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/delay.h>
18 #include <linux/security.h>
19 #include <linux/cred.h>
20 #include <linux/idr.h>
21 #include <linux/init.h> /* init_rootfs */
22 #include <linux/fs_struct.h> /* get_fs_root et.al. */
23 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
24 #include <linux/file.h>
25 #include <linux/uaccess.h>
26 #include <linux/proc_ns.h>
27 #include <linux/magic.h>
28 #include <linux/memblock.h>
29 #include <linux/task_work.h>
30 #include <linux/sched/task.h>
31 #include <uapi/linux/mount.h>
32 #include <linux/fs_context.h>
33 #include <linux/shmem_fs.h>
38 /* Maximum number of mounts in a mount namespace */
39 unsigned int sysctl_mount_max __read_mostly = 100000;
41 static unsigned int m_hash_mask __read_mostly;
42 static unsigned int m_hash_shift __read_mostly;
43 static unsigned int mp_hash_mask __read_mostly;
44 static unsigned int mp_hash_shift __read_mostly;
46 static __initdata unsigned long mhash_entries;
47 static int __init set_mhash_entries(char *str)
51 mhash_entries = simple_strtoul(str, &str, 0);
54 __setup("mhash_entries=", set_mhash_entries);
56 static __initdata unsigned long mphash_entries;
57 static int __init set_mphash_entries(char *str)
61 mphash_entries = simple_strtoul(str, &str, 0);
64 __setup("mphash_entries=", set_mphash_entries);
67 static DEFINE_IDA(mnt_id_ida);
68 static DEFINE_IDA(mnt_group_ida);
70 static struct hlist_head *mount_hashtable __read_mostly;
71 static struct hlist_head *mountpoint_hashtable __read_mostly;
72 static struct kmem_cache *mnt_cache __read_mostly;
73 static DECLARE_RWSEM(namespace_sem);
74 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
75 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
78 struct kobject *fs_kobj;
79 EXPORT_SYMBOL_GPL(fs_kobj);
82 * vfsmount lock may be taken for read to prevent changes to the
83 * vfsmount hash, ie. during mountpoint lookups or walking back
86 * It should be taken for write in all cases where the vfsmount
87 * tree or hash is modified or when a vfsmount structure is modified.
89 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
91 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
93 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
94 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> m_hash_shift);
96 return &mount_hashtable[tmp & m_hash_mask];
99 static inline struct hlist_head *mp_hash(struct dentry *dentry)
101 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
102 tmp = tmp + (tmp >> mp_hash_shift);
103 return &mountpoint_hashtable[tmp & mp_hash_mask];
106 static int mnt_alloc_id(struct mount *mnt)
108 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
116 static void mnt_free_id(struct mount *mnt)
118 ida_free(&mnt_id_ida, mnt->mnt_id);
122 * Allocate a new peer group ID
124 static int mnt_alloc_group_id(struct mount *mnt)
126 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
130 mnt->mnt_group_id = res;
135 * Release a peer group ID
137 void mnt_release_group_id(struct mount *mnt)
139 ida_free(&mnt_group_ida, mnt->mnt_group_id);
140 mnt->mnt_group_id = 0;
144 * vfsmount lock must be held for read
146 static inline void mnt_add_count(struct mount *mnt, int n)
149 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
158 * vfsmount lock must be held for write
160 int mnt_get_count(struct mount *mnt)
166 for_each_possible_cpu(cpu) {
167 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
172 return mnt->mnt_count;
176 static struct mount *alloc_vfsmnt(const char *name)
178 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
182 err = mnt_alloc_id(mnt);
187 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
188 if (!mnt->mnt_devname)
193 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
195 goto out_free_devname;
197 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
200 mnt->mnt_writers = 0;
203 INIT_HLIST_NODE(&mnt->mnt_hash);
204 INIT_LIST_HEAD(&mnt->mnt_child);
205 INIT_LIST_HEAD(&mnt->mnt_mounts);
206 INIT_LIST_HEAD(&mnt->mnt_list);
207 INIT_LIST_HEAD(&mnt->mnt_expire);
208 INIT_LIST_HEAD(&mnt->mnt_share);
209 INIT_LIST_HEAD(&mnt->mnt_slave_list);
210 INIT_LIST_HEAD(&mnt->mnt_slave);
211 INIT_HLIST_NODE(&mnt->mnt_mp_list);
212 INIT_LIST_HEAD(&mnt->mnt_umounting);
213 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
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) {
331 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
332 * be set to match its requirements. So we must not load that until
333 * MNT_WRITE_HOLD is cleared.
336 if (mnt_is_readonly(m)) {
337 mnt_dec_writers(mnt);
346 * mnt_want_write - get write access to a mount
347 * @m: the mount on which to take a write
349 * This tells the low-level filesystem that a write is about to be performed to
350 * it, and makes sure that writes are allowed (mount is read-write, filesystem
351 * is not frozen) before returning success. When the write operation is
352 * finished, mnt_drop_write() must be called. This is effectively a refcount.
354 int mnt_want_write(struct vfsmount *m)
358 sb_start_write(m->mnt_sb);
359 ret = __mnt_want_write(m);
361 sb_end_write(m->mnt_sb);
364 EXPORT_SYMBOL_GPL(mnt_want_write);
367 * mnt_clone_write - get write access to a mount
368 * @mnt: the mount on which to take a write
370 * This is effectively like mnt_want_write, except
371 * it must only be used to take an extra write reference
372 * on a mountpoint that we already know has a write reference
373 * on it. This allows some optimisation.
375 * After finished, mnt_drop_write must be called as usual to
376 * drop the reference.
378 int mnt_clone_write(struct vfsmount *mnt)
380 /* superblock may be r/o */
381 if (__mnt_is_readonly(mnt))
384 mnt_inc_writers(real_mount(mnt));
388 EXPORT_SYMBOL_GPL(mnt_clone_write);
391 * __mnt_want_write_file - get write access to a file's mount
392 * @file: the file who's mount on which to take a write
394 * This is like __mnt_want_write, but it takes a file and can
395 * do some optimisations if the file is open for write already
397 int __mnt_want_write_file(struct file *file)
399 if (!(file->f_mode & FMODE_WRITER))
400 return __mnt_want_write(file->f_path.mnt);
402 return mnt_clone_write(file->f_path.mnt);
406 * mnt_want_write_file - get write access to a file's mount
407 * @file: the file who's mount on which to take a write
409 * This is like mnt_want_write, but it takes a file and can
410 * do some optimisations if the file is open for write already
412 int mnt_want_write_file(struct file *file)
416 sb_start_write(file_inode(file)->i_sb);
417 ret = __mnt_want_write_file(file);
419 sb_end_write(file_inode(file)->i_sb);
422 EXPORT_SYMBOL_GPL(mnt_want_write_file);
425 * __mnt_drop_write - give up write access to a mount
426 * @mnt: the mount on which to give up write access
428 * Tells the low-level filesystem that we are done
429 * performing writes to it. Must be matched with
430 * __mnt_want_write() call above.
432 void __mnt_drop_write(struct vfsmount *mnt)
435 mnt_dec_writers(real_mount(mnt));
440 * mnt_drop_write - give up write access to a mount
441 * @mnt: the mount on which to give up write access
443 * Tells the low-level filesystem that we are done performing writes to it and
444 * also allows filesystem to be frozen again. Must be matched with
445 * mnt_want_write() call above.
447 void mnt_drop_write(struct vfsmount *mnt)
449 __mnt_drop_write(mnt);
450 sb_end_write(mnt->mnt_sb);
452 EXPORT_SYMBOL_GPL(mnt_drop_write);
454 void __mnt_drop_write_file(struct file *file)
456 __mnt_drop_write(file->f_path.mnt);
459 void mnt_drop_write_file(struct file *file)
461 __mnt_drop_write_file(file);
462 sb_end_write(file_inode(file)->i_sb);
464 EXPORT_SYMBOL(mnt_drop_write_file);
466 static int mnt_make_readonly(struct mount *mnt)
471 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
473 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
474 * should be visible before we do.
479 * With writers on hold, if this value is zero, then there are
480 * definitely no active writers (although held writers may subsequently
481 * increment the count, they'll have to wait, and decrement it after
482 * seeing MNT_READONLY).
484 * It is OK to have counter incremented on one CPU and decremented on
485 * another: the sum will add up correctly. The danger would be when we
486 * sum up each counter, if we read a counter before it is incremented,
487 * but then read another CPU's count which it has been subsequently
488 * decremented from -- we would see more decrements than we should.
489 * MNT_WRITE_HOLD protects against this scenario, because
490 * mnt_want_write first increments count, then smp_mb, then spins on
491 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
492 * we're counting up here.
494 if (mnt_get_writers(mnt) > 0)
497 mnt->mnt.mnt_flags |= MNT_READONLY;
499 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
500 * that become unheld will see MNT_READONLY.
503 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
508 static int __mnt_unmake_readonly(struct mount *mnt)
511 mnt->mnt.mnt_flags &= ~MNT_READONLY;
516 int sb_prepare_remount_readonly(struct super_block *sb)
521 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
522 if (atomic_long_read(&sb->s_remove_count))
526 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
527 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
528 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
530 if (mnt_get_writers(mnt) > 0) {
536 if (!err && atomic_long_read(&sb->s_remove_count))
540 sb->s_readonly_remount = 1;
543 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
544 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
545 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
552 static void free_vfsmnt(struct mount *mnt)
554 kfree_const(mnt->mnt_devname);
556 free_percpu(mnt->mnt_pcp);
558 kmem_cache_free(mnt_cache, mnt);
561 static void delayed_free_vfsmnt(struct rcu_head *head)
563 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
566 /* call under rcu_read_lock */
567 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
570 if (read_seqretry(&mount_lock, seq))
574 mnt = real_mount(bastard);
575 mnt_add_count(mnt, 1);
576 smp_mb(); // see mntput_no_expire()
577 if (likely(!read_seqretry(&mount_lock, seq)))
579 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
580 mnt_add_count(mnt, -1);
584 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
585 mnt_add_count(mnt, -1);
590 /* caller will mntput() */
594 /* call under rcu_read_lock */
595 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
597 int res = __legitimize_mnt(bastard, seq);
600 if (unlikely(res < 0)) {
609 * find the first mount at @dentry on vfsmount @mnt.
