is open for write already
387 int __mnt_want_write_file(struct file *file)
389 struct inode *inode = file_inode(file);
391 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
392 return __mnt_want_write(file->f_path.mnt);
394 return mnt_clone_write(file->f_path.mnt);
398 * mnt_want_write_file - get write access to a file's mount
399 * @file: the file who's mount on which to take a write
401 * This is like mnt_want_write, but it takes a file and can
402 * do some optimisations if the file is open for write already
404 int mnt_want_write_file(struct file *file)
408 sb_start_write(file->f_path.mnt->mnt_sb);
409 ret = __mnt_want_write_file(file);
411 sb_end_write(file->f_path.mnt->mnt_sb);
414 EXPORT_SYMBOL_GPL(mnt_want_write_file);
417 * __mnt_drop_write - give up write access to a mount
418 * @mnt: the mount on which to give up write access
420 * Tells the low-level filesystem that we are done
421 * performing writes to it. Must be matched with
422 * __mnt_want_write() call above.
424 void __mnt_drop_write(struct vfsmount *mnt)
427 mnt_dec_writers(real_mount(mnt));
432 * mnt_drop_write - give up write access to a mount
433 * @mnt: the mount on which to give up write access
435 * Tells the low-level filesystem that we are done performing writes to it and
436 * also allows filesystem to be frozen again. Must be matched with
437 * mnt_want_write() call above.
439 void mnt_drop_write(struct vfsmount *mnt)
441 __mnt_drop_write(mnt);
442 sb_end_write(mnt->mnt_sb);
444 EXPORT_SYMBOL_GPL(mnt_drop_write);
446 void __mnt_drop_write_file(struct file *file)
448 __mnt_drop_write(file->f_path.mnt);
451 void mnt_drop_write_file(struct file *file)
453 mnt_drop_write(file->f_path.mnt);
455 EXPORT_SYMBOL(mnt_drop_write_file);
457 static int mnt_make_readonly(struct mount *mnt)
461 br_write_lock(&vfsmount_lock);
462 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
464 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
465 * should be visible before we do.
470 * With writers on hold, if this value is zero, then there are
471 * definitely no active writers (although held writers may subsequently
472 * increment the count, they'll have to wait, and decrement it after
473 * seeing MNT_READONLY).
475 * It is OK to have counter incremented on one CPU and decremented on
476 * another: the sum will add up correctly. The danger would be when we
477 * sum up each counter, if we read a counter before it is incremented,
478 * but then read another CPU's count which it has been subsequently
479 * decremented from -- we would see more decrements than we should.
480 * MNT_WRITE_HOLD protects against this scenario, because
481 * mnt_want_write first increments count, then smp_mb, then spins on
482 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
483 * we're counting up here.
485 if (mnt_get_writers(mnt) > 0)
488 mnt->mnt.mnt_flags |= MNT_READONLY;
490 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
491 * that become unheld will see MNT_READONLY.
494 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
495 br_write_unlock(&vfsmount_lock);
499 static void __mnt_unmake_readonly(struct mount *mnt)
501 br_write_lock(&vfsmount_lock);
502 mnt->mnt.mnt_flags &= ~MNT_READONLY;
503 br_write_unlock(&vfsmount_lock);
506 int sb_prepare_remount_readonly(struct super_block *sb)
511 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
512 if (atomic_long_read(&sb->s_remove_count))
515 br_write_lock(&vfsmount_lock);
516 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
517 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
518 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
520 if (mnt_get_writers(mnt) > 0) {
526 if (!err && atomic_long_read(&sb->s_remove_count))
530 sb->s_readonly_remount = 1;
533 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
534 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
535 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
537 br_write_unlock(&vfsmount_lock);
542 static void free_vfsmnt(struct mount *mnt)
544 kfree(mnt->mnt_devname);
547 free_percpu(mnt->mnt_pcp);
549 kmem_cache_free(mnt_cache, mnt);
553 * find the first or last mount at @dentry on vfsmount @mnt depending on
554 * @dir. If @dir is set return the first mount else return the last mount.
555 * vfsmount_lock must be held for read or write.
557 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
560 struct list_head *head = mount_hashtable + hash(mnt, dentry);
561 struct list_head *tmp = head;
562 struct mount *p, *found = NULL;
565 tmp = dir ? tmp->next : tmp->prev;
569 p = list_entry(tmp, struct mount, mnt_hash);
570 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
579 * lookup_mnt - Return the first child mount mounted at path
581 * "First" means first mounted chronologically. If you create the
584 * mount /dev/sda1 /mnt
585 * mount /dev/sda2 /mnt
586 * mount /dev/sda3 /mnt
588 * Then lookup_mnt() on the base /mnt dentry in the root mount will
589 * return successively the root dentry and vfsmount of /dev/sda1, then
590 * /dev/sda2, then /dev/sda3, then NULL.
592 * lookup_mnt takes a reference to the found vfsmount.
