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);
1245 /* path_is_mountpoint() - Check if path is a mount in the current
1248 * d_mountpoint() can only be used reliably to establish if a dentry is
1249 * not mounted in any namespace and that common case is handled inline.
1250 * d_mountpoint() isn't aware of the possibility there may be multiple
1251 * mounts using a given dentry in a different namespace. This function
1252 * checks if the passed in path is a mountpoint rather than the dentry
1255 bool path_is_mountpoint(const struct path *path)
1260 if (!d_mountpoint(path->dentry))
1265 seq = read_seqbegin(&mount_lock);
1266 res = __path_is_mountpoint(path);
1267 } while (read_seqretry(&mount_lock, seq));
1272 EXPORT_SYMBOL(path_is_mountpoint);
1274 struct vfsmount *mnt_clone_internal(const struct path *path)
1277 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1280 p->mnt.mnt_flags |= MNT_INTERNAL;
1284 #ifdef CONFIG_PROC_FS
1285 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1286 struct list_head *p)
1288 struct mount *mnt, *ret = NULL;
1291 list_for_each_continue(p, &ns->list) {
1292 mnt = list_entry(p, typeof(*mnt), mnt_list);
1293 if (!mnt_is_cursor(mnt)) {
1303 /* iterator; we want it to have access to namespace_sem, thus here... */
1304 static void *m_start(struct seq_file *m, loff_t *pos)
1306 struct proc_mounts *p = m->private;
1307 struct list_head *prev;
1309 down_read(&namespace_sem);
1311 prev = &p->ns->list;
1313 prev = &p->cursor.mnt_list;
1315 /* Read after we'd reached the end? */
1316 if (list_empty(prev))
1320 return mnt_list_next(p->ns, prev);
1323 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1325 struct proc_mounts *p = m->private;
1326 struct mount *mnt = v;
1329 return mnt_list_next(p->ns, &mnt->mnt_list);
1332 static void m_stop(struct seq_file *m, void *v)
1334 struct proc_mounts *p = m->private;
1335 struct mount *mnt = v;
1337 lock_ns_list(p->ns);
1339 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1341 list_del_init(&p->cursor.mnt_list);
1342 unlock_ns_list(p->ns);
1343 up_read(&namespace_sem);
1346 static int m_show(struct seq_file *m, void *v)
1348 struct proc_mounts *p = m->private;
1349 struct mount *r = v;
1350 return p->show(m, &r->mnt);
1353 const struct seq_operations mounts_op = {
1360 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1362 down_read(&namespace_sem);
1364 list_del(&cursor->mnt_list);
1366 up_read(&namespace_sem);
1368 #endif /* CONFIG_PROC_FS */
1371 * may_umount_tree - check if a mount tree is busy
1372 * @mnt: root of mount tree
1374 * This is called to check if a tree of mounts has any
1375 * open files, pwds, chroots or sub mounts that are
1378 int may_umount_tree(struct vfsmount *m)
1380 struct mount *mnt = real_mount(m);
1381 int actual_refs = 0;
1382 int minimum_refs = 0;
1386 /* write lock needed for mnt_get_count */
1388 for (p = mnt; p; p = next_mnt(p, mnt)) {
1389 actual_refs += mnt_get_count(p);
1392 unlock_mount_hash();
1394 if (actual_refs > minimum_refs)
1400 EXPORT_SYMBOL(may_umount_tree);
1403 * may_umount - check if a mount point is busy
1404 * @mnt: root of mount
1406 * This is called to check if a mount point has any
1407 * open files, pwds, chroots or sub mounts. If the
1408 * mount has sub mounts this will return busy
1409 * regardless of whether the sub mounts are busy.
1411 * Doesn't take quota and stuff into account. IOW, in some cases it will
1412 * give false negatives. The main reason why it's here is that we need
1413 * a non-destructive way to look for easily umountable filesystems.
1415 int may_umount(struct vfsmount *mnt)
1418 down_read(&namespace_sem);
1420 if (propagate_mount_busy(real_mount(mnt), 2))
1422 unlock_mount_hash();
1423 up_read(&namespace_sem);
1427 EXPORT_SYMBOL(may_umount);
1429 static void namespace_unlock(void)
1431 struct hlist_head head;
1432 struct hlist_node *p;
1436 hlist_move_list(&unmounted, &head);
1437 list_splice_init(&ex_mountpoints, &list);
1439 up_write(&namespace_sem);
1441 shrink_dentry_list(&list);
1443 if (likely(hlist_empty(&head)))
1446 synchronize_rcu_expedited();
1448 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1449 hlist_del(&m->mnt_umount);
1454 static inline void namespace_lock(void)
1456 down_write(&namespace_sem);
1459 enum umount_tree_flags {
1461 UMOUNT_PROPAGATE = 2,
1462 UMOUNT_CONNECTED = 4,
1465 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1467 /* Leaving mounts connected is only valid for lazy umounts */
1468 if (how & UMOUNT_SYNC)
1471 /* A mount without a parent has nothing to be connected to */
1472 if (!mnt_has_parent(mnt))
1475 /* Because the reference counting rules change when mounts are
1476 * unmounted and connected, umounted mounts may not be
1477 * connected to mounted mounts.
1479 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1482 /* Has it been requested that the mount remain connected? */
1483 if (how & UMOUNT_CONNECTED)
1486 /* Is the mount locked such that it needs to remain connected? */
1487 if (IS_MNT_LOCKED(mnt))
1490 /* By default disconnect the mount */
1495 * mount_lock must be held
1496 * namespace_sem must be held for write
1498 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1500 LIST_HEAD(tmp_list);
1503 if (how & UMOUNT_PROPAGATE)
1504 propagate_mount_unlock(mnt);
1506 /* Gather the mounts to umount */
1507 for (p = mnt; p; p = next_mnt(p, mnt)) {
1508 p->mnt.mnt_flags |= MNT_UMOUNT;
1509 list_move(&p->mnt_list, &tmp_list);
1512 /* Hide the mounts from mnt_mounts */
1513 list_for_each_entry(p, &tmp_list, mnt_list) {
1514 list_del_init(&p->mnt_child);
1517 /* Add propogated mounts to the tmp_list */
1518 if (how & UMOUNT_PROPAGATE)
1519 propagate_umount(&tmp_list);
1521 while (!list_empty(&tmp_list)) {
1522 struct mnt_namespace *ns;
1524 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1525 list_del_init(&p->mnt_expire);
1526 list_del_init(&p->mnt_list);
1530 __touch_mnt_namespace(ns);
1533 if (how & UMOUNT_SYNC)
1534 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1536 disconnect = disconnect_mount(p, how);
1537 if (mnt_has_parent(p)) {
1538 mnt_add_count(p->mnt_parent, -1);
1540 /* Don't forget about p */
1541 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1546 change_mnt_propagation(p, MS_PRIVATE);
1548 hlist_add_head(&p->mnt_umount, &unmounted);
1552 static void shrink_submounts(struct mount *mnt);
1554 static int do_umount_root(struct super_block *sb)
1558 down_write(&sb->s_umount);
1559 if (!sb_rdonly(sb)) {
1560 struct fs_context *fc;
1562 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1567 ret = parse_monolithic_mount_data(fc, NULL);
1569 ret = reconfigure_super(fc);
1573 up_write(&sb->s_umount);
1577 static int do_umount(struct mount *mnt, int flags)
1579 struct super_block *sb = mnt->mnt.mnt_sb;
1582 retval = security_sb_umount(&mnt->mnt, flags);
1587 * Allow userspace to request a mountpoint be expired rather than
1588 * unmounting unconditionally. Unmount only happens if:
1589 * (1) the mark is already set (the mark is cleared by mntput())
1590 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1592 if (flags & MNT_EXPIRE) {
1593 if (&mnt->mnt == current->fs->root.mnt ||
1594 flags & (MNT_FORCE | MNT_DETACH))
1598 * probably don't strictly need the lock here if we examined
1599 * all race cases, but it's a slowpath.
