1 // SPDX-License-Identifier: GPL-2.0-only
5 * (C) Copyright Al Viro 2000, 2001
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/cred.h>
19 #include <linux/idr.h>
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/file.h>
24 #include <linux/uaccess.h>
25 #include <linux/proc_ns.h>
26 #include <linux/magic.h>
27 #include <linux/memblock.h>
28 #include <linux/task_work.h>
29 #include <linux/sched/task.h>
30 #include <uapi/linux/mount.h>
31 #include <linux/fs_context.h>
32 #include <linux/shmem_fs.h>
37 /* Maximum number of mounts in a mount namespace */
38 unsigned int sysctl_mount_max __read_mostly = 100000;
40 static unsigned int m_hash_mask __read_mostly;
41 static unsigned int m_hash_shift __read_mostly;
42 static unsigned int mp_hash_mask __read_mostly;
43 static unsigned int mp_hash_shift __read_mostly;
45 static __initdata unsigned long mhash_entries;
46 static int __init set_mhash_entries(char *str)
50 mhash_entries = simple_strtoul(str, &str, 0);
53 __setup("mhash_entries=", set_mhash_entries);
55 static __initdata unsigned long mphash_entries;
56 static int __init set_mphash_entries(char *str)
60 mphash_entries = simple_strtoul(str, &str, 0);
63 __setup("mphash_entries=", set_mphash_entries);
66 static DEFINE_IDA(mnt_id_ida);
67 static DEFINE_IDA(mnt_group_ida);
69 static struct hlist_head *mount_hashtable __read_mostly;
70 static struct hlist_head *mountpoint_hashtable __read_mostly;
71 static struct kmem_cache *mnt_cache __read_mostly;
72 static DECLARE_RWSEM(namespace_sem);
73 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
74 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
77 struct kobject *fs_kobj;
78 EXPORT_SYMBOL_GPL(fs_kobj);
81 * vfsmount lock may be taken for read to prevent changes to the
82 * vfsmount hash, ie. during mountpoint lookups or walking back
85 * It should be taken for write in all cases where the vfsmount
86 * tree or hash is modified or when a vfsmount structure is modified.
88 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
90 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
92 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
93 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
94 tmp = tmp + (tmp >> m_hash_shift);
95 return &mount_hashtable[tmp & m_hash_mask];
98 static inline struct hlist_head *mp_hash(struct dentry *dentry)
100 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
101 tmp = tmp + (tmp >> mp_hash_shift);
102 return &mountpoint_hashtable[tmp & mp_hash_mask];
105 static int mnt_alloc_id(struct mount *mnt)
107 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
115 static void mnt_free_id(struct mount *mnt)
117 ida_free(&mnt_id_ida, mnt->mnt_id);
121 * Allocate a new peer group ID
123 static int mnt_alloc_group_id(struct mount *mnt)
125 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
129 mnt->mnt_group_id = res;
134 * Release a peer group ID
136 void mnt_release_group_id(struct mount *mnt)
138 ida_free(&mnt_group_ida, mnt->mnt_group_id);
139 mnt->mnt_group_id = 0;
143 * vfsmount lock must be held for read
145 static inline void mnt_add_count(struct mount *mnt, int n)
148 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
157 * vfsmount lock must be held for write
159 int mnt_get_count(struct mount *mnt)
165 for_each_possible_cpu(cpu) {
166 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
171 return mnt->mnt_count;
175 static struct mount *alloc_vfsmnt(const char *name)
177 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
181 err = mnt_alloc_id(mnt);
186 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
192 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
194 goto out_free_devname;
196 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
199 mnt->mnt_writers = 0;
202 INIT_HLIST_NODE(&mnt->mnt_hash);
203 INIT_LIST_HEAD(&mnt->mnt_child);
204 INIT_LIST_HEAD(&mnt->mnt_mounts);
205 INIT_LIST_HEAD(&mnt->mnt_list);
206 INIT_LIST_HEAD(&mnt->mnt_expire);
207 INIT_LIST_HEAD(&mnt->mnt_share);
208 INIT_LIST_HEAD(&mnt->mnt_slave_list);
209 INIT_LIST_HEAD(&mnt->mnt_slave);
210 INIT_HLIST_NODE(&mnt->mnt_mp_list);
211 INIT_LIST_HEAD(&mnt->mnt_umounting);
212 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
213 mnt->mnt.mnt_userns = &init_user_ns;
219 kfree_const(mnt->mnt_devname);
224 kmem_cache_free(mnt_cache, mnt);
229 * Most r/o checks on a fs are for operations that take
230 * discrete amounts of time, like a write() or unlink().
231 * We must keep track of when those operations start
232 * (for permission checks) and when they end, so that
233 * we can determine when writes are able to occur to
237 * __mnt_is_readonly: check whether a mount is read-only
238 * @mnt: the mount to check for its write status
240 * This shouldn't be used directly ouside of the VFS.
241 * It does not guarantee that the filesystem will stay
242 * r/w, just that it is right *now*. This can not and
243 * should not be used in place of IS_RDONLY(inode).
244 * mnt_want/drop_write() will _keep_ the filesystem
247 bool __mnt_is_readonly(struct vfsmount *mnt)
249 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
251 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
253 static inline void mnt_inc_writers(struct mount *mnt)
256 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
262 static inline void mnt_dec_writers(struct mount *mnt)
265 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
271 static unsigned int mnt_get_writers(struct mount *mnt)
274 unsigned int count = 0;
277 for_each_possible_cpu(cpu) {
278 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
283 return mnt->mnt_writers;
287 static int mnt_is_readonly(struct vfsmount *mnt)
289 if (mnt->mnt_sb->s_readonly_remount)
291 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
293 return __mnt_is_readonly(mnt);
297 * Most r/o & frozen checks on a fs are for operations that take discrete
298 * amounts of time, like a write() or unlink(). We must keep track of when
299 * those operations start (for permission checks) and when they end, so that we
300 * can determine when writes are able to occur to a filesystem.
303 * __mnt_want_write - get write access to a mount without freeze protection
304 * @m: the mount on which to take a write
306 * This tells the low-level filesystem that a write is about to be performed to
307 * it, and makes sure that writes are allowed (mnt it read-write) before
308 * returning success. This operation does not protect against filesystem being
309 * frozen. When the write operation is finished, __mnt_drop_write() must be
310 * called. This is effectively a refcount.
312 int __mnt_want_write(struct vfsmount *m)
314 struct mount *mnt = real_mount(m);
318 mnt_inc_writers(mnt);
320 * The store to mnt_inc_writers must be visible before we pass
321 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
322 * incremented count after it has set MNT_WRITE_HOLD.
325 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
328 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
329 * be set to match its requirements. So we must not load that until
330 * MNT_WRITE_HOLD is cleared.
333 if (mnt_is_readonly(m)) {
334 mnt_dec_writers(mnt);
343 * mnt_want_write - get write access to a mount
344 * @m: the mount on which to take a write
346 * This tells the low-level filesystem that a write is about to be performed to
347 * it, and makes sure that writes are allowed (mount is read-write, filesystem
348 * is not frozen) before returning success. When the write operation is
349 * finished, mnt_drop_write() must be called. This is effectively a refcount.
351 int mnt_want_write(struct vfsmount *m)
355 sb_start_write(m->mnt_sb);
356 ret = __mnt_want_write(m);
358 sb_end_write(m->mnt_sb);
361 EXPORT_SYMBOL_GPL(mnt_want_write);
364 * mnt_clone_write - get write access to a mount
365 * @mnt: the mount on which to take a write
367 * This is effectively like mnt_want_write, except
368 * it must only be used to take an extra write reference
369 * on a mountpoint that we already know has a write reference
370 * on it. This allows some optimisation.
372 * After finished, mnt_drop_write must be called as usual to
373 * drop the reference.
375 int mnt_clone_write(struct vfsmount *mnt)
377 /* superblock may be r/o */
378 if (__mnt_is_readonly(mnt))
381 mnt_inc_writers(real_mount(mnt));
385 EXPORT_SYMBOL_GPL(mnt_clone_write);
388 * __mnt_want_write_file - get write access to a file's mount
389 * @file: the file who's mount on which to take a write
391 * This is like __mnt_want_write, but it takes a file and can
392 * do some optimisations if the file is open for write already
394 int __mnt_want_write_file(struct file *file)
396 if (!(file->f_mode & FMODE_WRITER))
397 return __mnt_want_write(file->f_path.mnt);
399 return mnt_clone_write(file->f_path.mnt);
403 * mnt_want_write_file - get write access to a file's mount
404 * @file: the file who's mount on which to take a write
406 * This is like mnt_want_write, but it takes a file and can
407 * do some optimisations if the file is open for write already
409 int mnt_want_write_file(struct file *file)
413 sb_start_write(file_inode(file)->i_sb);
414 ret = __mnt_want_write_file(file);
416 sb_end_write(file_inode(file)->i_sb);
419 EXPORT_SYMBOL_GPL(mnt_want_write_file);
422 * __mnt_drop_write - give up write access to a mount
423 * @mnt: the mount on which to give up write access
425 * Tells the low-level filesystem that we are done
426 * performing writes to it. Must be matched with
427 * __mnt_want_write() call above.
429 void __mnt_drop_write(struct vfsmount *mnt)
432 mnt_dec_writers(real_mount(mnt));
437 * mnt_drop_write - give up write access to a mount
438 * @mnt: the mount on which to give up write access
440 * Tells the low-level filesystem that we are done performing writes to it and
441 * also allows filesystem to be frozen again. Must be matched with
442 * mnt_want_write() call above.
444 void mnt_drop_write(struct vfsmount *mnt)
446 __mnt_drop_write(mnt);
447 sb_end_write(mnt->mnt_sb);
449 EXPORT_SYMBOL_GPL(mnt_drop_write);
451 void __mnt_drop_write_file(struct file *file)
453 __mnt_drop_write(file->f_path.mnt);
456 void mnt_drop_write_file(struct file *file)
458 __mnt_drop_write_file(file);
459 sb_end_write(file_inode(file)->i_sb);
461 EXPORT_SYMBOL(mnt_drop_write_file);
463 static int mnt_make_readonly(struct mount *mnt)
468 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
470 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
471 * should be visible before we do.
476 * With writers on hold, if this value is zero, then there are
477 * definitely no active writers (although held writers may subsequently
478 * increment the count, they'll have to wait, and decrement it after
479 * seeing MNT_READONLY).
481 * It is OK to have counter incremented on one CPU and decremented on
482 * another: the sum will add up correctly. The danger would be when we
483 * sum up each counter, if we read a counter before it is incremented,
484 * but then read another CPU's count which it has been subsequently
485 * decremented from -- we would see more decrements than we should.
