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 smp_mb(); // see mntput_no_expire()
607 if (likely(!read_seqretry(&mount_lock, seq)))
609 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
610 mnt_add_count(mnt, -1);
614 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
615 mnt_add_count(mnt, -1);
620 /* caller will mntput() */
624 /* call under rcu_read_lock */
625 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
627 int res = __legitimize_mnt(bastard, seq);
630 if (unlikely(res < 0)) {
639 * find the first mount at @dentry on vfsmount @mnt.
640 * call under rcu_read_lock()
642 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
644 struct hlist_head *head = m_hash(mnt, dentry);
647 hlist_for_each_entry_rcu(p, head, mnt_hash)
648 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
654 * lookup_mnt - Return the first child mount mounted at path
656 * "First" means first mounted chronologically. If you create the
659 * mount /dev/sda1 /mnt
660 * mount /dev/sda2 /mnt
661 * mount /dev/sda3 /mnt
663 * Then lookup_mnt() on the base /mnt dentry in the root mount will
664 * return successively the root dentry and vfsmount of /dev/sda1, then
665 * /dev/sda2, then /dev/sda3, then NULL.
667 * lookup_mnt takes a reference to the found vfsmount.
669 struct vfsmount *lookup_mnt(struct path *path)
671 struct mount *child_mnt;
677 seq = read_seqbegin(&mount_lock);
678 child_mnt = __lookup_mnt(path->mnt, path->dentry);
679 m = child_mnt ? &child_mnt->mnt : NULL;
680 } while (!legitimize_mnt(m, seq));
686 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
687 * current mount namespace.
689 * The common case is dentries are not mountpoints at all and that
690 * test is handled inline. For the slow case when we are actually
691 * dealing with a mountpoint of some kind, walk through all of the
692 * mounts in the current mount namespace and test to see if the dentry
695 * The mount_hashtable is not usable in the context because we
696 * need to identify all mounts that may be in the current mount
697 * namespace not just a mount that happens to have some specified
700 bool __is_local_mountpoint(struct dentry *dentry)
702 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
704 bool is_covered = false;
706 if (!d_mountpoint(dentry))
709 down_read(&namespace_sem);
710 list_for_each_entry(mnt, &ns->list, mnt_list) {
711 is_covered = (mnt->mnt_mountpoint == dentry);
715 up_read(&namespace_sem);
720 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
722 struct hlist_head *chain = mp_hash(dentry);
723 struct mountpoint *mp;
725 hlist_for_each_entry(mp, chain, m_hash) {
726 if (mp->m_dentry == dentry) {
727 /* might be worth a WARN_ON() */
728 if (d_unlinked(dentry))
729 return ERR_PTR(-ENOENT);
737 static struct mountpoint *get_mountpoint(struct dentry *dentry)
739 struct mountpoint *mp, *new = NULL;
742 if (d_mountpoint(dentry)) {
744 read_seqlock_excl(&mount_lock);
745 mp = lookup_mountpoint(dentry);
746 read_sequnlock_excl(&mount_lock);
752 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
754 return ERR_PTR(-ENOMEM);
757 /* Exactly one processes may set d_mounted */
758 ret = d_set_mounted(dentry);
760 /* Someone else set d_mounted? */
764 /* The dentry is not available as a mountpoint? */
769 /* Add the new mountpoint to the hash table */
770 read_seqlock_excl(&mount_lock);
771 new->m_dentry = dentry;
773 hlist_add_head(&new->m_hash, mp_hash(dentry));
774 INIT_HLIST_HEAD(&new->m_list);
775 read_sequnlock_excl(&mount_lock);
784 static void put_mountpoint(struct mountpoint *mp)
786 if (!--mp->m_count) {
787 struct dentry *dentry = mp->m_dentry;
788 BUG_ON(!hlist_empty(&mp->m_list));
789 spin_lock(&dentry->d_lock);
790 dentry->d_flags &= ~DCACHE_MOUNTED;
791 spin_unlock(&dentry->d_lock);
792 hlist_del(&mp->m_hash);
797 static inline int check_mnt(struct mount *mnt)
799 return mnt->mnt_ns == current->nsproxy->mnt_ns;
803 * vfsmount lock must be held for write
805 static void touch_mnt_namespace(struct mnt_namespace *ns)
809 wake_up_interruptible(&ns->poll);
814 * vfsmount lock must be held for write
816 static void __touch_mnt_namespace(struct mnt_namespace *ns)
818 if (ns && ns->event != event) {
820 wake_up_interruptible(&ns->poll);
825 * vfsmount lock must be held for write
827 static void unhash_mnt(struct mount *mnt)
829 mnt->mnt_parent = mnt;
830 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
831 list_del_init(&mnt->mnt_child);
832 hlist_del_init_rcu(&mnt->mnt_hash);
833 hlist_del_init(&mnt->mnt_mp_list);
834 put_mountpoint(mnt->mnt_mp);
839 * vfsmount lock must be held for write
841 static void detach_mnt(struct mount *mnt, struct path *old_path)
843 old_path->dentry = mnt->mnt_mountpoint;
844 old_path->mnt = &mnt->mnt_parent->mnt;
849 * vfsmount lock must be held for write
851 static void umount_mnt(struct mount *mnt)
853 /* old mountpoint will be dropped when we can do that */
854 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
859 * vfsmount lock must be held for write
861 void mnt_set_mountpoint(struct mount *mnt,
862 struct mountpoint *mp,
863 struct mount *child_mnt)
866 mnt_add_count(mnt, 1); /* essentially, that's mntget */
867 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
868 child_mnt->mnt_parent = mnt;
869 child_mnt->mnt_mp = mp;
870 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
873 static void __attach_mnt(struct mount *mnt, struct mount *parent)
875 hlist_add_head_rcu(&mnt->mnt_hash,
876 m_hash(&parent->mnt, mnt->mnt_mountpoint));
877 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
881 * vfsmount lock must be held for write
883 static void attach_mnt(struct mount *mnt,
884 struct mount *parent,
885 struct mountpoint *mp)
887 mnt_set_mountpoint(parent, mp, mnt);
888 __attach_mnt(mnt, parent);
891 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
893 struct mountpoint *old_mp = mnt->mnt_mp;
894 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
895 struct mount *old_parent = mnt->mnt_parent;
897 list_del_init(&mnt->mnt_child);
898 hlist_del_init(&mnt->mnt_mp_list);
899 hlist_del_init_rcu(&mnt->mnt_hash);
901 attach_mnt(mnt, parent, mp);
903 put_mountpoint(old_mp);
906 * Safely avoid even the suggestion this code might sleep or
907 * lock the mount hash by taking advantage of the knowledge that
908 * mnt_change_mountpoint will not release the final reference
911 * During mounting, the mount passed in as the parent mount will
912 * continue to use the old mountpoint and during unmounting, the
913 * old mountpoint will continue to exist until namespace_unlock,
914 * which happens well after mnt_change_mountpoint.
