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,
208 if (!mnt->mnt_devname)
213 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
215 goto out_free_devname;
217 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
220 mnt->mnt_writers = 0;
223 INIT_HLIST_NODE(&mnt->mnt_hash);
224 INIT_LIST_HEAD(&mnt->mnt_child);
225 INIT_LIST_HEAD(&mnt->mnt_mounts);
226 INIT_LIST_HEAD(&mnt->mnt_list);
227 INIT_LIST_HEAD(&mnt->mnt_expire);
228 INIT_LIST_HEAD(&mnt->mnt_share);
229 INIT_LIST_HEAD(&mnt->mnt_slave_list);
230 INIT_LIST_HEAD(&mnt->mnt_slave);
231 INIT_HLIST_NODE(&mnt->mnt_mp_list);
232 INIT_LIST_HEAD(&mnt->mnt_umounting);
233 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
234 mnt->mnt.mnt_userns = &init_user_ns;
240 kfree_const(mnt->mnt_devname);
245 kmem_cache_free(mnt_cache, mnt);
250 * Most r/o checks on a fs are for operations that take
251 * discrete amounts of time, like a write() or unlink().
252 * We must keep track of when those operations start
253 * (for permission checks) and when they end, so that
254 * we can determine when writes are able to occur to
258 * __mnt_is_readonly: check whether a mount is read-only
259 * @mnt: the mount to check for its write status
261 * This shouldn't be used directly ouside of the VFS.
262 * It does not guarantee that the filesystem will stay
263 * r/w, just that it is right *now*. This can not and
264 * should not be used in place of IS_RDONLY(inode).
265 * mnt_want/drop_write() will _keep_ the filesystem
268 bool __mnt_is_readonly(struct vfsmount *mnt)
270 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
272 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
274 static inline void mnt_inc_writers(struct mount *mnt)
277 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
283 static inline void mnt_dec_writers(struct mount *mnt)
286 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
292 static unsigned int mnt_get_writers(struct mount *mnt)
295 unsigned int count = 0;
298 for_each_possible_cpu(cpu) {
299 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
304 return mnt->mnt_writers;
308 static int mnt_is_readonly(struct vfsmount *mnt)
310 if (mnt->mnt_sb->s_readonly_remount)
312 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
314 return __mnt_is_readonly(mnt);
318 * Most r/o & frozen checks on a fs are for operations that take discrete
319 * amounts of time, like a write() or unlink(). We must keep track of when
320 * those operations start (for permission checks) and when they end, so that we
321 * can determine when writes are able to occur to a filesystem.
324 * __mnt_want_write - get write access to a mount without freeze protection
325 * @m: the mount on which to take a write
327 * This tells the low-level filesystem that a write is about to be performed to
328 * it, and makes sure that writes are allowed (mnt it read-write) before
329 * returning success. This operation does not protect against filesystem being
330 * frozen. When the write operation is finished, __mnt_drop_write() must be
331 * called. This is effectively a refcount.
333 int __mnt_want_write(struct vfsmount *m)
335 struct mount *mnt = real_mount(m);
339 mnt_inc_writers(mnt);
341 * The store to mnt_inc_writers must be visible before we pass
342 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
343 * incremented count after it has set MNT_WRITE_HOLD.
346 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
349 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
350 * be set to match its requirements. So we must not load that until
351 * MNT_WRITE_HOLD is cleared.
354 if (mnt_is_readonly(m)) {
355 mnt_dec_writers(mnt);
364 * mnt_want_write - get write access to a mount
365 * @m: the mount on which to take a write
367 * This tells the low-level filesystem that a write is about to be performed to
368 * it, and makes sure that writes are allowed (mount is read-write, filesystem
369 * is not frozen) before returning success. When the write operation is
370 * finished, mnt_drop_write() must be called. This is effectively a refcount.
372 int mnt_want_write(struct vfsmount *m)
376 sb_start_write(m->mnt_sb);
377 ret = __mnt_want_write(m);
379 sb_end_write(m->mnt_sb);
382 EXPORT_SYMBOL_GPL(mnt_want_write);
385 * __mnt_want_write_file - get write access to a file's mount
386 * @file: the file who's mount on which to take a write
388 * This is like __mnt_want_write, but if the file is already open for writing it
389 * skips incrementing mnt_writers (since the open file already has a reference)
390 * and instead only does the check for emergency r/o remounts. This must be
391 * paired with __mnt_drop_write_file.
393 int __mnt_want_write_file(struct file *file)
395 if (file->f_mode & FMODE_WRITER) {
397 * Superblock may have become readonly while there are still
398 * writable fd's, e.g. due to a fs error with errors=remount-ro
400 if (__mnt_is_readonly(file->f_path.mnt))
404 return __mnt_want_write(file->f_path.mnt);
408 * mnt_want_write_file - get write access to a file's mount
409 * @file: the file who's mount on which to take a write
411 * This is like mnt_want_write, but if the file is already open for writing it
412 * skips incrementing mnt_writers (since the open file already has a reference)
413 * and instead only does the freeze protection and the check for emergency r/o
414 * remounts. This must be paired with mnt_drop_write_file.
416 int mnt_want_write_file(struct file *file)
420 sb_start_write(file_inode(file)->i_sb);
421 ret = __mnt_want_write_file(file);
423 sb_end_write(file_inode(file)->i_sb);
426 EXPORT_SYMBOL_GPL(mnt_want_write_file);
429 * __mnt_drop_write - give up write access to a mount
430 * @mnt: the mount on which to give up write access
432 * Tells the low-level filesystem that we are done
433 * performing writes to it. Must be matched with
434 * __mnt_want_write() call above.
436 void __mnt_drop_write(struct vfsmount *mnt)
439 mnt_dec_writers(real_mount(mnt));
444 * mnt_drop_write - give up write access to a mount
445 * @mnt: the mount on which to give up write access
447 * Tells the low-level filesystem that we are done performing writes to it and
448 * also allows filesystem to be frozen again. Must be matched with
449 * mnt_want_write() call above.
451 void mnt_drop_write(struct vfsmount *mnt)
453 __mnt_drop_write(mnt);
454 sb_end_write(mnt->mnt_sb);
456 EXPORT_SYMBOL_GPL(mnt_drop_write);
458 void __mnt_drop_write_file(struct file *file)
460 if (!(file->f_mode & FMODE_WRITER))
461 __mnt_drop_write(file->f_path.mnt);
464 void mnt_drop_write_file(struct file *file)
466 __mnt_drop_write_file(file);
467 sb_end_write(file_inode(file)->i_sb);
469 EXPORT_SYMBOL(mnt_drop_write_file);
471 static inline int mnt_hold_writers(struct mount *mnt)
473 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
475 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
476 * should be visible before we do.
481 * With writers on hold, if this value is zero, then there are
482 * definitely no active writers (although held writers may subsequently
483 * increment the count, they'll have to wait, and decrement it after
484 * seeing MNT_READONLY).
486 * It is OK to have counter incremented on one CPU and decremented on
487 * another: the sum will add up correctly. The danger would be when we
488 * sum up each counter, if we read a counter before it is incremented,
489 * but then read another CPU's count which it has been subsequently
490 * decremented from -- we would see more decrements than we should.
491 * MNT_WRITE_HOLD protects against this scenario, because
492 * mnt_want_write first increments count, then smp_mb, then spins on
493 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
494 * we're counting up here.
496 if (mnt_get_writers(mnt) > 0)
502 static inline void mnt_unhold_writers(struct mount *mnt)
505 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
506 * that become unheld will see MNT_READONLY.
509 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
512 static int mnt_make_readonly(struct mount *mnt)
516 ret = mnt_hold_writers(mnt);
518 mnt->mnt.mnt_flags |= MNT_READONLY;
519 mnt_unhold_writers(mnt);
523 int sb_prepare_remount_readonly(struct super_block *sb)
528 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
529 if (atomic_long_read(&sb->s_remove_count))
533 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
534 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
535 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
537 if (mnt_get_writers(mnt) > 0) {
543 if (!err && atomic_long_read(&sb->s_remove_count))
547 sb->s_readonly_remount = 1;
550 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
551 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
552 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
559 static void free_vfsmnt(struct mount *mnt)
561 struct user_namespace *mnt_userns;
563 mnt_userns = mnt_user_ns(&mnt->mnt);
564 if (mnt_userns != &init_user_ns)
565 put_user_ns(mnt_userns);
566 kfree_const(mnt->mnt_devname);
568 free_percpu(mnt->mnt_pcp);
570 kmem_cache_free(mnt_cache, mnt);
573 static void delayed_free_vfsmnt(struct rcu_head *head)
575 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
578 /* call under rcu_read_lock */
579 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
582 if (read_seqretry(&mount_lock, seq))
586 mnt = real_mount(bastard);
587 mnt_add_count(mnt, 1);
588 smp_mb(); // see mntput_no_expire()
589 if (likely(!read_seqretry(&mount_lock, seq)))
591 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
592 mnt_add_count(mnt, -1);
596 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
597 mnt_add_count(mnt, -1);
602 /* caller will mntput() */
606 /* call under rcu_read_lock */
607 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
609 int res = __legitimize_mnt(bastard, seq);
612 if (unlikely(res < 0)) {
621 * find the first mount at @dentry on vfsmount @mnt.
622 * call under rcu_read_lock()
624 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
626 struct hlist_head *head = m_hash(mnt, dentry);
629 hlist_for_each_entry_rcu(p, head, mnt_hash)
630 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
636 * lookup_mnt - Return the first child mount mounted at path
638 * "First" means first mounted chronologically. If you create the
641 * mount /dev/sda1 /mnt
642 * mount /dev/sda2 /mnt
643 * mount /dev/sda3 /mnt
645 * Then lookup_mnt() on the base /mnt dentry in the root mount will
646 * return successively the root dentry and vfsmount of /dev/sda1, then
647 * /dev/sda2, then /dev/sda3, then NULL.
649 * lookup_mnt takes a reference to the found vfsmount.
651 struct vfsmount *lookup_mnt(const struct path *path)
653 struct mount *child_mnt;
659 seq = read_seqbegin(&mount_lock);
660 child_mnt = __lookup_mnt(path->mnt, path->dentry);
661 m = child_mnt ? &child_mnt->mnt : NULL;
662 } while (!legitimize_mnt(m, seq));
667 static inline void lock_ns_list(struct mnt_namespace *ns)
669 spin_lock(&ns->ns_lock);
672 static inline void unlock_ns_list(struct mnt_namespace *ns)
674 spin_unlock(&ns->ns_lock);
677 static inline bool mnt_is_cursor(struct mount *mnt)
679 return mnt->mnt.mnt_flags & MNT_CURSOR;
683 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
684 * current mount namespace.