594 struct vfsmount *lookup_mnt(struct path *path)
596 struct mount *child_mnt;
598 br_read_lock(&vfsmount_lock);
599 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
601 mnt_add_count(child_mnt, 1);
602 br_read_unlock(&vfsmount_lock);
603 return &child_mnt->mnt;
605 br_read_unlock(&vfsmount_lock);
610 static struct mountpoint *new_mountpoint(struct dentry *dentry)
612 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
613 struct mountpoint *mp;
616 list_for_each_entry(mp, chain, m_hash) {
617 if (mp->m_dentry == dentry) {
618 /* might be worth a WARN_ON() */
619 if (d_unlinked(dentry))
620 return ERR_PTR(-ENOENT);
626 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
628 return ERR_PTR(-ENOMEM);
630 ret = d_set_mounted(dentry);
636 mp->m_dentry = dentry;
638 list_add(&mp->m_hash, chain);
642 static void put_mountpoint(struct mountpoint *mp)
644 if (!--mp->m_count) {
645 struct dentry *dentry = mp->m_dentry;
646 spin_lock(&dentry->d_lock);
647 dentry->d_flags &= ~DCACHE_MOUNTED;
648 spin_unlock(&dentry->d_lock);
649 list_del(&mp->m_hash);
654 static inline int check_mnt(struct mount *mnt)
656 return mnt->mnt_ns == current->nsproxy->mnt_ns;
660 * vfsmount lock must be held for write
662 static void touch_mnt_namespace(struct mnt_namespace *ns)
666 wake_up_interruptible(&ns->poll);
671 * vfsmount lock must be held for write
673 static void __touch_mnt_namespace(struct mnt_namespace *ns)
675 if (ns && ns->event != event) {
677 wake_up_interruptible(&ns->poll);
682 * vfsmount lock must be held for write
684 static void detach_mnt(struct mount *mnt, struct path *old_path)
686 old_path->dentry = mnt->mnt_mountpoint;
687 old_path->mnt = &mnt->mnt_parent->mnt;
688 mnt->mnt_parent = mnt;
689 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
690 list_del_init(&mnt->mnt_child);
691 list_del_init(&mnt->mnt_hash);
692 put_mountpoint(mnt->mnt_mp);
697 * vfsmount lock must be held for write
699 void mnt_set_mountpoint(struct mount *mnt,
700 struct mountpoint *mp,
701 struct mount *child_mnt)
704 mnt_add_count(mnt, 1); /* essentially, that's mntget */
705 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
706 child_mnt->mnt_parent = mnt;
707 child_mnt->mnt_mp = mp;
711 * vfsmount lock must be held for write
713 static void attach_mnt(struct mount *mnt,
714 struct mount *parent,
715 struct mountpoint *mp)
717 mnt_set_mountpoint(parent, mp, mnt);
718 list_add_tail(&mnt->mnt_hash, mount_hashtable +
719 hash(&parent->mnt, mp->m_dentry));
720 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
724 * vfsmount lock must be held for write
726 static void commit_tree(struct mount *mnt)
728 struct mount *parent = mnt->mnt_parent;
731 struct mnt_namespace *n = parent->mnt_ns;
733 BUG_ON(parent == mnt);
735 list_add_tail(&head, &mnt->mnt_list);
736 list_for_each_entry(m, &head, mnt_list)
739 list_splice(&head, n->list.prev);
741 list_add_tail(&mnt->mnt_hash, mount_hashtable +
742 hash(&parent->mnt, mnt->mnt_mountpoint));
743 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
744 touch_mnt_namespace(n);
747 static struct mount *next_mnt(struct mount *p, struct mount *root)
749 struct list_head *next = p->mnt_mounts.next;
750 if (next == &p->mnt_mounts) {
754 next = p->mnt_child.next;
755 if (next != &p->mnt_parent->mnt_mounts)
760 return list_entry(next, struct mount, mnt_child);
763 static struct mount *skip_mnt_tree(struct mount *p)
765 struct list_head *prev = p->mnt_mounts.prev;
766 while (prev != &p->mnt_mounts) {
767 p = list_entry(prev, struct mount, mnt_child);
768 prev = p->mnt_mounts.prev;
774 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
780 return ERR_PTR(-ENODEV);
782 mnt = alloc_vfsmnt(name);
784 return ERR_PTR(-ENOMEM);
786 if (flags & MS_KERNMOUNT)
787 mnt->mnt.mnt_flags = MNT_INTERNAL;
789 root = mount_fs(type, flags, name, data);
792 return ERR_CAST(root);
795 mnt->mnt.mnt_root = root;
796 mnt->mnt.mnt_sb = root->d_sb;
797 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
798 mnt->mnt_parent = mnt;
799 br_write_lock(&vfsmount_lock);
800 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
801 br_write_unlock(&vfsmount_lock);
804 EXPORT_SYMBOL_GPL(vfs_kern_mount);
806 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
809 struct super_block *sb = old->mnt.mnt_sb;
813 mnt = alloc_vfsmnt(old->mnt_devname);
815 return ERR_PTR(-ENOMEM);
817 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
818 mnt->mnt_group_id = 0; /* not a peer of original */
820 mnt->mnt_group_id = old->mnt_group_id;
822 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
823 err = mnt_alloc_group_id(mnt);
828 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
829 /* Don't allow unprivileged users to change mount flags */
830 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
831 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
833 /* Don't allow unprivileged users to reveal what is under a mount */
834 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
835 mnt->mnt.mnt_flags |= MNT_LOCKED;
837 atomic_inc(&sb->s_active);
838 mnt->mnt.mnt_sb = sb;
839 mnt->mnt.mnt_root = dget(root);
840 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
841 mnt->mnt_parent = mnt;
842 br_write_lock(&vfsmount_lock);
843 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
844 br_write_unlock(&vfsmount_lock);
846 if ((flag & CL_SLAVE) ||
847 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
848 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
849 mnt->mnt_master = old;
850 CLEAR_MNT_SHARED(mnt);
851 } else if (!(flag & CL_PRIVATE)) {
852 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
853 list_add(&mnt->mnt_share, &old->mnt_share);
854 if (IS_MNT_SLAVE(old))
855 list_add(&mnt->mnt_slave, &old->mnt_slave);
856 mnt->mnt_master = old->mnt_master;
858 if (flag & CL_MAKE_SHARED)
861 /* stick the duplicate mount on the same expiry list
862 * as the original if that was on one */
863 if (flag & CL_EXPIRE) {
864 if (!list_empty(&old->mnt_expire))
865 list_add(&mnt->mnt_expire, &old->mnt_expire);
875 static void mntput_no_expire(struct mount *mnt)
879 br_read_lock(&vfsmount_lock);
880 if (likely(mnt->mnt_ns)) {
881 /* shouldn't be the last one */
882 mnt_add_count(mnt, -1);
883 br_read_unlock(&vfsmount_lock);
886 br_read_unlock(&vfsmount_lock);
888 br_write_lock(&vfsmount_lock);
889 mnt_add_count(mnt, -1);
890 if (mnt_get_count(mnt)) {
891 br_write_unlock(&vfsmount_lock);
895 mnt_add_count(mnt, -1);
896 if (likely(mnt_get_count(mnt)))
898 br_write_lock(&vfsmount_lock);
900 if (unlikely(mnt->mnt_pinned)) {
901 mnt_add_count(mnt, mnt->mnt_pinned + 1);
903 br_write_unlock(&vfsmount_lock);
904 acct_auto_close_mnt(&mnt->mnt);
908 list_del(&mnt->mnt_instance);
909 br_write_unlock(&vfsmount_lock);
912 * This probably indicates that somebody messed
913 * up a mnt_want/drop_write() pair. If this
914 * happens, the filesystem was probably unable
915 * to make r/w->r/o transitions.
918 * The locking used to deal with mnt_count decrement provides barriers,
919 * so mnt_get_writers() below is safe.
921 WARN_ON(mnt_get_writers(mnt));
922 fsnotify_vfsmount_delete(&mnt->mnt);
923 dput(mnt->mnt.mnt_root);
924 deactivate_super(mnt->mnt.mnt_sb);
928 void mntput(struct vfsmount *mnt)
931 struct mount *m = real_mount(mnt);
932 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
933 if (unlikely(m->mnt_expiry_mark))
934 m->mnt_expiry_mark = 0;
938 EXPORT_SYMBOL(mntput);
940 struct vfsmount *mntget(struct vfsmount *mnt)
943 mnt_add_count(real_mount(mnt), 1);
946 EXPORT_SYMBOL(mntget);
948 void mnt_pin(struct vfsmount *mnt)
950 br_write_lock(&vfsmount_lock);
951 real_mount(mnt)->mnt_pinned++;
952 br_write_unlock(&vfsmount_lock);
954 EXPORT_SYMBOL(mnt_pin);
956 void mnt_unpin(struct vfsmount *m)
958 struct mount *mnt = real_mount(m);
959 br_write_lock(&vfsmount_lock);
960 if (mnt->mnt_pinned) {
961 mnt_add_count(mnt, 1);
964 br_write_unlock(&vfsmount_lock);
966 EXPORT_SYMBOL(mnt_unpin);
968 static inline void mangle(struct seq_file *m, const char *s)
970 seq_escape(m, s, " \t\n\\");
974 * Simple .show_options callback for filesystems which don't want to
975 * implement more complex mount option showing.