916 spin_lock(&old_mountpoint->d_lock);
917 old_mountpoint->d_lockref.count--;
918 spin_unlock(&old_mountpoint->d_lock);
920 mnt_add_count(old_parent, -1);
924 * vfsmount lock must be held for write
926 static void commit_tree(struct mount *mnt)
928 struct mount *parent = mnt->mnt_parent;
931 struct mnt_namespace *n = parent->mnt_ns;
933 BUG_ON(parent == mnt);
935 list_add_tail(&head, &mnt->mnt_list);
936 list_for_each_entry(m, &head, mnt_list)
939 list_splice(&head, n->list.prev);
941 n->mounts += n->pending_mounts;
942 n->pending_mounts = 0;
944 __attach_mnt(mnt, parent);
945 touch_mnt_namespace(n);
948 static struct mount *next_mnt(struct mount *p, struct mount *root)
950 struct list_head *next = p->mnt_mounts.next;
951 if (next == &p->mnt_mounts) {
955 next = p->mnt_child.next;
956 if (next != &p->mnt_parent->mnt_mounts)
961 return list_entry(next, struct mount, mnt_child);
964 static struct mount *skip_mnt_tree(struct mount *p)
966 struct list_head *prev = p->mnt_mounts.prev;
967 while (prev != &p->mnt_mounts) {
968 p = list_entry(prev, struct mount, mnt_child);
969 prev = p->mnt_mounts.prev;
975 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
981 return ERR_PTR(-ENODEV);
983 mnt = alloc_vfsmnt(name);
985 return ERR_PTR(-ENOMEM);
987 if (flags & MS_KERNMOUNT)
988 mnt->mnt.mnt_flags = MNT_INTERNAL;
990 root = mount_fs(type, flags, name, data);
994 return ERR_CAST(root);
997 mnt->mnt.mnt_root = root;
998 mnt->mnt.mnt_sb = root->d_sb;
999 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1000 mnt->mnt_parent = mnt;
1002 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1003 unlock_mount_hash();
1006 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1008 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1011 struct super_block *sb = old->mnt.mnt_sb;
1015 mnt = alloc_vfsmnt(old->mnt_devname);
1017 return ERR_PTR(-ENOMEM);
1019 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1020 mnt->mnt_group_id = 0; /* not a peer of original */
1022 mnt->mnt_group_id = old->mnt_group_id;
1024 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1025 err = mnt_alloc_group_id(mnt);
1030 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1031 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1032 /* Don't allow unprivileged users to change mount flags */
1033 if (flag & CL_UNPRIVILEGED) {
1034 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1036 if (mnt->mnt.mnt_flags & MNT_READONLY)
1037 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1039 if (mnt->mnt.mnt_flags & MNT_NODEV)
1040 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1042 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1043 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1045 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1046 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1049 /* Don't allow unprivileged users to reveal what is under a mount */
1050 if ((flag & CL_UNPRIVILEGED) &&
1051 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1052 mnt->mnt.mnt_flags |= MNT_LOCKED;
1054 atomic_inc(&sb->s_active);
1055 mnt->mnt.mnt_sb = sb;
1056 mnt->mnt.mnt_root = dget(root);
1057 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1058 mnt->mnt_parent = mnt;
1060 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1061 unlock_mount_hash();
1063 if ((flag & CL_SLAVE) ||
1064 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1065 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1066 mnt->mnt_master = old;
1067 CLEAR_MNT_SHARED(mnt);
1068 } else if (!(flag & CL_PRIVATE)) {
1069 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1070 list_add(&mnt->mnt_share, &old->mnt_share);
1071 if (IS_MNT_SLAVE(old))
1072 list_add(&mnt->mnt_slave, &old->mnt_slave);
1073 mnt->mnt_master = old->mnt_master;
1075 if (flag & CL_MAKE_SHARED)
1076 set_mnt_shared(mnt);
1078 /* stick the duplicate mount on the same expiry list
1079 * as the original if that was on one */
1080 if (flag & CL_EXPIRE) {
1081 if (!list_empty(&old->mnt_expire))
1082 list_add(&mnt->mnt_expire, &old->mnt_expire);
1090 return ERR_PTR(err);
1093 static void cleanup_mnt(struct mount *mnt)
1096 * This probably indicates that somebody messed
1097 * up a mnt_want/drop_write() pair. If this
1098 * happens, the filesystem was probably unable
1099 * to make r/w->r/o transitions.
1102 * The locking used to deal with mnt_count decrement provides barriers,
1103 * so mnt_get_writers() below is safe.
1105 WARN_ON(mnt_get_writers(mnt));
1106 if (unlikely(mnt->mnt_pins.first))
1108 fsnotify_vfsmount_delete(&mnt->mnt);
1109 dput(mnt->mnt.mnt_root);
1110 deactivate_super(mnt->mnt.mnt_sb);
1112 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1115 static void __cleanup_mnt(struct rcu_head *head)
1117 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1120 static LLIST_HEAD(delayed_mntput_list);
1121 static void delayed_mntput(struct work_struct *unused)
1123 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1124 struct llist_node *next;
1126 for (; node; node = next) {
1127 next = llist_next(node);
1128 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1131 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1133 static void mntput_no_expire(struct mount *mnt)
1136 if (likely(READ_ONCE(mnt->mnt_ns))) {
1138 * Since we don't do lock_mount_hash() here,
1139 * ->mnt_ns can change under us. However, if it's
1140 * non-NULL, then there's a reference that won't
1141 * be dropped until after an RCU delay done after
1142 * turning ->mnt_ns NULL. So if we observe it
1143 * non-NULL under rcu_read_lock(), the reference
1144 * we are dropping is not the final one.
1146 mnt_add_count(mnt, -1);
1152 * make sure that if __legitimize_mnt() has not seen us grab
1153 * mount_lock, we'll see their refcount increment here.
1156 mnt_add_count(mnt, -1);
1157 if (mnt_get_count(mnt)) {
1159 unlock_mount_hash();
1162 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1164 unlock_mount_hash();
1167 mnt->mnt.mnt_flags |= MNT_DOOMED;
1170 list_del(&mnt->mnt_instance);
1172 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1173 struct mount *p, *tmp;
1174 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1178 unlock_mount_hash();
1180 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1181 struct task_struct *task = current;
1182 if (likely(!(task->flags & PF_KTHREAD))) {
1183 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1184 if (!task_work_add(task, &mnt->mnt_rcu, true))
1187 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1188 schedule_delayed_work(&delayed_mntput_work, 1);
1194 void mntput(struct vfsmount *mnt)
1197 struct mount *m = real_mount(mnt);
1198 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1199 if (unlikely(m->mnt_expiry_mark))
1200 m->mnt_expiry_mark = 0;
1201 mntput_no_expire(m);
1204 EXPORT_SYMBOL(mntput);
1206 struct vfsmount *mntget(struct vfsmount *mnt)
1209 mnt_add_count(real_mount(mnt), 1);
1212 EXPORT_SYMBOL(mntget);
1214 struct vfsmount *mnt_clone_internal(struct path *path)
1217 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1220 p->mnt.mnt_flags |= MNT_INTERNAL;
1224 static inline void mangle(struct seq_file *m, const char *s)
1226 seq_escape(m, s, " \t\n\\");
1230 * Simple .show_options callback for filesystems which don't want to
1231 * implement more complex mount option showing.
1233 * See also save_mount_options().
1235 int generic_show_options(struct seq_file *m, struct dentry *root)
1237 const char *options;
1240 options = rcu_dereference(root->d_sb->s_options);
1242 if (options != NULL && options[0]) {
1250 EXPORT_SYMBOL(generic_show_options);
1253 * If filesystem uses generic_show_options(), this function should be
1254 * called from the fill_super() callback.