610 * call under rcu_read_lock()
612 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
614 struct hlist_head *head = m_hash(mnt, dentry);
617 hlist_for_each_entry_rcu(p, head, mnt_hash)
618 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
624 * lookup_mnt - Return the first child mount mounted at path
626 * "First" means first mounted chronologically. If you create the
629 * mount /dev/sda1 /mnt
630 * mount /dev/sda2 /mnt
631 * mount /dev/sda3 /mnt
633 * Then lookup_mnt() on the base /mnt dentry in the root mount will
634 * return successively the root dentry and vfsmount of /dev/sda1, then
635 * /dev/sda2, then /dev/sda3, then NULL.
637 * lookup_mnt takes a reference to the found vfsmount.
639 struct vfsmount *lookup_mnt(const struct path *path)
641 struct mount *child_mnt;
647 seq = read_seqbegin(&mount_lock);
648 child_mnt = __lookup_mnt(path->mnt, path->dentry);
649 m = child_mnt ? &child_mnt->mnt : NULL;
650 } while (!legitimize_mnt(m, seq));
655 static inline void lock_ns_list(struct mnt_namespace *ns)
657 spin_lock(&ns->ns_lock);
660 static inline void unlock_ns_list(struct mnt_namespace *ns)
662 spin_unlock(&ns->ns_lock);
665 static inline bool mnt_is_cursor(struct mount *mnt)
667 return mnt->mnt.mnt_flags & MNT_CURSOR;
671 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
672 * current mount namespace.
674 * The common case is dentries are not mountpoints at all and that
675 * test is handled inline. For the slow case when we are actually
676 * dealing with a mountpoint of some kind, walk through all of the
677 * mounts in the current mount namespace and test to see if the dentry
680 * The mount_hashtable is not usable in the context because we
681 * need to identify all mounts that may be in the current mount
682 * namespace not just a mount that happens to have some specified
685 bool __is_local_mountpoint(struct dentry *dentry)
687 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
689 bool is_covered = false;
691 down_read(&namespace_sem);
693 list_for_each_entry(mnt, &ns->list, mnt_list) {
694 if (mnt_is_cursor(mnt))
696 is_covered = (mnt->mnt_mountpoint == dentry);
701 up_read(&namespace_sem);
706 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
708 struct hlist_head *chain = mp_hash(dentry);
709 struct mountpoint *mp;
711 hlist_for_each_entry(mp, chain, m_hash) {
712 if (mp->m_dentry == dentry) {
720 static struct mountpoint *get_mountpoint(struct dentry *dentry)
722 struct mountpoint *mp, *new = NULL;
725 if (d_mountpoint(dentry)) {
726 /* might be worth a WARN_ON() */
727 if (d_unlinked(dentry))
728 return ERR_PTR(-ENOENT);
730 read_seqlock_excl(&mount_lock);
731 mp = lookup_mountpoint(dentry);
732 read_sequnlock_excl(&mount_lock);
738 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
740 return ERR_PTR(-ENOMEM);
743 /* Exactly one processes may set d_mounted */
744 ret = d_set_mounted(dentry);
746 /* Someone else set d_mounted? */
750 /* The dentry is not available as a mountpoint? */
755 /* Add the new mountpoint to the hash table */
756 read_seqlock_excl(&mount_lock);
757 new->m_dentry = dget(dentry);
759 hlist_add_head(&new->m_hash, mp_hash(dentry));
760 INIT_HLIST_HEAD(&new->m_list);
761 read_sequnlock_excl(&mount_lock);
771 * vfsmount lock must be held. Additionally, the caller is responsible
772 * for serializing calls for given disposal list.
774 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
776 if (!--mp->m_count) {
777 struct dentry *dentry = mp->m_dentry;
778 BUG_ON(!hlist_empty(&mp->m_list));
779 spin_lock(&dentry->d_lock);
780 dentry->d_flags &= ~DCACHE_MOUNTED;
781 spin_unlock(&dentry->d_lock);
782 dput_to_list(dentry, list);
783 hlist_del(&mp->m_hash);
788 /* called with namespace_lock and vfsmount lock */
789 static void put_mountpoint(struct mountpoint *mp)
791 __put_mountpoint(mp, &ex_mountpoints);
794 static inline int check_mnt(struct mount *mnt)
796 return mnt->mnt_ns == current->nsproxy->mnt_ns;
800 * vfsmount lock must be held for write
802 static void touch_mnt_namespace(struct mnt_namespace *ns)
806 wake_up_interruptible(&ns->poll);
811 * vfsmount lock must be held for write
813 static void __touch_mnt_namespace(struct mnt_namespace *ns)
815 if (ns && ns->event != event) {
817 wake_up_interruptible(&ns->poll);
822 * vfsmount lock must be held for write
824 static struct mountpoint *unhash_mnt(struct mount *mnt)
826 struct mountpoint *mp;
827 mnt->mnt_parent = mnt;
828 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
829 list_del_init(&mnt->mnt_child);
830 hlist_del_init_rcu(&mnt->mnt_hash);
831 hlist_del_init(&mnt->mnt_mp_list);
838 * vfsmount lock must be held for write
840 static void umount_mnt(struct mount *mnt)
842 put_mountpoint(unhash_mnt(mnt));
846 * vfsmount lock must be held for write
848 void mnt_set_mountpoint(struct mount *mnt,
849 struct mountpoint *mp,
850 struct mount *child_mnt)
853 mnt_add_count(mnt, 1); /* essentially, that's mntget */
854 child_mnt->mnt_mountpoint = mp->m_dentry;
855 child_mnt->mnt_parent = mnt;
856 child_mnt->mnt_mp = mp;
857 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
860 static void __attach_mnt(struct mount *mnt, struct mount *parent)
862 hlist_add_head_rcu(&mnt->mnt_hash,
863 m_hash(&parent->mnt, mnt->mnt_mountpoint));
864 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
868 * vfsmount lock must be held for write
870 static void attach_mnt(struct mount *mnt,
871 struct mount *parent,
872 struct mountpoint *mp)
874 mnt_set_mountpoint(parent, mp, mnt);
875 __attach_mnt(mnt, parent);
878 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
880 struct mountpoint *old_mp = mnt->mnt_mp;
881 struct mount *old_parent = mnt->mnt_parent;
883 list_del_init(&mnt->mnt_child);
884 hlist_del_init(&mnt->mnt_mp_list);
885 hlist_del_init_rcu(&mnt->mnt_hash);
887 attach_mnt(mnt, parent, mp);
889 put_mountpoint(old_mp);
890 mnt_add_count(old_parent, -1);
894 * vfsmount lock must be held for write
896 static void commit_tree(struct mount *mnt)
898 struct mount *parent = mnt->mnt_parent;
901 struct mnt_namespace *n = parent->mnt_ns;
903 BUG_ON(parent == mnt);
905 list_add_tail(&head, &mnt->mnt_list);
906 list_for_each_entry(m, &head, mnt_list)
909 list_splice(&head, n->list.prev);
911 n->mounts += n->pending_mounts;
912 n->pending_mounts = 0;
914 __attach_mnt(mnt, parent);
915 touch_mnt_namespace(n);
918 static struct mount *next_mnt(struct mount *p, struct mount *root)
920 struct list_head *next = p->mnt_mounts.next;
921 if (next == &p->mnt_mounts) {
925 next = p->mnt_child.next;
926 if (next != &p->mnt_parent->mnt_mounts)
931 return list_entry(next, struct mount, mnt_child);
934 static struct mount *skip_mnt_tree(struct mount *p)
936 struct list_head *prev = p->mnt_mounts.prev;
937 while (prev != &p->mnt_mounts) {
938 p = list_entry(prev, struct mount, mnt_child);
939 prev = p->mnt_mounts.prev;
945 * vfs_create_mount - Create a mount for a configured superblock
946 * @fc: The configuration context with the superblock attached
948 * Create a mount to an already configured superblock. If necessary, the
949 * caller should invoke vfs_get_tree() before calling this.
951 * Note that this does not attach the mount to anything.
953 struct vfsmount *vfs_create_mount(struct fs_context *fc)
958 return ERR_PTR(-EINVAL);
960 mnt = alloc_vfsmnt(fc->source ?: "none");
962 return ERR_PTR(-ENOMEM);
964 if (fc->sb_flags & SB_KERNMOUNT)
965 mnt->mnt.mnt_flags = MNT_INTERNAL;
967 atomic_inc(&fc->root->d_sb->s_active);
968 mnt->mnt.mnt_sb = fc->root->d_sb;
969 mnt->mnt.mnt_root = dget(fc->root);
970 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
971 mnt->mnt_parent = mnt;
974 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
978 EXPORT_SYMBOL(vfs_create_mount);
980 struct vfsmount *fc_mount(struct fs_context *fc)
982 int err = vfs_get_tree(fc);
984 up_write(&fc->root->d_sb->s_umount);
985 return vfs_create_mount(fc);
989 EXPORT_SYMBOL(fc_mount);
991 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
992 int flags, const char *name,
995 struct fs_context *fc;
996 struct vfsmount *mnt;
1000 return ERR_PTR(-EINVAL);
1002 fc = fs_context_for_mount(type, flags);
1004 return ERR_CAST(fc);
1007 ret = vfs_parse_fs_string(fc, "source",
1008 name, strlen(name));
1010 ret = parse_monolithic_mount_data(fc, data);
1019 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1022 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1023 const char *name, void *data)
1025 /* Until it is worked out how to pass the user namespace
1026 * through from the parent mount to the submount don't support
1027 * unprivileged mounts with submounts.
1029 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1030 return ERR_PTR(-EPERM);
1032 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1034 EXPORT_SYMBOL_GPL(vfs_submount);
1036 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1039 struct super_block *sb = old->mnt.mnt_sb;
1043 mnt = alloc_vfsmnt(old->mnt_devname);
1045 return ERR_PTR(-ENOMEM);
1047 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1048 mnt->mnt_group_id = 0; /* not a peer of original */
1050 mnt->mnt_group_id = old->mnt_group_id;
1052 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1053 err = mnt_alloc_group_id(mnt);
1058 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1059 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1061 atomic_inc(&sb->s_active);
1062 mnt->mnt.mnt_sb = sb;
1063 mnt->mnt.mnt_root = dget(root);
1064 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1065 mnt->mnt_parent = mnt;
1067 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1068 unlock_mount_hash();
1070 if ((flag & CL_SLAVE) ||
1071 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1072 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1073 mnt->mnt_master = old;
1074 CLEAR_MNT_SHARED(mnt);
1075 } else if (!(flag & CL_PRIVATE)) {
1076 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1077 list_add(&mnt->mnt_share, &old->mnt_share);
1078 if (IS_MNT_SLAVE(old))
1079 list_add(&mnt->mnt_slave, &old->mnt_slave);
1080 mnt->mnt_master = old->mnt_master;
1082 CLEAR_MNT_SHARED(mnt);
1084 if (flag & CL_MAKE_SHARED)
1085 set_mnt_shared(mnt);
1087 /* stick the duplicate mount on the same expiry list
1088 * as the original if that was on one */
1089 if (flag & CL_EXPIRE) {
1090 if (!list_empty(&old->mnt_expire))
1091 list_add(&mnt->mnt_expire, &old->mnt_expire);
1099 return ERR_PTR(err);
1102 static void cleanup_mnt(struct mount *mnt)
1104 struct hlist_node *p;
1107 * The warning here probably indicates that somebody messed
1108 * up a mnt_want/drop_write() pair. If this happens, the
1109 * filesystem was probably unable to make r/w->r/o transitions.
