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/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 /* Maximum number of mounts in a mount namespace */
31 unsigned int sysctl_mount_max __read_mostly = 100000;
33 static unsigned int m_hash_mask __read_mostly;
34 static unsigned int m_hash_shift __read_mostly;
35 static unsigned int mp_hash_mask __read_mostly;
36 static unsigned int mp_hash_shift __read_mostly;
38 static __initdata unsigned long mhash_entries;
39 static int __init set_mhash_entries(char *str)
43 mhash_entries = simple_strtoul(str, &str, 0);
46 __setup("mhash_entries=", set_mhash_entries);
48 static __initdata unsigned long mphash_entries;
49 static int __init set_mphash_entries(char *str)
53 mphash_entries = simple_strtoul(str, &str, 0);
56 __setup("mphash_entries=", set_mphash_entries);
59 static DEFINE_IDA(mnt_id_ida);
60 static DEFINE_IDA(mnt_group_ida);
61 static DEFINE_SPINLOCK(mnt_id_lock);
62 static int mnt_id_start = 0;
63 static int mnt_group_start = 1;
65 static struct hlist_head *mount_hashtable __read_mostly;
66 static struct hlist_head *mountpoint_hashtable __read_mostly;
67 static struct kmem_cache *mnt_cache __read_mostly;
68 static DECLARE_RWSEM(namespace_sem);
71 struct kobject *fs_kobj;
72 EXPORT_SYMBOL_GPL(fs_kobj);
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
84 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
86 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
88 tmp = tmp + (tmp >> m_hash_shift);
89 return &mount_hashtable[tmp & m_hash_mask];
92 static inline struct hlist_head *mp_hash(struct dentry *dentry)
94 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> mp_hash_shift);
96 return &mountpoint_hashtable[tmp & mp_hash_mask];
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
103 static int mnt_alloc_id(struct mount *mnt)
108 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
109 spin_lock(&mnt_id_lock);
110 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
112 mnt_id_start = mnt->mnt_id + 1;
113 spin_unlock(&mnt_id_lock);
120 static void mnt_free_id(struct mount *mnt)
122 int id = mnt->mnt_id;
123 spin_lock(&mnt_id_lock);
124 ida_remove(&mnt_id_ida, id);
125 if (mnt_id_start > id)
127 spin_unlock(&mnt_id_lock);
131 * Allocate a new peer group ID
133 * mnt_group_ida is protected by namespace_sem
135 static int mnt_alloc_group_id(struct mount *mnt)
139 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
142 res = ida_get_new_above(&mnt_group_ida,
146 mnt_group_start = mnt->mnt_group_id + 1;
152 * Release a peer group ID
154 void mnt_release_group_id(struct mount *mnt)
156 int id = mnt->mnt_group_id;
157 ida_remove(&mnt_group_ida, id);
158 if (mnt_group_start > id)
159 mnt_group_start = id;
160 mnt->mnt_group_id = 0;
164 * vfsmount lock must be held for read
166 static inline void mnt_add_count(struct mount *mnt, int n)
169 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
178 * vfsmount lock must be held for write
180 unsigned int mnt_get_count(struct mount *mnt)
183 unsigned int count = 0;
186 for_each_possible_cpu(cpu) {
187 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
192 return mnt->mnt_count;
196 static void drop_mountpoint(struct fs_pin *p)
198 struct mount *m = container_of(p, struct mount, mnt_umount);
199 dput(m->mnt_ex_mountpoint);
204 static struct mount *alloc_vfsmnt(const char *name)
206 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
210 err = mnt_alloc_id(mnt);
215 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
216 if (!mnt->mnt_devname)
221 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
223 goto out_free_devname;
225 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
228 mnt->mnt_writers = 0;
231 INIT_HLIST_NODE(&mnt->mnt_hash);
232 INIT_LIST_HEAD(&mnt->mnt_child);
233 INIT_LIST_HEAD(&mnt->mnt_mounts);
234 INIT_LIST_HEAD(&mnt->mnt_list);
235 INIT_LIST_HEAD(&mnt->mnt_expire);
236 INIT_LIST_HEAD(&mnt->mnt_share);
237 INIT_LIST_HEAD(&mnt->mnt_slave_list);
238 INIT_LIST_HEAD(&mnt->mnt_slave);
239 INIT_HLIST_NODE(&mnt->mnt_mp_list);
240 INIT_LIST_HEAD(&mnt->mnt_umounting);
241 #ifdef CONFIG_FSNOTIFY
242 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
244 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
250 kfree_const(mnt->mnt_devname);
255 kmem_cache_free(mnt_cache, mnt);
260 * Most r/o checks on a fs are for operations that take
261 * discrete amounts of time, like a write() or unlink().
262 * We must keep track of when those operations start
263 * (for permission checks) and when they end, so that
264 * we can determine when writes are able to occur to
268 * __mnt_is_readonly: check whether a mount is read-only
269 * @mnt: the mount to check for its write status
271 * This shouldn't be used directly ouside of the VFS.
272 * It does not guarantee that the filesystem will stay
273 * r/w, just that it is right *now*. This can not and
274 * should not be used in place of IS_RDONLY(inode).
275 * mnt_want/drop_write() will _keep_ the filesystem
278 int __mnt_is_readonly(struct vfsmount *mnt)
280 if (mnt->mnt_flags & MNT_READONLY)
282 if (mnt->mnt_sb->s_flags & MS_RDONLY)
286 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
288 static inline void mnt_inc_writers(struct mount *mnt)
291 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
297 static inline void mnt_dec_writers(struct mount *mnt)
300 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
306 static unsigned int mnt_get_writers(struct mount *mnt)
309 unsigned int count = 0;
312 for_each_possible_cpu(cpu) {
313 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
318 return mnt->mnt_writers;
322 static int mnt_is_readonly(struct vfsmount *mnt)
324 if (mnt->mnt_sb->s_readonly_remount)
326 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
328 return __mnt_is_readonly(mnt);
332 * Most r/o & frozen checks on a fs are for operations that take discrete
333 * amounts of time, like a write() or unlink(). We must keep track of when
334 * those operations start (for permission checks) and when they end, so that we
335 * can determine when writes are able to occur to a filesystem.
338 * __mnt_want_write - get write access to a mount without freeze protection
339 * @m: the mount on which to take a write
341 * This tells the low-level filesystem that a write is about to be performed to
342 * it, and makes sure that writes are allowed (mnt it read-write) before
343 * returning success. This operation does not protect against filesystem being
344 * frozen. When the write operation is finished, __mnt_drop_write() must be
345 * called. This is effectively a refcount.
347 int __mnt_want_write(struct vfsmount *m)
349 struct mount *mnt = real_mount(m);
353 mnt_inc_writers(mnt);
355 * The store to mnt_inc_writers must be visible before we pass
356 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
357 * incremented count after it has set MNT_WRITE_HOLD.
360 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
363 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
364 * be set to match its requirements. So we must not load that until
365 * MNT_WRITE_HOLD is cleared.
368 if (mnt_is_readonly(m)) {
369 mnt_dec_writers(mnt);
378 * mnt_want_write - get write access to a mount
379 * @m: the mount on which to take a write
381 * This tells the low-level filesystem that a write is about to be performed to
382 * it, and makes sure that writes are allowed (mount is read-write, filesystem
383 * is not frozen) before returning success. When the write operation is
384 * finished, mnt_drop_write() must be called. This is effectively a refcount.
386 int mnt_want_write(struct vfsmount *m)
390 sb_start_write(m->mnt_sb);
391 ret = __mnt_want_write(m);
393 sb_end_write(m->mnt_sb);
396 EXPORT_SYMBOL_GPL(mnt_want_write);
399 * mnt_clone_write - get write access to a mount
400 * @mnt: the mount on which to take a write
402 * This is effectively like mnt_want_write, except
403 * it must only be used to take an extra write reference
404 * on a mountpoint that we already know has a write reference
405 * on it. This allows some optimisation.
407 * After finished, mnt_drop_write must be called as usual to
408 * drop the reference.
410 int mnt_clone_write(struct vfsmount *mnt)
412 /* superblock may be r/o */
413 if (__mnt_is_readonly(mnt))
416 mnt_inc_writers(real_mount(mnt));
420 EXPORT_SYMBOL_GPL(mnt_clone_write);
423 * __mnt_want_write_file - get write access to a file's mount
424 * @file: the file who's mount on which to take a write
426 * This is like __mnt_want_write, but it takes a file and can
427 * do some optimisations if the file is open for write already
429 int __mnt_want_write_file(struct file *file)
431 if (!(file->f_mode & FMODE_WRITER))
432 return __mnt_want_write(file->f_path.mnt);
434 return mnt_clone_write(file->f_path.mnt);
438 * mnt_want_write_file - get write access to a file's mount
439 * @file: the file who's mount on which to take a write
441 * This is like mnt_want_write, but it takes a file and can
442 * do some optimisations if the file is open for write already
444 int mnt_want_write_file(struct file *file)
448 sb_start_write(file->f_path.mnt->mnt_sb);
449 ret = __mnt_want_write_file(file);
451 sb_end_write(file->f_path.mnt->mnt_sb);
454 EXPORT_SYMBOL_GPL(mnt_want_write_file);
457 * __mnt_drop_write - give up write access to a mount
458 * @mnt: the mount on which to give up write access
460 * Tells the low-level filesystem that we are done
461 * performing writes to it. Must be matched with
462 * __mnt_want_write() call above.
