4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
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/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/memblock.h>
27 #include <linux/task_work.h>
28 #include <linux/sched/task.h>
29 #include <uapi/linux/mount.h>
30 #include <linux/fs_context.h>
35 /* Maximum number of mounts in a mount namespace */
36 unsigned int sysctl_mount_max __read_mostly = 100000;
38 static unsigned int m_hash_mask __read_mostly;
39 static unsigned int m_hash_shift __read_mostly;
40 static unsigned int mp_hash_mask __read_mostly;
41 static unsigned int mp_hash_shift __read_mostly;
43 static __initdata unsigned long mhash_entries;
44 static int __init set_mhash_entries(char *str)
48 mhash_entries = simple_strtoul(str, &str, 0);
51 __setup("mhash_entries=", set_mhash_entries);
53 static __initdata unsigned long mphash_entries;
54 static int __init set_mphash_entries(char *str)
58 mphash_entries = simple_strtoul(str, &str, 0);
61 __setup("mphash_entries=", set_mphash_entries);
64 static DEFINE_IDA(mnt_id_ida);
65 static DEFINE_IDA(mnt_group_ida);
67 static struct hlist_head *mount_hashtable __read_mostly;
68 static struct hlist_head *mountpoint_hashtable __read_mostly;
69 static struct kmem_cache *mnt_cache __read_mostly;
70 static DECLARE_RWSEM(namespace_sem);
73 struct kobject *fs_kobj;
74 EXPORT_SYMBOL_GPL(fs_kobj);
77 * vfsmount lock may be taken for read to prevent changes to the
78 * vfsmount hash, ie. during mountpoint lookups or walking back
81 * It should be taken for write in all cases where the vfsmount
82 * tree or hash is modified or when a vfsmount structure is modified.
84 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
86 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
88 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
89 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
90 tmp = tmp + (tmp >> m_hash_shift);
91 return &mount_hashtable[tmp & m_hash_mask];
94 static inline struct hlist_head *mp_hash(struct dentry *dentry)
96 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
97 tmp = tmp + (tmp >> mp_hash_shift);
98 return &mountpoint_hashtable[tmp & mp_hash_mask];
101 static int mnt_alloc_id(struct mount *mnt)
103 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
111 static void mnt_free_id(struct mount *mnt)
113 ida_free(&mnt_id_ida, mnt->mnt_id);
117 * Allocate a new peer group ID
119 static int mnt_alloc_group_id(struct mount *mnt)
121 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
125 mnt->mnt_group_id = res;
130 * Release a peer group ID
132 void mnt_release_group_id(struct mount *mnt)
134 ida_free(&mnt_group_ida, mnt->mnt_group_id);
135 mnt->mnt_group_id = 0;
139 * vfsmount lock must be held for read
141 static inline void mnt_add_count(struct mount *mnt, int n)
144 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
153 * vfsmount lock must be held for write
155 unsigned int mnt_get_count(struct mount *mnt)
158 unsigned int count = 0;
161 for_each_possible_cpu(cpu) {
162 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
167 return mnt->mnt_count;
171 static void drop_mountpoint(struct fs_pin *p)
173 struct mount *m = container_of(p, struct mount, mnt_umount);
174 dput(m->mnt_ex_mountpoint);
179 static struct mount *alloc_vfsmnt(const char *name)
181 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
185 err = mnt_alloc_id(mnt);
190 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
191 if (!mnt->mnt_devname)
196 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
198 goto out_free_devname;
200 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
203 mnt->mnt_writers = 0;
206 INIT_HLIST_NODE(&mnt->mnt_hash);
207 INIT_LIST_HEAD(&mnt->mnt_child);
208 INIT_LIST_HEAD(&mnt->mnt_mounts);
209 INIT_LIST_HEAD(&mnt->mnt_list);
210 INIT_LIST_HEAD(&mnt->mnt_expire);
211 INIT_LIST_HEAD(&mnt->mnt_share);
212 INIT_LIST_HEAD(&mnt->mnt_slave_list);
213 INIT_LIST_HEAD(&mnt->mnt_slave);
214 INIT_HLIST_NODE(&mnt->mnt_mp_list);
215 INIT_LIST_HEAD(&mnt->mnt_umounting);
216 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
222 kfree_const(mnt->mnt_devname);
227 kmem_cache_free(mnt_cache, mnt);
232 * Most r/o checks on a fs are for operations that take
233 * discrete amounts of time, like a write() or unlink().
234 * We must keep track of when those operations start
235 * (for permission checks) and when they end, so that
236 * we can determine when writes are able to occur to
240 * __mnt_is_readonly: check whether a mount is read-only
241 * @mnt: the mount to check for its write status
243 * This shouldn't be used directly ouside of the VFS.
244 * It does not guarantee that the filesystem will stay
245 * r/w, just that it is right *now*. This can not and
246 * should not be used in place of IS_RDONLY(inode).
247 * mnt_want/drop_write() will _keep_ the filesystem
250 bool __mnt_is_readonly(struct vfsmount *mnt)
252 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
254 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
256 static inline void mnt_inc_writers(struct mount *mnt)
259 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
265 static inline void mnt_dec_writers(struct mount *mnt)
268 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
274 static unsigned int mnt_get_writers(struct mount *mnt)
277 unsigned int count = 0;
280 for_each_possible_cpu(cpu) {
281 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
286 return mnt->mnt_writers;
290 static int mnt_is_readonly(struct vfsmount *mnt)
292 if (mnt->mnt_sb->s_readonly_remount)
294 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
296 return __mnt_is_readonly(mnt);
300 * Most r/o & frozen checks on a fs are for operations that take discrete
301 * amounts of time, like a write() or unlink(). We must keep track of when
302 * those operations start (for permission checks) and when they end, so that we
303 * can determine when writes are able to occur to a filesystem.
306 * __mnt_want_write - get write access to a mount without freeze protection
307 * @m: the mount on which to take a write
309 * This tells the low-level filesystem that a write is about to be performed to
310 * it, and makes sure that writes are allowed (mnt it read-write) before
311 * returning success. This operation does not protect against filesystem being
312 * frozen. When the write operation is finished, __mnt_drop_write() must be
313 * called. This is effectively a refcount.
315 int __mnt_want_write(struct vfsmount *m)
317 struct mount *mnt = real_mount(m);
321 mnt_inc_writers(mnt);
323 * The store to mnt_inc_writers must be visible before we pass
324 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
325 * incremented count after it has set MNT_WRITE_HOLD.
328 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
331 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
332 * be set to match its requirements. So we must not load that until
333 * MNT_WRITE_HOLD is cleared.
336 if (mnt_is_readonly(m)) {
337 mnt_dec_writers(mnt);
346 * mnt_want_write - get write access to a mount
347 * @m: the mount on which to take a write
349 * This tells the low-level filesystem that a write is about to be performed to
350 * it, and makes sure that writes are allowed (mount is read-write, filesystem
351 * is not frozen) before returning success. When the write operation is
352 * finished, mnt_drop_write() must be called. This is effectively a refcount.
354 int mnt_want_write(struct vfsmount *m)
358 sb_start_write(m->mnt_sb);
359 ret = __mnt_want_write(m);
361 sb_end_write(m->mnt_sb);
364 EXPORT_SYMBOL_GPL(mnt_want_write);
367 * mnt_clone_write - get write access to a mount
368 * @mnt: the mount on which to take a write
370 * This is effectively like mnt_want_write, except
371 * it must only be used to take an extra write reference
372 * on a mountpoint that we already know has a write reference
373 * on it. This allows some optimisation.
375 * After finished, mnt_drop_write must be called as usual to
376 * drop the reference.
378 int mnt_clone_write(struct vfsmount *mnt)
380 /* superblock may be r/o */
381 if (__mnt_is_readonly(mnt))
384 mnt_inc_writers(real_mount(mnt));
388 EXPORT_SYMBOL_GPL(mnt_clone_write);
391 * __mnt_want_write_file - get write access to a file's mount
392 * @file: the file who's mount on which to take a write
394 * This is like __mnt_want_write, but it takes a file and can
395 * do some optimisations if the file is open for write already
397 int __mnt_want_write_file(struct file *file)
399 if (!(file->f_mode & FMODE_WRITER))
400 return __mnt_want_write(file->f_path.mnt);
402 return mnt_clone_write(file->f_path.mnt);
406 * mnt_want_write_file - get write access to a file's mount
407 * @file: the file who's mount on which to take a write
409 * This is like mnt_want_write, but it takes a file and can
410 * do some optimisations if the file is open for write already
412 int mnt_want_write_file(struct file *file)
416 sb_start_write(file_inode(file)->i_sb);
417 ret = __mnt_want_write_file(file);
419 sb_end_write(file_inode(file)->i_sb);
422 EXPORT_SYMBOL_GPL(mnt_want_write_file);
425 * __mnt_drop_write - give up write access to a mount
426 * @mnt: the mount on which to give up write access
428 * Tells the low-level filesystem that we are done
429 * performing writes to it. Must be matched with
430 * __mnt_want_write() call above.
432 void __mnt_drop_write(struct vfsmount *mnt)
435 mnt_dec_writers(real_mount(mnt));
440 * mnt_drop_write - give up write access to a mount
441 * @mnt: the mount on which to give up write access
443 * Tells the low-level filesystem that we are done performing writes to it and
444 * also allows filesystem to be frozen again. Must be matched with
445 * mnt_want_write() call above.