686 * The common case is dentries are not mountpoints at all and that
687 * test is handled inline. For the slow case when we are actually
688 * dealing with a mountpoint of some kind, walk through all of the
689 * mounts in the current mount namespace and test to see if the dentry
692 * The mount_hashtable is not usable in the context because we
693 * need to identify all mounts that may be in the current mount
694 * namespace not just a mount that happens to have some specified
697 bool __is_local_mountpoint(struct dentry *dentry)
699 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
701 bool is_covered = false;
703 down_read(&namespace_sem);
705 list_for_each_entry(mnt, &ns->list, mnt_list) {
706 if (mnt_is_cursor(mnt))
708 is_covered = (mnt->mnt_mountpoint == dentry);
713 up_read(&namespace_sem);
718 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
720 struct hlist_head *chain = mp_hash(dentry);
721 struct mountpoint *mp;
723 hlist_for_each_entry(mp, chain, m_hash) {
724 if (mp->m_dentry == dentry) {
732 static struct mountpoint *get_mountpoint(struct dentry *dentry)
734 struct mountpoint *mp, *new = NULL;
737 if (d_mountpoint(dentry)) {
738 /* might be worth a WARN_ON() */
739 if (d_unlinked(dentry))
740 return ERR_PTR(-ENOENT);
742 read_seqlock_excl(&mount_lock);
743 mp = lookup_mountpoint(dentry);
744 read_sequnlock_excl(&mount_lock);
750 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
752 return ERR_PTR(-ENOMEM);
755 /* Exactly one processes may set d_mounted */
756 ret = d_set_mounted(dentry);
758 /* Someone else set d_mounted? */
762 /* The dentry is not available as a mountpoint? */
767 /* Add the new mountpoint to the hash table */
768 read_seqlock_excl(&mount_lock);
769 new->m_dentry = dget(dentry);
771 hlist_add_head(&new->m_hash, mp_hash(dentry));
772 INIT_HLIST_HEAD(&new->m_list);
773 read_sequnlock_excl(&mount_lock);
783 * vfsmount lock must be held. Additionally, the caller is responsible
784 * for serializing calls for given disposal list.
786 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
788 if (!--mp->m_count) {
789 struct dentry *dentry = mp->m_dentry;
790 BUG_ON(!hlist_empty(&mp->m_list));
791 spin_lock(&dentry->d_lock);
792 dentry->d_flags &= ~DCACHE_MOUNTED;
793 spin_unlock(&dentry->d_lock);
794 dput_to_list(dentry, list);
795 hlist_del(&mp->m_hash);
800 /* called with namespace_lock and vfsmount lock */
801 static void put_mountpoint(struct mountpoint *mp)
803 __put_mountpoint(mp, &ex_mountpoints);
806 static inline int check_mnt(struct mount *mnt)
808 return mnt->mnt_ns == current->nsproxy->mnt_ns;
812 * vfsmount lock must be held for write
814 static void touch_mnt_namespace(struct mnt_namespace *ns)
818 wake_up_interruptible(&ns->poll);
823 * vfsmount lock must be held for write
825 static void __touch_mnt_namespace(struct mnt_namespace *ns)
827 if (ns && ns->event != event) {
829 wake_up_interruptible(&ns->poll);
834 * vfsmount lock must be held for write
836 static struct mountpoint *unhash_mnt(struct mount *mnt)
838 struct mountpoint *mp;
839 mnt->mnt_parent = mnt;
840 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
841 list_del_init(&mnt->mnt_child);
842 hlist_del_init_rcu(&mnt->mnt_hash);
843 hlist_del_init(&mnt->mnt_mp_list);
850 * vfsmount lock must be held for write
852 static void umount_mnt(struct mount *mnt)
854 put_mountpoint(unhash_mnt(mnt));
858 * vfsmount lock must be held for write
860 void mnt_set_mountpoint(struct mount *mnt,
861 struct mountpoint *mp,
862 struct mount *child_mnt)
865 mnt_add_count(mnt, 1); /* essentially, that's mntget */
866 child_mnt->mnt_mountpoint = mp->m_dentry;
867 child_mnt->mnt_parent = mnt;
868 child_mnt->mnt_mp = mp;
869 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
872 static void __attach_mnt(struct mount *mnt, struct mount *parent)
874 hlist_add_head_rcu(&mnt->mnt_hash,
875 m_hash(&parent->mnt, mnt->mnt_mountpoint));
876 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
880 * vfsmount lock must be held for write
882 static void attach_mnt(struct mount *mnt,
883 struct mount *parent,
884 struct mountpoint *mp)
886 mnt_set_mountpoint(parent, mp, mnt);
887 __attach_mnt(mnt, parent);
890 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
892 struct mountpoint *old_mp = mnt->mnt_mp;
893 struct mount *old_parent = mnt->mnt_parent;
895 list_del_init(&mnt->mnt_child);
896 hlist_del_init(&mnt->mnt_mp_list);
897 hlist_del_init_rcu(&mnt->mnt_hash);
899 attach_mnt(mnt, parent, mp);
901 put_mountpoint(old_mp);
902 mnt_add_count(old_parent, -1);
906 * vfsmount lock must be held for write
908 static void commit_tree(struct mount *mnt)
910 struct mount *parent = mnt->mnt_parent;
913 struct mnt_namespace *n = parent->mnt_ns;
915 BUG_ON(parent == mnt);
917 list_add_tail(&head, &mnt->mnt_list);
918 list_for_each_entry(m, &head, mnt_list)
921 list_splice(&head, n->list.prev);
923 n->mounts += n->pending_mounts;
924 n->pending_mounts = 0;
926 __attach_mnt(mnt, parent);
927 touch_mnt_namespace(n);
930 static struct mount *next_mnt(struct mount *p, struct mount *root)
932 struct list_head *next = p->mnt_mounts.next;
933 if (next == &p->mnt_mounts) {
937 next = p->mnt_child.next;
938 if (next != &p->mnt_parent->mnt_mounts)
943 return list_entry(next, struct mount, mnt_child);
946 static struct mount *skip_mnt_tree(struct mount *p)
948 struct list_head *prev = p->mnt_mounts.prev;
949 while (prev != &p->mnt_mounts) {
950 p = list_entry(prev, struct mount, mnt_child);
951 prev = p->mnt_mounts.prev;
957 * vfs_create_mount - Create a mount for a configured superblock
958 * @fc: The configuration context with the superblock attached
960 * Create a mount to an already configured superblock. If necessary, the
961 * caller should invoke vfs_get_tree() before calling this.
963 * Note that this does not attach the mount to anything.
965 struct vfsmount *vfs_create_mount(struct fs_context *fc)
970 return ERR_PTR(-EINVAL);
972 mnt = alloc_vfsmnt(fc->source ?: "none");
974 return ERR_PTR(-ENOMEM);
976 if (fc->sb_flags & SB_KERNMOUNT)
977 mnt->mnt.mnt_flags = MNT_INTERNAL;
979 atomic_inc(&fc->root->d_sb->s_active);
980 mnt->mnt.mnt_sb = fc->root->d_sb;
981 mnt->mnt.mnt_root = dget(fc->root);
982 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
983 mnt->mnt_parent = mnt;
986 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
990 EXPORT_SYMBOL(vfs_create_mount);
992 struct vfsmount *fc_mount(struct fs_context *fc)
994 int err = vfs_get_tree(fc);
996 up_write(&fc->root->d_sb->s_umount);
997 return vfs_create_mount(fc);
1001 EXPORT_SYMBOL(fc_mount);
1003 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1004 int flags, const char *name,
1007 struct fs_context *fc;
1008 struct vfsmount *mnt;
1012 return ERR_PTR(-EINVAL);
1014 fc = fs_context_for_mount(type, flags);
1016 return ERR_CAST(fc);
1019 ret = vfs_parse_fs_string(fc, "source",
1020 name, strlen(name));
1022 ret = parse_monolithic_mount_data(fc, data);
1031 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1034 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1035 const char *name, void *data)
1037 /* Until it is worked out how to pass the user namespace
1038 * through from the parent mount to the submount don't support
1039 * unprivileged mounts with submounts.
1041 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1042 return ERR_PTR(-EPERM);
1044 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1046 EXPORT_SYMBOL_GPL(vfs_submount);
1048 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1051 struct super_block *sb = old->mnt.mnt_sb;
1055 mnt = alloc_vfsmnt(old->mnt_devname);
1057 return ERR_PTR(-ENOMEM);
1059 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1060 mnt->mnt_group_id = 0; /* not a peer of original */
1062 mnt->mnt_group_id = old->mnt_group_id;
1064 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1065 err = mnt_alloc_group_id(mnt);
1070 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1071 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1073 atomic_inc(&sb->s_active);
1074 mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1075 if (mnt->mnt.mnt_userns != &init_user_ns)
1076 mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1077 mnt->mnt.mnt_sb = sb;
1078 mnt->mnt.mnt_root = dget(root);
1079 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1080 mnt->mnt_parent = mnt;
1082 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1083 unlock_mount_hash();
1085 if ((flag & CL_SLAVE) ||
1086 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1087 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1088 mnt->mnt_master = old;
1089 CLEAR_MNT_SHARED(mnt);
1090 } else if (!(flag & CL_PRIVATE)) {
1091 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1092 list_add(&mnt->mnt_share, &old->mnt_share);
1093 if (IS_MNT_SLAVE(old))
1094 list_add(&mnt->mnt_slave, &old->mnt_slave);
1095 mnt->mnt_master = old->mnt_master;
1097 CLEAR_MNT_SHARED(mnt);
1099 if (flag & CL_MAKE_SHARED)
1100 set_mnt_shared(mnt);
1102 /* stick the duplicate mount on the same expiry list
1103 * as the original if that was on one */
1104 if (flag & CL_EXPIRE) {
1105 if (!list_empty(&old->mnt_expire))
1106 list_add(&mnt->mnt_expire, &old->mnt_expire);
1114 return ERR_PTR(err);
1117 static void cleanup_mnt(struct mount *mnt)
1119 struct hlist_node *p;
1122 * The warning here probably indicates that somebody messed
1123 * up a mnt_want/drop_write() pair. If this happens, the
1124 * filesystem was probably unable to make r/w->r/o transitions.
1125 * The locking used to deal with mnt_count decrement provides barriers,
1126 * so mnt_get_writers() below is safe.
1128 WARN_ON(mnt_get_writers(mnt));
1129 if (unlikely(mnt->mnt_pins.first))
1131 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1132 hlist_del(&m->mnt_umount);
1135 fsnotify_vfsmount_delete(&mnt->mnt);
1136 dput(mnt->mnt.mnt_root);
1137 deactivate_super(mnt->mnt.mnt_sb);
1139 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1142 static void __cleanup_mnt(struct rcu_head *head)
1144 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1147 static LLIST_HEAD(delayed_mntput_list);
1148 static void delayed_mntput(struct work_struct *unused)
1150 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1151 struct mount *m, *t;
1153 llist_for_each_entry_safe(m, t, node, mnt_llist)
1156 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1158 static void mntput_no_expire(struct mount *mnt)
1164 if (likely(READ_ONCE(mnt->mnt_ns))) {
1166 * Since we don't do lock_mount_hash() here,
1167 * ->mnt_ns can change under us. However, if it's
1168 * non-NULL, then there's a reference that won't
1169 * be dropped until after an RCU delay done after
1170 * turning ->mnt_ns NULL. So if we observe it
1171 * non-NULL under rcu_read_lock(), the reference
1172 * we are dropping is not the final one.
1174 mnt_add_count(mnt, -1);
1180 * make sure that if __legitimize_mnt() has not seen us grab
1181 * mount_lock, we'll see their refcount increment here.