977 * See also save_mount_options().
979 int generic_show_options(struct seq_file *m, struct dentry *root)
984 options = rcu_dereference(root->d_sb->s_options);
986 if (options != NULL && options[0]) {
994 EXPORT_SYMBOL(generic_show_options);
997 * If filesystem uses generic_show_options(), this function should be
998 * called from the fill_super() callback.
1000 * The .remount_fs callback usually needs to be handled in a special
1001 * way, to make sure, that previous options are not overwritten if the
1004 * Also note, that if the filesystem's .remount_fs function doesn't
1005 * reset all options to their default value, but changes only newly
1006 * given options, then the displayed options will not reflect reality
1009 void save_mount_options(struct super_block *sb, char *options)
1011 BUG_ON(sb->s_options);
1012 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1014 EXPORT_SYMBOL(save_mount_options);
1016 void replace_mount_options(struct super_block *sb, char *options)
1018 char *old = sb->s_options;
1019 rcu_assign_pointer(sb->s_options, options);
1025 EXPORT_SYMBOL(replace_mount_options);
1027 #ifdef CONFIG_PROC_FS
1028 /* iterator; we want it to have access to namespace_sem, thus here... */
1029 static void *m_start(struct seq_file *m, loff_t *pos)
1031 struct proc_mounts *p = proc_mounts(m);
1033 down_read(&namespace_sem);
1034 return seq_list_start(&p->ns->list, *pos);
1037 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1039 struct proc_mounts *p = proc_mounts(m);
1041 return seq_list_next(v, &p->ns->list, pos);
1044 static void m_stop(struct seq_file *m, void *v)
1046 up_read(&namespace_sem);
1049 static int m_show(struct seq_file *m, void *v)
1051 struct proc_mounts *p = proc_mounts(m);
1052 struct mount *r = list_entry(v, struct mount, mnt_list);
1053 return p->show(m, &r->mnt);
1056 const struct seq_operations mounts_op = {
1062 #endif /* CONFIG_PROC_FS */
1065 * may_umount_tree - check if a mount tree is busy
1066 * @mnt: root of mount tree
1068 * This is called to check if a tree of mounts has any
1069 * open files, pwds, chroots or sub mounts that are
1072 int may_umount_tree(struct vfsmount *m)
1074 struct mount *mnt = real_mount(m);
1075 int actual_refs = 0;
1076 int minimum_refs = 0;
1080 /* write lock needed for mnt_get_count */
1081 br_write_lock(&vfsmount_lock);
1082 for (p = mnt; p; p = next_mnt(p, mnt)) {
1083 actual_refs += mnt_get_count(p);
1086 br_write_unlock(&vfsmount_lock);
1088 if (actual_refs > minimum_refs)
1094 EXPORT_SYMBOL(may_umount_tree);
1097 * may_umount - check if a mount point is busy
1098 * @mnt: root of mount
1100 * This is called to check if a mount point has any
1101 * open files, pwds, chroots or sub mounts. If the
1102 * mount has sub mounts this will return busy
1103 * regardless of whether the sub mounts are busy.
1105 * Doesn't take quota and stuff into account. IOW, in some cases it will
1106 * give false negatives. The main reason why it's here is that we need
1107 * a non-destructive way to look for easily umountable filesystems.
1109 int may_umount(struct vfsmount *mnt)
1112 down_read(&namespace_sem);
1113 br_write_lock(&vfsmount_lock);
1114 if (propagate_mount_busy(real_mount(mnt), 2))
1116 br_write_unlock(&vfsmount_lock);
1117 up_read(&namespace_sem);
1121 EXPORT_SYMBOL(may_umount);
1123 static LIST_HEAD(unmounted); /* protected by namespace_sem */
1125 static void namespace_unlock(void)
1130 if (likely(list_empty(&unmounted))) {
1131 up_write(&namespace_sem);
1135 list_splice_init(&unmounted, &head);
1136 up_write(&namespace_sem);
1138 while (!list_empty(&head)) {
1139 mnt = list_first_entry(&head, struct mount, mnt_hash);
1140 list_del_init(&mnt->mnt_hash);
1141 if (mnt_has_parent(mnt)) {
1142 struct dentry *dentry;
1145 br_write_lock(&vfsmount_lock);
1146 dentry = mnt->mnt_mountpoint;
1147 m = mnt->mnt_parent;
1148 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1149 mnt->mnt_parent = mnt;
1151 br_write_unlock(&vfsmount_lock);
1159 static inline void namespace_lock(void)
1161 down_write(&namespace_sem);
1165 * vfsmount lock must be held for write
1166 * namespace_sem must be held for write
1168 void umount_tree(struct mount *mnt, int propagate)
1170 LIST_HEAD(tmp_list);
1173 for (p = mnt; p; p = next_mnt(p, mnt))
1174 list_move(&p->mnt_hash, &tmp_list);
1177 propagate_umount(&tmp_list);
1179 list_for_each_entry(p, &tmp_list, mnt_hash) {
1180 list_del_init(&p->mnt_expire);
1181 list_del_init(&p->mnt_list);
1182 __touch_mnt_namespace(p->mnt_ns);
1184 list_del_init(&p->mnt_child);
1185 if (mnt_has_parent(p)) {
1186 p->mnt_parent->mnt_ghosts++;
1187 put_mountpoint(p->mnt_mp);
1190 change_mnt_propagation(p, MS_PRIVATE);
1192 list_splice(&tmp_list, &unmounted);
1195 static void shrink_submounts(struct mount *mnt);
1197 static int do_umount(struct mount *mnt, int flags)
1199 struct super_block *sb = mnt->mnt.mnt_sb;
1202 retval = security_sb_umount(&mnt->mnt, flags);
1207 * Allow userspace to request a mountpoint be expired rather than
1208 * unmounting unconditionally. Unmount only happens if:
1209 * (1) the mark is already set (the mark is cleared by mntput())
1210 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1212 if (flags & MNT_EXPIRE) {
1213 if (&mnt->mnt == current->fs->root.mnt ||
1214 flags & (MNT_FORCE | MNT_DETACH))
1218 * probably don't strictly need the lock here if we examined
1219 * all race cases, but it's a slowpath.
1221 br_write_lock(&vfsmount_lock);
1222 if (mnt_get_count(mnt) != 2) {
1223 br_write_unlock(&vfsmount_lock);
1226 br_write_unlock(&vfsmount_lock);
1228 if (!xchg(&mnt->mnt_expiry_mark, 1))
1233 * If we may have to abort operations to get out of this
1234 * mount, and they will themselves hold resources we must
1235 * allow the fs to do things. In the Unix tradition of
1236 * 'Gee thats tricky lets do it in userspace' the umount_begin
1237 * might fail to complete on the first run through as other tasks
1238 * must return, and the like. Thats for the mount program to worry
1239 * about for the moment.