447 void mnt_drop_write(struct vfsmount *mnt)
449 __mnt_drop_write(mnt);
450 sb_end_write(mnt->mnt_sb);
452 EXPORT_SYMBOL_GPL(mnt_drop_write);
454 void __mnt_drop_write_file(struct file *file)
456 __mnt_drop_write(file->f_path.mnt);
459 void mnt_drop_write_file(struct file *file)
461 __mnt_drop_write_file(file);
462 sb_end_write(file_inode(file)->i_sb);
464 EXPORT_SYMBOL(mnt_drop_write_file);
466 static int mnt_make_readonly(struct mount *mnt)
471 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
473 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
474 * should be visible before we do.
479 * With writers on hold, if this value is zero, then there are
480 * definitely no active writers (although held writers may subsequently
481 * increment the count, they'll have to wait, and decrement it after
482 * seeing MNT_READONLY).
484 * It is OK to have counter incremented on one CPU and decremented on
485 * another: the sum will add up correctly. The danger would be when we
486 * sum up each counter, if we read a counter before it is incremented,
487 * but then read another CPU's count which it has been subsequently
488 * decremented from -- we would see more decrements than we should.
489 * MNT_WRITE_HOLD protects against this scenario, because
490 * mnt_want_write first increments count, then smp_mb, then spins on
491 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
492 * we're counting up here.
494 if (mnt_get_writers(mnt) > 0)
497 mnt->mnt.mnt_flags |= MNT_READONLY;
499 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
500 * that become unheld will see MNT_READONLY.
503 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
508 static int __mnt_unmake_readonly(struct mount *mnt)
511 mnt->mnt.mnt_flags &= ~MNT_READONLY;
516 int sb_prepare_remount_readonly(struct super_block *sb)
521 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
522 if (atomic_long_read(&sb->s_remove_count))
526 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
527 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
528 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
530 if (mnt_get_writers(mnt) > 0) {
536 if (!err && atomic_long_read(&sb->s_remove_count))
540 sb->s_readonly_remount = 1;
543 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
544 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
545 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
552 static void free_vfsmnt(struct mount *mnt)
554 kfree_const(mnt->mnt_devname);
556 free_percpu(mnt->mnt_pcp);
558 kmem_cache_free(mnt_cache, mnt);
561 static void delayed_free_vfsmnt(struct rcu_head *head)
563 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
566 /* call under rcu_read_lock */
567 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
570 if (read_seqretry(&mount_lock, seq))
574 mnt = real_mount(bastard);
575 mnt_add_count(mnt, 1);
576 smp_mb(); // see mntput_no_expire()
577 if (likely(!read_seqretry(&mount_lock, seq)))
579 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
580 mnt_add_count(mnt, -1);
584 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
585 mnt_add_count(mnt, -1);
590 /* caller will mntput() */
594 /* call under rcu_read_lock */
595 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
597 int res = __legitimize_mnt(bastard, seq);
600 if (unlikely(res < 0)) {
609 * find the first mount at @dentry on vfsmount @mnt.
610 * call under rcu_read_lock()
612 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
614 struct hlist_head *head = m_hash(mnt, dentry);
617 hlist_for_each_entry_rcu(p, head, mnt_hash)
618 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
624 * lookup_mnt - Return the first child mount mounted at path
626 * "First" means first mounted chronologically. If you create the
629 * mount /dev/sda1 /mnt
630 * mount /dev/sda2 /mnt
631 * mount /dev/sda3 /mnt
633 * Then lookup_mnt() on the base /mnt dentry in the root mount will
634 * return successively the root dentry and vfsmount of /dev/sda1, then
635 * /dev/sda2, then /dev/sda3, then NULL.
637 * lookup_mnt takes a reference to the found vfsmount.
639 struct vfsmount *lookup_mnt(const struct path *path)
641 struct mount *child_mnt;
647 seq = read_seqbegin(&mount_lock);
648 child_mnt = __lookup_mnt(path->mnt, path->dentry);
649 m = child_mnt ? &child_mnt->mnt : NULL;
650 } while (!legitimize_mnt(m, seq));
656 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
657 * current mount namespace.
659 * The common case is dentries are not mountpoints at all and that
660 * test is handled inline. For the slow case when we are actually
661 * dealing with a mountpoint of some kind, walk through all of the
662 * mounts in the current mount namespace and test to see if the dentry
665 * The mount_hashtable is not usable in the context because we
666 * need to identify all mounts that may be in the current mount
667 * namespace not just a mount that happens to have some specified
670 bool __is_local_mountpoint(struct dentry *dentry)
672 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
674 bool is_covered = false;
676 if (!d_mountpoint(dentry))
679 down_read(&namespace_sem);
680 list_for_each_entry(mnt, &ns->list, mnt_list) {
681 is_covered = (mnt->mnt_mountpoint == dentry);
685 up_read(&namespace_sem);
690 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
692 struct hlist_head *chain = mp_hash(dentry);
693 struct mountpoint *mp;
695 hlist_for_each_entry(mp, chain, m_hash) {
696 if (mp->m_dentry == dentry) {
704 static struct mountpoint *get_mountpoint(struct dentry *dentry)
706 struct mountpoint *mp, *new = NULL;
709 if (d_mountpoint(dentry)) {
710 /* might be worth a WARN_ON() */
711 if (d_unlinked(dentry))
712 return ERR_PTR(-ENOENT);
714 read_seqlock_excl(&mount_lock);
715 mp = lookup_mountpoint(dentry);
716 read_sequnlock_excl(&mount_lock);
722 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
724 return ERR_PTR(-ENOMEM);
727 /* Exactly one processes may set d_mounted */
728 ret = d_set_mounted(dentry);
730 /* Someone else set d_mounted? */
734 /* The dentry is not available as a mountpoint? */
739 /* Add the new mountpoint to the hash table */
740 read_seqlock_excl(&mount_lock);
741 new->m_dentry = dentry;
743 hlist_add_head(&new->m_hash, mp_hash(dentry));
744 INIT_HLIST_HEAD(&new->m_list);
745 read_sequnlock_excl(&mount_lock);
754 static void put_mountpoint(struct mountpoint *mp)
756 if (!--mp->m_count) {
757 struct dentry *dentry = mp->m_dentry;
758 BUG_ON(!hlist_empty(&mp->m_list));
759 spin_lock(&dentry->d_lock);
760 dentry->d_flags &= ~DCACHE_MOUNTED;
761 spin_unlock(&dentry->d_lock);
762 hlist_del(&mp->m_hash);
767 static inline int check_mnt(struct mount *mnt)
769 return mnt->mnt_ns == current->nsproxy->mnt_ns;
773 * vfsmount lock must be held for write
775 static void touch_mnt_namespace(struct mnt_namespace *ns)
779 wake_up_interruptible(&ns->poll);
784 * vfsmount lock must be held for write
786 static void __touch_mnt_namespace(struct mnt_namespace *ns)
788 if (ns && ns->event != event) {
790 wake_up_interruptible(&ns->poll);
795 * vfsmount lock must be held for write
797 static void unhash_mnt(struct mount *mnt)
799 mnt->mnt_parent = mnt;
800 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
801 list_del_init(&mnt->mnt_child);
802 hlist_del_init_rcu(&mnt->mnt_hash);
803 hlist_del_init(&mnt->mnt_mp_list);
804 put_mountpoint(mnt->mnt_mp);
809 * vfsmount lock must be held for write
811 static void detach_mnt(struct mount *mnt, struct path *old_path)
813 old_path->dentry = mnt->mnt_mountpoint;
814 old_path->mnt = &mnt->mnt_parent->mnt;
819 * vfsmount lock must be held for write
821 static void umount_mnt(struct mount *mnt)
823 /* old mountpoint will be dropped when we can do that */
824 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
829 * vfsmount lock must be held for write
831 void mnt_set_mountpoint(struct mount *mnt,
832 struct mountpoint *mp,
833 struct mount *child_mnt)
836 mnt_add_count(mnt, 1); /* essentially, that's mntget */
837 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
838 child_mnt->mnt_parent = mnt;
839 child_mnt->mnt_mp = mp;
840 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
843 static void __attach_mnt(struct mount *mnt, struct mount *parent)
845 hlist_add_head_rcu(&mnt->mnt_hash,
846 m_hash(&parent->mnt, mnt->mnt_mountpoint));
847 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
851 * vfsmount lock must be held for write
853 static void attach_mnt(struct mount *mnt,
854 struct mount *parent,
855 struct mountpoint *mp)
857 mnt_set_mountpoint(parent, mp, mnt);
858 __attach_mnt(mnt, parent);
861 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
863 struct mountpoint *old_mp = mnt->mnt_mp;
864 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
865 struct mount *old_parent = mnt->mnt_parent;
867 list_del_init(&mnt->mnt_child);
868 hlist_del_init(&mnt->mnt_mp_list);
869 hlist_del_init_rcu(&mnt->mnt_hash);
871 attach_mnt(mnt, parent, mp);
873 put_mountpoint(old_mp);
876 * Safely avoid even the suggestion this code might sleep or
877 * lock the mount hash by taking advantage of the knowledge that
878 * mnt_change_mountpoint will not release the final reference
881 * During mounting, the mount passed in as the parent mount will
882 * continue to use the old mountpoint and during unmounting, the
883 * old mountpoint will continue to exist until namespace_unlock,
884 * which happens well after mnt_change_mountpoint.
