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/idr.h>
19 #include <linux/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
40 mhash_entries = simple_strtoul(str, &str, 0);
43 __setup("mhash_entries=", set_mhash_entries);
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
50 mphash_entries = simple_strtoul(str, &str, 0);
53 __setup("mphash_entries=", set_mphash_entries);
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount *mnt)
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
117 static void mnt_free_id(struct mount *mnt)
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
124 spin_unlock(&mnt_id_lock);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount *mnt)
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start = mnt->mnt_group_id + 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount *mnt)
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount *mnt, int n)
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount *mnt)
180 unsigned int count = 0;
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
189 return mnt->mnt_count;
193 static void drop_mountpoint(struct fs_pin *p)
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
201 static struct mount *alloc_vfsmnt(const char *name)
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 err = mnt_alloc_id(mnt);
212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 if (!mnt->mnt_devname)
218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
220 goto out_free_devname;
222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225 mnt->mnt_writers = 0;
228 INIT_HLIST_NODE(&mnt->mnt_hash);
229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
246 kfree_const(mnt->mnt_devname);
251 kmem_cache_free(mnt_cache, mnt);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount *mnt)
276 if (mnt->mnt_flags & MNT_READONLY)
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
284 static inline void mnt_inc_writers(struct mount *mnt)
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
293 static inline void mnt_dec_writers(struct mount *mnt)
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
302 static unsigned int mnt_get_writers(struct mount *mnt)
305 unsigned int count = 0;
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
314 return mnt->mnt_writers;
318 static int mnt_is_readonly(struct vfsmount *mnt)
320 if (mnt->mnt_sb->s_readonly_remount)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount *m)
345 struct mount *mnt = real_mount(m);
349 mnt_inc_writers(mnt);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount *m)
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
389 sb_end_write(m->mnt_sb);
392 EXPORT_SYMBOL_GPL(mnt_want_write);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount *mnt)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
412 mnt_inc_writers(real_mount(mnt));
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file *file)
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
430 return mnt_clone_write(file->f_path.mnt);
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 int mnt_want_write_file(struct file *file)
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
447 sb_end_write(file->f_path.mnt->mnt_sb);
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
460 void __mnt_drop_write(struct vfsmount *mnt)
463 mnt_dec_writers(real_mount(mnt));
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
475 void mnt_drop_write(struct vfsmount *mnt)
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
482 void __mnt_drop_write_file(struct file *file)
484 __mnt_drop_write(file->f_path.mnt);
487 void mnt_drop_write_file(struct file *file)
489 mnt_drop_write(file->f_path.mnt);
491 EXPORT_SYMBOL(mnt_drop_write_file);
493 static int mnt_make_readonly(struct mount *mnt)
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
521 if (mnt_get_writers(mnt) > 0)
524 mnt->mnt.mnt_flags |= MNT_READONLY;
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
535 static void __mnt_unmake_readonly(struct mount *mnt)
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
542 int sb_prepare_remount_readonly(struct super_block *sb)
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
556 if (mnt_get_writers(mnt) > 0) {
562 if (!err && atomic_long_read(&sb->s_remove_count))
566 sb->s_readonly_remount = 1;
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
578 static void free_vfsmnt(struct mount *mnt)
580 kfree(mnt->mnt.data);
581 kfree_const(mnt->mnt_devname);
583 free_percpu(mnt->mnt_pcp);
585 kmem_cache_free(mnt_cache, mnt);
588 static void delayed_free_vfsmnt(struct rcu_head *head)
590 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
593 /* call under rcu_read_lock */
594 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
597 if (read_seqretry(&mount_lock, seq))
601 mnt = real_mount(bastard);
602 mnt_add_count(mnt, 1);
603 if (likely(!read_seqretry(&mount_lock, seq)))
605 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
606 mnt_add_count(mnt, -1);
612 /* call under rcu_read_lock */
613 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
615 int res = __legitimize_mnt(bastard, seq);
618 if (unlikely(res < 0)) {
627 * find the first mount at @dentry on vfsmount @mnt.
628 * call under rcu_read_lock()
630 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
632 struct hlist_head *head = m_hash(mnt, dentry);
635 hlist_for_each_entry_rcu(p, head, mnt_hash)
636 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
642 * find the last mount at @dentry on vfsmount @mnt.
643 * mount_lock must be held.
645 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
647 struct mount *p, *res = NULL;
648 p = __lookup_mnt(mnt, dentry);
651 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
653 hlist_for_each_entry_continue(p, mnt_hash) {
654 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
656 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
664 * lookup_mnt - Return the first child mount mounted at path
666 * "First" means first mounted chronologically. If you create the
669 * mount /dev/sda1 /mnt
670 * mount /dev/sda2 /mnt
671 * mount /dev/sda3 /mnt
673 * Then lookup_mnt() on the base /mnt dentry in the root mount will
674 * return successively the root dentry and vfsmount of /dev/sda1, then
675 * /dev/sda2, then /dev/sda3, then NULL.
677 * lookup_mnt takes a reference to the found vfsmount.
679 struct vfsmount *lookup_mnt(struct path *path)
681 struct mount *child_mnt;
687 seq = read_seqbegin(&mount_lock);
688 child_mnt = __lookup_mnt(path->mnt, path->dentry);
689 m = child_mnt ? &child_mnt->mnt : NULL;
690 } while (!legitimize_mnt(m, seq));
696 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
697 * current mount namespace.
699 * The common case is dentries are not mountpoints at all and that
700 * test is handled inline. For the slow case when we are actually
701 * dealing with a mountpoint of some kind, walk through all of the
702 * mounts in the current mount namespace and test to see if the dentry
705 * The mount_hashtable is not usable in the context because we
706 * need to identify all mounts that may be in the current mount
707 * namespace not just a mount that happens to have some specified
710 bool __is_local_mountpoint(struct dentry *dentry)
712 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
714 bool is_covered = false;
716 if (!d_mountpoint(dentry))
719 down_read(&namespace_sem);
720 list_for_each_entry(mnt, &ns->list, mnt_list) {
721 is_covered = (mnt->mnt_mountpoint == dentry);
725 up_read(&namespace_sem);
730 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
732 struct hlist_head *chain = mp_hash(dentry);
733 struct mountpoint *mp;
735 hlist_for_each_entry(mp, chain, m_hash) {
736 if (mp->m_dentry == dentry) {
737 /* might be worth a WARN_ON() */
738 if (d_unlinked(dentry))
739 return ERR_PTR(-ENOENT);
747 static struct mountpoint *get_mountpoint(struct dentry *dentry)
749 struct mountpoint *mp, *new = NULL;
752 if (d_mountpoint(dentry)) {
754 read_seqlock_excl(&mount_lock);
755 mp = lookup_mountpoint(dentry);
756 read_sequnlock_excl(&mount_lock);
762 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
764 return ERR_PTR(-ENOMEM);
767 /* Exactly one processes may set d_mounted */
768 ret = d_set_mounted(dentry);
770 /* Someone else set d_mounted? */
774 /* The dentry is not available as a mountpoint? */
779 /* Add the new mountpoint to the hash table */
780 read_seqlock_excl(&mount_lock);
781 new->m_dentry = dentry;
783 hlist_add_head(&new->m_hash, mp_hash(dentry));
784 INIT_HLIST_HEAD(&new->m_list);
785 read_sequnlock_excl(&mount_lock);
794 static void put_mountpoint(struct mountpoint *mp)
796 if (!--mp->m_count) {
797 struct dentry *dentry = mp->m_dentry;
798 BUG_ON(!hlist_empty(&mp->m_list));
799 spin_lock(&dentry->d_lock);
800 dentry->d_flags &= ~DCACHE_MOUNTED;
801 spin_unlock(&dentry->d_lock);
802 hlist_del(&mp->m_hash);
807 static inline int check_mnt(struct mount *mnt)