1602 if (mnt_get_count(mnt) != 2) {
1603 unlock_mount_hash();
1606 unlock_mount_hash();
1608 if (!xchg(&mnt->mnt_expiry_mark, 1))
1613 * If we may have to abort operations to get out of this
1614 * mount, and they will themselves hold resources we must
1615 * allow the fs to do things. In the Unix tradition of
1616 * 'Gee thats tricky lets do it in userspace' the umount_begin
1617 * might fail to complete on the first run through as other tasks
1618 * must return, and the like. Thats for the mount program to worry
1619 * about for the moment.
1622 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1623 sb->s_op->umount_begin(sb);
1627 * No sense to grab the lock for this test, but test itself looks
1628 * somewhat bogus. Suggestions for better replacement?
1629 * Ho-hum... In principle, we might treat that as umount + switch
1630 * to rootfs. GC would eventually take care of the old vfsmount.
1631 * Actually it makes sense, especially if rootfs would contain a
1632 * /reboot - static binary that would close all descriptors and
1633 * call reboot(9). Then init(8) could umount root and exec /reboot.
1635 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1637 * Special case for "unmounting" root ...
1638 * we just try to remount it readonly.
1640 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1642 return do_umount_root(sb);
1648 /* Recheck MNT_LOCKED with the locks held */
1650 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1654 if (flags & MNT_DETACH) {
1655 if (!list_empty(&mnt->mnt_list))
1656 umount_tree(mnt, UMOUNT_PROPAGATE);
1659 shrink_submounts(mnt);
1661 if (!propagate_mount_busy(mnt, 2)) {
1662 if (!list_empty(&mnt->mnt_list))
1663 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1668 unlock_mount_hash();
1674 * __detach_mounts - lazily unmount all mounts on the specified dentry
1676 * During unlink, rmdir, and d_drop it is possible to loose the path
1677 * to an existing mountpoint, and wind up leaking the mount.
1678 * detach_mounts allows lazily unmounting those mounts instead of
1681 * The caller may hold dentry->d_inode->i_mutex.
1683 void __detach_mounts(struct dentry *dentry)
1685 struct mountpoint *mp;
1690 mp = lookup_mountpoint(dentry);
1695 while (!hlist_empty(&mp->m_list)) {
1696 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1697 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1699 hlist_add_head(&mnt->mnt_umount, &unmounted);
1701 else umount_tree(mnt, UMOUNT_CONNECTED);
1705 unlock_mount_hash();
1710 * Is the caller allowed to modify his namespace?
1712 static inline bool may_mount(void)
1714 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1717 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1718 static inline bool may_mandlock(void)
1720 return capable(CAP_SYS_ADMIN);
1723 static inline bool may_mandlock(void)
1725 pr_warn("VFS: \"mand\" mount option not supported");
1730 static int can_umount(const struct path *path, int flags)
1732 struct mount *mnt = real_mount(path->mnt);
1736 if (path->dentry != path->mnt->mnt_root)
1738 if (!check_mnt(mnt))
1740 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1742 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1747 // caller is responsible for flags being sane
1748 int path_umount(struct path *path, int flags)
1750 struct mount *mnt = real_mount(path->mnt);
1753 ret = can_umount(path, flags);
1755 ret = do_umount(mnt, flags);
1757 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1759 mntput_no_expire(mnt);
1763 static int ksys_umount(char __user *name, int flags)
1765 int lookup_flags = LOOKUP_MOUNTPOINT;
1769 // basic validity checks done first
1770 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1773 if (!(flags & UMOUNT_NOFOLLOW))
1774 lookup_flags |= LOOKUP_FOLLOW;
1775 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1778 return path_umount(&path, flags);
1781 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1783 return ksys_umount(name, flags);
1786 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1789 * The 2.0 compatible umount. No flags.
1791 SYSCALL_DEFINE1(oldumount, char __user *, name)
1793 return ksys_umount(name, 0);
1798 static bool is_mnt_ns_file(struct dentry *dentry)
1800 /* Is this a proxy for a mount namespace? */
1801 return dentry->d_op == &ns_dentry_operations &&
1802 dentry->d_fsdata == &mntns_operations;
1805 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1807 return container_of(ns, struct mnt_namespace, ns);
1810 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1815 static bool mnt_ns_loop(struct dentry *dentry)
1817 /* Could bind mounting the mount namespace inode cause a
1818 * mount namespace loop?
1820 struct mnt_namespace *mnt_ns;
1821 if (!is_mnt_ns_file(dentry))
1824 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1825 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1828 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1831 struct mount *res, *p, *q, *r, *parent;
1833 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1834 return ERR_PTR(-EINVAL);
1836 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1837 return ERR_PTR(-EINVAL);
1839 res = q = clone_mnt(mnt, dentry, flag);
1843 q->mnt_mountpoint = mnt->mnt_mountpoint;
1846 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1848 if (!is_subdir(r->mnt_mountpoint, dentry))
1851 for (s = r; s; s = next_mnt(s, r)) {
1852 if (!(flag & CL_COPY_UNBINDABLE) &&
1853 IS_MNT_UNBINDABLE(s)) {
1854 if (s->mnt.mnt_flags & MNT_LOCKED) {
1855 /* Both unbindable and locked. */
1856 q = ERR_PTR(-EPERM);
1859 s = skip_mnt_tree(s);
1863 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1864 is_mnt_ns_file(s->mnt.mnt_root)) {
1865 s = skip_mnt_tree(s);
1868 while (p != s->mnt_parent) {
1874 q = clone_mnt(p, p->mnt.mnt_root, flag);
1878 list_add_tail(&q->mnt_list, &res->mnt_list);
1879 attach_mnt(q, parent, p->mnt_mp);
1880 unlock_mount_hash();
1887 umount_tree(res, UMOUNT_SYNC);
1888 unlock_mount_hash();
1893 /* Caller should check returned pointer for errors */
1895 struct vfsmount *collect_mounts(const struct path *path)
1899 if (!check_mnt(real_mount(path->mnt)))
1900 tree = ERR_PTR(-EINVAL);
1902 tree = copy_tree(real_mount(path->mnt), path->dentry,
1903 CL_COPY_ALL | CL_PRIVATE);
1906 return ERR_CAST(tree);
1910 static void free_mnt_ns(struct mnt_namespace *);
1911 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1913 void dissolve_on_fput(struct vfsmount *mnt)
1915 struct mnt_namespace *ns;
1918 ns = real_mount(mnt)->mnt_ns;
1921 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1925 unlock_mount_hash();
1931 void drop_collected_mounts(struct vfsmount *mnt)
1935 umount_tree(real_mount(mnt), 0);
1936 unlock_mount_hash();
1941 * clone_private_mount - create a private clone of a path
1943 * This creates a new vfsmount, which will be the clone of @path. The new will
1944 * not be attached anywhere in the namespace and will be private (i.e. changes
1945 * to the originating mount won't be propagated into this).
1947 * Release with mntput().
1949 struct vfsmount *clone_private_mount(const struct path *path)
1951 struct mount *old_mnt = real_mount(path->mnt);
1952 struct mount *new_mnt;
1954 if (IS_MNT_UNBINDABLE(old_mnt))
1955 return ERR_PTR(-EINVAL);
1957 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1958 if (IS_ERR(new_mnt))
1959 return ERR_CAST(new_mnt);
1961 /* Longterm mount to be removed by kern_unmount*() */
1962 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1964 return &new_mnt->mnt;
1966 EXPORT_SYMBOL_GPL(clone_private_mount);
1968 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1969 struct vfsmount *root)
1972 int res = f(root, arg);
1975 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1976 res = f(&mnt->mnt, arg);
1983 static void lock_mnt_tree(struct mount *mnt)
1987 for (p = mnt; p; p = next_mnt(p, mnt)) {
1988 int flags = p->mnt.mnt_flags;
1989 /* Don't allow unprivileged users to change mount flags */
1990 flags |= MNT_LOCK_ATIME;
1992 if (flags & MNT_READONLY)
1993 flags |= MNT_LOCK_READONLY;
1995 if (flags & MNT_NODEV)
1996 flags |= MNT_LOCK_NODEV;
1998 if (flags & MNT_NOSUID)
1999 flags |= MNT_LOCK_NOSUID;
2001 if (flags & MNT_NOEXEC)
2002 flags |= MNT_LOCK_NOEXEC;
2003 /* Don't allow unprivileged users to reveal what is under a mount */
2004 if (list_empty(&p->mnt_expire))
2005 flags |= MNT_LOCKED;
2006 p->mnt.mnt_flags = flags;
2010 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2014 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2015 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2016 mnt_release_group_id(p);
2020 static int invent_group_ids(struct mount *mnt, bool recurse)
2024 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2025 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2026 int err = mnt_alloc_group_id(p);
2028 cleanup_group_ids(mnt, p);
2037 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2039 unsigned int max = READ_ONCE(sysctl_mount_max);
2040 unsigned int mounts = 0, old, pending, sum;
2043 for (p = mnt; p; p = next_mnt(p, mnt))
2047 pending = ns->pending_mounts;
2048 sum = old + pending;
2052 (mounts > (max - sum)))
2055 ns->pending_mounts = pending + mounts;
2060 * @source_mnt : mount tree to be attached
2061 * @nd : place the mount tree @source_mnt is attached
2062 * @parent_nd : if non-null, detach the source_mnt from its parent and
2063 * store the parent mount and mountpoint dentry.