486 * MNT_WRITE_HOLD protects against this scenario, because
487 * mnt_want_write first increments count, then smp_mb, then spins on
488 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
489 * we're counting up here.
491 if (mnt_get_writers(mnt) > 0)
494 mnt->mnt.mnt_flags |= MNT_READONLY;
496 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
497 * that become unheld will see MNT_READONLY.
500 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
505 static int __mnt_unmake_readonly(struct mount *mnt)
508 mnt->mnt.mnt_flags &= ~MNT_READONLY;
513 int sb_prepare_remount_readonly(struct super_block *sb)
518 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
519 if (atomic_long_read(&sb->s_remove_count))
523 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
524 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
525 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
527 if (mnt_get_writers(mnt) > 0) {
533 if (!err && atomic_long_read(&sb->s_remove_count))
537 sb->s_readonly_remount = 1;
540 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
541 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
542 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
549 static void free_vfsmnt(struct mount *mnt)
551 struct user_namespace *mnt_userns;
553 mnt_userns = mnt_user_ns(&mnt->mnt);
554 if (mnt_userns != &init_user_ns)
555 put_user_ns(mnt_userns);
556 kfree_const(mnt->mnt_devname);
558 free_percpu(mnt->mnt_pcp);
560 kmem_cache_free(mnt_cache, mnt);
563 static void delayed_free_vfsmnt(struct rcu_head *head)
565 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
568 /* call under rcu_read_lock */
569 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
572 if (read_seqretry(&mount_lock, seq))
576 mnt = real_mount(bastard);
577 mnt_add_count(mnt, 1);
578 smp_mb(); // see mntput_no_expire()
579 if (likely(!read_seqretry(&mount_lock, seq)))
581 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
582 mnt_add_count(mnt, -1);
586 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
587 mnt_add_count(mnt, -1);
592 /* caller will mntput() */
596 /* call under rcu_read_lock */
597 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
599 int res = __legitimize_mnt(bastard, seq);
602 if (unlikely(res < 0)) {
611 * find the first mount at @dentry on vfsmount @mnt.
612 * call under rcu_read_lock()
614 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
616 struct hlist_head *head = m_hash(mnt, dentry);
619 hlist_for_each_entry_rcu(p, head, mnt_hash)
620 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
626 * lookup_mnt - Return the first child mount mounted at path
628 * "First" means first mounted chronologically. If you create the
631 * mount /dev/sda1 /mnt
632 * mount /dev/sda2 /mnt
633 * mount /dev/sda3 /mnt
635 * Then lookup_mnt() on the base /mnt dentry in the root mount will
636 * return successively the root dentry and vfsmount of /dev/sda1, then
637 * /dev/sda2, then /dev/sda3, then NULL.
639 * lookup_mnt takes a reference to the found vfsmount.
641 struct vfsmount *lookup_mnt(const struct path *path)
643 struct mount *child_mnt;
649 seq = read_seqbegin(&mount_lock);
650 child_mnt = __lookup_mnt(path->mnt, path->dentry);
651 m = child_mnt ? &child_mnt->mnt : NULL;
652 } while (!legitimize_mnt(m, seq));
657 static inline void lock_ns_list(struct mnt_namespace *ns)
659 spin_lock(&ns->ns_lock);
662 static inline void unlock_ns_list(struct mnt_namespace *ns)
664 spin_unlock(&ns->ns_lock);
667 static inline bool mnt_is_cursor(struct mount *mnt)
669 return mnt->mnt.mnt_flags & MNT_CURSOR;
673 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
674 * current mount namespace.
676 * The common case is dentries are not mountpoints at all and that
677 * test is handled inline. For the slow case when we are actually
678 * dealing with a mountpoint of some kind, walk through all of the
679 * mounts in the current mount namespace and test to see if the dentry
682 * The mount_hashtable is not usable in the context because we
683 * need to identify all mounts that may be in the current mount
684 * namespace not just a mount that happens to have some specified
687 bool __is_local_mountpoint(struct dentry *dentry)
689 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
691 bool is_covered = false;
693 down_read(&namespace_sem);
695 list_for_each_entry(mnt, &ns->list, mnt_list) {
696 if (mnt_is_cursor(mnt))
698 is_covered = (mnt->mnt_mountpoint == dentry);
703 up_read(&namespace_sem);
708 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
710 struct hlist_head *chain = mp_hash(dentry);
711 struct mountpoint *mp;
713 hlist_for_each_entry(mp, chain, m_hash) {
714 if (mp->m_dentry == dentry) {
722 static struct mountpoint *get_mountpoint(struct dentry *dentry)
724 struct mountpoint *mp, *new = NULL;
727 if (d_mountpoint(dentry)) {
728 /* might be worth a WARN_ON() */
729 if (d_unlinked(dentry))
730 return ERR_PTR(-ENOENT);
732 read_seqlock_excl(&mount_lock);
733 mp = lookup_mountpoint(dentry);
734 read_sequnlock_excl(&mount_lock);
740 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
742 return ERR_PTR(-ENOMEM);
745 /* Exactly one processes may set d_mounted */
746 ret = d_set_mounted(dentry);
748 /* Someone else set d_mounted? */
752 /* The dentry is not available as a mountpoint? */
757 /* Add the new mountpoint to the hash table */
758 read_seqlock_excl(&mount_lock);
759 new->m_dentry = dget(dentry);
761 hlist_add_head(&new->m_hash, mp_hash(dentry));
762 INIT_HLIST_HEAD(&new->m_list);
763 read_sequnlock_excl(&mount_lock);
773 * vfsmount lock must be held. Additionally, the caller is responsible
774 * for serializing calls for given disposal list.
776 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
778 if (!--mp->m_count) {
779 struct dentry *dentry = mp->m_dentry;
780 BUG_ON(!hlist_empty(&mp->m_list));
781 spin_lock(&dentry->d_lock);
782 dentry->d_flags &= ~DCACHE_MOUNTED;
783 spin_unlock(&dentry->d_lock);
784 dput_to_list(dentry, list);
785 hlist_del(&mp->m_hash);
790 /* called with namespace_lock and vfsmount lock */
791 static void put_mountpoint(struct mountpoint *mp)
793 __put_mountpoint(mp, &ex_mountpoints);
796 static inline int check_mnt(struct mount *mnt)
798 return mnt->mnt_ns == current->nsproxy->mnt_ns;
802 * vfsmount lock must be held for write
804 static void touch_mnt_namespace(struct mnt_namespace *ns)
808 wake_up_interruptible(&ns->poll);
813 * vfsmount lock must be held for write
815 static void __touch_mnt_namespace(struct mnt_namespace *ns)
817 if (ns && ns->event != event) {
819 wake_up_interruptible(&ns->poll);
824 * vfsmount lock must be held for write
826 static struct mountpoint *unhash_mnt(struct mount *mnt)
828 struct mountpoint *mp;
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);
840 * vfsmount lock must be held for write
842 static void umount_mnt(struct mount *mnt)
844 put_mountpoint(unhash_mnt(mnt));
848 * vfsmount lock must be held for write
850 void mnt_set_mountpoint(struct mount *mnt,
851 struct mountpoint *mp,
852 struct mount *child_mnt)
855 mnt_add_count(mnt, 1); /* essentially, that's mntget */
856 child_mnt->mnt_mountpoint = mp->m_dentry;
857 child_mnt->mnt_parent = mnt;
858 child_mnt->mnt_mp = mp;
859 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
862 static void __attach_mnt(struct mount *mnt, struct mount *parent)
864 hlist_add_head_rcu(&mnt->mnt_hash,
865 m_hash(&parent->mnt, mnt->mnt_mountpoint));
866 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
870 * vfsmount lock must be held for write
872 static void attach_mnt(struct mount *mnt,
873 struct mount *parent,
874 struct mountpoint *mp)
876 mnt_set_mountpoint(parent, mp, mnt);
877 __attach_mnt(mnt, parent);
880 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
882 struct mountpoint *old_mp = mnt->mnt_mp;
883 struct mount *old_parent = mnt->mnt_parent;
885 list_del_init(&mnt->mnt_child);
886 hlist_del_init(&mnt->mnt_mp_list);
887 hlist_del_init_rcu(&mnt->mnt_hash);
889 attach_mnt(mnt, parent, mp);
891 put_mountpoint(old_mp);
892 mnt_add_count(old_parent, -1);
896 * vfsmount lock must be held for write
898 static void commit_tree(struct mount *mnt)
900 struct mount *parent = mnt->mnt_parent;
903 struct mnt_namespace *n = parent->mnt_ns;
905 BUG_ON(parent == mnt);
907 list_add_tail(&head, &mnt->mnt_list);
908 list_for_each_entry(m, &head, mnt_list)
911 list_splice(&head, n->list.prev);
913 n->mounts += n->pending_mounts;
914 n->pending_mounts = 0;
916 __attach_mnt(mnt, parent);
917 touch_mnt_namespace(n);
920 static struct mount *next_mnt(struct mount *p, struct mount *root)
922 struct list_head *next = p->mnt_mounts.next;
923 if (next == &p->mnt_mounts) {
927 next = p->mnt_child.next;
928 if (next != &p->mnt_parent->mnt_mounts)
933 return list_entry(next, struct mount, mnt_child);
936 static struct mount *skip_mnt_tree(struct mount *p)
938 struct list_head *prev = p->mnt_mounts.prev;
939 while (prev != &p->mnt_mounts) {
940 p = list_entry(prev, struct mount, mnt_child);
941 prev = p->mnt_mounts.prev;
947 * vfs_create_mount - Create a mount for a configured superblock
948 * @fc: The configuration context with the superblock attached
950 * Create a mount to an already configured superblock. If necessary, the
951 * caller should invoke vfs_get_tree() before calling this.
953 * Note that this does not attach the mount to anything.
955 struct vfsmount *vfs_create_mount(struct fs_context *fc)
960 return ERR_PTR(-EINVAL);
962 mnt = alloc_vfsmnt(fc->source ?: "none");
964 return ERR_PTR(-ENOMEM);
966 if (fc->sb_flags & SB_KERNMOUNT)
967 mnt->mnt.mnt_flags = MNT_INTERNAL;
969 atomic_inc(&fc->root->d_sb->s_active);
970 mnt->mnt.mnt_sb = fc->root->d_sb;
971 mnt->mnt.mnt_root = dget(fc->root);
972 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
973 mnt->mnt_parent = mnt;
976 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
980 EXPORT_SYMBOL(vfs_create_mount);
982 struct vfsmount *fc_mount(struct fs_context *fc)
984 int err = vfs_get_tree(fc);
986 up_write(&fc->root->d_sb->s_umount);
987 return vfs_create_mount(fc);
991 EXPORT_SYMBOL(fc_mount);
993 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
994 int flags, const char *name,
997 struct fs_context *fc;
998 struct vfsmount *mnt;
1002 return ERR_PTR(-EINVAL);
1004 fc = fs_context_for_mount(type, flags);
1006 return ERR_CAST(fc);
1009 ret = vfs_parse_fs_string(fc, "source",
1010 name, strlen(name));
1012 ret = parse_monolithic_mount_data(fc, data);
1021 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1024 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1025 const char *name, void *data)
1027 /* Until it is worked out how to pass the user namespace
1028 * through from the parent mount to the submount don't support
1029 * unprivileged mounts with submounts.