1110 * The locking used to deal with mnt_count decrement provides barriers,
1111 * so mnt_get_writers() below is safe.
1113 WARN_ON(mnt_get_writers(mnt));
1114 if (unlikely(mnt->mnt_pins.first))
1116 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1117 hlist_del(&m->mnt_umount);
1120 fsnotify_vfsmount_delete(&mnt->mnt);
1121 dput(mnt->mnt.mnt_root);
1122 deactivate_super(mnt->mnt.mnt_sb);
1124 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1127 static void __cleanup_mnt(struct rcu_head *head)
1129 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1132 static LLIST_HEAD(delayed_mntput_list);
1133 static void delayed_mntput(struct work_struct *unused)
1135 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1136 struct mount *m, *t;
1138 llist_for_each_entry_safe(m, t, node, mnt_llist)
1141 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1143 static void mntput_no_expire(struct mount *mnt)
1149 if (likely(READ_ONCE(mnt->mnt_ns))) {
1151 * Since we don't do lock_mount_hash() here,
1152 * ->mnt_ns can change under us. However, if it's
1153 * non-NULL, then there's a reference that won't
1154 * be dropped until after an RCU delay done after
1155 * turning ->mnt_ns NULL. So if we observe it
1156 * non-NULL under rcu_read_lock(), the reference
1157 * we are dropping is not the final one.
1159 mnt_add_count(mnt, -1);
1165 * make sure that if __legitimize_mnt() has not seen us grab
1166 * mount_lock, we'll see their refcount increment here.
1169 mnt_add_count(mnt, -1);
1170 count = mnt_get_count(mnt);
1174 unlock_mount_hash();
1177 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1179 unlock_mount_hash();
1182 mnt->mnt.mnt_flags |= MNT_DOOMED;
1185 list_del(&mnt->mnt_instance);
1187 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1188 struct mount *p, *tmp;
1189 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1190 __put_mountpoint(unhash_mnt(p), &list);
1191 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1194 unlock_mount_hash();
1195 shrink_dentry_list(&list);
1197 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1198 struct task_struct *task = current;
1199 if (likely(!(task->flags & PF_KTHREAD))) {
1200 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1201 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1204 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1205 schedule_delayed_work(&delayed_mntput_work, 1);
1211 void mntput(struct vfsmount *mnt)
1214 struct mount *m = real_mount(mnt);
1215 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1216 if (unlikely(m->mnt_expiry_mark))
1217 m->mnt_expiry_mark = 0;
1218 mntput_no_expire(m);
1221 EXPORT_SYMBOL(mntput);
1223 struct vfsmount *mntget(struct vfsmount *mnt)
1226 mnt_add_count(real_mount(mnt), 1);
1229 EXPORT_SYMBOL(mntget);
1231 /* path_is_mountpoint() - Check if path is a mount in the current
1234 * d_mountpoint() can only be used reliably to establish if a dentry is
1235 * not mounted in any namespace and that common case is handled inline.
1236 * d_mountpoint() isn't aware of the possibility there may be multiple
1237 * mounts using a given dentry in a different namespace. This function
1238 * checks if the passed in path is a mountpoint rather than the dentry
1241 bool path_is_mountpoint(const struct path *path)
1246 if (!d_mountpoint(path->dentry))
1251 seq = read_seqbegin(&mount_lock);
1252 res = __path_is_mountpoint(path);
1253 } while (read_seqretry(&mount_lock, seq));
1258 EXPORT_SYMBOL(path_is_mountpoint);
1260 struct vfsmount *mnt_clone_internal(const struct path *path)
1263 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1266 p->mnt.mnt_flags |= MNT_INTERNAL;
1270 #ifdef CONFIG_PROC_FS
1271 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1272 struct list_head *p)
1274 struct mount *mnt, *ret = NULL;
1277 list_for_each_continue(p, &ns->list) {
1278 mnt = list_entry(p, typeof(*mnt), mnt_list);
1279 if (!mnt_is_cursor(mnt)) {
1289 /* iterator; we want it to have access to namespace_sem, thus here... */
1290 static void *m_start(struct seq_file *m, loff_t *pos)
1292 struct proc_mounts *p = m->private;
1293 struct list_head *prev;
1295 down_read(&namespace_sem);
1297 prev = &p->ns->list;
1299 prev = &p->cursor.mnt_list;
1301 /* Read after we'd reached the end? */
1302 if (list_empty(prev))
1306 return mnt_list_next(p->ns, prev);
1309 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1311 struct proc_mounts *p = m->private;
1312 struct mount *mnt = v;
1315 return mnt_list_next(p->ns, &mnt->mnt_list);
1318 static void m_stop(struct seq_file *m, void *v)
1320 struct proc_mounts *p = m->private;
1321 struct mount *mnt = v;
1323 lock_ns_list(p->ns);
1325 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1327 list_del_init(&p->cursor.mnt_list);
1328 unlock_ns_list(p->ns);
1329 up_read(&namespace_sem);
1332 static int m_show(struct seq_file *m, void *v)
1334 struct proc_mounts *p = m->private;
1335 struct mount *r = v;
1336 return p->show(m, &r->mnt);
1339 const struct seq_operations mounts_op = {
1346 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1348 down_read(&namespace_sem);
1350 list_del(&cursor->mnt_list);
1352 up_read(&namespace_sem);
1354 #endif /* CONFIG_PROC_FS */
1357 * may_umount_tree - check if a mount tree is busy
1358 * @mnt: root of mount tree
1360 * This is called to check if a tree of mounts has any
1361 * open files, pwds, chroots or sub mounts that are
1364 int may_umount_tree(struct vfsmount *m)
1366 struct mount *mnt = real_mount(m);
1367 int actual_refs = 0;
1368 int minimum_refs = 0;
1372 /* write lock needed for mnt_get_count */
1374 for (p = mnt; p; p = next_mnt(p, mnt)) {
1375 actual_refs += mnt_get_count(p);
1378 unlock_mount_hash();
1380 if (actual_refs > minimum_refs)
1386 EXPORT_SYMBOL(may_umount_tree);
1389 * may_umount - check if a mount point is busy
1390 * @mnt: root of mount
1392 * This is called to check if a mount point has any
1393 * open files, pwds, chroots or sub mounts. If the
1394 * mount has sub mounts this will return busy
1395 * regardless of whether the sub mounts are busy.
1397 * Doesn't take quota and stuff into account. IOW, in some cases it will
1398 * give false negatives. The main reason why it's here is that we need
1399 * a non-destructive way to look for easily umountable filesystems.
1401 int may_umount(struct vfsmount *mnt)
1404 down_read(&namespace_sem);
1406 if (propagate_mount_busy(real_mount(mnt), 2))
1408 unlock_mount_hash();
1409 up_read(&namespace_sem);
1413 EXPORT_SYMBOL(may_umount);
1415 static void namespace_unlock(void)
1417 struct hlist_head head;
1418 struct hlist_node *p;
1422 hlist_move_list(&unmounted, &head);
1423 list_splice_init(&ex_mountpoints, &list);
1425 up_write(&namespace_sem);
1427 shrink_dentry_list(&list);
1429 if (likely(hlist_empty(&head)))
1432 synchronize_rcu_expedited();
1434 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1435 hlist_del(&m->mnt_umount);
1440 static inline void namespace_lock(void)
1442 down_write(&namespace_sem);
1445 enum umount_tree_flags {
1447 UMOUNT_PROPAGATE = 2,
1448 UMOUNT_CONNECTED = 4,
1451 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1453 /* Leaving mounts connected is only valid for lazy umounts */
1454 if (how & UMOUNT_SYNC)
1457 /* A mount without a parent has nothing to be connected to */
1458 if (!mnt_has_parent(mnt))
1461 /* Because the reference counting rules change when mounts are
1462 * unmounted and connected, umounted mounts may not be
1463 * connected to mounted mounts.
1465 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1468 /* Has it been requested that the mount remain connected? */
1469 if (how & UMOUNT_CONNECTED)
1472 /* Is the mount locked such that it needs to remain connected? */
1473 if (IS_MNT_LOCKED(mnt))
1476 /* By default disconnect the mount */
1481 * mount_lock must be held
1482 * namespace_sem must be held for write
1484 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1486 LIST_HEAD(tmp_list);
1489 if (how & UMOUNT_PROPAGATE)
1490 propagate_mount_unlock(mnt);
1492 /* Gather the mounts to umount */
1493 for (p = mnt; p; p = next_mnt(p, mnt)) {
1494 p->mnt.mnt_flags |= MNT_UMOUNT;
1495 list_move(&p->mnt_list, &tmp_list);
1498 /* Hide the mounts from mnt_mounts */
1499 list_for_each_entry(p, &tmp_list, mnt_list) {
1500 list_del_init(&p->mnt_child);
1503 /* Add propogated mounts to the tmp_list */
1504 if (how & UMOUNT_PROPAGATE)
1505 propagate_umount(&tmp_list);
1507 while (!list_empty(&tmp_list)) {
1508 struct mnt_namespace *ns;
1510 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1511 list_del_init(&p->mnt_expire);
1512 list_del_init(&p->mnt_list);
1516 __touch_mnt_namespace(ns);
1519 if (how & UMOUNT_SYNC)
1520 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1522 disconnect = disconnect_mount(p, how);
1523 if (mnt_has_parent(p)) {
1524 mnt_add_count(p->mnt_parent, -1);
1526 /* Don't forget about p */
1527 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1532 change_mnt_propagation(p, MS_PRIVATE);
1534 hlist_add_head(&p->mnt_umount, &unmounted);
1538 static void shrink_submounts(struct mount *mnt);
1540 static int do_umount_root(struct super_block *sb)
1544 down_write(&sb->s_umount);
1545 if (!sb_rdonly(sb)) {
1546 struct fs_context *fc;
1548 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1553 ret = parse_monolithic_mount_data(fc, NULL);
1555 ret = reconfigure_super(fc);
1559 up_write(&sb->s_umount);
1563 static int do_umount(struct mount *mnt, int flags)
1565 struct super_block *sb = mnt->mnt.mnt_sb;
1568 retval = security_sb_umount(&mnt->mnt, flags);
1573 * Allow userspace to request a mountpoint be expired rather than
1574 * unmounting unconditionally. Unmount only happens if:
1575 * (1) the mark is already set (the mark is cleared by mntput())
1576 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1578 if (flags & MNT_EXPIRE) {
1579 if (&mnt->mnt == current->fs->root.mnt ||
1580 flags & (MNT_FORCE | MNT_DETACH))
1584 * probably don't strictly need the lock here if we examined
1585 * all race cases, but it's a slowpath.