2064 * (done when source_mnt is moved)
2066 * NOTE: in the table below explains the semantics when a source mount
2067 * of a given type is attached to a destination mount of a given type.
2068 * ---------------------------------------------------------------------------
2069 * | BIND MOUNT OPERATION |
2070 * |**************************************************************************
2071 * | source-->| shared | private | slave | unbindable |
2075 * |**************************************************************************
2076 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2078 * |non-shared| shared (+) | private | slave (*) | invalid |
2079 * ***************************************************************************
2080 * A bind operation clones the source mount and mounts the clone on the
2081 * destination mount.
2083 * (++) the cloned mount is propagated to all the mounts in the propagation
2084 * tree of the destination mount and the cloned mount is added to
2085 * the peer group of the source mount.
2086 * (+) the cloned mount is created under the destination mount and is marked
2087 * as shared. The cloned mount is added to the peer group of the source
2089 * (+++) the mount is propagated to all the mounts in the propagation tree
2090 * of the destination mount and the cloned mount is made slave
2091 * of the same master as that of the source mount. The cloned mount
2092 * is marked as 'shared and slave'.
2093 * (*) the cloned mount is made a slave of the same master as that of the
2096 * ---------------------------------------------------------------------------
2097 * | MOVE MOUNT OPERATION |
2098 * |**************************************************************************
2099 * | source-->| shared | private | slave | unbindable |
2103 * |**************************************************************************
2104 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2106 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2107 * ***************************************************************************
2109 * (+) the mount is moved to the destination. And is then propagated to
2110 * all the mounts in the propagation tree of the destination mount.
2111 * (+*) the mount is moved to the destination.
2112 * (+++) the mount is moved to the destination and is then propagated to
2113 * all the mounts belonging to the destination mount's propagation tree.
2114 * the mount is marked as 'shared and slave'.
2115 * (*) the mount continues to be a slave at the new location.
2117 * if the source mount is a tree, the operations explained above is
2118 * applied to each mount in the tree.
2119 * Must be called without spinlocks held, since this function can sleep
2122 static int attach_recursive_mnt(struct mount *source_mnt,
2123 struct mount *dest_mnt,
2124 struct mountpoint *dest_mp,
2127 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2128 HLIST_HEAD(tree_list);
2129 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2130 struct mountpoint *smp;
2131 struct mount *child, *p;
2132 struct hlist_node *n;
2135 /* Preallocate a mountpoint in case the new mounts need
2136 * to be tucked under other mounts.
2138 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2140 return PTR_ERR(smp);
2142 /* Is there space to add these mounts to the mount namespace? */
2144 err = count_mounts(ns, source_mnt);
2149 if (IS_MNT_SHARED(dest_mnt)) {
2150 err = invent_group_ids(source_mnt, true);
2153 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2156 goto out_cleanup_ids;
2157 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2163 unhash_mnt(source_mnt);
2164 attach_mnt(source_mnt, dest_mnt, dest_mp);
2165 touch_mnt_namespace(source_mnt->mnt_ns);
2167 if (source_mnt->mnt_ns) {
2168 /* move from anon - the caller will destroy */
2169 list_del_init(&source_mnt->mnt_ns->list);
2171 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2172 commit_tree(source_mnt);
2175 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2177 hlist_del_init(&child->mnt_hash);
2178 q = __lookup_mnt(&child->mnt_parent->mnt,
2179 child->mnt_mountpoint);
2181 mnt_change_mountpoint(child, smp, q);
2182 /* Notice when we are propagating across user namespaces */
2183 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2184 lock_mnt_tree(child);
2185 child->mnt.mnt_flags &= ~MNT_LOCKED;
2188 put_mountpoint(smp);
2189 unlock_mount_hash();
2194 while (!hlist_empty(&tree_list)) {
2195 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2196 child->mnt_parent->mnt_ns->pending_mounts = 0;
2197 umount_tree(child, UMOUNT_SYNC);
2199 unlock_mount_hash();
2200 cleanup_group_ids(source_mnt, NULL);
2202 ns->pending_mounts = 0;
2204 read_seqlock_excl(&mount_lock);
2205 put_mountpoint(smp);
2206 read_sequnlock_excl(&mount_lock);
2211 static struct mountpoint *lock_mount(struct path *path)
2213 struct vfsmount *mnt;
2214 struct dentry *dentry = path->dentry;
2216 inode_lock(dentry->d_inode);
2217 if (unlikely(cant_mount(dentry))) {
2218 inode_unlock(dentry->d_inode);
2219 return ERR_PTR(-ENOENT);
2222 mnt = lookup_mnt(path);
2224 struct mountpoint *mp = get_mountpoint(dentry);
2227 inode_unlock(dentry->d_inode);
2233 inode_unlock(path->dentry->d_inode);
2236 dentry = path->dentry = dget(mnt->mnt_root);
2240 static void unlock_mount(struct mountpoint *where)
2242 struct dentry *dentry = where->m_dentry;
2244 read_seqlock_excl(&mount_lock);
2245 put_mountpoint(where);
2246 read_sequnlock_excl(&mount_lock);
2249 inode_unlock(dentry->d_inode);
2252 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2254 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2257 if (d_is_dir(mp->m_dentry) !=
2258 d_is_dir(mnt->mnt.mnt_root))
2261 return attach_recursive_mnt(mnt, p, mp, false);
2265 * Sanity check the flags to change_mnt_propagation.
2268 static int flags_to_propagation_type(int ms_flags)
2270 int type = ms_flags & ~(MS_REC | MS_SILENT);
2272 /* Fail if any non-propagation flags are set */
2273 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2275 /* Only one propagation flag should be set */
2276 if (!is_power_of_2(type))
2282 * recursively change the type of the mountpoint.
2284 static int do_change_type(struct path *path, int ms_flags)
2287 struct mount *mnt = real_mount(path->mnt);
2288 int recurse = ms_flags & MS_REC;
2292 if (path->dentry != path->mnt->mnt_root)
2295 type = flags_to_propagation_type(ms_flags);
2300 if (type == MS_SHARED) {
2301 err = invent_group_ids(mnt, recurse);
2307 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2308 change_mnt_propagation(m, type);
2309 unlock_mount_hash();
2316 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2318 struct mount *child;
2319 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2320 if (!is_subdir(child->mnt_mountpoint, dentry))
2323 if (child->mnt.mnt_flags & MNT_LOCKED)
2329 static struct mount *__do_loopback(struct path *old_path, int recurse)
2331 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2333 if (IS_MNT_UNBINDABLE(old))
2336 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2339 if (!recurse && has_locked_children(old, old_path->dentry))
2343 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2345 mnt = clone_mnt(old, old_path->dentry, 0);
2348 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2354 * do loopback mount.