1031 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1032 return ERR_PTR(-EPERM);
1034 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1036 EXPORT_SYMBOL_GPL(vfs_submount);
1038 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1041 struct super_block *sb = old->mnt.mnt_sb;
1045 mnt = alloc_vfsmnt(old->mnt_devname);
1047 return ERR_PTR(-ENOMEM);
1049 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1050 mnt->mnt_group_id = 0; /* not a peer of original */
1052 mnt->mnt_group_id = old->mnt_group_id;
1054 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1055 err = mnt_alloc_group_id(mnt);
1060 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1061 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1063 atomic_inc(&sb->s_active);
1064 mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1065 if (mnt->mnt.mnt_userns != &init_user_ns)
1066 mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1067 mnt->mnt.mnt_sb = sb;
1068 mnt->mnt.mnt_root = dget(root);
1069 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1070 mnt->mnt_parent = mnt;
1072 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1073 unlock_mount_hash();
1075 if ((flag & CL_SLAVE) ||
1076 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1077 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1078 mnt->mnt_master = old;
1079 CLEAR_MNT_SHARED(mnt);
1080 } else if (!(flag & CL_PRIVATE)) {
1081 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1082 list_add(&mnt->mnt_share, &old->mnt_share);
1083 if (IS_MNT_SLAVE(old))
1084 list_add(&mnt->mnt_slave, &old->mnt_slave);
1085 mnt->mnt_master = old->mnt_master;
1087 CLEAR_MNT_SHARED(mnt);
1089 if (flag & CL_MAKE_SHARED)
1090 set_mnt_shared(mnt);
1092 /* stick the duplicate mount on the same expiry list
1093 * as the original if that was on one */
1094 if (flag & CL_EXPIRE) {
1095 if (!list_empty(&old->mnt_expire))
1096 list_add(&mnt->mnt_expire, &old->mnt_expire);
1104 return ERR_PTR(err);
1107 static void cleanup_mnt(struct mount *mnt)
1109 struct hlist_node *p;
1112 * The warning here probably indicates that somebody messed
1113 * up a mnt_want/drop_write() pair. If this happens, the
1114 * filesystem was probably unable to make r/w->r/o transitions.
1115 * The locking used to deal with mnt_count decrement provides barriers,
1116 * so mnt_get_writers() below is safe.
1118 WARN_ON(mnt_get_writers(mnt));
1119 if (unlikely(mnt->mnt_pins.first))
1121 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1122 hlist_del(&m->mnt_umount);
1125 fsnotify_vfsmount_delete(&mnt->mnt);
1126 dput(mnt->mnt.mnt_root);
1127 deactivate_super(mnt->mnt.mnt_sb);
1129 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1132 static void __cleanup_mnt(struct rcu_head *head)
1134 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1137 static LLIST_HEAD(delayed_mntput_list);
1138 static void delayed_mntput(struct work_struct *unused)
1140 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1141 struct mount *m, *t;
1143 llist_for_each_entry_safe(m, t, node, mnt_llist)
1146 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1148 static void mntput_no_expire(struct mount *mnt)
1154 if (likely(READ_ONCE(mnt->mnt_ns))) {
1156 * Since we don't do lock_mount_hash() here,
1157 * ->mnt_ns can change under us. However, if it's
1158 * non-NULL, then there's a reference that won't
1159 * be dropped until after an RCU delay done after
1160 * turning ->mnt_ns NULL. So if we observe it
1161 * non-NULL under rcu_read_lock(), the reference
1162 * we are dropping is not the final one.
1164 mnt_add_count(mnt, -1);
1170 * make sure that if __legitimize_mnt() has not seen us grab
1171 * mount_lock, we'll see their refcount increment here.
1174 mnt_add_count(mnt, -1);
1175 count = mnt_get_count(mnt);
1179 unlock_mount_hash();
1182 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1184 unlock_mount_hash();
1187 mnt->mnt.mnt_flags |= MNT_DOOMED;
1190 list_del(&mnt->mnt_instance);
1192 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1193 struct mount *p, *tmp;
1194 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1195 __put_mountpoint(unhash_mnt(p), &list);
1196 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1199 unlock_mount_hash();
1200 shrink_dentry_list(&list);
1202 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1203 struct task_struct *task = current;
1204 if (likely(!(task->flags & PF_KTHREAD))) {
1205 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1206 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1209 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1210 schedule_delayed_work(&delayed_mntput_work, 1);
1216 void mntput(struct vfsmount *mnt)
1219 struct mount *m = real_mount(mnt);
1220 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1221 if (unlikely(m->mnt_expiry_mark))
1222 m->mnt_expiry_mark = 0;
1223 mntput_no_expire(m);
1226 EXPORT_SYMBOL(mntput);
1228 struct vfsmount *mntget(struct vfsmount *mnt)
1231 mnt_add_count(real_mount(mnt), 1);
1234 EXPORT_SYMBOL(mntget);
1236 /* path_is_mountpoint() - Check if path is a mount in the current
1239 * d_mountpoint() can only be used reliably to establish if a dentry is
1240 * not mounted in any namespace and that common case is handled inline.
1241 * d_mountpoint() isn't aware of the possibility there may be multiple
1242 * mounts using a given dentry in a different namespace. This function
1243 * checks if the passed in path is a mountpoint rather than the dentry
1246 bool path_is_mountpoint(const struct path *path)
1251 if (!d_mountpoint(path->dentry))
1256 seq = read_seqbegin(&mount_lock);
1257 res = __path_is_mountpoint(path);
1258 } while (read_seqretry(&mount_lock, seq));
1263 EXPORT_SYMBOL(path_is_mountpoint);
1265 struct vfsmount *mnt_clone_internal(const struct path *path)
1268 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1271 p->mnt.mnt_flags |= MNT_INTERNAL;
1275 #ifdef CONFIG_PROC_FS
1276 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1277 struct list_head *p)
1279 struct mount *mnt, *ret = NULL;
1282 list_for_each_continue(p, &ns->list) {
1283 mnt = list_entry(p, typeof(*mnt), mnt_list);
1284 if (!mnt_is_cursor(mnt)) {
1294 /* iterator; we want it to have access to namespace_sem, thus here... */
1295 static void *m_start(struct seq_file *m, loff_t *pos)
1297 struct proc_mounts *p = m->private;
1298 struct list_head *prev;
1300 down_read(&namespace_sem);
1302 prev = &p->ns->list;
1304 prev = &p->cursor.mnt_list;
1306 /* Read after we'd reached the end? */
1307 if (list_empty(prev))
1311 return mnt_list_next(p->ns, prev);
1314 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1316 struct proc_mounts *p = m->private;
1317 struct mount *mnt = v;
1320 return mnt_list_next(p->ns, &mnt->mnt_list);
1323 static void m_stop(struct seq_file *m, void *v)
1325 struct proc_mounts *p = m->private;
1326 struct mount *mnt = v;
1328 lock_ns_list(p->ns);
1330 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1332 list_del_init(&p->cursor.mnt_list);
1333 unlock_ns_list(p->ns);
1334 up_read(&namespace_sem);
1337 static int m_show(struct seq_file *m, void *v)
1339 struct proc_mounts *p = m->private;
1340 struct mount *r = v;
1341 return p->show(m, &r->mnt);
1344 const struct seq_operations mounts_op = {
1351 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1353 down_read(&namespace_sem);
1355 list_del(&cursor->mnt_list);
1357 up_read(&namespace_sem);
1359 #endif /* CONFIG_PROC_FS */
1362 * may_umount_tree - check if a mount tree is busy
1363 * @mnt: root of mount tree
1365 * This is called to check if a tree of mounts has any
1366 * open files, pwds, chroots or sub mounts that are
1369 int may_umount_tree(struct vfsmount *m)
1371 struct mount *mnt = real_mount(m);
1372 int actual_refs = 0;
1373 int minimum_refs = 0;
1377 /* write lock needed for mnt_get_count */
1379 for (p = mnt; p; p = next_mnt(p, mnt)) {
1380 actual_refs += mnt_get_count(p);
1383 unlock_mount_hash();
1385 if (actual_refs > minimum_refs)
1391 EXPORT_SYMBOL(may_umount_tree);
1394 * may_umount - check if a mount point is busy
1395 * @mnt: root of mount
1397 * This is called to check if a mount point has any
1398 * open files, pwds, chroots or sub mounts. If the
1399 * mount has sub mounts this will return busy
1400 * regardless of whether the sub mounts are busy.
1402 * Doesn't take quota and stuff into account. IOW, in some cases it will
1403 * give false negatives. The main reason why it's here is that we need
1404 * a non-destructive way to look for easily umountable filesystems.
1406 int may_umount(struct vfsmount *mnt)
1409 down_read(&namespace_sem);
1411 if (propagate_mount_busy(real_mount(mnt), 2))
1413 unlock_mount_hash();
1414 up_read(&namespace_sem);
1418 EXPORT_SYMBOL(may_umount);
1420 static void namespace_unlock(void)
1422 struct hlist_head head;
1423 struct hlist_node *p;
1427 hlist_move_list(&unmounted, &head);
1428 list_splice_init(&ex_mountpoints, &list);
1430 up_write(&namespace_sem);
1432 shrink_dentry_list(&list);
1434 if (likely(hlist_empty(&head)))
1437 synchronize_rcu_expedited();
1439 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1440 hlist_del(&m->mnt_umount);
1445 static inline void namespace_lock(void)
1447 down_write(&namespace_sem);
1450 enum umount_tree_flags {
1452 UMOUNT_PROPAGATE = 2,
1453 UMOUNT_CONNECTED = 4,
1456 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1458 /* Leaving mounts connected is only valid for lazy umounts */
1459 if (how & UMOUNT_SYNC)
1462 /* A mount without a parent has nothing to be connected to */
1463 if (!mnt_has_parent(mnt))
1466 /* Because the reference counting rules change when mounts are
1467 * unmounted and connected, umounted mounts may not be
1468 * connected to mounted mounts.