1588 if (mnt_get_count(mnt) != 2) {
1589 unlock_mount_hash();
1592 unlock_mount_hash();
1594 if (!xchg(&mnt->mnt_expiry_mark, 1))
1599 * If we may have to abort operations to get out of this
1600 * mount, and they will themselves hold resources we must
1601 * allow the fs to do things. In the Unix tradition of
1602 * 'Gee thats tricky lets do it in userspace' the umount_begin
1603 * might fail to complete on the first run through as other tasks
1604 * must return, and the like. Thats for the mount program to worry
1605 * about for the moment.
1608 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1609 sb->s_op->umount_begin(sb);
1613 * No sense to grab the lock for this test, but test itself looks
1614 * somewhat bogus. Suggestions for better replacement?
1615 * Ho-hum... In principle, we might treat that as umount + switch
1616 * to rootfs. GC would eventually take care of the old vfsmount.
1617 * Actually it makes sense, especially if rootfs would contain a
1618 * /reboot - static binary that would close all descriptors and
1619 * call reboot(9). Then init(8) could umount root and exec /reboot.
1621 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1623 * Special case for "unmounting" root ...
1624 * we just try to remount it readonly.
1626 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1628 return do_umount_root(sb);
1634 /* Recheck MNT_LOCKED with the locks held */
1636 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1640 if (flags & MNT_DETACH) {
1641 if (!list_empty(&mnt->mnt_list))
1642 umount_tree(mnt, UMOUNT_PROPAGATE);
1645 shrink_submounts(mnt);
1647 if (!propagate_mount_busy(mnt, 2)) {
1648 if (!list_empty(&mnt->mnt_list))
1649 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1654 unlock_mount_hash();
1660 * __detach_mounts - lazily unmount all mounts on the specified dentry
1662 * During unlink, rmdir, and d_drop it is possible to loose the path
1663 * to an existing mountpoint, and wind up leaking the mount.
1664 * detach_mounts allows lazily unmounting those mounts instead of
1667 * The caller may hold dentry->d_inode->i_mutex.
1669 void __detach_mounts(struct dentry *dentry)
1671 struct mountpoint *mp;
1676 mp = lookup_mountpoint(dentry);
1681 while (!hlist_empty(&mp->m_list)) {
1682 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1683 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1685 hlist_add_head(&mnt->mnt_umount, &unmounted);
1687 else umount_tree(mnt, UMOUNT_CONNECTED);
1691 unlock_mount_hash();
1696 * Is the caller allowed to modify his namespace?
1698 static inline bool may_mount(void)
1700 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1703 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1704 static bool may_mandlock(void)
1706 pr_warn_once("======================================================\n"
1707 "WARNING: the mand mount option is being deprecated and\n"
1708 " will be removed in v5.15!\n"
1709 "======================================================\n");
1710 return capable(CAP_SYS_ADMIN);
1713 static inline bool may_mandlock(void)
1715 pr_warn("VFS: \"mand\" mount option not supported");
1720 static int can_umount(const struct path *path, int flags)
1722 struct mount *mnt = real_mount(path->mnt);
1726 if (path->dentry != path->mnt->mnt_root)
1728 if (!check_mnt(mnt))
1730 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1732 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1737 // caller is responsible for flags being sane
1738 int path_umount(struct path *path, int flags)
1740 struct mount *mnt = real_mount(path->mnt);
1743 ret = can_umount(path, flags);
1745 ret = do_umount(mnt, flags);
1747 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1749 mntput_no_expire(mnt);
1753 static int ksys_umount(char __user *name, int flags)
1755 int lookup_flags = LOOKUP_MOUNTPOINT;
1759 // basic validity checks done first
1760 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1763 if (!(flags & UMOUNT_NOFOLLOW))
1764 lookup_flags |= LOOKUP_FOLLOW;
1765 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1768 return path_umount(&path, flags);
1771 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1773 return ksys_umount(name, flags);
1776 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1779 * The 2.0 compatible umount. No flags.
1781 SYSCALL_DEFINE1(oldumount, char __user *, name)
1783 return ksys_umount(name, 0);
1788 static bool is_mnt_ns_file(struct dentry *dentry)
1790 /* Is this a proxy for a mount namespace? */
1791 return dentry->d_op == &ns_dentry_operations &&
1792 dentry->d_fsdata == &mntns_operations;
1795 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1797 return container_of(ns, struct mnt_namespace, ns);
1800 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1805 static bool mnt_ns_loop(struct dentry *dentry)
1807 /* Could bind mounting the mount namespace inode cause a
1808 * mount namespace loop?
1810 struct mnt_namespace *mnt_ns;
1811 if (!is_mnt_ns_file(dentry))
1814 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1815 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1818 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1821 struct mount *res, *p, *q, *r, *parent;
1823 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1824 return ERR_PTR(-EINVAL);
1826 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1827 return ERR_PTR(-EINVAL);
1829 res = q = clone_mnt(mnt, dentry, flag);
1833 q->mnt_mountpoint = mnt->mnt_mountpoint;
1836 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1838 if (!is_subdir(r->mnt_mountpoint, dentry))
1841 for (s = r; s; s = next_mnt(s, r)) {
1842 if (!(flag & CL_COPY_UNBINDABLE) &&
1843 IS_MNT_UNBINDABLE(s)) {
1844 if (s->mnt.mnt_flags & MNT_LOCKED) {
1845 /* Both unbindable and locked. */
1846 q = ERR_PTR(-EPERM);
1849 s = skip_mnt_tree(s);
1853 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1854 is_mnt_ns_file(s->mnt.mnt_root)) {
1855 s = skip_mnt_tree(s);
1858 while (p != s->mnt_parent) {
1864 q = clone_mnt(p, p->mnt.mnt_root, flag);
1868 list_add_tail(&q->mnt_list, &res->mnt_list);
1869 attach_mnt(q, parent, p->mnt_mp);
1870 unlock_mount_hash();
1877 umount_tree(res, UMOUNT_SYNC);
1878 unlock_mount_hash();
1883 /* Caller should check returned pointer for errors */
1885 struct vfsmount *collect_mounts(const struct path *path)
1889 if (!check_mnt(real_mount(path->mnt)))
1890 tree = ERR_PTR(-EINVAL);
1892 tree = copy_tree(real_mount(path->mnt), path->dentry,
1893 CL_COPY_ALL | CL_PRIVATE);
1896 return ERR_CAST(tree);
1899 EXPORT_SYMBOL_GPL(collect_mounts);
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();
1930 EXPORT_SYMBOL_GPL(drop_collected_mounts);
1932 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1934 struct mount *child;
1936 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1937 if (!is_subdir(child->mnt_mountpoint, dentry))
1940 if (child->mnt.mnt_flags & MNT_LOCKED)
1947 * clone_private_mount - create a private clone of a path
1949 * This creates a new vfsmount, which will be the clone of @path. The new will
1950 * not be attached anywhere in the namespace and will be private (i.e. changes
1951 * to the originating mount won't be propagated into this).
1953 * Release with mntput().
1955 struct vfsmount *clone_private_mount(const struct path *path)
1957 struct mount *old_mnt = real_mount(path->mnt);
1958 struct mount *new_mnt;
1960 down_read(&namespace_sem);
1961 if (IS_MNT_UNBINDABLE(old_mnt))
1964 if (!check_mnt(old_mnt))
1967 if (has_locked_children(old_mnt, path->dentry))
1970 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1971 up_read(&namespace_sem);
1973 if (IS_ERR(new_mnt))
1974 return ERR_CAST(new_mnt);
1976 /* Longterm mount to be removed by kern_unmount*() */
1977 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1979 return &new_mnt->mnt;
1982 up_read(&namespace_sem);
1983 return ERR_PTR(-EINVAL);
1985 EXPORT_SYMBOL_GPL(clone_private_mount);
1987 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1988 struct vfsmount *root)
1991 int res = f(root, arg);
1994 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1995 res = f(&mnt->mnt, arg);
2001 EXPORT_SYMBOL_GPL(iterate_mounts);
2003 static void lock_mnt_tree(struct mount *mnt)
2007 for (p = mnt; p; p = next_mnt(p, mnt)) {
2008 int flags = p->mnt.mnt_flags;
2009 /* Don't allow unprivileged users to change mount flags */
2010 flags |= MNT_LOCK_ATIME;
2012 if (flags & MNT_READONLY)
2013 flags |= MNT_LOCK_READONLY;
2015 if (flags & MNT_NODEV)
2016 flags |= MNT_LOCK_NODEV;
2018 if (flags & MNT_NOSUID)
2019 flags |= MNT_LOCK_NOSUID;
2021 if (flags & MNT_NOEXEC)
2022 flags |= MNT_LOCK_NOEXEC;
2023 /* Don't allow unprivileged users to reveal what is under a mount */
2024 if (list_empty(&p->mnt_expire))
2025 flags |= MNT_LOCKED;
2026 p->mnt.mnt_flags = flags;
2030 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2034 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2035 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2036 mnt_release_group_id(p);
2040 static int invent_group_ids(struct mount *mnt, bool recurse)
2044 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2045 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2046 int err = mnt_alloc_group_id(p);
2048 cleanup_group_ids(mnt, p);
2057 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2059 unsigned int max = READ_ONCE(sysctl_mount_max);
2060 unsigned int mounts = 0, old, pending, sum;
2063 for (p = mnt; p; p = next_mnt(p, mnt))
2067 pending = ns->pending_mounts;
2068 sum = old + pending;
2072 (mounts > (max - sum)))
2075 ns->pending_mounts = pending + mounts;
2080 * @source_mnt : mount tree to be attached
2081 * @nd : place the mount tree @source_mnt is attached
2082 * @parent_nd : if non-null, detach the source_mnt from its parent and
2083 * store the parent mount and mountpoint dentry.