2356 static int do_loopback(struct path *path, const char *old_name,
2359 struct path old_path;
2360 struct mount *mnt = NULL, *parent;
2361 struct mountpoint *mp;
2363 if (!old_name || !*old_name)
2365 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2370 if (mnt_ns_loop(old_path.dentry))
2373 mp = lock_mount(path);
2379 parent = real_mount(path->mnt);
2380 if (!check_mnt(parent))
2383 mnt = __do_loopback(&old_path, recurse);
2389 err = graft_tree(mnt, parent, mp);
2392 umount_tree(mnt, UMOUNT_SYNC);
2393 unlock_mount_hash();
2398 path_put(&old_path);
2402 static struct file *open_detached_copy(struct path *path, bool recursive)
2404 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2405 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2406 struct mount *mnt, *p;
2410 return ERR_CAST(ns);
2413 mnt = __do_loopback(path, recursive);
2417 return ERR_CAST(mnt);
2421 for (p = mnt; p; p = next_mnt(p, mnt)) {
2426 list_add_tail(&ns->list, &mnt->mnt_list);
2428 unlock_mount_hash();
2432 path->mnt = &mnt->mnt;
2433 file = dentry_open(path, O_PATH, current_cred());
2435 dissolve_on_fput(path->mnt);
2437 file->f_mode |= FMODE_NEED_UNMOUNT;
2441 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2445 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2446 bool detached = flags & OPEN_TREE_CLONE;
2450 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2452 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2453 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2457 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2460 if (flags & AT_NO_AUTOMOUNT)
2461 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2462 if (flags & AT_SYMLINK_NOFOLLOW)
2463 lookup_flags &= ~LOOKUP_FOLLOW;
2464 if (flags & AT_EMPTY_PATH)
2465 lookup_flags |= LOOKUP_EMPTY;
2467 if (detached && !may_mount())
2470 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2474 error = user_path_at(dfd, filename, lookup_flags, &path);
2475 if (unlikely(error)) {
2476 file = ERR_PTR(error);
2479 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2481 file = dentry_open(&path, O_PATH, current_cred());
2486 return PTR_ERR(file);
2488 fd_install(fd, file);
2493 * Don't allow locked mount flags to be cleared.
2495 * No locks need to be held here while testing the various MNT_LOCK
2496 * flags because those flags can never be cleared once they are set.
2498 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2500 unsigned int fl = mnt->mnt.mnt_flags;
2502 if ((fl & MNT_LOCK_READONLY) &&
2503 !(mnt_flags & MNT_READONLY))
2506 if ((fl & MNT_LOCK_NODEV) &&
2507 !(mnt_flags & MNT_NODEV))
2510 if ((fl & MNT_LOCK_NOSUID) &&
2511 !(mnt_flags & MNT_NOSUID))
2514 if ((fl & MNT_LOCK_NOEXEC) &&
2515 !(mnt_flags & MNT_NOEXEC))
2518 if ((fl & MNT_LOCK_ATIME) &&
2519 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2525 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2527 bool readonly_request = (mnt_flags & MNT_READONLY);
2529 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2532 if (readonly_request)
2533 return mnt_make_readonly(mnt);
2535 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2539 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2541 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2542 mnt->mnt.mnt_flags = mnt_flags;
2543 touch_mnt_namespace(mnt->mnt_ns);
2546 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2548 struct super_block *sb = mnt->mnt_sb;
2550 if (!__mnt_is_readonly(mnt) &&
2551 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2552 char *buf = (char *)__get_free_page(GFP_KERNEL);
2553 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2556 time64_to_tm(sb->s_time_max, 0, &tm);
2558 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2560 is_mounted(mnt) ? "remounted" : "mounted",
2562 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2564 free_page((unsigned long)buf);
2569 * Handle reconfiguration of the mountpoint only without alteration of the
2570 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2573 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2575 struct super_block *sb = path->mnt->mnt_sb;
2576 struct mount *mnt = real_mount(path->mnt);
2579 if (!check_mnt(mnt))
2582 if (path->dentry != mnt->mnt.mnt_root)
2585 if (!can_change_locked_flags(mnt, mnt_flags))
2589 * We're only checking whether the superblock is read-only not
2590 * changing it, so only take down_read(&sb->s_umount).
2592 down_read(&sb->s_umount);
2594 ret = change_mount_ro_state(mnt, mnt_flags);
2596 set_mount_attributes(mnt, mnt_flags);
2597 unlock_mount_hash();
2598 up_read(&sb->s_umount);
2600 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2606 * change filesystem flags. dir should be a physical root of filesystem.
2607 * If you've mounted a non-root directory somewhere and want to do remount
2608 * on it - tough luck.
2610 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2611 int mnt_flags, void *data)
2614 struct super_block *sb = path->mnt->mnt_sb;
2615 struct mount *mnt = real_mount(path->mnt);
2616 struct fs_context *fc;
2618 if (!check_mnt(mnt))
2621 if (path->dentry != path->mnt->mnt_root)
2624 if (!can_change_locked_flags(mnt, mnt_flags))
2627 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2632 err = parse_monolithic_mount_data(fc, data);
2634 down_write(&sb->s_umount);
2636 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2637 err = reconfigure_super(fc);
2640 set_mount_attributes(mnt, mnt_flags);
2641 unlock_mount_hash();
2644 up_write(&sb->s_umount);
2647 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2653 static inline int tree_contains_unbindable(struct mount *mnt)
2656 for (p = mnt; p; p = next_mnt(p, mnt)) {
2657 if (IS_MNT_UNBINDABLE(p))
2664 * Check that there aren't references to earlier/same mount namespaces in the
2665 * specified subtree. Such references can act as pins for mount namespaces
2666 * that aren't checked by the mount-cycle checking code, thereby allowing
2667 * cycles to be made.
2669 static bool check_for_nsfs_mounts(struct mount *subtree)
2675 for (p = subtree; p; p = next_mnt(p, subtree))
2676 if (mnt_ns_loop(p->mnt.mnt_root))
2681 unlock_mount_hash();
2685 static int do_move_mount(struct path *old_path, struct path *new_path)
2687 struct mnt_namespace *ns;
2690 struct mount *parent;
2691 struct mountpoint *mp, *old_mp;
2695 mp = lock_mount(new_path);
2699 old = real_mount(old_path->mnt);
2700 p = real_mount(new_path->mnt);
2701 parent = old->mnt_parent;
2702 attached = mnt_has_parent(old);
2703 old_mp = old->mnt_mp;
2707 /* The mountpoint must be in our namespace. */
2711 /* The thing moved must be mounted... */
2712 if (!is_mounted(&old->mnt))
2715 /* ... and either ours or the root of anon namespace */
2716 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2719 if (old->mnt.mnt_flags & MNT_LOCKED)
2722 if (old_path->dentry != old_path->mnt->mnt_root)
2725 if (d_is_dir(new_path->dentry) !=
2726 d_is_dir(old_path->dentry))
2729 * Don't move a mount residing in a shared parent.
2731 if (attached && IS_MNT_SHARED(parent))
2734 * Don't move a mount tree containing unbindable mounts to a destination
2735 * mount which is shared.
2737 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2740 if (!check_for_nsfs_mounts(old))
2742 for (; mnt_has_parent(p); p = p->mnt_parent)
2746 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2751 /* if the mount is moved, it should no longer be expire
2753 list_del_init(&old->mnt_expire);
2755 put_mountpoint(old_mp);
2760 mntput_no_expire(parent);
2767 static int do_move_mount_old(struct path *path, const char *old_name)
2769 struct path old_path;
2772 if (!old_name || !*old_name)
2775 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2779 err = do_move_mount(&old_path, path);
2780 path_put(&old_path);
2785 * add a mount into a namespace's mount tree
2787 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2788 struct path *path, int mnt_flags)
2790 struct mount *parent = real_mount(path->mnt);
2792 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2794 if (unlikely(!check_mnt(parent))) {
2795 /* that's acceptable only for automounts done in private ns */
2796 if (!(mnt_flags & MNT_SHRINKABLE))
2798 /* ... and for those we'd better have mountpoint still alive */
2799 if (!parent->mnt_ns)
2803 /* Refuse the same filesystem on the same mount point */
2804 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2805 path->mnt->mnt_root == path->dentry)
2808 if (d_is_symlink(newmnt->mnt.mnt_root))
2811 newmnt->mnt.mnt_flags = mnt_flags;
2812 return graft_tree(newmnt, parent, mp);
2815 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2818 * Create a new mount using a superblock configuration and request it
2819 * be added to the namespace tree.
