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 INIT_LIST_HEAD(&mnt->mnt_mp_list);
201 #ifdef CONFIG_FSNOTIFY
202 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
209 kfree(mnt->mnt_devname);
214 kmem_cache_free(mnt_cache, mnt);
219 * Most r/o checks on a fs are for operations that take
220 * discrete amounts of time, like a write() or unlink().
221 * We must keep track of when those operations start
222 * (for permission checks) and when they end, so that
223 * we can determine when writes are able to occur to
227 * __mnt_is_readonly: check whether a mount is read-only
228 * @mnt: the mount to check for its write status
230 * This shouldn't be used directly ouside of the VFS.
231 * It does not guarantee that the filesystem will stay
232 * r/w, just that it is right *now*. This can not and
233 * should not be used in place of IS_RDONLY(inode).
234 * mnt_want/drop_write() will _keep_ the filesystem
237 int __mnt_is_readonly(struct vfsmount *mnt)
239 if (mnt->mnt_flags & MNT_READONLY)
241 if (mnt->mnt_sb->s_flags & MS_RDONLY)
245 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
247 static inline void mnt_inc_writers(struct mount *mnt)
250 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
256 static inline void mnt_dec_writers(struct mount *mnt)
259 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
265 static unsigned int mnt_get_writers(struct mount *mnt)
268 unsigned int count = 0;
271 for_each_possible_cpu(cpu) {
272 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
277 return mnt->mnt_writers;
281 static int mnt_is_readonly(struct vfsmount *mnt)
283 if (mnt->mnt_sb->s_readonly_remount)
285 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
287 return __mnt_is_readonly(mnt);
291 * Most r/o & frozen checks on a fs are for operations that take discrete
292 * amounts of time, like a write() or unlink(). We must keep track of when
293 * those operations start (for permission checks) and when they end, so that we
294 * can determine when writes are able to occur to a filesystem.
297 * __mnt_want_write - get write access to a mount without freeze protection
298 * @m: the mount on which to take a write
300 * This tells the low-level filesystem that a write is about to be performed to
301 * it, and makes sure that writes are allowed (mnt it read-write) before
302 * returning success. This operation does not protect against filesystem being
303 * frozen. When the write operation is finished, __mnt_drop_write() must be
304 * called. This is effectively a refcount.
306 int __mnt_want_write(struct vfsmount *m)
308 struct mount *mnt = real_mount(m);
312 mnt_inc_writers(mnt);
314 * The store to mnt_inc_writers must be visible before we pass
315 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
316 * incremented count after it has set MNT_WRITE_HOLD.
319 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
322 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
323 * be set to match its requirements. So we must not load that until
324 * MNT_WRITE_HOLD is cleared.
327 if (mnt_is_readonly(m)) {
328 mnt_dec_writers(mnt);
337 * mnt_want_write - get write access to a mount
338 * @m: the mount on which to take a write
340 * This tells the low-level filesystem that a write is about to be performed to
341 * it, and makes sure that writes are allowed (mount is read-write, filesystem
342 * is not frozen) before returning success. When the write operation is
343 * finished, mnt_drop_write() must be called. This is effectively a refcount.
345 int mnt_want_write(struct vfsmount *m)
349 sb_start_write(m->mnt_sb);
350 ret = __mnt_want_write(m);
352 sb_end_write(m->mnt_sb);
355 EXPORT_SYMBOL_GPL(mnt_want_write);
358 * mnt_clone_write - get write access to a mount
359 * @mnt: the mount on which to take a write
361 * This is effectively like mnt_want_write, except
362 * it must only be used to take an extra write reference
363 * on a mountpoint that we already know has a write reference
364 * on it. This allows some optimisation.
366 * After finished, mnt_drop_write must be called as usual to
367 * drop the reference.
369 int mnt_clone_write(struct vfsmount *mnt)
371 /* superblock may be r/o */
372 if (__mnt_is_readonly(mnt))
375 mnt_inc_writers(real_mount(mnt));
379 EXPORT_SYMBOL_GPL(mnt_clone_write);
382 * __mnt_want_write_file - get write access to a file's mount
383 * @file: the file who's mount on which to take a write
385 * This is like __mnt_want_write, but it takes a file and can
386 * do some optimisations if the file is open for write already
388 int __mnt_want_write_file(struct file *file)
390 struct inode *inode = file_inode(file);
392 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
393 return __mnt_want_write(file->f_path.mnt);
395 return mnt_clone_write(file->f_path.mnt);
399 * mnt_want_write_file - get write access to a file's mount
400 * @file: the file who's mount on which to take a write
402 * This is like mnt_want_write, but it takes a file and can
403 * do some optimisations if the file is open for write already
405 int mnt_want_write_file(struct file *file)
409 sb_start_write(file->f_path.mnt->mnt_sb);
410 ret = __mnt_want_write_file(file);
412 sb_end_write(file->f_path.mnt->mnt_sb);
415 EXPORT_SYMBOL_GPL(mnt_want_write_file);
418 * __mnt_drop_write - give up write access to a mount
419 * @mnt: the mount on which to give up write access
421 * Tells the low-level filesystem that we are done
422 * performing writes to it. Must be matched with
423 * __mnt_want_write() call above.
425 void __mnt_drop_write(struct vfsmount *mnt)
428 mnt_dec_writers(real_mount(mnt));
433 * mnt_drop_write - give up write access to a mount
434 * @mnt: the mount on which to give up write access
436 * Tells the low-level filesystem that we are done performing writes to it and
437 * also allows filesystem to be frozen again. Must be matched with
438 * mnt_want_write() call above.
440 void mnt_drop_write(struct vfsmount *mnt)
442 __mnt_drop_write(mnt);
443 sb_end_write(mnt->mnt_sb);
445 EXPORT_SYMBOL_GPL(mnt_drop_write);
447 void __mnt_drop_write_file(struct file *file)
449 __mnt_drop_write(file->f_path.mnt);
452 void mnt_drop_write_file(struct file *file)
454 mnt_drop_write(file->f_path.mnt);
456 EXPORT_SYMBOL(mnt_drop_write_file);
458 static int mnt_make_readonly(struct mount *mnt)
462 br_write_lock(&vfsmount_lock);
463 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
465 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
466 * should be visible before we do.
471 * With writers on hold, if this value is zero, then there are
472 * definitely no active writers (although held writers may subsequently
473 * increment the count, they'll have to wait, and decrement it after
474 * seeing MNT_READONLY).
476 * It is OK to have counter incremented on one CPU and decremented on
477 * another: the sum will add up correctly. The danger would be when we
478 * sum up each counter, if we read a counter before it is incremented,
479 * but then read another CPU's count which it has been subsequently
480 * decremented from -- we would see more decrements than we should.
481 * MNT_WRITE_HOLD protects against this scenario, because
482 * mnt_want_write first increments count, then smp_mb, then spins on
483 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
484 * we're counting up here.
486 if (mnt_get_writers(mnt) > 0)
489 mnt->mnt.mnt_flags |= MNT_READONLY;
491 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
492 * that become unheld will see MNT_READONLY.