809 return mnt->mnt_ns == current->nsproxy->mnt_ns;
813 * vfsmount lock must be held for write
815 static void touch_mnt_namespace(struct mnt_namespace *ns)
819 wake_up_interruptible(&ns->poll);
824 * vfsmount lock must be held for write
826 static void __touch_mnt_namespace(struct mnt_namespace *ns)
828 if (ns && ns->event != event) {
830 wake_up_interruptible(&ns->poll);
835 * vfsmount lock must be held for write
837 static void unhash_mnt(struct mount *mnt)
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);
844 put_mountpoint(mnt->mnt_mp);
849 * vfsmount lock must be held for write
851 static void detach_mnt(struct mount *mnt, struct path *old_path)
853 old_path->dentry = mnt->mnt_mountpoint;
854 old_path->mnt = &mnt->mnt_parent->mnt;
859 * vfsmount lock must be held for write
861 static void umount_mnt(struct mount *mnt)
863 /* old mountpoint will be dropped when we can do that */
864 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
869 * vfsmount lock must be held for write
871 void mnt_set_mountpoint(struct mount *mnt,
872 struct mountpoint *mp,
873 struct mount *child_mnt)
876 mnt_add_count(mnt, 1); /* essentially, that's mntget */
877 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
878 child_mnt->mnt_parent = mnt;
879 child_mnt->mnt_mp = mp;
880 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
884 * vfsmount lock must be held for write
886 static void attach_mnt(struct mount *mnt,
887 struct mount *parent,
888 struct mountpoint *mp)
890 mnt_set_mountpoint(parent, mp, mnt);
891 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
892 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
895 static void attach_shadowed(struct mount *mnt,
896 struct mount *parent,
897 struct mount *shadows)
900 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
901 list_add(&mnt->mnt_child, &shadows->mnt_child);
903 hlist_add_head_rcu(&mnt->mnt_hash,
904 m_hash(&parent->mnt, mnt->mnt_mountpoint));
905 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
910 * vfsmount lock must be held for write
912 static void commit_tree(struct mount *mnt, struct mount *shadows)
914 struct mount *parent = mnt->mnt_parent;
917 struct mnt_namespace *n = parent->mnt_ns;
919 BUG_ON(parent == mnt);
921 list_add_tail(&head, &mnt->mnt_list);
922 list_for_each_entry(m, &head, mnt_list)
925 list_splice(&head, n->list.prev);
927 attach_shadowed(mnt, parent, shadows);
928 touch_mnt_namespace(n);
931 static struct mount *next_mnt(struct mount *p, struct mount *root)
933 struct list_head *next = p->mnt_mounts.next;
934 if (next == &p->mnt_mounts) {
938 next = p->mnt_child.next;
939 if (next != &p->mnt_parent->mnt_mounts)
944 return list_entry(next, struct mount, mnt_child);
947 static struct mount *skip_mnt_tree(struct mount *p)
949 struct list_head *prev = p->mnt_mounts.prev;
950 while (prev != &p->mnt_mounts) {
951 p = list_entry(prev, struct mount, mnt_child);
952 prev = p->mnt_mounts.prev;
958 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
964 return ERR_PTR(-ENODEV);
966 mnt = alloc_vfsmnt(name);
968 return ERR_PTR(-ENOMEM);
970 mnt->mnt.data = NULL;
971 if (type->alloc_mnt_data) {
972 mnt->mnt.data = type->alloc_mnt_data();
973 if (!mnt->mnt.data) {
976 return ERR_PTR(-ENOMEM);
979 if (flags & MS_KERNMOUNT)
980 mnt->mnt.mnt_flags = MNT_INTERNAL;
982 root = mount_fs(type, flags, name, &mnt->mnt, data);
984 kfree(mnt->mnt.data);
987 return ERR_CAST(root);
990 mnt->mnt.mnt_root = root;
991 mnt->mnt.mnt_sb = root->d_sb;
992 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
993 mnt->mnt_parent = mnt;
995 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
999 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1001 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1004 struct super_block *sb = old->mnt.mnt_sb;
1008 mnt = alloc_vfsmnt(old->mnt_devname);
1010 return ERR_PTR(-ENOMEM);
1012 if (sb->s_op->clone_mnt_data) {
1013 mnt->mnt.data = sb->s_op->clone_mnt_data(old->mnt.data);
1014 if (!mnt->mnt.data) {
1020 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1021 mnt->mnt_group_id = 0; /* not a peer of original */
1023 mnt->mnt_group_id = old->mnt_group_id;
1025 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1026 err = mnt_alloc_group_id(mnt);
1031 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1032 /* Don't allow unprivileged users to change mount flags */
1033 if (flag & CL_UNPRIVILEGED) {
1034 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1036 if (mnt->mnt.mnt_flags & MNT_READONLY)
1037 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1039 if (mnt->mnt.mnt_flags & MNT_NODEV)
1040 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1042 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1043 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1045 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1046 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1049 /* Don't allow unprivileged users to reveal what is under a mount */
1050 if ((flag & CL_UNPRIVILEGED) &&
1051 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1052 mnt->mnt.mnt_flags |= MNT_LOCKED;
1054 atomic_inc(&sb->s_active);
1055 mnt->mnt.mnt_sb = sb;
1056 mnt->mnt.mnt_root = dget(root);
1057 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1058 mnt->mnt_parent = mnt;
1060 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1061 unlock_mount_hash();
1063 if ((flag & CL_SLAVE) ||
1064 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1065 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1066 mnt->mnt_master = old;
1067 CLEAR_MNT_SHARED(mnt);
1068 } else if (!(flag & CL_PRIVATE)) {
1069 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1070 list_add(&mnt->mnt_share, &old->mnt_share);
1071 if (IS_MNT_SLAVE(old))
1072 list_add(&mnt->mnt_slave, &old->mnt_slave);
1073 mnt->mnt_master = old->mnt_master;
1075 if (flag & CL_MAKE_SHARED)
1076 set_mnt_shared(mnt);
1078 /* stick the duplicate mount on the same expiry list
1079 * as the original if that was on one */
1080 if (flag & CL_EXPIRE) {
1081 if (!list_empty(&old->mnt_expire))
1082 list_add(&mnt->mnt_expire, &old->mnt_expire);
1088 kfree(mnt->mnt.data);
1091 return ERR_PTR(err);
1094 static void cleanup_mnt(struct mount *mnt)
1097 * This probably indicates that somebody messed
1098 * up a mnt_want/drop_write() pair. If this
1099 * happens, the filesystem was probably unable
1100 * to make r/w->r/o transitions.
1103 * The locking used to deal with mnt_count decrement provides barriers,
1104 * so mnt_get_writers() below is safe.
1106 WARN_ON(mnt_get_writers(mnt));
1107 if (unlikely(mnt->mnt_pins.first))
1109 fsnotify_vfsmount_delete(&mnt->mnt);
1110 dput(mnt->mnt.mnt_root);
1111 deactivate_super(mnt->mnt.mnt_sb);
1113 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1116 static void __cleanup_mnt(struct rcu_head *head)
1118 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1121 static LLIST_HEAD(delayed_mntput_list);
1122 static void delayed_mntput(struct work_struct *unused)
1124 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1125 struct llist_node *next;
1127 for (; node; node = next) {
1128 next = llist_next(node);
1129 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1132 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1134 static void mntput_no_expire(struct mount *mnt)
1137 mnt_add_count(mnt, -1);
1138 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1143 if (mnt_get_count(mnt)) {
1145 unlock_mount_hash();
1148 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1150 unlock_mount_hash();
1153 mnt->mnt.mnt_flags |= MNT_DOOMED;
1156 list_del(&mnt->mnt_instance);
1158 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1159 struct mount *p, *tmp;
1160 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1164 unlock_mount_hash();
1166 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1167 struct task_struct *task = current;
1168 if (likely(!(task->flags & PF_KTHREAD))) {
1169 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1170 if (!task_work_add(task, &mnt->mnt_rcu, true))
1173 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1174 schedule_delayed_work(&delayed_mntput_work, 1);
1180 void mntput(struct vfsmount *mnt)
1183 struct mount *m = real_mount(mnt);
1184 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1185 if (unlikely(m->mnt_expiry_mark))
1186 m->mnt_expiry_mark = 0;
1187 mntput_no_expire(m);
1190 EXPORT_SYMBOL(mntput);
1192 struct vfsmount *mntget(struct vfsmount *mnt)
1195 mnt_add_count(real_mount(mnt), 1);
1198 EXPORT_SYMBOL(mntget);
1200 struct vfsmount *mnt_clone_internal(struct path *path)
1203 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1206 p->mnt.mnt_flags |= MNT_INTERNAL;
1210 static inline void mangle(struct seq_file *m, const char *s)
1212 seq_escape(m, s, " \t\n\\");
1216 * Simple .show_options callback for filesystems which don't want to
1217 * implement more complex mount option showing.