2821 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2822 unsigned int mnt_flags)
2824 struct vfsmount *mnt;
2825 struct mountpoint *mp;
2826 struct super_block *sb = fc->root->d_sb;
2829 error = security_sb_kern_mount(sb);
2830 if (!error && mount_too_revealing(sb, &mnt_flags))
2833 if (unlikely(error)) {
2838 up_write(&sb->s_umount);
2840 mnt = vfs_create_mount(fc);
2842 return PTR_ERR(mnt);
2844 mnt_warn_timestamp_expiry(mountpoint, mnt);
2846 mp = lock_mount(mountpoint);
2851 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2859 * create a new mount for userspace and request it to be added into the
2862 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2863 int mnt_flags, const char *name, void *data)
2865 struct file_system_type *type;
2866 struct fs_context *fc;
2867 const char *subtype = NULL;
2873 type = get_fs_type(fstype);
2877 if (type->fs_flags & FS_HAS_SUBTYPE) {
2878 subtype = strchr(fstype, '.');
2882 put_filesystem(type);
2888 fc = fs_context_for_mount(type, sb_flags);
2889 put_filesystem(type);
2894 err = vfs_parse_fs_string(fc, "subtype",
2895 subtype, strlen(subtype));
2897 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2899 err = parse_monolithic_mount_data(fc, data);
2900 if (!err && !mount_capable(fc))
2903 err = vfs_get_tree(fc);
2905 err = do_new_mount_fc(fc, path, mnt_flags);
2911 int finish_automount(struct vfsmount *m, struct path *path)
2913 struct dentry *dentry = path->dentry;
2914 struct mountpoint *mp;
2923 mnt = real_mount(m);
2924 /* The new mount record should have at least 2 refs to prevent it being
2925 * expired before we get a chance to add it
2927 BUG_ON(mnt_get_count(mnt) < 2);
2929 if (m->mnt_sb == path->mnt->mnt_sb &&
2930 m->mnt_root == dentry) {
2936 * we don't want to use lock_mount() - in this case finding something
2937 * that overmounts our mountpoint to be means "quitely drop what we've
2938 * got", not "try to mount it on top".
2940 inode_lock(dentry->d_inode);
2942 if (unlikely(cant_mount(dentry))) {
2944 goto discard_locked;
2947 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2950 goto discard_locked;
2953 mp = get_mountpoint(dentry);
2956 goto discard_locked;
2959 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2968 inode_unlock(dentry->d_inode);
2970 /* remove m from any expiration list it may be on */
2971 if (!list_empty(&mnt->mnt_expire)) {
2973 list_del_init(&mnt->mnt_expire);
2982 * mnt_set_expiry - Put a mount on an expiration list
2983 * @mnt: The mount to list.
2984 * @expiry_list: The list to add the mount to.
2986 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2990 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2994 EXPORT_SYMBOL(mnt_set_expiry);
2997 * process a list of expirable mountpoints with the intent of discarding any
2998 * mountpoints that aren't in use and haven't been touched since last we came
3001 void mark_mounts_for_expiry(struct list_head *mounts)
3003 struct mount *mnt, *next;
3004 LIST_HEAD(graveyard);
3006 if (list_empty(mounts))
3012 /* extract from the expiration list every vfsmount that matches the
3013 * following criteria:
3014 * - only referenced by its parent vfsmount
3015 * - still marked for expiry (marked on the last call here; marks are
3016 * cleared by mntput())
3018 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3019 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3020 propagate_mount_busy(mnt, 1))
3022 list_move(&mnt->mnt_expire, &graveyard);
3024 while (!list_empty(&graveyard)) {
3025 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3026 touch_mnt_namespace(mnt->mnt_ns);
3027 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3029 unlock_mount_hash();
3033 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3036 * Ripoff of 'select_parent()'
3038 * search the list of submounts for a given mountpoint, and move any
3039 * shrinkable submounts to the 'graveyard' list.
3041 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3043 struct mount *this_parent = parent;
3044 struct list_head *next;
3048 next = this_parent->mnt_mounts.next;
3050 while (next != &this_parent->mnt_mounts) {
3051 struct list_head *tmp = next;
3052 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3055 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3058 * Descend a level if the d_mounts list is non-empty.
3060 if (!list_empty(&mnt->mnt_mounts)) {
3065 if (!propagate_mount_busy(mnt, 1)) {
3066 list_move_tail(&mnt->mnt_expire, graveyard);
3071 * All done at this level ... ascend and resume the search
3073 if (this_parent != parent) {
3074 next = this_parent->mnt_child.next;
3075 this_parent = this_parent->mnt_parent;
3082 * process a list of expirable mountpoints with the intent of discarding any
3083 * submounts of a specific parent mountpoint
3085 * mount_lock must be held for write
3087 static void shrink_submounts(struct mount *mnt)
3089 LIST_HEAD(graveyard);
3092 /* extract submounts of 'mountpoint' from the expiration list */
3093 while (select_submounts(mnt, &graveyard)) {
3094 while (!list_empty(&graveyard)) {
3095 m = list_first_entry(&graveyard, struct mount,
3097 touch_mnt_namespace(m->mnt_ns);
3098 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3103 static void *copy_mount_options(const void __user * data)
3106 unsigned left, offset;
3111 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3113 return ERR_PTR(-ENOMEM);
3115 left = copy_from_user(copy, data, PAGE_SIZE);
3118 * Not all architectures have an exact copy_from_user(). Resort to
3121 offset = PAGE_SIZE - left;
3124 if (get_user(c, (const char __user *)data + offset))
3131 if (left == PAGE_SIZE) {
3133 return ERR_PTR(-EFAULT);
3139 static char *copy_mount_string(const void __user *data)
3141 return data ? strndup_user(data, PATH_MAX) : NULL;
3145 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3146 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3148 * data is a (void *) that can point to any structure up to
3149 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3150 * information (or be NULL).
3152 * Pre-0.97 versions of mount() didn't have a flags word.
3153 * When the flags word was introduced its top half was required
3154 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3155 * Therefore, if this magic number is present, it carries no information
3156 * and must be discarded.
3158 int path_mount(const char *dev_name, struct path *path,
3159 const char *type_page, unsigned long flags, void *data_page)
3161 unsigned int mnt_flags = 0, sb_flags;
3165 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3166 flags &= ~MS_MGC_MSK;
3168 /* Basic sanity checks */
3170 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3172 if (flags & MS_NOUSER)
3175 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3180 if ((flags & SB_MANDLOCK) && !may_mandlock())
3183 /* Default to relatime unless overriden */
3184 if (!(flags & MS_NOATIME))
3185 mnt_flags |= MNT_RELATIME;
3187 /* Separate the per-mountpoint flags */
3188 if (flags & MS_NOSUID)
3189 mnt_flags |= MNT_NOSUID;
3190 if (flags & MS_NODEV)
3191 mnt_flags |= MNT_NODEV;
3192 if (flags & MS_NOEXEC)
3193 mnt_flags |= MNT_NOEXEC;
3194 if (flags & MS_NOATIME)
3195 mnt_flags |= MNT_NOATIME;
3196 if (flags & MS_NODIRATIME)
3197 mnt_flags |= MNT_NODIRATIME;
3198 if (flags & MS_STRICTATIME)
3199 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3200 if (flags & MS_RDONLY)
3201 mnt_flags |= MNT_READONLY;
3202 if (flags & MS_NOSYMFOLLOW)
3203 mnt_flags |= MNT_NOSYMFOLLOW;
3205 /* The default atime for remount is preservation */
3206 if ((flags & MS_REMOUNT) &&
3207 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3208 MS_STRICTATIME)) == 0)) {
3209 mnt_flags &= ~MNT_ATIME_MASK;
3210 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3213 sb_flags = flags & (SB_RDONLY |
3222 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3223 return do_reconfigure_mnt(path, mnt_flags);
3224 if (flags & MS_REMOUNT)
3225 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3226 if (flags & MS_BIND)
3227 return do_loopback(path, dev_name, flags & MS_REC);
3228 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3229 return do_change_type(path, flags);
3230 if (flags & MS_MOVE)
3231 return do_move_mount_old(path, dev_name);
3233 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3237 long do_mount(const char *dev_name, const char __user *dir_name,
3238 const char *type_page, unsigned long flags, void *data_page)
3243 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3246 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3251 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3253 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3256 static void dec_mnt_namespaces(struct ucounts *ucounts)
3258 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3261 static void free_mnt_ns(struct mnt_namespace *ns)
3263 if (!is_anon_ns(ns))
3264 ns_free_inum(&ns->ns);
3265 dec_mnt_namespaces(ns->ucounts);
3266 put_user_ns(ns->user_ns);
3271 * Assign a sequence number so we can detect when we attempt to bind
3272 * mount a reference to an older mount namespace into the current
3273 * mount namespace, preventing reference counting loops. A 64bit
3274 * number incrementing at 10Ghz will take 12,427 years to wrap which
3275 * is effectively never, so we can ignore the possibility.