464 void __mnt_drop_write(struct vfsmount *mnt)
467 mnt_dec_writers(real_mount(mnt));
472 * mnt_drop_write - give up write access to a mount
473 * @mnt: the mount on which to give up write access
475 * Tells the low-level filesystem that we are done performing writes to it and
476 * also allows filesystem to be frozen again. Must be matched with
477 * mnt_want_write() call above.
479 void mnt_drop_write(struct vfsmount *mnt)
481 __mnt_drop_write(mnt);
482 sb_end_write(mnt->mnt_sb);
484 EXPORT_SYMBOL_GPL(mnt_drop_write);
486 void __mnt_drop_write_file(struct file *file)
488 __mnt_drop_write(file->f_path.mnt);
491 void mnt_drop_write_file(struct file *file)
493 mnt_drop_write(file->f_path.mnt);
495 EXPORT_SYMBOL(mnt_drop_write_file);
497 static int mnt_make_readonly(struct mount *mnt)
502 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
504 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
505 * should be visible before we do.
510 * With writers on hold, if this value is zero, then there are
511 * definitely no active writers (although held writers may subsequently
512 * increment the count, they'll have to wait, and decrement it after
513 * seeing MNT_READONLY).
515 * It is OK to have counter incremented on one CPU and decremented on
516 * another: the sum will add up correctly. The danger would be when we
517 * sum up each counter, if we read a counter before it is incremented,
518 * but then read another CPU's count which it has been subsequently
519 * decremented from -- we would see more decrements than we should.
520 * MNT_WRITE_HOLD protects against this scenario, because
521 * mnt_want_write first increments count, then smp_mb, then spins on
522 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
523 * we're counting up here.
525 if (mnt_get_writers(mnt) > 0)
528 mnt->mnt.mnt_flags |= MNT_READONLY;
530 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
531 * that become unheld will see MNT_READONLY.
534 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
539 static void __mnt_unmake_readonly(struct mount *mnt)
542 mnt->mnt.mnt_flags &= ~MNT_READONLY;
546 int sb_prepare_remount_readonly(struct super_block *sb)
551 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
552 if (atomic_long_read(&sb->s_remove_count))
556 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
557 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
558 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
560 if (mnt_get_writers(mnt) > 0) {
566 if (!err && atomic_long_read(&sb->s_remove_count))
570 sb->s_readonly_remount = 1;
573 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
574 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
575 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
582 static void free_vfsmnt(struct mount *mnt)
584 kfree_const(mnt->mnt_devname);
586 free_percpu(mnt->mnt_pcp);
588 kmem_cache_free(mnt_cache, mnt);
591 static void delayed_free_vfsmnt(struct rcu_head *head)
593 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
596 /* call under rcu_read_lock */
597 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
600 if (read_seqretry(&mount_lock, seq))
604 mnt = real_mount(bastard);
605 mnt_add_count(mnt, 1);
606 if (likely(!read_seqretry(&mount_lock, seq)))
608 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
609 mnt_add_count(mnt, -1);
615 /* call under rcu_read_lock */
616 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
618 int res = __legitimize_mnt(bastard, seq);
621 if (unlikely(res < 0)) {
630 * find the first mount at @dentry on vfsmount @mnt.
631 * call under rcu_read_lock()
633 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
635 struct hlist_head *head = m_hash(mnt, dentry);
638 hlist_for_each_entry_rcu(p, head, mnt_hash)
639 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
645 * lookup_mnt - Return the first child mount mounted at path
647 * "First" means first mounted chronologically. If you create the
650 * mount /dev/sda1 /mnt
651 * mount /dev/sda2 /mnt
652 * mount /dev/sda3 /mnt
654 * Then lookup_mnt() on the base /mnt dentry in the root mount will
655 * return successively the root dentry and vfsmount of /dev/sda1, then
656 * /dev/sda2, then /dev/sda3, then NULL.
658 * lookup_mnt takes a reference to the found vfsmount.
660 struct vfsmount *lookup_mnt(struct path *path)
662 struct mount *child_mnt;
668 seq = read_seqbegin(&mount_lock);
669 child_mnt = __lookup_mnt(path->mnt, path->dentry);
670 m = child_mnt ? &child_mnt->mnt : NULL;
671 } while (!legitimize_mnt(m, seq));
677 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
678 * current mount namespace.
680 * The common case is dentries are not mountpoints at all and that
681 * test is handled inline. For the slow case when we are actually
682 * dealing with a mountpoint of some kind, walk through all of the
683 * mounts in the current mount namespace and test to see if the dentry
686 * The mount_hashtable is not usable in the context because we
687 * need to identify all mounts that may be in the current mount
688 * namespace not just a mount that happens to have some specified
691 bool __is_local_mountpoint(struct dentry *dentry)
693 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
695 bool is_covered = false;
697 if (!d_mountpoint(dentry))
700 down_read(&namespace_sem);
701 list_for_each_entry(mnt, &ns->list, mnt_list) {
702 is_covered = (mnt->mnt_mountpoint == dentry);
706 up_read(&namespace_sem);
711 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
713 struct hlist_head *chain = mp_hash(dentry);
714 struct mountpoint *mp;
716 hlist_for_each_entry(mp, chain, m_hash) {
717 if (mp->m_dentry == dentry) {
718 /* might be worth a WARN_ON() */
719 if (d_unlinked(dentry))
720 return ERR_PTR(-ENOENT);
728 static struct mountpoint *get_mountpoint(struct dentry *dentry)
730 struct mountpoint *mp, *new = NULL;
733 if (d_mountpoint(dentry)) {
735 read_seqlock_excl(&mount_lock);
736 mp = lookup_mountpoint(dentry);
737 read_sequnlock_excl(&mount_lock);
743 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
745 return ERR_PTR(-ENOMEM);
748 /* Exactly one processes may set d_mounted */
749 ret = d_set_mounted(dentry);
751 /* Someone else set d_mounted? */
755 /* The dentry is not available as a mountpoint? */
760 /* Add the new mountpoint to the hash table */
761 read_seqlock_excl(&mount_lock);
762 new->m_dentry = dentry;
764 hlist_add_head(&new->m_hash, mp_hash(dentry));
765 INIT_HLIST_HEAD(&new->m_list);
766 read_sequnlock_excl(&mount_lock);
775 static void put_mountpoint(struct mountpoint *mp)
777 if (!--mp->m_count) {
778 struct dentry *dentry = mp->m_dentry;
779 BUG_ON(!hlist_empty(&mp->m_list));
780 spin_lock(&dentry->d_lock);
781 dentry->d_flags &= ~DCACHE_MOUNTED;
782 spin_unlock(&dentry->d_lock);
783 hlist_del(&mp->m_hash);
788 static inline int check_mnt(struct mount *mnt)
790 return mnt->mnt_ns == current->nsproxy->mnt_ns;
794 * vfsmount lock must be held for write
796 static void touch_mnt_namespace(struct mnt_namespace *ns)
800 wake_up_interruptible(&ns->poll);
805 * vfsmount lock must be held for write
807 static void __touch_mnt_namespace(struct mnt_namespace *ns)
809 if (ns && ns->event != event) {
811 wake_up_interruptible(&ns->poll);
816 * vfsmount lock must be held for write
818 static void unhash_mnt(struct mount *mnt)
820 mnt->mnt_parent = mnt;
821 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
822 list_del_init(&mnt->mnt_child);
823 hlist_del_init_rcu(&mnt->mnt_hash);
824 hlist_del_init(&mnt->mnt_mp_list);
825 put_mountpoint(mnt->mnt_mp);
830 * vfsmount lock must be held for write
832 static void detach_mnt(struct mount *mnt, struct path *old_path)
834 old_path->dentry = mnt->mnt_mountpoint;
835 old_path->mnt = &mnt->mnt_parent->mnt;
840 * vfsmount lock must be held for write
842 static void umount_mnt(struct mount *mnt)
844 /* old mountpoint will be dropped when we can do that */
845 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
850 * vfsmount lock must be held for write
852 void mnt_set_mountpoint(struct mount *mnt,
853 struct mountpoint *mp,
854 struct mount *child_mnt)
857 mnt_add_count(mnt, 1); /* essentially, that's mntget */
858 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
859 child_mnt->mnt_parent = mnt;
860 child_mnt->mnt_mp = mp;
861 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
864 static void __attach_mnt(struct mount *mnt, struct mount *parent)
866 hlist_add_head_rcu(&mnt->mnt_hash,
867 m_hash(&parent->mnt, mnt->mnt_mountpoint));
868 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
872 * vfsmount lock must be held for write
874 static void attach_mnt(struct mount *mnt,
875 struct mount *parent,
876 struct mountpoint *mp)
878 mnt_set_mountpoint(parent, mp, mnt);
879 __attach_mnt(mnt, parent);
882 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
884 struct mountpoint *old_mp = mnt->mnt_mp;
885 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
886 struct mount *old_parent = mnt->mnt_parent;
888 list_del_init(&mnt->mnt_child);
889 hlist_del_init(&mnt->mnt_mp_list);
890 hlist_del_init_rcu(&mnt->mnt_hash);
892 attach_mnt(mnt, parent, mp);
894 put_mountpoint(old_mp);
897 * Safely avoid even the suggestion this code might sleep or
898 * lock the mount hash by taking advantage of the knowledge that
899 * mnt_change_mountpoint will not release the final reference
902 * During mounting, the mount passed in as the parent mount will
903 * continue to use the old mountpoint and during unmounting, the
904 * old mountpoint will continue to exist until namespace_unlock,
905 * which happens well after mnt_change_mountpoint.