1470 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1473 /* Has it been requested that the mount remain connected? */
1474 if (how & UMOUNT_CONNECTED)
1477 /* Is the mount locked such that it needs to remain connected? */
1478 if (IS_MNT_LOCKED(mnt))
1481 /* By default disconnect the mount */
1486 * mount_lock must be held
1487 * namespace_sem must be held for write
1489 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1491 LIST_HEAD(tmp_list);
1494 if (how & UMOUNT_PROPAGATE)
1495 propagate_mount_unlock(mnt);
1497 /* Gather the mounts to umount */
1498 for (p = mnt; p; p = next_mnt(p, mnt)) {
1499 p->mnt.mnt_flags |= MNT_UMOUNT;
1500 list_move(&p->mnt_list, &tmp_list);
1503 /* Hide the mounts from mnt_mounts */
1504 list_for_each_entry(p, &tmp_list, mnt_list) {
1505 list_del_init(&p->mnt_child);
1508 /* Add propogated mounts to the tmp_list */
1509 if (how & UMOUNT_PROPAGATE)
1510 propagate_umount(&tmp_list);
1512 while (!list_empty(&tmp_list)) {
1513 struct mnt_namespace *ns;
1515 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1516 list_del_init(&p->mnt_expire);
1517 list_del_init(&p->mnt_list);
1521 __touch_mnt_namespace(ns);
1524 if (how & UMOUNT_SYNC)
1525 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1527 disconnect = disconnect_mount(p, how);
1528 if (mnt_has_parent(p)) {
1529 mnt_add_count(p->mnt_parent, -1);
1531 /* Don't forget about p */
1532 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1537 change_mnt_propagation(p, MS_PRIVATE);
1539 hlist_add_head(&p->mnt_umount, &unmounted);
1543 static void shrink_submounts(struct mount *mnt);
1545 static int do_umount_root(struct super_block *sb)
1549 down_write(&sb->s_umount);
1550 if (!sb_rdonly(sb)) {
1551 struct fs_context *fc;
1553 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1558 ret = parse_monolithic_mount_data(fc, NULL);
1560 ret = reconfigure_super(fc);
1564 up_write(&sb->s_umount);
1568 static int do_umount(struct mount *mnt, int flags)
1570 struct super_block *sb = mnt->mnt.mnt_sb;
1573 retval = security_sb_umount(&mnt->mnt, flags);
1578 * Allow userspace to request a mountpoint be expired rather than
1579 * unmounting unconditionally. Unmount only happens if:
1580 * (1) the mark is already set (the mark is cleared by mntput())
1581 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1583 if (flags & MNT_EXPIRE) {
1584 if (&mnt->mnt == current->fs->root.mnt ||
1585 flags & (MNT_FORCE | MNT_DETACH))
1589 * probably don't strictly need the lock here if we examined
1590 * all race cases, but it's a slowpath.
1593 if (mnt_get_count(mnt) != 2) {
1594 unlock_mount_hash();
1597 unlock_mount_hash();
1599 if (!xchg(&mnt->mnt_expiry_mark, 1))
1604 * If we may have to abort operations to get out of this
1605 * mount, and they will themselves hold resources we must
1606 * allow the fs to do things. In the Unix tradition of
1607 * 'Gee thats tricky lets do it in userspace' the umount_begin
1608 * might fail to complete on the first run through as other tasks
1609 * must return, and the like. Thats for the mount program to worry
1610 * about for the moment.
1613 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1614 sb->s_op->umount_begin(sb);
1618 * No sense to grab the lock for this test, but test itself looks
1619 * somewhat bogus. Suggestions for better replacement?
1620 * Ho-hum... In principle, we might treat that as umount + switch
1621 * to rootfs. GC would eventually take care of the old vfsmount.
1622 * Actually it makes sense, especially if rootfs would contain a
1623 * /reboot - static binary that would close all descriptors and
1624 * call reboot(9). Then init(8) could umount root and exec /reboot.
1626 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1628 * Special case for "unmounting" root ...
1629 * we just try to remount it readonly.
1631 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1633 return do_umount_root(sb);
1639 /* Recheck MNT_LOCKED with the locks held */
1641 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1645 if (flags & MNT_DETACH) {
1646 if (!list_empty(&mnt->mnt_list))
1647 umount_tree(mnt, UMOUNT_PROPAGATE);
1650 shrink_submounts(mnt);
1652 if (!propagate_mount_busy(mnt, 2)) {
1653 if (!list_empty(&mnt->mnt_list))
1654 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1659 unlock_mount_hash();
1665 * __detach_mounts - lazily unmount all mounts on the specified dentry
1667 * During unlink, rmdir, and d_drop it is possible to loose the path
1668 * to an existing mountpoint, and wind up leaking the mount.
1669 * detach_mounts allows lazily unmounting those mounts instead of
1672 * The caller may hold dentry->d_inode->i_mutex.
1674 void __detach_mounts(struct dentry *dentry)
1676 struct mountpoint *mp;
1681 mp = lookup_mountpoint(dentry);
1686 while (!hlist_empty(&mp->m_list)) {
1687 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1688 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1690 hlist_add_head(&mnt->mnt_umount, &unmounted);
1692 else umount_tree(mnt, UMOUNT_CONNECTED);
1696 unlock_mount_hash();
1701 * Is the caller allowed to modify his namespace?
1703 static inline bool may_mount(void)
1705 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1708 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1709 static inline bool may_mandlock(void)
1711 return capable(CAP_SYS_ADMIN);
1714 static inline bool may_mandlock(void)
1716 pr_warn("VFS: \"mand\" mount option not supported");
1721 static int can_umount(const struct path *path, int flags)
1723 struct mount *mnt = real_mount(path->mnt);
1727 if (path->dentry != path->mnt->mnt_root)
1729 if (!check_mnt(mnt))
1731 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1733 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1738 // caller is responsible for flags being sane
1739 int path_umount(struct path *path, int flags)
1741 struct mount *mnt = real_mount(path->mnt);
1744 ret = can_umount(path, flags);
1746 ret = do_umount(mnt, flags);
1748 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1750 mntput_no_expire(mnt);
1754 static int ksys_umount(char __user *name, int flags)
1756 int lookup_flags = LOOKUP_MOUNTPOINT;
1760 // basic validity checks done first
1761 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1764 if (!(flags & UMOUNT_NOFOLLOW))
1765 lookup_flags |= LOOKUP_FOLLOW;
1766 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1769 return path_umount(&path, flags);
1772 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1774 return ksys_umount(name, flags);
1777 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1780 * The 2.0 compatible umount. No flags.
1782 SYSCALL_DEFINE1(oldumount, char __user *, name)
1784 return ksys_umount(name, 0);
1789 static bool is_mnt_ns_file(struct dentry *dentry)
1791 /* Is this a proxy for a mount namespace? */
1792 return dentry->d_op == &ns_dentry_operations &&
1793 dentry->d_fsdata == &mntns_operations;
1796 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1798 return container_of(ns, struct mnt_namespace, ns);
1801 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1806 static bool mnt_ns_loop(struct dentry *dentry)
1808 /* Could bind mounting the mount namespace inode cause a
1809 * mount namespace loop?
1811 struct mnt_namespace *mnt_ns;
1812 if (!is_mnt_ns_file(dentry))
1815 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1816 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1819 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1822 struct mount *res, *p, *q, *r, *parent;
1824 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1825 return ERR_PTR(-EINVAL);
1827 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1828 return ERR_PTR(-EINVAL);
1830 res = q = clone_mnt(mnt, dentry, flag);
1834 q->mnt_mountpoint = mnt->mnt_mountpoint;
1837 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1839 if (!is_subdir(r->mnt_mountpoint, dentry))
1842 for (s = r; s; s = next_mnt(s, r)) {
1843 if (!(flag & CL_COPY_UNBINDABLE) &&
1844 IS_MNT_UNBINDABLE(s)) {
1845 if (s->mnt.mnt_flags & MNT_LOCKED) {
1846 /* Both unbindable and locked. */
1847 q = ERR_PTR(-EPERM);
1850 s = skip_mnt_tree(s);
1854 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1855 is_mnt_ns_file(s->mnt.mnt_root)) {
1856 s = skip_mnt_tree(s);
1859 while (p != s->mnt_parent) {
1865 q = clone_mnt(p, p->mnt.mnt_root, flag);
1869 list_add_tail(&q->mnt_list, &res->mnt_list);
1870 attach_mnt(q, parent, p->mnt_mp);
1871 unlock_mount_hash();
1878 umount_tree(res, UMOUNT_SYNC);
1879 unlock_mount_hash();
1884 /* Caller should check returned pointer for errors */
1886 struct vfsmount *collect_mounts(const struct path *path)
1890 if (!check_mnt(real_mount(path->mnt)))
1891 tree = ERR_PTR(-EINVAL);
1893 tree = copy_tree(real_mount(path->mnt), path->dentry,
1894 CL_COPY_ALL | CL_PRIVATE);
1897 return ERR_CAST(tree);
1901 static void free_mnt_ns(struct mnt_namespace *);
1902 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1904 void dissolve_on_fput(struct vfsmount *mnt)
1906 struct mnt_namespace *ns;
1909 ns = real_mount(mnt)->mnt_ns;
1912 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1916 unlock_mount_hash();
1922 void drop_collected_mounts(struct vfsmount *mnt)
1926 umount_tree(real_mount(mnt), 0);
1927 unlock_mount_hash();
1932 * clone_private_mount - create a private clone of a path
1934 * This creates a new vfsmount, which will be the clone of @path. The new will
1935 * not be attached anywhere in the namespace and will be private (i.e. changes
1936 * to the originating mount won't be propagated into this).
1938 * Release with mntput().