3277 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3279 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3281 struct mnt_namespace *new_ns;
3282 struct ucounts *ucounts;
3285 ucounts = inc_mnt_namespaces(user_ns);
3287 return ERR_PTR(-ENOSPC);
3289 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3291 dec_mnt_namespaces(ucounts);
3292 return ERR_PTR(-ENOMEM);
3295 ret = ns_alloc_inum(&new_ns->ns);
3298 dec_mnt_namespaces(ucounts);
3299 return ERR_PTR(ret);
3302 new_ns->ns.ops = &mntns_operations;
3304 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3305 refcount_set(&new_ns->ns.count, 1);
3306 INIT_LIST_HEAD(&new_ns->list);
3307 init_waitqueue_head(&new_ns->poll);
3308 spin_lock_init(&new_ns->ns_lock);
3309 new_ns->user_ns = get_user_ns(user_ns);
3310 new_ns->ucounts = ucounts;
3315 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3316 struct user_namespace *user_ns, struct fs_struct *new_fs)
3318 struct mnt_namespace *new_ns;
3319 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3320 struct mount *p, *q;
3327 if (likely(!(flags & CLONE_NEWNS))) {
3334 new_ns = alloc_mnt_ns(user_ns, false);
3339 /* First pass: copy the tree topology */
3340 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3341 if (user_ns != ns->user_ns)
3342 copy_flags |= CL_SHARED_TO_SLAVE;
3343 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3346 free_mnt_ns(new_ns);
3347 return ERR_CAST(new);
3349 if (user_ns != ns->user_ns) {
3352 unlock_mount_hash();
3355 list_add_tail(&new_ns->list, &new->mnt_list);
3358 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3359 * as belonging to new namespace. We have already acquired a private
3360 * fs_struct, so tsk->fs->lock is not needed.
3368 if (&p->mnt == new_fs->root.mnt) {
3369 new_fs->root.mnt = mntget(&q->mnt);
3372 if (&p->mnt == new_fs->pwd.mnt) {
3373 new_fs->pwd.mnt = mntget(&q->mnt);
3377 p = next_mnt(p, old);
3378 q = next_mnt(q, new);
3381 while (p->mnt.mnt_root != q->mnt.mnt_root)
3382 p = next_mnt(p, old);
3394 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3396 struct mount *mnt = real_mount(m);
3397 struct mnt_namespace *ns;
3398 struct super_block *s;
3402 ns = alloc_mnt_ns(&init_user_ns, true);
3405 return ERR_CAST(ns);
3410 list_add(&mnt->mnt_list, &ns->list);
3412 err = vfs_path_lookup(m->mnt_root, m,
3413 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3418 return ERR_PTR(err);
3420 /* trade a vfsmount reference for active sb one */
3421 s = path.mnt->mnt_sb;
3422 atomic_inc(&s->s_active);
3424 /* lock the sucker */
3425 down_write(&s->s_umount);
3426 /* ... and return the root of (sub)tree on it */
3429 EXPORT_SYMBOL(mount_subtree);
3431 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3432 char __user *, type, unsigned long, flags, void __user *, data)
3439 kernel_type = copy_mount_string(type);
3440 ret = PTR_ERR(kernel_type);
3441 if (IS_ERR(kernel_type))
3444 kernel_dev = copy_mount_string(dev_name);
3445 ret = PTR_ERR(kernel_dev);
3446 if (IS_ERR(kernel_dev))
3449 options = copy_mount_options(data);
3450 ret = PTR_ERR(options);
3451 if (IS_ERR(options))
3454 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3465 #define FSMOUNT_VALID_FLAGS \
3466 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3467 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME)
3469 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3471 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3472 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3474 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3476 unsigned int mnt_flags = 0;
3478 if (attr_flags & MOUNT_ATTR_RDONLY)
3479 mnt_flags |= MNT_READONLY;
3480 if (attr_flags & MOUNT_ATTR_NOSUID)
3481 mnt_flags |= MNT_NOSUID;
3482 if (attr_flags & MOUNT_ATTR_NODEV)
3483 mnt_flags |= MNT_NODEV;
3484 if (attr_flags & MOUNT_ATTR_NOEXEC)
3485 mnt_flags |= MNT_NOEXEC;
3486 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3487 mnt_flags |= MNT_NODIRATIME;
3493 * Create a kernel mount representation for a new, prepared superblock
3494 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3496 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3497 unsigned int, attr_flags)
3499 struct mnt_namespace *ns;
3500 struct fs_context *fc;
3502 struct path newmount;
3505 unsigned int mnt_flags = 0;
3511 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3514 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3517 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3519 switch (attr_flags & MOUNT_ATTR__ATIME) {
3520 case MOUNT_ATTR_STRICTATIME:
3522 case MOUNT_ATTR_NOATIME:
3523 mnt_flags |= MNT_NOATIME;
3525 case MOUNT_ATTR_RELATIME:
3526 mnt_flags |= MNT_RELATIME;
3537 if (f.file->f_op != &fscontext_fops)
3540 fc = f.file->private_data;
3542 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3546 /* There must be a valid superblock or we can't mount it */
3552 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3553 pr_warn("VFS: Mount too revealing\n");
3558 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3562 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3565 newmount.mnt = vfs_create_mount(fc);
3566 if (IS_ERR(newmount.mnt)) {
3567 ret = PTR_ERR(newmount.mnt);
3570 newmount.dentry = dget(fc->root);
3571 newmount.mnt->mnt_flags = mnt_flags;
3573 /* We've done the mount bit - now move the file context into more or
3574 * less the same state as if we'd done an fspick(). We don't want to
3575 * do any memory allocation or anything like that at this point as we
3576 * don't want to have to handle any errors incurred.
3578 vfs_clean_context(fc);
3580 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3585 mnt = real_mount(newmount.mnt);
3589 list_add(&mnt->mnt_list, &ns->list);
3590 mntget(newmount.mnt);
3592 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3593 * it, not just simply put it.
3595 file = dentry_open(&newmount, O_PATH, fc->cred);
3597 dissolve_on_fput(newmount.mnt);
3598 ret = PTR_ERR(file);
3601 file->f_mode |= FMODE_NEED_UNMOUNT;
3603 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3605 fd_install(ret, file);
3610 path_put(&newmount);
3612 mutex_unlock(&fc->uapi_mutex);
3619 * Move a mount from one place to another. In combination with
3620 * fsopen()/fsmount() this is used to install a new mount and in combination
3621 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3624 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3626 SYSCALL_DEFINE5(move_mount,
3627 int, from_dfd, const char __user *, from_pathname,
3628 int, to_dfd, const char __user *, to_pathname,
3629 unsigned int, flags)
3631 struct path from_path, to_path;
3632 unsigned int lflags;
3638 if (flags & ~MOVE_MOUNT__MASK)
3641 /* If someone gives a pathname, they aren't permitted to move
3642 * from an fd that requires unmount as we can't get at the flag
3643 * to clear it afterwards.
3646 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3647 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3648 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3650 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3655 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3656 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3657 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3659 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3663 ret = security_move_mount(&from_path, &to_path);
3667 ret = do_move_mount(&from_path, &to_path);
3672 path_put(&from_path);
3677 * Return true if path is reachable from root
3679 * namespace_sem or mount_lock is held
3681 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3682 const struct path *root)
3684 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3685 dentry = mnt->mnt_mountpoint;
3686 mnt = mnt->mnt_parent;
3688 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3691 bool path_is_under(const struct path *path1, const struct path *path2)
3694 read_seqlock_excl(&mount_lock);
3695 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3696 read_sequnlock_excl(&mount_lock);
3699 EXPORT_SYMBOL(path_is_under);
3702 * pivot_root Semantics:
3703 * Moves the root file system of the current process to the directory put_old,
3704 * makes new_root as the new root file system of the current process, and sets
3705 * root/cwd of all processes which had them on the current root to new_root.