907 spin_lock(&old_mountpoint->d_lock);
908 old_mountpoint->d_lockref.count--;
909 spin_unlock(&old_mountpoint->d_lock);
911 mnt_add_count(old_parent, -1);
915 * vfsmount lock must be held for write
917 static void commit_tree(struct mount *mnt)
919 struct mount *parent = mnt->mnt_parent;
922 struct mnt_namespace *n = parent->mnt_ns;
924 BUG_ON(parent == mnt);
926 list_add_tail(&head, &mnt->mnt_list);
927 list_for_each_entry(m, &head, mnt_list)
930 list_splice(&head, n->list.prev);
932 n->mounts += n->pending_mounts;
933 n->pending_mounts = 0;
935 __attach_mnt(mnt, parent);
936 touch_mnt_namespace(n);
939 static struct mount *next_mnt(struct mount *p, struct mount *root)
941 struct list_head *next = p->mnt_mounts.next;
942 if (next == &p->mnt_mounts) {
946 next = p->mnt_child.next;
947 if (next != &p->mnt_parent->mnt_mounts)
952 return list_entry(next, struct mount, mnt_child);
955 static struct mount *skip_mnt_tree(struct mount *p)
957 struct list_head *prev = p->mnt_mounts.prev;
958 while (prev != &p->mnt_mounts) {
959 p = list_entry(prev, struct mount, mnt_child);
960 prev = p->mnt_mounts.prev;
966 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
972 return ERR_PTR(-ENODEV);
974 mnt = alloc_vfsmnt(name);
976 return ERR_PTR(-ENOMEM);
978 if (flags & MS_KERNMOUNT)
979 mnt->mnt.mnt_flags = MNT_INTERNAL;
981 root = mount_fs(type, flags, name, data);
985 return ERR_CAST(root);
988 mnt->mnt.mnt_root = root;
989 mnt->mnt.mnt_sb = root->d_sb;
990 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
991 mnt->mnt_parent = mnt;
993 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
997 EXPORT_SYMBOL_GPL(vfs_kern_mount);
999 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1002 struct super_block *sb = old->mnt.mnt_sb;
1006 mnt = alloc_vfsmnt(old->mnt_devname);
1008 return ERR_PTR(-ENOMEM);
1010 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1011 mnt->mnt_group_id = 0; /* not a peer of original */
1013 mnt->mnt_group_id = old->mnt_group_id;
1015 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1016 err = mnt_alloc_group_id(mnt);
1021 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1022 /* Don't allow unprivileged users to change mount flags */
1023 if (flag & CL_UNPRIVILEGED) {
1024 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1026 if (mnt->mnt.mnt_flags & MNT_READONLY)
1027 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1029 if (mnt->mnt.mnt_flags & MNT_NODEV)
1030 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1032 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1033 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1035 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1036 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1039 /* Don't allow unprivileged users to reveal what is under a mount */
1040 if ((flag & CL_UNPRIVILEGED) &&
1041 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1042 mnt->mnt.mnt_flags |= MNT_LOCKED;
1044 atomic_inc(&sb->s_active);
1045 mnt->mnt.mnt_sb = sb;
1046 mnt->mnt.mnt_root = dget(root);
1047 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1048 mnt->mnt_parent = mnt;
1050 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1051 unlock_mount_hash();
1053 if ((flag & CL_SLAVE) ||
1054 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1055 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1056 mnt->mnt_master = old;
1057 CLEAR_MNT_SHARED(mnt);
1058 } else if (!(flag & CL_PRIVATE)) {
1059 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1060 list_add(&mnt->mnt_share, &old->mnt_share);
1061 if (IS_MNT_SLAVE(old))
1062 list_add(&mnt->mnt_slave, &old->mnt_slave);
1063 mnt->mnt_master = old->mnt_master;
1065 if (flag & CL_MAKE_SHARED)
1066 set_mnt_shared(mnt);
1068 /* stick the duplicate mount on the same expiry list
1069 * as the original if that was on one */
1070 if (flag & CL_EXPIRE) {
1071 if (!list_empty(&old->mnt_expire))
1072 list_add(&mnt->mnt_expire, &old->mnt_expire);
1080 return ERR_PTR(err);
1083 static void cleanup_mnt(struct mount *mnt)
1086 * This probably indicates that somebody messed
1087 * up a mnt_want/drop_write() pair. If this
1088 * happens, the filesystem was probably unable
1089 * to make r/w->r/o transitions.
1092 * The locking used to deal with mnt_count decrement provides barriers,
1093 * so mnt_get_writers() below is safe.
1095 WARN_ON(mnt_get_writers(mnt));
1096 if (unlikely(mnt->mnt_pins.first))
1098 fsnotify_vfsmount_delete(&mnt->mnt);
1099 dput(mnt->mnt.mnt_root);
1100 deactivate_super(mnt->mnt.mnt_sb);
1102 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1105 static void __cleanup_mnt(struct rcu_head *head)
1107 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1110 static LLIST_HEAD(delayed_mntput_list);
1111 static void delayed_mntput(struct work_struct *unused)
1113 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1114 struct llist_node *next;
1116 for (; node; node = next) {
1117 next = llist_next(node);
1118 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1121 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1123 static void mntput_no_expire(struct mount *mnt)
1126 mnt_add_count(mnt, -1);
1127 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1132 if (mnt_get_count(mnt)) {
1134 unlock_mount_hash();
1137 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1139 unlock_mount_hash();
1142 mnt->mnt.mnt_flags |= MNT_DOOMED;
1145 list_del(&mnt->mnt_instance);
1147 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1148 struct mount *p, *tmp;
1149 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1153 unlock_mount_hash();
1155 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1156 struct task_struct *task = current;
1157 if (likely(!(task->flags & PF_KTHREAD))) {
1158 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1159 if (!task_work_add(task, &mnt->mnt_rcu, true))
1162 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1163 schedule_delayed_work(&delayed_mntput_work, 1);
1169 void mntput(struct vfsmount *mnt)
1172 struct mount *m = real_mount(mnt);
1173 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1174 if (unlikely(m->mnt_expiry_mark))
1175 m->mnt_expiry_mark = 0;
1176 mntput_no_expire(m);
1179 EXPORT_SYMBOL(mntput);
1181 struct vfsmount *mntget(struct vfsmount *mnt)
1184 mnt_add_count(real_mount(mnt), 1);
1187 EXPORT_SYMBOL(mntget);
1189 struct vfsmount *mnt_clone_internal(struct path *path)
1192 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1195 p->mnt.mnt_flags |= MNT_INTERNAL;
1199 static inline void mangle(struct seq_file *m, const char *s)
1201 seq_escape(m, s, " \t\n\\");
1205 * Simple .show_options callback for filesystems which don't want to
1206 * implement more complex mount option showing.
1208 * See also save_mount_options().
1210 int generic_show_options(struct seq_file *m, struct dentry *root)
1212 const char *options;
1215 options = rcu_dereference(root->d_sb->s_options);
1217 if (options != NULL && options[0]) {
1225 EXPORT_SYMBOL(generic_show_options);
1228 * If filesystem uses generic_show_options(), this function should be
1229 * called from the fill_super() callback.