495 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
496 br_write_unlock(&vfsmount_lock);
500 static void __mnt_unmake_readonly(struct mount *mnt)
502 br_write_lock(&vfsmount_lock);
503 mnt->mnt.mnt_flags &= ~MNT_READONLY;
504 br_write_unlock(&vfsmount_lock);
507 int sb_prepare_remount_readonly(struct super_block *sb)
512 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
513 if (atomic_long_read(&sb->s_remove_count))
516 br_write_lock(&vfsmount_lock);
517 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
518 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
519 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
521 if (mnt_get_writers(mnt) > 0) {
527 if (!err && atomic_long_read(&sb->s_remove_count))
531 sb->s_readonly_remount = 1;
534 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
535 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
536 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
538 br_write_unlock(&vfsmount_lock);
543 static void free_vfsmnt(struct mount *mnt)
545 kfree(mnt->mnt_devname);
548 free_percpu(mnt->mnt_pcp);
550 kmem_cache_free(mnt_cache, mnt);
554 * find the first or last mount at @dentry on vfsmount @mnt depending on
555 * @dir. If @dir is set return the first mount else return the last mount.
556 * vfsmount_lock must be held for read or write.
558 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
561 struct list_head *head = mount_hashtable + hash(mnt, dentry);
562 struct list_head *tmp = head;
563 struct mount *p, *found = NULL;
566 tmp = dir ? tmp->next : tmp->prev;
570 p = list_entry(tmp, struct mount, mnt_hash);
571 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
580 * lookup_mnt - Return the first child mount mounted at path
582 * "First" means first mounted chronologically. If you create the
585 * mount /dev/sda1 /mnt
586 * mount /dev/sda2 /mnt
587 * mount /dev/sda3 /mnt
589 * Then lookup_mnt() on the base /mnt dentry in the root mount will
590 * return successively the root dentry and vfsmount of /dev/sda1, then
591 * /dev/sda2, then /dev/sda3, then NULL.
593 * lookup_mnt takes a reference to the found vfsmount.
595 struct vfsmount *lookup_mnt(struct path *path)
597 struct mount *child_mnt;
599 br_read_lock(&vfsmount_lock);
600 child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
602 mnt_add_count(child_mnt, 1);
603 br_read_unlock(&vfsmount_lock);
604 return &child_mnt->mnt;
606 br_read_unlock(&vfsmount_lock);
611 static struct mountpoint *new_mountpoint(struct dentry *dentry)
613 struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
614 struct mountpoint *mp;
617 list_for_each_entry(mp, chain, m_hash) {
618 if (mp->m_dentry == dentry) {
619 /* might be worth a WARN_ON() */
620 if (d_unlinked(dentry))
621 return ERR_PTR(-ENOENT);
627 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
629 return ERR_PTR(-ENOMEM);
631 ret = d_set_mounted(dentry);
637 mp->m_dentry = dentry;
639 list_add(&mp->m_hash, chain);
640 INIT_LIST_HEAD(&mp->m_list);
644 static void put_mountpoint(struct mountpoint *mp)
646 if (!--mp->m_count) {
647 struct dentry *dentry = mp->m_dentry;
648 BUG_ON(!list_empty(&mp->m_list));
649 spin_lock(&dentry->d_lock);
650 dentry->d_flags &= ~DCACHE_MOUNTED;
651 spin_unlock(&dentry->d_lock);
652 list_del(&mp->m_hash);
657 static inline int check_mnt(struct mount *mnt)
659 return mnt->mnt_ns == current->nsproxy->mnt_ns;
663 * vfsmount lock must be held for write
665 static void touch_mnt_namespace(struct mnt_namespace *ns)
669 wake_up_interruptible(&ns->poll);
674 * vfsmount lock must be held for write
676 static void __touch_mnt_namespace(struct mnt_namespace *ns)
678 if (ns && ns->event != event) {
680 wake_up_interruptible(&ns->poll);
685 * vfsmount lock must be held for write
687 static void detach_mnt(struct mount *mnt, struct path *old_path)
689 old_path->dentry = mnt->mnt_mountpoint;
690 old_path->mnt = &mnt->mnt_parent->mnt;
691 mnt->mnt_parent = mnt;
692 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
693 list_del_init(&mnt->mnt_child);
694 list_del_init(&mnt->mnt_hash);
695 list_del_init(&mnt->mnt_mp_list);
696 put_mountpoint(mnt->mnt_mp);
701 * vfsmount lock must be held for write
703 void mnt_set_mountpoint(struct mount *mnt,
704 struct mountpoint *mp,
705 struct mount *child_mnt)
708 mnt_add_count(mnt, 1); /* essentially, that's mntget */
709 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
710 child_mnt->mnt_parent = mnt;
711 child_mnt->mnt_mp = mp;
712 list_add_tail(&child_mnt->mnt_mp_list, &mp->m_list);
716 * vfsmount lock must be held for write
718 static void attach_mnt(struct mount *mnt,
719 struct mount *parent,
720 struct mountpoint *mp)
722 mnt_set_mountpoint(parent, mp, mnt);
723 list_add_tail(&mnt->mnt_hash, mount_hashtable +
724 hash(&parent->mnt, mp->m_dentry));
725 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
729 * vfsmount lock must be held for write
731 static void commit_tree(struct mount *mnt)
733 struct mount *parent = mnt->mnt_parent;
736 struct mnt_namespace *n = parent->mnt_ns;
738 BUG_ON(parent == mnt);
740 list_add_tail(&head, &mnt->mnt_list);
741 list_for_each_entry(m, &head, mnt_list)
744 list_splice(&head, n->list.prev);
746 list_add_tail(&mnt->mnt_hash, mount_hashtable +
747 hash(&parent->mnt, mnt->mnt_mountpoint));
748 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
749 touch_mnt_namespace(n);
752 static struct mount *next_mnt(struct mount *p, struct mount *root)
754 struct list_head *next = p->mnt_mounts.next;
755 if (next == &p->mnt_mounts) {
759 next = p->mnt_child.next;
760 if (next != &p->mnt_parent->mnt_mounts)
765 return list_entry(next, struct mount, mnt_child);
768 static struct mount *skip_mnt_tree(struct mount *p)
770 struct list_head *prev = p->mnt_mounts.prev;
771 while (prev != &p->mnt_mounts) {
772 p = list_entry(prev, struct mount, mnt_child);
773 prev = p->mnt_mounts.prev;
779 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
785 return ERR_PTR(-ENODEV);
787 mnt = alloc_vfsmnt(name);
789 return ERR_PTR(-ENOMEM);
791 if (flags & MS_KERNMOUNT)
792 mnt->mnt.mnt_flags = MNT_INTERNAL;
794 root = mount_fs(type, flags, name, data);
797 return ERR_CAST(root);
800 mnt->mnt.mnt_root = root;
801 mnt->mnt.mnt_sb = root->d_sb;
802 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
803 mnt->mnt_parent = mnt;
804 br_write_lock(&vfsmount_lock);
805 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
806 br_write_unlock(&vfsmount_lock);
809 EXPORT_SYMBOL_GPL(vfs_kern_mount);
811 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
814 struct super_block *sb = old->mnt.mnt_sb;
818 mnt = alloc_vfsmnt(old->mnt_devname);
820 return ERR_PTR(-ENOMEM);
822 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
823 mnt->mnt_group_id = 0; /* not a peer of original */
825 mnt->mnt_group_id = old->mnt_group_id;
827 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
828 err = mnt_alloc_group_id(mnt);
833 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
834 /* Don't allow unprivileged users to change mount flags */
835 if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
836 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
838 /* Don't allow unprivileged users to reveal what is under a mount */
839 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
840 mnt->mnt.mnt_flags |= MNT_LOCKED;
842 atomic_inc(&sb->s_active);
843 mnt->mnt.mnt_sb = sb;
844 mnt->mnt.mnt_root = dget(root);
845 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
846 mnt->mnt_parent = mnt;
847 br_write_lock(&vfsmount_lock);
848 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
849 br_write_unlock(&vfsmount_lock);
851 if ((flag & CL_SLAVE) ||
852 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
853 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
854 mnt->mnt_master = old;
855 CLEAR_MNT_SHARED(mnt);
856 } else if (!(flag & CL_PRIVATE)) {
857 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
858 list_add(&mnt->mnt_share, &old->mnt_share);
859 if (IS_MNT_SLAVE(old))
860 list_add(&mnt->mnt_slave, &old->mnt_slave);
861 mnt->mnt_master = old->mnt_master;
863 if (flag & CL_MAKE_SHARED)
866 /* stick the duplicate mount on the same expiry list
867 * as the original if that was on one */
868 if (flag & CL_EXPIRE) {
869 if (!list_empty(&old->mnt_expire))
870 list_add(&mnt->mnt_expire, &old->mnt_expire);
880 static inline void mntfree(struct mount *mnt)
882 struct vfsmount *m = &mnt->mnt;
883 struct super_block *sb = m->mnt_sb;
886 * This probably indicates that somebody messed
887 * up a mnt_want/drop_write() pair. If this
888 * happens, the filesystem was probably unable
889 * to make r/w->r/o transitions.