1940 struct vfsmount *clone_private_mount(const struct path *path)
1942 struct mount *old_mnt = real_mount(path->mnt);
1943 struct mount *new_mnt;
1945 if (IS_MNT_UNBINDABLE(old_mnt))
1946 return ERR_PTR(-EINVAL);
1948 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1949 if (IS_ERR(new_mnt))
1950 return ERR_CAST(new_mnt);
1952 /* Longterm mount to be removed by kern_unmount*() */
1953 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1955 return &new_mnt->mnt;
1957 EXPORT_SYMBOL_GPL(clone_private_mount);
1959 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1960 struct vfsmount *root)
1963 int res = f(root, arg);
1966 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1967 res = f(&mnt->mnt, arg);
1974 static void lock_mnt_tree(struct mount *mnt)
1978 for (p = mnt; p; p = next_mnt(p, mnt)) {
1979 int flags = p->mnt.mnt_flags;
1980 /* Don't allow unprivileged users to change mount flags */
1981 flags |= MNT_LOCK_ATIME;
1983 if (flags & MNT_READONLY)
1984 flags |= MNT_LOCK_READONLY;
1986 if (flags & MNT_NODEV)
1987 flags |= MNT_LOCK_NODEV;
1989 if (flags & MNT_NOSUID)
1990 flags |= MNT_LOCK_NOSUID;
1992 if (flags & MNT_NOEXEC)
1993 flags |= MNT_LOCK_NOEXEC;
1994 /* Don't allow unprivileged users to reveal what is under a mount */
1995 if (list_empty(&p->mnt_expire))
1996 flags |= MNT_LOCKED;
1997 p->mnt.mnt_flags = flags;
2001 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2005 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2006 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2007 mnt_release_group_id(p);
2011 static int invent_group_ids(struct mount *mnt, bool recurse)
2015 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2016 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2017 int err = mnt_alloc_group_id(p);
2019 cleanup_group_ids(mnt, p);
2028 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2030 unsigned int max = READ_ONCE(sysctl_mount_max);
2031 unsigned int mounts = 0, old, pending, sum;
2034 for (p = mnt; p; p = next_mnt(p, mnt))
2038 pending = ns->pending_mounts;
2039 sum = old + pending;
2043 (mounts > (max - sum)))
2046 ns->pending_mounts = pending + mounts;
2051 * @source_mnt : mount tree to be attached
2052 * @nd : place the mount tree @source_mnt is attached
2053 * @parent_nd : if non-null, detach the source_mnt from its parent and
2054 * store the parent mount and mountpoint dentry.
2055 * (done when source_mnt is moved)
2057 * NOTE: in the table below explains the semantics when a source mount
2058 * of a given type is attached to a destination mount of a given type.
2059 * ---------------------------------------------------------------------------
2060 * | BIND MOUNT OPERATION |
2061 * |**************************************************************************
2062 * | source-->| shared | private | slave | unbindable |
2066 * |**************************************************************************
2067 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2069 * |non-shared| shared (+) | private | slave (*) | invalid |
2070 * ***************************************************************************
2071 * A bind operation clones the source mount and mounts the clone on the
2072 * destination mount.
2074 * (++) the cloned mount is propagated to all the mounts in the propagation
2075 * tree of the destination mount and the cloned mount is added to
2076 * the peer group of the source mount.
2077 * (+) the cloned mount is created under the destination mount and is marked
2078 * as shared. The cloned mount is added to the peer group of the source
2080 * (+++) the mount is propagated to all the mounts in the propagation tree
2081 * of the destination mount and the cloned mount is made slave
2082 * of the same master as that of the source mount. The cloned mount
2083 * is marked as 'shared and slave'.
2084 * (*) the cloned mount is made a slave of the same master as that of the
2087 * ---------------------------------------------------------------------------
2088 * | MOVE MOUNT OPERATION |
2089 * |**************************************************************************
2090 * | source-->| shared | private | slave | unbindable |
2094 * |**************************************************************************
2095 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2097 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2098 * ***************************************************************************
2100 * (+) the mount is moved to the destination. And is then propagated to
2101 * all the mounts in the propagation tree of the destination mount.
2102 * (+*) the mount is moved to the destination.
2103 * (+++) the mount is moved to the destination and is then propagated to
2104 * all the mounts belonging to the destination mount's propagation tree.
2105 * the mount is marked as 'shared and slave'.
2106 * (*) the mount continues to be a slave at the new location.
2108 * if the source mount is a tree, the operations explained above is
2109 * applied to each mount in the tree.
2110 * Must be called without spinlocks held, since this function can sleep
2113 static int attach_recursive_mnt(struct mount *source_mnt,
2114 struct mount *dest_mnt,
2115 struct mountpoint *dest_mp,
2118 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2119 HLIST_HEAD(tree_list);
2120 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2121 struct mountpoint *smp;
2122 struct mount *child, *p;
2123 struct hlist_node *n;
2126 /* Preallocate a mountpoint in case the new mounts need
2127 * to be tucked under other mounts.
2129 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2131 return PTR_ERR(smp);
2133 /* Is there space to add these mounts to the mount namespace? */
2135 err = count_mounts(ns, source_mnt);
2140 if (IS_MNT_SHARED(dest_mnt)) {
2141 err = invent_group_ids(source_mnt, true);
2144 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2147 goto out_cleanup_ids;
2148 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2154 unhash_mnt(source_mnt);
2155 attach_mnt(source_mnt, dest_mnt, dest_mp);
2156 touch_mnt_namespace(source_mnt->mnt_ns);
2158 if (source_mnt->mnt_ns) {
2159 /* move from anon - the caller will destroy */
2160 list_del_init(&source_mnt->mnt_ns->list);
2162 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2163 commit_tree(source_mnt);
2166 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2168 hlist_del_init(&child->mnt_hash);
2169 q = __lookup_mnt(&child->mnt_parent->mnt,
2170 child->mnt_mountpoint);
2172 mnt_change_mountpoint(child, smp, q);
2173 /* Notice when we are propagating across user namespaces */
2174 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2175 lock_mnt_tree(child);
2176 child->mnt.mnt_flags &= ~MNT_LOCKED;
2179 put_mountpoint(smp);
2180 unlock_mount_hash();
2185 while (!hlist_empty(&tree_list)) {
2186 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2187 child->mnt_parent->mnt_ns->pending_mounts = 0;
2188 umount_tree(child, UMOUNT_SYNC);
2190 unlock_mount_hash();
2191 cleanup_group_ids(source_mnt, NULL);
2193 ns->pending_mounts = 0;
2195 read_seqlock_excl(&mount_lock);
2196 put_mountpoint(smp);
2197 read_sequnlock_excl(&mount_lock);
2202 static struct mountpoint *lock_mount(struct path *path)
2204 struct vfsmount *mnt;
2205 struct dentry *dentry = path->dentry;
2207 inode_lock(dentry->d_inode);
2208 if (unlikely(cant_mount(dentry))) {
2209 inode_unlock(dentry->d_inode);
2210 return ERR_PTR(-ENOENT);
2213 mnt = lookup_mnt(path);
2215 struct mountpoint *mp = get_mountpoint(dentry);
2218 inode_unlock(dentry->d_inode);
2224 inode_unlock(path->dentry->d_inode);
2227 dentry = path->dentry = dget(mnt->mnt_root);
2231 static void unlock_mount(struct mountpoint *where)
2233 struct dentry *dentry = where->m_dentry;
2235 read_seqlock_excl(&mount_lock);
2236 put_mountpoint(where);
2237 read_sequnlock_excl(&mount_lock);
2240 inode_unlock(dentry->d_inode);
2243 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2245 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2248 if (d_is_dir(mp->m_dentry) !=
2249 d_is_dir(mnt->mnt.mnt_root))
2252 return attach_recursive_mnt(mnt, p, mp, false);
2256 * Sanity check the flags to change_mnt_propagation.
2259 static int flags_to_propagation_type(int ms_flags)
2261 int type = ms_flags & ~(MS_REC | MS_SILENT);
2263 /* Fail if any non-propagation flags are set */
2264 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2266 /* Only one propagation flag should be set */
2267 if (!is_power_of_2(type))
2273 * recursively change the type of the mountpoint.
2275 static int do_change_type(struct path *path, int ms_flags)
2278 struct mount *mnt = real_mount(path->mnt);
2279 int recurse = ms_flags & MS_REC;
2283 if (path->dentry != path->mnt->mnt_root)
2286 type = flags_to_propagation_type(ms_flags);
2291 if (type == MS_SHARED) {
2292 err = invent_group_ids(mnt, recurse);
2298 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2299 change_mnt_propagation(m, type);
2300 unlock_mount_hash();
2307 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2309 struct mount *child;
2310 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2311 if (!is_subdir(child->mnt_mountpoint, dentry))
2314 if (child->mnt.mnt_flags & MNT_LOCKED)
2320 static struct mount *__do_loopback(struct path *old_path, int recurse)
2322 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2324 if (IS_MNT_UNBINDABLE(old))
2327 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2330 if (!recurse && has_locked_children(old, old_path->dentry))
2334 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2336 mnt = clone_mnt(old, old_path->dentry, 0);
2339 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2345 * do loopback mount.
2347 static int do_loopback(struct path *path, const char *old_name,
2350 struct path old_path;
2351 struct mount *mnt = NULL, *parent;
2352 struct mountpoint *mp;
2354 if (!old_name || !*old_name)
2356 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2361 if (mnt_ns_loop(old_path.dentry))
2364 mp = lock_mount(path);
2370 parent = real_mount(path->mnt);
2371 if (!check_mnt(parent))
2374 mnt = __do_loopback(&old_path, recurse);
2380 err = graft_tree(mnt, parent, mp);
2383 umount_tree(mnt, UMOUNT_SYNC);
2384 unlock_mount_hash();
2389 path_put(&old_path);
2393 static struct file *open_detached_copy(struct path *path, bool recursive)
2395 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2396 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2397 struct mount *mnt, *p;
2401 return ERR_CAST(ns);
2404 mnt = __do_loopback(path, recursive);
2408 return ERR_CAST(mnt);
2412 for (p = mnt; p; p = next_mnt(p, mnt)) {
2417 list_add_tail(&ns->list, &mnt->mnt_list);
2419 unlock_mount_hash();
2423 path->mnt = &mnt->mnt;
2424 file = dentry_open(path, O_PATH, current_cred());
2426 dissolve_on_fput(path->mnt);
2428 file->f_mode |= FMODE_NEED_UNMOUNT;
2432 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2436 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2437 bool detached = flags & OPEN_TREE_CLONE;
2441 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2443 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2444 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2448 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2451 if (flags & AT_NO_AUTOMOUNT)
2452 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2453 if (flags & AT_SYMLINK_NOFOLLOW)
2454 lookup_flags &= ~LOOKUP_FOLLOW;
2455 if (flags & AT_EMPTY_PATH)
2456 lookup_flags |= LOOKUP_EMPTY;
2458 if (detached && !may_mount())
2461 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2465 error = user_path_at(dfd, filename, lookup_flags, &path);
2466 if (unlikely(error)) {
2467 file = ERR_PTR(error);
2470 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2472 file = dentry_open(&path, O_PATH, current_cred());
2477 return PTR_ERR(file);
2479 fd_install(fd, file);
2484 * Don't allow locked mount flags to be cleared.
2486 * No locks need to be held here while testing the various MNT_LOCK
2487 * flags because those flags can never be cleared once they are set.