1231 * The .remount_fs callback usually needs to be handled in a special
1232 * way, to make sure, that previous options are not overwritten if the
1235 * Also note, that if the filesystem's .remount_fs function doesn't
1236 * reset all options to their default value, but changes only newly
1237 * given options, then the displayed options will not reflect reality
1240 void save_mount_options(struct super_block *sb, char *options)
1242 BUG_ON(sb->s_options);
1243 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1245 EXPORT_SYMBOL(save_mount_options);
1247 void replace_mount_options(struct super_block *sb, char *options)
1249 char *old = sb->s_options;
1250 rcu_assign_pointer(sb->s_options, options);
1256 EXPORT_SYMBOL(replace_mount_options);
1258 #ifdef CONFIG_PROC_FS
1259 /* iterator; we want it to have access to namespace_sem, thus here... */
1260 static void *m_start(struct seq_file *m, loff_t *pos)
1262 struct proc_mounts *p = m->private;
1264 down_read(&namespace_sem);
1265 if (p->cached_event == p->ns->event) {
1266 void *v = p->cached_mount;
1267 if (*pos == p->cached_index)
1269 if (*pos == p->cached_index + 1) {
1270 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1271 return p->cached_mount = v;
1275 p->cached_event = p->ns->event;
1276 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1277 p->cached_index = *pos;
1278 return p->cached_mount;
1281 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1283 struct proc_mounts *p = m->private;
1285 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1286 p->cached_index = *pos;
1287 return p->cached_mount;
1290 static void m_stop(struct seq_file *m, void *v)
1292 up_read(&namespace_sem);
1295 static int m_show(struct seq_file *m, void *v)
1297 struct proc_mounts *p = m->private;
1298 struct mount *r = list_entry(v, struct mount, mnt_list);
1299 return p->show(m, &r->mnt);
1302 const struct seq_operations mounts_op = {
1308 #endif /* CONFIG_PROC_FS */
1311 * may_umount_tree - check if a mount tree is busy
1312 * @mnt: root of mount tree
1314 * This is called to check if a tree of mounts has any
1315 * open files, pwds, chroots or sub mounts that are
1318 int may_umount_tree(struct vfsmount *m)
1320 struct mount *mnt = real_mount(m);
1321 int actual_refs = 0;
1322 int minimum_refs = 0;
1326 /* write lock needed for mnt_get_count */
1328 for (p = mnt; p; p = next_mnt(p, mnt)) {
1329 actual_refs += mnt_get_count(p);
1332 unlock_mount_hash();
1334 if (actual_refs > minimum_refs)
1340 EXPORT_SYMBOL(may_umount_tree);
1343 * may_umount - check if a mount point is busy
1344 * @mnt: root of mount
1346 * This is called to check if a mount point has any
1347 * open files, pwds, chroots or sub mounts. If the
1348 * mount has sub mounts this will return busy
1349 * regardless of whether the sub mounts are busy.
1351 * Doesn't take quota and stuff into account. IOW, in some cases it will
1352 * give false negatives. The main reason why it's here is that we need
1353 * a non-destructive way to look for easily umountable filesystems.
1355 int may_umount(struct vfsmount *mnt)
1358 down_read(&namespace_sem);
1360 if (propagate_mount_busy(real_mount(mnt), 2))
1362 unlock_mount_hash();
1363 up_read(&namespace_sem);
1367 EXPORT_SYMBOL(may_umount);
1369 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1371 static void namespace_unlock(void)
1373 struct hlist_head head;
1375 hlist_move_list(&unmounted, &head);
1377 up_write(&namespace_sem);
1379 if (likely(hlist_empty(&head)))
1384 group_pin_kill(&head);
1387 static inline void namespace_lock(void)
1389 down_write(&namespace_sem);
1392 enum umount_tree_flags {
1394 UMOUNT_PROPAGATE = 2,
1395 UMOUNT_CONNECTED = 4,
1398 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1400 /* Leaving mounts connected is only valid for lazy umounts */
1401 if (how & UMOUNT_SYNC)
1404 /* A mount without a parent has nothing to be connected to */
1405 if (!mnt_has_parent(mnt))
1408 /* Because the reference counting rules change when mounts are
1409 * unmounted and connected, umounted mounts may not be
1410 * connected to mounted mounts.
1412 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1415 /* Has it been requested that the mount remain connected? */
1416 if (how & UMOUNT_CONNECTED)
1419 /* Is the mount locked such that it needs to remain connected? */
1420 if (IS_MNT_LOCKED(mnt))
1423 /* By default disconnect the mount */
1428 * mount_lock must be held
1429 * namespace_sem must be held for write
1431 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1433 LIST_HEAD(tmp_list);
1436 if (how & UMOUNT_PROPAGATE)
1437 propagate_mount_unlock(mnt);
1439 /* Gather the mounts to umount */
1440 for (p = mnt; p; p = next_mnt(p, mnt)) {
1441 p->mnt.mnt_flags |= MNT_UMOUNT;
1442 list_move(&p->mnt_list, &tmp_list);
1445 /* Hide the mounts from mnt_mounts */
1446 list_for_each_entry(p, &tmp_list, mnt_list) {
1447 list_del_init(&p->mnt_child);
1450 /* Add propogated mounts to the tmp_list */
1451 if (how & UMOUNT_PROPAGATE)
1452 propagate_umount(&tmp_list);
1454 while (!list_empty(&tmp_list)) {
1455 struct mnt_namespace *ns;
1457 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1458 list_del_init(&p->mnt_expire);
1459 list_del_init(&p->mnt_list);
1463 __touch_mnt_namespace(ns);
1466 if (how & UMOUNT_SYNC)
1467 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1469 disconnect = disconnect_mount(p, how);
1471 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1472 disconnect ? &unmounted : NULL);
1473 if (mnt_has_parent(p)) {
1474 mnt_add_count(p->mnt_parent, -1);
1476 /* Don't forget about p */
1477 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1482 change_mnt_propagation(p, MS_PRIVATE);
1486 static void shrink_submounts(struct mount *mnt);
1488 static int do_umount(struct mount *mnt, int flags)
1490 struct super_block *sb = mnt->mnt.mnt_sb;
1493 retval = security_sb_umount(&mnt->mnt, flags);
1498 * Allow userspace to request a mountpoint be expired rather than
1499 * unmounting unconditionally. Unmount only happens if:
1500 * (1) the mark is already set (the mark is cleared by mntput())
1501 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1503 if (flags & MNT_EXPIRE) {
1504 if (&mnt->mnt == current->fs->root.mnt ||
1505 flags & (MNT_FORCE | MNT_DETACH))
1509 * probably don't strictly need the lock here if we examined
1510 * all race cases, but it's a slowpath.
1513 if (mnt_get_count(mnt) != 2) {
1514 unlock_mount_hash();
1517 unlock_mount_hash();
1519 if (!xchg(&mnt->mnt_expiry_mark, 1))
1524 * If we may have to abort operations to get out of this
1525 * mount, and they will themselves hold resources we must
1526 * allow the fs to do things. In the Unix tradition of
1527 * 'Gee thats tricky lets do it in userspace' the umount_begin
1528 * might fail to complete on the first run through as other tasks
1529 * must return, and the like. Thats for the mount program to worry
1530 * about for the moment.
1533 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1534 sb->s_op->umount_begin(sb);
1538 * No sense to grab the lock for this test, but test itself looks
1539 * somewhat bogus. Suggestions for better replacement?
1540 * Ho-hum... In principle, we might treat that as umount + switch
1541 * to rootfs. GC would eventually take care of the old vfsmount.
1542 * Actually it makes sense, especially if rootfs would contain a
1543 * /reboot - static binary that would close all descriptors and
1544 * call reboot(9). Then init(8) could umount root and exec /reboot.
1546 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1548 * Special case for "unmounting" root ...
1549 * we just try to remount it readonly.
1551 if (!capable(CAP_SYS_ADMIN))
1553 down_write(&sb->s_umount);
1554 if (!(sb->s_flags & MS_RDONLY))
1555 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1556 up_write(&sb->s_umount);
1564 if (flags & MNT_DETACH) {
1565 if (!list_empty(&mnt->mnt_list))
1566 umount_tree(mnt, UMOUNT_PROPAGATE);
1569 shrink_submounts(mnt);
1571 if (!propagate_mount_busy(mnt, 2)) {
1572 if (!list_empty(&mnt->mnt_list))
1573 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1577 unlock_mount_hash();
1583 * __detach_mounts - lazily unmount all mounts on the specified dentry
1585 * During unlink, rmdir, and d_drop it is possible to loose the path
1586 * to an existing mountpoint, and wind up leaking the mount.
1587 * detach_mounts allows lazily unmounting those mounts instead of
1590 * The caller may hold dentry->d_inode->i_mutex.
1592 void __detach_mounts(struct dentry *dentry)
1594 struct mountpoint *mp;
1599 mp = lookup_mountpoint(dentry);
1600 if (IS_ERR_OR_NULL(mp))
1604 while (!hlist_empty(&mp->m_list)) {
1605 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1606 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1607 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1610 else umount_tree(mnt, UMOUNT_CONNECTED);
1614 unlock_mount_hash();
1619 * Is the caller allowed to modify his namespace?
1621 static inline bool may_mount(void)
1623 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1627 * Now umount can handle mount points as well as block devices.
1628 * This is important for filesystems which use unnamed block devices.
1630 * We now support a flag for forced unmount like the other 'big iron'
1631 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1634 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1639 int lookup_flags = 0;
1641 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1647 if (!(flags & UMOUNT_NOFOLLOW))
1648 lookup_flags |= LOOKUP_FOLLOW;
1650 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1653 mnt = real_mount(path.mnt);
1655 if (path.dentry != path.mnt->mnt_root)
1657 if (!check_mnt(mnt))
1659 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1662 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1665 retval = do_umount(mnt, flags);
1667 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1669 mntput_no_expire(mnt);
1674 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1677 * The 2.0 compatible umount. No flags.