1184 mnt_add_count(mnt, -1);
1185 count = mnt_get_count(mnt);
1189 unlock_mount_hash();
1192 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1194 unlock_mount_hash();
1197 mnt->mnt.mnt_flags |= MNT_DOOMED;
1200 list_del(&mnt->mnt_instance);
1202 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1203 struct mount *p, *tmp;
1204 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1205 __put_mountpoint(unhash_mnt(p), &list);
1206 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1209 unlock_mount_hash();
1210 shrink_dentry_list(&list);
1212 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1213 struct task_struct *task = current;
1214 if (likely(!(task->flags & PF_KTHREAD))) {
1215 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1216 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1219 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1220 schedule_delayed_work(&delayed_mntput_work, 1);
1226 void mntput(struct vfsmount *mnt)
1229 struct mount *m = real_mount(mnt);
1230 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1231 if (unlikely(m->mnt_expiry_mark))
1232 m->mnt_expiry_mark = 0;
1233 mntput_no_expire(m);
1236 EXPORT_SYMBOL(mntput);
1238 struct vfsmount *mntget(struct vfsmount *mnt)
1241 mnt_add_count(real_mount(mnt), 1);
1244 EXPORT_SYMBOL(mntget);
1247 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1248 * @path: path to check
1250 * d_mountpoint() can only be used reliably to establish if a dentry is
1251 * not mounted in any namespace and that common case is handled inline.
1252 * d_mountpoint() isn't aware of the possibility there may be multiple
1253 * mounts using a given dentry in a different namespace. This function
1254 * checks if the passed in path is a mountpoint rather than the dentry
1257 bool path_is_mountpoint(const struct path *path)
1262 if (!d_mountpoint(path->dentry))
1267 seq = read_seqbegin(&mount_lock);
1268 res = __path_is_mountpoint(path);
1269 } while (read_seqretry(&mount_lock, seq));
1274 EXPORT_SYMBOL(path_is_mountpoint);
1276 struct vfsmount *mnt_clone_internal(const struct path *path)
1279 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1282 p->mnt.mnt_flags |= MNT_INTERNAL;
1286 #ifdef CONFIG_PROC_FS
1287 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1288 struct list_head *p)
1290 struct mount *mnt, *ret = NULL;
1293 list_for_each_continue(p, &ns->list) {
1294 mnt = list_entry(p, typeof(*mnt), mnt_list);
1295 if (!mnt_is_cursor(mnt)) {
1305 /* iterator; we want it to have access to namespace_sem, thus here... */
1306 static void *m_start(struct seq_file *m, loff_t *pos)
1308 struct proc_mounts *p = m->private;
1309 struct list_head *prev;
1311 down_read(&namespace_sem);
1313 prev = &p->ns->list;
1315 prev = &p->cursor.mnt_list;
1317 /* Read after we'd reached the end? */
1318 if (list_empty(prev))
1322 return mnt_list_next(p->ns, prev);
1325 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1327 struct proc_mounts *p = m->private;
1328 struct mount *mnt = v;
1331 return mnt_list_next(p->ns, &mnt->mnt_list);
1334 static void m_stop(struct seq_file *m, void *v)
1336 struct proc_mounts *p = m->private;
1337 struct mount *mnt = v;
1339 lock_ns_list(p->ns);
1341 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1343 list_del_init(&p->cursor.mnt_list);
1344 unlock_ns_list(p->ns);
1345 up_read(&namespace_sem);
1348 static int m_show(struct seq_file *m, void *v)
1350 struct proc_mounts *p = m->private;
1351 struct mount *r = v;
1352 return p->show(m, &r->mnt);
1355 const struct seq_operations mounts_op = {
1362 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1364 down_read(&namespace_sem);
1366 list_del(&cursor->mnt_list);
1368 up_read(&namespace_sem);
1370 #endif /* CONFIG_PROC_FS */
1373 * may_umount_tree - check if a mount tree is busy
1374 * @m: root of mount tree
1376 * This is called to check if a tree of mounts has any
1377 * open files, pwds, chroots or sub mounts that are
1380 int may_umount_tree(struct vfsmount *m)
1382 struct mount *mnt = real_mount(m);
1383 int actual_refs = 0;
1384 int minimum_refs = 0;
1388 /* write lock needed for mnt_get_count */
1390 for (p = mnt; p; p = next_mnt(p, mnt)) {
1391 actual_refs += mnt_get_count(p);
1394 unlock_mount_hash();
1396 if (actual_refs > minimum_refs)
1402 EXPORT_SYMBOL(may_umount_tree);
1405 * may_umount - check if a mount point is busy
1406 * @mnt: root of mount
1408 * This is called to check if a mount point has any
1409 * open files, pwds, chroots or sub mounts. If the
1410 * mount has sub mounts this will return busy
1411 * regardless of whether the sub mounts are busy.
1413 * Doesn't take quota and stuff into account. IOW, in some cases it will
1414 * give false negatives. The main reason why it's here is that we need
1415 * a non-destructive way to look for easily umountable filesystems.
1417 int may_umount(struct vfsmount *mnt)
1420 down_read(&namespace_sem);
1422 if (propagate_mount_busy(real_mount(mnt), 2))
1424 unlock_mount_hash();
1425 up_read(&namespace_sem);
1429 EXPORT_SYMBOL(may_umount);
1431 static void namespace_unlock(void)
1433 struct hlist_head head;
1434 struct hlist_node *p;
1438 hlist_move_list(&unmounted, &head);
1439 list_splice_init(&ex_mountpoints, &list);
1441 up_write(&namespace_sem);
1443 shrink_dentry_list(&list);
1445 if (likely(hlist_empty(&head)))
1448 synchronize_rcu_expedited();
1450 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1451 hlist_del(&m->mnt_umount);
1456 static inline void namespace_lock(void)
1458 down_write(&namespace_sem);
1461 enum umount_tree_flags {
1463 UMOUNT_PROPAGATE = 2,
1464 UMOUNT_CONNECTED = 4,
1467 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1469 /* Leaving mounts connected is only valid for lazy umounts */
1470 if (how & UMOUNT_SYNC)
1473 /* A mount without a parent has nothing to be connected to */
1474 if (!mnt_has_parent(mnt))
1477 /* Because the reference counting rules change when mounts are
1478 * unmounted and connected, umounted mounts may not be
1479 * connected to mounted mounts.
1481 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1484 /* Has it been requested that the mount remain connected? */
1485 if (how & UMOUNT_CONNECTED)
1488 /* Is the mount locked such that it needs to remain connected? */
1489 if (IS_MNT_LOCKED(mnt))
1492 /* By default disconnect the mount */
1497 * mount_lock must be held
1498 * namespace_sem must be held for write
1500 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1502 LIST_HEAD(tmp_list);
1505 if (how & UMOUNT_PROPAGATE)
1506 propagate_mount_unlock(mnt);
1508 /* Gather the mounts to umount */
1509 for (p = mnt; p; p = next_mnt(p, mnt)) {
1510 p->mnt.mnt_flags |= MNT_UMOUNT;
1511 list_move(&p->mnt_list, &tmp_list);
1514 /* Hide the mounts from mnt_mounts */
1515 list_for_each_entry(p, &tmp_list, mnt_list) {
1516 list_del_init(&p->mnt_child);
1519 /* Add propogated mounts to the tmp_list */
1520 if (how & UMOUNT_PROPAGATE)
1521 propagate_umount(&tmp_list);
1523 while (!list_empty(&tmp_list)) {
1524 struct mnt_namespace *ns;
1526 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1527 list_del_init(&p->mnt_expire);
1528 list_del_init(&p->mnt_list);
1532 __touch_mnt_namespace(ns);
1535 if (how & UMOUNT_SYNC)
1536 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1538 disconnect = disconnect_mount(p, how);
1539 if (mnt_has_parent(p)) {
1540 mnt_add_count(p->mnt_parent, -1);
1542 /* Don't forget about p */
1543 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1548 change_mnt_propagation(p, MS_PRIVATE);
1550 hlist_add_head(&p->mnt_umount, &unmounted);
1554 static void shrink_submounts(struct mount *mnt);
1556 static int do_umount_root(struct super_block *sb)
1560 down_write(&sb->s_umount);
1561 if (!sb_rdonly(sb)) {
1562 struct fs_context *fc;
1564 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1569 ret = parse_monolithic_mount_data(fc, NULL);
1571 ret = reconfigure_super(fc);
1575 up_write(&sb->s_umount);
1579 static int do_umount(struct mount *mnt, int flags)
1581 struct super_block *sb = mnt->mnt.mnt_sb;
1584 retval = security_sb_umount(&mnt->mnt, flags);
1589 * Allow userspace to request a mountpoint be expired rather than
1590 * unmounting unconditionally. Unmount only happens if:
1591 * (1) the mark is already set (the mark is cleared by mntput())
1592 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1594 if (flags & MNT_EXPIRE) {
1595 if (&mnt->mnt == current->fs->root.mnt ||
1596 flags & (MNT_FORCE | MNT_DETACH))
1600 * probably don't strictly need the lock here if we examined
1601 * all race cases, but it's a slowpath.
1604 if (mnt_get_count(mnt) != 2) {
1605 unlock_mount_hash();
1608 unlock_mount_hash();
1610 if (!xchg(&mnt->mnt_expiry_mark, 1))
1615 * If we may have to abort operations to get out of this
1616 * mount, and they will themselves hold resources we must
1617 * allow the fs to do things. In the Unix tradition of
1618 * 'Gee thats tricky lets do it in userspace' the umount_begin
1619 * might fail to complete on the first run through as other tasks
1620 * must return, and the like. Thats for the mount program to worry
1621 * about for the moment.
1624 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1625 sb->s_op->umount_begin(sb);
1629 * No sense to grab the lock for this test, but test itself looks
1630 * somewhat bogus. Suggestions for better replacement?
1631 * Ho-hum... In principle, we might treat that as umount + switch
1632 * to rootfs. GC would eventually take care of the old vfsmount.
1633 * Actually it makes sense, especially if rootfs would contain a
1634 * /reboot - static binary that would close all descriptors and
1635 * call reboot(9). Then init(8) could umount root and exec /reboot.
1637 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1639 * Special case for "unmounting" root ...
1640 * we just try to remount it readonly.
1642 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1644 return do_umount_root(sb);
1650 /* Recheck MNT_LOCKED with the locks held */
1652 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1656 if (flags & MNT_DETACH) {
1657 if (!list_empty(&mnt->mnt_list))
1658 umount_tree(mnt, UMOUNT_PROPAGATE);
1661 shrink_submounts(mnt);
1663 if (!propagate_mount_busy(mnt, 2)) {
1664 if (!list_empty(&mnt->mnt_list))
1665 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1670 unlock_mount_hash();
1676 * __detach_mounts - lazily unmount all mounts on the specified dentry
1678 * During unlink, rmdir, and d_drop it is possible to loose the path
1679 * to an existing mountpoint, and wind up leaking the mount.
1680 * detach_mounts allows lazily unmounting those mounts instead of
1683 * The caller may hold dentry->d_inode->i_mutex.