1219 * See also save_mount_options().
1221 int generic_show_options(struct seq_file *m, struct dentry *root)
1223 const char *options;
1226 options = rcu_dereference(root->d_sb->s_options);
1228 if (options != NULL && options[0]) {
1236 EXPORT_SYMBOL(generic_show_options);
1239 * If filesystem uses generic_show_options(), this function should be
1240 * called from the fill_super() callback.
1242 * The .remount_fs callback usually needs to be handled in a special
1243 * way, to make sure, that previous options are not overwritten if the
1246 * Also note, that if the filesystem's .remount_fs function doesn't
1247 * reset all options to their default value, but changes only newly
1248 * given options, then the displayed options will not reflect reality
1251 void save_mount_options(struct super_block *sb, char *options)
1253 BUG_ON(sb->s_options);
1254 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1256 EXPORT_SYMBOL(save_mount_options);
1258 void replace_mount_options(struct super_block *sb, char *options)
1260 char *old = sb->s_options;
1261 rcu_assign_pointer(sb->s_options, options);
1267 EXPORT_SYMBOL(replace_mount_options);
1269 #ifdef CONFIG_PROC_FS
1270 /* iterator; we want it to have access to namespace_sem, thus here... */
1271 static void *m_start(struct seq_file *m, loff_t *pos)
1273 struct proc_mounts *p = m->private;
1275 down_read(&namespace_sem);
1276 if (p->cached_event == p->ns->event) {
1277 void *v = p->cached_mount;
1278 if (*pos == p->cached_index)
1280 if (*pos == p->cached_index + 1) {
1281 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1282 return p->cached_mount = v;
1286 p->cached_event = p->ns->event;
1287 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1288 p->cached_index = *pos;
1289 return p->cached_mount;
1292 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1294 struct proc_mounts *p = m->private;
1296 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1297 p->cached_index = *pos;
1298 return p->cached_mount;
1301 static void m_stop(struct seq_file *m, void *v)
1303 up_read(&namespace_sem);
1306 static int m_show(struct seq_file *m, void *v)
1308 struct proc_mounts *p = m->private;
1309 struct mount *r = list_entry(v, struct mount, mnt_list);
1310 return p->show(m, &r->mnt);
1313 const struct seq_operations mounts_op = {
1319 #endif /* CONFIG_PROC_FS */
1322 * may_umount_tree - check if a mount tree is busy
1323 * @mnt: root of mount tree
1325 * This is called to check if a tree of mounts has any
1326 * open files, pwds, chroots or sub mounts that are
1329 int may_umount_tree(struct vfsmount *m)
1331 struct mount *mnt = real_mount(m);
1332 int actual_refs = 0;
1333 int minimum_refs = 0;
1337 /* write lock needed for mnt_get_count */
1339 for (p = mnt; p; p = next_mnt(p, mnt)) {
1340 actual_refs += mnt_get_count(p);
1343 unlock_mount_hash();
1345 if (actual_refs > minimum_refs)
1351 EXPORT_SYMBOL(may_umount_tree);
1354 * may_umount - check if a mount point is busy
1355 * @mnt: root of mount
1357 * This is called to check if a mount point has any
1358 * open files, pwds, chroots or sub mounts. If the
1359 * mount has sub mounts this will return busy
1360 * regardless of whether the sub mounts are busy.
1362 * Doesn't take quota and stuff into account. IOW, in some cases it will
1363 * give false negatives. The main reason why it's here is that we need
1364 * a non-destructive way to look for easily umountable filesystems.
1366 int may_umount(struct vfsmount *mnt)
1369 down_read(&namespace_sem);
1371 if (propagate_mount_busy(real_mount(mnt), 2))
1373 unlock_mount_hash();
1374 up_read(&namespace_sem);
1378 EXPORT_SYMBOL(may_umount);
1380 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1382 static void namespace_unlock(void)
1384 struct hlist_head head;
1386 hlist_move_list(&unmounted, &head);
1388 up_write(&namespace_sem);
1390 if (likely(hlist_empty(&head)))
1395 group_pin_kill(&head);
1398 static inline void namespace_lock(void)
1400 down_write(&namespace_sem);
1403 enum umount_tree_flags {
1405 UMOUNT_PROPAGATE = 2,
1406 UMOUNT_CONNECTED = 4,
1409 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1411 /* Leaving mounts connected is only valid for lazy umounts */
1412 if (how & UMOUNT_SYNC)
1415 /* A mount without a parent has nothing to be connected to */
1416 if (!mnt_has_parent(mnt))
1419 /* Because the reference counting rules change when mounts are
1420 * unmounted and connected, umounted mounts may not be
1421 * connected to mounted mounts.
1423 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1426 /* Has it been requested that the mount remain connected? */
1427 if (how & UMOUNT_CONNECTED)
1430 /* Is the mount locked such that it needs to remain connected? */
1431 if (IS_MNT_LOCKED(mnt))
1434 /* By default disconnect the mount */
1439 * mount_lock must be held
1440 * namespace_sem must be held for write
1442 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1444 LIST_HEAD(tmp_list);
1447 if (how & UMOUNT_PROPAGATE)
1448 propagate_mount_unlock(mnt);
1450 /* Gather the mounts to umount */
1451 for (p = mnt; p; p = next_mnt(p, mnt)) {
1452 p->mnt.mnt_flags |= MNT_UMOUNT;
1453 list_move(&p->mnt_list, &tmp_list);
1456 /* Hide the mounts from mnt_mounts */
1457 list_for_each_entry(p, &tmp_list, mnt_list) {
1458 list_del_init(&p->mnt_child);
1461 /* Add propogated mounts to the tmp_list */
1462 if (how & UMOUNT_PROPAGATE)
1463 propagate_umount(&tmp_list);
1465 while (!list_empty(&tmp_list)) {
1467 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1468 list_del_init(&p->mnt_expire);
1469 list_del_init(&p->mnt_list);
1470 __touch_mnt_namespace(p->mnt_ns);
1472 if (how & UMOUNT_SYNC)
1473 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1475 disconnect = disconnect_mount(p, how);
1477 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1478 disconnect ? &unmounted : NULL);
1479 if (mnt_has_parent(p)) {
1480 mnt_add_count(p->mnt_parent, -1);
1482 /* Don't forget about p */
1483 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1488 change_mnt_propagation(p, MS_PRIVATE);
1492 static void shrink_submounts(struct mount *mnt);
1494 static int do_umount(struct mount *mnt, int flags)
1496 struct super_block *sb = mnt->mnt.mnt_sb;
1499 retval = security_sb_umount(&mnt->mnt, flags);
1504 * Allow userspace to request a mountpoint be expired rather than
1505 * unmounting unconditionally. Unmount only happens if:
1506 * (1) the mark is already set (the mark is cleared by mntput())
1507 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1509 if (flags & MNT_EXPIRE) {
1510 if (&mnt->mnt == current->fs->root.mnt ||
1511 flags & (MNT_FORCE | MNT_DETACH))
1515 * probably don't strictly need the lock here if we examined
1516 * all race cases, but it's a slowpath.