1256 * The .remount_fs callback usually needs to be handled in a special
1257 * way, to make sure, that previous options are not overwritten if the
1260 * Also note, that if the filesystem's .remount_fs function doesn't
1261 * reset all options to their default value, but changes only newly
1262 * given options, then the displayed options will not reflect reality
1265 void save_mount_options(struct super_block *sb, char *options)
1267 BUG_ON(sb->s_options);
1268 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1270 EXPORT_SYMBOL(save_mount_options);
1272 void replace_mount_options(struct super_block *sb, char *options)
1274 char *old = sb->s_options;
1275 rcu_assign_pointer(sb->s_options, options);
1281 EXPORT_SYMBOL(replace_mount_options);
1283 #ifdef CONFIG_PROC_FS
1284 /* iterator; we want it to have access to namespace_sem, thus here... */
1285 static void *m_start(struct seq_file *m, loff_t *pos)
1287 struct proc_mounts *p = m->private;
1289 down_read(&namespace_sem);
1290 if (p->cached_event == p->ns->event) {
1291 void *v = p->cached_mount;
1292 if (*pos == p->cached_index)
1294 if (*pos == p->cached_index + 1) {
1295 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1296 return p->cached_mount = v;
1300 p->cached_event = p->ns->event;
1301 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1302 p->cached_index = *pos;
1303 return p->cached_mount;
1306 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1308 struct proc_mounts *p = m->private;
1310 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1311 p->cached_index = *pos;
1312 return p->cached_mount;
1315 static void m_stop(struct seq_file *m, void *v)
1317 up_read(&namespace_sem);
1320 static int m_show(struct seq_file *m, void *v)
1322 struct proc_mounts *p = m->private;
1323 struct mount *r = list_entry(v, struct mount, mnt_list);
1324 return p->show(m, &r->mnt);
1327 const struct seq_operations mounts_op = {
1333 #endif /* CONFIG_PROC_FS */
1336 * may_umount_tree - check if a mount tree is busy
1337 * @mnt: root of mount tree
1339 * This is called to check if a tree of mounts has any
1340 * open files, pwds, chroots or sub mounts that are
1343 int may_umount_tree(struct vfsmount *m)
1345 struct mount *mnt = real_mount(m);
1346 int actual_refs = 0;
1347 int minimum_refs = 0;
1351 /* write lock needed for mnt_get_count */
1353 for (p = mnt; p; p = next_mnt(p, mnt)) {
1354 actual_refs += mnt_get_count(p);
1357 unlock_mount_hash();
1359 if (actual_refs > minimum_refs)
1365 EXPORT_SYMBOL(may_umount_tree);
1368 * may_umount - check if a mount point is busy
1369 * @mnt: root of mount
1371 * This is called to check if a mount point has any
1372 * open files, pwds, chroots or sub mounts. If the
1373 * mount has sub mounts this will return busy
1374 * regardless of whether the sub mounts are busy.
1376 * Doesn't take quota and stuff into account. IOW, in some cases it will
1377 * give false negatives. The main reason why it's here is that we need
1378 * a non-destructive way to look for easily umountable filesystems.
1380 int may_umount(struct vfsmount *mnt)
1383 down_read(&namespace_sem);
1385 if (propagate_mount_busy(real_mount(mnt), 2))
1387 unlock_mount_hash();
1388 up_read(&namespace_sem);
1392 EXPORT_SYMBOL(may_umount);
1394 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1396 static void namespace_unlock(void)
1398 struct hlist_head head;
1400 hlist_move_list(&unmounted, &head);
1402 up_write(&namespace_sem);
1404 if (likely(hlist_empty(&head)))
1409 group_pin_kill(&head);
1412 static inline void namespace_lock(void)
1414 down_write(&namespace_sem);
1417 enum umount_tree_flags {
1419 UMOUNT_PROPAGATE = 2,
1420 UMOUNT_CONNECTED = 4,
1423 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1425 /* Leaving mounts connected is only valid for lazy umounts */
1426 if (how & UMOUNT_SYNC)
1429 /* A mount without a parent has nothing to be connected to */
1430 if (!mnt_has_parent(mnt))
1433 /* Because the reference counting rules change when mounts are
1434 * unmounted and connected, umounted mounts may not be
1435 * connected to mounted mounts.
1437 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1440 /* Has it been requested that the mount remain connected? */
1441 if (how & UMOUNT_CONNECTED)
1444 /* Is the mount locked such that it needs to remain connected? */
1445 if (IS_MNT_LOCKED(mnt))
1448 /* By default disconnect the mount */
1453 * mount_lock must be held
1454 * namespace_sem must be held for write
1456 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1458 LIST_HEAD(tmp_list);
1461 if (how & UMOUNT_PROPAGATE)
1462 propagate_mount_unlock(mnt);
1464 /* Gather the mounts to umount */
1465 for (p = mnt; p; p = next_mnt(p, mnt)) {
1466 p->mnt.mnt_flags |= MNT_UMOUNT;
1467 list_move(&p->mnt_list, &tmp_list);
1470 /* Hide the mounts from mnt_mounts */
1471 list_for_each_entry(p, &tmp_list, mnt_list) {
1472 list_del_init(&p->mnt_child);
1475 /* Add propogated mounts to the tmp_list */
1476 if (how & UMOUNT_PROPAGATE)
1477 propagate_umount(&tmp_list);
1479 while (!list_empty(&tmp_list)) {
1480 struct mnt_namespace *ns;
1482 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1483 list_del_init(&p->mnt_expire);
1484 list_del_init(&p->mnt_list);
1488 __touch_mnt_namespace(ns);
1491 if (how & UMOUNT_SYNC)
1492 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1494 disconnect = disconnect_mount(p, how);
1496 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1497 disconnect ? &unmounted : NULL);
1498 if (mnt_has_parent(p)) {
1499 mnt_add_count(p->mnt_parent, -1);
1501 /* Don't forget about p */
1502 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1507 change_mnt_propagation(p, MS_PRIVATE);
1511 static void shrink_submounts(struct mount *mnt);
1513 static int do_umount(struct mount *mnt, int flags)
1515 struct super_block *sb = mnt->mnt.mnt_sb;
1518 retval = security_sb_umount(&mnt->mnt, flags);
1523 * Allow userspace to request a mountpoint be expired rather than
1524 * unmounting unconditionally. Unmount only happens if:
1525 * (1) the mark is already set (the mark is cleared by mntput())
1526 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1528 if (flags & MNT_EXPIRE) {
1529 if (&mnt->mnt == current->fs->root.mnt ||
1530 flags & (MNT_FORCE | MNT_DETACH))
1534 * probably don't strictly need the lock here if we examined
1535 * all race cases, but it's a slowpath.
1538 if (mnt_get_count(mnt) != 2) {
1539 unlock_mount_hash();
1542 unlock_mount_hash();
1544 if (!xchg(&mnt->mnt_expiry_mark, 1))
1549 * If we may have to abort operations to get out of this
1550 * mount, and they will themselves hold resources we must
1551 * allow the fs to do things. In the Unix tradition of
1552 * 'Gee thats tricky lets do it in userspace' the umount_begin
1553 * might fail to complete on the first run through as other tasks
1554 * must return, and the like. Thats for the mount program to worry
1555 * about for the moment.
1558 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1559 sb->s_op->umount_begin(sb);
1563 * No sense to grab the lock for this test, but test itself looks
1564 * somewhat bogus. Suggestions for better replacement?
1565 * Ho-hum... In principle, we might treat that as umount + switch
1566 * to rootfs. GC would eventually take care of the old vfsmount.
1567 * Actually it makes sense, especially if rootfs would contain a
1568 * /reboot - static binary that would close all descriptors and
1569 * call reboot(9). Then init(8) could umount root and exec /reboot.
1571 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1573 * Special case for "unmounting" root ...
1574 * we just try to remount it readonly.
1576 if (!capable(CAP_SYS_ADMIN))
1578 down_write(&sb->s_umount);
1579 if (!(sb->s_flags & MS_RDONLY))
1580 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1581 up_write(&sb->s_umount);
1588 /* Recheck MNT_LOCKED with the locks held */
1590 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1594 if (flags & MNT_DETACH) {
1595 if (!list_empty(&mnt->mnt_list))
1596 umount_tree(mnt, UMOUNT_PROPAGATE);
1599 shrink_submounts(mnt);
1601 if (!propagate_mount_busy(mnt, 2)) {
1602 if (!list_empty(&mnt->mnt_list))
1603 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1608 unlock_mount_hash();
1614 * __detach_mounts - lazily unmount all mounts on the specified dentry
1616 * During unlink, rmdir, and d_drop it is possible to loose the path
1617 * to an existing mountpoint, and wind up leaking the mount.