1685 void __detach_mounts(struct dentry *dentry)
1687 struct mountpoint *mp;
1692 mp = lookup_mountpoint(dentry);
1697 while (!hlist_empty(&mp->m_list)) {
1698 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1699 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1701 hlist_add_head(&mnt->mnt_umount, &unmounted);
1703 else umount_tree(mnt, UMOUNT_CONNECTED);
1707 unlock_mount_hash();
1712 * Is the caller allowed to modify his namespace?
1714 static inline bool may_mount(void)
1716 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1719 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1720 static bool may_mandlock(void)
1722 pr_warn_once("======================================================\n"
1723 "WARNING: the mand mount option is being deprecated and\n"
1724 " will be removed in v5.15!\n"
1725 "======================================================\n");
1726 return capable(CAP_SYS_ADMIN);
1729 static inline bool may_mandlock(void)
1731 pr_warn("VFS: \"mand\" mount option not supported");
1736 static int can_umount(const struct path *path, int flags)
1738 struct mount *mnt = real_mount(path->mnt);
1742 if (path->dentry != path->mnt->mnt_root)
1744 if (!check_mnt(mnt))
1746 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1748 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1753 // caller is responsible for flags being sane
1754 int path_umount(struct path *path, int flags)
1756 struct mount *mnt = real_mount(path->mnt);
1759 ret = can_umount(path, flags);
1761 ret = do_umount(mnt, flags);
1763 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1765 mntput_no_expire(mnt);
1769 static int ksys_umount(char __user *name, int flags)
1771 int lookup_flags = LOOKUP_MOUNTPOINT;
1775 // basic validity checks done first
1776 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1779 if (!(flags & UMOUNT_NOFOLLOW))
1780 lookup_flags |= LOOKUP_FOLLOW;
1781 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1784 return path_umount(&path, flags);
1787 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1789 return ksys_umount(name, flags);
1792 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1795 * The 2.0 compatible umount. No flags.
1797 SYSCALL_DEFINE1(oldumount, char __user *, name)
1799 return ksys_umount(name, 0);
1804 static bool is_mnt_ns_file(struct dentry *dentry)
1806 /* Is this a proxy for a mount namespace? */
1807 return dentry->d_op == &ns_dentry_operations &&
1808 dentry->d_fsdata == &mntns_operations;
1811 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1813 return container_of(ns, struct mnt_namespace, ns);
1816 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1821 static bool mnt_ns_loop(struct dentry *dentry)
1823 /* Could bind mounting the mount namespace inode cause a
1824 * mount namespace loop?
1826 struct mnt_namespace *mnt_ns;
1827 if (!is_mnt_ns_file(dentry))
1830 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1831 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1834 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1837 struct mount *res, *p, *q, *r, *parent;
1839 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1840 return ERR_PTR(-EINVAL);
1842 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1843 return ERR_PTR(-EINVAL);
1845 res = q = clone_mnt(mnt, dentry, flag);
1849 q->mnt_mountpoint = mnt->mnt_mountpoint;
1852 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1854 if (!is_subdir(r->mnt_mountpoint, dentry))
1857 for (s = r; s; s = next_mnt(s, r)) {
1858 if (!(flag & CL_COPY_UNBINDABLE) &&
1859 IS_MNT_UNBINDABLE(s)) {
1860 if (s->mnt.mnt_flags & MNT_LOCKED) {
1861 /* Both unbindable and locked. */
1862 q = ERR_PTR(-EPERM);
1865 s = skip_mnt_tree(s);
1869 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1870 is_mnt_ns_file(s->mnt.mnt_root)) {
1871 s = skip_mnt_tree(s);
1874 while (p != s->mnt_parent) {
1880 q = clone_mnt(p, p->mnt.mnt_root, flag);
1884 list_add_tail(&q->mnt_list, &res->mnt_list);
1885 attach_mnt(q, parent, p->mnt_mp);
1886 unlock_mount_hash();
1893 umount_tree(res, UMOUNT_SYNC);
1894 unlock_mount_hash();
1899 /* Caller should check returned pointer for errors */
1901 struct vfsmount *collect_mounts(const struct path *path)
1905 if (!check_mnt(real_mount(path->mnt)))
1906 tree = ERR_PTR(-EINVAL);
1908 tree = copy_tree(real_mount(path->mnt), path->dentry,
1909 CL_COPY_ALL | CL_PRIVATE);
1912 return ERR_CAST(tree);
1916 static void free_mnt_ns(struct mnt_namespace *);
1917 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1919 void dissolve_on_fput(struct vfsmount *mnt)
1921 struct mnt_namespace *ns;
1924 ns = real_mount(mnt)->mnt_ns;
1927 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1931 unlock_mount_hash();
1937 void drop_collected_mounts(struct vfsmount *mnt)
1941 umount_tree(real_mount(mnt), 0);
1942 unlock_mount_hash();
1946 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1948 struct mount *child;
1950 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1951 if (!is_subdir(child->mnt_mountpoint, dentry))
1954 if (child->mnt.mnt_flags & MNT_LOCKED)
1961 * clone_private_mount - create a private clone of a path
1962 * @path: path to clone
1964 * This creates a new vfsmount, which will be the clone of @path. The new mount
1965 * will not be attached anywhere in the namespace and will be private (i.e.
1966 * changes to the originating mount won't be propagated into this).
1968 * Release with mntput().
1970 struct vfsmount *clone_private_mount(const struct path *path)
1972 struct mount *old_mnt = real_mount(path->mnt);
1973 struct mount *new_mnt;
1975 down_read(&namespace_sem);
1976 if (IS_MNT_UNBINDABLE(old_mnt))
1979 if (!check_mnt(old_mnt))
1982 if (has_locked_children(old_mnt, path->dentry))
1985 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1986 up_read(&namespace_sem);
1988 if (IS_ERR(new_mnt))
1989 return ERR_CAST(new_mnt);
1991 /* Longterm mount to be removed by kern_unmount*() */
1992 new_mnt->mnt_ns = MNT_NS_INTERNAL;
1994 return &new_mnt->mnt;
1997 up_read(&namespace_sem);
1998 return ERR_PTR(-EINVAL);
2000 EXPORT_SYMBOL_GPL(clone_private_mount);
2002 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2003 struct vfsmount *root)
2006 int res = f(root, arg);
2009 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2010 res = f(&mnt->mnt, arg);
2017 static void lock_mnt_tree(struct mount *mnt)
2021 for (p = mnt; p; p = next_mnt(p, mnt)) {
2022 int flags = p->mnt.mnt_flags;
2023 /* Don't allow unprivileged users to change mount flags */
2024 flags |= MNT_LOCK_ATIME;
2026 if (flags & MNT_READONLY)
2027 flags |= MNT_LOCK_READONLY;
2029 if (flags & MNT_NODEV)
2030 flags |= MNT_LOCK_NODEV;
2032 if (flags & MNT_NOSUID)
2033 flags |= MNT_LOCK_NOSUID;
2035 if (flags & MNT_NOEXEC)
2036 flags |= MNT_LOCK_NOEXEC;
2037 /* Don't allow unprivileged users to reveal what is under a mount */
2038 if (list_empty(&p->mnt_expire))
2039 flags |= MNT_LOCKED;
2040 p->mnt.mnt_flags = flags;
2044 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2048 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2049 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2050 mnt_release_group_id(p);
2054 static int invent_group_ids(struct mount *mnt, bool recurse)
2058 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2059 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2060 int err = mnt_alloc_group_id(p);
2062 cleanup_group_ids(mnt, p);
2071 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2073 unsigned int max = READ_ONCE(sysctl_mount_max);
2074 unsigned int mounts = 0, old, pending, sum;
2077 for (p = mnt; p; p = next_mnt(p, mnt))
2081 pending = ns->pending_mounts;
2082 sum = old + pending;
2086 (mounts > (max - sum)))
2089 ns->pending_mounts = pending + mounts;
2094 * @source_mnt : mount tree to be attached
2095 * @nd : place the mount tree @source_mnt is attached
2096 * @parent_nd : if non-null, detach the source_mnt from its parent and
2097 * store the parent mount and mountpoint dentry.
2098 * (done when source_mnt is moved)
2100 * NOTE: in the table below explains the semantics when a source mount
2101 * of a given type is attached to a destination mount of a given type.
2102 * ---------------------------------------------------------------------------
2103 * | BIND MOUNT OPERATION |
2104 * |**************************************************************************
2105 * | source-->| shared | private | slave | unbindable |
2109 * |**************************************************************************
2110 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2112 * |non-shared| shared (+) | private | slave (*) | invalid |
2113 * ***************************************************************************
2114 * A bind operation clones the source mount and mounts the clone on the
2115 * destination mount.
2117 * (++) the cloned mount is propagated to all the mounts in the propagation
2118 * tree of the destination mount and the cloned mount is added to
2119 * the peer group of the source mount.
2120 * (+) the cloned mount is created under the destination mount and is marked
2121 * as shared. The cloned mount is added to the peer group of the source
2123 * (+++) the mount is propagated to all the mounts in the propagation tree
2124 * of the destination mount and the cloned mount is made slave
2125 * of the same master as that of the source mount. The cloned mount
2126 * is marked as 'shared and slave'.
2127 * (*) the cloned mount is made a slave of the same master as that of the
2130 * ---------------------------------------------------------------------------
2131 * | MOVE MOUNT OPERATION |
2132 * |**************************************************************************
2133 * | source-->| shared | private | slave | unbindable |
2137 * |**************************************************************************
2138 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2140 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2141 * ***************************************************************************
2143 * (+) the mount is moved to the destination. And is then propagated to
2144 * all the mounts in the propagation tree of the destination mount.
2145 * (+*) the mount is moved to the destination.
2146 * (+++) the mount is moved to the destination and is then propagated to
2147 * all the mounts belonging to the destination mount's propagation tree.
2148 * the mount is marked as 'shared and slave'.
2149 * (*) the mount continues to be a slave at the new location.
2151 * if the source mount is a tree, the operations explained above is
2152 * applied to each mount in the tree.
2153 * Must be called without spinlocks held, since this function can sleep
2156 static int attach_recursive_mnt(struct mount *source_mnt,
2157 struct mount *dest_mnt,
2158 struct mountpoint *dest_mp,
2161 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2162 HLIST_HEAD(tree_list);
2163 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2164 struct mountpoint *smp;
2165 struct mount *child, *p;
2166 struct hlist_node *n;
2169 /* Preallocate a mountpoint in case the new mounts need
2170 * to be tucked under other mounts.