1519 if (mnt_get_count(mnt) != 2) {
1520 unlock_mount_hash();
1523 unlock_mount_hash();
1525 if (!xchg(&mnt->mnt_expiry_mark, 1))
1530 * If we may have to abort operations to get out of this
1531 * mount, and they will themselves hold resources we must
1532 * allow the fs to do things. In the Unix tradition of
1533 * 'Gee thats tricky lets do it in userspace' the umount_begin
1534 * might fail to complete on the first run through as other tasks
1535 * must return, and the like. Thats for the mount program to worry
1536 * about for the moment.
1539 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1540 sb->s_op->umount_begin(sb);
1544 * No sense to grab the lock for this test, but test itself looks
1545 * somewhat bogus. Suggestions for better replacement?
1546 * Ho-hum... In principle, we might treat that as umount + switch
1547 * to rootfs. GC would eventually take care of the old vfsmount.
1548 * Actually it makes sense, especially if rootfs would contain a
1549 * /reboot - static binary that would close all descriptors and
1550 * call reboot(9). Then init(8) could umount root and exec /reboot.
1552 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1554 * Special case for "unmounting" root ...
1555 * we just try to remount it readonly.
1557 if (!capable(CAP_SYS_ADMIN))
1559 down_write(&sb->s_umount);
1560 if (!(sb->s_flags & MS_RDONLY))
1561 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1562 up_write(&sb->s_umount);
1570 if (flags & MNT_DETACH) {
1571 if (!list_empty(&mnt->mnt_list))
1572 umount_tree(mnt, UMOUNT_PROPAGATE);
1575 shrink_submounts(mnt);
1577 if (!propagate_mount_busy(mnt, 2)) {
1578 if (!list_empty(&mnt->mnt_list))
1579 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1583 unlock_mount_hash();
1589 * __detach_mounts - lazily unmount all mounts on the specified dentry
1591 * During unlink, rmdir, and d_drop it is possible to loose the path
1592 * to an existing mountpoint, and wind up leaking the mount.
1593 * detach_mounts allows lazily unmounting those mounts instead of
1596 * The caller may hold dentry->d_inode->i_mutex.
1598 void __detach_mounts(struct dentry *dentry)
1600 struct mountpoint *mp;
1605 mp = lookup_mountpoint(dentry);
1606 if (IS_ERR_OR_NULL(mp))
1610 while (!hlist_empty(&mp->m_list)) {
1611 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1612 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1613 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1616 else umount_tree(mnt, UMOUNT_CONNECTED);
1620 unlock_mount_hash();
1625 * Is the caller allowed to modify his namespace?
1627 static inline bool may_mount(void)
1629 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1633 * Now umount can handle mount points as well as block devices.
1634 * This is important for filesystems which use unnamed block devices.
1636 * We now support a flag for forced unmount like the other 'big iron'
1637 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1640 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1645 int lookup_flags = 0;
1647 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1653 if (!(flags & UMOUNT_NOFOLLOW))
1654 lookup_flags |= LOOKUP_FOLLOW;
1656 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1659 mnt = real_mount(path.mnt);
1661 if (path.dentry != path.mnt->mnt_root)
1663 if (!check_mnt(mnt))
1665 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1668 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1671 retval = do_umount(mnt, flags);
1673 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1675 mntput_no_expire(mnt);
1680 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1683 * The 2.0 compatible umount. No flags.
1685 SYSCALL_DEFINE1(oldumount, char __user *, name)
1687 return sys_umount(name, 0);
1692 static bool is_mnt_ns_file(struct dentry *dentry)
1694 /* Is this a proxy for a mount namespace? */
1695 return dentry->d_op == &ns_dentry_operations &&
1696 dentry->d_fsdata == &mntns_operations;
1699 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1701 return container_of(ns, struct mnt_namespace, ns);
1704 static bool mnt_ns_loop(struct dentry *dentry)
1706 /* Could bind mounting the mount namespace inode cause a
1707 * mount namespace loop?
1709 struct mnt_namespace *mnt_ns;
1710 if (!is_mnt_ns_file(dentry))
1713 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1714 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1717 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1720 struct mount *res, *p, *q, *r, *parent;
1722 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1723 return ERR_PTR(-EINVAL);
1725 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1726 return ERR_PTR(-EINVAL);
1728 res = q = clone_mnt(mnt, dentry, flag);
1732 q->mnt_mountpoint = mnt->mnt_mountpoint;
1735 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1737 if (!is_subdir(r->mnt_mountpoint, dentry))
1740 for (s = r; s; s = next_mnt(s, r)) {
1741 struct mount *t = NULL;
1742 if (!(flag & CL_COPY_UNBINDABLE) &&
1743 IS_MNT_UNBINDABLE(s)) {
1744 s = skip_mnt_tree(s);
1747 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1748 is_mnt_ns_file(s->mnt.mnt_root)) {
1749 s = skip_mnt_tree(s);
1752 while (p != s->mnt_parent) {
1758 q = clone_mnt(p, p->mnt.mnt_root, flag);
1762 list_add_tail(&q->mnt_list, &res->mnt_list);
1763 mnt_set_mountpoint(parent, p->mnt_mp, q);
1764 if (!list_empty(&parent->mnt_mounts)) {
1765 t = list_last_entry(&parent->mnt_mounts,
1766 struct mount, mnt_child);
1767 if (t->mnt_mp != p->mnt_mp)
1770 attach_shadowed(q, parent, t);
1771 unlock_mount_hash();
1778 umount_tree(res, UMOUNT_SYNC);
1779 unlock_mount_hash();
1784 /* Caller should check returned pointer for errors */
1786 struct vfsmount *collect_mounts(struct path *path)
1790 if (!check_mnt(real_mount(path->mnt)))
1791 tree = ERR_PTR(-EINVAL);
1793 tree = copy_tree(real_mount(path->mnt), path->dentry,
1794 CL_COPY_ALL | CL_PRIVATE);
1797 return ERR_CAST(tree);
1801 void drop_collected_mounts(struct vfsmount *mnt)
1805 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1806 unlock_mount_hash();
1811 * clone_private_mount - create a private clone of a path
1813 * This creates a new vfsmount, which will be the clone of @path. The new will
1814 * not be attached anywhere in the namespace and will be private (i.e. changes
1815 * to the originating mount won't be propagated into this).
1817 * Release with mntput().
1819 struct vfsmount *clone_private_mount(struct path *path)
1821 struct mount *old_mnt = real_mount(path->mnt);
1822 struct mount *new_mnt;
1824 if (IS_MNT_UNBINDABLE(old_mnt))
1825 return ERR_PTR(-EINVAL);
1827 down_read(&namespace_sem);
1828 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1829 up_read(&namespace_sem);
1830 if (IS_ERR(new_mnt))
1831 return ERR_CAST(new_mnt);
1833 return &new_mnt->mnt;
1835 EXPORT_SYMBOL_GPL(clone_private_mount);
1837 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1838 struct vfsmount *root)
1841 int res = f(root, arg);
1844 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1845 res = f(&mnt->mnt, arg);
1852 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1856 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1857 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1858 mnt_release_group_id(p);
1862 static int invent_group_ids(struct mount *mnt, bool recurse)
1866 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1867 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1868 int err = mnt_alloc_group_id(p);
1870 cleanup_group_ids(mnt, p);
1880 * @source_mnt : mount tree to be attached
1881 * @nd : place the mount tree @source_mnt is attached
1882 * @parent_nd : if non-null, detach the source_mnt from its parent and
1883 * store the parent mount and mountpoint dentry.
