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/delay.h>
18 #include <linux/security.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/bootmem.h>
27 #include <linux/task_work.h>
31 static unsigned int m_hash_mask __read_mostly;
32 static unsigned int m_hash_shift __read_mostly;
33 static unsigned int mp_hash_mask __read_mostly;
34 static unsigned int mp_hash_shift __read_mostly;
36 static __initdata unsigned long mhash_entries;
37 static int __init set_mhash_entries(char *str)
41 mhash_entries = simple_strtoul(str, &str, 0);
44 __setup("mhash_entries=", set_mhash_entries);
46 static __initdata unsigned long mphash_entries;
47 static int __init set_mphash_entries(char *str)
51 mphash_entries = simple_strtoul(str, &str, 0);
54 __setup("mphash_entries=", set_mphash_entries);
57 static DEFINE_IDA(mnt_id_ida);
58 static DEFINE_IDA(mnt_group_ida);
59 static DEFINE_SPINLOCK(mnt_id_lock);
60 static int mnt_id_start = 0;
61 static int mnt_group_start = 1;
63 static struct hlist_head *mount_hashtable __read_mostly;
64 static struct hlist_head *mountpoint_hashtable __read_mostly;
65 static struct kmem_cache *mnt_cache __read_mostly;
66 static DECLARE_RWSEM(namespace_sem);
69 struct kobject *fs_kobj;
70 EXPORT_SYMBOL_GPL(fs_kobj);
73 * vfsmount lock may be taken for read to prevent changes to the
74 * vfsmount hash, ie. during mountpoint lookups or walking back
77 * It should be taken for write in all cases where the vfsmount
78 * tree or hash is modified or when a vfsmount structure is modified.
80 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
82 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
84 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
85 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
86 tmp = tmp + (tmp >> m_hash_shift);
87 return &mount_hashtable[tmp & m_hash_mask];
90 static inline struct hlist_head *mp_hash(struct dentry *dentry)
92 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
93 tmp = tmp + (tmp >> mp_hash_shift);
94 return &mountpoint_hashtable[tmp & mp_hash_mask];
98 * allocation is serialized by namespace_sem, but we need the spinlock to
99 * serialize with freeing.
101 static int mnt_alloc_id(struct mount *mnt)
106 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
107 spin_lock(&mnt_id_lock);
108 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
110 mnt_id_start = mnt->mnt_id + 1;
111 spin_unlock(&mnt_id_lock);
118 static void mnt_free_id(struct mount *mnt)
120 int id = mnt->mnt_id;
121 spin_lock(&mnt_id_lock);
122 ida_remove(&mnt_id_ida, id);
123 if (mnt_id_start > id)
125 spin_unlock(&mnt_id_lock);
129 * Allocate a new peer group ID
131 * mnt_group_ida is protected by namespace_sem
133 static int mnt_alloc_group_id(struct mount *mnt)
137 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
140 res = ida_get_new_above(&mnt_group_ida,
144 mnt_group_start = mnt->mnt_group_id + 1;
150 * Release a peer group ID
152 void mnt_release_group_id(struct mount *mnt)
154 int id = mnt->mnt_group_id;
155 ida_remove(&mnt_group_ida, id);
156 if (mnt_group_start > id)
157 mnt_group_start = id;
158 mnt->mnt_group_id = 0;
162 * vfsmount lock must be held for read
164 static inline void mnt_add_count(struct mount *mnt, int n)
167 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
176 * vfsmount lock must be held for write
178 unsigned int mnt_get_count(struct mount *mnt)
181 unsigned int count = 0;
184 for_each_possible_cpu(cpu) {
185 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
190 return mnt->mnt_count;
194 static void drop_mountpoint(struct fs_pin *p)
196 struct mount *m = container_of(p, struct mount, mnt_umount);
197 dput(m->mnt_ex_mountpoint);
202 static struct mount *alloc_vfsmnt(const char *name)
204 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
208 err = mnt_alloc_id(mnt);
213 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
214 if (!mnt->mnt_devname)
219 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 goto out_free_devname;
223 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
226 mnt->mnt_writers = 0;
229 INIT_HLIST_NODE(&mnt->mnt_hash);
230 INIT_LIST_HEAD(&mnt->mnt_child);
231 INIT_LIST_HEAD(&mnt->mnt_mounts);
232 INIT_LIST_HEAD(&mnt->mnt_list);
233 INIT_LIST_HEAD(&mnt->mnt_expire);
234 INIT_LIST_HEAD(&mnt->mnt_share);
235 INIT_LIST_HEAD(&mnt->mnt_slave_list);
236 INIT_LIST_HEAD(&mnt->mnt_slave);
237 INIT_HLIST_NODE(&mnt->mnt_mp_list);
238 #ifdef CONFIG_FSNOTIFY
239 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
241 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
247 kfree_const(mnt->mnt_devname);
252 kmem_cache_free(mnt_cache, mnt);
257 * Most r/o checks on a fs are for operations that take
258 * discrete amounts of time, like a write() or unlink().
259 * We must keep track of when those operations start
260 * (for permission checks) and when they end, so that
261 * we can determine when writes are able to occur to
265 * __mnt_is_readonly: check whether a mount is read-only
266 * @mnt: the mount to check for its write status
268 * This shouldn't be used directly ouside of the VFS.
269 * It does not guarantee that the filesystem will stay
270 * r/w, just that it is right *now*. This can not and
271 * should not be used in place of IS_RDONLY(inode).
272 * mnt_want/drop_write() will _keep_ the filesystem
275 int __mnt_is_readonly(struct vfsmount *mnt)
277 if (mnt->mnt_flags & MNT_READONLY)
279 if (mnt->mnt_sb->s_flags & MS_RDONLY)
283 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
285 static inline void mnt_inc_writers(struct mount *mnt)
288 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
294 static inline void mnt_dec_writers(struct mount *mnt)
297 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
303 static unsigned int mnt_get_writers(struct mount *mnt)
306 unsigned int count = 0;
309 for_each_possible_cpu(cpu) {
310 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
315 return mnt->mnt_writers;
319 static int mnt_is_readonly(struct vfsmount *mnt)
321 if (mnt->mnt_sb->s_readonly_remount)
323 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
325 return __mnt_is_readonly(mnt);
329 * Most r/o & frozen checks on a fs are for operations that take discrete
330 * amounts of time, like a write() or unlink(). We must keep track of when
331 * those operations start (for permission checks) and when they end, so that we
332 * can determine when writes are able to occur to a filesystem.