892 * The locking used to deal with mnt_count decrement provides barriers,
893 * so mnt_get_writers() below is safe.
895 WARN_ON(mnt_get_writers(mnt));
896 fsnotify_vfsmount_delete(m);
899 deactivate_super(sb);
902 static void mntput_no_expire(struct mount *mnt)
906 br_read_lock(&vfsmount_lock);
907 if (likely(mnt->mnt_ns)) {
908 /* shouldn't be the last one */
909 mnt_add_count(mnt, -1);
910 br_read_unlock(&vfsmount_lock);
913 br_read_unlock(&vfsmount_lock);
915 br_write_lock(&vfsmount_lock);
916 mnt_add_count(mnt, -1);
917 if (mnt_get_count(mnt)) {
918 br_write_unlock(&vfsmount_lock);
922 mnt_add_count(mnt, -1);
923 if (likely(mnt_get_count(mnt)))
925 br_write_lock(&vfsmount_lock);
927 if (unlikely(mnt->mnt_pinned)) {
928 mnt_add_count(mnt, mnt->mnt_pinned + 1);
930 br_write_unlock(&vfsmount_lock);
931 acct_auto_close_mnt(&mnt->mnt);
935 list_del(&mnt->mnt_instance);
936 br_write_unlock(&vfsmount_lock);
940 void mntput(struct vfsmount *mnt)
943 struct mount *m = real_mount(mnt);
944 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
945 if (unlikely(m->mnt_expiry_mark))
946 m->mnt_expiry_mark = 0;
950 EXPORT_SYMBOL(mntput);
952 struct vfsmount *mntget(struct vfsmount *mnt)
955 mnt_add_count(real_mount(mnt), 1);
958 EXPORT_SYMBOL(mntget);
960 void mnt_pin(struct vfsmount *mnt)
962 br_write_lock(&vfsmount_lock);
963 real_mount(mnt)->mnt_pinned++;
964 br_write_unlock(&vfsmount_lock);
966 EXPORT_SYMBOL(mnt_pin);
968 void mnt_unpin(struct vfsmount *m)
970 struct mount *mnt = real_mount(m);
971 br_write_lock(&vfsmount_lock);
972 if (mnt->mnt_pinned) {
973 mnt_add_count(mnt, 1);
976 br_write_unlock(&vfsmount_lock);
978 EXPORT_SYMBOL(mnt_unpin);
980 static inline void mangle(struct seq_file *m, const char *s)
982 seq_escape(m, s, " \t\n\\");
986 * Simple .show_options callback for filesystems which don't want to
987 * implement more complex mount option showing.
989 * See also save_mount_options().
991 int generic_show_options(struct seq_file *m, struct dentry *root)
996 options = rcu_dereference(root->d_sb->s_options);
998 if (options != NULL && options[0]) {
1006 EXPORT_SYMBOL(generic_show_options);
1009 * If filesystem uses generic_show_options(), this function should be
1010 * called from the fill_super() callback.
1012 * The .remount_fs callback usually needs to be handled in a special
1013 * way, to make sure, that previous options are not overwritten if the
1016 * Also note, that if the filesystem's .remount_fs function doesn't
1017 * reset all options to their default value, but changes only newly
1018 * given options, then the displayed options will not reflect reality
1021 void save_mount_options(struct super_block *sb, char *options)
1023 BUG_ON(sb->s_options);
1024 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1026 EXPORT_SYMBOL(save_mount_options);
1028 void replace_mount_options(struct super_block *sb, char *options)
1030 char *old = sb->s_options;
1031 rcu_assign_pointer(sb->s_options, options);
1037 EXPORT_SYMBOL(replace_mount_options);
1039 #ifdef CONFIG_PROC_FS
1040 /* iterator; we want it to have access to namespace_sem, thus here... */
1041 static void *m_start(struct seq_file *m, loff_t *pos)
1043 struct proc_mounts *p = proc_mounts(m);
1045 down_read(&namespace_sem);
1046 return seq_list_start(&p->ns->list, *pos);
1049 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1051 struct proc_mounts *p = proc_mounts(m);
1053 return seq_list_next(v, &p->ns->list, pos);
1056 static void m_stop(struct seq_file *m, void *v)
1058 up_read(&namespace_sem);
1061 static int m_show(struct seq_file *m, void *v)
1063 struct proc_mounts *p = proc_mounts(m);
1064 struct mount *r = list_entry(v, struct mount, mnt_list);
1065 return p->show(m, &r->mnt);
1068 const struct seq_operations mounts_op = {
1074 #endif /* CONFIG_PROC_FS */
1077 * may_umount_tree - check if a mount tree is busy
1078 * @mnt: root of mount tree
1080 * This is called to check if a tree of mounts has any
1081 * open files, pwds, chroots or sub mounts that are
1084 int may_umount_tree(struct vfsmount *m)
1086 struct mount *mnt = real_mount(m);
1087 int actual_refs = 0;
1088 int minimum_refs = 0;
1092 /* write lock needed for mnt_get_count */
1093 br_write_lock(&vfsmount_lock);
1094 for (p = mnt; p; p = next_mnt(p, mnt)) {
1095 actual_refs += mnt_get_count(p);
1098 br_write_unlock(&vfsmount_lock);
1100 if (actual_refs > minimum_refs)
1106 EXPORT_SYMBOL(may_umount_tree);
1109 * may_umount - check if a mount point is busy
1110 * @mnt: root of mount
1112 * This is called to check if a mount point has any
1113 * open files, pwds, chroots or sub mounts. If the
1114 * mount has sub mounts this will return busy
1115 * regardless of whether the sub mounts are busy.
1117 * Doesn't take quota and stuff into account. IOW, in some cases it will
1118 * give false negatives. The main reason why it's here is that we need
1119 * a non-destructive way to look for easily umountable filesystems.