1679 SYSCALL_DEFINE1(oldumount, char __user *, name)
1681 return sys_umount(name, 0);
1686 static bool is_mnt_ns_file(struct dentry *dentry)
1688 /* Is this a proxy for a mount namespace? */
1689 return dentry->d_op == &ns_dentry_operations &&
1690 dentry->d_fsdata == &mntns_operations;
1693 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1695 return container_of(ns, struct mnt_namespace, ns);
1698 static bool mnt_ns_loop(struct dentry *dentry)
1700 /* Could bind mounting the mount namespace inode cause a
1701 * mount namespace loop?
1703 struct mnt_namespace *mnt_ns;
1704 if (!is_mnt_ns_file(dentry))
1707 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1708 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1711 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1714 struct mount *res, *p, *q, *r, *parent;
1716 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1717 return ERR_PTR(-EINVAL);
1719 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1720 return ERR_PTR(-EINVAL);
1722 res = q = clone_mnt(mnt, dentry, flag);
1726 q->mnt_mountpoint = mnt->mnt_mountpoint;
1729 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1731 if (!is_subdir(r->mnt_mountpoint, dentry))
1734 for (s = r; s; s = next_mnt(s, r)) {
1735 if (!(flag & CL_COPY_UNBINDABLE) &&
1736 IS_MNT_UNBINDABLE(s)) {
1737 s = skip_mnt_tree(s);
1740 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1741 is_mnt_ns_file(s->mnt.mnt_root)) {
1742 s = skip_mnt_tree(s);
1745 while (p != s->mnt_parent) {
1751 q = clone_mnt(p, p->mnt.mnt_root, flag);
1755 list_add_tail(&q->mnt_list, &res->mnt_list);
1756 attach_mnt(q, parent, p->mnt_mp);
1757 unlock_mount_hash();
1764 umount_tree(res, UMOUNT_SYNC);
1765 unlock_mount_hash();
1770 /* Caller should check returned pointer for errors */
1772 struct vfsmount *collect_mounts(struct path *path)
1776 if (!check_mnt(real_mount(path->mnt)))
1777 tree = ERR_PTR(-EINVAL);
1779 tree = copy_tree(real_mount(path->mnt), path->dentry,
1780 CL_COPY_ALL | CL_PRIVATE);
1783 return ERR_CAST(tree);
1787 void drop_collected_mounts(struct vfsmount *mnt)
1791 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1792 unlock_mount_hash();
1797 * clone_private_mount - create a private clone of a path
1799 * This creates a new vfsmount, which will be the clone of @path. The new will
1800 * not be attached anywhere in the namespace and will be private (i.e. changes
1801 * to the originating mount won't be propagated into this).
1803 * Release with mntput().
1805 struct vfsmount *clone_private_mount(struct path *path)
1807 struct mount *old_mnt = real_mount(path->mnt);
1808 struct mount *new_mnt;
1810 if (IS_MNT_UNBINDABLE(old_mnt))
1811 return ERR_PTR(-EINVAL);
1813 down_read(&namespace_sem);
1814 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1815 up_read(&namespace_sem);
1816 if (IS_ERR(new_mnt))
1817 return ERR_CAST(new_mnt);
1819 return &new_mnt->mnt;
1821 EXPORT_SYMBOL_GPL(clone_private_mount);
1823 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1824 struct vfsmount *root)
1827 int res = f(root, arg);
1830 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1831 res = f(&mnt->mnt, arg);
1838 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1842 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1843 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1844 mnt_release_group_id(p);
1848 static int invent_group_ids(struct mount *mnt, bool recurse)
1852 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1853 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1854 int err = mnt_alloc_group_id(p);
1856 cleanup_group_ids(mnt, p);
1865 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1867 unsigned int max = READ_ONCE(sysctl_mount_max);
1868 unsigned int mounts = 0, old, pending, sum;
1871 for (p = mnt; p; p = next_mnt(p, mnt))
1875 pending = ns->pending_mounts;
1876 sum = old + pending;
1880 (mounts > (max - sum)))
1883 ns->pending_mounts = pending + mounts;
1888 * @source_mnt : mount tree to be attached
1889 * @nd : place the mount tree @source_mnt is attached
1890 * @parent_nd : if non-null, detach the source_mnt from its parent and
1891 * store the parent mount and mountpoint dentry.
1892 * (done when source_mnt is moved)
1894 * NOTE: in the table below explains the semantics when a source mount
1895 * of a given type is attached to a destination mount of a given type.
1896 * ---------------------------------------------------------------------------
1897 * | BIND MOUNT OPERATION |
1898 * |**************************************************************************
1899 * | source-->| shared | private | slave | unbindable |
1903 * |**************************************************************************
1904 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1906 * |non-shared| shared (+) | private | slave (*) | invalid |
1907 * ***************************************************************************
1908 * A bind operation clones the source mount and mounts the clone on the
1909 * destination mount.
1911 * (++) the cloned mount is propagated to all the mounts in the propagation
1912 * tree of the destination mount and the cloned mount is added to
1913 * the peer group of the source mount.
1914 * (+) the cloned mount is created under the destination mount and is marked
1915 * as shared. The cloned mount is added to the peer group of the source
1917 * (+++) the mount is propagated to all the mounts in the propagation tree
1918 * of the destination mount and the cloned mount is made slave
1919 * of the same master as that of the source mount. The cloned mount
1920 * is marked as 'shared and slave'.
1921 * (*) the cloned mount is made a slave of the same master as that of the
1924 * ---------------------------------------------------------------------------
1925 * | MOVE MOUNT OPERATION |
1926 * |**************************************************************************
1927 * | source-->| shared | private | slave | unbindable |
1931 * |**************************************************************************
1932 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1934 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1935 * ***************************************************************************
1937 * (+) the mount is moved to the destination. And is then propagated to
1938 * all the mounts in the propagation tree of the destination mount.
1939 * (+*) the mount is moved to the destination.
1940 * (+++) the mount is moved to the destination and is then propagated to
1941 * all the mounts belonging to the destination mount's propagation tree.
1942 * the mount is marked as 'shared and slave'.
1943 * (*) the mount continues to be a slave at the new location.
1945 * if the source mount is a tree, the operations explained above is
1946 * applied to each mount in the tree.
1947 * Must be called without spinlocks held, since this function can sleep
1950 static int attach_recursive_mnt(struct mount *source_mnt,
1951 struct mount *dest_mnt,
1952 struct mountpoint *dest_mp,
1953 struct path *parent_path)
1955 HLIST_HEAD(tree_list);
1956 struct mnt_namespace *ns = dest_mnt->mnt_ns;
1957 struct mountpoint *smp;
1958 struct mount *child, *p;
1959 struct hlist_node *n;
1962 /* Preallocate a mountpoint in case the new mounts need
1963 * to be tucked under other mounts.
1965 smp = get_mountpoint(source_mnt->mnt.mnt_root);
1967 return PTR_ERR(smp);
1969 /* Is there space to add these mounts to the mount namespace? */
1971 err = count_mounts(ns, source_mnt);
1976 if (IS_MNT_SHARED(dest_mnt)) {
1977 err = invent_group_ids(source_mnt, true);
1980 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1983 goto out_cleanup_ids;
1984 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1990 detach_mnt(source_mnt, parent_path);
1991 attach_mnt(source_mnt, dest_mnt, dest_mp);
1992 touch_mnt_namespace(source_mnt->mnt_ns);
1994 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1995 commit_tree(source_mnt);
1998 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2000 hlist_del_init(&child->mnt_hash);
2001 q = __lookup_mnt(&child->mnt_parent->mnt,
2002 child->mnt_mountpoint);
2004 mnt_change_mountpoint(child, smp, q);
2007 put_mountpoint(smp);
2008 unlock_mount_hash();
2013 while (!hlist_empty(&tree_list)) {
2014 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2015 child->mnt_parent->mnt_ns->pending_mounts = 0;
2016 umount_tree(child, UMOUNT_SYNC);
2018 unlock_mount_hash();
2019 cleanup_group_ids(source_mnt, NULL);
2021 ns->pending_mounts = 0;
2023 read_seqlock_excl(&mount_lock);
2024 put_mountpoint(smp);
2025 read_sequnlock_excl(&mount_lock);
2030 static struct mountpoint *lock_mount(struct path *path)
2032 struct vfsmount *mnt;
2033 struct dentry *dentry = path->dentry;
2035 mutex_lock(&dentry->d_inode->i_mutex);
2036 if (unlikely(cant_mount(dentry))) {
2037 mutex_unlock(&dentry->d_inode->i_mutex);
2038 return ERR_PTR(-ENOENT);
2041 mnt = lookup_mnt(path);
2043 struct mountpoint *mp = get_mountpoint(dentry);
2046 mutex_unlock(&dentry->d_inode->i_mutex);
2052 mutex_unlock(&path->dentry->d_inode->i_mutex);
2055 dentry = path->dentry = dget(mnt->mnt_root);
2059 static void unlock_mount(struct mountpoint *where)
2061 struct dentry *dentry = where->m_dentry;
2063 read_seqlock_excl(&mount_lock);
2064 put_mountpoint(where);
2065 read_sequnlock_excl(&mount_lock);
2068 mutex_unlock(&dentry->d_inode->i_mutex);
2071 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2073 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2076 if (d_is_dir(mp->m_dentry) !=
2077 d_is_dir(mnt->mnt.mnt_root))
2080 return attach_recursive_mnt(mnt, p, mp, NULL);
2084 * Sanity check the flags to change_mnt_propagation.
