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/proc_fs.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 unsigned int attr_set;
79 unsigned int attr_clr;
80 unsigned int propagation;
81 unsigned int lookup_flags;
83 struct user_namespace *mnt_userns;
87 struct kobject *fs_kobj;
88 EXPORT_SYMBOL_GPL(fs_kobj);
91 * vfsmount lock may be taken for read to prevent changes to the
92 * vfsmount hash, ie. during mountpoint lookups or walking back
95 * It should be taken for write in all cases where the vfsmount
96 * tree or hash is modified or when a vfsmount structure is modified.
98 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
100 static inline void lock_mount_hash(void)
102 write_seqlock(&mount_lock);
105 static inline void unlock_mount_hash(void)
107 write_sequnlock(&mount_lock);
110 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
112 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
113 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
114 tmp = tmp + (tmp >> m_hash_shift);
115 return &mount_hashtable[tmp & m_hash_mask];
118 static inline struct hlist_head *mp_hash(struct dentry *dentry)
120 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
121 tmp = tmp + (tmp >> mp_hash_shift);
122 return &mountpoint_hashtable[tmp & mp_hash_mask];
125 static int mnt_alloc_id(struct mount *mnt)
127 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
135 static void mnt_free_id(struct mount *mnt)
137 ida_free(&mnt_id_ida, mnt->mnt_id);
141 * Allocate a new peer group ID
143 static int mnt_alloc_group_id(struct mount *mnt)
145 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
149 mnt->mnt_group_id = res;
154 * Release a peer group ID
156 void mnt_release_group_id(struct mount *mnt)
158 ida_free(&mnt_group_ida, mnt->mnt_group_id);
159 mnt->mnt_group_id = 0;
163 * vfsmount lock must be held for read
165 static inline void mnt_add_count(struct mount *mnt, int n)
168 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
177 * vfsmount lock must be held for write
179 int mnt_get_count(struct mount *mnt)
185 for_each_possible_cpu(cpu) {
186 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191 return mnt->mnt_count;
195 static struct mount *alloc_vfsmnt(const char *name)
197 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
201 err = mnt_alloc_id(mnt);
206 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
207 if (!mnt->mnt_devname)
212 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
214 goto out_free_devname;
216 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
219 mnt->mnt_writers = 0;
222 INIT_HLIST_NODE(&mnt->mnt_hash);
223 INIT_LIST_HEAD(&mnt->mnt_child);
224 INIT_LIST_HEAD(&mnt->mnt_mounts);
225 INIT_LIST_HEAD(&mnt->mnt_list);
226 INIT_LIST_HEAD(&mnt->mnt_expire);
227 INIT_LIST_HEAD(&mnt->mnt_share);
228 INIT_LIST_HEAD(&mnt->mnt_slave_list);
229 INIT_LIST_HEAD(&mnt->mnt_slave);
230 INIT_HLIST_NODE(&mnt->mnt_mp_list);
231 INIT_LIST_HEAD(&mnt->mnt_umounting);
232 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
233 mnt->mnt.mnt_userns = &init_user_ns;
239 kfree_const(mnt->mnt_devname);
244 kmem_cache_free(mnt_cache, mnt);
249 * Most r/o checks on a fs are for operations that take
250 * discrete amounts of time, like a write() or unlink().
251 * We must keep track of when those operations start
252 * (for permission checks) and when they end, so that
253 * we can determine when writes are able to occur to
257 * __mnt_is_readonly: check whether a mount is read-only
258 * @mnt: the mount to check for its write status
260 * This shouldn't be used directly ouside of the VFS.
261 * It does not guarantee that the filesystem will stay
262 * r/w, just that it is right *now*. This can not and
263 * should not be used in place of IS_RDONLY(inode).
264 * mnt_want/drop_write() will _keep_ the filesystem
267 bool __mnt_is_readonly(struct vfsmount *mnt)
269 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
271 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
273 static inline void mnt_inc_writers(struct mount *mnt)
276 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
282 static inline void mnt_dec_writers(struct mount *mnt)
285 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
291 static unsigned int mnt_get_writers(struct mount *mnt)
294 unsigned int count = 0;
297 for_each_possible_cpu(cpu) {
298 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
303 return mnt->mnt_writers;
307 static int mnt_is_readonly(struct vfsmount *mnt)
309 if (mnt->mnt_sb->s_readonly_remount)
311 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
313 return __mnt_is_readonly(mnt);
317 * Most r/o & frozen checks on a fs are for operations that take discrete
318 * amounts of time, like a write() or unlink(). We must keep track of when
319 * those operations start (for permission checks) and when they end, so that we
320 * can determine when writes are able to occur to a filesystem.
323 * __mnt_want_write - get write access to a mount without freeze protection
324 * @m: the mount on which to take a write
326 * This tells the low-level filesystem that a write is about to be performed to
327 * it, and makes sure that writes are allowed (mnt it read-write) before
328 * returning success. This operation does not protect against filesystem being
329 * frozen. When the write operation is finished, __mnt_drop_write() must be
330 * called. This is effectively a refcount.
332 int __mnt_want_write(struct vfsmount *m)
334 struct mount *mnt = real_mount(m);
338 mnt_inc_writers(mnt);
340 * The store to mnt_inc_writers must be visible before we pass
341 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
342 * incremented count after it has set MNT_WRITE_HOLD.
345 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
348 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
349 * be set to match its requirements. So we must not load that until
350 * MNT_WRITE_HOLD is cleared.
353 if (mnt_is_readonly(m)) {
354 mnt_dec_writers(mnt);
363 * mnt_want_write - get write access to a mount
364 * @m: the mount on which to take a write
366 * This tells the low-level filesystem that a write is about to be performed to
367 * it, and makes sure that writes are allowed (mount is read-write, filesystem
368 * is not frozen) before returning success. When the write operation is
369 * finished, mnt_drop_write() must be called. This is effectively a refcount.
371 int mnt_want_write(struct vfsmount *m)
375 sb_start_write(m->mnt_sb);
376 ret = __mnt_want_write(m);
378 sb_end_write(m->mnt_sb);
381 EXPORT_SYMBOL_GPL(mnt_want_write);
384 * __mnt_want_write_file - get write access to a file's mount
385 * @file: the file who's mount on which to take a write
387 * This is like __mnt_want_write, but if the file is already open for writing it
388 * skips incrementing mnt_writers (since the open file already has a reference)
389 * and instead only does the check for emergency r/o remounts. This must be
390 * paired with __mnt_drop_write_file.
392 int __mnt_want_write_file(struct file *file)
394 if (file->f_mode & FMODE_WRITER) {
396 * Superblock may have become readonly while there are still
397 * writable fd's, e.g. due to a fs error with errors=remount-ro
399 if (__mnt_is_readonly(file->f_path.mnt))
403 return __mnt_want_write(file->f_path.mnt);
407 * mnt_want_write_file - get write access to a file's mount
408 * @file: the file who's mount on which to take a write
410 * This is like mnt_want_write, but if the file is already open for writing it
411 * skips incrementing mnt_writers (since the open file already has a reference)
412 * and instead only does the freeze protection and the check for emergency r/o
413 * remounts. This must be paired with mnt_drop_write_file.
415 int mnt_want_write_file(struct file *file)
419 sb_start_write(file_inode(file)->i_sb);
420 ret = __mnt_want_write_file(file);
422 sb_end_write(file_inode(file)->i_sb);
425 EXPORT_SYMBOL_GPL(mnt_want_write_file);
428 * __mnt_drop_write - give up write access to a mount
429 * @mnt: the mount on which to give up write access
431 * Tells the low-level filesystem that we are done
432 * performing writes to it. Must be matched with
433 * __mnt_want_write() call above.
435 void __mnt_drop_write(struct vfsmount *mnt)
438 mnt_dec_writers(real_mount(mnt));
443 * mnt_drop_write - give up write access to a mount
444 * @mnt: the mount on which to give up write access
446 * Tells the low-level filesystem that we are done performing writes to it and
447 * also allows filesystem to be frozen again. Must be matched with
448 * mnt_want_write() call above.
450 void mnt_drop_write(struct vfsmount *mnt)
452 __mnt_drop_write(mnt);
453 sb_end_write(mnt->mnt_sb);
455 EXPORT_SYMBOL_GPL(mnt_drop_write);
457 void __mnt_drop_write_file(struct file *file)
459 if (!(file->f_mode & FMODE_WRITER))
460 __mnt_drop_write(file->f_path.mnt);
463 void mnt_drop_write_file(struct file *file)
465 __mnt_drop_write_file(file);
466 sb_end_write(file_inode(file)->i_sb);
468 EXPORT_SYMBOL(mnt_drop_write_file);
470 static inline int mnt_hold_writers(struct mount *mnt)
472 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
474 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
475 * should be visible before we do.
480 * With writers on hold, if this value is zero, then there are
481 * definitely no active writers (although held writers may subsequently
482 * increment the count, they'll have to wait, and decrement it after
483 * seeing MNT_READONLY).
485 * It is OK to have counter incremented on one CPU and decremented on
486 * another: the sum will add up correctly. The danger would be when we
487 * sum up each counter, if we read a counter before it is incremented,
488 * but then read another CPU's count which it has been subsequently
489 * decremented from -- we would see more decrements than we should.
490 * MNT_WRITE_HOLD protects against this scenario, because
491 * mnt_want_write first increments count, then smp_mb, then spins on
492 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
493 * we're counting up here.
495 if (mnt_get_writers(mnt) > 0)
501 static inline void mnt_unhold_writers(struct mount *mnt)
504 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
505 * that become unheld will see MNT_READONLY.
508 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
511 static int mnt_make_readonly(struct mount *mnt)
515 ret = mnt_hold_writers(mnt);
517 mnt->mnt.mnt_flags |= MNT_READONLY;
518 mnt_unhold_writers(mnt);
522 int sb_prepare_remount_readonly(struct super_block *sb)
527 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
528 if (atomic_long_read(&sb->s_remove_count))
532 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
533 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
534 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
536 if (mnt_get_writers(mnt) > 0) {
542 if (!err && atomic_long_read(&sb->s_remove_count))
546 sb->s_readonly_remount = 1;
549 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
550 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
551 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
558 static void free_vfsmnt(struct mount *mnt)
560 struct user_namespace *mnt_userns;
562 mnt_userns = mnt_user_ns(&mnt->mnt);
563 if (mnt_userns != &init_user_ns)
564 put_user_ns(mnt_userns);
565 kfree_const(mnt->mnt_devname);
567 free_percpu(mnt->mnt_pcp);
569 kmem_cache_free(mnt_cache, mnt);
572 static void delayed_free_vfsmnt(struct rcu_head *head)
574 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
577 /* call under rcu_read_lock */
578 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
581 if (read_seqretry(&mount_lock, seq))
585 mnt = real_mount(bastard);
586 mnt_add_count(mnt, 1);
587 smp_mb(); // see mntput_no_expire()
588 if (likely(!read_seqretry(&mount_lock, seq)))
590 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
591 mnt_add_count(mnt, -1);
595 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
596 mnt_add_count(mnt, -1);
601 /* caller will mntput() */
605 /* call under rcu_read_lock */
606 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
608 int res = __legitimize_mnt(bastard, seq);
611 if (unlikely(res < 0)) {
620 * find the first mount at @dentry on vfsmount @mnt.