3708 * The new_root and put_old must be directories, and must not be on the
3709 * same file system as the current process root. The put_old must be
3710 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3711 * pointed to by put_old must yield the same directory as new_root. No other
3712 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3714 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3715 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3716 * in this situation.
3719 * - we don't move root/cwd if they are not at the root (reason: if something
3720 * cared enough to change them, it's probably wrong to force them elsewhere)
3721 * - it's okay to pick a root that isn't the root of a file system, e.g.
3722 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3723 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3726 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3727 const char __user *, put_old)
3729 struct path new, old, root;
3730 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3731 struct mountpoint *old_mp, *root_mp;
3737 error = user_path_at(AT_FDCWD, new_root,
3738 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3742 error = user_path_at(AT_FDCWD, put_old,
3743 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3747 error = security_sb_pivotroot(&old, &new);
3751 get_fs_root(current->fs, &root);
3752 old_mp = lock_mount(&old);
3753 error = PTR_ERR(old_mp);
3758 new_mnt = real_mount(new.mnt);
3759 root_mnt = real_mount(root.mnt);
3760 old_mnt = real_mount(old.mnt);
3761 ex_parent = new_mnt->mnt_parent;
3762 root_parent = root_mnt->mnt_parent;
3763 if (IS_MNT_SHARED(old_mnt) ||
3764 IS_MNT_SHARED(ex_parent) ||
3765 IS_MNT_SHARED(root_parent))
3767 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3769 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3772 if (d_unlinked(new.dentry))
3775 if (new_mnt == root_mnt || old_mnt == root_mnt)
3776 goto out4; /* loop, on the same file system */
3778 if (root.mnt->mnt_root != root.dentry)
3779 goto out4; /* not a mountpoint */
3780 if (!mnt_has_parent(root_mnt))
3781 goto out4; /* not attached */
3782 if (new.mnt->mnt_root != new.dentry)
3783 goto out4; /* not a mountpoint */
3784 if (!mnt_has_parent(new_mnt))
3785 goto out4; /* not attached */
3786 /* make sure we can reach put_old from new_root */
3787 if (!is_path_reachable(old_mnt, old.dentry, &new))
3789 /* make certain new is below the root */
3790 if (!is_path_reachable(new_mnt, new.dentry, &root))
3793 umount_mnt(new_mnt);
3794 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3795 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3796 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3797 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3799 /* mount old root on put_old */
3800 attach_mnt(root_mnt, old_mnt, old_mp);
3801 /* mount new_root on / */
3802 attach_mnt(new_mnt, root_parent, root_mp);
3803 mnt_add_count(root_parent, -1);
3804 touch_mnt_namespace(current->nsproxy->mnt_ns);
3805 /* A moved mount should not expire automatically */
3806 list_del_init(&new_mnt->mnt_expire);
3807 put_mountpoint(root_mp);
3808 unlock_mount_hash();
3809 chroot_fs_refs(&root, &new);
3812 unlock_mount(old_mp);
3814 mntput_no_expire(ex_parent);
3825 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3827 unsigned int flags = mnt->mnt.mnt_flags;
3829 /* flags to clear */
3830 flags &= ~kattr->attr_clr;
3831 /* flags to raise */
3832 flags |= kattr->attr_set;
3837 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3839 struct vfsmount *m = &mnt->mnt;
3841 if (!kattr->mnt_userns)
3845 * Once a mount has been idmapped we don't allow it to change its
3846 * mapping. It makes things simpler and callers can just create
3847 * another bind-mount they can idmap if they want to.
3849 if (mnt_user_ns(m) != &init_user_ns)
3852 /* The underlying filesystem doesn't support idmapped mounts yet. */
3853 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3856 /* We're not controlling the superblock. */
3857 if (!ns_capable(m->mnt_sb->s_user_ns, CAP_SYS_ADMIN))
3860 /* Mount has already been visible in the filesystem hierarchy. */
3861 if (!is_anon_ns(mnt->mnt_ns))
3867 static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3868 struct mount *mnt, int *err)
3870 struct mount *m = mnt, *last = NULL;
3872 if (!is_mounted(&m->mnt)) {
3877 if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3885 flags = recalc_flags(kattr, m);
3886 if (!can_change_locked_flags(m, flags)) {
3891 *err = can_idmap_mount(kattr, m);
3897 if ((kattr->attr_set & MNT_READONLY) &&
3898 !(m->mnt.mnt_flags & MNT_READONLY)) {
3899 *err = mnt_hold_writers(m);
3903 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3909 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3911 struct user_namespace *mnt_userns;
3913 if (!kattr->mnt_userns)
3916 mnt_userns = get_user_ns(kattr->mnt_userns);
3917 /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3918 smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3921 static void mount_setattr_commit(struct mount_kattr *kattr,
3922 struct mount *mnt, struct mount *last,
3925 struct mount *m = mnt;
3931 do_idmap_mount(kattr, m);
3932 flags = recalc_flags(kattr, m);
3933 WRITE_ONCE(m->mnt.mnt_flags, flags);
3937 * We either set MNT_READONLY above so make it visible
3938 * before ~MNT_WRITE_HOLD or we failed to recursively
3939 * apply mount options.
3941 if ((kattr->attr_set & MNT_READONLY) &&
3942 (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3943 mnt_unhold_writers(m);
3945 if (!err && kattr->propagation)
3946 change_mnt_propagation(m, kattr->propagation);
3949 * On failure, only cleanup until we found the first mount
3950 * we failed to handle.
3952 if (err && m == last)
3954 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3957 touch_mnt_namespace(mnt->mnt_ns);
3960 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3962 struct mount *mnt = real_mount(path->mnt), *last = NULL;
3965 if (path->dentry != mnt->mnt.mnt_root)
3968 if (kattr->propagation) {
3970 * Only take namespace_lock() if we're actually changing
3974 if (kattr->propagation == MS_SHARED) {
3975 err = invent_group_ids(mnt, kattr->recurse);
3986 * Get the mount tree in a shape where we can change mount
3987 * properties without failure.
3989 last = mount_setattr_prepare(kattr, mnt, &err);
3990 if (last) /* Commit all changes or revert to the old state. */
3991 mount_setattr_commit(kattr, mnt, last, err);
3993 unlock_mount_hash();
3995 if (kattr->propagation) {
3998 cleanup_group_ids(mnt, NULL);
4004 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4005 struct mount_kattr *kattr, unsigned int flags)
4008 struct ns_common *ns;
4009 struct user_namespace *mnt_userns;
4012 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4016 * We currently do not support clearing an idmapped mount. If this ever
4017 * is a use-case we can revisit this but for now let's keep it simple
4020 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4023 if (attr->userns_fd > INT_MAX)
4026 file = fget(attr->userns_fd);
4030 if (!proc_ns_file(file)) {
4035 ns = get_proc_ns(file_inode(file));
4036 if (ns->ops->type != CLONE_NEWUSER) {
4042 * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4043 * This is simpler than just having to treat NULL as unmapped. Users
4044 * wanting to idmap a mount to init_user_ns can just use a namespace
4045 * with an identity mapping.