2084 * (done when source_mnt is moved)
2086 * NOTE: in the table below explains the semantics when a source mount
2087 * of a given type is attached to a destination mount of a given type.
2088 * ---------------------------------------------------------------------------
2089 * | BIND MOUNT OPERATION |
2090 * |**************************************************************************
2091 * | source-->| shared | private | slave | unbindable |
2095 * |**************************************************************************
2096 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2098 * |non-shared| shared (+) | private | slave (*) | invalid |
2099 * ***************************************************************************
2100 * A bind operation clones the source mount and mounts the clone on the
2101 * destination mount.
2103 * (++) the cloned mount is propagated to all the mounts in the propagation
2104 * tree of the destination mount and the cloned mount is added to
2105 * the peer group of the source mount.
2106 * (+) the cloned mount is created under the destination mount and is marked
2107 * as shared. The cloned mount is added to the peer group of the source
2109 * (+++) the mount is propagated to all the mounts in the propagation tree
2110 * of the destination mount and the cloned mount is made slave
2111 * of the same master as that of the source mount. The cloned mount
2112 * is marked as 'shared and slave'.
2113 * (*) the cloned mount is made a slave of the same master as that of the
2116 * ---------------------------------------------------------------------------
2117 * | MOVE MOUNT OPERATION |
2118 * |**************************************************************************
2119 * | source-->| shared | private | slave | unbindable |
2123 * |**************************************************************************
2124 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2126 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2127 * ***************************************************************************
2129 * (+) the mount is moved to the destination. And is then propagated to
2130 * all the mounts in the propagation tree of the destination mount.
2131 * (+*) the mount is moved to the destination.
2132 * (+++) the mount is moved to the destination and is then propagated to
2133 * all the mounts belonging to the destination mount's propagation tree.
2134 * the mount is marked as 'shared and slave'.
2135 * (*) the mount continues to be a slave at the new location.
2137 * if the source mount is a tree, the operations explained above is
2138 * applied to each mount in the tree.
2139 * Must be called without spinlocks held, since this function can sleep
2142 static int attach_recursive_mnt(struct mount *source_mnt,
2143 struct mount *dest_mnt,
2144 struct mountpoint *dest_mp,
2147 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2148 HLIST_HEAD(tree_list);
2149 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2150 struct mountpoint *smp;
2151 struct mount *child, *p;
2152 struct hlist_node *n;
2155 /* Preallocate a mountpoint in case the new mounts need
2156 * to be tucked under other mounts.
2158 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2160 return PTR_ERR(smp);
2162 /* Is there space to add these mounts to the mount namespace? */
2164 err = count_mounts(ns, source_mnt);
2169 if (IS_MNT_SHARED(dest_mnt)) {
2170 err = invent_group_ids(source_mnt, true);
2173 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2176 goto out_cleanup_ids;
2177 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2183 unhash_mnt(source_mnt);
2184 attach_mnt(source_mnt, dest_mnt, dest_mp);
2185 touch_mnt_namespace(source_mnt->mnt_ns);
2187 if (source_mnt->mnt_ns) {
2188 /* move from anon - the caller will destroy */
2189 list_del_init(&source_mnt->mnt_ns->list);
2191 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2192 commit_tree(source_mnt);
2195 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2197 hlist_del_init(&child->mnt_hash);
2198 q = __lookup_mnt(&child->mnt_parent->mnt,
2199 child->mnt_mountpoint);
2201 mnt_change_mountpoint(child, smp, q);
2202 /* Notice when we are propagating across user namespaces */
2203 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2204 lock_mnt_tree(child);
2205 child->mnt.mnt_flags &= ~MNT_LOCKED;
2208 put_mountpoint(smp);
2209 unlock_mount_hash();
2214 while (!hlist_empty(&tree_list)) {
2215 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2216 child->mnt_parent->mnt_ns->pending_mounts = 0;
2217 umount_tree(child, UMOUNT_SYNC);
2219 unlock_mount_hash();
2220 cleanup_group_ids(source_mnt, NULL);
2222 ns->pending_mounts = 0;
2224 read_seqlock_excl(&mount_lock);
2225 put_mountpoint(smp);
2226 read_sequnlock_excl(&mount_lock);
2231 static struct mountpoint *lock_mount(struct path *path)
2233 struct vfsmount *mnt;
2234 struct dentry *dentry = path->dentry;
2236 inode_lock(dentry->d_inode);
2237 if (unlikely(cant_mount(dentry))) {
2238 inode_unlock(dentry->d_inode);
2239 return ERR_PTR(-ENOENT);
2242 mnt = lookup_mnt(path);
2244 struct mountpoint *mp = get_mountpoint(dentry);
2247 inode_unlock(dentry->d_inode);
2253 inode_unlock(path->dentry->d_inode);
2256 dentry = path->dentry = dget(mnt->mnt_root);
2260 static void unlock_mount(struct mountpoint *where)
2262 struct dentry *dentry = where->m_dentry;
2264 read_seqlock_excl(&mount_lock);
2265 put_mountpoint(where);
2266 read_sequnlock_excl(&mount_lock);
2269 inode_unlock(dentry->d_inode);
2272 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2274 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2277 if (d_is_dir(mp->m_dentry) !=
2278 d_is_dir(mnt->mnt.mnt_root))
2281 return attach_recursive_mnt(mnt, p, mp, false);
2285 * Sanity check the flags to change_mnt_propagation.
2288 static int flags_to_propagation_type(int ms_flags)
2290 int type = ms_flags & ~(MS_REC | MS_SILENT);
2292 /* Fail if any non-propagation flags are set */
2293 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2295 /* Only one propagation flag should be set */
2296 if (!is_power_of_2(type))
2302 * recursively change the type of the mountpoint.
2304 static int do_change_type(struct path *path, int ms_flags)
2307 struct mount *mnt = real_mount(path->mnt);
2308 int recurse = ms_flags & MS_REC;
2312 if (path->dentry != path->mnt->mnt_root)
2315 type = flags_to_propagation_type(ms_flags);
2320 if (type == MS_SHARED) {
2321 err = invent_group_ids(mnt, recurse);
2327 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2328 change_mnt_propagation(m, type);
2329 unlock_mount_hash();
2336 static struct mount *__do_loopback(struct path *old_path, int recurse)
2338 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2340 if (IS_MNT_UNBINDABLE(old))
2343 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2346 if (!recurse && has_locked_children(old, old_path->dentry))
2350 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2352 mnt = clone_mnt(old, old_path->dentry, 0);
2355 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2361 * do loopback mount.
2363 static int do_loopback(struct path *path, const char *old_name,
2366 struct path old_path;
2367 struct mount *mnt = NULL, *parent;
2368 struct mountpoint *mp;
2370 if (!old_name || !*old_name)
2372 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2377 if (mnt_ns_loop(old_path.dentry))
2380 mp = lock_mount(path);
2386 parent = real_mount(path->mnt);
2387 if (!check_mnt(parent))
2390 mnt = __do_loopback(&old_path, recurse);
2396 err = graft_tree(mnt, parent, mp);
2399 umount_tree(mnt, UMOUNT_SYNC);
2400 unlock_mount_hash();
2405 path_put(&old_path);
2409 static struct file *open_detached_copy(struct path *path, bool recursive)
2411 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2412 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2413 struct mount *mnt, *p;
2417 return ERR_CAST(ns);
2420 mnt = __do_loopback(path, recursive);
2424 return ERR_CAST(mnt);
2428 for (p = mnt; p; p = next_mnt(p, mnt)) {
2433 list_add_tail(&ns->list, &mnt->mnt_list);
2435 unlock_mount_hash();
2439 path->mnt = &mnt->mnt;
2440 file = dentry_open(path, O_PATH, current_cred());
2442 dissolve_on_fput(path->mnt);
2444 file->f_mode |= FMODE_NEED_UNMOUNT;
2448 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2452 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2453 bool detached = flags & OPEN_TREE_CLONE;
2457 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2459 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2460 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2464 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2467 if (flags & AT_NO_AUTOMOUNT)
2468 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2469 if (flags & AT_SYMLINK_NOFOLLOW)
2470 lookup_flags &= ~LOOKUP_FOLLOW;
2471 if (flags & AT_EMPTY_PATH)
2472 lookup_flags |= LOOKUP_EMPTY;
2474 if (detached && !may_mount())
2477 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2481 error = user_path_at(dfd, filename, lookup_flags, &path);
2482 if (unlikely(error)) {
2483 file = ERR_PTR(error);
2486 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2488 file = dentry_open(&path, O_PATH, current_cred());
2493 return PTR_ERR(file);
2495 fd_install(fd, file);
2500 * Don't allow locked mount flags to be cleared.
2502 * No locks need to be held here while testing the various MNT_LOCK
2503 * flags because those flags can never be cleared once they are set.
2505 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2507 unsigned int fl = mnt->mnt.mnt_flags;
2509 if ((fl & MNT_LOCK_READONLY) &&
2510 !(mnt_flags & MNT_READONLY))
2513 if ((fl & MNT_LOCK_NODEV) &&
2514 !(mnt_flags & MNT_NODEV))
2517 if ((fl & MNT_LOCK_NOSUID) &&
2518 !(mnt_flags & MNT_NOSUID))
2521 if ((fl & MNT_LOCK_NOEXEC) &&
2522 !(mnt_flags & MNT_NOEXEC))
2525 if ((fl & MNT_LOCK_ATIME) &&
2526 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2532 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2534 bool readonly_request = (mnt_flags & MNT_READONLY);
2536 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2539 if (readonly_request)
2540 return mnt_make_readonly(mnt);
2542 return __mnt_unmake_readonly(mnt);
2546 * Update the user-settable attributes on a mount. The caller must hold
2547 * sb->s_umount for writing.