886 spin_lock(&old_mountpoint->d_lock);
887 old_mountpoint->d_lockref.count--;
888 spin_unlock(&old_mountpoint->d_lock);
890 mnt_add_count(old_parent, -1);
894 * vfsmount lock must be held for write
896 static void commit_tree(struct mount *mnt)
898 struct mount *parent = mnt->mnt_parent;
901 struct mnt_namespace *n = parent->mnt_ns;
903 BUG_ON(parent == mnt);
905 list_add_tail(&head, &mnt->mnt_list);
906 list_for_each_entry(m, &head, mnt_list)
909 list_splice(&head, n->list.prev);
911 n->mounts += n->pending_mounts;
912 n->pending_mounts = 0;
914 __attach_mnt(mnt, parent);
915 touch_mnt_namespace(n);
918 static struct mount *next_mnt(struct mount *p, struct mount *root)
920 struct list_head *next = p->mnt_mounts.next;
921 if (next == &p->mnt_mounts) {
925 next = p->mnt_child.next;
926 if (next != &p->mnt_parent->mnt_mounts)
931 return list_entry(next, struct mount, mnt_child);
934 static struct mount *skip_mnt_tree(struct mount *p)
936 struct list_head *prev = p->mnt_mounts.prev;
937 while (prev != &p->mnt_mounts) {
938 p = list_entry(prev, struct mount, mnt_child);
939 prev = p->mnt_mounts.prev;
945 * vfs_create_mount - Create a mount for a configured superblock
946 * @fc: The configuration context with the superblock attached
948 * Create a mount to an already configured superblock. If necessary, the
949 * caller should invoke vfs_get_tree() before calling this.
951 * Note that this does not attach the mount to anything.
953 struct vfsmount *vfs_create_mount(struct fs_context *fc)
958 return ERR_PTR(-EINVAL);
960 mnt = alloc_vfsmnt(fc->source ?: "none");
962 return ERR_PTR(-ENOMEM);
964 if (fc->sb_flags & SB_KERNMOUNT)
965 mnt->mnt.mnt_flags = MNT_INTERNAL;
967 atomic_inc(&fc->root->d_sb->s_active);
968 mnt->mnt.mnt_sb = fc->root->d_sb;
969 mnt->mnt.mnt_root = dget(fc->root);
970 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
971 mnt->mnt_parent = mnt;
974 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
978 EXPORT_SYMBOL(vfs_create_mount);
980 struct vfsmount *fc_mount(struct fs_context *fc)
982 int err = vfs_get_tree(fc);
984 up_write(&fc->root->d_sb->s_umount);
985 return vfs_create_mount(fc);
989 EXPORT_SYMBOL(fc_mount);
991 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
992 int flags, const char *name,
995 struct fs_context *fc;
996 struct vfsmount *mnt;
1000 return ERR_PTR(-EINVAL);
1002 fc = fs_context_for_mount(type, flags);
1004 return ERR_CAST(fc);
1007 ret = vfs_parse_fs_string(fc, "source",
1008 name, strlen(name));
1010 ret = parse_monolithic_mount_data(fc, data);
1019 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1022 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1023 const char *name, void *data)
1025 /* Until it is worked out how to pass the user namespace
1026 * through from the parent mount to the submount don't support
1027 * unprivileged mounts with submounts.
1029 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1030 return ERR_PTR(-EPERM);
1032 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1034 EXPORT_SYMBOL_GPL(vfs_submount);
1036 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1039 struct super_block *sb = old->mnt.mnt_sb;
1043 mnt = alloc_vfsmnt(old->mnt_devname);
1045 return ERR_PTR(-ENOMEM);
1047 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1048 mnt->mnt_group_id = 0; /* not a peer of original */
1050 mnt->mnt_group_id = old->mnt_group_id;
1052 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1053 err = mnt_alloc_group_id(mnt);
1058 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1059 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1061 atomic_inc(&sb->s_active);
1062 mnt->mnt.mnt_sb = sb;
1063 mnt->mnt.mnt_root = dget(root);
1064 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1065 mnt->mnt_parent = mnt;
1067 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1068 unlock_mount_hash();
1070 if ((flag & CL_SLAVE) ||
1071 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1072 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1073 mnt->mnt_master = old;
1074 CLEAR_MNT_SHARED(mnt);
1075 } else if (!(flag & CL_PRIVATE)) {
1076 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1077 list_add(&mnt->mnt_share, &old->mnt_share);
1078 if (IS_MNT_SLAVE(old))
1079 list_add(&mnt->mnt_slave, &old->mnt_slave);
1080 mnt->mnt_master = old->mnt_master;
1082 CLEAR_MNT_SHARED(mnt);
1084 if (flag & CL_MAKE_SHARED)
1085 set_mnt_shared(mnt);
1087 /* stick the duplicate mount on the same expiry list
1088 * as the original if that was on one */
1089 if (flag & CL_EXPIRE) {
1090 if (!list_empty(&old->mnt_expire))
1091 list_add(&mnt->mnt_expire, &old->mnt_expire);
1099 return ERR_PTR(err);
1102 static void cleanup_mnt(struct mount *mnt)
1105 * This probably indicates that somebody messed
1106 * up a mnt_want/drop_write() pair. If this
1107 * happens, the filesystem was probably unable
1108 * to make r/w->r/o transitions.
1111 * The locking used to deal with mnt_count decrement provides barriers,
1112 * so mnt_get_writers() below is safe.
1114 WARN_ON(mnt_get_writers(mnt));
1115 if (unlikely(mnt->mnt_pins.first))
1117 fsnotify_vfsmount_delete(&mnt->mnt);
1118 dput(mnt->mnt.mnt_root);
1119 deactivate_super(mnt->mnt.mnt_sb);
1121 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1124 static void __cleanup_mnt(struct rcu_head *head)
1126 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1129 static LLIST_HEAD(delayed_mntput_list);
1130 static void delayed_mntput(struct work_struct *unused)
1132 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1133 struct mount *m, *t;
1135 llist_for_each_entry_safe(m, t, node, mnt_llist)
1138 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1140 static void mntput_no_expire(struct mount *mnt)
1143 if (likely(READ_ONCE(mnt->mnt_ns))) {
1145 * Since we don't do lock_mount_hash() here,
1146 * ->mnt_ns can change under us. However, if it's
1147 * non-NULL, then there's a reference that won't
1148 * be dropped until after an RCU delay done after
1149 * turning ->mnt_ns NULL. So if we observe it
1150 * non-NULL under rcu_read_lock(), the reference
1151 * we are dropping is not the final one.
1153 mnt_add_count(mnt, -1);
1159 * make sure that if __legitimize_mnt() has not seen us grab
1160 * mount_lock, we'll see their refcount increment here.
1163 mnt_add_count(mnt, -1);
1164 if (mnt_get_count(mnt)) {
1166 unlock_mount_hash();
1169 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1171 unlock_mount_hash();
1174 mnt->mnt.mnt_flags |= MNT_DOOMED;
1177 list_del(&mnt->mnt_instance);
1179 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1180 struct mount *p, *tmp;
1181 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1185 unlock_mount_hash();
1187 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1188 struct task_struct *task = current;
1189 if (likely(!(task->flags & PF_KTHREAD))) {
1190 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1191 if (!task_work_add(task, &mnt->mnt_rcu, true))
1194 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1195 schedule_delayed_work(&delayed_mntput_work, 1);
1201 void mntput(struct vfsmount *mnt)
1204 struct mount *m = real_mount(mnt);
1205 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1206 if (unlikely(m->mnt_expiry_mark))
1207 m->mnt_expiry_mark = 0;
1208 mntput_no_expire(m);
1211 EXPORT_SYMBOL(mntput);
1213 struct vfsmount *mntget(struct vfsmount *mnt)
1216 mnt_add_count(real_mount(mnt), 1);
1219 EXPORT_SYMBOL(mntget);
1221 /* path_is_mountpoint() - Check if path is a mount in the current
1224 * d_mountpoint() can only be used reliably to establish if a dentry is
1225 * not mounted in any namespace and that common case is handled inline.
1226 * d_mountpoint() isn't aware of the possibility there may be multiple
1227 * mounts using a given dentry in a different namespace. This function
1228 * checks if the passed in path is a mountpoint rather than the dentry
1231 bool path_is_mountpoint(const struct path *path)
1236 if (!d_mountpoint(path->dentry))
1241 seq = read_seqbegin(&mount_lock);
1242 res = __path_is_mountpoint(path);
1243 } while (read_seqretry(&mount_lock, seq));
1248 EXPORT_SYMBOL(path_is_mountpoint);
1250 struct vfsmount *mnt_clone_internal(const struct path *path)
1253 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1256 p->mnt.mnt_flags |= MNT_INTERNAL;
1260 #ifdef CONFIG_PROC_FS
1261 /* iterator; we want it to have access to namespace_sem, thus here... */
1262 static void *m_start(struct seq_file *m, loff_t *pos)
1264 struct proc_mounts *p = m->private;
1266 down_read(&namespace_sem);
1267 if (p->cached_event == p->ns->event) {
1268 void *v = p->cached_mount;
1269 if (*pos == p->cached_index)
1271 if (*pos == p->cached_index + 1) {
1272 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1273 return p->cached_mount = v;
1277 p->cached_event = p->ns->event;
1278 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1279 p->cached_index = *pos;
1280 return p->cached_mount;
1283 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1285 struct proc_mounts *p = m->private;
1287 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1288 p->cached_index = *pos;
1289 return p->cached_mount;
1292 static void m_stop(struct seq_file *m, void *v)
1294 up_read(&namespace_sem);
1297 static int m_show(struct seq_file *m, void *v)
1299 struct proc_mounts *p = m->private;
1300 struct mount *r = list_entry(v, struct mount, mnt_list);
1301 return p->show(m, &r->mnt);
1304 const struct seq_operations mounts_op = {
1310 #endif /* CONFIG_PROC_FS */
1313 * may_umount_tree - check if a mount tree is busy
1314 * @mnt: root of mount tree
1316 * This is called to check if a tree of mounts has any
1317 * open files, pwds, chroots or sub mounts that are
1320 int may_umount_tree(struct vfsmount *m)
1322 struct mount *mnt = real_mount(m);
1323 int actual_refs = 0;
1324 int minimum_refs = 0;
1328 /* write lock needed for mnt_get_count */
1330 for (p = mnt; p; p = next_mnt(p, mnt)) {
1331 actual_refs += mnt_get_count(p);
1334 unlock_mount_hash();
1336 if (actual_refs > minimum_refs)
1342 EXPORT_SYMBOL(may_umount_tree);
1345 * may_umount - check if a mount point is busy
1346 * @mnt: root of mount
1348 * This is called to check if a mount point has any
1349 * open files, pwds, chroots or sub mounts. If the
1350 * mount has sub mounts this will return busy
1351 * regardless of whether the sub mounts are busy.