621 * call under rcu_read_lock()
623 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
625 struct hlist_head *head = m_hash(mnt, dentry);
628 hlist_for_each_entry_rcu(p, head, mnt_hash)
629 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
635 * lookup_mnt - Return the first child mount mounted at path
637 * "First" means first mounted chronologically. If you create the
640 * mount /dev/sda1 /mnt
641 * mount /dev/sda2 /mnt
642 * mount /dev/sda3 /mnt
644 * Then lookup_mnt() on the base /mnt dentry in the root mount will
645 * return successively the root dentry and vfsmount of /dev/sda1, then
646 * /dev/sda2, then /dev/sda3, then NULL.
648 * lookup_mnt takes a reference to the found vfsmount.
650 struct vfsmount *lookup_mnt(const struct path *path)
652 struct mount *child_mnt;
658 seq = read_seqbegin(&mount_lock);
659 child_mnt = __lookup_mnt(path->mnt, path->dentry);
660 m = child_mnt ? &child_mnt->mnt : NULL;
661 } while (!legitimize_mnt(m, seq));
666 static inline void lock_ns_list(struct mnt_namespace *ns)
668 spin_lock(&ns->ns_lock);
671 static inline void unlock_ns_list(struct mnt_namespace *ns)
673 spin_unlock(&ns->ns_lock);
676 static inline bool mnt_is_cursor(struct mount *mnt)
678 return mnt->mnt.mnt_flags & MNT_CURSOR;
682 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
683 * current mount namespace.
685 * The common case is dentries are not mountpoints at all and that
686 * test is handled inline. For the slow case when we are actually
687 * dealing with a mountpoint of some kind, walk through all of the
688 * mounts in the current mount namespace and test to see if the dentry
691 * The mount_hashtable is not usable in the context because we
692 * need to identify all mounts that may be in the current mount
693 * namespace not just a mount that happens to have some specified
696 bool __is_local_mountpoint(struct dentry *dentry)
698 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
700 bool is_covered = false;
702 down_read(&namespace_sem);
704 list_for_each_entry(mnt, &ns->list, mnt_list) {
705 if (mnt_is_cursor(mnt))
707 is_covered = (mnt->mnt_mountpoint == dentry);
712 up_read(&namespace_sem);
717 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
719 struct hlist_head *chain = mp_hash(dentry);
720 struct mountpoint *mp;
722 hlist_for_each_entry(mp, chain, m_hash) {
723 if (mp->m_dentry == dentry) {
731 static struct mountpoint *get_mountpoint(struct dentry *dentry)
733 struct mountpoint *mp, *new = NULL;
736 if (d_mountpoint(dentry)) {
737 /* might be worth a WARN_ON() */
738 if (d_unlinked(dentry))
739 return ERR_PTR(-ENOENT);
741 read_seqlock_excl(&mount_lock);
742 mp = lookup_mountpoint(dentry);
743 read_sequnlock_excl(&mount_lock);
749 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
751 return ERR_PTR(-ENOMEM);
754 /* Exactly one processes may set d_mounted */
755 ret = d_set_mounted(dentry);
757 /* Someone else set d_mounted? */
761 /* The dentry is not available as a mountpoint? */
766 /* Add the new mountpoint to the hash table */
767 read_seqlock_excl(&mount_lock);
768 new->m_dentry = dget(dentry);
770 hlist_add_head(&new->m_hash, mp_hash(dentry));
771 INIT_HLIST_HEAD(&new->m_list);
772 read_sequnlock_excl(&mount_lock);
782 * vfsmount lock must be held. Additionally, the caller is responsible
783 * for serializing calls for given disposal list.
785 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
787 if (!--mp->m_count) {
788 struct dentry *dentry = mp->m_dentry;
789 BUG_ON(!hlist_empty(&mp->m_list));
790 spin_lock(&dentry->d_lock);
791 dentry->d_flags &= ~DCACHE_MOUNTED;
792 spin_unlock(&dentry->d_lock);
793 dput_to_list(dentry, list);
794 hlist_del(&mp->m_hash);
799 /* called with namespace_lock and vfsmount lock */
800 static void put_mountpoint(struct mountpoint *mp)
802 __put_mountpoint(mp, &ex_mountpoints);
805 static inline int check_mnt(struct mount *mnt)
807 return mnt->mnt_ns == current->nsproxy->mnt_ns;
811 * vfsmount lock must be held for write
813 static void touch_mnt_namespace(struct mnt_namespace *ns)
817 wake_up_interruptible(&ns->poll);
822 * vfsmount lock must be held for write
824 static void __touch_mnt_namespace(struct mnt_namespace *ns)
826 if (ns && ns->event != event) {
828 wake_up_interruptible(&ns->poll);
833 * vfsmount lock must be held for write
835 static struct mountpoint *unhash_mnt(struct mount *mnt)
837 struct mountpoint *mp;
838 mnt->mnt_parent = mnt;
839 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
840 list_del_init(&mnt->mnt_child);
841 hlist_del_init_rcu(&mnt->mnt_hash);
842 hlist_del_init(&mnt->mnt_mp_list);
849 * vfsmount lock must be held for write
851 static void umount_mnt(struct mount *mnt)
853 put_mountpoint(unhash_mnt(mnt));
857 * vfsmount lock must be held for write
859 void mnt_set_mountpoint(struct mount *mnt,
860 struct mountpoint *mp,
861 struct mount *child_mnt)
864 mnt_add_count(mnt, 1); /* essentially, that's mntget */
865 child_mnt->mnt_mountpoint = mp->m_dentry;
866 child_mnt->mnt_parent = mnt;
867 child_mnt->mnt_mp = mp;
868 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
871 static void __attach_mnt(struct mount *mnt, struct mount *parent)
873 hlist_add_head_rcu(&mnt->mnt_hash,
874 m_hash(&parent->mnt, mnt->mnt_mountpoint));
875 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
879 * vfsmount lock must be held for write
881 static void attach_mnt(struct mount *mnt,
882 struct mount *parent,
883 struct mountpoint *mp)
885 mnt_set_mountpoint(parent, mp, mnt);
886 __attach_mnt(mnt, parent);
889 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
891 struct mountpoint *old_mp = mnt->mnt_mp;
892 struct mount *old_parent = mnt->mnt_parent;
894 list_del_init(&mnt->mnt_child);
895 hlist_del_init(&mnt->mnt_mp_list);
896 hlist_del_init_rcu(&mnt->mnt_hash);
898 attach_mnt(mnt, parent, mp);
900 put_mountpoint(old_mp);
901 mnt_add_count(old_parent, -1);
905 * vfsmount lock must be held for write
907 static void commit_tree(struct mount *mnt)
909 struct mount *parent = mnt->mnt_parent;
912 struct mnt_namespace *n = parent->mnt_ns;
914 BUG_ON(parent == mnt);
916 list_add_tail(&head, &mnt->mnt_list);
917 list_for_each_entry(m, &head, mnt_list)
920 list_splice(&head, n->list.prev);
922 n->mounts += n->pending_mounts;
923 n->pending_mounts = 0;
925 __attach_mnt(mnt, parent);
926 touch_mnt_namespace(n);
929 static struct mount *next_mnt(struct mount *p, struct mount *root)
931 struct list_head *next = p->mnt_mounts.next;
932 if (next == &p->mnt_mounts) {
936 next = p->mnt_child.next;
937 if (next != &p->mnt_parent->mnt_mounts)
942 return list_entry(next, struct mount, mnt_child);
945 static struct mount *skip_mnt_tree(struct mount *p)
947 struct list_head *prev = p->mnt_mounts.prev;
948 while (prev != &p->mnt_mounts) {
949 p = list_entry(prev, struct mount, mnt_child);
950 prev = p->mnt_mounts.prev;
956 * vfs_create_mount - Create a mount for a configured superblock
957 * @fc: The configuration context with the superblock attached
959 * Create a mount to an already configured superblock. If necessary, the
960 * caller should invoke vfs_get_tree() before calling this.
962 * Note that this does not attach the mount to anything.
964 struct vfsmount *vfs_create_mount(struct fs_context *fc)
969 return ERR_PTR(-EINVAL);
971 mnt = alloc_vfsmnt(fc->source ?: "none");
973 return ERR_PTR(-ENOMEM);
975 if (fc->sb_flags & SB_KERNMOUNT)
976 mnt->mnt.mnt_flags = MNT_INTERNAL;
978 atomic_inc(&fc->root->d_sb->s_active);
979 mnt->mnt.mnt_sb = fc->root->d_sb;
980 mnt->mnt.mnt_root = dget(fc->root);
981 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
982 mnt->mnt_parent = mnt;
985 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
989 EXPORT_SYMBOL(vfs_create_mount);
991 struct vfsmount *fc_mount(struct fs_context *fc)
993 int err = vfs_get_tree(fc);
995 up_write(&fc->root->d_sb->s_umount);
996 return vfs_create_mount(fc);
1000 EXPORT_SYMBOL(fc_mount);
1002 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1003 int flags, const char *name,
1006 struct fs_context *fc;
1007 struct vfsmount *mnt;
1011 return ERR_PTR(-EINVAL);
1013 fc = fs_context_for_mount(type, flags);
1015 return ERR_CAST(fc);
1018 ret = vfs_parse_fs_string(fc, "source",
1019 name, strlen(name));
1021 ret = parse_monolithic_mount_data(fc, data);
1030 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1033 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1034 const char *name, void *data)
1036 /* Until it is worked out how to pass the user namespace
1037 * through from the parent mount to the submount don't support
1038 * unprivileged mounts with submounts.
1040 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1041 return ERR_PTR(-EPERM);
1043 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1045 EXPORT_SYMBOL_GPL(vfs_submount);
1047 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1050 struct super_block *sb = old->mnt.mnt_sb;
1054 mnt = alloc_vfsmnt(old->mnt_devname);
1056 return ERR_PTR(-ENOMEM);
1058 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1059 mnt->mnt_group_id = 0; /* not a peer of original */
1061 mnt->mnt_group_id = old->mnt_group_id;
1063 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1064 err = mnt_alloc_group_id(mnt);
1069 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1070 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1072 atomic_inc(&sb->s_active);
1073 mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1074 if (mnt->mnt.mnt_userns != &init_user_ns)
1075 mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1076 mnt->mnt.mnt_sb = sb;
1077 mnt->mnt.mnt_root = dget(root);
1078 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1079 mnt->mnt_parent = mnt;
1081 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1082 unlock_mount_hash();
1084 if ((flag & CL_SLAVE) ||
1085 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1086 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1087 mnt->mnt_master = old;
1088 CLEAR_MNT_SHARED(mnt);
1089 } else if (!(flag & CL_PRIVATE)) {
1090 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1091 list_add(&mnt->mnt_share, &old->mnt_share);
1092 if (IS_MNT_SLAVE(old))
1093 list_add(&mnt->mnt_slave, &old->mnt_slave);
1094 mnt->mnt_master = old->mnt_master;
1096 CLEAR_MNT_SHARED(mnt);
1098 if (flag & CL_MAKE_SHARED)
1099 set_mnt_shared(mnt);
1101 /* stick the duplicate mount on the same expiry list
1102 * as the original if that was on one */
1103 if (flag & CL_EXPIRE) {
1104 if (!list_empty(&old->mnt_expire))
1105 list_add(&mnt->mnt_expire, &old->mnt_expire);
1113 return ERR_PTR(err);
1116 static void cleanup_mnt(struct mount *mnt)
1118 struct hlist_node *p;
1121 * The warning here probably indicates that somebody messed
1122 * up a mnt_want/drop_write() pair. If this happens, the
1123 * filesystem was probably unable to make r/w->r/o transitions.