335 * __mnt_want_write - get write access to a mount without freeze protection
336 * @m: the mount on which to take a write
338 * This tells the low-level filesystem that a write is about to be performed to
339 * it, and makes sure that writes are allowed (mnt it read-write) before
340 * returning success. This operation does not protect against filesystem being
341 * frozen. When the write operation is finished, __mnt_drop_write() must be
342 * called. This is effectively a refcount.
344 int __mnt_want_write(struct vfsmount *m)
346 struct mount *mnt = real_mount(m);
350 mnt_inc_writers(mnt);
352 * The store to mnt_inc_writers must be visible before we pass
353 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
354 * incremented count after it has set MNT_WRITE_HOLD.
357 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
363 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
364 * be set to match its requirements. So we must not load that until
365 * MNT_WRITE_HOLD is cleared.
368 if (mnt_is_readonly(m)) {
369 mnt_dec_writers(mnt);
378 * mnt_want_write - get write access to a mount
379 * @m: the mount on which to take a write
381 * This tells the low-level filesystem that a write is about to be performed to
382 * it, and makes sure that writes are allowed (mount is read-write, filesystem
383 * is not frozen) before returning success. When the write operation is
384 * finished, mnt_drop_write() must be called. This is effectively a refcount.
386 int mnt_want_write(struct vfsmount *m)
390 sb_start_write(m->mnt_sb);
391 ret = __mnt_want_write(m);
393 sb_end_write(m->mnt_sb);
396 EXPORT_SYMBOL_GPL(mnt_want_write);
399 * mnt_clone_write - get write access to a mount
400 * @mnt: the mount on which to take a write
402 * This is effectively like mnt_want_write, except
403 * it must only be used to take an extra write reference
404 * on a mountpoint that we already know has a write reference
405 * on it. This allows some optimisation.
407 * After finished, mnt_drop_write must be called as usual to
408 * drop the reference.
410 int mnt_clone_write(struct vfsmount *mnt)
412 /* superblock may be r/o */
413 if (__mnt_is_readonly(mnt))
416 mnt_inc_writers(real_mount(mnt));
420 EXPORT_SYMBOL_GPL(mnt_clone_write);
423 * __mnt_want_write_file - get write access to a file's mount
424 * @file: the file who's mount on which to take a write
426 * This is like __mnt_want_write, but it takes a file and can
427 * do some optimisations if the file is open for write already
429 int __mnt_want_write_file(struct file *file)
431 if (!(file->f_mode & FMODE_WRITER))
432 return __mnt_want_write(file->f_path.mnt);
434 return mnt_clone_write(file->f_path.mnt);
438 * mnt_want_write_file - get write access to a file's mount
439 * @file: the file who's mount on which to take a write
441 * This is like mnt_want_write, but it takes a file and can
442 * do some optimisations if the file is open for write already
444 int mnt_want_write_file(struct file *file)
448 sb_start_write(file->f_path.mnt->mnt_sb);
449 ret = __mnt_want_write_file(file);
451 sb_end_write(file->f_path.mnt->mnt_sb);
454 EXPORT_SYMBOL_GPL(mnt_want_write_file);
457 * __mnt_drop_write - give up write access to a mount
458 * @mnt: the mount on which to give up write access
460 * Tells the low-level filesystem that we are done
461 * performing writes to it. Must be matched with
462 * __mnt_want_write() call above.
464 void __mnt_drop_write(struct vfsmount *mnt)
467 mnt_dec_writers(real_mount(mnt));
472 * mnt_drop_write - give up write access to a mount
473 * @mnt: the mount on which to give up write access
475 * Tells the low-level filesystem that we are done performing writes to it and
476 * also allows filesystem to be frozen again. Must be matched with
477 * mnt_want_write() call above.
479 void mnt_drop_write(struct vfsmount *mnt)
481 __mnt_drop_write(mnt);
482 sb_end_write(mnt->mnt_sb);
484 EXPORT_SYMBOL_GPL(mnt_drop_write);
486 void __mnt_drop_write_file(struct file *file)
488 __mnt_drop_write(file->f_path.mnt);
491 void mnt_drop_write_file(struct file *file)
493 mnt_drop_write(file->f_path.mnt);
495 EXPORT_SYMBOL(mnt_drop_write_file);
497 static int mnt_make_readonly(struct mount *mnt)
502 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
504 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
505 * should be visible before we do.
510 * With writers on hold, if this value is zero, then there are
511 * definitely no active writers (although held writers may subsequently
512 * increment the count, they'll have to wait, and decrement it after
513 * seeing MNT_READONLY).
515 * It is OK to have counter incremented on one CPU and decremented on
516 * another: the sum will add up correctly. The danger would be when we
517 * sum up each counter, if we read a counter before it is incremented,
518 * but then read another CPU's count which it has been subsequently
519 * decremented from -- we would see more decrements than we should.
520 * MNT_WRITE_HOLD protects against this scenario, because
521 * mnt_want_write first increments count, then smp_mb, then spins on
522 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
523 * we're counting up here.
525 if (mnt_get_writers(mnt) > 0)
528 mnt->mnt.mnt_flags |= MNT_READONLY;
530 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
531 * that become unheld will see MNT_READONLY.
534 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
539 static void __mnt_unmake_readonly(struct mount *mnt)
542 mnt->mnt.mnt_flags &= ~MNT_READONLY;
546 int sb_prepare_remount_readonly(struct super_block *sb)
551 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
552 if (atomic_long_read(&sb->s_remove_count))
556 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
557 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
558 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
560 if (mnt_get_writers(mnt) > 0) {
566 if (!err && atomic_long_read(&sb->s_remove_count))
570 sb->s_readonly_remount = 1;
573 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
574 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
575 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
582 static void free_vfsmnt(struct mount *mnt)
584 kfree_const(mnt->mnt_devname);
586 free_percpu(mnt->mnt_pcp);
588 kmem_cache_free(mnt_cache, mnt);
591 static void delayed_free_vfsmnt(struct rcu_head *head)
593 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
596 /* call under rcu_read_lock */
597 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
600 if (read_seqretry(&mount_lock, seq))
604 mnt = real_mount(bastard);
605 mnt_add_count(mnt, 1);
606 if (likely(!read_seqretry(&mount_lock, seq)))
608 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
609 mnt_add_count(mnt, -1);
619 * find the first mount at @dentry on vfsmount @mnt.