1353 * Doesn't take quota and stuff into account. IOW, in some cases it will
1354 * give false negatives. The main reason why it's here is that we need
1355 * a non-destructive way to look for easily umountable filesystems.
1357 int may_umount(struct vfsmount *mnt)
1360 down_read(&namespace_sem);
1362 if (propagate_mount_busy(real_mount(mnt), 2))
1364 unlock_mount_hash();
1365 up_read(&namespace_sem);
1369 EXPORT_SYMBOL(may_umount);
1371 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1373 static void namespace_unlock(void)
1375 struct hlist_head head;
1377 hlist_move_list(&unmounted, &head);
1379 up_write(&namespace_sem);
1381 if (likely(hlist_empty(&head)))
1384 synchronize_rcu_expedited();
1386 group_pin_kill(&head);
1389 static inline void namespace_lock(void)
1391 down_write(&namespace_sem);
1394 enum umount_tree_flags {
1396 UMOUNT_PROPAGATE = 2,
1397 UMOUNT_CONNECTED = 4,
1400 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1402 /* Leaving mounts connected is only valid for lazy umounts */
1403 if (how & UMOUNT_SYNC)
1406 /* A mount without a parent has nothing to be connected to */
1407 if (!mnt_has_parent(mnt))
1410 /* Because the reference counting rules change when mounts are
1411 * unmounted and connected, umounted mounts may not be
1412 * connected to mounted mounts.
1414 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1417 /* Has it been requested that the mount remain connected? */
1418 if (how & UMOUNT_CONNECTED)
1421 /* Is the mount locked such that it needs to remain connected? */
1422 if (IS_MNT_LOCKED(mnt))
1425 /* By default disconnect the mount */
1430 * mount_lock must be held
1431 * namespace_sem must be held for write
1433 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1435 LIST_HEAD(tmp_list);
1438 if (how & UMOUNT_PROPAGATE)
1439 propagate_mount_unlock(mnt);
1441 /* Gather the mounts to umount */
1442 for (p = mnt; p; p = next_mnt(p, mnt)) {
1443 p->mnt.mnt_flags |= MNT_UMOUNT;
1444 list_move(&p->mnt_list, &tmp_list);
1447 /* Hide the mounts from mnt_mounts */
1448 list_for_each_entry(p, &tmp_list, mnt_list) {
1449 list_del_init(&p->mnt_child);
1452 /* Add propogated mounts to the tmp_list */
1453 if (how & UMOUNT_PROPAGATE)
1454 propagate_umount(&tmp_list);
1456 while (!list_empty(&tmp_list)) {
1457 struct mnt_namespace *ns;
1459 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1460 list_del_init(&p->mnt_expire);
1461 list_del_init(&p->mnt_list);
1465 __touch_mnt_namespace(ns);
1468 if (how & UMOUNT_SYNC)
1469 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1471 disconnect = disconnect_mount(p, how);
1473 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1474 disconnect ? &unmounted : NULL);
1475 if (mnt_has_parent(p)) {
1476 mnt_add_count(p->mnt_parent, -1);
1478 /* Don't forget about p */
1479 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1484 change_mnt_propagation(p, MS_PRIVATE);
1488 static void shrink_submounts(struct mount *mnt);
1490 static int do_umount_root(struct super_block *sb)
1494 down_write(&sb->s_umount);
1495 if (!sb_rdonly(sb)) {
1496 struct fs_context *fc;
1498 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1503 ret = parse_monolithic_mount_data(fc, NULL);
1505 ret = reconfigure_super(fc);
1509 up_write(&sb->s_umount);
1513 static int do_umount(struct mount *mnt, int flags)
1515 struct super_block *sb = mnt->mnt.mnt_sb;
1518 retval = security_sb_umount(&mnt->mnt, flags);
1523 * Allow userspace to request a mountpoint be expired rather than
1524 * unmounting unconditionally. Unmount only happens if:
1525 * (1) the mark is already set (the mark is cleared by mntput())
1526 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1528 if (flags & MNT_EXPIRE) {
1529 if (&mnt->mnt == current->fs->root.mnt ||
1530 flags & (MNT_FORCE | MNT_DETACH))
1534 * probably don't strictly need the lock here if we examined
1535 * all race cases, but it's a slowpath.
1538 if (mnt_get_count(mnt) != 2) {
1539 unlock_mount_hash();
1542 unlock_mount_hash();
1544 if (!xchg(&mnt->mnt_expiry_mark, 1))
1549 * If we may have to abort operations to get out of this
1550 * mount, and they will themselves hold resources we must
1551 * allow the fs to do things. In the Unix tradition of
1552 * 'Gee thats tricky lets do it in userspace' the umount_begin
1553 * might fail to complete on the first run through as other tasks
1554 * must return, and the like. Thats for the mount program to worry
1555 * about for the moment.
1558 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1559 sb->s_op->umount_begin(sb);
1563 * No sense to grab the lock for this test, but test itself looks
1564 * somewhat bogus. Suggestions for better replacement?
1565 * Ho-hum... In principle, we might treat that as umount + switch
1566 * to rootfs. GC would eventually take care of the old vfsmount.
1567 * Actually it makes sense, especially if rootfs would contain a
1568 * /reboot - static binary that would close all descriptors and
1569 * call reboot(9). Then init(8) could umount root and exec /reboot.
1571 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1573 * Special case for "unmounting" root ...
1574 * we just try to remount it readonly.
1576 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1578 return do_umount_root(sb);
1584 /* Recheck MNT_LOCKED with the locks held */
1586 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1590 if (flags & MNT_DETACH) {
1591 if (!list_empty(&mnt->mnt_list))
1592 umount_tree(mnt, UMOUNT_PROPAGATE);
1595 shrink_submounts(mnt);
1597 if (!propagate_mount_busy(mnt, 2)) {
1598 if (!list_empty(&mnt->mnt_list))
1599 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1604 unlock_mount_hash();
1610 * __detach_mounts - lazily unmount all mounts on the specified dentry
1612 * During unlink, rmdir, and d_drop it is possible to loose the path
1613 * to an existing mountpoint, and wind up leaking the mount.
1614 * detach_mounts allows lazily unmounting those mounts instead of
1617 * The caller may hold dentry->d_inode->i_mutex.
1619 void __detach_mounts(struct dentry *dentry)
1621 struct mountpoint *mp;
1626 mp = lookup_mountpoint(dentry);
1627 if (IS_ERR_OR_NULL(mp))
1631 while (!hlist_empty(&mp->m_list)) {
1632 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1633 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1634 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1637 else umount_tree(mnt, UMOUNT_CONNECTED);
1641 unlock_mount_hash();
1646 * Is the caller allowed to modify his namespace?
1648 static inline bool may_mount(void)
1650 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1653 static inline bool may_mandlock(void)
1655 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1658 return capable(CAP_SYS_ADMIN);
1662 * Now umount can handle mount points as well as block devices.
1663 * This is important for filesystems which use unnamed block devices.
1665 * We now support a flag for forced unmount like the other 'big iron'
1666 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1669 int ksys_umount(char __user *name, int flags)
1674 int lookup_flags = 0;
1676 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1682 if (!(flags & UMOUNT_NOFOLLOW))
1683 lookup_flags |= LOOKUP_FOLLOW;
1685 lookup_flags |= LOOKUP_NO_EVAL;
1687 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1690 mnt = real_mount(path.mnt);
1692 if (path.dentry != path.mnt->mnt_root)
1694 if (!check_mnt(mnt))
1696 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1699 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1702 retval = do_umount(mnt, flags);
1704 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1706 mntput_no_expire(mnt);
1711 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1713 return ksys_umount(name, flags);
1716 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1719 * The 2.0 compatible umount. No flags.
1721 SYSCALL_DEFINE1(oldumount, char __user *, name)
1723 return ksys_umount(name, 0);
1728 static bool is_mnt_ns_file(struct dentry *dentry)
1730 /* Is this a proxy for a mount namespace? */
1731 return dentry->d_op == &ns_dentry_operations &&
1732 dentry->d_fsdata == &mntns_operations;
1735 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1737 return container_of(ns, struct mnt_namespace, ns);
1740 static bool mnt_ns_loop(struct dentry *dentry)
1742 /* Could bind mounting the mount namespace inode cause a
1743 * mount namespace loop?