2549 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2552 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2553 mnt->mnt.mnt_flags = mnt_flags;
2554 touch_mnt_namespace(mnt->mnt_ns);
2555 unlock_mount_hash();
2558 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2560 struct super_block *sb = mnt->mnt_sb;
2562 if (!__mnt_is_readonly(mnt) &&
2563 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2564 char *buf = (char *)__get_free_page(GFP_KERNEL);
2565 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2568 time64_to_tm(sb->s_time_max, 0, &tm);
2570 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2572 is_mounted(mnt) ? "remounted" : "mounted",
2574 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2576 free_page((unsigned long)buf);
2581 * Handle reconfiguration of the mountpoint only without alteration of the
2582 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2585 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2587 struct super_block *sb = path->mnt->mnt_sb;
2588 struct mount *mnt = real_mount(path->mnt);
2591 if (!check_mnt(mnt))
2594 if (path->dentry != mnt->mnt.mnt_root)
2597 if (!can_change_locked_flags(mnt, mnt_flags))
2600 down_write(&sb->s_umount);
2601 ret = change_mount_ro_state(mnt, mnt_flags);
2603 set_mount_attributes(mnt, mnt_flags);
2604 up_write(&sb->s_umount);
2606 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2612 * change filesystem flags. dir should be a physical root of filesystem.
2613 * If you've mounted a non-root directory somewhere and want to do remount
2614 * on it - tough luck.
2616 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2617 int mnt_flags, void *data)
2620 struct super_block *sb = path->mnt->mnt_sb;
2621 struct mount *mnt = real_mount(path->mnt);
2622 struct fs_context *fc;
2624 if (!check_mnt(mnt))
2627 if (path->dentry != path->mnt->mnt_root)
2630 if (!can_change_locked_flags(mnt, mnt_flags))
2633 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2638 err = parse_monolithic_mount_data(fc, data);
2640 down_write(&sb->s_umount);
2642 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2643 err = reconfigure_super(fc);
2645 set_mount_attributes(mnt, mnt_flags);
2647 up_write(&sb->s_umount);
2650 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2656 static inline int tree_contains_unbindable(struct mount *mnt)
2659 for (p = mnt; p; p = next_mnt(p, mnt)) {
2660 if (IS_MNT_UNBINDABLE(p))
2667 * Check that there aren't references to earlier/same mount namespaces in the
2668 * specified subtree. Such references can act as pins for mount namespaces
2669 * that aren't checked by the mount-cycle checking code, thereby allowing
2670 * cycles to be made.
2672 static bool check_for_nsfs_mounts(struct mount *subtree)
2678 for (p = subtree; p; p = next_mnt(p, subtree))
2679 if (mnt_ns_loop(p->mnt.mnt_root))
2684 unlock_mount_hash();
2688 static int do_move_mount(struct path *old_path, struct path *new_path)
2690 struct mnt_namespace *ns;
2693 struct mount *parent;
2694 struct mountpoint *mp, *old_mp;
2698 mp = lock_mount(new_path);
2702 old = real_mount(old_path->mnt);
2703 p = real_mount(new_path->mnt);
2704 parent = old->mnt_parent;
2705 attached = mnt_has_parent(old);
2706 old_mp = old->mnt_mp;
2710 /* The mountpoint must be in our namespace. */
2714 /* The thing moved must be mounted... */
2715 if (!is_mounted(&old->mnt))
2718 /* ... and either ours or the root of anon namespace */
2719 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2722 if (old->mnt.mnt_flags & MNT_LOCKED)
2725 if (old_path->dentry != old_path->mnt->mnt_root)
2728 if (d_is_dir(new_path->dentry) !=
2729 d_is_dir(old_path->dentry))
2732 * Don't move a mount residing in a shared parent.
2734 if (attached && IS_MNT_SHARED(parent))
2737 * Don't move a mount tree containing unbindable mounts to a destination
2738 * mount which is shared.
2740 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2743 if (!check_for_nsfs_mounts(old))
2745 for (; mnt_has_parent(p); p = p->mnt_parent)
2749 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2754 /* if the mount is moved, it should no longer be expire
2756 list_del_init(&old->mnt_expire);
2758 put_mountpoint(old_mp);
2763 mntput_no_expire(parent);
2770 static int do_move_mount_old(struct path *path, const char *old_name)
2772 struct path old_path;
2775 if (!old_name || !*old_name)
2778 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2782 err = do_move_mount(&old_path, path);
2783 path_put(&old_path);
2788 * add a mount into a namespace's mount tree
2790 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2791 struct path *path, int mnt_flags)
2793 struct mount *parent = real_mount(path->mnt);
2795 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2797 if (unlikely(!check_mnt(parent))) {
2798 /* that's acceptable only for automounts done in private ns */
2799 if (!(mnt_flags & MNT_SHRINKABLE))
2801 /* ... and for those we'd better have mountpoint still alive */
2802 if (!parent->mnt_ns)
2806 /* Refuse the same filesystem on the same mount point */
2807 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2808 path->mnt->mnt_root == path->dentry)
2811 if (d_is_symlink(newmnt->mnt.mnt_root))
2814 newmnt->mnt.mnt_flags = mnt_flags;
2815 return graft_tree(newmnt, parent, mp);
2818 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2821 * Create a new mount using a superblock configuration and request it
2822 * be added to the namespace tree.
2824 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2825 unsigned int mnt_flags)
2827 struct vfsmount *mnt;
2828 struct mountpoint *mp;
2829 struct super_block *sb = fc->root->d_sb;
2832 error = security_sb_kern_mount(sb);
2833 if (!error && mount_too_revealing(sb, &mnt_flags))
2836 if (unlikely(error)) {
2841 up_write(&sb->s_umount);
2843 mnt = vfs_create_mount(fc);
2845 return PTR_ERR(mnt);
2847 mnt_warn_timestamp_expiry(mountpoint, mnt);
2849 mp = lock_mount(mountpoint);
2854 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2862 * create a new mount for userspace and request it to be added into the
2865 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2866 int mnt_flags, const char *name, void *data)
2868 struct file_system_type *type;
2869 struct fs_context *fc;
2870 const char *subtype = NULL;
2876 type = get_fs_type(fstype);
2880 if (type->fs_flags & FS_HAS_SUBTYPE) {
2881 subtype = strchr(fstype, '.');
2885 put_filesystem(type);
2891 fc = fs_context_for_mount(type, sb_flags);
2892 put_filesystem(type);
2897 err = vfs_parse_fs_string(fc, "subtype",
2898 subtype, strlen(subtype));
2900 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2902 err = parse_monolithic_mount_data(fc, data);
2903 if (!err && !mount_capable(fc))
2906 err = vfs_get_tree(fc);
2908 err = do_new_mount_fc(fc, path, mnt_flags);
2914 int finish_automount(struct vfsmount *m, struct path *path)
2916 struct dentry *dentry = path->dentry;
2917 struct mountpoint *mp;
2926 mnt = real_mount(m);
2927 /* The new mount record should have at least 2 refs to prevent it being
2928 * expired before we get a chance to add it
2930 BUG_ON(mnt_get_count(mnt) < 2);
2932 if (m->mnt_sb == path->mnt->mnt_sb &&
2933 m->mnt_root == dentry) {
2939 * we don't want to use lock_mount() - in this case finding something
2940 * that overmounts our mountpoint to be means "quitely drop what we've
2941 * got", not "try to mount it on top".
2943 inode_lock(dentry->d_inode);
2945 if (unlikely(cant_mount(dentry))) {
2947 goto discard_locked;
2950 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2953 goto discard_locked;
2956 mp = get_mountpoint(dentry);
2959 goto discard_locked;
2962 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2971 inode_unlock(dentry->d_inode);
2973 /* remove m from any expiration list it may be on */
2974 if (!list_empty(&mnt->mnt_expire)) {
2976 list_del_init(&mnt->mnt_expire);
2985 * mnt_set_expiry - Put a mount on an expiration list
2986 * @mnt: The mount to list.
2987 * @expiry_list: The list to add the mount to.
2989 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2993 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2997 EXPORT_SYMBOL(mnt_set_expiry);
3000 * process a list of expirable mountpoints with the intent of discarding any
3001 * mountpoints that aren't in use and haven't been touched since last we came
3004 void mark_mounts_for_expiry(struct list_head *mounts)
3006 struct mount *mnt, *next;
3007 LIST_HEAD(graveyard);
3009 if (list_empty(mounts))
3015 /* extract from the expiration list every vfsmount that matches the
3016 * following criteria:
3017 * - only referenced by its parent vfsmount
3018 * - still marked for expiry (marked on the last call here; marks are
3019 * cleared by mntput())
3021 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3022 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3023 propagate_mount_busy(mnt, 1))
3025 list_move(&mnt->mnt_expire, &graveyard);
3027 while (!list_empty(&graveyard)) {
3028 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3029 touch_mnt_namespace(mnt->mnt_ns);
3030 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3032 unlock_mount_hash();
3036 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3039 * Ripoff of 'select_parent()'
3041 * search the list of submounts for a given mountpoint, and move any
3042 * shrinkable submounts to the 'graveyard' list.
3044 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3046 struct mount *this_parent = parent;
3047 struct list_head *next;
3051 next = this_parent->mnt_mounts.next;
3053 while (next != &this_parent->mnt_mounts) {
3054 struct list_head *tmp = next;
3055 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3058 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3061 * Descend a level if the d_mounts list is non-empty.
3063 if (!list_empty(&mnt->mnt_mounts)) {
3068 if (!propagate_mount_busy(mnt, 1)) {
3069 list_move_tail(&mnt->mnt_expire, graveyard);
3074 * All done at this level ... ascend and resume the search
3076 if (this_parent != parent) {
3077 next = this_parent->mnt_child.next;
3078 this_parent = this_parent->mnt_parent;
3085 * process a list of expirable mountpoints with the intent of discarding any
3086 * submounts of a specific parent mountpoint
3088 * mount_lock must be held for write
3090 static void shrink_submounts(struct mount *mnt)
3092 LIST_HEAD(graveyard);
3095 /* extract submounts of 'mountpoint' from the expiration list */
3096 while (select_submounts(mnt, &graveyard)) {
3097 while (!list_empty(&graveyard)) {
3098 m = list_first_entry(&graveyard, struct mount,
3100 touch_mnt_namespace(m->mnt_ns);
3101 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3106 static void *copy_mount_options(const void __user * data)
3109 unsigned left, offset;
3114 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3116 return ERR_PTR(-ENOMEM);
3118 left = copy_from_user(copy, data, PAGE_SIZE);
3121 * Not all architectures have an exact copy_from_user(). Resort to
3124 offset = PAGE_SIZE - left;
3127 if (get_user(c, (const char __user *)data + offset))
3134 if (left == PAGE_SIZE) {
3136 return ERR_PTR(-EFAULT);
3142 static char *copy_mount_string(const void __user *data)
3144 return data ? strndup_user(data, PATH_MAX) : NULL;
3148 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3149 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3151 * data is a (void *) that can point to any structure up to
3152 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3153 * information (or be NULL).