1121 int may_umount(struct vfsmount *mnt)
1124 down_read(&namespace_sem);
1125 br_write_lock(&vfsmount_lock);
1126 if (propagate_mount_busy(real_mount(mnt), 2))
1128 br_write_unlock(&vfsmount_lock);
1129 up_read(&namespace_sem);
1133 EXPORT_SYMBOL(may_umount);
1135 static LIST_HEAD(unmounted); /* protected by namespace_sem */
1137 static void namespace_unlock(void)
1142 if (likely(list_empty(&unmounted))) {
1143 up_write(&namespace_sem);
1147 list_splice_init(&unmounted, &head);
1148 up_write(&namespace_sem);
1150 while (!list_empty(&head)) {
1151 mnt = list_first_entry(&head, struct mount, mnt_hash);
1152 list_del_init(&mnt->mnt_hash);
1153 if (mnt_has_parent(mnt)) {
1154 struct dentry *dentry;
1157 br_write_lock(&vfsmount_lock);
1158 dentry = mnt->mnt_mountpoint;
1159 m = mnt->mnt_parent;
1160 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1161 mnt->mnt_parent = mnt;
1163 br_write_unlock(&vfsmount_lock);
1171 static inline void namespace_lock(void)
1173 down_write(&namespace_sem);
1177 * vfsmount lock must be held for write
1178 * namespace_sem must be held for write
1180 void umount_tree(struct mount *mnt, int propagate)
1182 LIST_HEAD(tmp_list);
1185 for (p = mnt; p; p = next_mnt(p, mnt))
1186 list_move(&p->mnt_hash, &tmp_list);
1189 propagate_umount(&tmp_list);
1191 list_for_each_entry(p, &tmp_list, mnt_hash) {
1192 list_del_init(&p->mnt_expire);
1193 list_del_init(&p->mnt_list);
1194 __touch_mnt_namespace(p->mnt_ns);
1196 list_del_init(&p->mnt_child);
1197 if (mnt_has_parent(p)) {
1198 p->mnt_parent->mnt_ghosts++;
1199 list_del_init(&p->mnt_mp_list);
1200 put_mountpoint(p->mnt_mp);
1203 change_mnt_propagation(p, MS_PRIVATE);
1205 list_splice(&tmp_list, &unmounted);
1208 static void shrink_submounts(struct mount *mnt);
1210 static int do_umount(struct mount *mnt, int flags)
1212 struct super_block *sb = mnt->mnt.mnt_sb;
1215 retval = security_sb_umount(&mnt->mnt, flags);
1220 * Allow userspace to request a mountpoint be expired rather than
1221 * unmounting unconditionally. Unmount only happens if:
1222 * (1) the mark is already set (the mark is cleared by mntput())
1223 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1225 if (flags & MNT_EXPIRE) {
1226 if (&mnt->mnt == current->fs->root.mnt ||
1227 flags & (MNT_FORCE | MNT_DETACH))
1231 * probably don't strictly need the lock here if we examined
1232 * all race cases, but it's a slowpath.
1234 br_write_lock(&vfsmount_lock);
1235 if (mnt_get_count(mnt) != 2) {
1236 br_write_unlock(&vfsmount_lock);
1239 br_write_unlock(&vfsmount_lock);
1241 if (!xchg(&mnt->mnt_expiry_mark, 1))
1246 * If we may have to abort operations to get out of this
1247 * mount, and they will themselves hold resources we must
1248 * allow the fs to do things. In the Unix tradition of
1249 * 'Gee thats tricky lets do it in userspace' the umount_begin
1250 * might fail to complete on the first run through as other tasks
1251 * must return, and the like. Thats for the mount program to worry
1252 * about for the moment.
1255 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1256 sb->s_op->umount_begin(sb);
1260 * No sense to grab the lock for this test, but test itself looks
1261 * somewhat bogus. Suggestions for better replacement?
1262 * Ho-hum... In principle, we might treat that as umount + switch
1263 * to rootfs. GC would eventually take care of the old vfsmount.
1264 * Actually it makes sense, especially if rootfs would contain a
1265 * /reboot - static binary that would close all descriptors and
1266 * call reboot(9). Then init(8) could umount root and exec /reboot.
1268 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1270 * Special case for "unmounting" root ...
1271 * we just try to remount it readonly.
1273 down_write(&sb->s_umount);
1274 if (!(sb->s_flags & MS_RDONLY))
1275 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1276 up_write(&sb->s_umount);
1281 br_write_lock(&vfsmount_lock);
1284 if (!(flags & MNT_DETACH))
1285 shrink_submounts(mnt);
1288 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1289 if (!list_empty(&mnt->mnt_list))
1290 umount_tree(mnt, 1);
1293 br_write_unlock(&vfsmount_lock);
1298 void detach_mounts(struct dentry *dentry)
1300 struct mountpoint *mp;
1304 if (!d_mountpoint(dentry))
1307 mp = new_mountpoint(dentry);
1311 br_write_lock(&vfsmount_lock);
1312 while (!list_empty(&mp->m_list)) {
1313 mnt = list_first_entry(&mp->m_list, struct mount, mnt_mp_list);
1314 umount_tree(mnt, 1);
1316 br_write_unlock(&vfsmount_lock);
1323 * Is the caller allowed to modify his namespace?
1325 static inline bool may_mount(void)
1327 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1331 * Now umount can handle mount points as well as block devices.
1332 * This is important for filesystems which use unnamed block devices.
1334 * We now support a flag for forced unmount like the other 'big iron'
1335 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1338 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1343 int lookup_flags = 0;
1345 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1351 if (!(flags & UMOUNT_NOFOLLOW))
1352 lookup_flags |= LOOKUP_FOLLOW;
1354 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1357 mnt = real_mount(path.mnt);
1359 if (path.dentry != path.mnt->mnt_root)
1361 if (!check_mnt(mnt))
1363 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1366 retval = do_umount(mnt, flags);
1368 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1370 mntput_no_expire(mnt);
1375 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1378 * The 2.0 compatible umount. No flags.
1380 SYSCALL_DEFINE1(oldumount, char __user *, name)
1382 return sys_umount(name, 0);
1387 static bool is_mnt_ns_file(struct dentry *dentry)
1389 /* Is this a proxy for a mount namespace? */
1390 struct inode *inode = dentry->d_inode;
1393 if (!proc_ns_inode(inode))
1396 ei = get_proc_ns(inode);
1397 if (ei->ns_ops != &mntns_operations)
1403 static bool mnt_ns_loop(struct dentry *dentry)
1405 /* Could bind mounting the mount namespace inode cause a
1406 * mount namespace loop?
1408 struct mnt_namespace *mnt_ns;
1409 if (!is_mnt_ns_file(dentry))
1412 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1413 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1416 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1419 struct mount *res, *p, *q, *r, *parent;
1421 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1422 return ERR_PTR(-EINVAL);
1424 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1425 return ERR_PTR(-EINVAL);
1427 res = q = clone_mnt(mnt, dentry, flag);
1431 q->mnt.mnt_flags &= ~MNT_LOCKED;
1432 q->mnt_mountpoint = mnt->mnt_mountpoint;
1435 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1437 if (!is_subdir(r->mnt_mountpoint, dentry))
1440 for (s = r; s; s = next_mnt(s, r)) {
1441 if (!(flag & CL_COPY_UNBINDABLE) &&
1442 IS_MNT_UNBINDABLE(s)) {
1443 s = skip_mnt_tree(s);
1446 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1447 is_mnt_ns_file(s->mnt.mnt_root)) {
1448 s = skip_mnt_tree(s);
1451 while (p != s->mnt_parent) {
1457 q = clone_mnt(p, p->mnt.mnt_root, flag);
1460 br_write_lock(&vfsmount_lock);
1461 list_add_tail(&q->mnt_list, &res->mnt_list);
1462 attach_mnt(q, parent, p->mnt_mp);
1463 br_write_unlock(&vfsmount_lock);
1469 br_write_lock(&vfsmount_lock);
1470 umount_tree(res, 0);
1471 br_write_unlock(&vfsmount_lock);
1476 /* Caller should check returned pointer for errors */
1478 struct vfsmount *collect_mounts(struct path *path)
1482 tree = copy_tree(real_mount(path->mnt), path->dentry,
1483 CL_COPY_ALL | CL_PRIVATE);
1486 return ERR_CAST(tree);
1490 void drop_collected_mounts(struct vfsmount *mnt)
1493 br_write_lock(&vfsmount_lock);
1494 umount_tree(real_mount(mnt), 0);
1495 br_write_unlock(&vfsmount_lock);
1499 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1500 struct vfsmount *root)
1503 int res = f(root, arg);
1506 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1507 res = f(&mnt->mnt, arg);
1514 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1518 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1519 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1520 mnt_release_group_id(p);
1524 static int invent_group_ids(struct mount *mnt, bool recurse)
1528 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1529 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1530 int err = mnt_alloc_group_id(p);
1532 cleanup_group_ids(mnt, p);
1542 * @source_mnt : mount tree to be attached
1543 * @nd : place the mount tree @source_mnt is attached
1544 * @parent_nd : if non-null, detach the source_mnt from its parent and
1545 * store the parent mount and mountpoint dentry.