620 * call under rcu_read_lock()
622 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
624 struct hlist_head *head = m_hash(mnt, dentry);
627 hlist_for_each_entry_rcu(p, head, mnt_hash)
628 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
634 * find the last mount at @dentry on vfsmount @mnt.
635 * mount_lock must be held.
637 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
639 struct mount *p, *res = NULL;
640 p = __lookup_mnt(mnt, dentry);
643 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
645 hlist_for_each_entry_continue(p, mnt_hash) {
646 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
648 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
656 * lookup_mnt - Return the first child mount mounted at path
658 * "First" means first mounted chronologically. If you create the
661 * mount /dev/sda1 /mnt
662 * mount /dev/sda2 /mnt
663 * mount /dev/sda3 /mnt
665 * Then lookup_mnt() on the base /mnt dentry in the root mount will
666 * return successively the root dentry and vfsmount of /dev/sda1, then
667 * /dev/sda2, then /dev/sda3, then NULL.
669 * lookup_mnt takes a reference to the found vfsmount.
671 struct vfsmount *lookup_mnt(struct path *path)
673 struct mount *child_mnt;
679 seq = read_seqbegin(&mount_lock);
680 child_mnt = __lookup_mnt(path->mnt, path->dentry);
681 m = child_mnt ? &child_mnt->mnt : NULL;
682 } while (!legitimize_mnt(m, seq));
688 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
689 * current mount namespace.
691 * The common case is dentries are not mountpoints at all and that
692 * test is handled inline. For the slow case when we are actually
693 * dealing with a mountpoint of some kind, walk through all of the
694 * mounts in the current mount namespace and test to see if the dentry
697 * The mount_hashtable is not usable in the context because we
698 * need to identify all mounts that may be in the current mount
699 * namespace not just a mount that happens to have some specified
702 bool __is_local_mountpoint(struct dentry *dentry)
704 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
706 bool is_covered = false;
708 if (!d_mountpoint(dentry))
711 down_read(&namespace_sem);
712 list_for_each_entry(mnt, &ns->list, mnt_list) {
713 is_covered = (mnt->mnt_mountpoint == dentry);
717 up_read(&namespace_sem);
722 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
724 struct hlist_head *chain = mp_hash(dentry);
725 struct mountpoint *mp;
727 hlist_for_each_entry(mp, chain, m_hash) {
728 if (mp->m_dentry == dentry) {
729 /* might be worth a WARN_ON() */
730 if (d_unlinked(dentry))
731 return ERR_PTR(-ENOENT);
739 static struct mountpoint *new_mountpoint(struct dentry *dentry)
741 struct hlist_head *chain = mp_hash(dentry);
742 struct mountpoint *mp;
745 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
747 return ERR_PTR(-ENOMEM);
749 ret = d_set_mounted(dentry);
755 mp->m_dentry = dentry;
757 hlist_add_head(&mp->m_hash, chain);
758 INIT_HLIST_HEAD(&mp->m_list);
762 static void put_mountpoint(struct mountpoint *mp)
764 if (!--mp->m_count) {
765 struct dentry *dentry = mp->m_dentry;
766 BUG_ON(!hlist_empty(&mp->m_list));
767 spin_lock(&dentry->d_lock);
768 dentry->d_flags &= ~DCACHE_MOUNTED;
769 spin_unlock(&dentry->d_lock);
770 hlist_del(&mp->m_hash);
775 static inline int check_mnt(struct mount *mnt)
777 return mnt->mnt_ns == current->nsproxy->mnt_ns;
781 * vfsmount lock must be held for write
783 static void touch_mnt_namespace(struct mnt_namespace *ns)
787 wake_up_interruptible(&ns->poll);
792 * vfsmount lock must be held for write
794 static void __touch_mnt_namespace(struct mnt_namespace *ns)
796 if (ns && ns->event != event) {
798 wake_up_interruptible(&ns->poll);
803 * vfsmount lock must be held for write
805 static void unhash_mnt(struct mount *mnt)
807 mnt->mnt_parent = mnt;
808 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
809 list_del_init(&mnt->mnt_child);
810 hlist_del_init_rcu(&mnt->mnt_hash);
811 hlist_del_init(&mnt->mnt_mp_list);
812 put_mountpoint(mnt->mnt_mp);
817 * vfsmount lock must be held for write
819 static void detach_mnt(struct mount *mnt, struct path *old_path)
821 old_path->dentry = mnt->mnt_mountpoint;
822 old_path->mnt = &mnt->mnt_parent->mnt;
827 * vfsmount lock must be held for write
829 static void umount_mnt(struct mount *mnt)
831 /* old mountpoint will be dropped when we can do that */
832 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
837 * vfsmount lock must be held for write
839 void mnt_set_mountpoint(struct mount *mnt,
840 struct mountpoint *mp,
841 struct mount *child_mnt)
844 mnt_add_count(mnt, 1); /* essentially, that's mntget */
845 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
846 child_mnt->mnt_parent = mnt;
847 child_mnt->mnt_mp = mp;
848 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
852 * vfsmount lock must be held for write
854 static void attach_mnt(struct mount *mnt,
855 struct mount *parent,
856 struct mountpoint *mp)
858 mnt_set_mountpoint(parent, mp, mnt);
859 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
860 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
863 static void attach_shadowed(struct mount *mnt,
864 struct mount *parent,
865 struct mount *shadows)
868 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
869 list_add(&mnt->mnt_child, &shadows->mnt_child);
871 hlist_add_head_rcu(&mnt->mnt_hash,
872 m_hash(&parent->mnt, mnt->mnt_mountpoint));
873 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
878 * vfsmount lock must be held for write
880 static void commit_tree(struct mount *mnt, struct mount *shadows)
882 struct mount *parent = mnt->mnt_parent;
885 struct mnt_namespace *n = parent->mnt_ns;
887 BUG_ON(parent == mnt);
889 list_add_tail(&head, &mnt->mnt_list);
890 list_for_each_entry(m, &head, mnt_list)
893 list_splice(&head, n->list.prev);
895 attach_shadowed(mnt, parent, shadows);
896 touch_mnt_namespace(n);
899 static struct mount *next_mnt(struct mount *p, struct mount *root)
901 struct list_head *next = p->mnt_mounts.next;
902 if (next == &p->mnt_mounts) {
906 next = p->mnt_child.next;
907 if (next != &p->mnt_parent->mnt_mounts)
912 return list_entry(next, struct mount, mnt_child);