2172 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2174 return PTR_ERR(smp);
2176 /* Is there space to add these mounts to the mount namespace? */
2178 err = count_mounts(ns, source_mnt);
2183 if (IS_MNT_SHARED(dest_mnt)) {
2184 err = invent_group_ids(source_mnt, true);
2187 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2190 goto out_cleanup_ids;
2191 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2197 unhash_mnt(source_mnt);
2198 attach_mnt(source_mnt, dest_mnt, dest_mp);
2199 touch_mnt_namespace(source_mnt->mnt_ns);
2201 if (source_mnt->mnt_ns) {
2202 /* move from anon - the caller will destroy */
2203 list_del_init(&source_mnt->mnt_ns->list);
2205 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2206 commit_tree(source_mnt);
2209 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2211 hlist_del_init(&child->mnt_hash);
2212 q = __lookup_mnt(&child->mnt_parent->mnt,
2213 child->mnt_mountpoint);
2215 mnt_change_mountpoint(child, smp, q);
2216 /* Notice when we are propagating across user namespaces */
2217 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2218 lock_mnt_tree(child);
2219 child->mnt.mnt_flags &= ~MNT_LOCKED;
2222 put_mountpoint(smp);
2223 unlock_mount_hash();
2228 while (!hlist_empty(&tree_list)) {
2229 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2230 child->mnt_parent->mnt_ns->pending_mounts = 0;
2231 umount_tree(child, UMOUNT_SYNC);
2233 unlock_mount_hash();
2234 cleanup_group_ids(source_mnt, NULL);
2236 ns->pending_mounts = 0;
2238 read_seqlock_excl(&mount_lock);
2239 put_mountpoint(smp);
2240 read_sequnlock_excl(&mount_lock);
2245 static struct mountpoint *lock_mount(struct path *path)
2247 struct vfsmount *mnt;
2248 struct dentry *dentry = path->dentry;
2250 inode_lock(dentry->d_inode);
2251 if (unlikely(cant_mount(dentry))) {
2252 inode_unlock(dentry->d_inode);
2253 return ERR_PTR(-ENOENT);
2256 mnt = lookup_mnt(path);
2258 struct mountpoint *mp = get_mountpoint(dentry);
2261 inode_unlock(dentry->d_inode);
2267 inode_unlock(path->dentry->d_inode);
2270 dentry = path->dentry = dget(mnt->mnt_root);
2274 static void unlock_mount(struct mountpoint *where)
2276 struct dentry *dentry = where->m_dentry;
2278 read_seqlock_excl(&mount_lock);
2279 put_mountpoint(where);
2280 read_sequnlock_excl(&mount_lock);
2283 inode_unlock(dentry->d_inode);
2286 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2288 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2291 if (d_is_dir(mp->m_dentry) !=
2292 d_is_dir(mnt->mnt.mnt_root))
2295 return attach_recursive_mnt(mnt, p, mp, false);
2299 * Sanity check the flags to change_mnt_propagation.
2302 static int flags_to_propagation_type(int ms_flags)
2304 int type = ms_flags & ~(MS_REC | MS_SILENT);
2306 /* Fail if any non-propagation flags are set */
2307 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2309 /* Only one propagation flag should be set */
2310 if (!is_power_of_2(type))
2316 * recursively change the type of the mountpoint.
2318 static int do_change_type(struct path *path, int ms_flags)
2321 struct mount *mnt = real_mount(path->mnt);
2322 int recurse = ms_flags & MS_REC;
2326 if (path->dentry != path->mnt->mnt_root)
2329 type = flags_to_propagation_type(ms_flags);
2334 if (type == MS_SHARED) {
2335 err = invent_group_ids(mnt, recurse);
2341 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2342 change_mnt_propagation(m, type);
2343 unlock_mount_hash();
2350 static struct mount *__do_loopback(struct path *old_path, int recurse)
2352 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2354 if (IS_MNT_UNBINDABLE(old))
2357 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2360 if (!recurse && has_locked_children(old, old_path->dentry))
2364 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2366 mnt = clone_mnt(old, old_path->dentry, 0);
2369 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2375 * do loopback mount.
2377 static int do_loopback(struct path *path, const char *old_name,
2380 struct path old_path;
2381 struct mount *mnt = NULL, *parent;
2382 struct mountpoint *mp;
2384 if (!old_name || !*old_name)
2386 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2391 if (mnt_ns_loop(old_path.dentry))
2394 mp = lock_mount(path);
2400 parent = real_mount(path->mnt);
2401 if (!check_mnt(parent))
2404 mnt = __do_loopback(&old_path, recurse);
2410 err = graft_tree(mnt, parent, mp);
2413 umount_tree(mnt, UMOUNT_SYNC);
2414 unlock_mount_hash();
2419 path_put(&old_path);
2423 static struct file *open_detached_copy(struct path *path, bool recursive)
2425 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2426 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2427 struct mount *mnt, *p;
2431 return ERR_CAST(ns);
2434 mnt = __do_loopback(path, recursive);
2438 return ERR_CAST(mnt);
2442 for (p = mnt; p; p = next_mnt(p, mnt)) {
2447 list_add_tail(&ns->list, &mnt->mnt_list);
2449 unlock_mount_hash();
2453 path->mnt = &mnt->mnt;
2454 file = dentry_open(path, O_PATH, current_cred());
2456 dissolve_on_fput(path->mnt);
2458 file->f_mode |= FMODE_NEED_UNMOUNT;
2462 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2466 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2467 bool detached = flags & OPEN_TREE_CLONE;
2471 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2473 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2474 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2478 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2481 if (flags & AT_NO_AUTOMOUNT)
2482 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2483 if (flags & AT_SYMLINK_NOFOLLOW)
2484 lookup_flags &= ~LOOKUP_FOLLOW;
2485 if (flags & AT_EMPTY_PATH)
2486 lookup_flags |= LOOKUP_EMPTY;
2488 if (detached && !may_mount())
2491 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2495 error = user_path_at(dfd, filename, lookup_flags, &path);
2496 if (unlikely(error)) {
2497 file = ERR_PTR(error);
2500 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2502 file = dentry_open(&path, O_PATH, current_cred());
2507 return PTR_ERR(file);
2509 fd_install(fd, file);
2514 * Don't allow locked mount flags to be cleared.
2516 * No locks need to be held here while testing the various MNT_LOCK
2517 * flags because those flags can never be cleared once they are set.
2519 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2521 unsigned int fl = mnt->mnt.mnt_flags;
2523 if ((fl & MNT_LOCK_READONLY) &&
2524 !(mnt_flags & MNT_READONLY))
2527 if ((fl & MNT_LOCK_NODEV) &&
2528 !(mnt_flags & MNT_NODEV))
2531 if ((fl & MNT_LOCK_NOSUID) &&
2532 !(mnt_flags & MNT_NOSUID))
2535 if ((fl & MNT_LOCK_NOEXEC) &&
2536 !(mnt_flags & MNT_NOEXEC))
2539 if ((fl & MNT_LOCK_ATIME) &&
2540 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2546 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2548 bool readonly_request = (mnt_flags & MNT_READONLY);
2550 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2553 if (readonly_request)
2554 return mnt_make_readonly(mnt);
2556 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2560 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2562 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2563 mnt->mnt.mnt_flags = mnt_flags;
2564 touch_mnt_namespace(mnt->mnt_ns);
2567 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2569 struct super_block *sb = mnt->mnt_sb;
2571 if (!__mnt_is_readonly(mnt) &&
2572 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2573 char *buf = (char *)__get_free_page(GFP_KERNEL);
2574 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2577 time64_to_tm(sb->s_time_max, 0, &tm);
2579 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2581 is_mounted(mnt) ? "remounted" : "mounted",
2583 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2585 free_page((unsigned long)buf);
2590 * Handle reconfiguration of the mountpoint only without alteration of the
2591 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2594 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2596 struct super_block *sb = path->mnt->mnt_sb;
2597 struct mount *mnt = real_mount(path->mnt);
2600 if (!check_mnt(mnt))
2603 if (path->dentry != mnt->mnt.mnt_root)
2606 if (!can_change_locked_flags(mnt, mnt_flags))
2610 * We're only checking whether the superblock is read-only not
2611 * changing it, so only take down_read(&sb->s_umount).
2613 down_read(&sb->s_umount);
2615 ret = change_mount_ro_state(mnt, mnt_flags);
2617 set_mount_attributes(mnt, mnt_flags);
2618 unlock_mount_hash();
2619 up_read(&sb->s_umount);
2621 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2627 * change filesystem flags. dir should be a physical root of filesystem.
2628 * If you've mounted a non-root directory somewhere and want to do remount
2629 * on it - tough luck.
2631 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2632 int mnt_flags, void *data)
2635 struct super_block *sb = path->mnt->mnt_sb;
2636 struct mount *mnt = real_mount(path->mnt);
2637 struct fs_context *fc;
2639 if (!check_mnt(mnt))
2642 if (path->dentry != path->mnt->mnt_root)
2645 if (!can_change_locked_flags(mnt, mnt_flags))
2648 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2653 err = parse_monolithic_mount_data(fc, data);
2655 down_write(&sb->s_umount);
2657 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2658 err = reconfigure_super(fc);
2661 set_mount_attributes(mnt, mnt_flags);
2662 unlock_mount_hash();
2665 up_write(&sb->s_umount);
2668 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2674 static inline int tree_contains_unbindable(struct mount *mnt)
2677 for (p = mnt; p; p = next_mnt(p, mnt)) {
2678 if (IS_MNT_UNBINDABLE(p))
2685 * Check that there aren't references to earlier/same mount namespaces in the
2686 * specified subtree. Such references can act as pins for mount namespaces
2687 * that aren't checked by the mount-cycle checking code, thereby allowing
2688 * cycles to be made.
2690 static bool check_for_nsfs_mounts(struct mount *subtree)
2696 for (p = subtree; p; p = next_mnt(p, subtree))
2697 if (mnt_ns_loop(p->mnt.mnt_root))
2702 unlock_mount_hash();
2706 static int do_move_mount(struct path *old_path, struct path *new_path)
2708 struct mnt_namespace *ns;
2711 struct mount *parent;
2712 struct mountpoint *mp, *old_mp;
2716 mp = lock_mount(new_path);
2720 old = real_mount(old_path->mnt);
2721 p = real_mount(new_path->mnt);
2722 parent = old->mnt_parent;
2723 attached = mnt_has_parent(old);
2724 old_mp = old->mnt_mp;
2728 /* The mountpoint must be in our namespace. */
2732 /* The thing moved must be mounted... */
2733 if (!is_mounted(&old->mnt))
2736 /* ... and either ours or the root of anon namespace */
2737 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2740 if (old->mnt.mnt_flags & MNT_LOCKED)
2743 if (old_path->dentry != old_path->mnt->mnt_root)
2746 if (d_is_dir(new_path->dentry) !=
2747 d_is_dir(old_path->dentry))
2750 * Don't move a mount residing in a shared parent.
2752 if (attached && IS_MNT_SHARED(parent))
2755 * Don't move a mount tree containing unbindable mounts to a destination
2756 * mount which is shared.
2758 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2761 if (!check_for_nsfs_mounts(old))
2763 for (; mnt_has_parent(p); p = p->mnt_parent)
2767 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2772 /* if the mount is moved, it should no longer be expire
2774 list_del_init(&old->mnt_expire);
2776 put_mountpoint(old_mp);
2781 mntput_no_expire(parent);
2788 static int do_move_mount_old(struct path *path, const char *old_name)
2790 struct path old_path;
2793 if (!old_name || !*old_name)
2796 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2800 err = do_move_mount(&old_path, path);
2801 path_put(&old_path);
2806 * add a mount into a namespace's mount tree
2808 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2809 struct path *path, int mnt_flags)
2811 struct mount *parent = real_mount(path->mnt);
2813 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2815 if (unlikely(!check_mnt(parent))) {
2816 /* that's acceptable only for automounts done in private ns */
2817 if (!(mnt_flags & MNT_SHRINKABLE))
2819 /* ... and for those we'd better have mountpoint still alive */
2820 if (!parent->mnt_ns)
2824 /* Refuse the same filesystem on the same mount point */
2825 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2826 path->mnt->mnt_root == path->dentry)
2829 if (d_is_symlink(newmnt->mnt.mnt_root))
2832 newmnt->mnt.mnt_flags = mnt_flags;
2833 return graft_tree(newmnt, parent, mp);
2836 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2839 * Create a new mount using a superblock configuration and request it
2840 * be added to the namespace tree.