1618 * detach_mounts allows lazily unmounting those mounts instead of
1621 * The caller may hold dentry->d_inode->i_mutex.
1623 void __detach_mounts(struct dentry *dentry)
1625 struct mountpoint *mp;
1630 mp = lookup_mountpoint(dentry);
1631 if (IS_ERR_OR_NULL(mp))
1635 while (!hlist_empty(&mp->m_list)) {
1636 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1637 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1638 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1641 else umount_tree(mnt, UMOUNT_CONNECTED);
1645 unlock_mount_hash();
1650 * Is the caller allowed to modify his namespace?
1652 static inline bool may_mount(void)
1654 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1658 * Now umount can handle mount points as well as block devices.
1659 * This is important for filesystems which use unnamed block devices.
1661 * We now support a flag for forced unmount like the other 'big iron'
1662 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1665 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1670 int lookup_flags = 0;
1672 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1678 if (!(flags & UMOUNT_NOFOLLOW))
1679 lookup_flags |= LOOKUP_FOLLOW;
1681 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1684 mnt = real_mount(path.mnt);
1686 if (path.dentry != path.mnt->mnt_root)
1688 if (!check_mnt(mnt))
1690 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1693 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1696 retval = do_umount(mnt, flags);
1698 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1700 mntput_no_expire(mnt);
1705 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1708 * The 2.0 compatible umount. No flags.
1710 SYSCALL_DEFINE1(oldumount, char __user *, name)
1712 return sys_umount(name, 0);
1717 static bool is_mnt_ns_file(struct dentry *dentry)
1719 /* Is this a proxy for a mount namespace? */
1720 return dentry->d_op == &ns_dentry_operations &&
1721 dentry->d_fsdata == &mntns_operations;
1724 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1726 return container_of(ns, struct mnt_namespace, ns);
1729 static bool mnt_ns_loop(struct dentry *dentry)
1731 /* Could bind mounting the mount namespace inode cause a
1732 * mount namespace loop?
1734 struct mnt_namespace *mnt_ns;
1735 if (!is_mnt_ns_file(dentry))
1738 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1739 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1742 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1745 struct mount *res, *p, *q, *r, *parent;
1747 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1748 return ERR_PTR(-EINVAL);
1750 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1751 return ERR_PTR(-EINVAL);
1753 res = q = clone_mnt(mnt, dentry, flag);
1757 q->mnt_mountpoint = mnt->mnt_mountpoint;
1760 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1762 if (!is_subdir(r->mnt_mountpoint, dentry))
1765 for (s = r; s; s = next_mnt(s, r)) {
1766 if (!(flag & CL_COPY_UNBINDABLE) &&
1767 IS_MNT_UNBINDABLE(s)) {
1768 if (s->mnt.mnt_flags & MNT_LOCKED) {
1769 /* Both unbindable and locked. */
1770 q = ERR_PTR(-EPERM);
1773 s = skip_mnt_tree(s);
1777 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1778 is_mnt_ns_file(s->mnt.mnt_root)) {
1779 s = skip_mnt_tree(s);
1782 while (p != s->mnt_parent) {
1788 q = clone_mnt(p, p->mnt.mnt_root, flag);
1792 list_add_tail(&q->mnt_list, &res->mnt_list);
1793 attach_mnt(q, parent, p->mnt_mp);
1794 unlock_mount_hash();
1801 umount_tree(res, UMOUNT_SYNC);
1802 unlock_mount_hash();
1807 /* Caller should check returned pointer for errors */
1809 struct vfsmount *collect_mounts(struct path *path)
1813 if (!check_mnt(real_mount(path->mnt)))
1814 tree = ERR_PTR(-EINVAL);
1816 tree = copy_tree(real_mount(path->mnt), path->dentry,
1817 CL_COPY_ALL | CL_PRIVATE);
1820 return ERR_CAST(tree);
1824 void drop_collected_mounts(struct vfsmount *mnt)
1828 umount_tree(real_mount(mnt), 0);
1829 unlock_mount_hash();
1833 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1835 struct mount *child;
1837 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1838 if (!is_subdir(child->mnt_mountpoint, dentry))
1841 if (child->mnt.mnt_flags & MNT_LOCKED)
1848 * clone_private_mount - create a private clone of a path
1850 * This creates a new vfsmount, which will be the clone of @path. The new will
1851 * not be attached anywhere in the namespace and will be private (i.e. changes
1852 * to the originating mount won't be propagated into this).
1854 * Release with mntput().
1856 struct vfsmount *clone_private_mount(struct path *path)
1858 struct mount *old_mnt = real_mount(path->mnt);
1859 struct mount *new_mnt;
1861 down_read(&namespace_sem);
1862 if (IS_MNT_UNBINDABLE(old_mnt))
1865 if (!check_mnt(old_mnt))
1868 if (has_locked_children(old_mnt, path->dentry))
1871 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1872 up_read(&namespace_sem);
1874 if (IS_ERR(new_mnt))
1875 return ERR_CAST(new_mnt);
1877 return &new_mnt->mnt;
1880 up_read(&namespace_sem);
1881 return ERR_PTR(-EINVAL);
1883 EXPORT_SYMBOL_GPL(clone_private_mount);
1885 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1886 struct vfsmount *root)
1889 int res = f(root, arg);
1892 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1893 res = f(&mnt->mnt, arg);
1900 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1904 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1905 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1906 mnt_release_group_id(p);
1910 static int invent_group_ids(struct mount *mnt, bool recurse)
1914 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1915 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1916 int err = mnt_alloc_group_id(p);
1918 cleanup_group_ids(mnt, p);
1927 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1929 unsigned int max = READ_ONCE(sysctl_mount_max);
1930 unsigned int mounts = 0, old, pending, sum;
1933 for (p = mnt; p; p = next_mnt(p, mnt))
1937 pending = ns->pending_mounts;
1938 sum = old + pending;
1942 (mounts > (max - sum)))
1945 ns->pending_mounts = pending + mounts;
1950 * @source_mnt : mount tree to be attached
1951 * @nd : place the mount tree @source_mnt is attached
1952 * @parent_nd : if non-null, detach the source_mnt from its parent and
1953 * store the parent mount and mountpoint dentry.
1954 * (done when source_mnt is moved)
1956 * NOTE: in the table below explains the semantics when a source mount
1957 * of a given type is attached to a destination mount of a given type.
1958 * ---------------------------------------------------------------------------
1959 * | BIND MOUNT OPERATION |
1960 * |**************************************************************************
1961 * | source-->| shared | private | slave | unbindable |
1965 * |**************************************************************************
1966 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1968 * |non-shared| shared (+) | private | slave (*) | invalid |
1969 * ***************************************************************************
1970 * A bind operation clones the source mount and mounts the clone on the
1971 * destination mount.
1973 * (++) the cloned mount is propagated to all the mounts in the propagation
1974 * tree of the destination mount and the cloned mount is added to
1975 * the peer group of the source mount.
1976 * (+) the cloned mount is created under the destination mount and is marked
1977 * as shared. The cloned mount is added to the peer group of the source
1979 * (+++) the mount is propagated to all the mounts in the propagation tree
1980 * of the destination mount and the cloned mount is made slave
1981 * of the same master as that of the source mount. The cloned mount
1982 * is marked as 'shared and slave'.