1546 * (done when source_mnt is moved)
1548 * NOTE: in the table below explains the semantics when a source mount
1549 * of a given type is attached to a destination mount of a given type.
1550 * ---------------------------------------------------------------------------
1551 * | BIND MOUNT OPERATION |
1552 * |**************************************************************************
1553 * | source-->| shared | private | slave | unbindable |
1557 * |**************************************************************************
1558 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1560 * |non-shared| shared (+) | private | slave (*) | invalid |
1561 * ***************************************************************************
1562 * A bind operation clones the source mount and mounts the clone on the
1563 * destination mount.
1565 * (++) the cloned mount is propagated to all the mounts in the propagation
1566 * tree of the destination mount and the cloned mount is added to
1567 * the peer group of the source mount.
1568 * (+) the cloned mount is created under the destination mount and is marked
1569 * as shared. The cloned mount is added to the peer group of the source
1571 * (+++) the mount is propagated to all the mounts in the propagation tree
1572 * of the destination mount and the cloned mount is made slave
1573 * of the same master as that of the source mount. The cloned mount
1574 * is marked as 'shared and slave'.
1575 * (*) the cloned mount is made a slave of the same master as that of the
1578 * ---------------------------------------------------------------------------
1579 * | MOVE MOUNT OPERATION |
1580 * |**************************************************************************
1581 * | source-->| shared | private | slave | unbindable |
1585 * |**************************************************************************
1586 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1588 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1589 * ***************************************************************************
1591 * (+) the mount is moved to the destination. And is then propagated to
1592 * all the mounts in the propagation tree of the destination mount.
1593 * (+*) the mount is moved to the destination.
1594 * (+++) the mount is moved to the destination and is then propagated to
1595 * all the mounts belonging to the destination mount's propagation tree.
1596 * the mount is marked as 'shared and slave'.
1597 * (*) the mount continues to be a slave at the new location.
1599 * if the source mount is a tree, the operations explained above is
1600 * applied to each mount in the tree.
1601 * Must be called without spinlocks held, since this function can sleep
1604 static int attach_recursive_mnt(struct mount *source_mnt,
1605 struct mount *dest_mnt,
1606 struct mountpoint *dest_mp,
1607 struct path *parent_path)
1609 LIST_HEAD(tree_list);
1610 struct mount *child, *p;
1613 if (IS_MNT_SHARED(dest_mnt)) {
1614 err = invent_group_ids(source_mnt, true);
1618 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1620 goto out_cleanup_ids;
1622 br_write_lock(&vfsmount_lock);
1624 if (IS_MNT_SHARED(dest_mnt)) {
1625 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1629 detach_mnt(source_mnt, parent_path);
1630 attach_mnt(source_mnt, dest_mnt, dest_mp);
1631 touch_mnt_namespace(source_mnt->mnt_ns);
1633 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1634 commit_tree(source_mnt);
1637 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1638 list_del_init(&child->mnt_hash);
1641 br_write_unlock(&vfsmount_lock);
1646 if (IS_MNT_SHARED(dest_mnt))
1647 cleanup_group_ids(source_mnt, NULL);
1652 static struct mountpoint *lock_mount(struct path *path)
1654 struct vfsmount *mnt;
1655 struct dentry *dentry = path->dentry;
1657 mutex_lock(&dentry->d_inode->i_mutex);
1658 if (unlikely(cant_mount(dentry))) {
1659 mutex_unlock(&dentry->d_inode->i_mutex);
1660 return ERR_PTR(-ENOENT);
1663 mnt = lookup_mnt(path);
1665 struct mountpoint *mp = new_mountpoint(dentry);
1668 mutex_unlock(&dentry->d_inode->i_mutex);
1674 mutex_unlock(&path->dentry->d_inode->i_mutex);
1677 dentry = path->dentry = dget(mnt->mnt_root);
1681 static void unlock_mount(struct mountpoint *where)
1683 struct dentry *dentry = where->m_dentry;
1684 put_mountpoint(where);
1686 mutex_unlock(&dentry->d_inode->i_mutex);
1689 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1691 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1694 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1695 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1698 return attach_recursive_mnt(mnt, p, mp, NULL);
1702 * Sanity check the flags to change_mnt_propagation.
1705 static int flags_to_propagation_type(int flags)
1707 int type = flags & ~(MS_REC | MS_SILENT);
1709 /* Fail if any non-propagation flags are set */
1710 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1712 /* Only one propagation flag should be set */
1713 if (!is_power_of_2(type))
1719 * recursively change the type of the mountpoint.
1721 static int do_change_type(struct path *path, int flag)
1724 struct mount *mnt = real_mount(path->mnt);
1725 int recurse = flag & MS_REC;
1729 if (path->dentry != path->mnt->mnt_root)
1732 type = flags_to_propagation_type(flag);
1737 if (type == MS_SHARED) {
1738 err = invent_group_ids(mnt, recurse);
1743 br_write_lock(&vfsmount_lock);
1744 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1745 change_mnt_propagation(m, type);
1746 br_write_unlock(&vfsmount_lock);
1753 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1755 struct mount *child;
1756 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1757 if (!is_subdir(child->mnt_mountpoint, dentry))
1760 if (child->mnt.mnt_flags & MNT_LOCKED)
1767 * do loopback mount.
1769 static int do_loopback(struct path *path, const char *old_name,
1772 struct path old_path;
1773 struct mount *mnt = NULL, *old, *parent;
1774 struct mountpoint *mp;
1776 if (!old_name || !*old_name)
1778 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1783 if (mnt_ns_loop(old_path.dentry))
1786 mp = lock_mount(path);
1791 old = real_mount(old_path.mnt);
1792 parent = real_mount(path->mnt);
1795 if (IS_MNT_UNBINDABLE(old))
1798 if (!check_mnt(parent) || !check_mnt(old))
1801 if (!recurse && has_locked_children(old, old_path.dentry))
1805 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1807 mnt = clone_mnt(old, old_path.dentry, 0);
1814 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1816 err = graft_tree(mnt, parent, mp);
1818 br_write_lock(&vfsmount_lock);
1819 umount_tree(mnt, 0);
1820 br_write_unlock(&vfsmount_lock);
1825 path_put(&old_path);
1829 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1832 int readonly_request = 0;
1834 if (ms_flags & MS_RDONLY)
1835 readonly_request = 1;
1836 if (readonly_request == __mnt_is_readonly(mnt))
1839 if (mnt->mnt_flags & MNT_LOCK_READONLY)
1842 if (readonly_request)
1843 error = mnt_make_readonly(real_mount(mnt));
1845 __mnt_unmake_readonly(real_mount(mnt));
1850 * change filesystem flags. dir should be a physical root of filesystem.