915 static struct mount *skip_mnt_tree(struct mount *p)
917 struct list_head *prev = p->mnt_mounts.prev;
918 while (prev != &p->mnt_mounts) {
919 p = list_entry(prev, struct mount, mnt_child);
920 prev = p->mnt_mounts.prev;
926 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
932 return ERR_PTR(-ENODEV);
934 mnt = alloc_vfsmnt(name);
936 return ERR_PTR(-ENOMEM);
938 if (flags & MS_KERNMOUNT)
939 mnt->mnt.mnt_flags = MNT_INTERNAL;
941 root = mount_fs(type, flags, name, data);
945 return ERR_CAST(root);
948 mnt->mnt.mnt_root = root;
949 mnt->mnt.mnt_sb = root->d_sb;
950 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
951 mnt->mnt_parent = mnt;
953 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
957 EXPORT_SYMBOL_GPL(vfs_kern_mount);
959 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
962 struct super_block *sb = old->mnt.mnt_sb;
966 mnt = alloc_vfsmnt(old->mnt_devname);
968 return ERR_PTR(-ENOMEM);
970 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
971 mnt->mnt_group_id = 0; /* not a peer of original */
973 mnt->mnt_group_id = old->mnt_group_id;
975 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
976 err = mnt_alloc_group_id(mnt);
981 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
982 /* Don't allow unprivileged users to change mount flags */
983 if (flag & CL_UNPRIVILEGED) {
984 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
986 if (mnt->mnt.mnt_flags & MNT_READONLY)
987 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
989 if (mnt->mnt.mnt_flags & MNT_NODEV)
990 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
992 if (mnt->mnt.mnt_flags & MNT_NOSUID)
993 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
995 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
996 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
999 /* Don't allow unprivileged users to reveal what is under a mount */
1000 if ((flag & CL_UNPRIVILEGED) &&
1001 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1002 mnt->mnt.mnt_flags |= MNT_LOCKED;
1004 atomic_inc(&sb->s_active);
1005 mnt->mnt.mnt_sb = sb;
1006 mnt->mnt.mnt_root = dget(root);
1007 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1008 mnt->mnt_parent = mnt;
1010 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1011 unlock_mount_hash();
1013 if ((flag & CL_SLAVE) ||
1014 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1015 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1016 mnt->mnt_master = old;
1017 CLEAR_MNT_SHARED(mnt);
1018 } else if (!(flag & CL_PRIVATE)) {
1019 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1020 list_add(&mnt->mnt_share, &old->mnt_share);
1021 if (IS_MNT_SLAVE(old))
1022 list_add(&mnt->mnt_slave, &old->mnt_slave);
1023 mnt->mnt_master = old->mnt_master;
1025 if (flag & CL_MAKE_SHARED)
1026 set_mnt_shared(mnt);
1028 /* stick the duplicate mount on the same expiry list
1029 * as the original if that was on one */
1030 if (flag & CL_EXPIRE) {
1031 if (!list_empty(&old->mnt_expire))
1032 list_add(&mnt->mnt_expire, &old->mnt_expire);
1040 return ERR_PTR(err);
1043 static void cleanup_mnt(struct mount *mnt)
1046 * This probably indicates that somebody messed
1047 * up a mnt_want/drop_write() pair. If this
1048 * happens, the filesystem was probably unable
1049 * to make r/w->r/o transitions.
1052 * The locking used to deal with mnt_count decrement provides barriers,
1053 * so mnt_get_writers() below is safe.
1055 WARN_ON(mnt_get_writers(mnt));
1056 if (unlikely(mnt->mnt_pins.first))
1058 fsnotify_vfsmount_delete(&mnt->mnt);
1059 dput(mnt->mnt.mnt_root);
1060 deactivate_super(mnt->mnt.mnt_sb);
1062 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1065 static void __cleanup_mnt(struct rcu_head *head)
1067 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1070 static LLIST_HEAD(delayed_mntput_list);
1071 static void delayed_mntput(struct work_struct *unused)
1073 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1074 struct llist_node *next;
1076 for (; node; node = next) {
1077 next = llist_next(node);
1078 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1081 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1083 static void mntput_no_expire(struct mount *mnt)
1086 mnt_add_count(mnt, -1);
1087 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1092 if (mnt_get_count(mnt)) {
1094 unlock_mount_hash();
1097 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1099 unlock_mount_hash();
1102 mnt->mnt.mnt_flags |= MNT_DOOMED;
1105 list_del(&mnt->mnt_instance);
1107 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1108 struct mount *p, *tmp;
1109 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1113 unlock_mount_hash();
1115 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1116 struct task_struct *task = current;
1117 if (likely(!(task->flags & PF_KTHREAD))) {
1118 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1119 if (!task_work_add(task, &mnt->mnt_rcu, true))
1122 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1123 schedule_delayed_work(&delayed_mntput_work, 1);
1129 void mntput(struct vfsmount *mnt)
1132 struct mount *m = real_mount(mnt);
1133 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1134 if (unlikely(m->mnt_expiry_mark))
1135 m->mnt_expiry_mark = 0;
1136 mntput_no_expire(m);
1139 EXPORT_SYMBOL(mntput);
1141 struct vfsmount *mntget(struct vfsmount *mnt)
1144 mnt_add_count(real_mount(mnt), 1);
1147 EXPORT_SYMBOL(mntget);
1149 struct vfsmount *mnt_clone_internal(struct path *path)
1152 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1155 p->mnt.mnt_flags |= MNT_INTERNAL;
1159 static inline void mangle(struct seq_file *m, const char *s)
1161 seq_escape(m, s, " \t\n\\");
1165 * Simple .show_options callback for filesystems which don't want to
1166 * implement more complex mount option showing.
1168 * See also save_mount_options().
1170 int generic_show_options(struct seq_file *m, struct dentry *root)
1172 const char *options;
1175 options = rcu_dereference(root->d_sb->s_options);
1177 if (options != NULL && options[0]) {
1185 EXPORT_SYMBOL(generic_show_options);
1188 * If filesystem uses generic_show_options(), this function should be
1189 * called from the fill_super() callback.