1983 * (*) the cloned mount is made a slave of the same master as that of the
1986 * ---------------------------------------------------------------------------
1987 * | MOVE MOUNT OPERATION |
1988 * |**************************************************************************
1989 * | source-->| shared | private | slave | unbindable |
1993 * |**************************************************************************
1994 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1996 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1997 * ***************************************************************************
1999 * (+) the mount is moved to the destination. And is then propagated to
2000 * all the mounts in the propagation tree of the destination mount.
2001 * (+*) the mount is moved to the destination.
2002 * (+++) the mount is moved to the destination and is then propagated to
2003 * all the mounts belonging to the destination mount's propagation tree.
2004 * the mount is marked as 'shared and slave'.
2005 * (*) the mount continues to be a slave at the new location.
2007 * if the source mount is a tree, the operations explained above is
2008 * applied to each mount in the tree.
2009 * Must be called without spinlocks held, since this function can sleep
2012 static int attach_recursive_mnt(struct mount *source_mnt,
2013 struct mount *dest_mnt,
2014 struct mountpoint *dest_mp,
2015 struct path *parent_path)
2017 HLIST_HEAD(tree_list);
2018 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2019 struct mountpoint *smp;
2020 struct mount *child, *p;
2021 struct hlist_node *n;
2024 /* Preallocate a mountpoint in case the new mounts need
2025 * to be tucked under other mounts.
2027 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2029 return PTR_ERR(smp);
2031 /* Is there space to add these mounts to the mount namespace? */
2033 err = count_mounts(ns, source_mnt);
2038 if (IS_MNT_SHARED(dest_mnt)) {
2039 err = invent_group_ids(source_mnt, true);
2042 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2045 goto out_cleanup_ids;
2046 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2052 detach_mnt(source_mnt, parent_path);
2053 attach_mnt(source_mnt, dest_mnt, dest_mp);
2054 touch_mnt_namespace(source_mnt->mnt_ns);
2056 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2057 commit_tree(source_mnt);
2060 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2062 hlist_del_init(&child->mnt_hash);
2063 q = __lookup_mnt(&child->mnt_parent->mnt,
2064 child->mnt_mountpoint);
2066 mnt_change_mountpoint(child, smp, q);
2069 put_mountpoint(smp);
2070 unlock_mount_hash();
2075 while (!hlist_empty(&tree_list)) {
2076 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2077 child->mnt_parent->mnt_ns->pending_mounts = 0;
2078 umount_tree(child, UMOUNT_SYNC);
2080 unlock_mount_hash();
2081 cleanup_group_ids(source_mnt, NULL);
2083 ns->pending_mounts = 0;
2085 read_seqlock_excl(&mount_lock);
2086 put_mountpoint(smp);
2087 read_sequnlock_excl(&mount_lock);
2092 static struct mountpoint *lock_mount(struct path *path)
2094 struct vfsmount *mnt;
2095 struct dentry *dentry = path->dentry;
2097 mutex_lock(&dentry->d_inode->i_mutex);
2098 if (unlikely(cant_mount(dentry))) {
2099 mutex_unlock(&dentry->d_inode->i_mutex);
2100 return ERR_PTR(-ENOENT);
2103 mnt = lookup_mnt(path);
2105 struct mountpoint *mp = get_mountpoint(dentry);
2108 mutex_unlock(&dentry->d_inode->i_mutex);
2114 mutex_unlock(&path->dentry->d_inode->i_mutex);
2117 dentry = path->dentry = dget(mnt->mnt_root);
2121 static void unlock_mount(struct mountpoint *where)
2123 struct dentry *dentry = where->m_dentry;
2125 read_seqlock_excl(&mount_lock);
2126 put_mountpoint(where);
2127 read_sequnlock_excl(&mount_lock);
2130 mutex_unlock(&dentry->d_inode->i_mutex);
2133 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2135 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2138 if (d_is_dir(mp->m_dentry) !=
2139 d_is_dir(mnt->mnt.mnt_root))
2142 return attach_recursive_mnt(mnt, p, mp, NULL);
2146 * Sanity check the flags to change_mnt_propagation.
2149 static int flags_to_propagation_type(int flags)
2151 int type = flags & ~(MS_REC | MS_SILENT);
2153 /* Fail if any non-propagation flags are set */
2154 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2156 /* Only one propagation flag should be set */
2157 if (!is_power_of_2(type))
2163 * recursively change the type of the mountpoint.
2165 static int do_change_type(struct path *path, int flag)
2168 struct mount *mnt = real_mount(path->mnt);
2169 int recurse = flag & MS_REC;
2173 if (path->dentry != path->mnt->mnt_root)
2176 type = flags_to_propagation_type(flag);
2181 if (type == MS_SHARED) {
2182 err = invent_group_ids(mnt, recurse);
2188 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2189 change_mnt_propagation(m, type);
2190 unlock_mount_hash();
2198 * do loopback mount.
2200 static int do_loopback(struct path *path, const char *old_name,
2203 struct path old_path;
2204 struct mount *mnt = NULL, *old, *parent;
2205 struct mountpoint *mp;
2207 if (!old_name || !*old_name)
2209 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2214 if (mnt_ns_loop(old_path.dentry))
2217 mp = lock_mount(path);
2222 old = real_mount(old_path.mnt);
2223 parent = real_mount(path->mnt);
2226 if (IS_MNT_UNBINDABLE(old))
2229 if (!check_mnt(parent))
2232 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2235 if (!recurse && has_locked_children(old, old_path.dentry))
2239 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2241 mnt = clone_mnt(old, old_path.dentry, 0);
2248 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2250 err = graft_tree(mnt, parent, mp);
2253 umount_tree(mnt, UMOUNT_SYNC);
2254 unlock_mount_hash();
2259 path_put(&old_path);
2263 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2266 int readonly_request = 0;
2268 if (ms_flags & MS_RDONLY)
2269 readonly_request = 1;
2270 if (readonly_request == __mnt_is_readonly(mnt))
2273 if (readonly_request)
2274 error = mnt_make_readonly(real_mount(mnt));
2276 __mnt_unmake_readonly(real_mount(mnt));
2281 * change filesystem flags. dir should be a physical root of filesystem.
2282 * If you've mounted a non-root directory somewhere and want to do remount
2283 * on it - tough luck.
2285 static int do_remount(struct path *path, int flags, int mnt_flags,
2289 struct super_block *sb = path->mnt->mnt_sb;
2290 struct mount *mnt = real_mount(path->mnt);
2292 if (!check_mnt(mnt))
2295 if (path->dentry != path->mnt->mnt_root)
2298 /* Don't allow changing of locked mnt flags.
2300 * No locks need to be held here while testing the various
2301 * MNT_LOCK flags because those flags can never be cleared
2302 * once they are set.