3155 * Pre-0.97 versions of mount() didn't have a flags word.
3156 * When the flags word was introduced its top half was required
3157 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3158 * Therefore, if this magic number is present, it carries no information
3159 * and must be discarded.
3161 int path_mount(const char *dev_name, struct path *path,
3162 const char *type_page, unsigned long flags, void *data_page)
3164 unsigned int mnt_flags = 0, sb_flags;
3168 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3169 flags &= ~MS_MGC_MSK;
3171 /* Basic sanity checks */
3173 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3175 if (flags & MS_NOUSER)
3178 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3183 if ((flags & SB_MANDLOCK) && !may_mandlock())
3186 /* Default to relatime unless overriden */
3187 if (!(flags & MS_NOATIME))
3188 mnt_flags |= MNT_RELATIME;
3190 /* Separate the per-mountpoint flags */
3191 if (flags & MS_NOSUID)
3192 mnt_flags |= MNT_NOSUID;
3193 if (flags & MS_NODEV)
3194 mnt_flags |= MNT_NODEV;
3195 if (flags & MS_NOEXEC)
3196 mnt_flags |= MNT_NOEXEC;
3197 if (flags & MS_NOATIME)
3198 mnt_flags |= MNT_NOATIME;
3199 if (flags & MS_NODIRATIME)
3200 mnt_flags |= MNT_NODIRATIME;
3201 if (flags & MS_STRICTATIME)
3202 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3203 if (flags & MS_RDONLY)
3204 mnt_flags |= MNT_READONLY;
3205 if (flags & MS_NOSYMFOLLOW)
3206 mnt_flags |= MNT_NOSYMFOLLOW;
3208 /* The default atime for remount is preservation */
3209 if ((flags & MS_REMOUNT) &&
3210 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3211 MS_STRICTATIME)) == 0)) {
3212 mnt_flags &= ~MNT_ATIME_MASK;
3213 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3216 sb_flags = flags & (SB_RDONLY |
3225 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3226 return do_reconfigure_mnt(path, mnt_flags);
3227 if (flags & MS_REMOUNT)
3228 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3229 if (flags & MS_BIND)
3230 return do_loopback(path, dev_name, flags & MS_REC);
3231 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3232 return do_change_type(path, flags);
3233 if (flags & MS_MOVE)
3234 return do_move_mount_old(path, dev_name);
3236 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3240 long do_mount(const char *dev_name, const char __user *dir_name,
3241 const char *type_page, unsigned long flags, void *data_page)
3246 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3249 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3254 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3256 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3259 static void dec_mnt_namespaces(struct ucounts *ucounts)
3261 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3264 static void free_mnt_ns(struct mnt_namespace *ns)
3266 if (!is_anon_ns(ns))
3267 ns_free_inum(&ns->ns);
3268 dec_mnt_namespaces(ns->ucounts);
3269 put_user_ns(ns->user_ns);
3274 * Assign a sequence number so we can detect when we attempt to bind
3275 * mount a reference to an older mount namespace into the current
3276 * mount namespace, preventing reference counting loops. A 64bit
3277 * number incrementing at 10Ghz will take 12,427 years to wrap which
3278 * is effectively never, so we can ignore the possibility.
3280 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3282 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3284 struct mnt_namespace *new_ns;
3285 struct ucounts *ucounts;
3288 ucounts = inc_mnt_namespaces(user_ns);
3290 return ERR_PTR(-ENOSPC);
3292 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3294 dec_mnt_namespaces(ucounts);
3295 return ERR_PTR(-ENOMEM);
3298 ret = ns_alloc_inum(&new_ns->ns);
3301 dec_mnt_namespaces(ucounts);
3302 return ERR_PTR(ret);
3305 new_ns->ns.ops = &mntns_operations;
3307 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3308 atomic_set(&new_ns->count, 1);
3309 INIT_LIST_HEAD(&new_ns->list);
3310 init_waitqueue_head(&new_ns->poll);
3311 spin_lock_init(&new_ns->ns_lock);
3312 new_ns->user_ns = get_user_ns(user_ns);
3313 new_ns->ucounts = ucounts;
3318 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3319 struct user_namespace *user_ns, struct fs_struct *new_fs)
3321 struct mnt_namespace *new_ns;
3322 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3323 struct mount *p, *q;
3330 if (likely(!(flags & CLONE_NEWNS))) {
3337 new_ns = alloc_mnt_ns(user_ns, false);
3342 /* First pass: copy the tree topology */
3343 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3344 if (user_ns != ns->user_ns)
3345 copy_flags |= CL_SHARED_TO_SLAVE;
3346 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3349 free_mnt_ns(new_ns);
3350 return ERR_CAST(new);
3352 if (user_ns != ns->user_ns) {
3355 unlock_mount_hash();
3358 list_add_tail(&new_ns->list, &new->mnt_list);
3361 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3362 * as belonging to new namespace. We have already acquired a private
3363 * fs_struct, so tsk->fs->lock is not needed.
3371 if (&p->mnt == new_fs->root.mnt) {
3372 new_fs->root.mnt = mntget(&q->mnt);
3375 if (&p->mnt == new_fs->pwd.mnt) {
3376 new_fs->pwd.mnt = mntget(&q->mnt);
3380 p = next_mnt(p, old);
3381 q = next_mnt(q, new);
3384 while (p->mnt.mnt_root != q->mnt.mnt_root)
3385 p = next_mnt(p, old);
3397 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3399 struct mount *mnt = real_mount(m);
3400 struct mnt_namespace *ns;
3401 struct super_block *s;
3405 ns = alloc_mnt_ns(&init_user_ns, true);
3408 return ERR_CAST(ns);
3413 list_add(&mnt->mnt_list, &ns->list);
3415 err = vfs_path_lookup(m->mnt_root, m,
3416 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3421 return ERR_PTR(err);
3423 /* trade a vfsmount reference for active sb one */
3424 s = path.mnt->mnt_sb;
3425 atomic_inc(&s->s_active);
3427 /* lock the sucker */
3428 down_write(&s->s_umount);
3429 /* ... and return the root of (sub)tree on it */
3432 EXPORT_SYMBOL(mount_subtree);
3434 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3435 char __user *, type, unsigned long, flags, void __user *, data)
3442 kernel_type = copy_mount_string(type);
3443 ret = PTR_ERR(kernel_type);
3444 if (IS_ERR(kernel_type))
3447 kernel_dev = copy_mount_string(dev_name);
3448 ret = PTR_ERR(kernel_dev);
3449 if (IS_ERR(kernel_dev))
3452 options = copy_mount_options(data);
3453 ret = PTR_ERR(options);
3454 if (IS_ERR(options))
3457 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3469 * Create a kernel mount representation for a new, prepared superblock
3470 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3472 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3473 unsigned int, attr_flags)
3475 struct mnt_namespace *ns;
3476 struct fs_context *fc;
3478 struct path newmount;
3481 unsigned int mnt_flags = 0;
3487 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3490 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3495 MOUNT_ATTR_NODIRATIME))
3498 if (attr_flags & MOUNT_ATTR_RDONLY)
3499 mnt_flags |= MNT_READONLY;
3500 if (attr_flags & MOUNT_ATTR_NOSUID)
3501 mnt_flags |= MNT_NOSUID;
3502 if (attr_flags & MOUNT_ATTR_NODEV)
3503 mnt_flags |= MNT_NODEV;
3504 if (attr_flags & MOUNT_ATTR_NOEXEC)
3505 mnt_flags |= MNT_NOEXEC;
3506 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3507 mnt_flags |= MNT_NODIRATIME;
3509 switch (attr_flags & MOUNT_ATTR__ATIME) {
3510 case MOUNT_ATTR_STRICTATIME:
3512 case MOUNT_ATTR_NOATIME:
3513 mnt_flags |= MNT_NOATIME;
3515 case MOUNT_ATTR_RELATIME:
3516 mnt_flags |= MNT_RELATIME;
3527 if (f.file->f_op != &fscontext_fops)
3530 fc = f.file->private_data;
3532 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3536 /* There must be a valid superblock or we can't mount it */
3542 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3543 pr_warn("VFS: Mount too revealing\n");
3548 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3552 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3555 newmount.mnt = vfs_create_mount(fc);
3556 if (IS_ERR(newmount.mnt)) {
3557 ret = PTR_ERR(newmount.mnt);
3560 newmount.dentry = dget(fc->root);
3561 newmount.mnt->mnt_flags = mnt_flags;
3563 /* We've done the mount bit - now move the file context into more or
3564 * less the same state as if we'd done an fspick(). We don't want to
3565 * do any memory allocation or anything like that at this point as we
3566 * don't want to have to handle any errors incurred.
3568 vfs_clean_context(fc);
3570 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3575 mnt = real_mount(newmount.mnt);
3579 list_add(&mnt->mnt_list, &ns->list);
3580 mntget(newmount.mnt);
3582 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3583 * it, not just simply put it.
3585 file = dentry_open(&newmount, O_PATH, fc->cred);
3587 dissolve_on_fput(newmount.mnt);
3588 ret = PTR_ERR(file);
3591 file->f_mode |= FMODE_NEED_UNMOUNT;
3593 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3595 fd_install(ret, file);
3600 path_put(&newmount);
3602 mutex_unlock(&fc->uapi_mutex);
3609 * Move a mount from one place to another. In combination with
3610 * fsopen()/fsmount() this is used to install a new mount and in combination
3611 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3614 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3616 SYSCALL_DEFINE5(move_mount,
3617 int, from_dfd, const char __user *, from_pathname,
3618 int, to_dfd, const char __user *, to_pathname,
3619 unsigned int, flags)
3621 struct path from_path, to_path;
3622 unsigned int lflags;
3628 if (flags & ~MOVE_MOUNT__MASK)
3631 /* If someone gives a pathname, they aren't permitted to move
3632 * from an fd that requires unmount as we can't get at the flag
3633 * to clear it afterwards.