2087 static int flags_to_propagation_type(int flags)
2089 int type = flags & ~(MS_REC | MS_SILENT);
2091 /* Fail if any non-propagation flags are set */
2092 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2094 /* Only one propagation flag should be set */
2095 if (!is_power_of_2(type))
2101 * recursively change the type of the mountpoint.
2103 static int do_change_type(struct path *path, int flag)
2106 struct mount *mnt = real_mount(path->mnt);
2107 int recurse = flag & MS_REC;
2111 if (path->dentry != path->mnt->mnt_root)
2114 type = flags_to_propagation_type(flag);
2119 if (type == MS_SHARED) {
2120 err = invent_group_ids(mnt, recurse);
2126 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2127 change_mnt_propagation(m, type);
2128 unlock_mount_hash();
2135 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2137 struct mount *child;
2138 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2139 if (!is_subdir(child->mnt_mountpoint, dentry))
2142 if (child->mnt.mnt_flags & MNT_LOCKED)
2149 * do loopback mount.
2151 static int do_loopback(struct path *path, const char *old_name,
2154 struct path old_path;
2155 struct mount *mnt = NULL, *old, *parent;
2156 struct mountpoint *mp;
2158 if (!old_name || !*old_name)
2160 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2165 if (mnt_ns_loop(old_path.dentry))
2168 mp = lock_mount(path);
2173 old = real_mount(old_path.mnt);
2174 parent = real_mount(path->mnt);
2177 if (IS_MNT_UNBINDABLE(old))
2180 if (!check_mnt(parent))
2183 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2186 if (!recurse && has_locked_children(old, old_path.dentry))
2190 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2192 mnt = clone_mnt(old, old_path.dentry, 0);
2199 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2201 err = graft_tree(mnt, parent, mp);
2204 umount_tree(mnt, UMOUNT_SYNC);
2205 unlock_mount_hash();
2210 path_put(&old_path);
2214 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2217 int readonly_request = 0;
2219 if (ms_flags & MS_RDONLY)
2220 readonly_request = 1;
2221 if (readonly_request == __mnt_is_readonly(mnt))
2224 if (readonly_request)
2225 error = mnt_make_readonly(real_mount(mnt));
2227 __mnt_unmake_readonly(real_mount(mnt));
2232 * change filesystem flags. dir should be a physical root of filesystem.
2233 * If you've mounted a non-root directory somewhere and want to do remount
2234 * on it - tough luck.
2236 static int do_remount(struct path *path, int flags, int mnt_flags,
2240 struct super_block *sb = path->mnt->mnt_sb;
2241 struct mount *mnt = real_mount(path->mnt);
2243 if (!check_mnt(mnt))
2246 if (path->dentry != path->mnt->mnt_root)
2249 /* Don't allow changing of locked mnt flags.
2251 * No locks need to be held here while testing the various
2252 * MNT_LOCK flags because those flags can never be cleared
2253 * once they are set.
2255 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2256 !(mnt_flags & MNT_READONLY)) {
2259 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2260 !(mnt_flags & MNT_NODEV)) {
2261 /* Was the nodev implicitly added in mount? */
2262 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2263 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2264 mnt_flags |= MNT_NODEV;
2269 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2270 !(mnt_flags & MNT_NOSUID)) {
2273 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2274 !(mnt_flags & MNT_NOEXEC)) {
2277 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2278 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2282 err = security_sb_remount(sb, data);
2286 down_write(&sb->s_umount);
2287 if (flags & MS_BIND)
2288 err = change_mount_flags(path->mnt, flags);
2289 else if (!capable(CAP_SYS_ADMIN))
2292 err = do_remount_sb(sb, flags, data, 0);
2295 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2296 mnt->mnt.mnt_flags = mnt_flags;
2297 touch_mnt_namespace(mnt->mnt_ns);
2298 unlock_mount_hash();
2300 up_write(&sb->s_umount);
2304 static inline int tree_contains_unbindable(struct mount *mnt)
2307 for (p = mnt; p; p = next_mnt(p, mnt)) {
2308 if (IS_MNT_UNBINDABLE(p))
2314 static int do_move_mount(struct path *path, const char *old_name)
2316 struct path old_path, parent_path;
2319 struct mountpoint *mp;
2321 if (!old_name || !*old_name)
2323 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2327 mp = lock_mount(path);
2332 old = real_mount(old_path.mnt);
2333 p = real_mount(path->mnt);
2336 if (!check_mnt(p) || !check_mnt(old))
2339 if (old->mnt.mnt_flags & MNT_LOCKED)
2343 if (old_path.dentry != old_path.mnt->mnt_root)
2346 if (!mnt_has_parent(old))
2349 if (d_is_dir(path->dentry) !=
2350 d_is_dir(old_path.dentry))
2353 * Don't move a mount residing in a shared parent.
2355 if (IS_MNT_SHARED(old->mnt_parent))
2358 * Don't move a mount tree containing unbindable mounts to a destination
2359 * mount which is shared.
2361 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2364 for (; mnt_has_parent(p); p = p->mnt_parent)
2368 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2372 /* if the mount is moved, it should no longer be expire
2374 list_del_init(&old->mnt_expire);
2379 path_put(&parent_path);
2380 path_put(&old_path);
2384 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2387 const char *subtype = strchr(fstype, '.');
2396 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2398 if (!mnt->mnt_sb->s_subtype)
2404 return ERR_PTR(err);
2408 * add a mount into a namespace's mount tree
2410 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2412 struct mountpoint *mp;
2413 struct mount *parent;
2416 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2418 mp = lock_mount(path);
2422 parent = real_mount(path->mnt);
2424 if (unlikely(!check_mnt(parent))) {
2425 /* that's acceptable only for automounts done in private ns */
2426 if (!(mnt_flags & MNT_SHRINKABLE))
2428 /* ... and for those we'd better have mountpoint still alive */
2429 if (!parent->mnt_ns)
2433 /* Refuse the same filesystem on the same mount point */
2435 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2436 path->mnt->mnt_root == path->dentry)
2440 if (d_is_symlink(newmnt->mnt.mnt_root))
2443 newmnt->mnt.mnt_flags = mnt_flags;
2444 err = graft_tree(newmnt, parent, mp);
2451 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2454 * create a new mount for userspace and request it to be added into the
2457 static int do_new_mount(struct path *path, const char *fstype, int flags,
2458 int mnt_flags, const char *name, void *data)
2460 struct file_system_type *type;
2461 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2462 struct vfsmount *mnt;
2468 type = get_fs_type(fstype);
2472 if (user_ns != &init_user_ns) {
2473 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2474 put_filesystem(type);
2477 /* Only in special cases allow devices from mounts
2478 * created outside the initial user namespace.
2480 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2482 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2484 if (type->fs_flags & FS_USERNS_VISIBLE) {
2485 if (!fs_fully_visible(type, &mnt_flags)) {
2486 put_filesystem(type);
2492 mnt = vfs_kern_mount(type, flags, name, data);
2493 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2494 !mnt->mnt_sb->s_subtype)
2495 mnt = fs_set_subtype(mnt, fstype);
2497 put_filesystem(type);
2499 return PTR_ERR(mnt);
2501 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2507 int finish_automount(struct vfsmount *m, struct path *path)
2509 struct mount *mnt = real_mount(m);
2511 /* The new mount record should have at least 2 refs to prevent it being
2512 * expired before we get a chance to add it
2514 BUG_ON(mnt_get_count(mnt) < 2);
2516 if (m->mnt_sb == path->mnt->mnt_sb &&
2517 m->mnt_root == path->dentry) {
2522 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2526 /* remove m from any expiration list it may be on */
2527 if (!list_empty(&mnt->mnt_expire)) {
2529 list_del_init(&mnt->mnt_expire);
2538 * mnt_set_expiry - Put a mount on an expiration list
2539 * @mnt: The mount to list.
2540 * @expiry_list: The list to add the mount to.
2542 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2546 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2550 EXPORT_SYMBOL(mnt_set_expiry);
2553 * process a list of expirable mountpoints with the intent of discarding any
2554 * mountpoints that aren't in use and haven't been touched since last we came
2557 void mark_mounts_for_expiry(struct list_head *mounts)
2559 struct mount *mnt, *next;
2560 LIST_HEAD(graveyard);
2562 if (list_empty(mounts))
2568 /* extract from the expiration list every vfsmount that matches the
2569 * following criteria:
2570 * - only referenced by its parent vfsmount
2571 * - still marked for expiry (marked on the last call here; marks are
2572 * cleared by mntput())
2574 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2575 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2576 propagate_mount_busy(mnt, 1))
2578 list_move(&mnt->mnt_expire, &graveyard);
2580 while (!list_empty(&graveyard)) {
2581 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2582 touch_mnt_namespace(mnt->mnt_ns);
2583 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2585 unlock_mount_hash();
2589 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2592 * Ripoff of 'select_parent()'
2594 * search the list of submounts for a given mountpoint, and move any
2595 * shrinkable submounts to the 'graveyard' list.