1745 struct mnt_namespace *mnt_ns;
1746 if (!is_mnt_ns_file(dentry))
1749 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1750 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1753 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1756 struct mount *res, *p, *q, *r, *parent;
1758 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1759 return ERR_PTR(-EINVAL);
1761 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1762 return ERR_PTR(-EINVAL);
1764 res = q = clone_mnt(mnt, dentry, flag);
1768 q->mnt_mountpoint = mnt->mnt_mountpoint;
1771 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1773 if (!is_subdir(r->mnt_mountpoint, dentry))
1776 for (s = r; s; s = next_mnt(s, r)) {
1777 if (!(flag & CL_COPY_UNBINDABLE) &&
1778 IS_MNT_UNBINDABLE(s)) {
1779 if (s->mnt.mnt_flags & MNT_LOCKED) {
1780 /* Both unbindable and locked. */
1781 q = ERR_PTR(-EPERM);
1784 s = skip_mnt_tree(s);
1788 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1789 is_mnt_ns_file(s->mnt.mnt_root)) {
1790 s = skip_mnt_tree(s);
1793 while (p != s->mnt_parent) {
1799 q = clone_mnt(p, p->mnt.mnt_root, flag);
1803 list_add_tail(&q->mnt_list, &res->mnt_list);
1804 attach_mnt(q, parent, p->mnt_mp);
1805 unlock_mount_hash();
1812 umount_tree(res, UMOUNT_SYNC);
1813 unlock_mount_hash();
1818 /* Caller should check returned pointer for errors */
1820 struct vfsmount *collect_mounts(const struct path *path)
1824 if (!check_mnt(real_mount(path->mnt)))
1825 tree = ERR_PTR(-EINVAL);
1827 tree = copy_tree(real_mount(path->mnt), path->dentry,
1828 CL_COPY_ALL | CL_PRIVATE);
1831 return ERR_CAST(tree);
1835 void drop_collected_mounts(struct vfsmount *mnt)
1839 umount_tree(real_mount(mnt), 0);
1840 unlock_mount_hash();
1845 * clone_private_mount - create a private clone of a path
1847 * This creates a new vfsmount, which will be the clone of @path. The new will
1848 * not be attached anywhere in the namespace and will be private (i.e. changes
1849 * to the originating mount won't be propagated into this).
1851 * Release with mntput().
1853 struct vfsmount *clone_private_mount(const struct path *path)
1855 struct mount *old_mnt = real_mount(path->mnt);
1856 struct mount *new_mnt;
1858 if (IS_MNT_UNBINDABLE(old_mnt))
1859 return ERR_PTR(-EINVAL);
1861 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1862 if (IS_ERR(new_mnt))
1863 return ERR_CAST(new_mnt);
1865 return &new_mnt->mnt;
1867 EXPORT_SYMBOL_GPL(clone_private_mount);
1869 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1870 struct vfsmount *root)
1873 int res = f(root, arg);
1876 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1877 res = f(&mnt->mnt, arg);
1884 static void lock_mnt_tree(struct mount *mnt)
1888 for (p = mnt; p; p = next_mnt(p, mnt)) {
1889 int flags = p->mnt.mnt_flags;
1890 /* Don't allow unprivileged users to change mount flags */
1891 flags |= MNT_LOCK_ATIME;
1893 if (flags & MNT_READONLY)
1894 flags |= MNT_LOCK_READONLY;
1896 if (flags & MNT_NODEV)
1897 flags |= MNT_LOCK_NODEV;
1899 if (flags & MNT_NOSUID)
1900 flags |= MNT_LOCK_NOSUID;
1902 if (flags & MNT_NOEXEC)
1903 flags |= MNT_LOCK_NOEXEC;
1904 /* Don't allow unprivileged users to reveal what is under a mount */
1905 if (list_empty(&p->mnt_expire))
1906 flags |= MNT_LOCKED;
1907 p->mnt.mnt_flags = flags;
1911 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1915 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1916 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1917 mnt_release_group_id(p);
1921 static int invent_group_ids(struct mount *mnt, bool recurse)
1925 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1926 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1927 int err = mnt_alloc_group_id(p);
1929 cleanup_group_ids(mnt, p);
1938 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1940 unsigned int max = READ_ONCE(sysctl_mount_max);
1941 unsigned int mounts = 0, old, pending, sum;
1944 for (p = mnt; p; p = next_mnt(p, mnt))
1948 pending = ns->pending_mounts;
1949 sum = old + pending;
1953 (mounts > (max - sum)))
1956 ns->pending_mounts = pending + mounts;
1961 * @source_mnt : mount tree to be attached
1962 * @nd : place the mount tree @source_mnt is attached
1963 * @parent_nd : if non-null, detach the source_mnt from its parent and
1964 * store the parent mount and mountpoint dentry.
1965 * (done when source_mnt is moved)
1967 * NOTE: in the table below explains the semantics when a source mount
1968 * of a given type is attached to a destination mount of a given type.
1969 * ---------------------------------------------------------------------------
1970 * | BIND MOUNT OPERATION |
1971 * |**************************************************************************
1972 * | source-->| shared | private | slave | unbindable |
1976 * |**************************************************************************
1977 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1979 * |non-shared| shared (+) | private | slave (*) | invalid |
1980 * ***************************************************************************
1981 * A bind operation clones the source mount and mounts the clone on the
1982 * destination mount.
1984 * (++) the cloned mount is propagated to all the mounts in the propagation
1985 * tree of the destination mount and the cloned mount is added to
1986 * the peer group of the source mount.
1987 * (+) the cloned mount is created under the destination mount and is marked
1988 * as shared. The cloned mount is added to the peer group of the source
1990 * (+++) the mount is propagated to all the mounts in the propagation tree
1991 * of the destination mount and the cloned mount is made slave
1992 * of the same master as that of the source mount. The cloned mount
1993 * is marked as 'shared and slave'.
1994 * (*) the cloned mount is made a slave of the same master as that of the
1997 * ---------------------------------------------------------------------------
1998 * | MOVE MOUNT OPERATION |
1999 * |**************************************************************************
2000 * | source-->| shared | private | slave | unbindable |
2004 * |**************************************************************************
2005 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2007 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2008 * ***************************************************************************
2010 * (+) the mount is moved to the destination. And is then propagated to
2011 * all the mounts in the propagation tree of the destination mount.
2012 * (+*) the mount is moved to the destination.
2013 * (+++) the mount is moved to the destination and is then propagated to
2014 * all the mounts belonging to the destination mount's propagation tree.
2015 * the mount is marked as 'shared and slave'.
2016 * (*) the mount continues to be a slave at the new location.
2018 * if the source mount is a tree, the operations explained above is
2019 * applied to each mount in the tree.
2020 * Must be called without spinlocks held, since this function can sleep
2023 static int attach_recursive_mnt(struct mount *source_mnt,
2024 struct mount *dest_mnt,
2025 struct mountpoint *dest_mp,
2026 struct path *parent_path)
2028 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2029 HLIST_HEAD(tree_list);
2030 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2031 struct mountpoint *smp;
2032 struct mount *child, *p;
2033 struct hlist_node *n;
2036 /* Preallocate a mountpoint in case the new mounts need
2037 * to be tucked under other mounts.
2039 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2041 return PTR_ERR(smp);
2043 /* Is there space to add these mounts to the mount namespace? */
2045 err = count_mounts(ns, source_mnt);
2050 if (IS_MNT_SHARED(dest_mnt)) {
2051 err = invent_group_ids(source_mnt, true);
2054 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2057 goto out_cleanup_ids;
2058 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2064 detach_mnt(source_mnt, parent_path);
2065 attach_mnt(source_mnt, dest_mnt, dest_mp);
2066 touch_mnt_namespace(source_mnt->mnt_ns);
2068 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2069 commit_tree(source_mnt);
2072 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2074 hlist_del_init(&child->mnt_hash);
2075 q = __lookup_mnt(&child->mnt_parent->mnt,
2076 child->mnt_mountpoint);
2078 mnt_change_mountpoint(child, smp, q);
2079 /* Notice when we are propagating across user namespaces */
2080 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2081 lock_mnt_tree(child);
2084 put_mountpoint(smp);
2085 unlock_mount_hash();
2090 while (!hlist_empty(&tree_list)) {
2091 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2092 child->mnt_parent->mnt_ns->pending_mounts = 0;
2093 umount_tree(child, UMOUNT_SYNC);
2095 unlock_mount_hash();
2096 cleanup_group_ids(source_mnt, NULL);
2098 ns->pending_mounts = 0;
2100 read_seqlock_excl(&mount_lock);
2101 put_mountpoint(smp);
2102 read_sequnlock_excl(&mount_lock);
2107 static struct mountpoint *lock_mount(struct path *path)
2109 struct vfsmount *mnt;
2110 struct dentry *dentry = path->dentry;
2112 inode_lock(dentry->d_inode);
2113 if (unlikely(cant_mount(dentry))) {
2114 inode_unlock(dentry->d_inode);
2115 return ERR_PTR(-ENOENT);
2118 mnt = lookup_mnt(path);
2120 struct mountpoint *mp = get_mountpoint(dentry);
2123 inode_unlock(dentry->d_inode);
2129 inode_unlock(path->dentry->d_inode);
2132 dentry = path->dentry = dget(mnt->mnt_root);
2136 static void unlock_mount(struct mountpoint *where)
2138 struct dentry *dentry = where->m_dentry;
2140 read_seqlock_excl(&mount_lock);
2141 put_mountpoint(where);
2142 read_sequnlock_excl(&mount_lock);
2145 inode_unlock(dentry->d_inode);
2148 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2150 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2153 if (d_is_dir(mp->m_dentry) !=
2154 d_is_dir(mnt->mnt.mnt_root))
2157 return attach_recursive_mnt(mnt, p, mp, NULL);
2161 * Sanity check the flags to change_mnt_propagation.
2164 static int flags_to_propagation_type(int ms_flags)
2166 int type = ms_flags & ~(MS_REC | MS_SILENT);
2168 /* Fail if any non-propagation flags are set */
2169 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2171 /* Only one propagation flag should be set */
2172 if (!is_power_of_2(type))
2178 * recursively change the type of the mountpoint.