1242 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1243 sb->s_op->umount_begin(sb);
1247 * No sense to grab the lock for this test, but test itself looks
1248 * somewhat bogus. Suggestions for better replacement?
1249 * Ho-hum... In principle, we might treat that as umount + switch
1250 * to rootfs. GC would eventually take care of the old vfsmount.
1251 * Actually it makes sense, especially if rootfs would contain a
1252 * /reboot - static binary that would close all descriptors and
1253 * call reboot(9). Then init(8) could umount root and exec /reboot.
1255 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1257 * Special case for "unmounting" root ...
1258 * we just try to remount it readonly.
1260 down_write(&sb->s_umount);
1261 if (!(sb->s_flags & MS_RDONLY))
1262 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1263 up_write(&sb->s_umount);
1268 br_write_lock(&vfsmount_lock);
1271 if (!(flags & MNT_DETACH))
1272 shrink_submounts(mnt);
1275 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1276 if (!list_empty(&mnt->mnt_list))
1277 umount_tree(mnt, 1);
1280 br_write_unlock(&vfsmount_lock);
1286 * Is the caller allowed to modify his namespace?
1288 static inline bool may_mount(void)
1290 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1294 * Now umount can handle mount points as well as block devices.
1295 * This is important for filesystems which use unnamed block devices.
1297 * We now support a flag for forced unmount like the other 'big iron'
1298 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1301 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1306 int lookup_flags = 0;
1308 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1314 if (!(flags & UMOUNT_NOFOLLOW))
1315 lookup_flags |= LOOKUP_FOLLOW;
1317 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1320 mnt = real_mount(path.mnt);
1322 if (path.dentry != path.mnt->mnt_root)
1324 if (!check_mnt(mnt))
1326 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1329 retval = do_umount(mnt, flags);
1331 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1333 mntput_no_expire(mnt);
1338 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1341 * The 2.0 compatible umount. No flags.
1343 SYSCALL_DEFINE1(oldumount, char __user *, name)
1345 return sys_umount(name, 0);
1350 static bool is_mnt_ns_file(struct dentry *dentry)
1352 /* Is this a proxy for a mount namespace? */
1353 struct inode *inode = dentry->d_inode;
1356 if (!proc_ns_inode(inode))
1359 ei = get_proc_ns(inode);
1360 if (ei->ns_ops != &mntns_operations)
1366 static bool mnt_ns_loop(struct dentry *dentry)
1368 /* Could bind mounting the mount namespace inode cause a
1369 * mount namespace loop?
1371 struct mnt_namespace *mnt_ns;
1372 if (!is_mnt_ns_file(dentry))
1375 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1376 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1379 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1382 struct mount *res, *p, *q, *r, *parent;
1384 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1385 return ERR_PTR(-EINVAL);
1387 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1388 return ERR_PTR(-EINVAL);
1390 res = q = clone_mnt(mnt, dentry, flag);
1394 q->mnt.mnt_flags &= ~MNT_LOCKED;
1395 q->mnt_mountpoint = mnt->mnt_mountpoint;
1398 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1400 if (!is_subdir(r->mnt_mountpoint, dentry))
1403 for (s = r; s; s = next_mnt(s, r)) {
1404 if (!(flag & CL_COPY_UNBINDABLE) &&
1405 IS_MNT_UNBINDABLE(s)) {
1406 s = skip_mnt_tree(s);
1409 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1410 is_mnt_ns_file(s->mnt.mnt_root)) {
1411 s = skip_mnt_tree(s);
1414 while (p != s->mnt_parent) {
1420 q = clone_mnt(p, p->mnt.mnt_root, flag);
1423 br_write_lock(&vfsmount_lock);
1424 list_add_tail(&q->mnt_list, &res->mnt_list);
1425 attach_mnt(q, parent, p->mnt_mp);
1426 br_write_unlock(&vfsmount_lock);
1432 br_write_lock(&vfsmount_lock);
1433 umount_tree(res, 0);
1434 br_write_unlock(&vfsmount_lock);
1439 /* Caller should check returned pointer for errors */
1441 struct vfsmount *collect_mounts(struct path *path)
1445 tree = copy_tree(real_mount(path->mnt), path->dentry,
1446 CL_COPY_ALL | CL_PRIVATE);
1449 return ERR_CAST(tree);
1453 void drop_collected_mounts(struct vfsmount *mnt)
1456 br_write_lock(&vfsmount_lock);
1457 umount_tree(real_mount(mnt), 0);
1458 br_write_unlock(&vfsmount_lock);
1462 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1463 struct vfsmount *root)
1466 int res = f(root, arg);
1469 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1470 res = f(&mnt->mnt, arg);
1477 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1481 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1482 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1483 mnt_release_group_id(p);
1487 static int invent_group_ids(struct mount *mnt, bool recurse)
1491 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1492 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1493 int err = mnt_alloc_group_id(p);
1495 cleanup_group_ids(mnt, p);
1505 * @source_mnt : mount tree to be attached
1506 * @nd : place the mount tree @source_mnt is attached
1507 * @parent_nd : if non-null, detach the source_mnt from its parent and
1508 * store the parent mount and mountpoint dentry.
1509 * (done when source_mnt is moved)
1511 * NOTE: in the table below explains the semantics when a source mount
1512 * of a given type is attached to a destination mount of a given type.
1513 * ---------------------------------------------------------------------------
1514 * | BIND MOUNT OPERATION |
1515 * |**************************************************************************
1516 * | source-->| shared | private | slave | unbindable |
1520 * |**************************************************************************
1521 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1523 * |non-shared| shared (+) | private | slave (*) | invalid |
1524 * ***************************************************************************
1525 * A bind operation clones the source mount and mounts the clone on the
1526 * destination mount.
1528 * (++) the cloned mount is propagated to all the mounts in the propagation
1529 * tree of the destination mount and the cloned mount is added to
1530 * the peer group of the source mount.
1531 * (+) the cloned mount is created under the destination mount and is marked
1532 * as shared. The cloned mount is added to the peer group of the source
1534 * (+++) the mount is propagated to all the mounts in the propagation tree
1535 * of the destination mount and the cloned mount is made slave
1536 * of the same master as that of the source mount. The cloned mount
1537 * is marked as 'shared and slave'.
1538 * (*) the cloned mount is made a slave of the same master as that of the
1541 * ---------------------------------------------------------------------------
1542 * | MOVE MOUNT OPERATION |
1543 * |**************************************************************************
1544 * | source-->| shared | private | slave | unbindable |
1548 * |**************************************************************************
1549 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1551 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1552 * ***************************************************************************
1554 * (+) the mount is moved to the destination. And is then propagated to
1555 * all the mounts in the propagation tree of the destination mount.