1884 * (done when source_mnt is moved)
1886 * NOTE: in the table below explains the semantics when a source mount
1887 * of a given type is attached to a destination mount of a given type.
1888 * ---------------------------------------------------------------------------
1889 * | BIND MOUNT OPERATION |
1890 * |**************************************************************************
1891 * | source-->| shared | private | slave | unbindable |
1895 * |**************************************************************************
1896 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1898 * |non-shared| shared (+) | private | slave (*) | invalid |
1899 * ***************************************************************************
1900 * A bind operation clones the source mount and mounts the clone on the
1901 * destination mount.
1903 * (++) the cloned mount is propagated to all the mounts in the propagation
1904 * tree of the destination mount and the cloned mount is added to
1905 * the peer group of the source mount.
1906 * (+) the cloned mount is created under the destination mount and is marked
1907 * as shared. The cloned mount is added to the peer group of the source
1909 * (+++) the mount is propagated to all the mounts in the propagation tree
1910 * of the destination mount and the cloned mount is made slave
1911 * of the same master as that of the source mount. The cloned mount
1912 * is marked as 'shared and slave'.
1913 * (*) the cloned mount is made a slave of the same master as that of the
1916 * ---------------------------------------------------------------------------
1917 * | MOVE MOUNT OPERATION |
1918 * |**************************************************************************
1919 * | source-->| shared | private | slave | unbindable |
1923 * |**************************************************************************
1924 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1926 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1927 * ***************************************************************************
1929 * (+) the mount is moved to the destination. And is then propagated to
1930 * all the mounts in the propagation tree of the destination mount.
1931 * (+*) the mount is moved to the destination.
1932 * (+++) the mount is moved to the destination and is then propagated to
1933 * all the mounts belonging to the destination mount's propagation tree.
1934 * the mount is marked as 'shared and slave'.
1935 * (*) the mount continues to be a slave at the new location.
1937 * if the source mount is a tree, the operations explained above is
1938 * applied to each mount in the tree.
1939 * Must be called without spinlocks held, since this function can sleep
1942 static int attach_recursive_mnt(struct mount *source_mnt,
1943 struct mount *dest_mnt,
1944 struct mountpoint *dest_mp,
1945 struct path *parent_path)
1947 HLIST_HEAD(tree_list);
1948 struct mount *child, *p;
1949 struct hlist_node *n;
1952 if (IS_MNT_SHARED(dest_mnt)) {
1953 err = invent_group_ids(source_mnt, true);
1956 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1959 goto out_cleanup_ids;
1960 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1966 detach_mnt(source_mnt, parent_path);
1967 attach_mnt(source_mnt, dest_mnt, dest_mp);
1968 touch_mnt_namespace(source_mnt->mnt_ns);
1970 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1971 commit_tree(source_mnt, NULL);
1974 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1976 hlist_del_init(&child->mnt_hash);
1977 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1978 child->mnt_mountpoint);
1979 commit_tree(child, q);
1981 unlock_mount_hash();
1986 while (!hlist_empty(&tree_list)) {
1987 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1988 umount_tree(child, UMOUNT_SYNC);
1990 unlock_mount_hash();
1991 cleanup_group_ids(source_mnt, NULL);
1996 static struct mountpoint *lock_mount(struct path *path)
1998 struct vfsmount *mnt;
1999 struct dentry *dentry = path->dentry;
2001 mutex_lock(&dentry->d_inode->i_mutex);
2002 if (unlikely(cant_mount(dentry))) {
2003 mutex_unlock(&dentry->d_inode->i_mutex);
2004 return ERR_PTR(-ENOENT);
2007 mnt = lookup_mnt(path);
2009 struct mountpoint *mp = get_mountpoint(dentry);
2012 mutex_unlock(&dentry->d_inode->i_mutex);
2018 mutex_unlock(&path->dentry->d_inode->i_mutex);
2021 dentry = path->dentry = dget(mnt->mnt_root);
2025 static void unlock_mount(struct mountpoint *where)
2027 struct dentry *dentry = where->m_dentry;
2029 read_seqlock_excl(&mount_lock);
2030 put_mountpoint(where);
2031 read_sequnlock_excl(&mount_lock);
2034 mutex_unlock(&dentry->d_inode->i_mutex);
2037 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2039 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2042 if (d_is_dir(mp->m_dentry) !=
2043 d_is_dir(mnt->mnt.mnt_root))
2046 return attach_recursive_mnt(mnt, p, mp, NULL);
2050 * Sanity check the flags to change_mnt_propagation.
2053 static int flags_to_propagation_type(int flags)
2055 int type = flags & ~(MS_REC | MS_SILENT);
2057 /* Fail if any non-propagation flags are set */
2058 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2060 /* Only one propagation flag should be set */
2061 if (!is_power_of_2(type))
2067 * recursively change the type of the mountpoint.
2069 static int do_change_type(struct path *path, int flag)
2072 struct mount *mnt = real_mount(path->mnt);
2073 int recurse = flag & MS_REC;
2077 if (path->dentry != path->mnt->mnt_root)
2080 type = flags_to_propagation_type(flag);
2085 if (type == MS_SHARED) {
2086 err = invent_group_ids(mnt, recurse);
2092 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2093 change_mnt_propagation(m, type);
2094 unlock_mount_hash();
2101 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2103 struct mount *child;
2104 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2105 if (!is_subdir(child->mnt_mountpoint, dentry))
2108 if (child->mnt.mnt_flags & MNT_LOCKED)
2115 * do loopback mount.
2117 static int do_loopback(struct path *path, const char *old_name,
2120 struct path old_path;
2121 struct mount *mnt = NULL, *old, *parent;
2122 struct mountpoint *mp;
2124 if (!old_name || !*old_name)
2126 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2131 if (mnt_ns_loop(old_path.dentry))
2134 mp = lock_mount(path);
2139 old = real_mount(old_path.mnt);
2140 parent = real_mount(path->mnt);
2143 if (IS_MNT_UNBINDABLE(old))
2146 if (!check_mnt(parent))
2149 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2152 if (!recurse && has_locked_children(old, old_path.dentry))
2156 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2158 mnt = clone_mnt(old, old_path.dentry, 0);
2165 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2167 err = graft_tree(mnt, parent, mp);
2170 umount_tree(mnt, UMOUNT_SYNC);
2171 unlock_mount_hash();
2176 path_put(&old_path);
2180 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2183 int readonly_request = 0;
2185 if (ms_flags & MS_RDONLY)
2186 readonly_request = 1;
2187 if (readonly_request == __mnt_is_readonly(mnt))
2190 if (readonly_request)
2191 error = mnt_make_readonly(real_mount(mnt));
2193 __mnt_unmake_readonly(real_mount(mnt));
2198 * change filesystem flags. dir should be a physical root of filesystem.
2199 * If you've mounted a non-root directory somewhere and want to do remount
2200 * on it - tough luck.
2202 static int do_remount(struct path *path, int flags, int mnt_flags,
2206 struct super_block *sb = path->mnt->mnt_sb;
2207 struct mount *mnt = real_mount(path->mnt);
2209 if (!check_mnt(mnt))
2212 if (path->dentry != path->mnt->mnt_root)
2215 /* Don't allow changing of locked mnt flags.
2217 * No locks need to be held here while testing the various
2218 * MNT_LOCK flags because those flags can never be cleared
2219 * once they are set.