4047 mnt_userns = container_of(ns, struct user_namespace, ns);
4048 if (mnt_userns == &init_user_ns) {
4052 kattr->mnt_userns = get_user_ns(mnt_userns);
4059 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4060 struct mount_kattr *kattr, unsigned int flags)
4062 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4064 if (flags & AT_NO_AUTOMOUNT)
4065 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4066 if (flags & AT_SYMLINK_NOFOLLOW)
4067 lookup_flags &= ~LOOKUP_FOLLOW;
4068 if (flags & AT_EMPTY_PATH)
4069 lookup_flags |= LOOKUP_EMPTY;
4071 *kattr = (struct mount_kattr) {
4072 .lookup_flags = lookup_flags,
4073 .recurse = !!(flags & AT_RECURSIVE),
4076 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4078 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4080 kattr->propagation = attr->propagation;
4082 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4085 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4086 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4089 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4090 * users wanting to transition to a different atime setting cannot
4091 * simply specify the atime setting in @attr_set, but must also
4092 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4093 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4094 * @attr_clr and that @attr_set can't have any atime bits set if
4095 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4097 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4098 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4102 * Clear all previous time settings as they are mutually
4105 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4106 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4107 case MOUNT_ATTR_RELATIME:
4108 kattr->attr_set |= MNT_RELATIME;
4110 case MOUNT_ATTR_NOATIME:
4111 kattr->attr_set |= MNT_NOATIME;
4113 case MOUNT_ATTR_STRICTATIME:
4119 if (attr->attr_set & MOUNT_ATTR__ATIME)
4123 return build_mount_idmapped(attr, usize, kattr, flags);
4126 static void finish_mount_kattr(struct mount_kattr *kattr)
4128 put_user_ns(kattr->mnt_userns);
4129 kattr->mnt_userns = NULL;
4132 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4133 unsigned int, flags, struct mount_attr __user *, uattr,
4138 struct mount_attr attr;
4139 struct mount_kattr kattr;
4141 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4143 if (flags & ~(AT_EMPTY_PATH |
4145 AT_SYMLINK_NOFOLLOW |
4149 if (unlikely(usize > PAGE_SIZE))
4151 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4157 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4161 /* Don't bother walking through the mounts if this is a nop. */
4162 if (attr.attr_set == 0 &&
4163 attr.attr_clr == 0 &&
4164 attr.propagation == 0)
4167 err = build_mount_kattr(&attr, usize, &kattr, flags);
4171 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4175 err = do_mount_setattr(&target, &kattr);
4176 finish_mount_kattr(&kattr);
4181 static void __init init_mount_tree(void)
4183 struct vfsmount *mnt;
4185 struct mnt_namespace *ns;
4188 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4190 panic("Can't create rootfs");
4192 ns = alloc_mnt_ns(&init_user_ns, false);
4194 panic("Can't allocate initial namespace");
4195 m = real_mount(mnt);
4199 list_add(&m->mnt_list, &ns->list);
4200 init_task.nsproxy->mnt_ns = ns;
4204 root.dentry = mnt->mnt_root;
4205 mnt->mnt_flags |= MNT_LOCKED;
4207 set_fs_pwd(current->fs, &root);
4208 set_fs_root(current->fs, &root);
4211 void __init mnt_init(void)
4215 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4216 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4218 mount_hashtable = alloc_large_system_hash("Mount-cache",
4219 sizeof(struct hlist_head),
4222 &m_hash_shift, &m_hash_mask, 0, 0);
4223 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4224 sizeof(struct hlist_head),
4227 &mp_hash_shift, &mp_hash_mask, 0, 0);
4229 if (!mount_hashtable || !mountpoint_hashtable)
4230 panic("Failed to allocate mount hash table\n");
4236 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4238 fs_kobj = kobject_create_and_add("fs", NULL);
4240 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4246 void put_mnt_ns(struct mnt_namespace *ns)
4248 if (!refcount_dec_and_test(&ns->ns.count))
4250 drop_collected_mounts(&ns->root->mnt);
4254 struct vfsmount *kern_mount(struct file_system_type *type)
4256 struct vfsmount *mnt;
4257 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4260 * it is a longterm mount, don't release mnt until
4261 * we unmount before file sys is unregistered
4263 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4267 EXPORT_SYMBOL_GPL(kern_mount);
4269 void kern_unmount(struct vfsmount *mnt)
4271 /* release long term mount so mount point can be released */
4272 if (!IS_ERR_OR_NULL(mnt)) {
4273 real_mount(mnt)->mnt_ns = NULL;
4274 synchronize_rcu(); /* yecchhh... */
4278 EXPORT_SYMBOL(kern_unmount);
4280 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4284 for (i = 0; i < num; i++)
4286 real_mount(mnt[i])->mnt_ns = NULL;
4287 synchronize_rcu_expedited();
4288 for (i = 0; i < num; i++)
4291 EXPORT_SYMBOL(kern_unmount_array);
4293 bool our_mnt(struct vfsmount *mnt)
4295 return check_mnt(real_mount(mnt));
4298 bool current_chrooted(void)
4300 /* Does the current process have a non-standard root */
4301 struct path ns_root;
4302 struct path fs_root;
4305 /* Find the namespace root */
4306 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4307 ns_root.dentry = ns_root.mnt->mnt_root;
4309 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4312 get_fs_root(current->fs, &fs_root);
4314 chrooted = !path_equal(&fs_root, &ns_root);
4322 static bool mnt_already_visible(struct mnt_namespace *ns,
4323 const struct super_block *sb,
4326 int new_flags = *new_mnt_flags;
4328 bool visible = false;
4330 down_read(&namespace_sem);
4332 list_for_each_entry(mnt, &ns->list, mnt_list) {
4333 struct mount *child;
4336 if (mnt_is_cursor(mnt))
4339 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4342 /* This mount is not fully visible if it's root directory
4343 * is not the root directory of the filesystem.
4345 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4348 /* A local view of the mount flags */
4349 mnt_flags = mnt->mnt.mnt_flags;
4351 /* Don't miss readonly hidden in the superblock flags */
4352 if (sb_rdonly(mnt->mnt.mnt_sb))
4353 mnt_flags |= MNT_LOCK_READONLY;
4355 /* Verify the mount flags are equal to or more permissive
4356 * than the proposed new mount.
4358 if ((mnt_flags & MNT_LOCK_READONLY) &&
4359 !(new_flags & MNT_READONLY))
4361 if ((mnt_flags & MNT_LOCK_ATIME) &&
4362 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4365 /* This mount is not fully visible if there are any
4366 * locked child mounts that cover anything except for
4367 * empty directories.
4369 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4370 struct inode *inode = child->mnt_mountpoint->d_inode;
4371 /* Only worry about locked mounts */
4372 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4374 /* Is the directory permanetly empty? */
4375 if (!is_empty_dir_inode(inode))
4378 /* Preserve the locked attributes */
4379 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4387 up_read(&namespace_sem);
4391 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4393 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4394 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4395 unsigned long s_iflags;
4397 if (ns->user_ns == &init_user_ns)
4400 /* Can this filesystem be too revealing? */
4401 s_iflags = sb->s_iflags;
4402 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4405 if ((s_iflags & required_iflags) != required_iflags) {
4406 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4411 return !mnt_already_visible(ns, sb, new_mnt_flags);
4414 bool mnt_may_suid(struct vfsmount *mnt)
4417 * Foreign mounts (accessed via fchdir or through /proc
4418 * symlinks) are always treated as if they are nosuid. This
4419 * prevents namespaces from trusting potentially unsafe
4420 * suid/sgid bits, file caps, or security labels that originate
4421 * in other namespaces.
4423 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4424 current_in_userns(mnt->mnt_sb->s_user_ns);
4427 static struct ns_common *mntns_get(struct task_struct *task)
4429 struct ns_common *ns = NULL;
4430 struct nsproxy *nsproxy;
4433 nsproxy = task->nsproxy;
4435 ns = &nsproxy->mnt_ns->ns;
4436 get_mnt_ns(to_mnt_ns(ns));
4443 static void mntns_put(struct ns_common *ns)
4445 put_mnt_ns(to_mnt_ns(ns));
4448 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4450 struct nsproxy *nsproxy = nsset->nsproxy;
4451 struct fs_struct *fs = nsset->fs;
4452 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4453 struct user_namespace *user_ns = nsset->cred->user_ns;
4457 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4458 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4459 !ns_capable(user_ns, CAP_SYS_ADMIN))
4462 if (is_anon_ns(mnt_ns))
4469 old_mnt_ns = nsproxy->mnt_ns;
4470 nsproxy->mnt_ns = mnt_ns;
4473 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4474 "/", LOOKUP_DOWN, &root);
4476 /* revert to old namespace */
4477 nsproxy->mnt_ns = old_mnt_ns;
4482 put_mnt_ns(old_mnt_ns);
4484 /* Update the pwd and root */
4485 set_fs_pwd(fs, &root);
4486 set_fs_root(fs, &root);
4492 static struct user_namespace *mntns_owner(struct ns_common *ns)
4494 return to_mnt_ns(ns)->user_ns;
4497 const struct proc_ns_operations mntns_operations = {
4499 .type = CLONE_NEWNS,
4502 .install = mntns_install,
4503 .owner = mntns_owner,