2180 static int do_change_type(struct path *path, int ms_flags)
2183 struct mount *mnt = real_mount(path->mnt);
2184 int recurse = ms_flags & MS_REC;
2188 if (path->dentry != path->mnt->mnt_root)
2191 type = flags_to_propagation_type(ms_flags);
2196 if (type == MS_SHARED) {
2197 err = invent_group_ids(mnt, recurse);
2203 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2204 change_mnt_propagation(m, type);
2205 unlock_mount_hash();
2212 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2214 struct mount *child;
2215 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2216 if (!is_subdir(child->mnt_mountpoint, dentry))
2219 if (child->mnt.mnt_flags & MNT_LOCKED)
2226 * do loopback mount.
2228 static int do_loopback(struct path *path, const char *old_name,
2231 struct path old_path;
2232 struct mount *mnt = NULL, *old, *parent;
2233 struct mountpoint *mp;
2235 if (!old_name || !*old_name)
2237 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2242 if (mnt_ns_loop(old_path.dentry))
2245 mp = lock_mount(path);
2250 old = real_mount(old_path.mnt);
2251 parent = real_mount(path->mnt);
2254 if (IS_MNT_UNBINDABLE(old))
2257 if (!check_mnt(parent))
2260 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2263 if (!recurse && has_locked_children(old, old_path.dentry))
2267 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2269 mnt = clone_mnt(old, old_path.dentry, 0);
2276 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2278 err = graft_tree(mnt, parent, mp);
2281 umount_tree(mnt, UMOUNT_SYNC);
2282 unlock_mount_hash();
2287 path_put(&old_path);
2292 * Don't allow locked mount flags to be cleared.
2294 * No locks need to be held here while testing the various MNT_LOCK
2295 * flags because those flags can never be cleared once they are set.
2297 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2299 unsigned int fl = mnt->mnt.mnt_flags;
2301 if ((fl & MNT_LOCK_READONLY) &&
2302 !(mnt_flags & MNT_READONLY))
2305 if ((fl & MNT_LOCK_NODEV) &&
2306 !(mnt_flags & MNT_NODEV))
2309 if ((fl & MNT_LOCK_NOSUID) &&
2310 !(mnt_flags & MNT_NOSUID))
2313 if ((fl & MNT_LOCK_NOEXEC) &&
2314 !(mnt_flags & MNT_NOEXEC))
2317 if ((fl & MNT_LOCK_ATIME) &&
2318 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2324 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2326 bool readonly_request = (mnt_flags & MNT_READONLY);
2328 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2331 if (readonly_request)
2332 return mnt_make_readonly(mnt);
2334 return __mnt_unmake_readonly(mnt);
2338 * Update the user-settable attributes on a mount. The caller must hold
2339 * sb->s_umount for writing.
2341 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2344 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2345 mnt->mnt.mnt_flags = mnt_flags;
2346 touch_mnt_namespace(mnt->mnt_ns);
2347 unlock_mount_hash();
2351 * Handle reconfiguration of the mountpoint only without alteration of the
2352 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2355 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2357 struct super_block *sb = path->mnt->mnt_sb;
2358 struct mount *mnt = real_mount(path->mnt);
2361 if (!check_mnt(mnt))
2364 if (path->dentry != mnt->mnt.mnt_root)
2367 if (!can_change_locked_flags(mnt, mnt_flags))
2370 down_write(&sb->s_umount);
2371 ret = change_mount_ro_state(mnt, mnt_flags);
2373 set_mount_attributes(mnt, mnt_flags);
2374 up_write(&sb->s_umount);
2379 * change filesystem flags. dir should be a physical root of filesystem.
2380 * If you've mounted a non-root directory somewhere and want to do remount
2381 * on it - tough luck.
2383 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2384 int mnt_flags, void *data)
2387 struct super_block *sb = path->mnt->mnt_sb;
2388 struct mount *mnt = real_mount(path->mnt);
2389 struct fs_context *fc;
2391 if (!check_mnt(mnt))
2394 if (path->dentry != path->mnt->mnt_root)
2397 if (!can_change_locked_flags(mnt, mnt_flags))
2400 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2404 err = parse_monolithic_mount_data(fc, data);
2406 down_write(&sb->s_umount);
2408 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2409 err = reconfigure_super(fc);
2411 set_mount_attributes(mnt, mnt_flags);
2413 up_write(&sb->s_umount);
2419 static inline int tree_contains_unbindable(struct mount *mnt)
2422 for (p = mnt; p; p = next_mnt(p, mnt)) {
2423 if (IS_MNT_UNBINDABLE(p))
2429 static int do_move_mount(struct path *path, const char *old_name)
2431 struct path old_path, parent_path;
2434 struct mountpoint *mp;
2436 if (!old_name || !*old_name)
2438 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2442 mp = lock_mount(path);
2447 old = real_mount(old_path.mnt);
2448 p = real_mount(path->mnt);
2451 if (!check_mnt(p) || !check_mnt(old))
2454 if (old->mnt.mnt_flags & MNT_LOCKED)
2458 if (old_path.dentry != old_path.mnt->mnt_root)
2461 if (!mnt_has_parent(old))
2464 if (d_is_dir(path->dentry) !=
2465 d_is_dir(old_path.dentry))
2468 * Don't move a mount residing in a shared parent.
2470 if (IS_MNT_SHARED(old->mnt_parent))
2473 * Don't move a mount tree containing unbindable mounts to a destination
2474 * mount which is shared.
2476 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2479 for (; mnt_has_parent(p); p = p->mnt_parent)
2483 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2487 /* if the mount is moved, it should no longer be expire
2489 list_del_init(&old->mnt_expire);
2494 path_put(&parent_path);
2495 path_put(&old_path);
2500 * add a mount into a namespace's mount tree
2502 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2504 struct mountpoint *mp;
2505 struct mount *parent;
2508 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2510 mp = lock_mount(path);
2514 parent = real_mount(path->mnt);
2516 if (unlikely(!check_mnt(parent))) {
2517 /* that's acceptable only for automounts done in private ns */
2518 if (!(mnt_flags & MNT_SHRINKABLE))
2520 /* ... and for those we'd better have mountpoint still alive */
2521 if (!parent->mnt_ns)
2525 /* Refuse the same filesystem on the same mount point */
2527 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2528 path->mnt->mnt_root == path->dentry)
2532 if (d_is_symlink(newmnt->mnt.mnt_root))
2535 newmnt->mnt.mnt_flags = mnt_flags;
2536 err = graft_tree(newmnt, parent, mp);
2543 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2546 * Create a new mount using a superblock configuration and request it
2547 * be added to the namespace tree.
2549 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2550 unsigned int mnt_flags)
2552 struct vfsmount *mnt;
2553 struct super_block *sb = fc->root->d_sb;
2556 error = security_sb_kern_mount(sb);
2557 if (!error && mount_too_revealing(sb, &mnt_flags))
2560 if (unlikely(error)) {
2565 up_write(&sb->s_umount);
2567 mnt = vfs_create_mount(fc);
2569 return PTR_ERR(mnt);
2571 error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags);
2578 * create a new mount for userspace and request it to be added into the
2581 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2582 int mnt_flags, const char *name, void *data)
2584 struct file_system_type *type;
2585 struct fs_context *fc;
2586 const char *subtype = NULL;
2592 type = get_fs_type(fstype);
2596 if (type->fs_flags & FS_HAS_SUBTYPE) {
2597 subtype = strchr(fstype, '.');
2601 put_filesystem(type);
2609 fc = fs_context_for_mount(type, sb_flags);
2610 put_filesystem(type);
2615 err = vfs_parse_fs_string(fc, "subtype",
2616 subtype, strlen(subtype));
2618 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2620 err = parse_monolithic_mount_data(fc, data);
2622 err = vfs_get_tree(fc);
2624 err = do_new_mount_fc(fc, path, mnt_flags);
2630 int finish_automount(struct vfsmount *m, struct path *path)
2632 struct mount *mnt = real_mount(m);
2634 /* The new mount record should have at least 2 refs to prevent it being
2635 * expired before we get a chance to add it
2637 BUG_ON(mnt_get_count(mnt) < 2);
2639 if (m->mnt_sb == path->mnt->mnt_sb &&
2640 m->mnt_root == path->dentry) {
2645 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2649 /* remove m from any expiration list it may be on */
2650 if (!list_empty(&mnt->mnt_expire)) {
2652 list_del_init(&mnt->mnt_expire);
2661 * mnt_set_expiry - Put a mount on an expiration list
2662 * @mnt: The mount to list.
2663 * @expiry_list: The list to add the mount to.
2665 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2669 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2673 EXPORT_SYMBOL(mnt_set_expiry);
2676 * process a list of expirable mountpoints with the intent of discarding any
2677 * mountpoints that aren't in use and haven't been touched since last we came
2680 void mark_mounts_for_expiry(struct list_head *mounts)
2682 struct mount *mnt, *next;
2683 LIST_HEAD(graveyard);
2685 if (list_empty(mounts))
2691 /* extract from the expiration list every vfsmount that matches the
2692 * following criteria:
2693 * - only referenced by its parent vfsmount
2694 * - still marked for expiry (marked on the last call here; marks are
2695 * cleared by mntput())
2697 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2698 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2699 propagate_mount_busy(mnt, 1))
2701 list_move(&mnt->mnt_expire, &graveyard);
2703 while (!list_empty(&graveyard)) {
2704 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2705 touch_mnt_namespace(mnt->mnt_ns);
2706 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2708 unlock_mount_hash();
2712 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2715 * Ripoff of 'select_parent()'
2717 * search the list of submounts for a given mountpoint, and move any
2718 * shrinkable submounts to the 'graveyard' list.