1851 * If you've mounted a non-root directory somewhere and want to do remount
1852 * on it - tough luck.
1854 static int do_remount(struct path *path, int flags, int mnt_flags,
1858 struct super_block *sb = path->mnt->mnt_sb;
1859 struct mount *mnt = real_mount(path->mnt);
1861 if (!check_mnt(mnt))
1864 if (path->dentry != path->mnt->mnt_root)
1867 err = security_sb_remount(sb, data);
1871 down_write(&sb->s_umount);
1872 if (flags & MS_BIND)
1873 err = change_mount_flags(path->mnt, flags);
1874 else if (!capable(CAP_SYS_ADMIN))
1877 err = do_remount_sb(sb, flags, data, 0);
1879 br_write_lock(&vfsmount_lock);
1880 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1881 mnt->mnt.mnt_flags = mnt_flags;
1882 touch_mnt_namespace(mnt->mnt_ns);
1883 br_write_unlock(&vfsmount_lock);
1885 up_write(&sb->s_umount);
1889 static inline int tree_contains_unbindable(struct mount *mnt)
1892 for (p = mnt; p; p = next_mnt(p, mnt)) {
1893 if (IS_MNT_UNBINDABLE(p))
1899 static int do_move_mount(struct path *path, const char *old_name)
1901 struct path old_path, parent_path;
1904 struct mountpoint *mp;
1906 if (!old_name || !*old_name)
1908 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1912 mp = lock_mount(path);
1917 old = real_mount(old_path.mnt);
1918 p = real_mount(path->mnt);
1921 if (!check_mnt(p) || !check_mnt(old))
1924 if (old->mnt.mnt_flags & MNT_LOCKED)
1928 if (old_path.dentry != old_path.mnt->mnt_root)
1931 if (!mnt_has_parent(old))
1934 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1935 S_ISDIR(old_path.dentry->d_inode->i_mode))
1938 * Don't move a mount residing in a shared parent.
1940 if (IS_MNT_SHARED(old->mnt_parent))
1943 * Don't move a mount tree containing unbindable mounts to a destination
1944 * mount which is shared.
1946 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1949 for (; mnt_has_parent(p); p = p->mnt_parent)
1953 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
1957 /* if the mount is moved, it should no longer be expire
1959 list_del_init(&old->mnt_expire);
1964 path_put(&parent_path);
1965 path_put(&old_path);
1969 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1972 const char *subtype = strchr(fstype, '.');
1981 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1983 if (!mnt->mnt_sb->s_subtype)
1989 return ERR_PTR(err);
1993 * add a mount into a namespace's mount tree
1995 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1997 struct mountpoint *mp;
1998 struct mount *parent;
2001 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
2003 mp = lock_mount(path);
2007 parent = real_mount(path->mnt);
2009 if (unlikely(!check_mnt(parent))) {
2010 /* that's acceptable only for automounts done in private ns */
2011 if (!(mnt_flags & MNT_SHRINKABLE))
2013 /* ... and for those we'd better have mountpoint still alive */
2014 if (!parent->mnt_ns)
2018 /* Refuse the same filesystem on the same mount point */
2020 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2021 path->mnt->mnt_root == path->dentry)
2025 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2028 newmnt->mnt.mnt_flags = mnt_flags;
2029 err = graft_tree(newmnt, parent, mp);
2037 * create a new mount for userspace and request it to be added into the
2040 static int do_new_mount(struct path *path, const char *fstype, int flags,
2041 int mnt_flags, const char *name, void *data)
2043 struct file_system_type *type;
2044 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2045 struct vfsmount *mnt;
2051 type = get_fs_type(fstype);
2055 if (user_ns != &init_user_ns) {
2056 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2057 put_filesystem(type);
2060 /* Only in special cases allow devices from mounts
2061 * created outside the initial user namespace.
2063 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2065 mnt_flags |= MNT_NODEV;
2069 mnt = vfs_kern_mount(type, flags, name, data);
2070 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2071 !mnt->mnt_sb->s_subtype)
2072 mnt = fs_set_subtype(mnt, fstype);
2074 put_filesystem(type);
2076 return PTR_ERR(mnt);
2078 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2084 int finish_automount(struct vfsmount *m, struct path *path)
2086 struct mount *mnt = real_mount(m);
2088 /* The new mount record should have at least 2 refs to prevent it being
2089 * expired before we get a chance to add it
2091 BUG_ON(mnt_get_count(mnt) < 2);
2093 if (m->mnt_sb == path->mnt->mnt_sb &&
2094 m->mnt_root == path->dentry) {
2099 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2103 /* remove m from any expiration list it may be on */
2104 if (!list_empty(&mnt->mnt_expire)) {
2106 br_write_lock(&vfsmount_lock);
2107 list_del_init(&mnt->mnt_expire);
2108 br_write_unlock(&vfsmount_lock);
2117 * mnt_set_expiry - Put a mount on an expiration list
2118 * @mnt: The mount to list.
2119 * @expiry_list: The list to add the mount to.
2121 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2124 br_write_lock(&vfsmount_lock);
2126 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2128 br_write_unlock(&vfsmount_lock);
2131 EXPORT_SYMBOL(mnt_set_expiry);
2134 * process a list of expirable mountpoints with the intent of discarding any
2135 * mountpoints that aren't in use and haven't been touched since last we came
2138 void mark_mounts_for_expiry(struct list_head *mounts)
2140 struct mount *mnt, *next;
2141 LIST_HEAD(graveyard);
2143 if (list_empty(mounts))
2147 br_write_lock(&vfsmount_lock);
2149 /* extract from the expiration list every vfsmount that matches the
2150 * following criteria:
2151 * - only referenced by its parent vfsmount
2152 * - still marked for expiry (marked on the last call here; marks are
2153 * cleared by mntput())
2155 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2156 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2157 propagate_mount_busy(mnt, 1))
2159 list_move(&mnt->mnt_expire, &graveyard);
2161 while (!list_empty(&graveyard)) {
2162 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2163 touch_mnt_namespace(mnt->mnt_ns);
2164 umount_tree(mnt, 1);
2166 br_write_unlock(&vfsmount_lock);
2170 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2173 * Ripoff of 'select_parent()'
2175 * search the list of submounts for a given mountpoint, and move any
2176 * shrinkable submounts to the 'graveyard' list.
2178 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2180 struct mount *this_parent = parent;
2181 struct list_head *next;
2185 next = this_parent->mnt_mounts.next;
2187 while (next != &this_parent->mnt_mounts) {
2188 struct list_head *tmp = next;
2189 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2192 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2195 * Descend a level if the d_mounts list is non-empty.
2197 if (!list_empty(&mnt->mnt_mounts)) {
2202 if (!propagate_mount_busy(mnt, 1)) {
2203 list_move_tail(&mnt->mnt_expire, graveyard);
2208 * All done at this level ... ascend and resume the search
2210 if (this_parent != parent) {
2211 next = this_parent->mnt_child.next;
2212 this_parent = this_parent->mnt_parent;
2219 * process a list of expirable mountpoints with the intent of discarding any
2220 * submounts of a specific parent mountpoint
2222 * vfsmount_lock must be held for write
2224 static void shrink_submounts(struct mount *mnt)
2226 LIST_HEAD(graveyard);
2229 /* extract submounts of 'mountpoint' from the expiration list */
2230 while (select_submounts(mnt, &graveyard)) {
2231 while (!list_empty(&graveyard)) {
2232 m = list_first_entry(&graveyard, struct mount,
2234 touch_mnt_namespace(m->mnt_ns);
2241 * Some copy_from_user() implementations do not return the exact number of
2242 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2243 * Note that this function differs from copy_from_user() in that it will oops
2244 * on bad values of `to', rather than returning a short copy.