2842 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2843 unsigned int mnt_flags)
2845 struct vfsmount *mnt;
2846 struct mountpoint *mp;
2847 struct super_block *sb = fc->root->d_sb;
2850 error = security_sb_kern_mount(sb);
2851 if (!error && mount_too_revealing(sb, &mnt_flags))
2854 if (unlikely(error)) {
2859 up_write(&sb->s_umount);
2861 mnt = vfs_create_mount(fc);
2863 return PTR_ERR(mnt);
2865 mnt_warn_timestamp_expiry(mountpoint, mnt);
2867 mp = lock_mount(mountpoint);
2872 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2880 * create a new mount for userspace and request it to be added into the
2883 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2884 int mnt_flags, const char *name, void *data)
2886 struct file_system_type *type;
2887 struct fs_context *fc;
2888 const char *subtype = NULL;
2894 type = get_fs_type(fstype);
2898 if (type->fs_flags & FS_HAS_SUBTYPE) {
2899 subtype = strchr(fstype, '.');
2903 put_filesystem(type);
2909 fc = fs_context_for_mount(type, sb_flags);
2910 put_filesystem(type);
2915 err = vfs_parse_fs_string(fc, "subtype",
2916 subtype, strlen(subtype));
2918 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2920 err = parse_monolithic_mount_data(fc, data);
2921 if (!err && !mount_capable(fc))
2924 err = vfs_get_tree(fc);
2926 err = do_new_mount_fc(fc, path, mnt_flags);
2932 int finish_automount(struct vfsmount *m, struct path *path)
2934 struct dentry *dentry = path->dentry;
2935 struct mountpoint *mp;
2944 mnt = real_mount(m);
2945 /* The new mount record should have at least 2 refs to prevent it being
2946 * expired before we get a chance to add it
2948 BUG_ON(mnt_get_count(mnt) < 2);
2950 if (m->mnt_sb == path->mnt->mnt_sb &&
2951 m->mnt_root == dentry) {
2957 * we don't want to use lock_mount() - in this case finding something
2958 * that overmounts our mountpoint to be means "quitely drop what we've
2959 * got", not "try to mount it on top".
2961 inode_lock(dentry->d_inode);
2963 if (unlikely(cant_mount(dentry))) {
2965 goto discard_locked;
2968 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2971 goto discard_locked;
2974 mp = get_mountpoint(dentry);
2977 goto discard_locked;
2980 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2989 inode_unlock(dentry->d_inode);
2991 /* remove m from any expiration list it may be on */
2992 if (!list_empty(&mnt->mnt_expire)) {
2994 list_del_init(&mnt->mnt_expire);
3003 * mnt_set_expiry - Put a mount on an expiration list
3004 * @mnt: The mount to list.
3005 * @expiry_list: The list to add the mount to.
3007 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3011 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3015 EXPORT_SYMBOL(mnt_set_expiry);
3018 * process a list of expirable mountpoints with the intent of discarding any
3019 * mountpoints that aren't in use and haven't been touched since last we came
3022 void mark_mounts_for_expiry(struct list_head *mounts)
3024 struct mount *mnt, *next;
3025 LIST_HEAD(graveyard);
3027 if (list_empty(mounts))
3033 /* extract from the expiration list every vfsmount that matches the
3034 * following criteria:
3035 * - only referenced by its parent vfsmount
3036 * - still marked for expiry (marked on the last call here; marks are
3037 * cleared by mntput())
3039 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3040 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3041 propagate_mount_busy(mnt, 1))
3043 list_move(&mnt->mnt_expire, &graveyard);
3045 while (!list_empty(&graveyard)) {
3046 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3047 touch_mnt_namespace(mnt->mnt_ns);
3048 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3050 unlock_mount_hash();
3054 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3057 * Ripoff of 'select_parent()'
3059 * search the list of submounts for a given mountpoint, and move any
3060 * shrinkable submounts to the 'graveyard' list.
3062 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3064 struct mount *this_parent = parent;
3065 struct list_head *next;
3069 next = this_parent->mnt_mounts.next;
3071 while (next != &this_parent->mnt_mounts) {
3072 struct list_head *tmp = next;
3073 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3076 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3079 * Descend a level if the d_mounts list is non-empty.
3081 if (!list_empty(&mnt->mnt_mounts)) {
3086 if (!propagate_mount_busy(mnt, 1)) {
3087 list_move_tail(&mnt->mnt_expire, graveyard);
3092 * All done at this level ... ascend and resume the search
3094 if (this_parent != parent) {
3095 next = this_parent->mnt_child.next;
3096 this_parent = this_parent->mnt_parent;
3103 * process a list of expirable mountpoints with the intent of discarding any
3104 * submounts of a specific parent mountpoint
3106 * mount_lock must be held for write
3108 static void shrink_submounts(struct mount *mnt)
3110 LIST_HEAD(graveyard);
3113 /* extract submounts of 'mountpoint' from the expiration list */
3114 while (select_submounts(mnt, &graveyard)) {
3115 while (!list_empty(&graveyard)) {
3116 m = list_first_entry(&graveyard, struct mount,
3118 touch_mnt_namespace(m->mnt_ns);
3119 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3124 static void *copy_mount_options(const void __user * data)
3127 unsigned left, offset;
3132 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3134 return ERR_PTR(-ENOMEM);
3136 left = copy_from_user(copy, data, PAGE_SIZE);
3139 * Not all architectures have an exact copy_from_user(). Resort to
3142 offset = PAGE_SIZE - left;
3145 if (get_user(c, (const char __user *)data + offset))
3152 if (left == PAGE_SIZE) {
3154 return ERR_PTR(-EFAULT);
3160 static char *copy_mount_string(const void __user *data)
3162 return data ? strndup_user(data, PATH_MAX) : NULL;
3166 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3167 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3169 * data is a (void *) that can point to any structure up to
3170 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3171 * information (or be NULL).
3173 * Pre-0.97 versions of mount() didn't have a flags word.
3174 * When the flags word was introduced its top half was required
3175 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3176 * Therefore, if this magic number is present, it carries no information
3177 * and must be discarded.
3179 int path_mount(const char *dev_name, struct path *path,
3180 const char *type_page, unsigned long flags, void *data_page)
3182 unsigned int mnt_flags = 0, sb_flags;
3186 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3187 flags &= ~MS_MGC_MSK;
3189 /* Basic sanity checks */
3191 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3193 if (flags & MS_NOUSER)
3196 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3201 if ((flags & SB_MANDLOCK) && !may_mandlock())
3204 /* Default to relatime unless overriden */
3205 if (!(flags & MS_NOATIME))
3206 mnt_flags |= MNT_RELATIME;
3208 /* Separate the per-mountpoint flags */
3209 if (flags & MS_NOSUID)
3210 mnt_flags |= MNT_NOSUID;
3211 if (flags & MS_NODEV)
3212 mnt_flags |= MNT_NODEV;
3213 if (flags & MS_NOEXEC)
3214 mnt_flags |= MNT_NOEXEC;
3215 if (flags & MS_NOATIME)
3216 mnt_flags |= MNT_NOATIME;
3217 if (flags & MS_NODIRATIME)
3218 mnt_flags |= MNT_NODIRATIME;
3219 if (flags & MS_STRICTATIME)
3220 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3221 if (flags & MS_RDONLY)
3222 mnt_flags |= MNT_READONLY;
3223 if (flags & MS_NOSYMFOLLOW)
3224 mnt_flags |= MNT_NOSYMFOLLOW;
3226 /* The default atime for remount is preservation */
3227 if ((flags & MS_REMOUNT) &&
3228 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3229 MS_STRICTATIME)) == 0)) {
3230 mnt_flags &= ~MNT_ATIME_MASK;
3231 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3234 sb_flags = flags & (SB_RDONLY |
3243 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3244 return do_reconfigure_mnt(path, mnt_flags);
3245 if (flags & MS_REMOUNT)
3246 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3247 if (flags & MS_BIND)
3248 return do_loopback(path, dev_name, flags & MS_REC);
3249 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3250 return do_change_type(path, flags);
3251 if (flags & MS_MOVE)
3252 return do_move_mount_old(path, dev_name);
3254 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3258 long do_mount(const char *dev_name, const char __user *dir_name,
3259 const char *type_page, unsigned long flags, void *data_page)
3264 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3267 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3272 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3274 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3277 static void dec_mnt_namespaces(struct ucounts *ucounts)
3279 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3282 static void free_mnt_ns(struct mnt_namespace *ns)
3284 if (!is_anon_ns(ns))
3285 ns_free_inum(&ns->ns);
3286 dec_mnt_namespaces(ns->ucounts);
3287 put_user_ns(ns->user_ns);
3292 * Assign a sequence number so we can detect when we attempt to bind
3293 * mount a reference to an older mount namespace into the current
3294 * mount namespace, preventing reference counting loops. A 64bit
3295 * number incrementing at 10Ghz will take 12,427 years to wrap which
3296 * is effectively never, so we can ignore the possibility.
3298 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3300 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3302 struct mnt_namespace *new_ns;
3303 struct ucounts *ucounts;
3306 ucounts = inc_mnt_namespaces(user_ns);
3308 return ERR_PTR(-ENOSPC);
3310 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3312 dec_mnt_namespaces(ucounts);
3313 return ERR_PTR(-ENOMEM);
3316 ret = ns_alloc_inum(&new_ns->ns);
3319 dec_mnt_namespaces(ucounts);
3320 return ERR_PTR(ret);
3323 new_ns->ns.ops = &mntns_operations;
3325 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3326 refcount_set(&new_ns->ns.count, 1);
3327 INIT_LIST_HEAD(&new_ns->list);
3328 init_waitqueue_head(&new_ns->poll);
3329 spin_lock_init(&new_ns->ns_lock);
3330 new_ns->user_ns = get_user_ns(user_ns);
3331 new_ns->ucounts = ucounts;
3336 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3337 struct user_namespace *user_ns, struct fs_struct *new_fs)
3339 struct mnt_namespace *new_ns;
3340 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3341 struct mount *p, *q;
3348 if (likely(!(flags & CLONE_NEWNS))) {
3355 new_ns = alloc_mnt_ns(user_ns, false);
3360 /* First pass: copy the tree topology */
3361 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3362 if (user_ns != ns->user_ns)
3363 copy_flags |= CL_SHARED_TO_SLAVE;
3364 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3367 free_mnt_ns(new_ns);
3368 return ERR_CAST(new);
3370 if (user_ns != ns->user_ns) {
3373 unlock_mount_hash();
3376 list_add_tail(&new_ns->list, &new->mnt_list);
3379 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3380 * as belonging to new namespace. We have already acquired a private
3381 * fs_struct, so tsk->fs->lock is not needed.