1191 * The .remount_fs callback usually needs to be handled in a special
1192 * way, to make sure, that previous options are not overwritten if the
1195 * Also note, that if the filesystem's .remount_fs function doesn't
1196 * reset all options to their default value, but changes only newly
1197 * given options, then the displayed options will not reflect reality
1200 void save_mount_options(struct super_block *sb, char *options)
1202 BUG_ON(sb->s_options);
1203 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1205 EXPORT_SYMBOL(save_mount_options);
1207 void replace_mount_options(struct super_block *sb, char *options)
1209 char *old = sb->s_options;
1210 rcu_assign_pointer(sb->s_options, options);
1216 EXPORT_SYMBOL(replace_mount_options);
1218 #ifdef CONFIG_PROC_FS
1219 /* iterator; we want it to have access to namespace_sem, thus here... */
1220 static void *m_start(struct seq_file *m, loff_t *pos)
1222 struct proc_mounts *p = proc_mounts(m);
1224 down_read(&namespace_sem);
1225 if (p->cached_event == p->ns->event) {
1226 void *v = p->cached_mount;
1227 if (*pos == p->cached_index)
1229 if (*pos == p->cached_index + 1) {
1230 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1231 return p->cached_mount = v;
1235 p->cached_event = p->ns->event;
1236 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1237 p->cached_index = *pos;
1238 return p->cached_mount;
1241 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1243 struct proc_mounts *p = proc_mounts(m);
1245 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1246 p->cached_index = *pos;
1247 return p->cached_mount;
1250 static void m_stop(struct seq_file *m, void *v)
1252 up_read(&namespace_sem);
1255 static int m_show(struct seq_file *m, void *v)
1257 struct proc_mounts *p = proc_mounts(m);
1258 struct mount *r = list_entry(v, struct mount, mnt_list);
1259 return p->show(m, &r->mnt);
1262 const struct seq_operations mounts_op = {
1268 #endif /* CONFIG_PROC_FS */
1271 * may_umount_tree - check if a mount tree is busy
1272 * @mnt: root of mount tree
1274 * This is called to check if a tree of mounts has any
1275 * open files, pwds, chroots or sub mounts that are
1278 int may_umount_tree(struct vfsmount *m)
1280 struct mount *mnt = real_mount(m);
1281 int actual_refs = 0;
1282 int minimum_refs = 0;
1286 /* write lock needed for mnt_get_count */
1288 for (p = mnt; p; p = next_mnt(p, mnt)) {
1289 actual_refs += mnt_get_count(p);
1292 unlock_mount_hash();
1294 if (actual_refs > minimum_refs)
1300 EXPORT_SYMBOL(may_umount_tree);
1303 * may_umount - check if a mount point is busy
1304 * @mnt: root of mount
1306 * This is called to check if a mount point has any
1307 * open files, pwds, chroots or sub mounts. If the
1308 * mount has sub mounts this will return busy
1309 * regardless of whether the sub mounts are busy.
1311 * Doesn't take quota and stuff into account. IOW, in some cases it will
1312 * give false negatives. The main reason why it's here is that we need
1313 * a non-destructive way to look for easily umountable filesystems.
1315 int may_umount(struct vfsmount *mnt)
1318 down_read(&namespace_sem);
1320 if (propagate_mount_busy(real_mount(mnt), 2))
1322 unlock_mount_hash();
1323 up_read(&namespace_sem);
1327 EXPORT_SYMBOL(may_umount);
1329 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1331 static void namespace_unlock(void)
1333 struct hlist_head head;
1335 hlist_move_list(&unmounted, &head);
1337 up_write(&namespace_sem);
1339 if (likely(hlist_empty(&head)))
1344 group_pin_kill(&head);
1347 static inline void namespace_lock(void)
1349 down_write(&namespace_sem);
1352 enum umount_tree_flags {
1354 UMOUNT_PROPAGATE = 2,
1355 UMOUNT_CONNECTED = 4,
1358 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1360 /* Leaving mounts connected is only valid for lazy umounts */
1361 if (how & UMOUNT_SYNC)
1364 /* A mount without a parent has nothing to be connected to */
1365 if (!mnt_has_parent(mnt))
1368 /* Because the reference counting rules change when mounts are
1369 * unmounted and connected, umounted mounts may not be
1370 * connected to mounted mounts.
1372 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1375 /* Has it been requested that the mount remain connected? */
1376 if (how & UMOUNT_CONNECTED)
1379 /* Is the mount locked such that it needs to remain connected? */
1380 if (IS_MNT_LOCKED(mnt))
1383 /* By default disconnect the mount */
1388 * mount_lock must be held
1389 * namespace_sem must be held for write
1391 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1393 LIST_HEAD(tmp_list);
1396 if (how & UMOUNT_PROPAGATE)
1397 propagate_mount_unlock(mnt);
1399 /* Gather the mounts to umount */
1400 for (p = mnt; p; p = next_mnt(p, mnt)) {
1401 p->mnt.mnt_flags |= MNT_UMOUNT;
1402 list_move(&p->mnt_list, &tmp_list);
1405 /* Hide the mounts from mnt_mounts */
1406 list_for_each_entry(p, &tmp_list, mnt_list) {
1407 list_del_init(&p->mnt_child);
1410 /* Add propogated mounts to the tmp_list */
1411 if (how & UMOUNT_PROPAGATE)
1412 propagate_umount(&tmp_list);
1414 while (!list_empty(&tmp_list)) {
1416 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1417 list_del_init(&p->mnt_expire);
1418 list_del_init(&p->mnt_list);
1419 __touch_mnt_namespace(p->mnt_ns);
1421 if (how & UMOUNT_SYNC)
1422 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1424 disconnect = disconnect_mount(p, how);
1426 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1427 disconnect ? &unmounted : NULL);
1428 if (mnt_has_parent(p)) {
1429 mnt_add_count(p->mnt_parent, -1);
1431 /* Don't forget about p */
1432 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1437 change_mnt_propagation(p, MS_PRIVATE);
1441 static void shrink_submounts(struct mount *mnt);
1443 static int do_umount(struct mount *mnt, int flags)
1445 struct super_block *sb = mnt->mnt.mnt_sb;
1448 retval = security_sb_umount(&mnt->mnt, flags);
1453 * Allow userspace to request a mountpoint be expired rather than
1454 * unmounting unconditionally. Unmount only happens if:
1455 * (1) the mark is already set (the mark is cleared by mntput())
1456 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1458 if (flags & MNT_EXPIRE) {
1459 if (&mnt->mnt == current->fs->root.mnt ||
1460 flags & (MNT_FORCE | MNT_DETACH))
1464 * probably don't strictly need the lock here if we examined
1465 * all race cases, but it's a slowpath.