2597 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2599 struct mount *this_parent = parent;
2600 struct list_head *next;
2604 next = this_parent->mnt_mounts.next;
2606 while (next != &this_parent->mnt_mounts) {
2607 struct list_head *tmp = next;
2608 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2611 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2614 * Descend a level if the d_mounts list is non-empty.
2616 if (!list_empty(&mnt->mnt_mounts)) {
2621 if (!propagate_mount_busy(mnt, 1)) {
2622 list_move_tail(&mnt->mnt_expire, graveyard);
2627 * All done at this level ... ascend and resume the search
2629 if (this_parent != parent) {
2630 next = this_parent->mnt_child.next;
2631 this_parent = this_parent->mnt_parent;
2638 * process a list of expirable mountpoints with the intent of discarding any
2639 * submounts of a specific parent mountpoint
2641 * mount_lock must be held for write
2643 static void shrink_submounts(struct mount *mnt)
2645 LIST_HEAD(graveyard);
2648 /* extract submounts of 'mountpoint' from the expiration list */
2649 while (select_submounts(mnt, &graveyard)) {
2650 while (!list_empty(&graveyard)) {
2651 m = list_first_entry(&graveyard, struct mount,
2653 touch_mnt_namespace(m->mnt_ns);
2654 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2660 * Some copy_from_user() implementations do not return the exact number of
2661 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2662 * Note that this function differs from copy_from_user() in that it will oops
2663 * on bad values of `to', rather than returning a short copy.
2665 static long exact_copy_from_user(void *to, const void __user * from,
2669 const char __user *f = from;
2672 if (!access_ok(VERIFY_READ, from, n))
2676 if (__get_user(c, f)) {
2687 int copy_mount_options(const void __user * data, unsigned long *where)
2697 if (!(page = __get_free_page(GFP_KERNEL)))
2700 /* We only care that *some* data at the address the user
2701 * gave us is valid. Just in case, we'll zero
2702 * the remainder of the page.
2704 /* copy_from_user cannot cross TASK_SIZE ! */
2705 size = TASK_SIZE - (unsigned long)data;
2706 if (size > PAGE_SIZE)
2709 i = size - exact_copy_from_user((void *)page, data, size);
2715 memset((char *)page + i, 0, PAGE_SIZE - i);
2720 char *copy_mount_string(const void __user *data)
2722 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2726 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2727 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2729 * data is a (void *) that can point to any structure up to
2730 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2731 * information (or be NULL).
2733 * Pre-0.97 versions of mount() didn't have a flags word.
2734 * When the flags word was introduced its top half was required
2735 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2736 * Therefore, if this magic number is present, it carries no information
2737 * and must be discarded.
2739 long do_mount(const char *dev_name, const char __user *dir_name,
2740 const char *type_page, unsigned long flags, void *data_page)
2747 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2748 flags &= ~MS_MGC_MSK;
2750 /* Basic sanity checks */
2752 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2754 /* ... and get the mountpoint */
2755 retval = user_path(dir_name, &path);
2759 retval = security_sb_mount(dev_name, &path,
2760 type_page, flags, data_page);
2761 if (!retval && !may_mount())
2766 /* Default to relatime unless overriden */
2767 if (!(flags & MS_NOATIME))
2768 mnt_flags |= MNT_RELATIME;
2770 /* Separate the per-mountpoint flags */
2771 if (flags & MS_NOSUID)
2772 mnt_flags |= MNT_NOSUID;
2773 if (flags & MS_NODEV)
2774 mnt_flags |= MNT_NODEV;
2775 if (flags & MS_NOEXEC)
2776 mnt_flags |= MNT_NOEXEC;
2777 if (flags & MS_NOATIME)
2778 mnt_flags |= MNT_NOATIME;
2779 if (flags & MS_NODIRATIME)
2780 mnt_flags |= MNT_NODIRATIME;
2781 if (flags & MS_STRICTATIME)
2782 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2783 if (flags & MS_RDONLY)
2784 mnt_flags |= MNT_READONLY;
2786 /* The default atime for remount is preservation */
2787 if ((flags & MS_REMOUNT) &&
2788 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2789 MS_STRICTATIME)) == 0)) {
2790 mnt_flags &= ~MNT_ATIME_MASK;
2791 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2794 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2795 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2798 if (flags & MS_REMOUNT)
2799 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2801 else if (flags & MS_BIND)
2802 retval = do_loopback(&path, dev_name, flags & MS_REC);
2803 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2804 retval = do_change_type(&path, flags);
2805 else if (flags & MS_MOVE)
2806 retval = do_move_mount(&path, dev_name);
2808 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2809 dev_name, data_page);
2815 static void free_mnt_ns(struct mnt_namespace *ns)
2817 ns_free_inum(&ns->ns);
2818 put_user_ns(ns->user_ns);
2823 * Assign a sequence number so we can detect when we attempt to bind
2824 * mount a reference to an older mount namespace into the current
2825 * mount namespace, preventing reference counting loops. A 64bit
2826 * number incrementing at 10Ghz will take 12,427 years to wrap which
2827 * is effectively never, so we can ignore the possibility.
2829 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2831 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2833 struct mnt_namespace *new_ns;
2836 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2838 return ERR_PTR(-ENOMEM);
2839 ret = ns_alloc_inum(&new_ns->ns);
2842 return ERR_PTR(ret);
2844 new_ns->ns.ops = &mntns_operations;
2845 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2846 atomic_set(&new_ns->count, 1);
2847 new_ns->root = NULL;
2848 INIT_LIST_HEAD(&new_ns->list);
2849 init_waitqueue_head(&new_ns->poll);
2851 new_ns->user_ns = get_user_ns(user_ns);
2853 new_ns->pending_mounts = 0;
2857 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2858 struct user_namespace *user_ns, struct fs_struct *new_fs)
2860 struct mnt_namespace *new_ns;
2861 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2862 struct mount *p, *q;
2869 if (likely(!(flags & CLONE_NEWNS))) {
2876 new_ns = alloc_mnt_ns(user_ns);
2881 /* First pass: copy the tree topology */
2882 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2883 if (user_ns != ns->user_ns)
2884 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2885 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2888 free_mnt_ns(new_ns);
2889 return ERR_CAST(new);
2892 list_add_tail(&new_ns->list, &new->mnt_list);
2895 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2896 * as belonging to new namespace. We have already acquired a private
2897 * fs_struct, so tsk->fs->lock is not needed.
2905 if (&p->mnt == new_fs->root.mnt) {
2906 new_fs->root.mnt = mntget(&q->mnt);
2909 if (&p->mnt == new_fs->pwd.mnt) {
2910 new_fs->pwd.mnt = mntget(&q->mnt);
2914 p = next_mnt(p, old);
2915 q = next_mnt(q, new);
2918 while (p->mnt.mnt_root != q->mnt.mnt_root)
2919 p = next_mnt(p, old);
2932 * create_mnt_ns - creates a private namespace and adds a root filesystem
2933 * @mnt: pointer to the new root filesystem mountpoint
2935 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2937 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2938 if (!IS_ERR(new_ns)) {
2939 struct mount *mnt = real_mount(m);
2940 mnt->mnt_ns = new_ns;
2943 list_add(&mnt->mnt_list, &new_ns->list);
2950 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2952 struct mnt_namespace *ns;
2953 struct super_block *s;
2957 ns = create_mnt_ns(mnt);
2959 return ERR_CAST(ns);
2961 err = vfs_path_lookup(mnt->mnt_root, mnt,
2962 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2967 return ERR_PTR(err);
2969 /* trade a vfsmount reference for active sb one */
2970 s = path.mnt->mnt_sb;
2971 atomic_inc(&s->s_active);
2973 /* lock the sucker */
2974 down_write(&s->s_umount);
2975 /* ... and return the root of (sub)tree on it */
2978 EXPORT_SYMBOL(mount_subtree);
2980 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2981 char __user *, type, unsigned long, flags, void __user *, data)
2986 unsigned long data_page;
2988 kernel_type = copy_mount_string(type);
2989 ret = PTR_ERR(kernel_type);
2990 if (IS_ERR(kernel_type))
2993 kernel_dev = copy_mount_string(dev_name);
2994 ret = PTR_ERR(kernel_dev);
2995 if (IS_ERR(kernel_dev))
2998 ret = copy_mount_options(data, &data_page);
3002 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
3003 (void *) data_page);
3005 free_page(data_page);
3015 * Return true if path is reachable from root
3017 * namespace_sem or mount_lock is held
3019 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3020 const struct path *root)
3022 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3023 dentry = mnt->mnt_mountpoint;
3024 mnt = mnt->mnt_parent;
3026 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3029 int path_is_under(struct path *path1, struct path *path2)
3032 read_seqlock_excl(&mount_lock);
3033 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3034 read_sequnlock_excl(&mount_lock);
3037 EXPORT_SYMBOL(path_is_under);
3040 * pivot_root Semantics:
3041 * Moves the root file system of the current process to the directory put_old,
3042 * makes new_root as the new root file system of the current process, and sets
3043 * root/cwd of all processes which had them on the current root to new_root.