1124 * The locking used to deal with mnt_count decrement provides barriers,
1125 * so mnt_get_writers() below is safe.
1127 WARN_ON(mnt_get_writers(mnt));
1128 if (unlikely(mnt->mnt_pins.first))
1130 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1131 hlist_del(&m->mnt_umount);
1134 fsnotify_vfsmount_delete(&mnt->mnt);
1135 dput(mnt->mnt.mnt_root);
1136 deactivate_super(mnt->mnt.mnt_sb);
1138 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1141 static void __cleanup_mnt(struct rcu_head *head)
1143 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1146 static LLIST_HEAD(delayed_mntput_list);
1147 static void delayed_mntput(struct work_struct *unused)
1149 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1150 struct mount *m, *t;
1152 llist_for_each_entry_safe(m, t, node, mnt_llist)
1155 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1157 static void mntput_no_expire(struct mount *mnt)
1163 if (likely(READ_ONCE(mnt->mnt_ns))) {
1165 * Since we don't do lock_mount_hash() here,
1166 * ->mnt_ns can change under us. However, if it's
1167 * non-NULL, then there's a reference that won't
1168 * be dropped until after an RCU delay done after
1169 * turning ->mnt_ns NULL. So if we observe it
1170 * non-NULL under rcu_read_lock(), the reference
1171 * we are dropping is not the final one.
1173 mnt_add_count(mnt, -1);
1179 * make sure that if __legitimize_mnt() has not seen us grab
1180 * mount_lock, we'll see their refcount increment here.
1183 mnt_add_count(mnt, -1);
1184 count = mnt_get_count(mnt);
1188 unlock_mount_hash();
1191 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1193 unlock_mount_hash();
1196 mnt->mnt.mnt_flags |= MNT_DOOMED;
1199 list_del(&mnt->mnt_instance);
1201 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1202 struct mount *p, *tmp;
1203 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1204 __put_mountpoint(unhash_mnt(p), &list);
1205 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1208 unlock_mount_hash();
1209 shrink_dentry_list(&list);
1211 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1212 struct task_struct *task = current;
1213 if (likely(!(task->flags & PF_KTHREAD))) {
1214 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1215 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1218 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1219 schedule_delayed_work(&delayed_mntput_work, 1);
1225 void mntput(struct vfsmount *mnt)
1228 struct mount *m = real_mount(mnt);
1229 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1230 if (unlikely(m->mnt_expiry_mark))
1231 m->mnt_expiry_mark = 0;
1232 mntput_no_expire(m);
1235 EXPORT_SYMBOL(mntput);
1237 struct vfsmount *mntget(struct vfsmount *mnt)
1240 mnt_add_count(real_mount(mnt), 1);
1243 EXPORT_SYMBOL(mntget);
1246 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1247 * @path: path to check
1249 * d_mountpoint() can only be used reliably to establish if a dentry is
1250 * not mounted in any namespace and that common case is handled inline.
1251 * d_mountpoint() isn't aware of the possibility there may be multiple
1252 * mounts using a given dentry in a different namespace. This function
1253 * checks if the passed in path is a mountpoint rather than the dentry
1256 bool path_is_mountpoint(const struct path *path)
1261 if (!d_mountpoint(path->dentry))
1266 seq = read_seqbegin(&mount_lock);
1267 res = __path_is_mountpoint(path);
1268 } while (read_seqretry(&mount_lock, seq));
1273 EXPORT_SYMBOL(path_is_mountpoint);
1275 struct vfsmount *mnt_clone_internal(const struct path *path)
1278 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1281 p->mnt.mnt_flags |= MNT_INTERNAL;
1285 #ifdef CONFIG_PROC_FS
1286 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1287 struct list_head *p)
1289 struct mount *mnt, *ret = NULL;
1292 list_for_each_continue(p, &ns->list) {
1293 mnt = list_entry(p, typeof(*mnt), mnt_list);
1294 if (!mnt_is_cursor(mnt)) {
1304 /* iterator; we want it to have access to namespace_sem, thus here... */
1305 static void *m_start(struct seq_file *m, loff_t *pos)
1307 struct proc_mounts *p = m->private;
1308 struct list_head *prev;
1310 down_read(&namespace_sem);
1312 prev = &p->ns->list;
1314 prev = &p->cursor.mnt_list;
1316 /* Read after we'd reached the end? */
1317 if (list_empty(prev))
1321 return mnt_list_next(p->ns, prev);
1324 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1326 struct proc_mounts *p = m->private;
1327 struct mount *mnt = v;
1330 return mnt_list_next(p->ns, &mnt->mnt_list);
1333 static void m_stop(struct seq_file *m, void *v)
1335 struct proc_mounts *p = m->private;
1336 struct mount *mnt = v;
1338 lock_ns_list(p->ns);
1340 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1342 list_del_init(&p->cursor.mnt_list);
1343 unlock_ns_list(p->ns);
1344 up_read(&namespace_sem);
1347 static int m_show(struct seq_file *m, void *v)
1349 struct proc_mounts *p = m->private;
1350 struct mount *r = v;
1351 return p->show(m, &r->mnt);
1354 const struct seq_operations mounts_op = {
1361 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1363 down_read(&namespace_sem);
1365 list_del(&cursor->mnt_list);
1367 up_read(&namespace_sem);
1369 #endif /* CONFIG_PROC_FS */
1372 * may_umount_tree - check if a mount tree is busy
1373 * @m: root of mount tree
1375 * This is called to check if a tree of mounts has any
1376 * open files, pwds, chroots or sub mounts that are
1379 int may_umount_tree(struct vfsmount *m)
1381 struct mount *mnt = real_mount(m);
1382 int actual_refs = 0;
1383 int minimum_refs = 0;
1387 /* write lock needed for mnt_get_count */
1389 for (p = mnt; p; p = next_mnt(p, mnt)) {
1390 actual_refs += mnt_get_count(p);
1393 unlock_mount_hash();
1395 if (actual_refs > minimum_refs)
1401 EXPORT_SYMBOL(may_umount_tree);
1404 * may_umount - check if a mount point is busy
1405 * @mnt: root of mount
1407 * This is called to check if a mount point has any
1408 * open files, pwds, chroots or sub mounts. If the
1409 * mount has sub mounts this will return busy
1410 * regardless of whether the sub mounts are busy.
1412 * Doesn't take quota and stuff into account. IOW, in some cases it will
1413 * give false negatives. The main reason why it's here is that we need
1414 * a non-destructive way to look for easily umountable filesystems.
1416 int may_umount(struct vfsmount *mnt)
1419 down_read(&namespace_sem);
1421 if (propagate_mount_busy(real_mount(mnt), 2))
1423 unlock_mount_hash();
1424 up_read(&namespace_sem);
1428 EXPORT_SYMBOL(may_umount);
1430 static void namespace_unlock(void)
1432 struct hlist_head head;
1433 struct hlist_node *p;
1437 hlist_move_list(&unmounted, &head);
1438 list_splice_init(&ex_mountpoints, &list);
1440 up_write(&namespace_sem);
1442 shrink_dentry_list(&list);
1444 if (likely(hlist_empty(&head)))
1447 synchronize_rcu_expedited();
1449 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1450 hlist_del(&m->mnt_umount);
1455 static inline void namespace_lock(void)
1457 down_write(&namespace_sem);
1460 enum umount_tree_flags {
1462 UMOUNT_PROPAGATE = 2,
1463 UMOUNT_CONNECTED = 4,
1466 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1468 /* Leaving mounts connected is only valid for lazy umounts */
1469 if (how & UMOUNT_SYNC)
1472 /* A mount without a parent has nothing to be connected to */
1473 if (!mnt_has_parent(mnt))
1476 /* Because the reference counting rules change when mounts are
1477 * unmounted and connected, umounted mounts may not be
1478 * connected to mounted mounts.
1480 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1483 /* Has it been requested that the mount remain connected? */
1484 if (how & UMOUNT_CONNECTED)
1487 /* Is the mount locked such that it needs to remain connected? */
1488 if (IS_MNT_LOCKED(mnt))
1491 /* By default disconnect the mount */
1496 * mount_lock must be held
1497 * namespace_sem must be held for write
1499 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1501 LIST_HEAD(tmp_list);
1504 if (how & UMOUNT_PROPAGATE)
1505 propagate_mount_unlock(mnt);
1507 /* Gather the mounts to umount */
1508 for (p = mnt; p; p = next_mnt(p, mnt)) {
1509 p->mnt.mnt_flags |= MNT_UMOUNT;
1510 list_move(&p->mnt_list, &tmp_list);
1513 /* Hide the mounts from mnt_mounts */
1514 list_for_each_entry(p, &tmp_list, mnt_list) {
1515 list_del_init(&p->mnt_child);
1518 /* Add propogated mounts to the tmp_list */
1519 if (how & UMOUNT_PROPAGATE)
1520 propagate_umount(&tmp_list);
1522 while (!list_empty(&tmp_list)) {
1523 struct mnt_namespace *ns;
1525 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1526 list_del_init(&p->mnt_expire);
1527 list_del_init(&p->mnt_list);
1531 __touch_mnt_namespace(ns);
1534 if (how & UMOUNT_SYNC)
1535 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1537 disconnect = disconnect_mount(p, how);
1538 if (mnt_has_parent(p)) {
1539 mnt_add_count(p->mnt_parent, -1);
1541 /* Don't forget about p */
1542 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1547 change_mnt_propagation(p, MS_PRIVATE);
1549 hlist_add_head(&p->mnt_umount, &unmounted);
1553 static void shrink_submounts(struct mount *mnt);
1555 static int do_umount_root(struct super_block *sb)
1559 down_write(&sb->s_umount);
1560 if (!sb_rdonly(sb)) {
1561 struct fs_context *fc;
1563 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1568 ret = parse_monolithic_mount_data(fc, NULL);
1570 ret = reconfigure_super(fc);
1574 up_write(&sb->s_umount);
1578 static int do_umount(struct mount *mnt, int flags)
1580 struct super_block *sb = mnt->mnt.mnt_sb;
1583 retval = security_sb_umount(&mnt->mnt, flags);
1588 * Allow userspace to request a mountpoint be expired rather than
1589 * unmounting unconditionally. Unmount only happens if:
1590 * (1) the mark is already set (the mark is cleared by mntput())
1591 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1593 if (flags & MNT_EXPIRE) {
1594 if (&mnt->mnt == current->fs->root.mnt ||
1595 flags & (MNT_FORCE | MNT_DETACH))
1599 * probably don't strictly need the lock here if we examined
1600 * all race cases, but it's a slowpath.
1603 if (mnt_get_count(mnt) != 2) {
1604 unlock_mount_hash();
1607 unlock_mount_hash();
1609 if (!xchg(&mnt->mnt_expiry_mark, 1))
1614 * If we may have to abort operations to get out of this
1615 * mount, and they will themselves hold resources we must
1616 * allow the fs to do things. In the Unix tradition of
1617 * 'Gee thats tricky lets do it in userspace' the umount_begin
1618 * might fail to complete on the first run through as other tasks
1619 * must return, and the like. Thats for the mount program to worry
1620 * about for the moment.