2246 static long exact_copy_from_user(void *to, const void __user * from,
2250 const char __user *f = from;
2253 if (!access_ok(VERIFY_READ, from, n))
2257 if (__get_user(c, f)) {
2268 int copy_mount_options(const void __user * data, unsigned long *where)
2278 if (!(page = __get_free_page(GFP_KERNEL)))
2281 /* We only care that *some* data at the address the user
2282 * gave us is valid. Just in case, we'll zero
2283 * the remainder of the page.
2285 /* copy_from_user cannot cross TASK_SIZE ! */
2286 size = TASK_SIZE - (unsigned long)data;
2287 if (size > PAGE_SIZE)
2290 i = size - exact_copy_from_user((void *)page, data, size);
2296 memset((char *)page + i, 0, PAGE_SIZE - i);
2301 int copy_mount_string(const void __user *data, char **where)
2310 tmp = strndup_user(data, PAGE_SIZE);
2312 return PTR_ERR(tmp);
2319 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2320 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2322 * data is a (void *) that can point to any structure up to
2323 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2324 * information (or be NULL).
2326 * Pre-0.97 versions of mount() didn't have a flags word.
2327 * When the flags word was introduced its top half was required
2328 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2329 * Therefore, if this magic number is present, it carries no information
2330 * and must be discarded.
2332 long do_mount(const char *dev_name, const char *dir_name,
2333 const char *type_page, unsigned long flags, void *data_page)
2340 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2341 flags &= ~MS_MGC_MSK;
2343 /* Basic sanity checks */
2345 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2349 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2351 /* ... and get the mountpoint */
2352 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2356 retval = security_sb_mount(dev_name, &path,
2357 type_page, flags, data_page);
2358 if (!retval && !may_mount())
2363 /* Default to relatime unless overriden */
2364 if (!(flags & MS_NOATIME))
2365 mnt_flags |= MNT_RELATIME;
2367 /* Separate the per-mountpoint flags */
2368 if (flags & MS_NOSUID)
2369 mnt_flags |= MNT_NOSUID;
2370 if (flags & MS_NODEV)
2371 mnt_flags |= MNT_NODEV;
2372 if (flags & MS_NOEXEC)
2373 mnt_flags |= MNT_NOEXEC;
2374 if (flags & MS_NOATIME)
2375 mnt_flags |= MNT_NOATIME;
2376 if (flags & MS_NODIRATIME)
2377 mnt_flags |= MNT_NODIRATIME;
2378 if (flags & MS_STRICTATIME)
2379 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2380 if (flags & MS_RDONLY)
2381 mnt_flags |= MNT_READONLY;
2383 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2384 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2387 if (flags & MS_REMOUNT)
2388 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2390 else if (flags & MS_BIND)
2391 retval = do_loopback(&path, dev_name, flags & MS_REC);
2392 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2393 retval = do_change_type(&path, flags);
2394 else if (flags & MS_MOVE)
2395 retval = do_move_mount(&path, dev_name);
2397 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2398 dev_name, data_page);
2404 static void free_mnt_ns(struct mnt_namespace *ns)
2406 proc_free_inum(ns->proc_inum);
2407 put_user_ns(ns->user_ns);
2412 * Assign a sequence number so we can detect when we attempt to bind
2413 * mount a reference to an older mount namespace into the current
2414 * mount namespace, preventing reference counting loops. A 64bit
2415 * number incrementing at 10Ghz will take 12,427 years to wrap which
2416 * is effectively never, so we can ignore the possibility.
2418 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2420 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2422 struct mnt_namespace *new_ns;
2425 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2427 return ERR_PTR(-ENOMEM);
2428 ret = proc_alloc_inum(&new_ns->proc_inum);
2431 return ERR_PTR(ret);
2433 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2434 atomic_set(&new_ns->count, 1);
2435 new_ns->root = NULL;
2436 INIT_LIST_HEAD(&new_ns->list);
2437 init_waitqueue_head(&new_ns->poll);
2439 new_ns->user_ns = get_user_ns(user_ns);
2444 * Allocate a new namespace structure and populate it with contents
2445 * copied from the namespace of the passed in task structure.
2447 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2448 struct user_namespace *user_ns, struct fs_struct *fs)
2450 struct mnt_namespace *new_ns;
2451 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2452 struct mount *p, *q;
2453 struct mount *old = mnt_ns->root;
2457 new_ns = alloc_mnt_ns(user_ns);
2462 /* First pass: copy the tree topology */
2463 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2464 if (user_ns != mnt_ns->user_ns)
2465 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2466 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2469 free_mnt_ns(new_ns);
2470 return ERR_CAST(new);
2473 list_add_tail(&new_ns->list, &new->mnt_list);
2476 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2477 * as belonging to new namespace. We have already acquired a private
2478 * fs_struct, so tsk->fs->lock is not needed.
2485 if (&p->mnt == fs->root.mnt) {
2486 fs->root.mnt = mntget(&q->mnt);
2489 if (&p->mnt == fs->pwd.mnt) {
2490 fs->pwd.mnt = mntget(&q->mnt);
2494 p = next_mnt(p, old);
2495 q = next_mnt(q, new);
2498 while (p->mnt.mnt_root != q->mnt.mnt_root)
2499 p = next_mnt(p, old);
2511 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2512 struct user_namespace *user_ns, struct fs_struct *new_fs)
2514 struct mnt_namespace *new_ns;
2519 if (!(flags & CLONE_NEWNS))
2522 new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2529 * create_mnt_ns - creates a private namespace and adds a root filesystem
2530 * @mnt: pointer to the new root filesystem mountpoint
2532 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2534 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2535 if (!IS_ERR(new_ns)) {
2536 struct mount *mnt = real_mount(m);
2537 mnt->mnt_ns = new_ns;
2539 list_add(&mnt->mnt_list, &new_ns->list);
2546 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2548 struct mnt_namespace *ns;
2549 struct super_block *s;
2553 ns = create_mnt_ns(mnt);
2555 return ERR_CAST(ns);
2557 err = vfs_path_lookup(mnt->mnt_root, mnt,
2558 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2563 return ERR_PTR(err);
2565 /* trade a vfsmount reference for active sb one */
2566 s = path.mnt->mnt_sb;
2567 atomic_inc(&s->s_active);
2569 /* lock the sucker */
2570 down_write(&s->s_umount);
2571 /* ... and return the root of (sub)tree on it */
2574 EXPORT_SYMBOL(mount_subtree);
2576 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2577 char __user *, type, unsigned long, flags, void __user *, data)
2581 struct filename *kernel_dir;
2583 unsigned long data_page;
2585 ret = copy_mount_string(type, &kernel_type);
2589 kernel_dir = getname(dir_name);
2590 if (IS_ERR(kernel_dir)) {
2591 ret = PTR_ERR(kernel_dir);
2595 ret = copy_mount_string(dev_name, &kernel_dev);
2599 ret = copy_mount_options(data, &data_page);
2603 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2604 (void *) data_page);
2606 free_page(data_page);
2610 putname(kernel_dir);
2618 * Return true if path is reachable from root
2620 * namespace_sem or vfsmount_lock is held
2622 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2623 const struct path *root)
2625 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2626 dentry = mnt->mnt_mountpoint;
2627 mnt = mnt->mnt_parent;
2629 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2632 int path_is_under(struct path *path1, struct path *path2)
2635 br_read_lock(&vfsmount_lock);
2636 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2637 br_read_unlock(&vfsmount_lock);
2640 EXPORT_SYMBOL(path_is_under);
2643 * pivot_root Semantics:
2644 * Moves the root file system of the current process to the directory put_old,
2645 * makes new_root as the new root file system of the current process, and sets
2646 * root/cwd of all processes which had them on the current root to new_root.