2221 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2222 !(mnt_flags & MNT_READONLY)) {
2225 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2226 !(mnt_flags & MNT_NODEV)) {
2227 /* Was the nodev implicitly added in mount? */
2228 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2229 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2230 mnt_flags |= MNT_NODEV;
2235 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2236 !(mnt_flags & MNT_NOSUID)) {
2239 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2240 !(mnt_flags & MNT_NOEXEC)) {
2243 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2244 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2248 err = security_sb_remount(sb, data);
2252 down_write(&sb->s_umount);
2253 if (flags & MS_BIND)
2254 err = change_mount_flags(path->mnt, flags);
2255 else if (!capable(CAP_SYS_ADMIN))
2258 err = do_remount_sb2(path->mnt, sb, flags, data, 0);
2261 propagate_remount(mnt);
2262 unlock_mount_hash();
2267 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2268 mnt->mnt.mnt_flags = mnt_flags;
2269 touch_mnt_namespace(mnt->mnt_ns);
2270 unlock_mount_hash();
2272 up_write(&sb->s_umount);
2276 static inline int tree_contains_unbindable(struct mount *mnt)
2279 for (p = mnt; p; p = next_mnt(p, mnt)) {
2280 if (IS_MNT_UNBINDABLE(p))
2286 static int do_move_mount(struct path *path, const char *old_name)
2288 struct path old_path, parent_path;
2291 struct mountpoint *mp;
2293 if (!old_name || !*old_name)
2295 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2299 mp = lock_mount(path);
2304 old = real_mount(old_path.mnt);
2305 p = real_mount(path->mnt);
2308 if (!check_mnt(p) || !check_mnt(old))
2311 if (old->mnt.mnt_flags & MNT_LOCKED)
2315 if (old_path.dentry != old_path.mnt->mnt_root)
2318 if (!mnt_has_parent(old))
2321 if (d_is_dir(path->dentry) !=
2322 d_is_dir(old_path.dentry))
2325 * Don't move a mount residing in a shared parent.
2327 if (IS_MNT_SHARED(old->mnt_parent))
2330 * Don't move a mount tree containing unbindable mounts to a destination
2331 * mount which is shared.
2333 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2336 for (; mnt_has_parent(p); p = p->mnt_parent)
2340 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2344 /* if the mount is moved, it should no longer be expire
2346 list_del_init(&old->mnt_expire);
2351 path_put(&parent_path);
2352 path_put(&old_path);
2356 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2359 const char *subtype = strchr(fstype, '.');
2368 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2370 if (!mnt->mnt_sb->s_subtype)
2376 return ERR_PTR(err);
2380 * add a mount into a namespace's mount tree
2382 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2384 struct mountpoint *mp;
2385 struct mount *parent;
2388 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2390 mp = lock_mount(path);
2394 parent = real_mount(path->mnt);
2396 if (unlikely(!check_mnt(parent))) {
2397 /* that's acceptable only for automounts done in private ns */
2398 if (!(mnt_flags & MNT_SHRINKABLE))
2400 /* ... and for those we'd better have mountpoint still alive */
2401 if (!parent->mnt_ns)
2405 /* Refuse the same filesystem on the same mount point */
2407 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2408 path->mnt->mnt_root == path->dentry)
2412 if (d_is_symlink(newmnt->mnt.mnt_root))
2415 newmnt->mnt.mnt_flags = mnt_flags;
2416 err = graft_tree(newmnt, parent, mp);
2423 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2426 * create a new mount for userspace and request it to be added into the
2429 static int do_new_mount(struct path *path, const char *fstype, int flags,
2430 int mnt_flags, const char *name, void *data)
2432 struct file_system_type *type;
2433 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2434 struct vfsmount *mnt;
2440 type = get_fs_type(fstype);
2444 if (user_ns != &init_user_ns) {
2445 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2446 put_filesystem(type);
2449 /* Only in special cases allow devices from mounts
2450 * created outside the initial user namespace.
2452 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2454 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2456 if (type->fs_flags & FS_USERNS_VISIBLE) {
2457 if (!fs_fully_visible(type, &mnt_flags)) {
2458 put_filesystem(type);
2464 mnt = vfs_kern_mount(type, flags, name, data);
2465 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2466 !mnt->mnt_sb->s_subtype)
2467 mnt = fs_set_subtype(mnt, fstype);
2469 put_filesystem(type);
2471 return PTR_ERR(mnt);
2473 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2479 int finish_automount(struct vfsmount *m, struct path *path)
2481 struct mount *mnt = real_mount(m);
2483 /* The new mount record should have at least 2 refs to prevent it being
2484 * expired before we get a chance to add it
2486 BUG_ON(mnt_get_count(mnt) < 2);
2488 if (m->mnt_sb == path->mnt->mnt_sb &&
2489 m->mnt_root == path->dentry) {
2494 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2498 /* remove m from any expiration list it may be on */
2499 if (!list_empty(&mnt->mnt_expire)) {
2501 list_del_init(&mnt->mnt_expire);
2510 * mnt_set_expiry - Put a mount on an expiration list
2511 * @mnt: The mount to list.
2512 * @expiry_list: The list to add the mount to.
2514 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2518 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2522 EXPORT_SYMBOL(mnt_set_expiry);
2525 * process a list of expirable mountpoints with the intent of discarding any
2526 * mountpoints that aren't in use and haven't been touched since last we came
2529 void mark_mounts_for_expiry(struct list_head *mounts)
2531 struct mount *mnt, *next;
2532 LIST_HEAD(graveyard);
2534 if (list_empty(mounts))
2540 /* extract from the expiration list every vfsmount that matches the
2541 * following criteria:
2542 * - only referenced by its parent vfsmount
2543 * - still marked for expiry (marked on the last call here; marks are
2544 * cleared by mntput())
2546 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2547 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2548 propagate_mount_busy(mnt, 1))
2550 list_move(&mnt->mnt_expire, &graveyard);
2552 while (!list_empty(&graveyard)) {
2553 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2554 touch_mnt_namespace(mnt->mnt_ns);
2555 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2557 unlock_mount_hash();
2561 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2564 * Ripoff of 'select_parent()'
2566 * search the list of submounts for a given mountpoint, and move any
2567 * shrinkable submounts to the 'graveyard' list.
2569 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2571 struct mount *this_parent = parent;
2572 struct list_head *next;
2576 next = this_parent->mnt_mounts.next;
2578 while (next != &this_parent->mnt_mounts) {
2579 struct list_head *tmp = next;
2580 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2583 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2586 * Descend a level if the d_mounts list is non-empty.
2588 if (!list_empty(&mnt->mnt_mounts)) {
2593 if (!propagate_mount_busy(mnt, 1)) {
2594 list_move_tail(&mnt->mnt_expire, graveyard);
2599 * All done at this level ... ascend and resume the search
2601 if (this_parent != parent) {
2602 next = this_parent->mnt_child.next;
2603 this_parent = this_parent->mnt_parent;
2610 * process a list of expirable mountpoints with the intent of discarding any
2611 * submounts of a specific parent mountpoint
2613 * mount_lock must be held for write
2615 static void shrink_submounts(struct mount *mnt)
2617 LIST_HEAD(graveyard);
2620 /* extract submounts of 'mountpoint' from the expiration list */
2621 while (select_submounts(mnt, &graveyard)) {
2622 while (!list_empty(&graveyard)) {
2623 m = list_first_entry(&graveyard, struct mount,
2625 touch_mnt_namespace(m->mnt_ns);
2626 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2632 * Some copy_from_user() implementations do not return the exact number of
2633 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2634 * Note that this function differs from copy_from_user() in that it will oops
2635 * on bad values of `to', rather than returning a short copy.
2637 static long exact_copy_from_user(void *to, const void __user * from,
2641 const char __user *f = from;
2644 if (!access_ok(VERIFY_READ, from, n))
2648 if (__get_user(c, f)) {
2659 int copy_mount_options(const void __user * data, unsigned long *where)
2669 if (!(page = __get_free_page(GFP_KERNEL)))
2672 /* We only care that *some* data at the address the user
2673 * gave us is valid. Just in case, we'll zero
2674 * the remainder of the page.