1623 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1624 sb->s_op->umount_begin(sb);
1628 * No sense to grab the lock for this test, but test itself looks
1629 * somewhat bogus. Suggestions for better replacement?
1630 * Ho-hum... In principle, we might treat that as umount + switch
1631 * to rootfs. GC would eventually take care of the old vfsmount.
1632 * Actually it makes sense, especially if rootfs would contain a
1633 * /reboot - static binary that would close all descriptors and
1634 * call reboot(9). Then init(8) could umount root and exec /reboot.
1636 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1638 * Special case for "unmounting" root ...
1639 * we just try to remount it readonly.
1641 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1643 return do_umount_root(sb);
1649 /* Recheck MNT_LOCKED with the locks held */
1651 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1655 if (flags & MNT_DETACH) {
1656 if (!list_empty(&mnt->mnt_list))
1657 umount_tree(mnt, UMOUNT_PROPAGATE);
1660 shrink_submounts(mnt);
1662 if (!propagate_mount_busy(mnt, 2)) {
1663 if (!list_empty(&mnt->mnt_list))
1664 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1669 unlock_mount_hash();
1675 * __detach_mounts - lazily unmount all mounts on the specified dentry
1677 * During unlink, rmdir, and d_drop it is possible to loose the path
1678 * to an existing mountpoint, and wind up leaking the mount.
1679 * detach_mounts allows lazily unmounting those mounts instead of
1682 * The caller may hold dentry->d_inode->i_mutex.
1684 void __detach_mounts(struct dentry *dentry)
1686 struct mountpoint *mp;
1691 mp = lookup_mountpoint(dentry);
1696 while (!hlist_empty(&mp->m_list)) {
1697 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1698 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1700 hlist_add_head(&mnt->mnt_umount, &unmounted);
1702 else umount_tree(mnt, UMOUNT_CONNECTED);
1706 unlock_mount_hash();
1711 * Is the caller allowed to modify his namespace?
1713 static inline bool may_mount(void)
1715 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1718 static void warn_mandlock(void)
1720 pr_warn_once("=======================================================\n"
1721 "WARNING: The mand mount option has been deprecated and\n"
1722 " and is ignored by this kernel. Remove the mand\n"
1723 " option from the mount to silence this warning.\n"
1724 "=======================================================\n");
1727 static int can_umount(const struct path *path, int flags)
1729 struct mount *mnt = real_mount(path->mnt);
1733 if (path->dentry != path->mnt->mnt_root)
1735 if (!check_mnt(mnt))
1737 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1739 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1744 // caller is responsible for flags being sane
1745 int path_umount(struct path *path, int flags)
1747 struct mount *mnt = real_mount(path->mnt);
1750 ret = can_umount(path, flags);
1752 ret = do_umount(mnt, flags);
1754 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1756 mntput_no_expire(mnt);
1760 static int ksys_umount(char __user *name, int flags)
1762 int lookup_flags = LOOKUP_MOUNTPOINT;
1766 // basic validity checks done first
1767 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1770 if (!(flags & UMOUNT_NOFOLLOW))
1771 lookup_flags |= LOOKUP_FOLLOW;
1772 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1775 return path_umount(&path, flags);
1778 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1780 return ksys_umount(name, flags);
1783 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1786 * The 2.0 compatible umount. No flags.
1788 SYSCALL_DEFINE1(oldumount, char __user *, name)
1790 return ksys_umount(name, 0);
1795 static bool is_mnt_ns_file(struct dentry *dentry)
1797 /* Is this a proxy for a mount namespace? */
1798 return dentry->d_op == &ns_dentry_operations &&
1799 dentry->d_fsdata == &mntns_operations;
1802 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1804 return container_of(ns, struct mnt_namespace, ns);
1807 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1812 static bool mnt_ns_loop(struct dentry *dentry)
1814 /* Could bind mounting the mount namespace inode cause a
1815 * mount namespace loop?
1817 struct mnt_namespace *mnt_ns;
1818 if (!is_mnt_ns_file(dentry))
1821 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1822 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1825 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1828 struct mount *res, *p, *q, *r, *parent;
1830 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1831 return ERR_PTR(-EINVAL);
1833 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1834 return ERR_PTR(-EINVAL);
1836 res = q = clone_mnt(mnt, dentry, flag);
1840 q->mnt_mountpoint = mnt->mnt_mountpoint;
1843 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1845 if (!is_subdir(r->mnt_mountpoint, dentry))
1848 for (s = r; s; s = next_mnt(s, r)) {
1849 if (!(flag & CL_COPY_UNBINDABLE) &&
1850 IS_MNT_UNBINDABLE(s)) {
1851 if (s->mnt.mnt_flags & MNT_LOCKED) {
1852 /* Both unbindable and locked. */
1853 q = ERR_PTR(-EPERM);
1856 s = skip_mnt_tree(s);
1860 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1861 is_mnt_ns_file(s->mnt.mnt_root)) {
1862 s = skip_mnt_tree(s);
1865 while (p != s->mnt_parent) {
1871 q = clone_mnt(p, p->mnt.mnt_root, flag);
1875 list_add_tail(&q->mnt_list, &res->mnt_list);
1876 attach_mnt(q, parent, p->mnt_mp);
1877 unlock_mount_hash();
1884 umount_tree(res, UMOUNT_SYNC);
1885 unlock_mount_hash();
1890 /* Caller should check returned pointer for errors */
1892 struct vfsmount *collect_mounts(const struct path *path)
1896 if (!check_mnt(real_mount(path->mnt)))
1897 tree = ERR_PTR(-EINVAL);
1899 tree = copy_tree(real_mount(path->mnt), path->dentry,
1900 CL_COPY_ALL | CL_PRIVATE);
1903 return ERR_CAST(tree);
1907 static void free_mnt_ns(struct mnt_namespace *);
1908 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1910 void dissolve_on_fput(struct vfsmount *mnt)
1912 struct mnt_namespace *ns;
1915 ns = real_mount(mnt)->mnt_ns;
1918 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1922 unlock_mount_hash();
1928 void drop_collected_mounts(struct vfsmount *mnt)
1932 umount_tree(real_mount(mnt), 0);
1933 unlock_mount_hash();
1937 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1939 struct mount *child;
1941 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1942 if (!is_subdir(child->mnt_mountpoint, dentry))
1945 if (child->mnt.mnt_flags & MNT_LOCKED)
1952 * clone_private_mount - create a private clone of a path
1953 * @path: path to clone
1955 * This creates a new vfsmount, which will be the clone of @path. The new mount
1956 * will not be attached anywhere in the namespace and will be private (i.e.
1957 * changes to the originating mount won't be propagated into this).
1959 * Release with mntput().
1961 struct vfsmount *clone_private_mount(const struct path *path)
1963 struct mount *old_mnt = real_mount(path->mnt);
1964 struct mount *new_mnt;
1966 down_read(&namespace_sem);
1967 if (IS_MNT_UNBINDABLE(old_mnt))
1970 if (!check_mnt(old_mnt))
1973 if (has_locked_children(old_mnt, path->dentry))
1976 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1977 up_read(&namespace_sem);
1979 if (IS_ERR(new_mnt))
1980 return ERR_CAST(new_mnt);
1982 /* Longterm mount to be removed by kern_unmount*() */
1983 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1985 return &new_mnt->mnt;
1988 up_read(&namespace_sem);
1989 return ERR_PTR(-EINVAL);
1991 EXPORT_SYMBOL_GPL(clone_private_mount);
1993 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1994 struct vfsmount *root)
1997 int res = f(root, arg);
2000 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2001 res = f(&mnt->mnt, arg);
2008 static void lock_mnt_tree(struct mount *mnt)
2012 for (p = mnt; p; p = next_mnt(p, mnt)) {
2013 int flags = p->mnt.mnt_flags;
2014 /* Don't allow unprivileged users to change mount flags */
2015 flags |= MNT_LOCK_ATIME;
2017 if (flags & MNT_READONLY)
2018 flags |= MNT_LOCK_READONLY;
2020 if (flags & MNT_NODEV)
2021 flags |= MNT_LOCK_NODEV;
2023 if (flags & MNT_NOSUID)
2024 flags |= MNT_LOCK_NOSUID;
2026 if (flags & MNT_NOEXEC)
2027 flags |= MNT_LOCK_NOEXEC;
2028 /* Don't allow unprivileged users to reveal what is under a mount */
2029 if (list_empty(&p->mnt_expire))
2030 flags |= MNT_LOCKED;
2031 p->mnt.mnt_flags = flags;
2035 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2039 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2040 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2041 mnt_release_group_id(p);
2045 static int invent_group_ids(struct mount *mnt, bool recurse)
2049 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2050 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2051 int err = mnt_alloc_group_id(p);
2053 cleanup_group_ids(mnt, p);
2062 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2064 unsigned int max = READ_ONCE(sysctl_mount_max);
2065 unsigned int mounts = 0, old, pending, sum;
2068 for (p = mnt; p; p = next_mnt(p, mnt))
2072 pending = ns->pending_mounts;
2073 sum = old + pending;
2077 (mounts > (max - sum)))
2080 ns->pending_mounts = pending + mounts;
2085 * @source_mnt : mount tree to be attached
2086 * @nd : place the mount tree @source_mnt is attached
2087 * @parent_nd : if non-null, detach the source_mnt from its parent and
2088 * store the parent mount and mountpoint dentry.
2089 * (done when source_mnt is moved)
2091 * NOTE: in the table below explains the semantics when a source mount
2092 * of a given type is attached to a destination mount of a given type.
2093 * ---------------------------------------------------------------------------
2094 * | BIND MOUNT OPERATION |
2095 * |**************************************************************************
2096 * | source-->| shared | private | slave | unbindable |
2100 * |**************************************************************************
2101 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2103 * |non-shared| shared (+) | private | slave (*) | invalid |
2104 * ***************************************************************************
2105 * A bind operation clones the source mount and mounts the clone on the
2106 * destination mount.
2108 * (++) the cloned mount is propagated to all the mounts in the propagation
2109 * tree of the destination mount and the cloned mount is added to
2110 * the peer group of the source mount.
2111 * (+) the cloned mount is created under the destination mount and is marked
2112 * as shared. The cloned mount is added to the peer group of the source
2114 * (+++) the mount is propagated to all the mounts in the propagation tree
2115 * of the destination mount and the cloned mount is made slave
2116 * of the same master as that of the source mount. The cloned mount
2117 * is marked as 'shared and slave'.
2118 * (*) the cloned mount is made a slave of the same master as that of the
2121 * ---------------------------------------------------------------------------
2122 * | MOVE MOUNT OPERATION |
2123 * |**************************************************************************
2124 * | source-->| shared | private | slave | unbindable |
2128 * |**************************************************************************
2129 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2131 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2132 * ***************************************************************************
2134 * (+) the mount is moved to the destination. And is then propagated to
2135 * all the mounts in the propagation tree of the destination mount.
2136 * (+*) the mount is moved to the destination.
2137 * (+++) the mount is moved to the destination and is then propagated to
2138 * all the mounts belonging to the destination mount's propagation tree.
2139 * the mount is marked as 'shared and slave'.
2140 * (*) the mount continues to be a slave at the new location.
2142 * if the source mount is a tree, the operations explained above is
2143 * applied to each mount in the tree.