2489 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2491 unsigned int fl = mnt->mnt.mnt_flags;
2493 if ((fl & MNT_LOCK_READONLY) &&
2494 !(mnt_flags & MNT_READONLY))
2497 if ((fl & MNT_LOCK_NODEV) &&
2498 !(mnt_flags & MNT_NODEV))
2501 if ((fl & MNT_LOCK_NOSUID) &&
2502 !(mnt_flags & MNT_NOSUID))
2505 if ((fl & MNT_LOCK_NOEXEC) &&
2506 !(mnt_flags & MNT_NOEXEC))
2509 if ((fl & MNT_LOCK_ATIME) &&
2510 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2516 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2518 bool readonly_request = (mnt_flags & MNT_READONLY);
2520 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2523 if (readonly_request)
2524 return mnt_make_readonly(mnt);
2526 return __mnt_unmake_readonly(mnt);
2530 * Update the user-settable attributes on a mount. The caller must hold
2531 * sb->s_umount for writing.
2533 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2536 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2537 mnt->mnt.mnt_flags = mnt_flags;
2538 touch_mnt_namespace(mnt->mnt_ns);
2539 unlock_mount_hash();
2542 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2544 struct super_block *sb = mnt->mnt_sb;
2546 if (!__mnt_is_readonly(mnt) &&
2547 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2548 char *buf = (char *)__get_free_page(GFP_KERNEL);
2549 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2552 time64_to_tm(sb->s_time_max, 0, &tm);
2554 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2556 is_mounted(mnt) ? "remounted" : "mounted",
2558 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2560 free_page((unsigned long)buf);
2565 * Handle reconfiguration of the mountpoint only without alteration of the
2566 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2569 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2571 struct super_block *sb = path->mnt->mnt_sb;
2572 struct mount *mnt = real_mount(path->mnt);
2575 if (!check_mnt(mnt))
2578 if (path->dentry != mnt->mnt.mnt_root)
2581 if (!can_change_locked_flags(mnt, mnt_flags))
2584 down_write(&sb->s_umount);
2585 ret = change_mount_ro_state(mnt, mnt_flags);
2587 set_mount_attributes(mnt, mnt_flags);
2588 up_write(&sb->s_umount);
2590 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2596 * change filesystem flags. dir should be a physical root of filesystem.
2597 * If you've mounted a non-root directory somewhere and want to do remount
2598 * on it - tough luck.
2600 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2601 int mnt_flags, void *data)
2604 struct super_block *sb = path->mnt->mnt_sb;
2605 struct mount *mnt = real_mount(path->mnt);
2606 struct fs_context *fc;
2608 if (!check_mnt(mnt))
2611 if (path->dentry != path->mnt->mnt_root)
2614 if (!can_change_locked_flags(mnt, mnt_flags))
2617 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2622 err = parse_monolithic_mount_data(fc, data);
2624 down_write(&sb->s_umount);
2626 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2627 err = reconfigure_super(fc);
2629 set_mount_attributes(mnt, mnt_flags);
2631 up_write(&sb->s_umount);
2634 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2640 static inline int tree_contains_unbindable(struct mount *mnt)
2643 for (p = mnt; p; p = next_mnt(p, mnt)) {
2644 if (IS_MNT_UNBINDABLE(p))
2651 * Check that there aren't references to earlier/same mount namespaces in the
2652 * specified subtree. Such references can act as pins for mount namespaces
2653 * that aren't checked by the mount-cycle checking code, thereby allowing
2654 * cycles to be made.
2656 static bool check_for_nsfs_mounts(struct mount *subtree)
2662 for (p = subtree; p; p = next_mnt(p, subtree))
2663 if (mnt_ns_loop(p->mnt.mnt_root))
2668 unlock_mount_hash();
2672 static int do_move_mount(struct path *old_path, struct path *new_path)
2674 struct mnt_namespace *ns;
2677 struct mount *parent;
2678 struct mountpoint *mp, *old_mp;
2682 mp = lock_mount(new_path);
2686 old = real_mount(old_path->mnt);
2687 p = real_mount(new_path->mnt);
2688 parent = old->mnt_parent;
2689 attached = mnt_has_parent(old);
2690 old_mp = old->mnt_mp;
2694 /* The mountpoint must be in our namespace. */
2698 /* The thing moved must be mounted... */
2699 if (!is_mounted(&old->mnt))
2702 /* ... and either ours or the root of anon namespace */
2703 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2706 if (old->mnt.mnt_flags & MNT_LOCKED)
2709 if (old_path->dentry != old_path->mnt->mnt_root)
2712 if (d_is_dir(new_path->dentry) !=
2713 d_is_dir(old_path->dentry))
2716 * Don't move a mount residing in a shared parent.
2718 if (attached && IS_MNT_SHARED(parent))
2721 * Don't move a mount tree containing unbindable mounts to a destination
2722 * mount which is shared.
2724 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2727 if (!check_for_nsfs_mounts(old))
2729 for (; mnt_has_parent(p); p = p->mnt_parent)
2733 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2738 /* if the mount is moved, it should no longer be expire
2740 list_del_init(&old->mnt_expire);
2742 put_mountpoint(old_mp);
2747 mntput_no_expire(parent);
2754 static int do_move_mount_old(struct path *path, const char *old_name)
2756 struct path old_path;
2759 if (!old_name || !*old_name)
2762 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2766 err = do_move_mount(&old_path, path);
2767 path_put(&old_path);
2772 * add a mount into a namespace's mount tree
2774 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2775 struct path *path, int mnt_flags)
2777 struct mount *parent = real_mount(path->mnt);
2779 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2781 if (unlikely(!check_mnt(parent))) {
2782 /* that's acceptable only for automounts done in private ns */
2783 if (!(mnt_flags & MNT_SHRINKABLE))
2785 /* ... and for those we'd better have mountpoint still alive */
2786 if (!parent->mnt_ns)
2790 /* Refuse the same filesystem on the same mount point */
2791 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2792 path->mnt->mnt_root == path->dentry)
2795 if (d_is_symlink(newmnt->mnt.mnt_root))
2798 newmnt->mnt.mnt_flags = mnt_flags;
2799 return graft_tree(newmnt, parent, mp);
2802 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2805 * Create a new mount using a superblock configuration and request it
2806 * be added to the namespace tree.
2808 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2809 unsigned int mnt_flags)
2811 struct vfsmount *mnt;
2812 struct mountpoint *mp;
2813 struct super_block *sb = fc->root->d_sb;
2816 error = security_sb_kern_mount(sb);
2817 if (!error && mount_too_revealing(sb, &mnt_flags))
2820 if (unlikely(error)) {
2825 up_write(&sb->s_umount);
2827 mnt = vfs_create_mount(fc);
2829 return PTR_ERR(mnt);
2831 mnt_warn_timestamp_expiry(mountpoint, mnt);
2833 mp = lock_mount(mountpoint);
2838 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2846 * create a new mount for userspace and request it to be added into the
2849 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2850 int mnt_flags, const char *name, void *data)
2852 struct file_system_type *type;
2853 struct fs_context *fc;
2854 const char *subtype = NULL;
2860 type = get_fs_type(fstype);
2864 if (type->fs_flags & FS_HAS_SUBTYPE) {
2865 subtype = strchr(fstype, '.');
2869 put_filesystem(type);
2875 fc = fs_context_for_mount(type, sb_flags);
2876 put_filesystem(type);
2881 err = vfs_parse_fs_string(fc, "subtype",
2882 subtype, strlen(subtype));
2884 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2886 err = parse_monolithic_mount_data(fc, data);
2887 if (!err && !mount_capable(fc))
2890 err = vfs_get_tree(fc);
2892 err = do_new_mount_fc(fc, path, mnt_flags);
2898 int finish_automount(struct vfsmount *m, struct path *path)
2900 struct dentry *dentry = path->dentry;
2901 struct mountpoint *mp;
2910 mnt = real_mount(m);
2911 /* The new mount record should have at least 2 refs to prevent it being
2912 * expired before we get a chance to add it
2914 BUG_ON(mnt_get_count(mnt) < 2);
2916 if (m->mnt_sb == path->mnt->mnt_sb &&
2917 m->mnt_root == dentry) {
2923 * we don't want to use lock_mount() - in this case finding something
2924 * that overmounts our mountpoint to be means "quitely drop what we've
2925 * got", not "try to mount it on top".
2927 inode_lock(dentry->d_inode);
2929 if (unlikely(cant_mount(dentry))) {
2931 goto discard_locked;
2934 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2937 goto discard_locked;
2940 mp = get_mountpoint(dentry);
2943 goto discard_locked;
2946 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2955 inode_unlock(dentry->d_inode);
2957 /* remove m from any expiration list it may be on */
2958 if (!list_empty(&mnt->mnt_expire)) {
2960 list_del_init(&mnt->mnt_expire);
2969 * mnt_set_expiry - Put a mount on an expiration list
2970 * @mnt: The mount to list.
2971 * @expiry_list: The list to add the mount to.
2973 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2977 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2981 EXPORT_SYMBOL(mnt_set_expiry);
2984 * process a list of expirable mountpoints with the intent of discarding any
2985 * mountpoints that aren't in use and haven't been touched since last we came
2988 void mark_mounts_for_expiry(struct list_head *mounts)
2990 struct mount *mnt, *next;
2991 LIST_HEAD(graveyard);
2993 if (list_empty(mounts))
2999 /* extract from the expiration list every vfsmount that matches the
3000 * following criteria:
3001 * - only referenced by its parent vfsmount
3002 * - still marked for expiry (marked on the last call here; marks are
3003 * cleared by mntput())
3005 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3006 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3007 propagate_mount_busy(mnt, 1))
3009 list_move(&mnt->mnt_expire, &graveyard);
3011 while (!list_empty(&graveyard)) {
3012 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3013 touch_mnt_namespace(mnt->mnt_ns);
3014 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3016 unlock_mount_hash();
3020 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3023 * Ripoff of 'select_parent()'
3025 * search the list of submounts for a given mountpoint, and move any
3026 * shrinkable submounts to the 'graveyard' list.
3028 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3030 struct mount *this_parent = parent;
3031 struct list_head *next;
3035 next = this_parent->mnt_mounts.next;
3037 while (next != &this_parent->mnt_mounts) {
3038 struct list_head *tmp = next;
3039 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3042 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3045 * Descend a level if the d_mounts list is non-empty.