1468 if (mnt_get_count(mnt) != 2) {
1469 unlock_mount_hash();
1472 unlock_mount_hash();
1474 if (!xchg(&mnt->mnt_expiry_mark, 1))
1479 * If we may have to abort operations to get out of this
1480 * mount, and they will themselves hold resources we must
1481 * allow the fs to do things. In the Unix tradition of
1482 * 'Gee thats tricky lets do it in userspace' the umount_begin
1483 * might fail to complete on the first run through as other tasks
1484 * must return, and the like. Thats for the mount program to worry
1485 * about for the moment.
1488 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1489 sb->s_op->umount_begin(sb);
1493 * No sense to grab the lock for this test, but test itself looks
1494 * somewhat bogus. Suggestions for better replacement?
1495 * Ho-hum... In principle, we might treat that as umount + switch
1496 * to rootfs. GC would eventually take care of the old vfsmount.
1497 * Actually it makes sense, especially if rootfs would contain a
1498 * /reboot - static binary that would close all descriptors and
1499 * call reboot(9). Then init(8) could umount root and exec /reboot.
1501 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1503 * Special case for "unmounting" root ...
1504 * we just try to remount it readonly.
1506 if (!capable(CAP_SYS_ADMIN))
1508 down_write(&sb->s_umount);
1509 if (!(sb->s_flags & MS_RDONLY))
1510 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1511 up_write(&sb->s_umount);
1519 if (flags & MNT_DETACH) {
1520 if (!list_empty(&mnt->mnt_list))
1521 umount_tree(mnt, UMOUNT_PROPAGATE);
1524 shrink_submounts(mnt);
1526 if (!propagate_mount_busy(mnt, 2)) {
1527 if (!list_empty(&mnt->mnt_list))
1528 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1532 unlock_mount_hash();
1538 * __detach_mounts - lazily unmount all mounts on the specified dentry
1540 * During unlink, rmdir, and d_drop it is possible to loose the path
1541 * to an existing mountpoint, and wind up leaking the mount.
1542 * detach_mounts allows lazily unmounting those mounts instead of
1545 * The caller may hold dentry->d_inode->i_mutex.
1547 void __detach_mounts(struct dentry *dentry)
1549 struct mountpoint *mp;
1553 mp = lookup_mountpoint(dentry);
1554 if (IS_ERR_OR_NULL(mp))
1558 while (!hlist_empty(&mp->m_list)) {
1559 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1560 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1561 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1564 else umount_tree(mnt, UMOUNT_CONNECTED);
1566 unlock_mount_hash();
1573 * Is the caller allowed to modify his namespace?
1575 static inline bool may_mount(void)
1577 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1581 * Now umount can handle mount points as well as block devices.
1582 * This is important for filesystems which use unnamed block devices.
1584 * We now support a flag for forced unmount like the other 'big iron'
1585 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1588 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1593 int lookup_flags = 0;
1595 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1601 if (!(flags & UMOUNT_NOFOLLOW))
1602 lookup_flags |= LOOKUP_FOLLOW;
1604 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1607 mnt = real_mount(path.mnt);
1609 if (path.dentry != path.mnt->mnt_root)
1611 if (!check_mnt(mnt))
1613 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1616 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1619 retval = do_umount(mnt, flags);
1621 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1623 mntput_no_expire(mnt);
1628 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1631 * The 2.0 compatible umount. No flags.
1633 SYSCALL_DEFINE1(oldumount, char __user *, name)
1635 return sys_umount(name, 0);
1640 static bool is_mnt_ns_file(struct dentry *dentry)
1642 /* Is this a proxy for a mount namespace? */
1643 return dentry->d_op == &ns_dentry_operations &&
1644 dentry->d_fsdata == &mntns_operations;
1647 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1649 return container_of(ns, struct mnt_namespace, ns);
1652 static bool mnt_ns_loop(struct dentry *dentry)
1654 /* Could bind mounting the mount namespace inode cause a
1655 * mount namespace loop?
1657 struct mnt_namespace *mnt_ns;
1658 if (!is_mnt_ns_file(dentry))
1661 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1662 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1665 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1668 struct mount *res, *p, *q, *r, *parent;
1670 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1671 return ERR_PTR(-EINVAL);
1673 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1674 return ERR_PTR(-EINVAL);
1676 res = q = clone_mnt(mnt, dentry, flag);
1680 q->mnt_mountpoint = mnt->mnt_mountpoint;
1683 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1685 if (!is_subdir(r->mnt_mountpoint, dentry))
1688 for (s = r; s; s = next_mnt(s, r)) {
1689 struct mount *t = NULL;
1690 if (!(flag & CL_COPY_UNBINDABLE) &&
1691 IS_MNT_UNBINDABLE(s)) {
1692 s = skip_mnt_tree(s);
1695 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1696 is_mnt_ns_file(s->mnt.mnt_root)) {
1697 s = skip_mnt_tree(s);
1700 while (p != s->mnt_parent) {
1706 q = clone_mnt(p, p->mnt.mnt_root, flag);
1710 list_add_tail(&q->mnt_list, &res->mnt_list);
1711 mnt_set_mountpoint(parent, p->mnt_mp, q);
1712 if (!list_empty(&parent->mnt_mounts)) {
1713 t = list_last_entry(&parent->mnt_mounts,
1714 struct mount, mnt_child);
1715 if (t->mnt_mp != p->mnt_mp)
1718 attach_shadowed(q, parent, t);
1719 unlock_mount_hash();
1726 umount_tree(res, UMOUNT_SYNC);
1727 unlock_mount_hash();
1732 /* Caller should check returned pointer for errors */
1734 struct vfsmount *collect_mounts(struct path *path)
1738 if (!check_mnt(real_mount(path->mnt)))
1739 tree = ERR_PTR(-EINVAL);
1741 tree = copy_tree(real_mount(path->mnt), path->dentry,
1742 CL_COPY_ALL | CL_PRIVATE);
1745 return ERR_CAST(tree);
1749 void drop_collected_mounts(struct vfsmount *mnt)
1753 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1754 unlock_mount_hash();
1759 * clone_private_mount - create a private clone of a path
1761 * This creates a new vfsmount, which will be the clone of @path. The new will
1762 * not be attached anywhere in the namespace and will be private (i.e. changes
1763 * to the originating mount won't be propagated into this).