1556 * (+*) the mount is moved to the destination.
1557 * (+++) the mount is moved to the destination and is then propagated to
1558 * all the mounts belonging to the destination mount's propagation tree.
1559 * the mount is marked as 'shared and slave'.
1560 * (*) the mount continues to be a slave at the new location.
1562 * if the source mount is a tree, the operations explained above is
1563 * applied to each mount in the tree.
1564 * Must be called without spinlocks held, since this function can sleep
1567 static int attach_recursive_mnt(struct mount *source_mnt,
1568 struct mount *dest_mnt,
1569 struct mountpoint *dest_mp,
1570 struct path *parent_path)
1572 LIST_HEAD(tree_list);
1573 struct mount *child, *p;
1576 if (IS_MNT_SHARED(dest_mnt)) {
1577 err = invent_group_ids(source_mnt, true);
1581 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1583 goto out_cleanup_ids;
1585 br_write_lock(&vfsmount_lock);
1587 if (IS_MNT_SHARED(dest_mnt)) {
1588 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1592 detach_mnt(source_mnt, parent_path);
1593 attach_mnt(source_mnt, dest_mnt, dest_mp);
1594 touch_mnt_namespace(source_mnt->mnt_ns);
1596 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1597 commit_tree(source_mnt);
1600 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1601 list_del_init(&child->mnt_hash);
1604 br_write_unlock(&vfsmount_lock);
1609 if (IS_MNT_SHARED(dest_mnt))
1610 cleanup_group_ids(source_mnt, NULL);
1615 static struct mountpoint *lock_mount(struct path *path)
1617 struct vfsmount *mnt;
1618 struct dentry *dentry = path->dentry;
1620 mutex_lock(&dentry->d_inode->i_mutex);
1621 if (unlikely(cant_mount(dentry))) {
1622 mutex_unlock(&dentry->d_inode->i_mutex);
1623 return ERR_PTR(-ENOENT);
1626 mnt = lookup_mnt(path);
1628 struct mountpoint *mp = new_mountpoint(dentry);
1631 mutex_unlock(&dentry->d_inode->i_mutex);
1637 mutex_unlock(&path->dentry->d_inode->i_mutex);
1640 dentry = path->dentry = dget(mnt->mnt_root);
1644 static void unlock_mount(struct mountpoint *where)
1646 struct dentry *dentry = where->m_dentry;
1647 put_mountpoint(where);
1649 mutex_unlock(&dentry->d_inode->i_mutex);
1652 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1654 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1657 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1658 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1661 return attach_recursive_mnt(mnt, p, mp, NULL);
1665 * Sanity check the flags to change_mnt_propagation.
1668 static int flags_to_propagation_type(int flags)
1670 int type = flags & ~(MS_REC | MS_SILENT);
1672 /* Fail if any non-propagation flags are set */
1673 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1675 /* Only one propagation flag should be set */
1676 if (!is_power_of_2(type))
1682 * recursively change the type of the mountpoint.
1684 static int do_change_type(struct path *path, int flag)
1687 struct mount *mnt = real_mount(path->mnt);
1688 int recurse = flag & MS_REC;
1692 if (path->dentry != path->mnt->mnt_root)
1695 type = flags_to_propagation_type(flag);
1700 if (type == MS_SHARED) {
1701 err = invent_group_ids(mnt, recurse);
1706 br_write_lock(&vfsmount_lock);
1707 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1708 change_mnt_propagation(m, type);
1709 br_write_unlock(&vfsmount_lock);
1716 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1718 struct mount *child;
1719 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1720 if (!is_subdir(child->mnt_mountpoint, dentry))
1723 if (child->mnt.mnt_flags & MNT_LOCKED)
1730 * do loopback mount.
1732 static int do_loopback(struct path *path, const char *old_name,
1735 struct path old_path;
1736 struct mount *mnt = NULL, *old, *parent;
1737 struct mountpoint *mp;
1739 if (!old_name || !*old_name)
1741 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1746 if (mnt_ns_loop(old_path.dentry))
1749 mp = lock_mount(path);
1754 old = real_mount(old_path.mnt);
1755 parent = real_mount(path->mnt);
1758 if (IS_MNT_UNBINDABLE(old))
1761 if (!check_mnt(parent) || !check_mnt(old))
1764 if (!recurse && has_locked_children(old, old_path.dentry))
1768 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1770 mnt = clone_mnt(old, old_path.dentry, 0);
1777 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1779 err = graft_tree(mnt, parent, mp);
1781 br_write_lock(&vfsmount_lock);
1782 umount_tree(mnt, 0);
1783 br_write_unlock(&vfsmount_lock);
1788 path_put(&old_path);
1792 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1795 int readonly_request = 0;
1797 if (ms_flags & MS_RDONLY)
1798 readonly_request = 1;
1799 if (readonly_request == __mnt_is_readonly(mnt))
1802 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1805 if (readonly_request)
1806 error = mnt_make_readonly(real_mount(mnt));
1808 __mnt_unmake_readonly(real_mount(mnt));
1813 * change filesystem flags. dir should be a physical root of filesystem.
1814 * If you've mounted a non-root directory somewhere and want to do remount
1815 * on it - tough luck.
1817 static int do_remount(struct path *path, int flags, int mnt_flags,
1821 struct super_block *sb = path->mnt->mnt_sb;
1822 struct mount *mnt = real_mount(path->mnt);
1824 if (!check_mnt(mnt))
1827 if (path->dentry != path->mnt->mnt_root)
1830 err = security_sb_remount(sb, data);
1834 down_write(&sb->s_umount);
1835 if (flags & MS_BIND)
1836 err = change_mount_flags(path->mnt, flags);
1837 else if (!capable(CAP_SYS_ADMIN))
1840 err = do_remount_sb(sb, flags, data, 0);
1842 br_write_lock(&vfsmount_lock);
1843 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1844 mnt->mnt.mnt_flags = mnt_flags;
1845 touch_mnt_namespace(mnt->mnt_ns);
1846 br_write_unlock(&vfsmount_lock);
1848 up_write(&sb->s_umount);
1852 static inline int tree_contains_unbindable(struct mount *mnt)
1855 for (p = mnt; p; p = next_mnt(p, mnt)) {
1856 if (IS_MNT_UNBINDABLE(p))
1862 static int do_move_mount(struct path *path, const char *old_name)
1864 struct path old_path, parent_path;
1867 struct mountpoint *mp;
1869 if (!old_name || !*old_name)
1871 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1875 mp = lock_mount(path);
1880 old = real_mount(old_path.mnt);
1881 p = real_mount(path->mnt);
1884 if (!check_mnt(p) || !check_mnt(old))
1887 if (old->mnt.mnt_flags & MNT_LOCKED)
1891 if (old_path.dentry != old_path.mnt->mnt_root)
1894 if (!mnt_has_parent(old))
1897 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1898 S_ISDIR(old_path.dentry->d_inode->i_mode))
1901 * Don't move a mount residing in a shared parent.