3047 if (!list_empty(&mnt->mnt_mounts)) {
3052 if (!propagate_mount_busy(mnt, 1)) {
3053 list_move_tail(&mnt->mnt_expire, graveyard);
3058 * All done at this level ... ascend and resume the search
3060 if (this_parent != parent) {
3061 next = this_parent->mnt_child.next;
3062 this_parent = this_parent->mnt_parent;
3069 * process a list of expirable mountpoints with the intent of discarding any
3070 * submounts of a specific parent mountpoint
3072 * mount_lock must be held for write
3074 static void shrink_submounts(struct mount *mnt)
3076 LIST_HEAD(graveyard);
3079 /* extract submounts of 'mountpoint' from the expiration list */
3080 while (select_submounts(mnt, &graveyard)) {
3081 while (!list_empty(&graveyard)) {
3082 m = list_first_entry(&graveyard, struct mount,
3084 touch_mnt_namespace(m->mnt_ns);
3085 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3090 static void *copy_mount_options(const void __user * data)
3093 unsigned left, offset;
3098 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3100 return ERR_PTR(-ENOMEM);
3102 left = copy_from_user(copy, data, PAGE_SIZE);
3105 * Not all architectures have an exact copy_from_user(). Resort to
3108 offset = PAGE_SIZE - left;
3111 if (get_user(c, (const char __user *)data + offset))
3118 if (left == PAGE_SIZE) {
3120 return ERR_PTR(-EFAULT);
3126 static char *copy_mount_string(const void __user *data)
3128 return data ? strndup_user(data, PATH_MAX) : NULL;
3132 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3133 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3135 * data is a (void *) that can point to any structure up to
3136 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3137 * information (or be NULL).
3139 * Pre-0.97 versions of mount() didn't have a flags word.
3140 * When the flags word was introduced its top half was required
3141 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3142 * Therefore, if this magic number is present, it carries no information
3143 * and must be discarded.
3145 int path_mount(const char *dev_name, struct path *path,
3146 const char *type_page, unsigned long flags, void *data_page)
3148 unsigned int mnt_flags = 0, sb_flags;
3152 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3153 flags &= ~MS_MGC_MSK;
3155 /* Basic sanity checks */
3157 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3159 if (flags & MS_NOUSER)
3162 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3167 if ((flags & SB_MANDLOCK) && !may_mandlock())
3170 /* Default to relatime unless overriden */
3171 if (!(flags & MS_NOATIME))
3172 mnt_flags |= MNT_RELATIME;
3174 /* Separate the per-mountpoint flags */
3175 if (flags & MS_NOSUID)
3176 mnt_flags |= MNT_NOSUID;
3177 if (flags & MS_NODEV)
3178 mnt_flags |= MNT_NODEV;
3179 if (flags & MS_NOEXEC)
3180 mnt_flags |= MNT_NOEXEC;
3181 if (flags & MS_NOATIME)
3182 mnt_flags |= MNT_NOATIME;
3183 if (flags & MS_NODIRATIME)
3184 mnt_flags |= MNT_NODIRATIME;
3185 if (flags & MS_STRICTATIME)
3186 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3187 if (flags & MS_RDONLY)
3188 mnt_flags |= MNT_READONLY;
3189 if (flags & MS_NOSYMFOLLOW)
3190 mnt_flags |= MNT_NOSYMFOLLOW;
3192 /* The default atime for remount is preservation */
3193 if ((flags & MS_REMOUNT) &&
3194 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3195 MS_STRICTATIME)) == 0)) {
3196 mnt_flags &= ~MNT_ATIME_MASK;
3197 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3200 sb_flags = flags & (SB_RDONLY |
3209 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3210 return do_reconfigure_mnt(path, mnt_flags);
3211 if (flags & MS_REMOUNT)
3212 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3213 if (flags & MS_BIND)
3214 return do_loopback(path, dev_name, flags & MS_REC);
3215 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3216 return do_change_type(path, flags);
3217 if (flags & MS_MOVE)
3218 return do_move_mount_old(path, dev_name);
3220 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3224 long do_mount(const char *dev_name, const char __user *dir_name,
3225 const char *type_page, unsigned long flags, void *data_page)
3230 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3233 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3238 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3240 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3243 static void dec_mnt_namespaces(struct ucounts *ucounts)
3245 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3248 static void free_mnt_ns(struct mnt_namespace *ns)
3250 if (!is_anon_ns(ns))
3251 ns_free_inum(&ns->ns);
3252 dec_mnt_namespaces(ns->ucounts);
3253 put_user_ns(ns->user_ns);
3258 * Assign a sequence number so we can detect when we attempt to bind
3259 * mount a reference to an older mount namespace into the current
3260 * mount namespace, preventing reference counting loops. A 64bit
3261 * number incrementing at 10Ghz will take 12,427 years to wrap which
3262 * is effectively never, so we can ignore the possibility.
3264 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3266 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3268 struct mnt_namespace *new_ns;
3269 struct ucounts *ucounts;
3272 ucounts = inc_mnt_namespaces(user_ns);
3274 return ERR_PTR(-ENOSPC);
3276 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3278 dec_mnt_namespaces(ucounts);
3279 return ERR_PTR(-ENOMEM);
3282 ret = ns_alloc_inum(&new_ns->ns);
3285 dec_mnt_namespaces(ucounts);
3286 return ERR_PTR(ret);
3289 new_ns->ns.ops = &mntns_operations;
3291 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3292 refcount_set(&new_ns->ns.count, 1);
3293 INIT_LIST_HEAD(&new_ns->list);
3294 init_waitqueue_head(&new_ns->poll);
3295 spin_lock_init(&new_ns->ns_lock);
3296 new_ns->user_ns = get_user_ns(user_ns);
3297 new_ns->ucounts = ucounts;
3302 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3303 struct user_namespace *user_ns, struct fs_struct *new_fs)
3305 struct mnt_namespace *new_ns;
3306 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3307 struct mount *p, *q;
3314 if (likely(!(flags & CLONE_NEWNS))) {
3321 new_ns = alloc_mnt_ns(user_ns, false);
3326 /* First pass: copy the tree topology */
3327 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3328 if (user_ns != ns->user_ns)
3329 copy_flags |= CL_SHARED_TO_SLAVE;
3330 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3333 free_mnt_ns(new_ns);
3334 return ERR_CAST(new);
3336 if (user_ns != ns->user_ns) {
3339 unlock_mount_hash();
3342 list_add_tail(&new_ns->list, &new->mnt_list);
3345 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3346 * as belonging to new namespace. We have already acquired a private
3347 * fs_struct, so tsk->fs->lock is not needed.
3355 if (&p->mnt == new_fs->root.mnt) {
3356 new_fs->root.mnt = mntget(&q->mnt);
3359 if (&p->mnt == new_fs->pwd.mnt) {
3360 new_fs->pwd.mnt = mntget(&q->mnt);
3364 p = next_mnt(p, old);
3365 q = next_mnt(q, new);
3368 while (p->mnt.mnt_root != q->mnt.mnt_root)
3369 p = next_mnt(p, old);
3381 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3383 struct mount *mnt = real_mount(m);
3384 struct mnt_namespace *ns;
3385 struct super_block *s;
3389 ns = alloc_mnt_ns(&init_user_ns, true);
3392 return ERR_CAST(ns);
3397 list_add(&mnt->mnt_list, &ns->list);
3399 err = vfs_path_lookup(m->mnt_root, m,
3400 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3405 return ERR_PTR(err);
3407 /* trade a vfsmount reference for active sb one */
3408 s = path.mnt->mnt_sb;
3409 atomic_inc(&s->s_active);
3411 /* lock the sucker */
3412 down_write(&s->s_umount);
3413 /* ... and return the root of (sub)tree on it */
3416 EXPORT_SYMBOL(mount_subtree);
3418 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3419 char __user *, type, unsigned long, flags, void __user *, data)
3426 kernel_type = copy_mount_string(type);
3427 ret = PTR_ERR(kernel_type);
3428 if (IS_ERR(kernel_type))
3431 kernel_dev = copy_mount_string(dev_name);
3432 ret = PTR_ERR(kernel_dev);
3433 if (IS_ERR(kernel_dev))
3436 options = copy_mount_options(data);
3437 ret = PTR_ERR(options);
3438 if (IS_ERR(options))
3441 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3453 * Create a kernel mount representation for a new, prepared superblock
3454 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3456 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3457 unsigned int, attr_flags)
3459 struct mnt_namespace *ns;
3460 struct fs_context *fc;
3462 struct path newmount;
3465 unsigned int mnt_flags = 0;
3471 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3474 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3479 MOUNT_ATTR_NODIRATIME))
3482 if (attr_flags & MOUNT_ATTR_RDONLY)
3483 mnt_flags |= MNT_READONLY;
3484 if (attr_flags & MOUNT_ATTR_NOSUID)
3485 mnt_flags |= MNT_NOSUID;
3486 if (attr_flags & MOUNT_ATTR_NODEV)
3487 mnt_flags |= MNT_NODEV;
3488 if (attr_flags & MOUNT_ATTR_NOEXEC)
3489 mnt_flags |= MNT_NOEXEC;
3490 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3491 mnt_flags |= MNT_NODIRATIME;
3493 switch (attr_flags & MOUNT_ATTR__ATIME) {
3494 case MOUNT_ATTR_STRICTATIME:
3496 case MOUNT_ATTR_NOATIME:
3497 mnt_flags |= MNT_NOATIME;
3499 case MOUNT_ATTR_RELATIME:
3500 mnt_flags |= MNT_RELATIME;
3511 if (f.file->f_op != &fscontext_fops)
3514 fc = f.file->private_data;
3516 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3520 /* There must be a valid superblock or we can't mount it */
3526 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3527 pr_warn("VFS: Mount too revealing\n");
3532 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3536 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3539 newmount.mnt = vfs_create_mount(fc);
3540 if (IS_ERR(newmount.mnt)) {
3541 ret = PTR_ERR(newmount.mnt);
3544 newmount.dentry = dget(fc->root);
3545 newmount.mnt->mnt_flags = mnt_flags;
3547 /* We've done the mount bit - now move the file context into more or
3548 * less the same state as if we'd done an fspick(). We don't want to
3549 * do any memory allocation or anything like that at this point as we
3550 * don't want to have to handle any errors incurred.
3552 vfs_clean_context(fc);
3554 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3559 mnt = real_mount(newmount.mnt);
3563 list_add(&mnt->mnt_list, &ns->list);
3564 mntget(newmount.mnt);
3566 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3567 * it, not just simply put it.
3569 file = dentry_open(&newmount, O_PATH, fc->cred);
3571 dissolve_on_fput(newmount.mnt);
3572 ret = PTR_ERR(file);
3575 file->f_mode |= FMODE_NEED_UNMOUNT;
3577 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3579 fd_install(ret, file);
3584 path_put(&newmount);
3586 mutex_unlock(&fc->uapi_mutex);
3593 * Move a mount from one place to another. In combination with
3594 * fsopen()/fsmount() this is used to install a new mount and in combination
3595 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3598 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3600 SYSCALL_DEFINE5(move_mount,
3601 int, from_dfd, const char __user *, from_pathname,
3602 int, to_dfd, const char __user *, to_pathname,
3603 unsigned int, flags)
3605 struct path from_path, to_path;
3606 unsigned int lflags;
3612 if (flags & ~MOVE_MOUNT__MASK)
3615 /* If someone gives a pathname, they aren't permitted to move
3616 * from an fd that requires unmount as we can't get at the flag
3617 * to clear it afterwards.