2304 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2305 !(mnt_flags & MNT_READONLY)) {
2308 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2309 !(mnt_flags & MNT_NODEV)) {
2310 /* Was the nodev implicitly added in mount? */
2311 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2312 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2313 mnt_flags |= MNT_NODEV;
2318 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2319 !(mnt_flags & MNT_NOSUID)) {
2322 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2323 !(mnt_flags & MNT_NOEXEC)) {
2326 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2327 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2331 err = security_sb_remount(sb, data);
2335 down_write(&sb->s_umount);
2336 if (flags & MS_BIND)
2337 err = change_mount_flags(path->mnt, flags);
2338 else if (!capable(CAP_SYS_ADMIN))
2341 err = do_remount_sb(sb, flags, data, 0);
2344 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2345 mnt->mnt.mnt_flags = mnt_flags;
2346 touch_mnt_namespace(mnt->mnt_ns);
2347 unlock_mount_hash();
2349 up_write(&sb->s_umount);
2353 static inline int tree_contains_unbindable(struct mount *mnt)
2356 for (p = mnt; p; p = next_mnt(p, mnt)) {
2357 if (IS_MNT_UNBINDABLE(p))
2363 static int do_move_mount(struct path *path, const char *old_name)
2365 struct path old_path, parent_path;
2368 struct mountpoint *mp;
2370 if (!old_name || !*old_name)
2372 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2376 mp = lock_mount(path);
2381 old = real_mount(old_path.mnt);
2382 p = real_mount(path->mnt);
2385 if (!check_mnt(p) || !check_mnt(old))
2388 if (old->mnt.mnt_flags & MNT_LOCKED)
2392 if (old_path.dentry != old_path.mnt->mnt_root)
2395 if (!mnt_has_parent(old))
2398 if (d_is_dir(path->dentry) !=
2399 d_is_dir(old_path.dentry))
2402 * Don't move a mount residing in a shared parent.
2404 if (IS_MNT_SHARED(old->mnt_parent))
2407 * Don't move a mount tree containing unbindable mounts to a destination
2408 * mount which is shared.
2410 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2413 for (; mnt_has_parent(p); p = p->mnt_parent)
2417 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2421 /* if the mount is moved, it should no longer be expire
2423 list_del_init(&old->mnt_expire);
2428 path_put(&parent_path);
2429 path_put(&old_path);
2433 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2436 const char *subtype = strchr(fstype, '.');
2445 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2447 if (!mnt->mnt_sb->s_subtype)
2453 return ERR_PTR(err);
2457 * add a mount into a namespace's mount tree
2459 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2461 struct mountpoint *mp;
2462 struct mount *parent;
2465 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2467 mp = lock_mount(path);
2471 parent = real_mount(path->mnt);
2473 if (unlikely(!check_mnt(parent))) {
2474 /* that's acceptable only for automounts done in private ns */
2475 if (!(mnt_flags & MNT_SHRINKABLE))
2477 /* ... and for those we'd better have mountpoint still alive */
2478 if (!parent->mnt_ns)
2482 /* Refuse the same filesystem on the same mount point */
2484 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2485 path->mnt->mnt_root == path->dentry)
2489 if (d_is_symlink(newmnt->mnt.mnt_root))
2492 newmnt->mnt.mnt_flags = mnt_flags;
2493 err = graft_tree(newmnt, parent, mp);
2500 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2503 * create a new mount for userspace and request it to be added into the
2506 static int do_new_mount(struct path *path, const char *fstype, int flags,
2507 int mnt_flags, const char *name, void *data)
2509 struct file_system_type *type;
2510 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2511 struct vfsmount *mnt;
2517 type = get_fs_type(fstype);
2521 if (user_ns != &init_user_ns) {
2522 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2523 put_filesystem(type);
2526 /* Only in special cases allow devices from mounts
2527 * created outside the initial user namespace.
2529 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2531 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2533 if (type->fs_flags & FS_USERNS_VISIBLE) {
2534 if (!fs_fully_visible(type, &mnt_flags)) {
2535 put_filesystem(type);
2541 mnt = vfs_kern_mount(type, flags, name, data);
2542 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2543 !mnt->mnt_sb->s_subtype)
2544 mnt = fs_set_subtype(mnt, fstype);
2546 put_filesystem(type);
2548 return PTR_ERR(mnt);
2550 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2556 int finish_automount(struct vfsmount *m, struct path *path)
2558 struct mount *mnt = real_mount(m);
2560 /* The new mount record should have at least 2 refs to prevent it being
2561 * expired before we get a chance to add it
2563 BUG_ON(mnt_get_count(mnt) < 2);
2565 if (m->mnt_sb == path->mnt->mnt_sb &&
2566 m->mnt_root == path->dentry) {
2571 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2575 /* remove m from any expiration list it may be on */
2576 if (!list_empty(&mnt->mnt_expire)) {
2578 list_del_init(&mnt->mnt_expire);
2587 * mnt_set_expiry - Put a mount on an expiration list
2588 * @mnt: The mount to list.
2589 * @expiry_list: The list to add the mount to.
2591 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2595 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2599 EXPORT_SYMBOL(mnt_set_expiry);
2602 * process a list of expirable mountpoints with the intent of discarding any
2603 * mountpoints that aren't in use and haven't been touched since last we came
2606 void mark_mounts_for_expiry(struct list_head *mounts)
2608 struct mount *mnt, *next;
2609 LIST_HEAD(graveyard);
2611 if (list_empty(mounts))
2617 /* extract from the expiration list every vfsmount that matches the
2618 * following criteria:
2619 * - only referenced by its parent vfsmount
2620 * - still marked for expiry (marked on the last call here; marks are
2621 * cleared by mntput())
2623 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2624 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2625 propagate_mount_busy(mnt, 1))
2627 list_move(&mnt->mnt_expire, &graveyard);
2629 while (!list_empty(&graveyard)) {
2630 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2631 touch_mnt_namespace(mnt->mnt_ns);
2632 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2634 unlock_mount_hash();
2638 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2641 * Ripoff of 'select_parent()'
2643 * search the list of submounts for a given mountpoint, and move any
2644 * shrinkable submounts to the 'graveyard' list.
2646 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2648 struct mount *this_parent = parent;
2649 struct list_head *next;
2653 next = this_parent->mnt_mounts.next;
2655 while (next != &this_parent->mnt_mounts) {
2656 struct list_head *tmp = next;
2657 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2660 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2663 * Descend a level if the d_mounts list is non-empty.
2665 if (!list_empty(&mnt->mnt_mounts)) {
2670 if (!propagate_mount_busy(mnt, 1)) {
2671 list_move_tail(&mnt->mnt_expire, graveyard);
2676 * All done at this level ... ascend and resume the search
2678 if (this_parent != parent) {
2679 next = this_parent->mnt_child.next;
2680 this_parent = this_parent->mnt_parent;
2687 * process a list of expirable mountpoints with the intent of discarding any
2688 * submounts of a specific parent mountpoint
2690 * mount_lock must be held for write
2692 static void shrink_submounts(struct mount *mnt)
2694 LIST_HEAD(graveyard);
2697 /* extract submounts of 'mountpoint' from the expiration list */
2698 while (select_submounts(mnt, &graveyard)) {
2699 while (!list_empty(&graveyard)) {
2700 m = list_first_entry(&graveyard, struct mount,
2702 touch_mnt_namespace(m->mnt_ns);
2703 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2709 * Some copy_from_user() implementations do not return the exact number of
2710 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2711 * Note that this function differs from copy_from_user() in that it will oops
2712 * on bad values of `to', rather than returning a short copy.
2714 static long exact_copy_from_user(void *to, const void __user * from,
2718 const char __user *f = from;
2721 if (!access_ok(VERIFY_READ, from, n))
2725 if (__get_user(c, f)) {
2736 int copy_mount_options(const void __user * data, unsigned long *where)
2746 if (!(page = __get_free_page(GFP_KERNEL)))
2749 /* We only care that *some* data at the address the user
2750 * gave us is valid. Just in case, we'll zero
2751 * the remainder of the page.
2753 /* copy_from_user cannot cross TASK_SIZE ! */
2754 size = TASK_SIZE - (unsigned long)data;
2755 if (size > PAGE_SIZE)
2758 i = size - exact_copy_from_user((void *)page, data, size);
2764 memset((char *)page + i, 0, PAGE_SIZE - i);
2769 char *copy_mount_string(const void __user *data)
2771 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2775 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2776 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2778 * data is a (void *) that can point to any structure up to
2779 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2780 * information (or be NULL).