3389 if (&p->mnt == new_fs->root.mnt) {
3390 new_fs->root.mnt = mntget(&q->mnt);
3393 if (&p->mnt == new_fs->pwd.mnt) {
3394 new_fs->pwd.mnt = mntget(&q->mnt);
3398 p = next_mnt(p, old);
3399 q = next_mnt(q, new);
3402 while (p->mnt.mnt_root != q->mnt.mnt_root)
3403 p = next_mnt(p, old);
3415 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3417 struct mount *mnt = real_mount(m);
3418 struct mnt_namespace *ns;
3419 struct super_block *s;
3423 ns = alloc_mnt_ns(&init_user_ns, true);
3426 return ERR_CAST(ns);
3431 list_add(&mnt->mnt_list, &ns->list);
3433 err = vfs_path_lookup(m->mnt_root, m,
3434 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3439 return ERR_PTR(err);
3441 /* trade a vfsmount reference for active sb one */
3442 s = path.mnt->mnt_sb;
3443 atomic_inc(&s->s_active);
3445 /* lock the sucker */
3446 down_write(&s->s_umount);
3447 /* ... and return the root of (sub)tree on it */
3450 EXPORT_SYMBOL(mount_subtree);
3452 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3453 char __user *, type, unsigned long, flags, void __user *, data)
3460 kernel_type = copy_mount_string(type);
3461 ret = PTR_ERR(kernel_type);
3462 if (IS_ERR(kernel_type))
3465 kernel_dev = copy_mount_string(dev_name);
3466 ret = PTR_ERR(kernel_dev);
3467 if (IS_ERR(kernel_dev))
3470 options = copy_mount_options(data);
3471 ret = PTR_ERR(options);
3472 if (IS_ERR(options))
3475 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3486 #define FSMOUNT_VALID_FLAGS \
3487 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3488 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3489 MOUNT_ATTR_NOSYMFOLLOW)
3491 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3493 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3494 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3496 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3498 unsigned int mnt_flags = 0;
3500 if (attr_flags & MOUNT_ATTR_RDONLY)
3501 mnt_flags |= MNT_READONLY;
3502 if (attr_flags & MOUNT_ATTR_NOSUID)
3503 mnt_flags |= MNT_NOSUID;
3504 if (attr_flags & MOUNT_ATTR_NODEV)
3505 mnt_flags |= MNT_NODEV;
3506 if (attr_flags & MOUNT_ATTR_NOEXEC)
3507 mnt_flags |= MNT_NOEXEC;
3508 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3509 mnt_flags |= MNT_NODIRATIME;
3510 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3511 mnt_flags |= MNT_NOSYMFOLLOW;
3517 * Create a kernel mount representation for a new, prepared superblock
3518 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3520 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3521 unsigned int, attr_flags)
3523 struct mnt_namespace *ns;
3524 struct fs_context *fc;
3526 struct path newmount;
3529 unsigned int mnt_flags = 0;
3535 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3538 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3541 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3543 switch (attr_flags & MOUNT_ATTR__ATIME) {
3544 case MOUNT_ATTR_STRICTATIME:
3546 case MOUNT_ATTR_NOATIME:
3547 mnt_flags |= MNT_NOATIME;
3549 case MOUNT_ATTR_RELATIME:
3550 mnt_flags |= MNT_RELATIME;
3561 if (f.file->f_op != &fscontext_fops)
3564 fc = f.file->private_data;
3566 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3570 /* There must be a valid superblock or we can't mount it */
3576 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3577 pr_warn("VFS: Mount too revealing\n");
3582 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3586 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3589 newmount.mnt = vfs_create_mount(fc);
3590 if (IS_ERR(newmount.mnt)) {
3591 ret = PTR_ERR(newmount.mnt);
3594 newmount.dentry = dget(fc->root);
3595 newmount.mnt->mnt_flags = mnt_flags;
3597 /* We've done the mount bit - now move the file context into more or
3598 * less the same state as if we'd done an fspick(). We don't want to
3599 * do any memory allocation or anything like that at this point as we
3600 * don't want to have to handle any errors incurred.
3602 vfs_clean_context(fc);
3604 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3609 mnt = real_mount(newmount.mnt);
3613 list_add(&mnt->mnt_list, &ns->list);
3614 mntget(newmount.mnt);
3616 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3617 * it, not just simply put it.
3619 file = dentry_open(&newmount, O_PATH, fc->cred);
3621 dissolve_on_fput(newmount.mnt);
3622 ret = PTR_ERR(file);
3625 file->f_mode |= FMODE_NEED_UNMOUNT;
3627 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3629 fd_install(ret, file);
3634 path_put(&newmount);
3636 mutex_unlock(&fc->uapi_mutex);
3643 * Move a mount from one place to another. In combination with
3644 * fsopen()/fsmount() this is used to install a new mount and in combination
3645 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3648 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3650 SYSCALL_DEFINE5(move_mount,
3651 int, from_dfd, const char __user *, from_pathname,
3652 int, to_dfd, const char __user *, to_pathname,
3653 unsigned int, flags)
3655 struct path from_path, to_path;
3656 unsigned int lflags;
3662 if (flags & ~MOVE_MOUNT__MASK)
3665 /* If someone gives a pathname, they aren't permitted to move
3666 * from an fd that requires unmount as we can't get at the flag
3667 * to clear it afterwards.
3670 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3671 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3672 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3674 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3679 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3680 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3681 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3683 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3687 ret = security_move_mount(&from_path, &to_path);
3691 ret = do_move_mount(&from_path, &to_path);
3696 path_put(&from_path);
3701 * Return true if path is reachable from root
3703 * namespace_sem or mount_lock is held
3705 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3706 const struct path *root)
3708 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3709 dentry = mnt->mnt_mountpoint;
3710 mnt = mnt->mnt_parent;
3712 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3715 bool path_is_under(const struct path *path1, const struct path *path2)
3718 read_seqlock_excl(&mount_lock);
3719 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3720 read_sequnlock_excl(&mount_lock);
3723 EXPORT_SYMBOL(path_is_under);
3726 * pivot_root Semantics:
3727 * Moves the root file system of the current process to the directory put_old,
3728 * makes new_root as the new root file system of the current process, and sets
3729 * root/cwd of all processes which had them on the current root to new_root.
3732 * The new_root and put_old must be directories, and must not be on the
3733 * same file system as the current process root. The put_old must be
3734 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3735 * pointed to by put_old must yield the same directory as new_root. No other
3736 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3738 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3739 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3740 * in this situation.
3743 * - we don't move root/cwd if they are not at the root (reason: if something
3744 * cared enough to change them, it's probably wrong to force them elsewhere)
3745 * - it's okay to pick a root that isn't the root of a file system, e.g.
3746 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3747 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3750 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3751 const char __user *, put_old)
3753 struct path new, old, root;
3754 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3755 struct mountpoint *old_mp, *root_mp;
3761 error = user_path_at(AT_FDCWD, new_root,
3762 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3766 error = user_path_at(AT_FDCWD, put_old,
3767 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3771 error = security_sb_pivotroot(&old, &new);
3775 get_fs_root(current->fs, &root);
3776 old_mp = lock_mount(&old);
3777 error = PTR_ERR(old_mp);
3782 new_mnt = real_mount(new.mnt);
3783 root_mnt = real_mount(root.mnt);
3784 old_mnt = real_mount(old.mnt);
3785 ex_parent = new_mnt->mnt_parent;
3786 root_parent = root_mnt->mnt_parent;
3787 if (IS_MNT_SHARED(old_mnt) ||
3788 IS_MNT_SHARED(ex_parent) ||
3789 IS_MNT_SHARED(root_parent))
3791 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3793 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3796 if (d_unlinked(new.dentry))
3799 if (new_mnt == root_mnt || old_mnt == root_mnt)
3800 goto out4; /* loop, on the same file system */
3802 if (root.mnt->mnt_root != root.dentry)
3803 goto out4; /* not a mountpoint */
3804 if (!mnt_has_parent(root_mnt))
3805 goto out4; /* not attached */
3806 if (new.mnt->mnt_root != new.dentry)
3807 goto out4; /* not a mountpoint */
3808 if (!mnt_has_parent(new_mnt))
3809 goto out4; /* not attached */
3810 /* make sure we can reach put_old from new_root */
3811 if (!is_path_reachable(old_mnt, old.dentry, &new))
3813 /* make certain new is below the root */
3814 if (!is_path_reachable(new_mnt, new.dentry, &root))
3817 umount_mnt(new_mnt);
3818 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3819 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3820 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3821 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3823 /* mount old root on put_old */
3824 attach_mnt(root_mnt, old_mnt, old_mp);
3825 /* mount new_root on / */
3826 attach_mnt(new_mnt, root_parent, root_mp);
3827 mnt_add_count(root_parent, -1);
3828 touch_mnt_namespace(current->nsproxy->mnt_ns);
3829 /* A moved mount should not expire automatically */
3830 list_del_init(&new_mnt->mnt_expire);
3831 put_mountpoint(root_mp);
3832 unlock_mount_hash();
3833 chroot_fs_refs(&root, &new);
3836 unlock_mount(old_mp);
3838 mntput_no_expire(ex_parent);
3849 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3851 unsigned int flags = mnt->mnt.mnt_flags;
3853 /* flags to clear */
3854 flags &= ~kattr->attr_clr;
3855 /* flags to raise */
3856 flags |= kattr->attr_set;
3861 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3863 struct vfsmount *m = &mnt->mnt;
3865 if (!kattr->mnt_userns)
3869 * Once a mount has been idmapped we don't allow it to change its
3870 * mapping. It makes things simpler and callers can just create
3871 * another bind-mount they can idmap if they want to.
3873 if (mnt_user_ns(m) != &init_user_ns)
3876 /* The underlying filesystem doesn't support idmapped mounts yet. */
3877 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3880 /* Don't yet support filesystem mountable in user namespaces. */
3881 if (m->mnt_sb->s_user_ns != &init_user_ns)
3884 /* We're not controlling the superblock. */
3885 if (!capable(CAP_SYS_ADMIN))
3888 /* Mount has already been visible in the filesystem hierarchy. */
3889 if (!is_anon_ns(mnt->mnt_ns))
3895 static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3896 struct mount *mnt, int *err)
3898 struct mount *m = mnt, *last = NULL;
3900 if (!is_mounted(&m->mnt)) {
3905 if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3913 flags = recalc_flags(kattr, m);
3914 if (!can_change_locked_flags(m, flags)) {
3919 *err = can_idmap_mount(kattr, m);
3925 if ((kattr->attr_set & MNT_READONLY) &&
3926 !(m->mnt.mnt_flags & MNT_READONLY)) {
3927 *err = mnt_hold_writers(m);
3931 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3937 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3939 struct user_namespace *mnt_userns;
3941 if (!kattr->mnt_userns)
3944 mnt_userns = get_user_ns(kattr->mnt_userns);
3945 /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3946 smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3949 static void mount_setattr_commit(struct mount_kattr *kattr,
3950 struct mount *mnt, struct mount *last,
3953 struct mount *m = mnt;
3959 do_idmap_mount(kattr, m);
3960 flags = recalc_flags(kattr, m);
3961 WRITE_ONCE(m->mnt.mnt_flags, flags);
3965 * We either set MNT_READONLY above so make it visible
3966 * before ~MNT_WRITE_HOLD or we failed to recursively
3967 * apply mount options.