3620 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3621 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3622 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3624 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3629 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3630 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3631 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3633 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3637 ret = security_move_mount(&from_path, &to_path);
3641 ret = do_move_mount(&from_path, &to_path);
3646 path_put(&from_path);
3651 * Return true if path is reachable from root
3653 * namespace_sem or mount_lock is held
3655 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3656 const struct path *root)
3658 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3659 dentry = mnt->mnt_mountpoint;
3660 mnt = mnt->mnt_parent;
3662 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3665 bool path_is_under(const struct path *path1, const struct path *path2)
3668 read_seqlock_excl(&mount_lock);
3669 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3670 read_sequnlock_excl(&mount_lock);
3673 EXPORT_SYMBOL(path_is_under);
3676 * pivot_root Semantics:
3677 * Moves the root file system of the current process to the directory put_old,
3678 * makes new_root as the new root file system of the current process, and sets
3679 * root/cwd of all processes which had them on the current root to new_root.
3682 * The new_root and put_old must be directories, and must not be on the
3683 * same file system as the current process root. The put_old must be
3684 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3685 * pointed to by put_old must yield the same directory as new_root. No other
3686 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3688 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3689 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3690 * in this situation.
3693 * - we don't move root/cwd if they are not at the root (reason: if something
3694 * cared enough to change them, it's probably wrong to force them elsewhere)
3695 * - it's okay to pick a root that isn't the root of a file system, e.g.
3696 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3697 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3700 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3701 const char __user *, put_old)
3703 struct path new, old, root;
3704 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3705 struct mountpoint *old_mp, *root_mp;
3711 error = user_path_at(AT_FDCWD, new_root,
3712 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3716 error = user_path_at(AT_FDCWD, put_old,
3717 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3721 error = security_sb_pivotroot(&old, &new);
3725 get_fs_root(current->fs, &root);
3726 old_mp = lock_mount(&old);
3727 error = PTR_ERR(old_mp);
3732 new_mnt = real_mount(new.mnt);
3733 root_mnt = real_mount(root.mnt);
3734 old_mnt = real_mount(old.mnt);
3735 ex_parent = new_mnt->mnt_parent;
3736 root_parent = root_mnt->mnt_parent;
3737 if (IS_MNT_SHARED(old_mnt) ||
3738 IS_MNT_SHARED(ex_parent) ||
3739 IS_MNT_SHARED(root_parent))
3741 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3743 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3746 if (d_unlinked(new.dentry))
3749 if (new_mnt == root_mnt || old_mnt == root_mnt)
3750 goto out4; /* loop, on the same file system */
3752 if (root.mnt->mnt_root != root.dentry)
3753 goto out4; /* not a mountpoint */
3754 if (!mnt_has_parent(root_mnt))
3755 goto out4; /* not attached */
3756 if (new.mnt->mnt_root != new.dentry)
3757 goto out4; /* not a mountpoint */
3758 if (!mnt_has_parent(new_mnt))
3759 goto out4; /* not attached */
3760 /* make sure we can reach put_old from new_root */
3761 if (!is_path_reachable(old_mnt, old.dentry, &new))
3763 /* make certain new is below the root */
3764 if (!is_path_reachable(new_mnt, new.dentry, &root))
3767 umount_mnt(new_mnt);
3768 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3769 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3770 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3771 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3773 /* mount old root on put_old */
3774 attach_mnt(root_mnt, old_mnt, old_mp);
3775 /* mount new_root on / */
3776 attach_mnt(new_mnt, root_parent, root_mp);
3777 mnt_add_count(root_parent, -1);
3778 touch_mnt_namespace(current->nsproxy->mnt_ns);
3779 /* A moved mount should not expire automatically */
3780 list_del_init(&new_mnt->mnt_expire);
3781 put_mountpoint(root_mp);
3782 unlock_mount_hash();
3783 chroot_fs_refs(&root, &new);
3786 unlock_mount(old_mp);
3788 mntput_no_expire(ex_parent);
3799 static void __init init_mount_tree(void)
3801 struct vfsmount *mnt;
3803 struct mnt_namespace *ns;
3806 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3808 panic("Can't create rootfs");
3810 ns = alloc_mnt_ns(&init_user_ns, false);
3812 panic("Can't allocate initial namespace");
3813 m = real_mount(mnt);
3817 list_add(&m->mnt_list, &ns->list);
3818 init_task.nsproxy->mnt_ns = ns;
3822 root.dentry = mnt->mnt_root;
3823 mnt->mnt_flags |= MNT_LOCKED;
3825 set_fs_pwd(current->fs, &root);
3826 set_fs_root(current->fs, &root);
3829 void __init mnt_init(void)
3833 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3834 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3836 mount_hashtable = alloc_large_system_hash("Mount-cache",
3837 sizeof(struct hlist_head),
3840 &m_hash_shift, &m_hash_mask, 0, 0);
3841 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3842 sizeof(struct hlist_head),
3845 &mp_hash_shift, &mp_hash_mask, 0, 0);
3847 if (!mount_hashtable || !mountpoint_hashtable)
3848 panic("Failed to allocate mount hash table\n");
3854 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3856 fs_kobj = kobject_create_and_add("fs", NULL);
3858 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3864 void put_mnt_ns(struct mnt_namespace *ns)
3866 if (!refcount_dec_and_test(&ns->ns.count))
3868 drop_collected_mounts(&ns->root->mnt);
3872 struct vfsmount *kern_mount(struct file_system_type *type)
3874 struct vfsmount *mnt;
3875 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3878 * it is a longterm mount, don't release mnt until
3879 * we unmount before file sys is unregistered
3881 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3885 EXPORT_SYMBOL_GPL(kern_mount);
3887 void kern_unmount(struct vfsmount *mnt)
3889 /* release long term mount so mount point can be released */
3890 if (!IS_ERR_OR_NULL(mnt)) {
3891 real_mount(mnt)->mnt_ns = NULL;
3892 synchronize_rcu(); /* yecchhh... */
3896 EXPORT_SYMBOL(kern_unmount);
3898 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
3902 for (i = 0; i < num; i++)
3904 real_mount(mnt[i])->mnt_ns = NULL;
3905 synchronize_rcu_expedited();
3906 for (i = 0; i < num; i++)
3909 EXPORT_SYMBOL(kern_unmount_array);
3911 bool our_mnt(struct vfsmount *mnt)
3913 return check_mnt(real_mount(mnt));
3916 bool current_chrooted(void)
3918 /* Does the current process have a non-standard root */
3919 struct path ns_root;
3920 struct path fs_root;
3923 /* Find the namespace root */
3924 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3925 ns_root.dentry = ns_root.mnt->mnt_root;
3927 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3930 get_fs_root(current->fs, &fs_root);
3932 chrooted = !path_equal(&fs_root, &ns_root);
3940 static bool mnt_already_visible(struct mnt_namespace *ns,
3941 const struct super_block *sb,
3944 int new_flags = *new_mnt_flags;
3946 bool visible = false;
3948 down_read(&namespace_sem);
3950 list_for_each_entry(mnt, &ns->list, mnt_list) {
3951 struct mount *child;
3954 if (mnt_is_cursor(mnt))
3957 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3960 /* This mount is not fully visible if it's root directory
3961 * is not the root directory of the filesystem.
3963 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3966 /* A local view of the mount flags */
3967 mnt_flags = mnt->mnt.mnt_flags;
3969 /* Don't miss readonly hidden in the superblock flags */
3970 if (sb_rdonly(mnt->mnt.mnt_sb))
3971 mnt_flags |= MNT_LOCK_READONLY;
3973 /* Verify the mount flags are equal to or more permissive
3974 * than the proposed new mount.
3976 if ((mnt_flags & MNT_LOCK_READONLY) &&
3977 !(new_flags & MNT_READONLY))
3979 if ((mnt_flags & MNT_LOCK_ATIME) &&
3980 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3983 /* This mount is not fully visible if there are any
3984 * locked child mounts that cover anything except for
3985 * empty directories.
3987 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3988 struct inode *inode = child->mnt_mountpoint->d_inode;
3989 /* Only worry about locked mounts */
3990 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3992 /* Is the directory permanetly empty? */
3993 if (!is_empty_dir_inode(inode))
3996 /* Preserve the locked attributes */
3997 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4005 up_read(&namespace_sem);
4009 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4011 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4012 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4013 unsigned long s_iflags;
4015 if (ns->user_ns == &init_user_ns)
4018 /* Can this filesystem be too revealing? */
4019 s_iflags = sb->s_iflags;
4020 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4023 if ((s_iflags & required_iflags) != required_iflags) {
4024 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4029 return !mnt_already_visible(ns, sb, new_mnt_flags);
4032 bool mnt_may_suid(struct vfsmount *mnt)
4035 * Foreign mounts (accessed via fchdir or through /proc
4036 * symlinks) are always treated as if they are nosuid. This
4037 * prevents namespaces from trusting potentially unsafe
4038 * suid/sgid bits, file caps, or security labels that originate
4039 * in other namespaces.
4041 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4042 current_in_userns(mnt->mnt_sb->s_user_ns);
4045 static struct ns_common *mntns_get(struct task_struct *task)
4047 struct ns_common *ns = NULL;
4048 struct nsproxy *nsproxy;
4051 nsproxy = task->nsproxy;
4053 ns = &nsproxy->mnt_ns->ns;
4054 get_mnt_ns(to_mnt_ns(ns));
4061 static void mntns_put(struct ns_common *ns)
4063 put_mnt_ns(to_mnt_ns(ns));
4066 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4068 struct nsproxy *nsproxy = nsset->nsproxy;
4069 struct fs_struct *fs = nsset->fs;
4070 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4071 struct user_namespace *user_ns = nsset->cred->user_ns;
4075 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4076 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4077 !ns_capable(user_ns, CAP_SYS_ADMIN))
4080 if (is_anon_ns(mnt_ns))
4087 old_mnt_ns = nsproxy->mnt_ns;
4088 nsproxy->mnt_ns = mnt_ns;
4091 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4092 "/", LOOKUP_DOWN, &root);
4094 /* revert to old namespace */
4095 nsproxy->mnt_ns = old_mnt_ns;
4100 put_mnt_ns(old_mnt_ns);
4102 /* Update the pwd and root */
4103 set_fs_pwd(fs, &root);
4104 set_fs_root(fs, &root);
4110 static struct user_namespace *mntns_owner(struct ns_common *ns)
4112 return to_mnt_ns(ns)->user_ns;
4115 const struct proc_ns_operations mntns_operations = {
4117 .type = CLONE_NEWNS,
4120 .install = mntns_install,
4121 .owner = mntns_owner,