3636 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3637 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3638 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3640 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3645 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3646 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3647 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3649 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3653 ret = security_move_mount(&from_path, &to_path);
3657 ret = do_move_mount(&from_path, &to_path);
3662 path_put(&from_path);
3667 * Return true if path is reachable from root
3669 * namespace_sem or mount_lock is held
3671 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3672 const struct path *root)
3674 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3675 dentry = mnt->mnt_mountpoint;
3676 mnt = mnt->mnt_parent;
3678 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3681 bool path_is_under(const struct path *path1, const struct path *path2)
3684 read_seqlock_excl(&mount_lock);
3685 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3686 read_sequnlock_excl(&mount_lock);
3689 EXPORT_SYMBOL(path_is_under);
3692 * pivot_root Semantics:
3693 * Moves the root file system of the current process to the directory put_old,
3694 * makes new_root as the new root file system of the current process, and sets
3695 * root/cwd of all processes which had them on the current root to new_root.
3698 * The new_root and put_old must be directories, and must not be on the
3699 * same file system as the current process root. The put_old must be
3700 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3701 * pointed to by put_old must yield the same directory as new_root. No other
3702 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3704 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3705 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3706 * in this situation.
3709 * - we don't move root/cwd if they are not at the root (reason: if something
3710 * cared enough to change them, it's probably wrong to force them elsewhere)
3711 * - it's okay to pick a root that isn't the root of a file system, e.g.
3712 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3713 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3716 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3717 const char __user *, put_old)
3719 struct path new, old, root;
3720 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3721 struct mountpoint *old_mp, *root_mp;
3727 error = user_path_at(AT_FDCWD, new_root,
3728 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3732 error = user_path_at(AT_FDCWD, put_old,
3733 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3737 error = security_sb_pivotroot(&old, &new);
3741 get_fs_root(current->fs, &root);
3742 old_mp = lock_mount(&old);
3743 error = PTR_ERR(old_mp);
3748 new_mnt = real_mount(new.mnt);
3749 root_mnt = real_mount(root.mnt);
3750 old_mnt = real_mount(old.mnt);
3751 ex_parent = new_mnt->mnt_parent;
3752 root_parent = root_mnt->mnt_parent;
3753 if (IS_MNT_SHARED(old_mnt) ||
3754 IS_MNT_SHARED(ex_parent) ||
3755 IS_MNT_SHARED(root_parent))
3757 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3759 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3762 if (d_unlinked(new.dentry))
3765 if (new_mnt == root_mnt || old_mnt == root_mnt)
3766 goto out4; /* loop, on the same file system */
3768 if (root.mnt->mnt_root != root.dentry)
3769 goto out4; /* not a mountpoint */
3770 if (!mnt_has_parent(root_mnt))
3771 goto out4; /* not attached */
3772 if (new.mnt->mnt_root != new.dentry)
3773 goto out4; /* not a mountpoint */
3774 if (!mnt_has_parent(new_mnt))
3775 goto out4; /* not attached */
3776 /* make sure we can reach put_old from new_root */
3777 if (!is_path_reachable(old_mnt, old.dentry, &new))
3779 /* make certain new is below the root */
3780 if (!is_path_reachable(new_mnt, new.dentry, &root))
3783 umount_mnt(new_mnt);
3784 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3785 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3786 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3787 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3789 /* mount old root on put_old */
3790 attach_mnt(root_mnt, old_mnt, old_mp);
3791 /* mount new_root on / */
3792 attach_mnt(new_mnt, root_parent, root_mp);
3793 mnt_add_count(root_parent, -1);
3794 touch_mnt_namespace(current->nsproxy->mnt_ns);
3795 /* A moved mount should not expire automatically */
3796 list_del_init(&new_mnt->mnt_expire);
3797 put_mountpoint(root_mp);
3798 unlock_mount_hash();
3799 chroot_fs_refs(&root, &new);
3802 unlock_mount(old_mp);
3804 mntput_no_expire(ex_parent);
3815 static void __init init_mount_tree(void)
3817 struct vfsmount *mnt;
3819 struct mnt_namespace *ns;
3822 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3824 panic("Can't create rootfs");
3826 ns = alloc_mnt_ns(&init_user_ns, false);
3828 panic("Can't allocate initial namespace");
3829 m = real_mount(mnt);
3833 list_add(&m->mnt_list, &ns->list);
3834 init_task.nsproxy->mnt_ns = ns;
3838 root.dentry = mnt->mnt_root;
3839 mnt->mnt_flags |= MNT_LOCKED;
3841 set_fs_pwd(current->fs, &root);
3842 set_fs_root(current->fs, &root);
3845 void __init mnt_init(void)
3849 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3850 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3852 mount_hashtable = alloc_large_system_hash("Mount-cache",
3853 sizeof(struct hlist_head),
3856 &m_hash_shift, &m_hash_mask, 0, 0);
3857 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3858 sizeof(struct hlist_head),
3861 &mp_hash_shift, &mp_hash_mask, 0, 0);
3863 if (!mount_hashtable || !mountpoint_hashtable)
3864 panic("Failed to allocate mount hash table\n");
3870 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3872 fs_kobj = kobject_create_and_add("fs", NULL);
3874 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3880 void put_mnt_ns(struct mnt_namespace *ns)
3882 if (!atomic_dec_and_test(&ns->count))
3884 drop_collected_mounts(&ns->root->mnt);
3888 struct vfsmount *kern_mount(struct file_system_type *type)
3890 struct vfsmount *mnt;
3891 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3894 * it is a longterm mount, don't release mnt until
3895 * we unmount before file sys is unregistered
3897 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3901 EXPORT_SYMBOL_GPL(kern_mount);
3903 void kern_unmount(struct vfsmount *mnt)
3905 /* release long term mount so mount point can be released */
3906 if (!IS_ERR_OR_NULL(mnt)) {
3907 real_mount(mnt)->mnt_ns = NULL;
3908 synchronize_rcu(); /* yecchhh... */
3912 EXPORT_SYMBOL(kern_unmount);
3914 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
3918 for (i = 0; i < num; i++)
3920 real_mount(mnt[i])->mnt_ns = NULL;
3921 synchronize_rcu_expedited();
3922 for (i = 0; i < num; i++)
3925 EXPORT_SYMBOL(kern_unmount_array);
3927 bool our_mnt(struct vfsmount *mnt)
3929 return check_mnt(real_mount(mnt));
3932 bool current_chrooted(void)
3934 /* Does the current process have a non-standard root */
3935 struct path ns_root;
3936 struct path fs_root;
3939 /* Find the namespace root */
3940 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3941 ns_root.dentry = ns_root.mnt->mnt_root;
3943 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3946 get_fs_root(current->fs, &fs_root);
3948 chrooted = !path_equal(&fs_root, &ns_root);
3956 static bool mnt_already_visible(struct mnt_namespace *ns,
3957 const struct super_block *sb,
3960 int new_flags = *new_mnt_flags;
3962 bool visible = false;
3964 down_read(&namespace_sem);
3966 list_for_each_entry(mnt, &ns->list, mnt_list) {
3967 struct mount *child;
3970 if (mnt_is_cursor(mnt))
3973 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3976 /* This mount is not fully visible if it's root directory
3977 * is not the root directory of the filesystem.
3979 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3982 /* A local view of the mount flags */
3983 mnt_flags = mnt->mnt.mnt_flags;
3985 /* Don't miss readonly hidden in the superblock flags */
3986 if (sb_rdonly(mnt->mnt.mnt_sb))
3987 mnt_flags |= MNT_LOCK_READONLY;
3989 /* Verify the mount flags are equal to or more permissive
3990 * than the proposed new mount.
3992 if ((mnt_flags & MNT_LOCK_READONLY) &&
3993 !(new_flags & MNT_READONLY))
3995 if ((mnt_flags & MNT_LOCK_ATIME) &&
3996 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3999 /* This mount is not fully visible if there are any
4000 * locked child mounts that cover anything except for
4001 * empty directories.
4003 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4004 struct inode *inode = child->mnt_mountpoint->d_inode;
4005 /* Only worry about locked mounts */
4006 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4008 /* Is the directory permanetly empty? */
4009 if (!is_empty_dir_inode(inode))
4012 /* Preserve the locked attributes */
4013 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4021 up_read(&namespace_sem);
4025 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4027 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4028 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4029 unsigned long s_iflags;
4031 if (ns->user_ns == &init_user_ns)
4034 /* Can this filesystem be too revealing? */
4035 s_iflags = sb->s_iflags;
4036 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4039 if ((s_iflags & required_iflags) != required_iflags) {
4040 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4045 return !mnt_already_visible(ns, sb, new_mnt_flags);
4048 bool mnt_may_suid(struct vfsmount *mnt)
4051 * Foreign mounts (accessed via fchdir or through /proc
4052 * symlinks) are always treated as if they are nosuid. This
4053 * prevents namespaces from trusting potentially unsafe
4054 * suid/sgid bits, file caps, or security labels that originate
4055 * in other namespaces.
4057 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4058 current_in_userns(mnt->mnt_sb->s_user_ns);
4061 static struct ns_common *mntns_get(struct task_struct *task)
4063 struct ns_common *ns = NULL;
4064 struct nsproxy *nsproxy;
4067 nsproxy = task->nsproxy;
4069 ns = &nsproxy->mnt_ns->ns;
4070 get_mnt_ns(to_mnt_ns(ns));
4077 static void mntns_put(struct ns_common *ns)
4079 put_mnt_ns(to_mnt_ns(ns));
4082 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4084 struct nsproxy *nsproxy = nsset->nsproxy;
4085 struct fs_struct *fs = nsset->fs;
4086 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4087 struct user_namespace *user_ns = nsset->cred->user_ns;
4091 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4092 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4093 !ns_capable(user_ns, CAP_SYS_ADMIN))
4096 if (is_anon_ns(mnt_ns))
4103 old_mnt_ns = nsproxy->mnt_ns;
4104 nsproxy->mnt_ns = mnt_ns;
4107 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4108 "/", LOOKUP_DOWN, &root);
4110 /* revert to old namespace */
4111 nsproxy->mnt_ns = old_mnt_ns;
4116 put_mnt_ns(old_mnt_ns);
4118 /* Update the pwd and root */
4119 set_fs_pwd(fs, &root);
4120 set_fs_root(fs, &root);
4126 static struct user_namespace *mntns_owner(struct ns_common *ns)
4128 return to_mnt_ns(ns)->user_ns;
4131 const struct proc_ns_operations mntns_operations = {
4133 .type = CLONE_NEWNS,
4136 .install = mntns_install,
4137 .owner = mntns_owner,