1765 * Release with mntput().
1767 struct vfsmount *clone_private_mount(struct path *path)
1769 struct mount *old_mnt = real_mount(path->mnt);
1770 struct mount *new_mnt;
1772 if (IS_MNT_UNBINDABLE(old_mnt))
1773 return ERR_PTR(-EINVAL);
1775 down_read(&namespace_sem);
1776 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1777 up_read(&namespace_sem);
1778 if (IS_ERR(new_mnt))
1779 return ERR_CAST(new_mnt);
1781 return &new_mnt->mnt;
1783 EXPORT_SYMBOL_GPL(clone_private_mount);
1785 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1786 struct vfsmount *root)
1789 int res = f(root, arg);
1792 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1793 res = f(&mnt->mnt, arg);
1800 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1804 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1805 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1806 mnt_release_group_id(p);
1810 static int invent_group_ids(struct mount *mnt, bool recurse)
1814 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1815 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1816 int err = mnt_alloc_group_id(p);
1818 cleanup_group_ids(mnt, p);
1828 * @source_mnt : mount tree to be attached
1829 * @nd : place the mount tree @source_mnt is attached
1830 * @parent_nd : if non-null, detach the source_mnt from its parent and
1831 * store the parent mount and mountpoint dentry.
1832 * (done when source_mnt is moved)
1834 * NOTE: in the table below explains the semantics when a source mount
1835 * of a given type is attached to a destination mount of a given type.
1836 * ---------------------------------------------------------------------------
1837 * | BIND MOUNT OPERATION |
1838 * |**************************************************************************
1839 * | source-->| shared | private | slave | unbindable |
1843 * |**************************************************************************
1844 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1846 * |non-shared| shared (+) | private | slave (*) | invalid |
1847 * ***************************************************************************
1848 * A bind operation clones the source mount and mounts the clone on the
1849 * destination mount.
1851 * (++) the cloned mount is propagated to all the mounts in the propagation
1852 * tree of the destination mount and the cloned mount is added to
1853 * the peer group of the source mount.
1854 * (+) the cloned mount is created under the destination mount and is marked
1855 * as shared. The cloned mount is added to the peer group of the source
1857 * (+++) the mount is propagated to all the mounts in the propagation tree
1858 * of the destination mount and the cloned mount is made slave
1859 * of the same master as that of the source mount. The cloned mount
1860 * is marked as 'shared and slave'.
1861 * (*) the cloned mount is made a slave of the same master as that of the
1864 * ---------------------------------------------------------------------------
1865 * | MOVE MOUNT OPERATION |
1866 * |**************************************************************************
1867 * | source-->| shared | private | slave | unbindable |
1871 * |**************************************************************************
1872 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1874 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1875 * ***************************************************************************
1877 * (+) the mount is moved to the destination. And is then propagated to
1878 * all the mounts in the propagation tree of the destination mount.
1879 * (+*) the mount is moved to the destination.
1880 * (+++) the mount is moved to the destination and is then propagated to
1881 * all the mounts belonging to the destination mount's propagation tree.
1882 * the mount is marked as 'shared and slave'.
1883 * (*) the mount continues to be a slave at the new location.
1885 * if the source mount is a tree, the operations explained above is
1886 * applied to each mount in the tree.
1887 * Must be called without spinlocks held, since this function can sleep
1890 static int attach_recursive_mnt(struct mount *source_mnt,
1891 struct mount *dest_mnt,
1892 struct mountpoint *dest_mp,
1893 struct path *parent_path)
1895 HLIST_HEAD(tree_list);
1896 struct mount *child, *p;
1897 struct hlist_node *n;
1900 if (IS_MNT_SHARED(dest_mnt)) {
1901 err = invent_group_ids(source_mnt, true);
1904 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1907 goto out_cleanup_ids;
1908 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1914 detach_mnt(source_mnt, parent_path);
1915 attach_mnt(source_mnt, dest_mnt, dest_mp);
1916 touch_mnt_namespace(source_mnt->mnt_ns);
1918 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1919 commit_tree(source_mnt, NULL);
1922 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1924 hlist_del_init(&child->mnt_hash);
1925 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1926 child->mnt_mountpoint);
1927 commit_tree(child, q);
1929 unlock_mount_hash();
1934 while (!hlist_empty(&tree_list)) {
1935 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1936 umount_tree(child, UMOUNT_SYNC);
1938 unlock_mount_hash();
1939 cleanup_group_ids(source_mnt, NULL);
1944 static struct mountpoint *lock_mount(struct path *path)
1946 struct vfsmount *mnt;
1947 struct dentry *dentry = path->dentry;
1949 mutex_lock(&dentry->d_inode->i_mutex);
1950 if (unlikely(cant_mount(dentry))) {
1951 mutex_unlock(&dentry->d_inode->i_mutex);
1952 return ERR_PTR(-ENOENT);
1955 mnt = lookup_mnt(path);
1957 struct mountpoint *mp = lookup_mountpoint(dentry);
1959 mp = new_mountpoint(dentry);
1962 mutex_unlock(&dentry->d_inode->i_mutex);
1968 mutex_unlock(&path->dentry->d_inode->i_mutex);
1971 dentry = path->dentry = dget(mnt->mnt_root);
1975 static void unlock_mount(struct mountpoint *where)
1977 struct dentry *dentry = where->m_dentry;
1978 put_mountpoint(where);
1980 mutex_unlock(&dentry->d_inode->i_mutex);
1983 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1985 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1988 if (d_is_dir(mp->m_dentry) !=
1989 d_is_dir(mnt->mnt.mnt_root))
1992 return attach_recursive_mnt(mnt, p, mp, NULL);
1996 * Sanity check the flags to change_mnt_propagation.
1999 static int flags_to_propagation_type(int flags)
2001 int type = flags & ~(MS_REC | MS_SILENT);
2003 /* Fail if any non-propagation flags are set */
2004 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2006 /* Only one propagation flag should be set */
2007 if (!is_power_of_2(type))
2013 * recursively change the type of the mountpoint.
2015 static int do_change_type(struct path *path, int flag)
2018 struct mount *mnt = real_mount(path->mnt);
2019 int recurse = flag & MS_REC;
2023 if (path->dentry != path->mnt->mnt_root)
2026 type = flags_to_propagation_type(flag);
2031 if (type == MS_SHARED) {
2032 err = invent_group_ids(mnt, recurse);
2038 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2039 change_mnt_propagation(m, type);
2040 unlock_mount_hash();
2047 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2049 struct mount *child;
2050 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2051 if (!is_subdir(child->mnt_mountpoint, dentry))
2054 if (child->mnt.mnt_flags & MNT_LOCKED)
2061 * do loopback mount.