2720 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2722 struct mount *this_parent = parent;
2723 struct list_head *next;
2727 next = this_parent->mnt_mounts.next;
2729 while (next != &this_parent->mnt_mounts) {
2730 struct list_head *tmp = next;
2731 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2734 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2737 * Descend a level if the d_mounts list is non-empty.
2739 if (!list_empty(&mnt->mnt_mounts)) {
2744 if (!propagate_mount_busy(mnt, 1)) {
2745 list_move_tail(&mnt->mnt_expire, graveyard);
2750 * All done at this level ... ascend and resume the search
2752 if (this_parent != parent) {
2753 next = this_parent->mnt_child.next;
2754 this_parent = this_parent->mnt_parent;
2761 * process a list of expirable mountpoints with the intent of discarding any
2762 * submounts of a specific parent mountpoint
2764 * mount_lock must be held for write
2766 static void shrink_submounts(struct mount *mnt)
2768 LIST_HEAD(graveyard);
2771 /* extract submounts of 'mountpoint' from the expiration list */
2772 while (select_submounts(mnt, &graveyard)) {
2773 while (!list_empty(&graveyard)) {
2774 m = list_first_entry(&graveyard, struct mount,
2776 touch_mnt_namespace(m->mnt_ns);
2777 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2783 * Some copy_from_user() implementations do not return the exact number of
2784 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2785 * Note that this function differs from copy_from_user() in that it will oops
2786 * on bad values of `to', rather than returning a short copy.
2788 static long exact_copy_from_user(void *to, const void __user * from,
2792 const char __user *f = from;
2795 if (!access_ok(from, n))
2799 if (__get_user(c, f)) {
2810 void *copy_mount_options(const void __user * data)
2819 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2821 return ERR_PTR(-ENOMEM);
2823 /* We only care that *some* data at the address the user
2824 * gave us is valid. Just in case, we'll zero
2825 * the remainder of the page.
2827 /* copy_from_user cannot cross TASK_SIZE ! */
2828 size = TASK_SIZE - (unsigned long)data;
2829 if (size > PAGE_SIZE)
2832 i = size - exact_copy_from_user(copy, data, size);
2835 return ERR_PTR(-EFAULT);
2838 memset(copy + i, 0, PAGE_SIZE - i);
2842 char *copy_mount_string(const void __user *data)
2844 return data ? strndup_user(data, PATH_MAX) : NULL;
2848 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2849 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2851 * data is a (void *) that can point to any structure up to
2852 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2853 * information (or be NULL).
2855 * Pre-0.97 versions of mount() didn't have a flags word.
2856 * When the flags word was introduced its top half was required
2857 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2858 * Therefore, if this magic number is present, it carries no information
2859 * and must be discarded.
2861 long do_mount(const char *dev_name, const char __user *dir_name,
2862 const char *type_page, unsigned long flags, void *data_page)
2865 unsigned int mnt_flags = 0, sb_flags;
2869 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2870 flags &= ~MS_MGC_MSK;
2872 /* Basic sanity checks */
2874 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2876 if (flags & MS_NOUSER)
2879 /* ... and get the mountpoint */
2880 retval = user_path(dir_name, &path);
2884 retval = security_sb_mount(dev_name, &path,
2885 type_page, flags, data_page);
2886 if (!retval && !may_mount())
2888 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
2893 /* Default to relatime unless overriden */
2894 if (!(flags & MS_NOATIME))
2895 mnt_flags |= MNT_RELATIME;
2897 /* Separate the per-mountpoint flags */
2898 if (flags & MS_NOSUID)
2899 mnt_flags |= MNT_NOSUID;
2900 if (flags & MS_NODEV)
2901 mnt_flags |= MNT_NODEV;
2902 if (flags & MS_NOEXEC)
2903 mnt_flags |= MNT_NOEXEC;
2904 if (flags & MS_NOATIME)
2905 mnt_flags |= MNT_NOATIME;
2906 if (flags & MS_NODIRATIME)
2907 mnt_flags |= MNT_NODIRATIME;
2908 if (flags & MS_STRICTATIME)
2909 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2910 if (flags & MS_RDONLY)
2911 mnt_flags |= MNT_READONLY;
2913 /* The default atime for remount is preservation */
2914 if ((flags & MS_REMOUNT) &&
2915 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2916 MS_STRICTATIME)) == 0)) {
2917 mnt_flags &= ~MNT_ATIME_MASK;
2918 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2921 sb_flags = flags & (SB_RDONLY |
2930 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
2931 retval = do_reconfigure_mnt(&path, mnt_flags);
2932 else if (flags & MS_REMOUNT)
2933 retval = do_remount(&path, flags, sb_flags, mnt_flags,
2935 else if (flags & MS_BIND)
2936 retval = do_loopback(&path, dev_name, flags & MS_REC);
2937 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2938 retval = do_change_type(&path, flags);
2939 else if (flags & MS_MOVE)
2940 retval = do_move_mount(&path, dev_name);
2942 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2943 dev_name, data_page);
2949 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2951 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2954 static void dec_mnt_namespaces(struct ucounts *ucounts)
2956 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2959 static void free_mnt_ns(struct mnt_namespace *ns)
2961 if (!is_anon_ns(ns))
2962 ns_free_inum(&ns->ns);
2963 dec_mnt_namespaces(ns->ucounts);
2964 put_user_ns(ns->user_ns);
2969 * Assign a sequence number so we can detect when we attempt to bind
2970 * mount a reference to an older mount namespace into the current
2971 * mount namespace, preventing reference counting loops. A 64bit
2972 * number incrementing at 10Ghz will take 12,427 years to wrap which
2973 * is effectively never, so we can ignore the possibility.
2975 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2977 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
2979 struct mnt_namespace *new_ns;
2980 struct ucounts *ucounts;
2983 ucounts = inc_mnt_namespaces(user_ns);
2985 return ERR_PTR(-ENOSPC);
2987 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2989 dec_mnt_namespaces(ucounts);
2990 return ERR_PTR(-ENOMEM);
2993 ret = ns_alloc_inum(&new_ns->ns);
2996 dec_mnt_namespaces(ucounts);
2997 return ERR_PTR(ret);
3000 new_ns->ns.ops = &mntns_operations;
3002 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3003 atomic_set(&new_ns->count, 1);
3004 INIT_LIST_HEAD(&new_ns->list);
3005 init_waitqueue_head(&new_ns->poll);
3006 new_ns->user_ns = get_user_ns(user_ns);
3007 new_ns->ucounts = ucounts;
3012 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3013 struct user_namespace *user_ns, struct fs_struct *new_fs)
3015 struct mnt_namespace *new_ns;
3016 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3017 struct mount *p, *q;
3024 if (likely(!(flags & CLONE_NEWNS))) {
3031 new_ns = alloc_mnt_ns(user_ns, false);
3036 /* First pass: copy the tree topology */
3037 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3038 if (user_ns != ns->user_ns)
3039 copy_flags |= CL_SHARED_TO_SLAVE;
3040 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3043 free_mnt_ns(new_ns);
3044 return ERR_CAST(new);
3046 if (user_ns != ns->user_ns) {
3049 unlock_mount_hash();
3052 list_add_tail(&new_ns->list, &new->mnt_list);
3055 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3056 * as belonging to new namespace. We have already acquired a private
3057 * fs_struct, so tsk->fs->lock is not needed.
3065 if (&p->mnt == new_fs->root.mnt) {
3066 new_fs->root.mnt = mntget(&q->mnt);
3069 if (&p->mnt == new_fs->pwd.mnt) {
3070 new_fs->pwd.mnt = mntget(&q->mnt);
3074 p = next_mnt(p, old);
3075 q = next_mnt(q, new);
3078 while (p->mnt.mnt_root != q->mnt.mnt_root)
3079 p = next_mnt(p, old);
3091 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3093 struct mount *mnt = real_mount(m);
3094 struct mnt_namespace *ns;
3095 struct super_block *s;
3099 ns = alloc_mnt_ns(&init_user_ns, true);
3102 return ERR_CAST(ns);
3107 list_add(&mnt->mnt_list, &ns->list);
3109 err = vfs_path_lookup(m->mnt_root, m,
3110 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3115 return ERR_PTR(err);
3117 /* trade a vfsmount reference for active sb one */
3118 s = path.mnt->mnt_sb;
3119 atomic_inc(&s->s_active);
3121 /* lock the sucker */
3122 down_write(&s->s_umount);
3123 /* ... and return the root of (sub)tree on it */
3126 EXPORT_SYMBOL(mount_subtree);
3128 int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type,
3129 unsigned long flags, void __user *data)
3136 kernel_type = copy_mount_string(type);
3137 ret = PTR_ERR(kernel_type);
3138 if (IS_ERR(kernel_type))
3141 kernel_dev = copy_mount_string(dev_name);
3142 ret = PTR_ERR(kernel_dev);
3143 if (IS_ERR(kernel_dev))
3146 options = copy_mount_options(data);
3147 ret = PTR_ERR(options);
3148 if (IS_ERR(options))
3151 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3162 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3163 char __user *, type, unsigned long, flags, void __user *, data)
3165 return ksys_mount(dev_name, dir_name, type, flags, data);
3169 * Return true if path is reachable from root
3171 * namespace_sem or mount_lock is held
3173 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3174 const struct path *root)
3176 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3177 dentry = mnt->mnt_mountpoint;
3178 mnt = mnt->mnt_parent;
3180 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3183 bool path_is_under(const struct path *path1, const struct path *path2)
3186 read_seqlock_excl(&mount_lock);
3187 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3188 read_sequnlock_excl(&mount_lock);
3191 EXPORT_SYMBOL(path_is_under);
3194 * pivot_root Semantics:
3195 * Moves the root file system of the current process to the directory put_old,
3196 * makes new_root as the new root file system of the current process, and sets
3197 * root/cwd of all processes which had them on the current root to new_root.