2144 * Must be called without spinlocks held, since this function can sleep
2147 static int attach_recursive_mnt(struct mount *source_mnt,
2148 struct mount *dest_mnt,
2149 struct mountpoint *dest_mp,
2152 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2153 HLIST_HEAD(tree_list);
2154 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2155 struct mountpoint *smp;
2156 struct mount *child, *p;
2157 struct hlist_node *n;
2160 /* Preallocate a mountpoint in case the new mounts need
2161 * to be tucked under other mounts.
2163 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2165 return PTR_ERR(smp);
2167 /* Is there space to add these mounts to the mount namespace? */
2169 err = count_mounts(ns, source_mnt);
2174 if (IS_MNT_SHARED(dest_mnt)) {
2175 err = invent_group_ids(source_mnt, true);
2178 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2181 goto out_cleanup_ids;
2182 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2188 unhash_mnt(source_mnt);
2189 attach_mnt(source_mnt, dest_mnt, dest_mp);
2190 touch_mnt_namespace(source_mnt->mnt_ns);
2192 if (source_mnt->mnt_ns) {
2193 /* move from anon - the caller will destroy */
2194 list_del_init(&source_mnt->mnt_ns->list);
2196 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2197 commit_tree(source_mnt);
2200 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2202 hlist_del_init(&child->mnt_hash);
2203 q = __lookup_mnt(&child->mnt_parent->mnt,
2204 child->mnt_mountpoint);
2206 mnt_change_mountpoint(child, smp, q);
2207 /* Notice when we are propagating across user namespaces */
2208 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2209 lock_mnt_tree(child);
2210 child->mnt.mnt_flags &= ~MNT_LOCKED;
2213 put_mountpoint(smp);
2214 unlock_mount_hash();
2219 while (!hlist_empty(&tree_list)) {
2220 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2221 child->mnt_parent->mnt_ns->pending_mounts = 0;
2222 umount_tree(child, UMOUNT_SYNC);
2224 unlock_mount_hash();
2225 cleanup_group_ids(source_mnt, NULL);
2227 ns->pending_mounts = 0;
2229 read_seqlock_excl(&mount_lock);
2230 put_mountpoint(smp);
2231 read_sequnlock_excl(&mount_lock);
2236 static struct mountpoint *lock_mount(struct path *path)
2238 struct vfsmount *mnt;
2239 struct dentry *dentry = path->dentry;
2241 inode_lock(dentry->d_inode);
2242 if (unlikely(cant_mount(dentry))) {
2243 inode_unlock(dentry->d_inode);
2244 return ERR_PTR(-ENOENT);
2247 mnt = lookup_mnt(path);
2249 struct mountpoint *mp = get_mountpoint(dentry);
2252 inode_unlock(dentry->d_inode);
2258 inode_unlock(path->dentry->d_inode);
2261 dentry = path->dentry = dget(mnt->mnt_root);
2265 static void unlock_mount(struct mountpoint *where)
2267 struct dentry *dentry = where->m_dentry;
2269 read_seqlock_excl(&mount_lock);
2270 put_mountpoint(where);
2271 read_sequnlock_excl(&mount_lock);
2274 inode_unlock(dentry->d_inode);
2277 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2279 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2282 if (d_is_dir(mp->m_dentry) !=
2283 d_is_dir(mnt->mnt.mnt_root))
2286 return attach_recursive_mnt(mnt, p, mp, false);
2290 * Sanity check the flags to change_mnt_propagation.
2293 static int flags_to_propagation_type(int ms_flags)
2295 int type = ms_flags & ~(MS_REC | MS_SILENT);
2297 /* Fail if any non-propagation flags are set */
2298 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2300 /* Only one propagation flag should be set */
2301 if (!is_power_of_2(type))
2307 * recursively change the type of the mountpoint.
2309 static int do_change_type(struct path *path, int ms_flags)
2312 struct mount *mnt = real_mount(path->mnt);
2313 int recurse = ms_flags & MS_REC;
2317 if (path->dentry != path->mnt->mnt_root)
2320 type = flags_to_propagation_type(ms_flags);
2325 if (type == MS_SHARED) {
2326 err = invent_group_ids(mnt, recurse);
2332 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2333 change_mnt_propagation(m, type);
2334 unlock_mount_hash();
2341 static struct mount *__do_loopback(struct path *old_path, int recurse)
2343 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2345 if (IS_MNT_UNBINDABLE(old))
2348 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2351 if (!recurse && has_locked_children(old, old_path->dentry))
2355 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2357 mnt = clone_mnt(old, old_path->dentry, 0);
2360 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2366 * do loopback mount.
2368 static int do_loopback(struct path *path, const char *old_name,
2371 struct path old_path;
2372 struct mount *mnt = NULL, *parent;
2373 struct mountpoint *mp;
2375 if (!old_name || !*old_name)
2377 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2382 if (mnt_ns_loop(old_path.dentry))
2385 mp = lock_mount(path);
2391 parent = real_mount(path->mnt);
2392 if (!check_mnt(parent))
2395 mnt = __do_loopback(&old_path, recurse);
2401 err = graft_tree(mnt, parent, mp);
2404 umount_tree(mnt, UMOUNT_SYNC);
2405 unlock_mount_hash();
2410 path_put(&old_path);
2414 static struct file *open_detached_copy(struct path *path, bool recursive)
2416 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2417 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2418 struct mount *mnt, *p;
2422 return ERR_CAST(ns);
2425 mnt = __do_loopback(path, recursive);
2429 return ERR_CAST(mnt);
2433 for (p = mnt; p; p = next_mnt(p, mnt)) {
2438 list_add_tail(&ns->list, &mnt->mnt_list);
2440 unlock_mount_hash();
2444 path->mnt = &mnt->mnt;
2445 file = dentry_open(path, O_PATH, current_cred());
2447 dissolve_on_fput(path->mnt);
2449 file->f_mode |= FMODE_NEED_UNMOUNT;
2453 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2457 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2458 bool detached = flags & OPEN_TREE_CLONE;
2462 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2464 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2465 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2469 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2472 if (flags & AT_NO_AUTOMOUNT)
2473 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2474 if (flags & AT_SYMLINK_NOFOLLOW)
2475 lookup_flags &= ~LOOKUP_FOLLOW;
2476 if (flags & AT_EMPTY_PATH)
2477 lookup_flags |= LOOKUP_EMPTY;
2479 if (detached && !may_mount())
2482 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2486 error = user_path_at(dfd, filename, lookup_flags, &path);
2487 if (unlikely(error)) {
2488 file = ERR_PTR(error);
2491 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2493 file = dentry_open(&path, O_PATH, current_cred());
2498 return PTR_ERR(file);
2500 fd_install(fd, file);
2505 * Don't allow locked mount flags to be cleared.
2507 * No locks need to be held here while testing the various MNT_LOCK
2508 * flags because those flags can never be cleared once they are set.
2510 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2512 unsigned int fl = mnt->mnt.mnt_flags;
2514 if ((fl & MNT_LOCK_READONLY) &&
2515 !(mnt_flags & MNT_READONLY))
2518 if ((fl & MNT_LOCK_NODEV) &&
2519 !(mnt_flags & MNT_NODEV))
2522 if ((fl & MNT_LOCK_NOSUID) &&
2523 !(mnt_flags & MNT_NOSUID))
2526 if ((fl & MNT_LOCK_NOEXEC) &&
2527 !(mnt_flags & MNT_NOEXEC))
2530 if ((fl & MNT_LOCK_ATIME) &&
2531 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2537 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2539 bool readonly_request = (mnt_flags & MNT_READONLY);
2541 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2544 if (readonly_request)
2545 return mnt_make_readonly(mnt);
2547 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2551 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2553 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2554 mnt->mnt.mnt_flags = mnt_flags;
2555 touch_mnt_namespace(mnt->mnt_ns);
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))
2601 * We're only checking whether the superblock is read-only not
2602 * changing it, so only take down_read(&sb->s_umount).
2604 down_read(&sb->s_umount);
2606 ret = change_mount_ro_state(mnt, mnt_flags);
2608 set_mount_attributes(mnt, mnt_flags);
2609 unlock_mount_hash();
2610 up_read(&sb->s_umount);
2612 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2618 * change filesystem flags. dir should be a physical root of filesystem.
2619 * If you've mounted a non-root directory somewhere and want to do remount
2620 * on it - tough luck.
2622 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2623 int mnt_flags, void *data)
2626 struct super_block *sb = path->mnt->mnt_sb;
2627 struct mount *mnt = real_mount(path->mnt);
2628 struct fs_context *fc;
2630 if (!check_mnt(mnt))
2633 if (path->dentry != path->mnt->mnt_root)
2636 if (!can_change_locked_flags(mnt, mnt_flags))
2639 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2644 err = parse_monolithic_mount_data(fc, data);
2646 down_write(&sb->s_umount);
2648 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2649 err = reconfigure_super(fc);
2652 set_mount_attributes(mnt, mnt_flags);
2653 unlock_mount_hash();
2656 up_write(&sb->s_umount);
2659 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2665 static inline int tree_contains_unbindable(struct mount *mnt)
2668 for (p = mnt; p; p = next_mnt(p, mnt)) {
2669 if (IS_MNT_UNBINDABLE(p))
2676 * Check that there aren't references to earlier/same mount namespaces in the
2677 * specified subtree. Such references can act as pins for mount namespaces
2678 * that aren't checked by the mount-cycle checking code, thereby allowing
2679 * cycles to be made.
2681 static bool check_for_nsfs_mounts(struct mount *subtree)
2687 for (p = subtree; p; p = next_mnt(p, subtree))
2688 if (mnt_ns_loop(p->mnt.mnt_root))
2693 unlock_mount_hash();
2697 static int do_move_mount(struct path *old_path, struct path *new_path)
2699 struct mnt_namespace *ns;
2702 struct mount *parent;
2703 struct mountpoint *mp, *old_mp;
2707 mp = lock_mount(new_path);
2711 old = real_mount(old_path->mnt);
2712 p = real_mount(new_path->mnt);
2713 parent = old->mnt_parent;
2714 attached = mnt_has_parent(old);
2715 old_mp = old->mnt_mp;
2719 /* The mountpoint must be in our namespace. */
2723 /* The thing moved must be mounted... */
2724 if (!is_mounted(&old->mnt))
2727 /* ... and either ours or the root of anon namespace */
2728 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2731 if (old->mnt.mnt_flags & MNT_LOCKED)
2734 if (old_path->dentry != old_path->mnt->mnt_root)
2737 if (d_is_dir(new_path->dentry) !=
2738 d_is_dir(old_path->dentry))
2741 * Don't move a mount residing in a shared parent.
2743 if (attached && IS_MNT_SHARED(parent))
2746 * Don't move a mount tree containing unbindable mounts to a destination
2747 * mount which is shared.