1903 if (IS_MNT_SHARED(old->mnt_parent))
1906 * Don't move a mount tree containing unbindable mounts to a destination
1907 * mount which is shared.
1909 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1912 for (; mnt_has_parent(p); p = p->mnt_parent)
1916 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
1920 /* if the mount is moved, it should no longer be expire
1922 list_del_init(&old->mnt_expire);
1927 path_put(&parent_path);
1928 path_put(&old_path);
1932 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1935 const char *subtype = strchr(fstype, '.');
1944 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1946 if (!mnt->mnt_sb->s_subtype)
1952 return ERR_PTR(err);
1956 * add a mount into a namespace's mount tree
1958 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1960 struct mountpoint *mp;
1961 struct mount *parent;
1964 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1966 mp = lock_mount(path);
1970 parent = real_mount(path->mnt);
1972 if (unlikely(!check_mnt(parent))) {
1973 /* that's acceptable only for automounts done in private ns */
1974 if (!(mnt_flags & MNT_SHRINKABLE))
1976 /* ... and for those we'd better have mountpoint still alive */
1977 if (!parent->mnt_ns)
1981 /* Refuse the same filesystem on the same mount point */
1983 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1984 path->mnt->mnt_root == path->dentry)
1988 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1991 newmnt->mnt.mnt_flags = mnt_flags;
1992 err = graft_tree(newmnt, parent, mp);
2000 * create a new mount for userspace and request it to be added into the
2003 static int do_new_mount(struct path *path, const char *fstype, int flags,
2004 int mnt_flags, const char *name, void *data)
2006 struct file_system_type *type;
2007 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2008 struct vfsmount *mnt;
2014 type = get_fs_type(fstype);
2018 if (user_ns != &init_user_ns) {
2019 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2020 put_filesystem(type);
2023 /* Only in special cases allow devices from mounts
2024 * created outside the initial user namespace.
2026 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2028 mnt_flags |= MNT_NODEV;
2032 mnt = vfs_kern_mount(type, flags, name, data);
2033 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2034 !mnt->mnt_sb->s_subtype)
2035 mnt = fs_set_subtype(mnt, fstype);
2037 put_filesystem(type);
2039 return PTR_ERR(mnt);
2041 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2047 int finish_automount(struct vfsmount *m, struct path *path)
2049 struct mount *mnt = real_mount(m);
2051 /* The new mount record should have at least 2 refs to prevent it being
2052 * expired before we get a chance to add it
2054 BUG_ON(mnt_get_count(mnt) < 2);
2056 if (m->mnt_sb == path->mnt->mnt_sb &&
2057 m->mnt_root == path->dentry) {
2062 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2066 /* remove m from any expiration list it may be on */
2067 if (!list_empty(&mnt->mnt_expire)) {
2069 br_write_lock(&vfsmount_lock);
2070 list_del_init(&mnt->mnt_expire);
2071 br_write_unlock(&vfsmount_lock);
2080 * mnt_set_expiry - Put a mount on an expiration list
2081 * @mnt: The mount to list.
2082 * @expiry_list: The list to add the mount to.
2084 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2087 br_write_lock(&vfsmount_lock);
2089 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2091 br_write_unlock(&vfsmount_lock);
2094 EXPORT_SYMBOL(mnt_set_expiry);
2097 * process a list of expirable mountpoints with the intent of discarding any
2098 * mountpoints that aren't in use and haven't been touched since last we came
2101 void mark_mounts_for_expiry(struct list_head *mounts)
2103 struct mount *mnt, *next;
2104 LIST_HEAD(graveyard);
2106 if (list_empty(mounts))
2110 br_write_lock(&vfsmount_lock);
2112 /* extract from the expiration list every vfsmount that matches the
2113 * following criteria:
2114 * - only referenced by its parent vfsmount
2115 * - still marked for expiry (marked on the last call here; marks are
2116 * cleared by mntput())
2118 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2119 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2120 propagate_mount_busy(mnt, 1))
2122 list_move(&mnt->mnt_expire, &graveyard);
2124 while (!list_empty(&graveyard)) {
2125 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2126 touch_mnt_namespace(mnt->mnt_ns);
2127 umount_tree(mnt, 1);
2129 br_write_unlock(&vfsmount_lock);
2133 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2136 * Ripoff of 'select_parent()'
2138 * search the list of submounts for a given mountpoint, and move any
2139 * shrinkable submounts to the 'graveyard' list.
2141 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2143 struct mount *this_parent = parent;
2144 struct list_head *next;
2148 next = this_parent->mnt_mounts.next;
2150 while (next != &this_parent->mnt_mounts) {
2151 struct list_head *tmp = next;
2152 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2155 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2158 * Descend a level if the d_mounts list is non-empty.
2160 if (!list_empty(&mnt->mnt_mounts)) {
2165 if (!propagate_mount_busy(mnt, 1)) {
2166 list_move_tail(&mnt->mnt_expire, graveyard);
2171 * All done at this level ... ascend and resume the search
2173 if (this_parent != parent) {
2174 next = this_parent->mnt_child.next;
2175 this_parent = this_parent->mnt_parent;
2182 * process a list of expirable mountpoints with the intent of discarding any
2183 * submounts of a specific parent mountpoint
2185 * vfsmount_lock must be held for write
2187 static void shrink_submounts(struct mount *mnt)
2189 LIST_HEAD(graveyard);
2192 /* extract submounts of 'mountpoint' from the expiration list */
2193 while (select_submounts(mnt, &graveyard)) {
2194 while (!list_empty(&graveyard)) {
2195 m = list_first_entry(&graveyard, struct mount,
2197 touch_mnt_namespace(m->mnt_ns);
2204 * Some copy_from_user() implementations do not return the exact number of
2205 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2206 * Note that this function differs from copy_from_user() in that it will oops
2207 * on bad values of `to', rather than returning a short copy.
2209 static long exact_copy_from_user(void *to, const void __user * from,
2213 const char __user *f = from;
2216 if (!access_ok(VERIFY_READ, from, n))
2220 if (__get_user(c, f)) {
2231 int copy_mount_options(const void __user * data, unsigned long *where)
2241 if (!(page = __get_free_page(GFP_KERNEL)))
2244 /* We only care that *some* data at the address the user
2245 * gave us is valid. Just in case, we'll zero
2246 * the remainder of the page.
2248 /* copy_from_user cannot cross TASK_SIZE ! */
2249 size = TASK_SIZE - (unsigned long)data;
2250 if (size > PAGE_SIZE)
2253 i = size - exact_copy_from_user((void *)page, data, size);
2259 memset((char *)page + i, 0, PAGE_SIZE - i);
2264 int copy_mount_string(const void __user *data, char **where)
2273 tmp = strndup_user(data, PAGE_SIZE);
2275 return PTR_ERR(tmp);
2282 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2283 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2285 * data is a (void *) that can point to any structure up to
2286 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2287 * information (or be NULL).