2649 * The new_root and put_old must be directories, and must not be on the
2650 * same file system as the current process root. The put_old must be
2651 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2652 * pointed to by put_old must yield the same directory as new_root. No other
2653 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2655 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2656 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2657 * in this situation.
2660 * - we don't move root/cwd if they are not at the root (reason: if something
2661 * cared enough to change them, it's probably wrong to force them elsewhere)
2662 * - it's okay to pick a root that isn't the root of a file system, e.g.
2663 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2664 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2667 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2668 const char __user *, put_old)
2670 struct path new, old, parent_path, root_parent, root;
2671 struct mount *new_mnt, *root_mnt, *old_mnt;
2672 struct mountpoint *old_mp, *root_mp;
2678 error = user_path_dir(new_root, &new);
2682 error = user_path_dir(put_old, &old);
2686 error = security_sb_pivotroot(&old, &new);
2690 get_fs_root(current->fs, &root);
2691 old_mp = lock_mount(&old);
2692 error = PTR_ERR(old_mp);
2697 new_mnt = real_mount(new.mnt);
2698 root_mnt = real_mount(root.mnt);
2699 old_mnt = real_mount(old.mnt);
2700 if (IS_MNT_SHARED(old_mnt) ||
2701 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2702 IS_MNT_SHARED(root_mnt->mnt_parent))
2704 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2706 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2709 if (d_unlinked(new.dentry))
2712 if (new_mnt == root_mnt || old_mnt == root_mnt)
2713 goto out4; /* loop, on the same file system */
2715 if (root.mnt->mnt_root != root.dentry)
2716 goto out4; /* not a mountpoint */
2717 if (!mnt_has_parent(root_mnt))
2718 goto out4; /* not attached */
2719 root_mp = root_mnt->mnt_mp;
2720 if (new.mnt->mnt_root != new.dentry)
2721 goto out4; /* not a mountpoint */
2722 if (!mnt_has_parent(new_mnt))
2723 goto out4; /* not attached */
2724 /* make sure we can reach put_old from new_root */
2725 if (!is_path_reachable(old_mnt, old.dentry, &new))
2727 root_mp->m_count++; /* pin it so it won't go away */
2728 br_write_lock(&vfsmount_lock);
2729 detach_mnt(new_mnt, &parent_path);
2730 detach_mnt(root_mnt, &root_parent);
2731 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2732 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2733 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2735 /* mount old root on put_old */
2736 attach_mnt(root_mnt, old_mnt, old_mp);
2737 /* mount new_root on / */
2738 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2739 touch_mnt_namespace(current->nsproxy->mnt_ns);
2740 br_write_unlock(&vfsmount_lock);
2741 chroot_fs_refs(&root, &new);
2742 put_mountpoint(root_mp);
2745 unlock_mount(old_mp);
2747 path_put(&root_parent);
2748 path_put(&parent_path);
2760 static void __init init_mount_tree(void)
2762 struct vfsmount *mnt;
2763 struct mnt_namespace *ns;
2765 struct file_system_type *type;
2767 type = get_fs_type("rootfs");
2769 panic("Can't find rootfs type");
2770 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2771 put_filesystem(type);
2773 panic("Can't create rootfs");
2775 ns = create_mnt_ns(mnt);
2777 panic("Can't allocate initial namespace");
2779 init_task.nsproxy->mnt_ns = ns;
2783 root.dentry = mnt->mnt_root;
2785 set_fs_pwd(current->fs, &root);
2786 set_fs_root(current->fs, &root);
2789 void __init mnt_init(void)
2794 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2795 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2797 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2798 mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2800 if (!mount_hashtable || !mountpoint_hashtable)
2801 panic("Failed to allocate mount hash table\n");
2803 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2805 for (u = 0; u < HASH_SIZE; u++)
2806 INIT_LIST_HEAD(&mount_hashtable[u]);
2807 for (u = 0; u < HASH_SIZE; u++)
2808 INIT_LIST_HEAD(&mountpoint_hashtable[u]);
2810 br_lock_init(&vfsmount_lock);
2814 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2816 fs_kobj = kobject_create_and_add("fs", NULL);
2818 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2823 void put_mnt_ns(struct mnt_namespace *ns)
2825 if (!atomic_dec_and_test(&ns->count))
2827 drop_collected_mounts(&ns->root->mnt);
2831 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2833 struct vfsmount *mnt;
2834 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2837 * it is a longterm mount, don't release mnt until
2838 * we unmount before file sys is unregistered
2840 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2844 EXPORT_SYMBOL_GPL(kern_mount_data);
2846 void kern_unmount(struct vfsmount *mnt)
2848 /* release long term mount so mount point can be released */
2849 if (!IS_ERR_OR_NULL(mnt)) {
2850 br_write_lock(&vfsmount_lock);
2851 real_mount(mnt)->mnt_ns = NULL;
2852 br_write_unlock(&vfsmount_lock);
2856 EXPORT_SYMBOL(kern_unmount);
2858 bool our_mnt(struct vfsmount *mnt)
2860 return check_mnt(real_mount(mnt));
2863 bool current_chrooted(void)
2865 /* Does the current process have a non-standard root */
2866 struct path ns_root;
2867 struct path fs_root;
2870 /* Find the namespace root */
2871 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
2872 ns_root.dentry = ns_root.mnt->mnt_root;
2874 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2877 get_fs_root(current->fs, &fs_root);
2879 chrooted = !path_equal(&fs_root, &ns_root);
2887 bool fs_fully_visible(struct file_system_type *type)
2889 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2891 bool visible = false;
2896 down_read(&namespace_sem);
2897 list_for_each_entry(mnt, &ns->list, mnt_list) {
2898 struct mount *child;
2899 if (mnt->mnt.mnt_sb->s_type != type)
2902 /* This mount is not fully visible if there are any child mounts
2903 * that cover anything except for empty directories.
2905 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2906 struct inode *inode = child->mnt_mountpoint->d_inode;
2907 if (!S_ISDIR(inode->i_mode))
2909 if (inode->i_nlink != 2)
2917 up_read(&namespace_sem);
2921 static void *mntns_get(struct task_struct *task)
2923 struct mnt_namespace *ns = NULL;
2924 struct nsproxy *nsproxy;
2927 nsproxy = task_nsproxy(task);
2929 ns = nsproxy->mnt_ns;
2937 static void mntns_put(void *ns)
2942 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2944 struct fs_struct *fs = current->fs;
2945 struct mnt_namespace *mnt_ns = ns;
2948 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2949 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2950 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2957 put_mnt_ns(nsproxy->mnt_ns);
2958 nsproxy->mnt_ns = mnt_ns;
2961 root.mnt = &mnt_ns->root->mnt;
2962 root.dentry = mnt_ns->root->mnt.mnt_root;
2964 while(d_mountpoint(root.dentry) && follow_down_one(&root))
2967 /* Update the pwd and root */
2968 set_fs_pwd(fs, &root);
2969 set_fs_root(fs, &root);
2975 static unsigned int mntns_inum(void *ns)
2977 struct mnt_namespace *mnt_ns = ns;
2978 return mnt_ns->proc_inum;
2981 const struct proc_ns_operations mntns_operations = {
2983 .type = CLONE_NEWNS,
2986 .install = mntns_install,