3046 * The new_root and put_old must be directories, and must not be on the
3047 * same file system as the current process root. The put_old must be
3048 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3049 * pointed to by put_old must yield the same directory as new_root. No other
3050 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3052 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3053 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3054 * in this situation.
3057 * - we don't move root/cwd if they are not at the root (reason: if something
3058 * cared enough to change them, it's probably wrong to force them elsewhere)
3059 * - it's okay to pick a root that isn't the root of a file system, e.g.
3060 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3061 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3064 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3065 const char __user *, put_old)
3067 struct path new, old, parent_path, root_parent, root;
3068 struct mount *new_mnt, *root_mnt, *old_mnt;
3069 struct mountpoint *old_mp, *root_mp;
3075 error = user_path_dir(new_root, &new);
3079 error = user_path_dir(put_old, &old);
3083 error = security_sb_pivotroot(&old, &new);
3087 get_fs_root(current->fs, &root);
3088 old_mp = lock_mount(&old);
3089 error = PTR_ERR(old_mp);
3094 new_mnt = real_mount(new.mnt);
3095 root_mnt = real_mount(root.mnt);
3096 old_mnt = real_mount(old.mnt);
3097 if (IS_MNT_SHARED(old_mnt) ||
3098 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3099 IS_MNT_SHARED(root_mnt->mnt_parent))
3101 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3103 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3106 if (d_unlinked(new.dentry))
3109 if (new_mnt == root_mnt || old_mnt == root_mnt)
3110 goto out4; /* loop, on the same file system */
3112 if (root.mnt->mnt_root != root.dentry)
3113 goto out4; /* not a mountpoint */
3114 if (!mnt_has_parent(root_mnt))
3115 goto out4; /* not attached */
3116 root_mp = root_mnt->mnt_mp;
3117 if (new.mnt->mnt_root != new.dentry)
3118 goto out4; /* not a mountpoint */
3119 if (!mnt_has_parent(new_mnt))
3120 goto out4; /* not attached */
3121 /* make sure we can reach put_old from new_root */
3122 if (!is_path_reachable(old_mnt, old.dentry, &new))
3124 /* make certain new is below the root */
3125 if (!is_path_reachable(new_mnt, new.dentry, &root))
3127 root_mp->m_count++; /* pin it so it won't go away */
3129 detach_mnt(new_mnt, &parent_path);
3130 detach_mnt(root_mnt, &root_parent);
3131 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3132 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3133 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3135 /* mount old root on put_old */
3136 attach_mnt(root_mnt, old_mnt, old_mp);
3137 /* mount new_root on / */
3138 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3139 touch_mnt_namespace(current->nsproxy->mnt_ns);
3140 /* A moved mount should not expire automatically */
3141 list_del_init(&new_mnt->mnt_expire);
3142 put_mountpoint(root_mp);
3143 unlock_mount_hash();
3144 chroot_fs_refs(&root, &new);
3147 unlock_mount(old_mp);
3149 path_put(&root_parent);
3150 path_put(&parent_path);
3162 static void __init init_mount_tree(void)
3164 struct vfsmount *mnt;
3165 struct mnt_namespace *ns;
3167 struct file_system_type *type;
3169 type = get_fs_type("rootfs");
3171 panic("Can't find rootfs type");
3172 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3173 put_filesystem(type);
3175 panic("Can't create rootfs");
3177 ns = create_mnt_ns(mnt);
3179 panic("Can't allocate initial namespace");
3181 init_task.nsproxy->mnt_ns = ns;
3185 root.dentry = mnt->mnt_root;
3186 mnt->mnt_flags |= MNT_LOCKED;
3188 set_fs_pwd(current->fs, &root);
3189 set_fs_root(current->fs, &root);
3192 void __init mnt_init(void)
3197 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3198 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3200 mount_hashtable = alloc_large_system_hash("Mount-cache",
3201 sizeof(struct hlist_head),
3204 &m_hash_shift, &m_hash_mask, 0, 0);
3205 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3206 sizeof(struct hlist_head),
3209 &mp_hash_shift, &mp_hash_mask, 0, 0);
3211 if (!mount_hashtable || !mountpoint_hashtable)
3212 panic("Failed to allocate mount hash table\n");
3214 for (u = 0; u <= m_hash_mask; u++)
3215 INIT_HLIST_HEAD(&mount_hashtable[u]);
3216 for (u = 0; u <= mp_hash_mask; u++)
3217 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3223 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3225 fs_kobj = kobject_create_and_add("fs", NULL);
3227 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3232 void put_mnt_ns(struct mnt_namespace *ns)
3234 if (!atomic_dec_and_test(&ns->count))
3236 drop_collected_mounts(&ns->root->mnt);
3240 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3242 struct vfsmount *mnt;
3243 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3246 * it is a longterm mount, don't release mnt until
3247 * we unmount before file sys is unregistered
3249 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3253 EXPORT_SYMBOL_GPL(kern_mount_data);
3255 void kern_unmount(struct vfsmount *mnt)
3257 /* release long term mount so mount point can be released */
3258 if (!IS_ERR_OR_NULL(mnt)) {
3259 real_mount(mnt)->mnt_ns = NULL;
3260 synchronize_rcu(); /* yecchhh... */
3264 EXPORT_SYMBOL(kern_unmount);
3266 bool our_mnt(struct vfsmount *mnt)
3268 return check_mnt(real_mount(mnt));
3271 bool current_chrooted(void)
3273 /* Does the current process have a non-standard root */
3274 struct path ns_root;
3275 struct path fs_root;
3278 /* Find the namespace root */
3279 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3280 ns_root.dentry = ns_root.mnt->mnt_root;
3282 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3285 get_fs_root(current->fs, &fs_root);
3287 chrooted = !path_equal(&fs_root, &ns_root);
3295 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3297 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3298 int new_flags = *new_mnt_flags;
3300 bool visible = false;
3305 down_read(&namespace_sem);
3306 list_for_each_entry(mnt, &ns->list, mnt_list) {
3307 struct mount *child;
3310 if (mnt->mnt.mnt_sb->s_type != type)
3313 /* This mount is not fully visible if it's root directory
3314 * is not the root directory of the filesystem.
3316 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3319 /* Read the mount flags and filter out flags that
3320 * may safely be ignored.
3322 mnt_flags = mnt->mnt.mnt_flags;
3323 if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3324 mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3326 /* Don't miss readonly hidden in the superblock flags */
3327 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3328 mnt_flags |= MNT_LOCK_READONLY;
3330 /* Verify the mount flags are equal to or more permissive
3331 * than the proposed new mount.
3333 if ((mnt_flags & MNT_LOCK_READONLY) &&
3334 !(new_flags & MNT_READONLY))
3336 if ((mnt_flags & MNT_LOCK_NODEV) &&
3337 !(new_flags & MNT_NODEV))
3339 if ((mnt_flags & MNT_LOCK_NOSUID) &&
3340 !(new_flags & MNT_NOSUID))
3342 if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3343 !(new_flags & MNT_NOEXEC))
3345 if ((mnt_flags & MNT_LOCK_ATIME) &&
3346 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3349 /* This mount is not fully visible if there are any
3350 * locked child mounts that cover anything except for
3351 * empty directories.
3353 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3354 struct inode *inode = child->mnt_mountpoint->d_inode;
3355 /* Only worry about locked mounts */
3356 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3358 /* Is the directory permanetly empty? */
3359 if (!is_empty_dir_inode(inode))
3362 /* Preserve the locked attributes */
3363 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3373 up_read(&namespace_sem);
3377 static struct ns_common *mntns_get(struct task_struct *task)
3379 struct ns_common *ns = NULL;
3380 struct nsproxy *nsproxy;
3383 nsproxy = task->nsproxy;
3385 ns = &nsproxy->mnt_ns->ns;
3386 get_mnt_ns(to_mnt_ns(ns));
3393 static void mntns_put(struct ns_common *ns)
3395 put_mnt_ns(to_mnt_ns(ns));
3398 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3400 struct fs_struct *fs = current->fs;
3401 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3404 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3405 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3406 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3413 put_mnt_ns(nsproxy->mnt_ns);
3414 nsproxy->mnt_ns = mnt_ns;
3417 root.mnt = &mnt_ns->root->mnt;
3418 root.dentry = mnt_ns->root->mnt.mnt_root;
3420 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3423 /* Update the pwd and root */
3424 set_fs_pwd(fs, &root);
3425 set_fs_root(fs, &root);
3431 const struct proc_ns_operations mntns_operations = {
3433 .type = CLONE_NEWNS,
3436 .install = mntns_install,