2289 * Pre-0.97 versions of mount() didn't have a flags word.
2290 * When the flags word was introduced its top half was required
2291 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2292 * Therefore, if this magic number is present, it carries no information
2293 * and must be discarded.
2295 long do_mount(const char *dev_name, const char *dir_name,
2296 const char *type_page, unsigned long flags, void *data_page)
2303 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2304 flags &= ~MS_MGC_MSK;
2306 /* Basic sanity checks */
2308 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2312 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2314 /* ... and get the mountpoint */
2315 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2319 retval = security_sb_mount(dev_name, &path,
2320 type_page, flags, data_page);
2321 if (!retval && !may_mount())
2326 /* Default to relatime unless overriden */
2327 if (!(flags & MS_NOATIME))
2328 mnt_flags |= MNT_RELATIME;
2330 /* Separate the per-mountpoint flags */
2331 if (flags & MS_NOSUID)
2332 mnt_flags |= MNT_NOSUID;
2333 if (flags & MS_NODEV)
2334 mnt_flags |= MNT_NODEV;
2335 if (flags & MS_NOEXEC)
2336 mnt_flags |= MNT_NOEXEC;
2337 if (flags & MS_NOATIME)
2338 mnt_flags |= MNT_NOATIME;
2339 if (flags & MS_NODIRATIME)
2340 mnt_flags |= MNT_NODIRATIME;
2341 if (flags & MS_STRICTATIME)
2342 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2343 if (flags & MS_RDONLY)
2344 mnt_flags |= MNT_READONLY;
2346 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2347 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2350 if (flags & MS_REMOUNT)
2351 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2353 else if (flags & MS_BIND)
2354 retval = do_loopback(&path, dev_name, flags & MS_REC);
2355 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2356 retval = do_change_type(&path, flags);
2357 else if (flags & MS_MOVE)
2358 retval = do_move_mount(&path, dev_name);
2360 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2361 dev_name, data_page);
2367 static void free_mnt_ns(struct mnt_namespace *ns)
2369 proc_free_inum(ns->proc_inum);
2370 put_user_ns(ns->user_ns);
2375 * Assign a sequence number so we can detect when we attempt to bind
2376 * mount a reference to an older mount namespace into the current
2377 * mount namespace, preventing reference counting loops. A 64bit
2378 * number incrementing at 10Ghz will take 12,427 years to wrap which
2379 * is effectively never, so we can ignore the possibility.
2381 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2383 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2385 struct mnt_namespace *new_ns;
2388 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2390 return ERR_PTR(-ENOMEM);
2391 ret = proc_alloc_inum(&new_ns->proc_inum);
2394 return ERR_PTR(ret);
2396 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2397 atomic_set(&new_ns->count, 1);
2398 new_ns->root = NULL;
2399 INIT_LIST_HEAD(&new_ns->list);
2400 init_waitqueue_head(&new_ns->poll);
2402 new_ns->user_ns = get_user_ns(user_ns);
2407 * Allocate a new namespace structure and populate it with contents
2408 * copied from the namespace of the passed in task structure.
2410 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2411 struct user_namespace *user_ns, struct fs_struct *fs)
2413 struct mnt_namespace *new_ns;
2414 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2415 struct mount *p, *q;
2416 struct mount *old = mnt_ns->root;
2420 new_ns = alloc_mnt_ns(user_ns);
2425 /* First pass: copy the tree topology */
2426 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2427 if (user_ns != mnt_ns->user_ns)
2428 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2429 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2432 free_mnt_ns(new_ns);
2433 return ERR_CAST(new);
2436 list_add_tail(&new_ns->list, &new->mnt_list);
2439 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2440 * as belonging to new namespace. We have already acquired a private
2441 * fs_struct, so tsk->fs->lock is not needed.
2448 if (&p->mnt == fs->root.mnt) {
2449 fs->root.mnt = mntget(&q->mnt);
2452 if (&p->mnt == fs->pwd.mnt) {
2453 fs->pwd.mnt = mntget(&q->mnt);
2457 p = next_mnt(p, old);
2458 q = next_mnt(q, new);
2461 while (p->mnt.mnt_root != q->mnt.mnt_root)
2462 p = next_mnt(p, old);
2474 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2475 struct user_namespace *user_ns, struct fs_struct *new_fs)
2477 struct mnt_namespace *new_ns;
2482 if (!(flags & CLONE_NEWNS))
2485 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2492 * create_mnt_ns - creates a private namespace and adds a root filesystem
2493 * @mnt: pointer to the new root filesystem mountpoint
2495 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2497 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2498 if (!IS_ERR(new_ns)) {
2499 struct mount *mnt = real_mount(m);
2500 mnt->mnt_ns = new_ns;
2502 list_add(&mnt->mnt_list, &new_ns->list);
2509 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2511 struct mnt_namespace *ns;
2512 struct super_block *s;
2516 ns = create_mnt_ns(mnt);
2518 return ERR_CAST(ns);
2520 err = vfs_path_lookup(mnt->mnt_root, mnt,
2521 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2526 return ERR_PTR(err);
2528 /* trade a vfsmount reference for active sb one */
2529 s = path.mnt->mnt_sb;
2530 atomic_inc(&s->s_active);
2532 /* lock the sucker */
2533 down_write(&s->s_umount);
2534 /* ... and return the root of (sub)tree on it */
2537 EXPORT_SYMBOL(mount_subtree);
2539 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2540 char __user *, type, unsigned long, flags, void __user *, data)
2544 struct filename *kernel_dir;
2546 unsigned long data_page;
2548 ret = copy_mount_string(type, &kernel_type);
2552 kernel_dir = getname(dir_name);
2553 if (IS_ERR(kernel_dir)) {
2554 ret = PTR_ERR(kernel_dir);
2558 ret = copy_mount_string(dev_name, &kernel_dev);
2562 ret = copy_mount_options(data, &data_page);
2566 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2567 (void *) data_page);
2569 free_page(data_page);
2573 putname(kernel_dir);
2581 * Return true if path is reachable from root
2583 * namespace_sem or vfsmount_lock is held
2585 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2586 const struct path *root)
2588 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2589 dentry = mnt->mnt_mountpoint;
2590 mnt = mnt->mnt_parent;
2592 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2595 int path_is_under(struct path *path1, struct path *path2)
2598 br_read_lock(&vfsmount_lock);
2599 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2600 br_read_unlock(&vfsmount_lock);
2603 EXPORT_SYMBOL(path_is_under);
2606 * pivot_root Semantics:
2607 * Moves the root file system of the current process to the directory put_old,
2608 * makes new_root as the new root file system of the current process, and sets
2609 * root/cwd of all processes which had them on the current root to new_root.
2612 * The new_root and put_old must be directories, and must not be on the
2613 * same file system as the current process root. The put_old must be
2614 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2615 * pointed to by put_old must yield the same directory as new_root. No other
2616 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2618 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2619 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2620 * in this situation.