2782 * Pre-0.97 versions of mount() didn't have a flags word.
2783 * When the flags word was introduced its top half was required
2784 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2785 * Therefore, if this magic number is present, it carries no information
2786 * and must be discarded.
2788 long do_mount(const char *dev_name, const char __user *dir_name,
2789 const char *type_page, unsigned long flags, void *data_page)
2796 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2797 flags &= ~MS_MGC_MSK;
2799 /* Basic sanity checks */
2801 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2803 /* ... and get the mountpoint */
2804 retval = user_path(dir_name, &path);
2808 retval = security_sb_mount(dev_name, &path,
2809 type_page, flags, data_page);
2810 if (!retval && !may_mount())
2815 /* Default to relatime unless overriden */
2816 if (!(flags & MS_NOATIME))
2817 mnt_flags |= MNT_RELATIME;
2819 /* Separate the per-mountpoint flags */
2820 if (flags & MS_NOSUID)
2821 mnt_flags |= MNT_NOSUID;
2822 if (flags & MS_NODEV)
2823 mnt_flags |= MNT_NODEV;
2824 if (flags & MS_NOEXEC)
2825 mnt_flags |= MNT_NOEXEC;
2826 if (flags & MS_NOATIME)
2827 mnt_flags |= MNT_NOATIME;
2828 if (flags & MS_NODIRATIME)
2829 mnt_flags |= MNT_NODIRATIME;
2830 if (flags & MS_STRICTATIME)
2831 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2832 if (flags & MS_RDONLY)
2833 mnt_flags |= MNT_READONLY;
2835 /* The default atime for remount is preservation */
2836 if ((flags & MS_REMOUNT) &&
2837 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2838 MS_STRICTATIME)) == 0)) {
2839 mnt_flags &= ~MNT_ATIME_MASK;
2840 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2843 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2844 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2847 if (flags & MS_REMOUNT)
2848 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2850 else if (flags & MS_BIND)
2851 retval = do_loopback(&path, dev_name, flags & MS_REC);
2852 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2853 retval = do_change_type(&path, flags);
2854 else if (flags & MS_MOVE)
2855 retval = do_move_mount(&path, dev_name);
2857 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2858 dev_name, data_page);
2864 static void free_mnt_ns(struct mnt_namespace *ns)
2866 ns_free_inum(&ns->ns);
2867 put_user_ns(ns->user_ns);
2872 * Assign a sequence number so we can detect when we attempt to bind
2873 * mount a reference to an older mount namespace into the current
2874 * mount namespace, preventing reference counting loops. A 64bit
2875 * number incrementing at 10Ghz will take 12,427 years to wrap which
2876 * is effectively never, so we can ignore the possibility.
2878 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2880 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2882 struct mnt_namespace *new_ns;
2885 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2887 return ERR_PTR(-ENOMEM);
2888 ret = ns_alloc_inum(&new_ns->ns);
2891 return ERR_PTR(ret);
2893 new_ns->ns.ops = &mntns_operations;
2894 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2895 atomic_set(&new_ns->count, 1);
2896 new_ns->root = NULL;
2897 INIT_LIST_HEAD(&new_ns->list);
2898 init_waitqueue_head(&new_ns->poll);
2900 new_ns->user_ns = get_user_ns(user_ns);
2902 new_ns->pending_mounts = 0;
2906 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2907 struct user_namespace *user_ns, struct fs_struct *new_fs)
2909 struct mnt_namespace *new_ns;
2910 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2911 struct mount *p, *q;
2918 if (likely(!(flags & CLONE_NEWNS))) {
2925 new_ns = alloc_mnt_ns(user_ns);
2930 /* First pass: copy the tree topology */
2931 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2932 if (user_ns != ns->user_ns)
2933 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2934 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2937 free_mnt_ns(new_ns);
2938 return ERR_CAST(new);
2941 list_add_tail(&new_ns->list, &new->mnt_list);
2944 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2945 * as belonging to new namespace. We have already acquired a private
2946 * fs_struct, so tsk->fs->lock is not needed.
2954 if (&p->mnt == new_fs->root.mnt) {
2955 new_fs->root.mnt = mntget(&q->mnt);
2958 if (&p->mnt == new_fs->pwd.mnt) {
2959 new_fs->pwd.mnt = mntget(&q->mnt);
2963 p = next_mnt(p, old);
2964 q = next_mnt(q, new);
2967 while (p->mnt.mnt_root != q->mnt.mnt_root)
2968 p = next_mnt(p, old);
2981 * create_mnt_ns - creates a private namespace and adds a root filesystem
2982 * @mnt: pointer to the new root filesystem mountpoint
2984 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2986 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2987 if (!IS_ERR(new_ns)) {
2988 struct mount *mnt = real_mount(m);
2989 mnt->mnt_ns = new_ns;
2992 list_add(&mnt->mnt_list, &new_ns->list);
2999 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3001 struct mnt_namespace *ns;
3002 struct super_block *s;
3006 ns = create_mnt_ns(mnt);
3008 return ERR_CAST(ns);
3010 err = vfs_path_lookup(mnt->mnt_root, mnt,
3011 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3016 return ERR_PTR(err);
3018 /* trade a vfsmount reference for active sb one */
3019 s = path.mnt->mnt_sb;
3020 atomic_inc(&s->s_active);
3022 /* lock the sucker */
3023 down_write(&s->s_umount);
3024 /* ... and return the root of (sub)tree on it */
3027 EXPORT_SYMBOL(mount_subtree);
3029 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3030 char __user *, type, unsigned long, flags, void __user *, data)
3035 unsigned long data_page;
3037 kernel_type = copy_mount_string(type);
3038 ret = PTR_ERR(kernel_type);
3039 if (IS_ERR(kernel_type))
3042 kernel_dev = copy_mount_string(dev_name);
3043 ret = PTR_ERR(kernel_dev);
3044 if (IS_ERR(kernel_dev))
3047 ret = copy_mount_options(data, &data_page);
3051 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
3052 (void *) data_page);
3054 free_page(data_page);
3064 * Return true if path is reachable from root
3066 * namespace_sem or mount_lock is held
3068 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3069 const struct path *root)
3071 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3072 dentry = mnt->mnt_mountpoint;
3073 mnt = mnt->mnt_parent;
3075 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3078 int path_is_under(struct path *path1, struct path *path2)
3081 read_seqlock_excl(&mount_lock);
3082 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3083 read_sequnlock_excl(&mount_lock);
3086 EXPORT_SYMBOL(path_is_under);
3089 * pivot_root Semantics:
3090 * Moves the root file system of the current process to the directory put_old,
3091 * makes new_root as the new root file system of the current process, and sets
3092 * root/cwd of all processes which had them on the current root to new_root.
3095 * The new_root and put_old must be directories, and must not be on the
3096 * same file system as the current process root. The put_old must be
3097 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3098 * pointed to by put_old must yield the same directory as new_root. No other
3099 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3101 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3102 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3103 * in this situation.
3106 * - we don't move root/cwd if they are not at the root (reason: if something
3107 * cared enough to change them, it's probably wrong to force them elsewhere)
3108 * - it's okay to pick a root that isn't the root of a file system, e.g.