2749 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2752 if (!check_for_nsfs_mounts(old))
2754 for (; mnt_has_parent(p); p = p->mnt_parent)
2758 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2763 /* if the mount is moved, it should no longer be expire
2765 list_del_init(&old->mnt_expire);
2767 put_mountpoint(old_mp);
2772 mntput_no_expire(parent);
2779 static int do_move_mount_old(struct path *path, const char *old_name)
2781 struct path old_path;
2784 if (!old_name || !*old_name)
2787 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2791 err = do_move_mount(&old_path, path);
2792 path_put(&old_path);
2797 * add a mount into a namespace's mount tree
2799 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2800 struct path *path, int mnt_flags)
2802 struct mount *parent = real_mount(path->mnt);
2804 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2806 if (unlikely(!check_mnt(parent))) {
2807 /* that's acceptable only for automounts done in private ns */
2808 if (!(mnt_flags & MNT_SHRINKABLE))
2810 /* ... and for those we'd better have mountpoint still alive */
2811 if (!parent->mnt_ns)
2815 /* Refuse the same filesystem on the same mount point */
2816 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2817 path->mnt->mnt_root == path->dentry)
2820 if (d_is_symlink(newmnt->mnt.mnt_root))
2823 newmnt->mnt.mnt_flags = mnt_flags;
2824 return graft_tree(newmnt, parent, mp);
2827 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2830 * Create a new mount using a superblock configuration and request it
2831 * be added to the namespace tree.
2833 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2834 unsigned int mnt_flags)
2836 struct vfsmount *mnt;
2837 struct mountpoint *mp;
2838 struct super_block *sb = fc->root->d_sb;
2841 error = security_sb_kern_mount(sb);
2842 if (!error && mount_too_revealing(sb, &mnt_flags))
2845 if (unlikely(error)) {
2850 up_write(&sb->s_umount);
2852 mnt = vfs_create_mount(fc);
2854 return PTR_ERR(mnt);
2856 mnt_warn_timestamp_expiry(mountpoint, mnt);
2858 mp = lock_mount(mountpoint);
2863 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2871 * create a new mount for userspace and request it to be added into the
2874 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2875 int mnt_flags, const char *name, void *data)
2877 struct file_system_type *type;
2878 struct fs_context *fc;
2879 const char *subtype = NULL;
2885 type = get_fs_type(fstype);
2889 if (type->fs_flags & FS_HAS_SUBTYPE) {
2890 subtype = strchr(fstype, '.');
2894 put_filesystem(type);
2900 fc = fs_context_for_mount(type, sb_flags);
2901 put_filesystem(type);
2906 err = vfs_parse_fs_string(fc, "subtype",
2907 subtype, strlen(subtype));
2909 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2911 err = parse_monolithic_mount_data(fc, data);
2912 if (!err && !mount_capable(fc))
2915 err = vfs_get_tree(fc);
2917 err = do_new_mount_fc(fc, path, mnt_flags);
2923 int finish_automount(struct vfsmount *m, struct path *path)
2925 struct dentry *dentry = path->dentry;
2926 struct mountpoint *mp;
2935 mnt = real_mount(m);
2936 /* The new mount record should have at least 2 refs to prevent it being
2937 * expired before we get a chance to add it
2939 BUG_ON(mnt_get_count(mnt) < 2);
2941 if (m->mnt_sb == path->mnt->mnt_sb &&
2942 m->mnt_root == dentry) {
2948 * we don't want to use lock_mount() - in this case finding something
2949 * that overmounts our mountpoint to be means "quitely drop what we've
2950 * got", not "try to mount it on top".
2952 inode_lock(dentry->d_inode);
2954 if (unlikely(cant_mount(dentry))) {
2956 goto discard_locked;
2959 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2962 goto discard_locked;
2965 mp = get_mountpoint(dentry);
2968 goto discard_locked;
2971 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2980 inode_unlock(dentry->d_inode);
2982 /* remove m from any expiration list it may be on */
2983 if (!list_empty(&mnt->mnt_expire)) {
2985 list_del_init(&mnt->mnt_expire);
2994 * mnt_set_expiry - Put a mount on an expiration list
2995 * @mnt: The mount to list.
2996 * @expiry_list: The list to add the mount to.
2998 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3002 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3006 EXPORT_SYMBOL(mnt_set_expiry);
3009 * process a list of expirable mountpoints with the intent of discarding any
3010 * mountpoints that aren't in use and haven't been touched since last we came
3013 void mark_mounts_for_expiry(struct list_head *mounts)
3015 struct mount *mnt, *next;
3016 LIST_HEAD(graveyard);
3018 if (list_empty(mounts))
3024 /* extract from the expiration list every vfsmount that matches the
3025 * following criteria:
3026 * - only referenced by its parent vfsmount
3027 * - still marked for expiry (marked on the last call here; marks are
3028 * cleared by mntput())
3030 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3031 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3032 propagate_mount_busy(mnt, 1))
3034 list_move(&mnt->mnt_expire, &graveyard);
3036 while (!list_empty(&graveyard)) {
3037 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3038 touch_mnt_namespace(mnt->mnt_ns);
3039 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3041 unlock_mount_hash();
3045 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3048 * Ripoff of 'select_parent()'
3050 * search the list of submounts for a given mountpoint, and move any
3051 * shrinkable submounts to the 'graveyard' list.
3053 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3055 struct mount *this_parent = parent;
3056 struct list_head *next;
3060 next = this_parent->mnt_mounts.next;
3062 while (next != &this_parent->mnt_mounts) {
3063 struct list_head *tmp = next;
3064 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3067 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3070 * Descend a level if the d_mounts list is non-empty.
3072 if (!list_empty(&mnt->mnt_mounts)) {
3077 if (!propagate_mount_busy(mnt, 1)) {
3078 list_move_tail(&mnt->mnt_expire, graveyard);
3083 * All done at this level ... ascend and resume the search
3085 if (this_parent != parent) {
3086 next = this_parent->mnt_child.next;
3087 this_parent = this_parent->mnt_parent;
3094 * process a list of expirable mountpoints with the intent of discarding any
3095 * submounts of a specific parent mountpoint
3097 * mount_lock must be held for write
3099 static void shrink_submounts(struct mount *mnt)
3101 LIST_HEAD(graveyard);
3104 /* extract submounts of 'mountpoint' from the expiration list */
3105 while (select_submounts(mnt, &graveyard)) {
3106 while (!list_empty(&graveyard)) {
3107 m = list_first_entry(&graveyard, struct mount,
3109 touch_mnt_namespace(m->mnt_ns);
3110 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3115 static void *copy_mount_options(const void __user * data)
3118 unsigned left, offset;
3123 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3125 return ERR_PTR(-ENOMEM);
3127 left = copy_from_user(copy, data, PAGE_SIZE);
3130 * Not all architectures have an exact copy_from_user(). Resort to
3133 offset = PAGE_SIZE - left;
3136 if (get_user(c, (const char __user *)data + offset))
3143 if (left == PAGE_SIZE) {
3145 return ERR_PTR(-EFAULT);
3151 static char *copy_mount_string(const void __user *data)
3153 return data ? strndup_user(data, PATH_MAX) : NULL;
3157 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3158 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3160 * data is a (void *) that can point to any structure up to
3161 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3162 * information (or be NULL).
3164 * Pre-0.97 versions of mount() didn't have a flags word.
3165 * When the flags word was introduced its top half was required
3166 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3167 * Therefore, if this magic number is present, it carries no information
3168 * and must be discarded.
3170 int path_mount(const char *dev_name, struct path *path,
3171 const char *type_page, unsigned long flags, void *data_page)
3173 unsigned int mnt_flags = 0, sb_flags;
3177 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3178 flags &= ~MS_MGC_MSK;
3180 /* Basic sanity checks */
3182 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3184 if (flags & MS_NOUSER)
3187 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3192 if (flags & SB_MANDLOCK)
3195 /* Default to relatime unless overriden */
3196 if (!(flags & MS_NOATIME))
3197 mnt_flags |= MNT_RELATIME;
3199 /* Separate the per-mountpoint flags */
3200 if (flags & MS_NOSUID)
3201 mnt_flags |= MNT_NOSUID;
3202 if (flags & MS_NODEV)
3203 mnt_flags |= MNT_NODEV;
3204 if (flags & MS_NOEXEC)
3205 mnt_flags |= MNT_NOEXEC;
3206 if (flags & MS_NOATIME)
3207 mnt_flags |= MNT_NOATIME;
3208 if (flags & MS_NODIRATIME)
3209 mnt_flags |= MNT_NODIRATIME;
3210 if (flags & MS_STRICTATIME)
3211 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3212 if (flags & MS_RDONLY)
3213 mnt_flags |= MNT_READONLY;
3214 if (flags & MS_NOSYMFOLLOW)
3215 mnt_flags |= MNT_NOSYMFOLLOW;
3217 /* The default atime for remount is preservation */
3218 if ((flags & MS_REMOUNT) &&
3219 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3220 MS_STRICTATIME)) == 0)) {
3221 mnt_flags &= ~MNT_ATIME_MASK;
3222 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3225 sb_flags = flags & (SB_RDONLY |
3234 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3235 return do_reconfigure_mnt(path, mnt_flags);
3236 if (flags & MS_REMOUNT)
3237 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3238 if (flags & MS_BIND)
3239 return do_loopback(path, dev_name, flags & MS_REC);
3240 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3241 return do_change_type(path, flags);
3242 if (flags & MS_MOVE)
3243 return do_move_mount_old(path, dev_name);
3245 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3249 long do_mount(const char *dev_name, const char __user *dir_name,
3250 const char *type_page, unsigned long flags, void *data_page)
3255 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3258 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3263 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3265 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3268 static void dec_mnt_namespaces(struct ucounts *ucounts)
3270 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3273 static void free_mnt_ns(struct mnt_namespace *ns)
3275 if (!is_anon_ns(ns))
3276 ns_free_inum(&ns->ns);
3277 dec_mnt_namespaces(ns->ucounts);
3278 put_user_ns(ns->user_ns);
3283 * Assign a sequence number so we can detect when we attempt to bind
3284 * mount a reference to an older mount namespace into the current
3285 * mount namespace, preventing reference counting loops. A 64bit
3286 * number incrementing at 10Ghz will take 12,427 years to wrap which
3287 * is effectively never, so we can ignore the possibility.
3289 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3291 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3293 struct mnt_namespace *new_ns;
3294 struct ucounts *ucounts;
3297 ucounts = inc_mnt_namespaces(user_ns);
3299 return ERR_PTR(-ENOSPC);
3301 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3303 dec_mnt_namespaces(ucounts);
3304 return ERR_PTR(-ENOMEM);
3307 ret = ns_alloc_inum(&new_ns->ns);
3310 dec_mnt_namespaces(ucounts);
3311 return ERR_PTR(ret);
3314 new_ns->ns.ops = &mntns_operations;
3316 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3317 refcount_set(&new_ns->ns.count, 1);
3318 INIT_LIST_HEAD(&new_ns->list);
3319 init_waitqueue_head(&new_ns->poll);
3320 spin_lock_init(&new_ns->ns_lock);
3321 new_ns->user_ns = get_user_ns(user_ns);
3322 new_ns->ucounts = ucounts;
3327 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3328 struct user_namespace *user_ns, struct fs_struct *new_fs)
3330 struct mnt_namespace *new_ns;
3331 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3332 struct mount *p, *q;
3339 if (likely(!(flags & CLONE_NEWNS))) {
3346 new_ns = alloc_mnt_ns(user_ns, false);
3351 /* First pass: copy the tree topology */
3352 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3353 if (user_ns != ns->user_ns)
3354 copy_flags |= CL_SHARED_TO_SLAVE;
3355 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3358 free_mnt_ns(new_ns);
3359 return ERR_CAST(new);
3361 if (user_ns != ns->user_ns) {
3364 unlock_mount_hash();
3367 list_add_tail(&new_ns->list, &new->mnt_list);
3370 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3371 * as belonging to new namespace. We have already acquired a private
3372 * fs_struct, so tsk->fs->lock is not needed.