2623 * - we don't move root/cwd if they are not at the root (reason: if something
2624 * cared enough to change them, it's probably wrong to force them elsewhere)
2625 * - it's okay to pick a root that isn't the root of a file system, e.g.
2626 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2627 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2630 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2631 const char __user *, put_old)
2633 struct path new, old, parent_path, root_parent, root;
2634 struct mount *new_mnt, *root_mnt, *old_mnt;
2635 struct mountpoint *old_mp, *root_mp;
2641 error = user_path_dir(new_root, &new);
2645 error = user_path_dir(put_old, &old);
2649 error = security_sb_pivotroot(&old, &new);
2653 get_fs_root(current->fs, &root);
2654 old_mp = lock_mount(&old);
2655 error = PTR_ERR(old_mp);
2660 new_mnt = real_mount(new.mnt);
2661 root_mnt = real_mount(root.mnt);
2662 old_mnt = real_mount(old.mnt);
2663 if (IS_MNT_SHARED(old_mnt) ||
2664 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2665 IS_MNT_SHARED(root_mnt->mnt_parent))
2667 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2669 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2672 if (d_unlinked(new.dentry))
2675 if (new_mnt == root_mnt || old_mnt == root_mnt)
2676 goto out4; /* loop, on the same file system */
2678 if (root.mnt->mnt_root != root.dentry)
2679 goto out4; /* not a mountpoint */
2680 if (!mnt_has_parent(root_mnt))
2681 goto out4; /* not attached */
2682 root_mp = root_mnt->mnt_mp;
2683 if (new.mnt->mnt_root != new.dentry)
2684 goto out4; /* not a mountpoint */
2685 if (!mnt_has_parent(new_mnt))
2686 goto out4; /* not attached */
2687 /* make sure we can reach put_old from new_root */
2688 if (!is_path_reachable(old_mnt, old.dentry, &new))
2690 root_mp->m_count++; /* pin it so it won't go away */
2691 br_write_lock(&vfsmount_lock);
2692 detach_mnt(new_mnt, &parent_path);
2693 detach_mnt(root_mnt, &root_parent);
2694 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2695 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2696 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2698 /* mount old root on put_old */
2699 attach_mnt(root_mnt, old_mnt, old_mp);
2700 /* mount new_root on / */
2701 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2702 touch_mnt_namespace(current->nsproxy->mnt_ns);
2703 br_write_unlock(&vfsmount_lock);
2704 chroot_fs_refs(&root, &new);
2705 put_mountpoint(root_mp);
2708 unlock_mount(old_mp);
2710 path_put(&root_parent);
2711 path_put(&parent_path);
2723 static void __init init_mount_tree(void)
2725 struct vfsmount *mnt;
2726 struct mnt_namespace *ns;
2728 struct file_system_type *type;
2730 type = get_fs_type("rootfs");
2732 panic("Can't find rootfs type");
2733 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2734 put_filesystem(type);
2736 panic("Can't create rootfs");
2738 ns = create_mnt_ns(mnt);
2740 panic("Can't allocate initial namespace");
2742 init_task.nsproxy->mnt_ns = ns;
2746 root.dentry = mnt->mnt_root;
2748 set_fs_pwd(current->fs, &root);
2749 set_fs_root(current->fs, &root);
2752 void __init mnt_init(void)
2757 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2758 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2760 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2761 mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2763 if (!mount_hashtable || !mountpoint_hashtable)
2764 panic("Failed to allocate mount hash table\n");
2766 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2768 for (u = 0; u < HASH_SIZE; u++)
2769 INIT_LIST_HEAD(&mount_hashtable[u]);
2770 for (u = 0; u < HASH_SIZE; u++)
2771 INIT_LIST_HEAD(&mountpoint_hashtable[u]);
2773 br_lock_init(&vfsmount_lock);
2777 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2779 fs_kobj = kobject_create_and_add("fs", NULL);
2781 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2786 void put_mnt_ns(struct mnt_namespace *ns)
2788 if (!atomic_dec_and_test(&ns->count))
2790 drop_collected_mounts(&ns->root->mnt);
2794 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2796 struct vfsmount *mnt;
2797 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2800 * it is a longterm mount, don't release mnt until
2801 * we unmount before file sys is unregistered
2803 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2807 EXPORT_SYMBOL_GPL(kern_mount_data);
2809 void kern_unmount(struct vfsmount *mnt)
2811 /* release long term mount so mount point can be released */
2812 if (!IS_ERR_OR_NULL(mnt)) {
2813 br_write_lock(&vfsmount_lock);
2814 real_mount(mnt)->mnt_ns = NULL;
2815 br_write_unlock(&vfsmount_lock);
2819 EXPORT_SYMBOL(kern_unmount);
2821 bool our_mnt(struct vfsmount *mnt)
2823 return check_mnt(real_mount(mnt));
2826 bool current_chrooted(void)
2828 /* Does the current process have a non-standard root */
2829 struct path ns_root;
2830 struct path fs_root;
2833 /* Find the namespace root */
2834 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
2835 ns_root.dentry = ns_root.mnt->mnt_root;
2837 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2840 get_fs_root(current->fs, &fs_root);
2842 chrooted = !path_equal(&fs_root, &ns_root);
2850 bool fs_fully_visible(struct file_system_type *type)
2852 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2854 bool visible = false;
2859 down_read(&namespace_sem);
2860 list_for_each_entry(mnt, &ns->list, mnt_list) {
2861 struct mount *child;
2862 if (mnt->mnt.mnt_sb->s_type != type)
2865 /* This mount is not fully visible if there are any child mounts
2866 * that cover anything except for empty directories.
2868 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2869 struct inode *inode = child->mnt_mountpoint->d_inode;
2870 if (!S_ISDIR(inode->i_mode))
2872 if (inode->i_nlink != 2)
2880 up_read(&namespace_sem);
2884 static void *mntns_get(struct task_struct *task)
2886 struct mnt_namespace *ns = NULL;
2887 struct nsproxy *nsproxy;
2890 nsproxy = task_nsproxy(task);
2892 ns = nsproxy->mnt_ns;
2900 static void mntns_put(void *ns)
2905 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2907 struct fs_struct *fs = current->fs;
2908 struct mnt_namespace *mnt_ns = ns;
2911 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2912 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2913 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2920 put_mnt_ns(nsproxy->mnt_ns);
2921 nsproxy->mnt_ns = mnt_ns;
2924 root.mnt = &mnt_ns->root->mnt;
2925 root.dentry = mnt_ns->root->mnt.mnt_root;
2927 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2930 /* Update the pwd and root */
2931 set_fs_pwd(fs, &root);
2932 set_fs_root(fs, &root);
2938 static unsigned int mntns_inum(void *ns)
2940 struct mnt_namespace *mnt_ns = ns;
2941 return mnt_ns->proc_inum;
2944 const struct proc_ns_operations mntns_operations = {
2946 .type = CLONE_NEWNS,
2949 .install = mntns_install,