2676 /* copy_from_user cannot cross TASK_SIZE ! */
2677 size = TASK_SIZE - (unsigned long)data;
2678 if (size > PAGE_SIZE)
2681 i = size - exact_copy_from_user((void *)page, data, size);
2687 memset((char *)page + i, 0, PAGE_SIZE - i);
2692 char *copy_mount_string(const void __user *data)
2694 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2698 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2699 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2701 * data is a (void *) that can point to any structure up to
2702 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2703 * information (or be NULL).
2705 * Pre-0.97 versions of mount() didn't have a flags word.
2706 * When the flags word was introduced its top half was required
2707 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2708 * Therefore, if this magic number is present, it carries no information
2709 * and must be discarded.
2711 long do_mount(const char *dev_name, const char __user *dir_name,
2712 const char *type_page, unsigned long flags, void *data_page)
2719 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2720 flags &= ~MS_MGC_MSK;
2722 /* Basic sanity checks */
2724 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2726 /* ... and get the mountpoint */
2727 retval = user_path(dir_name, &path);
2731 retval = security_sb_mount(dev_name, &path,
2732 type_page, flags, data_page);
2733 if (!retval && !may_mount())
2738 /* Default to relatime unless overriden */
2739 if (!(flags & MS_NOATIME))
2740 mnt_flags |= MNT_RELATIME;
2742 /* Separate the per-mountpoint flags */
2743 if (flags & MS_NOSUID)
2744 mnt_flags |= MNT_NOSUID;
2745 if (flags & MS_NODEV)
2746 mnt_flags |= MNT_NODEV;
2747 if (flags & MS_NOEXEC)
2748 mnt_flags |= MNT_NOEXEC;
2749 if (flags & MS_NOATIME)
2750 mnt_flags |= MNT_NOATIME;
2751 if (flags & MS_NODIRATIME)
2752 mnt_flags |= MNT_NODIRATIME;
2753 if (flags & MS_STRICTATIME)
2754 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2755 if (flags & MS_RDONLY)
2756 mnt_flags |= MNT_READONLY;
2758 /* The default atime for remount is preservation */
2759 if ((flags & MS_REMOUNT) &&
2760 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2761 MS_STRICTATIME)) == 0)) {
2762 mnt_flags &= ~MNT_ATIME_MASK;
2763 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2766 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2767 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2770 if (flags & MS_REMOUNT)
2771 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2773 else if (flags & MS_BIND)
2774 retval = do_loopback(&path, dev_name, flags & MS_REC);
2775 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2776 retval = do_change_type(&path, flags);
2777 else if (flags & MS_MOVE)
2778 retval = do_move_mount(&path, dev_name);
2780 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2781 dev_name, data_page);
2787 static void free_mnt_ns(struct mnt_namespace *ns)
2789 ns_free_inum(&ns->ns);
2790 put_user_ns(ns->user_ns);
2795 * Assign a sequence number so we can detect when we attempt to bind
2796 * mount a reference to an older mount namespace into the current
2797 * mount namespace, preventing reference counting loops. A 64bit
2798 * number incrementing at 10Ghz will take 12,427 years to wrap which
2799 * is effectively never, so we can ignore the possibility.
2801 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2803 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2805 struct mnt_namespace *new_ns;
2808 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2810 return ERR_PTR(-ENOMEM);
2811 ret = ns_alloc_inum(&new_ns->ns);
2814 return ERR_PTR(ret);
2816 new_ns->ns.ops = &mntns_operations;
2817 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2818 atomic_set(&new_ns->count, 1);
2819 new_ns->root = NULL;
2820 INIT_LIST_HEAD(&new_ns->list);
2821 init_waitqueue_head(&new_ns->poll);
2823 new_ns->user_ns = get_user_ns(user_ns);
2827 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2828 struct user_namespace *user_ns, struct fs_struct *new_fs)
2830 struct mnt_namespace *new_ns;
2831 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2832 struct mount *p, *q;
2839 if (likely(!(flags & CLONE_NEWNS))) {
2846 new_ns = alloc_mnt_ns(user_ns);
2851 /* First pass: copy the tree topology */
2852 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2853 if (user_ns != ns->user_ns)
2854 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2855 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2858 free_mnt_ns(new_ns);
2859 return ERR_CAST(new);
2862 list_add_tail(&new_ns->list, &new->mnt_list);
2865 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2866 * as belonging to new namespace. We have already acquired a private
2867 * fs_struct, so tsk->fs->lock is not needed.
2874 if (&p->mnt == new_fs->root.mnt) {
2875 new_fs->root.mnt = mntget(&q->mnt);
2878 if (&p->mnt == new_fs->pwd.mnt) {
2879 new_fs->pwd.mnt = mntget(&q->mnt);
2883 p = next_mnt(p, old);
2884 q = next_mnt(q, new);
2887 while (p->mnt.mnt_root != q->mnt.mnt_root)
2888 p = next_mnt(p, old);
2901 * create_mnt_ns - creates a private namespace and adds a root filesystem
2902 * @mnt: pointer to the new root filesystem mountpoint
2904 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2906 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2907 if (!IS_ERR(new_ns)) {
2908 struct mount *mnt = real_mount(m);
2909 mnt->mnt_ns = new_ns;
2911 list_add(&mnt->mnt_list, &new_ns->list);
2918 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2920 struct mnt_namespace *ns;
2921 struct super_block *s;
2925 ns = create_mnt_ns(mnt);
2927 return ERR_CAST(ns);
2929 err = vfs_path_lookup(mnt->mnt_root, mnt,
2930 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2935 return ERR_PTR(err);
2937 /* trade a vfsmount reference for active sb one */
2938 s = path.mnt->mnt_sb;
2939 atomic_inc(&s->s_active);
2941 /* lock the sucker */
2942 down_write(&s->s_umount);
2943 /* ... and return the root of (sub)tree on it */
2946 EXPORT_SYMBOL(mount_subtree);
2948 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2949 char __user *, type, unsigned long, flags, void __user *, data)
2954 unsigned long data_page;
2956 kernel_type = copy_mount_string(type);
2957 ret = PTR_ERR(kernel_type);
2958 if (IS_ERR(kernel_type))
2961 kernel_dev = copy_mount_string(dev_name);
2962 ret = PTR_ERR(kernel_dev);
2963 if (IS_ERR(kernel_dev))
2966 ret = copy_mount_options(data, &data_page);
2970 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2971 (void *) data_page);
2973 free_page(data_page);
2983 * Return true if path is reachable from root
2985 * namespace_sem or mount_lock is held
2987 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2988 const struct path *root)
2990 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2991 dentry = mnt->mnt_mountpoint;
2992 mnt = mnt->mnt_parent;
2994 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2997 int path_is_under(struct path *path1, struct path *path2)
3000 read_seqlock_excl(&mount_lock);
3001 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3002 read_sequnlock_excl(&mount_lock);
3005 EXPORT_SYMBOL(path_is_under);
3008 * pivot_root Semantics:
3009 * Moves the root file system of the current process to the directory put_old,
3010 * makes new_root as the new root file system of the current process, and sets
3011 * root/cwd of all processes which had them on the current root to new_root.
3014 * The new_root and put_old must be directories, and must not be on the
3015 * same file system as the current process root. The put_old must be
3016 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3017 * pointed to by put_old must yield the same directory as new_root. No other
3018 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3020 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3021 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3022 * in this situation.
3025 * - we don't move root/cwd if they are not at the root (reason: if something
3026 * cared enough to change them, it's probably wrong to force them elsewhere)
3027 * - it's okay to pick a root that isn't the root of a file system, e.g.