3380 if (&p->mnt == new_fs->root.mnt) {
3381 new_fs->root.mnt = mntget(&q->mnt);
3384 if (&p->mnt == new_fs->pwd.mnt) {
3385 new_fs->pwd.mnt = mntget(&q->mnt);
3389 p = next_mnt(p, old);
3390 q = next_mnt(q, new);
3393 while (p->mnt.mnt_root != q->mnt.mnt_root)
3394 p = next_mnt(p, old);
3406 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3408 struct mount *mnt = real_mount(m);
3409 struct mnt_namespace *ns;
3410 struct super_block *s;
3414 ns = alloc_mnt_ns(&init_user_ns, true);
3417 return ERR_CAST(ns);
3422 list_add(&mnt->mnt_list, &ns->list);
3424 err = vfs_path_lookup(m->mnt_root, m,
3425 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3430 return ERR_PTR(err);
3432 /* trade a vfsmount reference for active sb one */
3433 s = path.mnt->mnt_sb;
3434 atomic_inc(&s->s_active);
3436 /* lock the sucker */
3437 down_write(&s->s_umount);
3438 /* ... and return the root of (sub)tree on it */
3441 EXPORT_SYMBOL(mount_subtree);
3443 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3444 char __user *, type, unsigned long, flags, void __user *, data)
3451 kernel_type = copy_mount_string(type);
3452 ret = PTR_ERR(kernel_type);
3453 if (IS_ERR(kernel_type))
3456 kernel_dev = copy_mount_string(dev_name);
3457 ret = PTR_ERR(kernel_dev);
3458 if (IS_ERR(kernel_dev))
3461 options = copy_mount_options(data);
3462 ret = PTR_ERR(options);
3463 if (IS_ERR(options))
3466 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3477 #define FSMOUNT_VALID_FLAGS \
3478 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3479 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3480 MOUNT_ATTR_NOSYMFOLLOW)
3482 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3484 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3485 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3487 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3489 unsigned int mnt_flags = 0;
3491 if (attr_flags & MOUNT_ATTR_RDONLY)
3492 mnt_flags |= MNT_READONLY;
3493 if (attr_flags & MOUNT_ATTR_NOSUID)
3494 mnt_flags |= MNT_NOSUID;
3495 if (attr_flags & MOUNT_ATTR_NODEV)
3496 mnt_flags |= MNT_NODEV;
3497 if (attr_flags & MOUNT_ATTR_NOEXEC)
3498 mnt_flags |= MNT_NOEXEC;
3499 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3500 mnt_flags |= MNT_NODIRATIME;
3501 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3502 mnt_flags |= MNT_NOSYMFOLLOW;
3508 * Create a kernel mount representation for a new, prepared superblock
3509 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3511 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3512 unsigned int, attr_flags)
3514 struct mnt_namespace *ns;
3515 struct fs_context *fc;
3517 struct path newmount;
3520 unsigned int mnt_flags = 0;
3526 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3529 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3532 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3534 switch (attr_flags & MOUNT_ATTR__ATIME) {
3535 case MOUNT_ATTR_STRICTATIME:
3537 case MOUNT_ATTR_NOATIME:
3538 mnt_flags |= MNT_NOATIME;
3540 case MOUNT_ATTR_RELATIME:
3541 mnt_flags |= MNT_RELATIME;
3552 if (f.file->f_op != &fscontext_fops)
3555 fc = f.file->private_data;
3557 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3561 /* There must be a valid superblock or we can't mount it */
3567 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3568 pr_warn("VFS: Mount too revealing\n");
3573 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3576 if (fc->sb_flags & SB_MANDLOCK)
3579 newmount.mnt = vfs_create_mount(fc);
3580 if (IS_ERR(newmount.mnt)) {
3581 ret = PTR_ERR(newmount.mnt);
3584 newmount.dentry = dget(fc->root);
3585 newmount.mnt->mnt_flags = mnt_flags;
3587 /* We've done the mount bit - now move the file context into more or
3588 * less the same state as if we'd done an fspick(). We don't want to
3589 * do any memory allocation or anything like that at this point as we
3590 * don't want to have to handle any errors incurred.
3592 vfs_clean_context(fc);
3594 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3599 mnt = real_mount(newmount.mnt);
3603 list_add(&mnt->mnt_list, &ns->list);
3604 mntget(newmount.mnt);
3606 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3607 * it, not just simply put it.
3609 file = dentry_open(&newmount, O_PATH, fc->cred);
3611 dissolve_on_fput(newmount.mnt);
3612 ret = PTR_ERR(file);
3615 file->f_mode |= FMODE_NEED_UNMOUNT;
3617 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3619 fd_install(ret, file);
3624 path_put(&newmount);
3626 mutex_unlock(&fc->uapi_mutex);
3633 * Move a mount from one place to another. In combination with
3634 * fsopen()/fsmount() this is used to install a new mount and in combination
3635 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3638 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3640 SYSCALL_DEFINE5(move_mount,
3641 int, from_dfd, const char __user *, from_pathname,
3642 int, to_dfd, const char __user *, to_pathname,
3643 unsigned int, flags)
3645 struct path from_path, to_path;
3646 unsigned int lflags;
3652 if (flags & ~MOVE_MOUNT__MASK)
3655 /* If someone gives a pathname, they aren't permitted to move
3656 * from an fd that requires unmount as we can't get at the flag
3657 * to clear it afterwards.
3660 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3661 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3662 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3664 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3669 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3670 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3671 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3673 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3677 ret = security_move_mount(&from_path, &to_path);
3681 ret = do_move_mount(&from_path, &to_path);
3686 path_put(&from_path);
3691 * Return true if path is reachable from root
3693 * namespace_sem or mount_lock is held
3695 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3696 const struct path *root)
3698 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3699 dentry = mnt->mnt_mountpoint;
3700 mnt = mnt->mnt_parent;
3702 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3705 bool path_is_under(const struct path *path1, const struct path *path2)
3708 read_seqlock_excl(&mount_lock);
3709 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3710 read_sequnlock_excl(&mount_lock);
3713 EXPORT_SYMBOL(path_is_under);
3716 * pivot_root Semantics:
3717 * Moves the root file system of the current process to the directory put_old,
3718 * makes new_root as the new root file system of the current process, and sets
3719 * root/cwd of all processes which had them on the current root to new_root.
3722 * The new_root and put_old must be directories, and must not be on the
3723 * same file system as the current process root. The put_old must be
3724 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3725 * pointed to by put_old must yield the same directory as new_root. No other
3726 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3728 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3729 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3730 * in this situation.
3733 * - we don't move root/cwd if they are not at the root (reason: if something
3734 * cared enough to change them, it's probably wrong to force them elsewhere)
3735 * - it's okay to pick a root that isn't the root of a file system, e.g.
3736 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3737 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3740 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3741 const char __user *, put_old)
3743 struct path new, old, root;
3744 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3745 struct mountpoint *old_mp, *root_mp;
3751 error = user_path_at(AT_FDCWD, new_root,
3752 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3756 error = user_path_at(AT_FDCWD, put_old,
3757 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3761 error = security_sb_pivotroot(&old, &new);
3765 get_fs_root(current->fs, &root);
3766 old_mp = lock_mount(&old);
3767 error = PTR_ERR(old_mp);
3772 new_mnt = real_mount(new.mnt);
3773 root_mnt = real_mount(root.mnt);
3774 old_mnt = real_mount(old.mnt);
3775 ex_parent = new_mnt->mnt_parent;
3776 root_parent = root_mnt->mnt_parent;
3777 if (IS_MNT_SHARED(old_mnt) ||
3778 IS_MNT_SHARED(ex_parent) ||
3779 IS_MNT_SHARED(root_parent))
3781 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3783 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3786 if (d_unlinked(new.dentry))
3789 if (new_mnt == root_mnt || old_mnt == root_mnt)
3790 goto out4; /* loop, on the same file system */
3792 if (root.mnt->mnt_root != root.dentry)
3793 goto out4; /* not a mountpoint */
3794 if (!mnt_has_parent(root_mnt))
3795 goto out4; /* not attached */
3796 if (new.mnt->mnt_root != new.dentry)
3797 goto out4; /* not a mountpoint */
3798 if (!mnt_has_parent(new_mnt))
3799 goto out4; /* not attached */
3800 /* make sure we can reach put_old from new_root */
3801 if (!is_path_reachable(old_mnt, old.dentry, &new))
3803 /* make certain new is below the root */
3804 if (!is_path_reachable(new_mnt, new.dentry, &root))
3807 umount_mnt(new_mnt);
3808 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3809 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3810 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3811 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3813 /* mount old root on put_old */
3814 attach_mnt(root_mnt, old_mnt, old_mp);
3815 /* mount new_root on / */
3816 attach_mnt(new_mnt, root_parent, root_mp);
3817 mnt_add_count(root_parent, -1);
3818 touch_mnt_namespace(current->nsproxy->mnt_ns);
3819 /* A moved mount should not expire automatically */
3820 list_del_init(&new_mnt->mnt_expire);
3821 put_mountpoint(root_mp);
3822 unlock_mount_hash();
3823 chroot_fs_refs(&root, &new);
3826 unlock_mount(old_mp);
3828 mntput_no_expire(ex_parent);
3839 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3841 unsigned int flags = mnt->mnt.mnt_flags;
3843 /* flags to clear */
3844 flags &= ~kattr->attr_clr;
3845 /* flags to raise */
3846 flags |= kattr->attr_set;
3851 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3853 struct vfsmount *m = &mnt->mnt;
3855 if (!kattr->mnt_userns)
3859 * Once a mount has been idmapped we don't allow it to change its
3860 * mapping. It makes things simpler and callers can just create
3861 * another bind-mount they can idmap if they want to.
3863 if (mnt_user_ns(m) != &init_user_ns)
3866 /* The underlying filesystem doesn't support idmapped mounts yet. */
3867 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3870 /* Don't yet support filesystem mountable in user namespaces. */
3871 if (m->mnt_sb->s_user_ns != &init_user_ns)
3874 /* We're not controlling the superblock. */
3875 if (!capable(CAP_SYS_ADMIN))
3878 /* Mount has already been visible in the filesystem hierarchy. */
3879 if (!is_anon_ns(mnt->mnt_ns))
3885 static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3886 struct mount *mnt, int *err)
3888 struct mount *m = mnt, *last = NULL;
3890 if (!is_mounted(&m->mnt)) {
3895 if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3903 flags = recalc_flags(kattr, m);
3904 if (!can_change_locked_flags(m, flags)) {
3909 *err = can_idmap_mount(kattr, m);
3915 if ((kattr->attr_set & MNT_READONLY) &&
3916 !(m->mnt.mnt_flags & MNT_READONLY)) {
3917 *err = mnt_hold_writers(m);
3921 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3927 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3929 struct user_namespace *mnt_userns;
3931 if (!kattr->mnt_userns)
3934 mnt_userns = get_user_ns(kattr->mnt_userns);
3935 /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3936 smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3939 static void mount_setattr_commit(struct mount_kattr *kattr,
3940 struct mount *mnt, struct mount *last,
3943 struct mount *m = mnt;
3949 do_idmap_mount(kattr, m);
3950 flags = recalc_flags(kattr, m);
3951 WRITE_ONCE(m->mnt.mnt_flags, flags);
3955 * We either set MNT_READONLY above so make it visible
3956 * before ~MNT_WRITE_HOLD or we failed to recursively
3957 * apply mount options.