3200 * The new_root and put_old must be directories, and must not be on the
3201 * same file system as the current process root. The put_old must be
3202 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3203 * pointed to by put_old must yield the same directory as new_root. No other
3204 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3206 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3207 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3208 * in this situation.
3211 * - we don't move root/cwd if they are not at the root (reason: if something
3212 * cared enough to change them, it's probably wrong to force them elsewhere)
3213 * - it's okay to pick a root that isn't the root of a file system, e.g.
3214 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3215 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3218 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3219 const char __user *, put_old)
3221 struct path new, old, parent_path, root_parent, root;
3222 struct mount *new_mnt, *root_mnt, *old_mnt;
3223 struct mountpoint *old_mp, *root_mp;
3229 error = user_path_dir(new_root, &new);
3233 error = user_path_dir(put_old, &old);
3237 error = security_sb_pivotroot(&old, &new);
3241 get_fs_root(current->fs, &root);
3242 old_mp = lock_mount(&old);
3243 error = PTR_ERR(old_mp);
3248 new_mnt = real_mount(new.mnt);
3249 root_mnt = real_mount(root.mnt);
3250 old_mnt = real_mount(old.mnt);
3251 if (IS_MNT_SHARED(old_mnt) ||
3252 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3253 IS_MNT_SHARED(root_mnt->mnt_parent))
3255 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3257 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3260 if (d_unlinked(new.dentry))
3263 if (new_mnt == root_mnt || old_mnt == root_mnt)
3264 goto out4; /* loop, on the same file system */
3266 if (root.mnt->mnt_root != root.dentry)
3267 goto out4; /* not a mountpoint */
3268 if (!mnt_has_parent(root_mnt))
3269 goto out4; /* not attached */
3270 root_mp = root_mnt->mnt_mp;
3271 if (new.mnt->mnt_root != new.dentry)
3272 goto out4; /* not a mountpoint */
3273 if (!mnt_has_parent(new_mnt))
3274 goto out4; /* not attached */
3275 /* make sure we can reach put_old from new_root */
3276 if (!is_path_reachable(old_mnt, old.dentry, &new))
3278 /* make certain new is below the root */
3279 if (!is_path_reachable(new_mnt, new.dentry, &root))
3281 root_mp->m_count++; /* pin it so it won't go away */
3283 detach_mnt(new_mnt, &parent_path);
3284 detach_mnt(root_mnt, &root_parent);
3285 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3286 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3287 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3289 /* mount old root on put_old */
3290 attach_mnt(root_mnt, old_mnt, old_mp);
3291 /* mount new_root on / */
3292 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3293 touch_mnt_namespace(current->nsproxy->mnt_ns);
3294 /* A moved mount should not expire automatically */
3295 list_del_init(&new_mnt->mnt_expire);
3296 put_mountpoint(root_mp);
3297 unlock_mount_hash();
3298 chroot_fs_refs(&root, &new);
3301 unlock_mount(old_mp);
3303 path_put(&root_parent);
3304 path_put(&parent_path);
3316 static void __init init_mount_tree(void)
3318 struct vfsmount *mnt;
3320 struct mnt_namespace *ns;
3322 struct file_system_type *type;
3324 type = get_fs_type("rootfs");
3326 panic("Can't find rootfs type");
3327 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3328 put_filesystem(type);
3330 panic("Can't create rootfs");
3332 ns = alloc_mnt_ns(&init_user_ns, false);
3334 panic("Can't allocate initial namespace");
3335 m = real_mount(mnt);
3339 list_add(&m->mnt_list, &ns->list);
3340 init_task.nsproxy->mnt_ns = ns;
3344 root.dentry = mnt->mnt_root;
3345 mnt->mnt_flags |= MNT_LOCKED;
3347 set_fs_pwd(current->fs, &root);
3348 set_fs_root(current->fs, &root);
3351 void __init mnt_init(void)
3355 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3356 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3358 mount_hashtable = alloc_large_system_hash("Mount-cache",
3359 sizeof(struct hlist_head),
3362 &m_hash_shift, &m_hash_mask, 0, 0);
3363 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3364 sizeof(struct hlist_head),
3367 &mp_hash_shift, &mp_hash_mask, 0, 0);
3369 if (!mount_hashtable || !mountpoint_hashtable)
3370 panic("Failed to allocate mount hash table\n");
3376 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3378 fs_kobj = kobject_create_and_add("fs", NULL);
3380 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3385 void put_mnt_ns(struct mnt_namespace *ns)
3387 if (!atomic_dec_and_test(&ns->count))
3389 drop_collected_mounts(&ns->root->mnt);
3393 struct vfsmount *kern_mount(struct file_system_type *type)
3395 struct vfsmount *mnt;
3396 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3399 * it is a longterm mount, don't release mnt until
3400 * we unmount before file sys is unregistered
3402 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3406 EXPORT_SYMBOL_GPL(kern_mount);
3408 void kern_unmount(struct vfsmount *mnt)
3410 /* release long term mount so mount point can be released */
3411 if (!IS_ERR_OR_NULL(mnt)) {
3412 real_mount(mnt)->mnt_ns = NULL;
3413 synchronize_rcu(); /* yecchhh... */
3417 EXPORT_SYMBOL(kern_unmount);
3419 bool our_mnt(struct vfsmount *mnt)
3421 return check_mnt(real_mount(mnt));
3424 bool current_chrooted(void)
3426 /* Does the current process have a non-standard root */
3427 struct path ns_root;
3428 struct path fs_root;
3431 /* Find the namespace root */
3432 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3433 ns_root.dentry = ns_root.mnt->mnt_root;
3435 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3438 get_fs_root(current->fs, &fs_root);
3440 chrooted = !path_equal(&fs_root, &ns_root);
3448 static bool mnt_already_visible(struct mnt_namespace *ns,
3449 const struct super_block *sb,
3452 int new_flags = *new_mnt_flags;
3454 bool visible = false;
3456 down_read(&namespace_sem);
3457 list_for_each_entry(mnt, &ns->list, mnt_list) {
3458 struct mount *child;
3461 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3464 /* This mount is not fully visible if it's root directory
3465 * is not the root directory of the filesystem.
3467 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3470 /* A local view of the mount flags */
3471 mnt_flags = mnt->mnt.mnt_flags;
3473 /* Don't miss readonly hidden in the superblock flags */
3474 if (sb_rdonly(mnt->mnt.mnt_sb))
3475 mnt_flags |= MNT_LOCK_READONLY;
3477 /* Verify the mount flags are equal to or more permissive
3478 * than the proposed new mount.
3480 if ((mnt_flags & MNT_LOCK_READONLY) &&
3481 !(new_flags & MNT_READONLY))
3483 if ((mnt_flags & MNT_LOCK_ATIME) &&
3484 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3487 /* This mount is not fully visible if there are any
3488 * locked child mounts that cover anything except for
3489 * empty directories.
3491 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3492 struct inode *inode = child->mnt_mountpoint->d_inode;
3493 /* Only worry about locked mounts */
3494 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3496 /* Is the directory permanetly empty? */
3497 if (!is_empty_dir_inode(inode))
3500 /* Preserve the locked attributes */
3501 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3508 up_read(&namespace_sem);
3512 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3514 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3515 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3516 unsigned long s_iflags;
3518 if (ns->user_ns == &init_user_ns)
3521 /* Can this filesystem be too revealing? */
3522 s_iflags = sb->s_iflags;
3523 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3526 if ((s_iflags & required_iflags) != required_iflags) {
3527 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3532 return !mnt_already_visible(ns, sb, new_mnt_flags);
3535 bool mnt_may_suid(struct vfsmount *mnt)
3538 * Foreign mounts (accessed via fchdir or through /proc
3539 * symlinks) are always treated as if they are nosuid. This
3540 * prevents namespaces from trusting potentially unsafe
3541 * suid/sgid bits, file caps, or security labels that originate
3542 * in other namespaces.
3544 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3545 current_in_userns(mnt->mnt_sb->s_user_ns);
3548 static struct ns_common *mntns_get(struct task_struct *task)
3550 struct ns_common *ns = NULL;
3551 struct nsproxy *nsproxy;
3554 nsproxy = task->nsproxy;
3556 ns = &nsproxy->mnt_ns->ns;
3557 get_mnt_ns(to_mnt_ns(ns));
3564 static void mntns_put(struct ns_common *ns)
3566 put_mnt_ns(to_mnt_ns(ns));
3569 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3571 struct fs_struct *fs = current->fs;
3572 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3576 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3577 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3578 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3581 if (is_anon_ns(mnt_ns))
3588 old_mnt_ns = nsproxy->mnt_ns;
3589 nsproxy->mnt_ns = mnt_ns;
3592 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3593 "/", LOOKUP_DOWN, &root);
3595 /* revert to old namespace */
3596 nsproxy->mnt_ns = old_mnt_ns;
3601 put_mnt_ns(old_mnt_ns);
3603 /* Update the pwd and root */
3604 set_fs_pwd(fs, &root);
3605 set_fs_root(fs, &root);
3611 static struct user_namespace *mntns_owner(struct ns_common *ns)
3613 return to_mnt_ns(ns)->user_ns;
3616 const struct proc_ns_operations mntns_operations = {
3618 .type = CLONE_NEWNS,
3621 .install = mntns_install,
3622 .owner = mntns_owner,