3028 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3029 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3032 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3033 const char __user *, put_old)
3035 struct path new, old, parent_path, root_parent, root;
3036 struct mount *new_mnt, *root_mnt, *old_mnt;
3037 struct mountpoint *old_mp, *root_mp;
3043 error = user_path_dir(new_root, &new);
3047 error = user_path_dir(put_old, &old);
3051 error = security_sb_pivotroot(&old, &new);
3055 get_fs_root(current->fs, &root);
3056 old_mp = lock_mount(&old);
3057 error = PTR_ERR(old_mp);
3062 new_mnt = real_mount(new.mnt);
3063 root_mnt = real_mount(root.mnt);
3064 old_mnt = real_mount(old.mnt);
3065 if (IS_MNT_SHARED(old_mnt) ||
3066 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3067 IS_MNT_SHARED(root_mnt->mnt_parent))
3069 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3071 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3074 if (d_unlinked(new.dentry))
3077 if (new_mnt == root_mnt || old_mnt == root_mnt)
3078 goto out4; /* loop, on the same file system */
3080 if (root.mnt->mnt_root != root.dentry)
3081 goto out4; /* not a mountpoint */
3082 if (!mnt_has_parent(root_mnt))
3083 goto out4; /* not attached */
3084 root_mp = root_mnt->mnt_mp;
3085 if (new.mnt->mnt_root != new.dentry)
3086 goto out4; /* not a mountpoint */
3087 if (!mnt_has_parent(new_mnt))
3088 goto out4; /* not attached */
3089 /* make sure we can reach put_old from new_root */
3090 if (!is_path_reachable(old_mnt, old.dentry, &new))
3092 /* make certain new is below the root */
3093 if (!is_path_reachable(new_mnt, new.dentry, &root))
3095 root_mp->m_count++; /* pin it so it won't go away */
3097 detach_mnt(new_mnt, &parent_path);
3098 detach_mnt(root_mnt, &root_parent);
3099 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3100 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3101 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3103 /* mount old root on put_old */
3104 attach_mnt(root_mnt, old_mnt, old_mp);
3105 /* mount new_root on / */
3106 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3107 touch_mnt_namespace(current->nsproxy->mnt_ns);
3108 /* A moved mount should not expire automatically */
3109 list_del_init(&new_mnt->mnt_expire);
3110 put_mountpoint(root_mp);
3111 unlock_mount_hash();
3112 chroot_fs_refs(&root, &new);
3115 unlock_mount(old_mp);
3117 path_put(&root_parent);
3118 path_put(&parent_path);
3130 static void __init init_mount_tree(void)
3132 struct vfsmount *mnt;
3133 struct mnt_namespace *ns;
3135 struct file_system_type *type;
3137 type = get_fs_type("rootfs");
3139 panic("Can't find rootfs type");
3140 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3141 put_filesystem(type);
3143 panic("Can't create rootfs");
3145 ns = create_mnt_ns(mnt);
3147 panic("Can't allocate initial namespace");
3149 init_task.nsproxy->mnt_ns = ns;
3153 root.dentry = mnt->mnt_root;
3154 mnt->mnt_flags |= MNT_LOCKED;
3156 set_fs_pwd(current->fs, &root);
3157 set_fs_root(current->fs, &root);
3160 void __init mnt_init(void)
3165 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3166 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3168 mount_hashtable = alloc_large_system_hash("Mount-cache",
3169 sizeof(struct hlist_head),
3172 &m_hash_shift, &m_hash_mask, 0, 0);
3173 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3174 sizeof(struct hlist_head),
3177 &mp_hash_shift, &mp_hash_mask, 0, 0);
3179 if (!mount_hashtable || !mountpoint_hashtable)
3180 panic("Failed to allocate mount hash table\n");
3182 for (u = 0; u <= m_hash_mask; u++)
3183 INIT_HLIST_HEAD(&mount_hashtable[u]);
3184 for (u = 0; u <= mp_hash_mask; u++)
3185 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3191 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3193 fs_kobj = kobject_create_and_add("fs", NULL);
3195 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3200 void put_mnt_ns(struct mnt_namespace *ns)
3202 if (!atomic_dec_and_test(&ns->count))
3204 drop_collected_mounts(&ns->root->mnt);
3208 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3210 struct vfsmount *mnt;
3211 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3214 * it is a longterm mount, don't release mnt until
3215 * we unmount before file sys is unregistered
3217 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3221 EXPORT_SYMBOL_GPL(kern_mount_data);
3223 void kern_unmount(struct vfsmount *mnt)
3225 /* release long term mount so mount point can be released */
3226 if (!IS_ERR_OR_NULL(mnt)) {
3227 real_mount(mnt)->mnt_ns = NULL;
3228 synchronize_rcu(); /* yecchhh... */
3232 EXPORT_SYMBOL(kern_unmount);
3234 bool our_mnt(struct vfsmount *mnt)
3236 return check_mnt(real_mount(mnt));
3239 bool current_chrooted(void)
3241 /* Does the current process have a non-standard root */
3242 struct path ns_root;
3243 struct path fs_root;
3246 /* Find the namespace root */
3247 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3248 ns_root.dentry = ns_root.mnt->mnt_root;
3250 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3253 get_fs_root(current->fs, &fs_root);
3255 chrooted = !path_equal(&fs_root, &ns_root);
3263 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3265 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3266 int new_flags = *new_mnt_flags;
3268 bool visible = false;
3273 down_read(&namespace_sem);
3274 list_for_each_entry(mnt, &ns->list, mnt_list) {
3275 struct mount *child;
3278 if (mnt->mnt.mnt_sb->s_type != type)
3281 /* This mount is not fully visible if it's root directory
3282 * is not the root directory of the filesystem.
3284 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3287 /* Read the mount flags and filter out flags that
3288 * may safely be ignored.
3290 mnt_flags = mnt->mnt.mnt_flags;
3291 if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3292 mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3294 /* Don't miss readonly hidden in the superblock flags */
3295 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3296 mnt_flags |= MNT_LOCK_READONLY;
3298 /* Verify the mount flags are equal to or more permissive
3299 * than the proposed new mount.
3301 if ((mnt_flags & MNT_LOCK_READONLY) &&
3302 !(new_flags & MNT_READONLY))
3304 if ((mnt_flags & MNT_LOCK_NODEV) &&
3305 !(new_flags & MNT_NODEV))
3307 if ((mnt_flags & MNT_LOCK_NOSUID) &&
3308 !(new_flags & MNT_NOSUID))
3310 if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3311 !(new_flags & MNT_NOEXEC))
3313 if ((mnt_flags & MNT_LOCK_ATIME) &&
3314 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3317 /* This mount is not fully visible if there are any
3318 * locked child mounts that cover anything except for
3319 * empty directories.
3321 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3322 struct inode *inode = child->mnt_mountpoint->d_inode;
3323 /* Only worry about locked mounts */
3324 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3326 /* Is the directory permanetly empty? */
3327 if (!is_empty_dir_inode(inode))
3330 /* Preserve the locked attributes */
3331 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3341 up_read(&namespace_sem);
3345 static struct ns_common *mntns_get(struct task_struct *task)
3347 struct ns_common *ns = NULL;
3348 struct nsproxy *nsproxy;
3351 nsproxy = task->nsproxy;
3353 ns = &nsproxy->mnt_ns->ns;
3354 get_mnt_ns(to_mnt_ns(ns));
3361 static void mntns_put(struct ns_common *ns)
3363 put_mnt_ns(to_mnt_ns(ns));
3366 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3368 struct fs_struct *fs = current->fs;
3369 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3372 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3373 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3374 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3381 put_mnt_ns(nsproxy->mnt_ns);
3382 nsproxy->mnt_ns = mnt_ns;
3385 root.mnt = &mnt_ns->root->mnt;
3386 root.dentry = mnt_ns->root->mnt.mnt_root;
3388 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3391 /* Update the pwd and root */
3392 set_fs_pwd(fs, &root);
3393 set_fs_root(fs, &root);
3399 const struct proc_ns_operations mntns_operations = {
3401 .type = CLONE_NEWNS,
3404 .install = mntns_install,