3109 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3110 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3113 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3114 const char __user *, put_old)
3116 struct path new, old, parent_path, root_parent, root;
3117 struct mount *new_mnt, *root_mnt, *old_mnt;
3118 struct mountpoint *old_mp, *root_mp;
3124 error = user_path_dir(new_root, &new);
3128 error = user_path_dir(put_old, &old);
3132 error = security_sb_pivotroot(&old, &new);
3136 get_fs_root(current->fs, &root);
3137 old_mp = lock_mount(&old);
3138 error = PTR_ERR(old_mp);
3143 new_mnt = real_mount(new.mnt);
3144 root_mnt = real_mount(root.mnt);
3145 old_mnt = real_mount(old.mnt);
3146 if (IS_MNT_SHARED(old_mnt) ||
3147 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3148 IS_MNT_SHARED(root_mnt->mnt_parent))
3150 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3152 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3155 if (d_unlinked(new.dentry))
3158 if (new_mnt == root_mnt || old_mnt == root_mnt)
3159 goto out4; /* loop, on the same file system */
3161 if (root.mnt->mnt_root != root.dentry)
3162 goto out4; /* not a mountpoint */
3163 if (!mnt_has_parent(root_mnt))
3164 goto out4; /* not attached */
3165 root_mp = root_mnt->mnt_mp;
3166 if (new.mnt->mnt_root != new.dentry)
3167 goto out4; /* not a mountpoint */
3168 if (!mnt_has_parent(new_mnt))
3169 goto out4; /* not attached */
3170 /* make sure we can reach put_old from new_root */
3171 if (!is_path_reachable(old_mnt, old.dentry, &new))
3173 /* make certain new is below the root */
3174 if (!is_path_reachable(new_mnt, new.dentry, &root))
3177 root_mp->m_count++; /* pin it so it won't go away */
3178 detach_mnt(new_mnt, &parent_path);
3179 detach_mnt(root_mnt, &root_parent);
3180 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3181 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3182 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3184 /* mount old root on put_old */
3185 attach_mnt(root_mnt, old_mnt, old_mp);
3186 /* mount new_root on / */
3187 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3188 touch_mnt_namespace(current->nsproxy->mnt_ns);
3189 /* A moved mount should not expire automatically */
3190 list_del_init(&new_mnt->mnt_expire);
3191 put_mountpoint(root_mp);
3192 unlock_mount_hash();
3193 chroot_fs_refs(&root, &new);
3196 unlock_mount(old_mp);
3198 path_put(&root_parent);
3199 path_put(&parent_path);
3211 static void __init init_mount_tree(void)
3213 struct vfsmount *mnt;
3214 struct mnt_namespace *ns;
3216 struct file_system_type *type;
3218 type = get_fs_type("rootfs");
3220 panic("Can't find rootfs type");
3221 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3222 put_filesystem(type);
3224 panic("Can't create rootfs");
3226 ns = create_mnt_ns(mnt);
3228 panic("Can't allocate initial namespace");
3230 init_task.nsproxy->mnt_ns = ns;
3234 root.dentry = mnt->mnt_root;
3235 mnt->mnt_flags |= MNT_LOCKED;
3237 set_fs_pwd(current->fs, &root);
3238 set_fs_root(current->fs, &root);
3241 void __init mnt_init(void)
3246 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3247 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3249 mount_hashtable = alloc_large_system_hash("Mount-cache",
3250 sizeof(struct hlist_head),
3253 &m_hash_shift, &m_hash_mask, 0, 0);
3254 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3255 sizeof(struct hlist_head),
3258 &mp_hash_shift, &mp_hash_mask, 0, 0);
3260 if (!mount_hashtable || !mountpoint_hashtable)
3261 panic("Failed to allocate mount hash table\n");
3263 for (u = 0; u <= m_hash_mask; u++)
3264 INIT_HLIST_HEAD(&mount_hashtable[u]);
3265 for (u = 0; u <= mp_hash_mask; u++)
3266 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3272 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3274 fs_kobj = kobject_create_and_add("fs", NULL);
3276 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3281 void put_mnt_ns(struct mnt_namespace *ns)
3283 if (!atomic_dec_and_test(&ns->count))
3285 drop_collected_mounts(&ns->root->mnt);
3289 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3291 struct vfsmount *mnt;
3292 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3295 * it is a longterm mount, don't release mnt until
3296 * we unmount before file sys is unregistered
3298 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3302 EXPORT_SYMBOL_GPL(kern_mount_data);
3304 void kern_unmount(struct vfsmount *mnt)
3306 /* release long term mount so mount point can be released */
3307 if (!IS_ERR_OR_NULL(mnt)) {
3308 real_mount(mnt)->mnt_ns = NULL;
3309 synchronize_rcu(); /* yecchhh... */
3313 EXPORT_SYMBOL(kern_unmount);
3315 bool our_mnt(struct vfsmount *mnt)
3317 return check_mnt(real_mount(mnt));
3320 bool current_chrooted(void)
3322 /* Does the current process have a non-standard root */
3323 struct path ns_root;
3324 struct path fs_root;
3327 /* Find the namespace root */
3328 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3329 ns_root.dentry = ns_root.mnt->mnt_root;
3331 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3334 get_fs_root(current->fs, &fs_root);
3336 chrooted = !path_equal(&fs_root, &ns_root);
3344 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3346 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3347 int new_flags = *new_mnt_flags;
3349 bool visible = false;
3354 down_read(&namespace_sem);
3355 list_for_each_entry(mnt, &ns->list, mnt_list) {
3356 struct mount *child;
3359 if (mnt->mnt.mnt_sb->s_type != type)
3362 /* This mount is not fully visible if it's root directory
3363 * is not the root directory of the filesystem.
3365 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3368 /* Read the mount flags and filter out flags that
3369 * may safely be ignored.
3371 mnt_flags = mnt->mnt.mnt_flags;
3372 if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3373 mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3375 /* Don't miss readonly hidden in the superblock flags */
3376 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3377 mnt_flags |= MNT_LOCK_READONLY;
3379 /* Verify the mount flags are equal to or more permissive
3380 * than the proposed new mount.
3382 if ((mnt_flags & MNT_LOCK_READONLY) &&
3383 !(new_flags & MNT_READONLY))
3385 if ((mnt_flags & MNT_LOCK_NODEV) &&
3386 !(new_flags & MNT_NODEV))
3388 if ((mnt_flags & MNT_LOCK_NOSUID) &&
3389 !(new_flags & MNT_NOSUID))
3391 if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3392 !(new_flags & MNT_NOEXEC))
3394 if ((mnt_flags & MNT_LOCK_ATIME) &&
3395 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3398 /* This mount is not fully visible if there are any
3399 * locked child mounts that cover anything except for
3400 * empty directories.
3402 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3403 struct inode *inode = child->mnt_mountpoint->d_inode;
3404 /* Only worry about locked mounts */
3405 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3407 /* Is the directory permanetly empty? */
3408 if (!is_empty_dir_inode(inode))
3411 /* Preserve the locked attributes */
3412 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3422 up_read(&namespace_sem);
3426 static struct ns_common *mntns_get(struct task_struct *task)
3428 struct ns_common *ns = NULL;
3429 struct nsproxy *nsproxy;
3432 nsproxy = task->nsproxy;
3434 ns = &nsproxy->mnt_ns->ns;
3435 get_mnt_ns(to_mnt_ns(ns));
3442 static void mntns_put(struct ns_common *ns)
3444 put_mnt_ns(to_mnt_ns(ns));
3447 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3449 struct fs_struct *fs = current->fs;
3450 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3453 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3454 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3455 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3462 put_mnt_ns(nsproxy->mnt_ns);
3463 nsproxy->mnt_ns = mnt_ns;
3466 root.mnt = &mnt_ns->root->mnt;
3467 root.dentry = mnt_ns->root->mnt.mnt_root;
3469 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3472 /* Update the pwd and root */
3473 set_fs_pwd(fs, &root);
3474 set_fs_root(fs, &root);
3480 const struct proc_ns_operations mntns_operations = {
3482 .type = CLONE_NEWNS,
3485 .install = mntns_install,