3969 if ((kattr->attr_set & MNT_READONLY) &&
3970 (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3971 mnt_unhold_writers(m);
3973 if (!err && kattr->propagation)
3974 change_mnt_propagation(m, kattr->propagation);
3977 * On failure, only cleanup until we found the first mount
3978 * we failed to handle.
3980 if (err && m == last)
3982 } while (kattr->recurse && (m = next_mnt(m, mnt)));
3985 touch_mnt_namespace(mnt->mnt_ns);
3988 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3990 struct mount *mnt = real_mount(path->mnt), *last = NULL;
3993 if (path->dentry != mnt->mnt.mnt_root)
3996 if (kattr->propagation) {
3998 * Only take namespace_lock() if we're actually changing
4002 if (kattr->propagation == MS_SHARED) {
4003 err = invent_group_ids(mnt, kattr->recurse);
4014 * Get the mount tree in a shape where we can change mount
4015 * properties without failure.
4017 last = mount_setattr_prepare(kattr, mnt, &err);
4018 if (last) /* Commit all changes or revert to the old state. */
4019 mount_setattr_commit(kattr, mnt, last, err);
4021 unlock_mount_hash();
4023 if (kattr->propagation) {
4026 cleanup_group_ids(mnt, NULL);
4032 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4033 struct mount_kattr *kattr, unsigned int flags)
4036 struct ns_common *ns;
4037 struct user_namespace *mnt_userns;
4040 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4044 * We currently do not support clearing an idmapped mount. If this ever
4045 * is a use-case we can revisit this but for now let's keep it simple
4048 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4051 if (attr->userns_fd > INT_MAX)
4054 file = fget(attr->userns_fd);
4058 if (!proc_ns_file(file)) {
4063 ns = get_proc_ns(file_inode(file));
4064 if (ns->ops->type != CLONE_NEWUSER) {
4070 * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4071 * This is simpler than just having to treat NULL as unmapped. Users
4072 * wanting to idmap a mount to init_user_ns can just use a namespace
4073 * with an identity mapping.
4075 mnt_userns = container_of(ns, struct user_namespace, ns);
4076 if (mnt_userns == &init_user_ns) {
4080 kattr->mnt_userns = get_user_ns(mnt_userns);
4087 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4088 struct mount_kattr *kattr, unsigned int flags)
4090 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4092 if (flags & AT_NO_AUTOMOUNT)
4093 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4094 if (flags & AT_SYMLINK_NOFOLLOW)
4095 lookup_flags &= ~LOOKUP_FOLLOW;
4096 if (flags & AT_EMPTY_PATH)
4097 lookup_flags |= LOOKUP_EMPTY;
4099 *kattr = (struct mount_kattr) {
4100 .lookup_flags = lookup_flags,
4101 .recurse = !!(flags & AT_RECURSIVE),
4104 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4106 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4108 kattr->propagation = attr->propagation;
4110 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4113 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4114 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4117 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4118 * users wanting to transition to a different atime setting cannot
4119 * simply specify the atime setting in @attr_set, but must also
4120 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4121 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4122 * @attr_clr and that @attr_set can't have any atime bits set if
4123 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4125 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4126 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4130 * Clear all previous time settings as they are mutually
4133 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4134 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4135 case MOUNT_ATTR_RELATIME:
4136 kattr->attr_set |= MNT_RELATIME;
4138 case MOUNT_ATTR_NOATIME:
4139 kattr->attr_set |= MNT_NOATIME;
4141 case MOUNT_ATTR_STRICTATIME:
4147 if (attr->attr_set & MOUNT_ATTR__ATIME)
4151 return build_mount_idmapped(attr, usize, kattr, flags);
4154 static void finish_mount_kattr(struct mount_kattr *kattr)
4156 put_user_ns(kattr->mnt_userns);
4157 kattr->mnt_userns = NULL;
4160 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4161 unsigned int, flags, struct mount_attr __user *, uattr,
4166 struct mount_attr attr;
4167 struct mount_kattr kattr;
4169 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4171 if (flags & ~(AT_EMPTY_PATH |
4173 AT_SYMLINK_NOFOLLOW |
4177 if (unlikely(usize > PAGE_SIZE))
4179 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4185 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4189 /* Don't bother walking through the mounts if this is a nop. */
4190 if (attr.attr_set == 0 &&
4191 attr.attr_clr == 0 &&
4192 attr.propagation == 0)
4195 err = build_mount_kattr(&attr, usize, &kattr, flags);
4199 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4203 err = do_mount_setattr(&target, &kattr);
4204 finish_mount_kattr(&kattr);
4209 static void __init init_mount_tree(void)
4211 struct vfsmount *mnt;
4213 struct mnt_namespace *ns;
4216 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4218 panic("Can't create rootfs");
4220 ns = alloc_mnt_ns(&init_user_ns, false);
4222 panic("Can't allocate initial namespace");
4223 m = real_mount(mnt);
4227 list_add(&m->mnt_list, &ns->list);
4228 init_task.nsproxy->mnt_ns = ns;
4232 root.dentry = mnt->mnt_root;
4233 mnt->mnt_flags |= MNT_LOCKED;
4235 set_fs_pwd(current->fs, &root);
4236 set_fs_root(current->fs, &root);
4239 void __init mnt_init(void)
4243 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4244 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4246 mount_hashtable = alloc_large_system_hash("Mount-cache",
4247 sizeof(struct hlist_head),
4250 &m_hash_shift, &m_hash_mask, 0, 0);
4251 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4252 sizeof(struct hlist_head),
4255 &mp_hash_shift, &mp_hash_mask, 0, 0);
4257 if (!mount_hashtable || !mountpoint_hashtable)
4258 panic("Failed to allocate mount hash table\n");
4264 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4266 fs_kobj = kobject_create_and_add("fs", NULL);
4268 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4274 void put_mnt_ns(struct mnt_namespace *ns)
4276 if (!refcount_dec_and_test(&ns->ns.count))
4278 drop_collected_mounts(&ns->root->mnt);
4282 struct vfsmount *kern_mount(struct file_system_type *type)
4284 struct vfsmount *mnt;
4285 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4288 * it is a longterm mount, don't release mnt until
4289 * we unmount before file sys is unregistered
4291 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4295 EXPORT_SYMBOL_GPL(kern_mount);
4297 void kern_unmount(struct vfsmount *mnt)
4299 /* release long term mount so mount point can be released */
4300 if (!IS_ERR_OR_NULL(mnt)) {
4301 real_mount(mnt)->mnt_ns = NULL;
4302 synchronize_rcu(); /* yecchhh... */
4306 EXPORT_SYMBOL(kern_unmount);
4308 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4312 for (i = 0; i < num; i++)
4314 real_mount(mnt[i])->mnt_ns = NULL;
4315 synchronize_rcu_expedited();
4316 for (i = 0; i < num; i++)
4319 EXPORT_SYMBOL(kern_unmount_array);
4321 bool our_mnt(struct vfsmount *mnt)
4323 return check_mnt(real_mount(mnt));
4326 bool current_chrooted(void)
4328 /* Does the current process have a non-standard root */
4329 struct path ns_root;
4330 struct path fs_root;
4333 /* Find the namespace root */
4334 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4335 ns_root.dentry = ns_root.mnt->mnt_root;
4337 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4340 get_fs_root(current->fs, &fs_root);
4342 chrooted = !path_equal(&fs_root, &ns_root);
4350 static bool mnt_already_visible(struct mnt_namespace *ns,
4351 const struct super_block *sb,
4354 int new_flags = *new_mnt_flags;
4356 bool visible = false;
4358 down_read(&namespace_sem);
4360 list_for_each_entry(mnt, &ns->list, mnt_list) {
4361 struct mount *child;
4364 if (mnt_is_cursor(mnt))
4367 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4370 /* This mount is not fully visible if it's root directory
4371 * is not the root directory of the filesystem.
4373 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4376 /* A local view of the mount flags */
4377 mnt_flags = mnt->mnt.mnt_flags;
4379 /* Don't miss readonly hidden in the superblock flags */
4380 if (sb_rdonly(mnt->mnt.mnt_sb))
4381 mnt_flags |= MNT_LOCK_READONLY;
4383 /* Verify the mount flags are equal to or more permissive
4384 * than the proposed new mount.
4386 if ((mnt_flags & MNT_LOCK_READONLY) &&
4387 !(new_flags & MNT_READONLY))
4389 if ((mnt_flags & MNT_LOCK_ATIME) &&
4390 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4393 /* This mount is not fully visible if there are any
4394 * locked child mounts that cover anything except for
4395 * empty directories.
4397 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4398 struct inode *inode = child->mnt_mountpoint->d_inode;
4399 /* Only worry about locked mounts */
4400 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4402 /* Is the directory permanetly empty? */
4403 if (!is_empty_dir_inode(inode))
4406 /* Preserve the locked attributes */
4407 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4415 up_read(&namespace_sem);
4419 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4421 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4422 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4423 unsigned long s_iflags;
4425 if (ns->user_ns == &init_user_ns)
4428 /* Can this filesystem be too revealing? */
4429 s_iflags = sb->s_iflags;
4430 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4433 if ((s_iflags & required_iflags) != required_iflags) {
4434 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4439 return !mnt_already_visible(ns, sb, new_mnt_flags);
4442 bool mnt_may_suid(struct vfsmount *mnt)
4445 * Foreign mounts (accessed via fchdir or through /proc
4446 * symlinks) are always treated as if they are nosuid. This
4447 * prevents namespaces from trusting potentially unsafe
4448 * suid/sgid bits, file caps, or security labels that originate
4449 * in other namespaces.
4451 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4452 current_in_userns(mnt->mnt_sb->s_user_ns);
4455 static struct ns_common *mntns_get(struct task_struct *task)
4457 struct ns_common *ns = NULL;
4458 struct nsproxy *nsproxy;
4461 nsproxy = task->nsproxy;
4463 ns = &nsproxy->mnt_ns->ns;
4464 get_mnt_ns(to_mnt_ns(ns));
4471 static void mntns_put(struct ns_common *ns)
4473 put_mnt_ns(to_mnt_ns(ns));
4476 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4478 struct nsproxy *nsproxy = nsset->nsproxy;
4479 struct fs_struct *fs = nsset->fs;
4480 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4481 struct user_namespace *user_ns = nsset->cred->user_ns;
4485 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4486 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4487 !ns_capable(user_ns, CAP_SYS_ADMIN))
4490 if (is_anon_ns(mnt_ns))
4497 old_mnt_ns = nsproxy->mnt_ns;
4498 nsproxy->mnt_ns = mnt_ns;
4501 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4502 "/", LOOKUP_DOWN, &root);
4504 /* revert to old namespace */
4505 nsproxy->mnt_ns = old_mnt_ns;
4510 put_mnt_ns(old_mnt_ns);
4512 /* Update the pwd and root */
4513 set_fs_pwd(fs, &root);
4514 set_fs_root(fs, &root);
4520 static struct user_namespace *mntns_owner(struct ns_common *ns)
4522 return to_mnt_ns(ns)->user_ns;
4525 const struct proc_ns_operations mntns_operations = {
4527 .type = CLONE_NEWNS,
4530 .install = mntns_install,
4531 .owner = mntns_owner,