3959 if ((kattr->attr_set & MNT_READONLY) &&
3960 (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3961 mnt_unhold_writers(m);
3963 if (!err && kattr->propagation)
3964 change_mnt_propagation(m, kattr->propagation);
3967 * On failure, only cleanup until we found the first mount
3968 * we failed to handle.
3970 if (err && m == last)
3972 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3975 touch_mnt_namespace(mnt->mnt_ns);
3978 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3980 struct mount *mnt = real_mount(path->mnt), *last = NULL;
3983 if (path->dentry != mnt->mnt.mnt_root)
3986 if (kattr->propagation) {
3988 * Only take namespace_lock() if we're actually changing
3992 if (kattr->propagation == MS_SHARED) {
3993 err = invent_group_ids(mnt, kattr->recurse);
4004 * Get the mount tree in a shape where we can change mount
4005 * properties without failure.
4007 last = mount_setattr_prepare(kattr, mnt, &err);
4008 if (last) /* Commit all changes or revert to the old state. */
4009 mount_setattr_commit(kattr, mnt, last, err);
4011 unlock_mount_hash();
4013 if (kattr->propagation) {
4016 cleanup_group_ids(mnt, NULL);
4022 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4023 struct mount_kattr *kattr, unsigned int flags)
4026 struct ns_common *ns;
4027 struct user_namespace *mnt_userns;
4030 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4034 * We currently do not support clearing an idmapped mount. If this ever
4035 * is a use-case we can revisit this but for now let's keep it simple
4038 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4041 if (attr->userns_fd > INT_MAX)
4044 file = fget(attr->userns_fd);
4048 if (!proc_ns_file(file)) {
4053 ns = get_proc_ns(file_inode(file));
4054 if (ns->ops->type != CLONE_NEWUSER) {
4060 * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4061 * This is simpler than just having to treat NULL as unmapped. Users
4062 * wanting to idmap a mount to init_user_ns can just use a namespace
4063 * with an identity mapping.
4065 mnt_userns = container_of(ns, struct user_namespace, ns);
4066 if (mnt_userns == &init_user_ns) {
4070 kattr->mnt_userns = get_user_ns(mnt_userns);
4077 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4078 struct mount_kattr *kattr, unsigned int flags)
4080 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4082 if (flags & AT_NO_AUTOMOUNT)
4083 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4084 if (flags & AT_SYMLINK_NOFOLLOW)
4085 lookup_flags &= ~LOOKUP_FOLLOW;
4086 if (flags & AT_EMPTY_PATH)
4087 lookup_flags |= LOOKUP_EMPTY;
4089 *kattr = (struct mount_kattr) {
4090 .lookup_flags = lookup_flags,
4091 .recurse = !!(flags & AT_RECURSIVE),
4094 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4096 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4098 kattr->propagation = attr->propagation;
4100 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4103 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4104 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4107 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4108 * users wanting to transition to a different atime setting cannot
4109 * simply specify the atime setting in @attr_set, but must also
4110 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4111 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4112 * @attr_clr and that @attr_set can't have any atime bits set if
4113 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4115 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4116 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4120 * Clear all previous time settings as they are mutually
4123 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4124 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4125 case MOUNT_ATTR_RELATIME:
4126 kattr->attr_set |= MNT_RELATIME;
4128 case MOUNT_ATTR_NOATIME:
4129 kattr->attr_set |= MNT_NOATIME;
4131 case MOUNT_ATTR_STRICTATIME:
4137 if (attr->attr_set & MOUNT_ATTR__ATIME)
4141 return build_mount_idmapped(attr, usize, kattr, flags);
4144 static void finish_mount_kattr(struct mount_kattr *kattr)
4146 put_user_ns(kattr->mnt_userns);
4147 kattr->mnt_userns = NULL;
4150 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4151 unsigned int, flags, struct mount_attr __user *, uattr,
4156 struct mount_attr attr;
4157 struct mount_kattr kattr;
4159 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4161 if (flags & ~(AT_EMPTY_PATH |
4163 AT_SYMLINK_NOFOLLOW |
4167 if (unlikely(usize > PAGE_SIZE))
4169 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4175 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4179 /* Don't bother walking through the mounts if this is a nop. */
4180 if (attr.attr_set == 0 &&
4181 attr.attr_clr == 0 &&
4182 attr.propagation == 0)
4185 err = build_mount_kattr(&attr, usize, &kattr, flags);
4189 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4193 err = do_mount_setattr(&target, &kattr);
4194 finish_mount_kattr(&kattr);
4199 static void __init init_mount_tree(void)
4201 struct vfsmount *mnt;
4203 struct mnt_namespace *ns;
4206 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4208 panic("Can't create rootfs");
4210 ns = alloc_mnt_ns(&init_user_ns, false);
4212 panic("Can't allocate initial namespace");
4213 m = real_mount(mnt);
4217 list_add(&m->mnt_list, &ns->list);
4218 init_task.nsproxy->mnt_ns = ns;
4222 root.dentry = mnt->mnt_root;
4223 mnt->mnt_flags |= MNT_LOCKED;
4225 set_fs_pwd(current->fs, &root);
4226 set_fs_root(current->fs, &root);
4229 void __init mnt_init(void)
4233 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4234 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4236 mount_hashtable = alloc_large_system_hash("Mount-cache",
4237 sizeof(struct hlist_head),
4240 &m_hash_shift, &m_hash_mask, 0, 0);
4241 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4242 sizeof(struct hlist_head),
4245 &mp_hash_shift, &mp_hash_mask, 0, 0);
4247 if (!mount_hashtable || !mountpoint_hashtable)
4248 panic("Failed to allocate mount hash table\n");
4254 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4256 fs_kobj = kobject_create_and_add("fs", NULL);
4258 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4264 void put_mnt_ns(struct mnt_namespace *ns)
4266 if (!refcount_dec_and_test(&ns->ns.count))
4268 drop_collected_mounts(&ns->root->mnt);
4272 struct vfsmount *kern_mount(struct file_system_type *type)
4274 struct vfsmount *mnt;
4275 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4278 * it is a longterm mount, don't release mnt until
4279 * we unmount before file sys is unregistered
4281 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4285 EXPORT_SYMBOL_GPL(kern_mount);
4287 void kern_unmount(struct vfsmount *mnt)
4289 /* release long term mount so mount point can be released */
4290 if (!IS_ERR_OR_NULL(mnt)) {
4291 real_mount(mnt)->mnt_ns = NULL;
4292 synchronize_rcu(); /* yecchhh... */
4296 EXPORT_SYMBOL(kern_unmount);
4298 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4302 for (i = 0; i < num; i++)
4304 real_mount(mnt[i])->mnt_ns = NULL;
4305 synchronize_rcu_expedited();
4306 for (i = 0; i < num; i++)
4309 EXPORT_SYMBOL(kern_unmount_array);
4311 bool our_mnt(struct vfsmount *mnt)
4313 return check_mnt(real_mount(mnt));
4316 bool current_chrooted(void)
4318 /* Does the current process have a non-standard root */
4319 struct path ns_root;
4320 struct path fs_root;
4323 /* Find the namespace root */
4324 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4325 ns_root.dentry = ns_root.mnt->mnt_root;
4327 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4330 get_fs_root(current->fs, &fs_root);
4332 chrooted = !path_equal(&fs_root, &ns_root);
4340 static bool mnt_already_visible(struct mnt_namespace *ns,
4341 const struct super_block *sb,
4344 int new_flags = *new_mnt_flags;
4346 bool visible = false;
4348 down_read(&namespace_sem);
4350 list_for_each_entry(mnt, &ns->list, mnt_list) {
4351 struct mount *child;
4354 if (mnt_is_cursor(mnt))
4357 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4360 /* This mount is not fully visible if it's root directory
4361 * is not the root directory of the filesystem.
4363 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4366 /* A local view of the mount flags */
4367 mnt_flags = mnt->mnt.mnt_flags;
4369 /* Don't miss readonly hidden in the superblock flags */
4370 if (sb_rdonly(mnt->mnt.mnt_sb))
4371 mnt_flags |= MNT_LOCK_READONLY;
4373 /* Verify the mount flags are equal to or more permissive
4374 * than the proposed new mount.
4376 if ((mnt_flags & MNT_LOCK_READONLY) &&
4377 !(new_flags & MNT_READONLY))
4379 if ((mnt_flags & MNT_LOCK_ATIME) &&
4380 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4383 /* This mount is not fully visible if there are any
4384 * locked child mounts that cover anything except for
4385 * empty directories.
4387 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4388 struct inode *inode = child->mnt_mountpoint->d_inode;
4389 /* Only worry about locked mounts */
4390 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4392 /* Is the directory permanetly empty? */
4393 if (!is_empty_dir_inode(inode))
4396 /* Preserve the locked attributes */
4397 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4405 up_read(&namespace_sem);
4409 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4411 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4412 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4413 unsigned long s_iflags;
4415 if (ns->user_ns == &init_user_ns)
4418 /* Can this filesystem be too revealing? */
4419 s_iflags = sb->s_iflags;
4420 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4423 if ((s_iflags & required_iflags) != required_iflags) {
4424 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4429 return !mnt_already_visible(ns, sb, new_mnt_flags);
4432 bool mnt_may_suid(struct vfsmount *mnt)
4435 * Foreign mounts (accessed via fchdir or through /proc
4436 * symlinks) are always treated as if they are nosuid. This
4437 * prevents namespaces from trusting potentially unsafe
4438 * suid/sgid bits, file caps, or security labels that originate
4439 * in other namespaces.
4441 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4442 current_in_userns(mnt->mnt_sb->s_user_ns);
4445 static struct ns_common *mntns_get(struct task_struct *task)
4447 struct ns_common *ns = NULL;
4448 struct nsproxy *nsproxy;
4451 nsproxy = task->nsproxy;
4453 ns = &nsproxy->mnt_ns->ns;
4454 get_mnt_ns(to_mnt_ns(ns));
4461 static void mntns_put(struct ns_common *ns)
4463 put_mnt_ns(to_mnt_ns(ns));
4466 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4468 struct nsproxy *nsproxy = nsset->nsproxy;
4469 struct fs_struct *fs = nsset->fs;
4470 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4471 struct user_namespace *user_ns = nsset->cred->user_ns;
4475 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4476 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4477 !ns_capable(user_ns, CAP_SYS_ADMIN))
4480 if (is_anon_ns(mnt_ns))
4487 old_mnt_ns = nsproxy->mnt_ns;
4488 nsproxy->mnt_ns = mnt_ns;
4491 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4492 "/", LOOKUP_DOWN, &root);
4494 /* revert to old namespace */
4495 nsproxy->mnt_ns = old_mnt_ns;
4500 put_mnt_ns(old_mnt_ns);
4502 /* Update the pwd and root */
4503 set_fs_pwd(fs, &root);
4504 set_fs_root(fs, &root);
4510 static struct user_namespace *mntns_owner(struct ns_common *ns)
4512 return to_mnt_ns(ns)->user_ns;
4515 const struct proc_ns_operations mntns_operations = {
4517 .type = CLONE_NEWNS,
4520 .install = mntns_install,
4521 .owner = mntns_owner,