2063 static int do_loopback(struct path *path, const char *old_name,
2066 struct path old_path;
2067 struct mount *mnt = NULL, *old, *parent;
2068 struct mountpoint *mp;
2070 if (!old_name || !*old_name)
2072 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2077 if (mnt_ns_loop(old_path.dentry))
2080 mp = lock_mount(path);
2085 old = real_mount(old_path.mnt);
2086 parent = real_mount(path->mnt);
2089 if (IS_MNT_UNBINDABLE(old))
2092 if (!check_mnt(parent))
2095 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2098 if (!recurse && has_locked_children(old, old_path.dentry))
2102 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2104 mnt = clone_mnt(old, old_path.dentry, 0);
2111 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2113 err = graft_tree(mnt, parent, mp);
2116 umount_tree(mnt, UMOUNT_SYNC);
2117 unlock_mount_hash();
2122 path_put(&old_path);
2126 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2129 int readonly_request = 0;
2131 if (ms_flags & MS_RDONLY)
2132 readonly_request = 1;
2133 if (readonly_request == __mnt_is_readonly(mnt))
2136 if (readonly_request)
2137 error = mnt_make_readonly(real_mount(mnt));
2139 __mnt_unmake_readonly(real_mount(mnt));
2144 * change filesystem flags. dir should be a physical root of filesystem.
2145 * If you've mounted a non-root directory somewhere and want to do remount
2146 * on it - tough luck.
2148 static int do_remount(struct path *path, int flags, int mnt_flags,
2152 struct super_block *sb = path->mnt->mnt_sb;
2153 struct mount *mnt = real_mount(path->mnt);
2155 if (!check_mnt(mnt))
2158 if (path->dentry != path->mnt->mnt_root)
2161 /* Don't allow changing of locked mnt flags.
2163 * No locks need to be held here while testing the various
2164 * MNT_LOCK flags because those flags can never be cleared
2165 * once they are set.
2167 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2168 !(mnt_flags & MNT_READONLY)) {
2171 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2172 !(mnt_flags & MNT_NODEV)) {
2173 /* Was the nodev implicitly added in mount? */
2174 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2175 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2176 mnt_flags |= MNT_NODEV;
2181 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2182 !(mnt_flags & MNT_NOSUID)) {
2185 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2186 !(mnt_flags & MNT_NOEXEC)) {
2189 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2190 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2194 err = security_sb_remount(sb, data);
2198 down_write(&sb->s_umount);
2199 if (flags & MS_BIND)
2200 err = change_mount_flags(path->mnt, flags);
2201 else if (!capable(CAP_SYS_ADMIN))
2204 err = do_remount_sb(sb, flags, data, 0);
2207 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2208 mnt->mnt.mnt_flags = mnt_flags;
2209 touch_mnt_namespace(mnt->mnt_ns);
2210 unlock_mount_hash();
2212 up_write(&sb->s_umount);
2216 static inline int tree_contains_unbindable(struct mount *mnt)
2219 for (p = mnt; p; p = next_mnt(p, mnt)) {
2220 if (IS_MNT_UNBINDABLE(p))
2226 static int do_move_mount(struct path *path, const char *old_name)
2228 struct path old_path, parent_path;
2231 struct mountpoint *mp;
2233 if (!old_name || !*old_name)
2235 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2239 mp = lock_mount(path);
2244 old = real_mount(old_path.mnt);
2245 p = real_mount(path->mnt);
2248 if (!check_mnt(p) || !check_mnt(old))
2251 if (old->mnt.mnt_flags & MNT_LOCKED)
2255 if (old_path.dentry != old_path.mnt->mnt_root)
2258 if (!mnt_has_parent(old))
2261 if (d_is_dir(path->dentry) !=
2262 d_is_dir(old_path.dentry))
2265 * Don't move a mount residing in a shared parent.
2267 if (IS_MNT_SHARED(old->mnt_parent))
2270 * Don't move a mount tree containing unbindable mounts to a destination
2271 * mount which is shared.
2273 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2276 for (; mnt_has_parent(p); p = p->mnt_parent)
2280 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2284 /* if the mount is moved, it should no longer be expire
2286 list_del_init(&old->mnt_expire);
2291 path_put(&parent_path);
2292 path_put(&old_path);
2296 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2299 const char *subtype = strchr(fstype, '.');
2308 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2310 if (!mnt->mnt_sb->s_subtype)
2316 return ERR_PTR(err);
2320 * add a mount into a namespace's mount tree
2322 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2324 struct mountpoint *mp;
2325 struct mount *parent;
2328 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2330 mp = lock_mount(path);
2334 parent = real_mount(path->mnt);
2336 if (unlikely(!check_mnt(parent))) {
2337 /* that's acceptable only for automounts done in private ns */
2338 if (!(mnt_flags & MNT_SHRINKABLE))
2340 /* ... and for those we'd better have mountpoint still alive */
2341 if (!parent->mnt_ns)
2345 /* Refuse the same filesystem on the same mount point */
2347 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2348 path->mnt->mnt_root == path->dentry)
2352 if (d_is_symlink(newmnt->mnt.mnt_root))
2355 newmnt->mnt.mnt_flags = mnt_flags;
2356 err = graft_tree(newmnt, parent, mp);
2363 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2366 * create a new mount for userspace and request it to be added into the
2369 static int do_new_mount(struct path *path, const char *fstype, int flags,
2370 int mnt_flags, const char *name, void *data)
2372 struct file_system_type *type;
2373 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2374 struct vfsmount *mnt;
2380 type = get_fs_type(fstype);
2384 if (user_ns != &init_user_ns) {
2385 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2386 put_filesystem(type);
2389 /* Only in special cases allow devices from mounts
2390 * created outside the initial user namespace.
2392 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2394 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2396 if (type->fs_flags & FS_USERNS_VISIBLE) {
2397 if (!fs_fully_visible(type, &mnt_flags))
2402 mnt = vfs_kern_mount(type, flags, name, data);
2403 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2404 !mnt->mnt_sb->s_subtype)
2405 mnt = fs_set_subtype(mnt, fstype);
2407 put_filesystem(type);
2409 return PTR_ERR(mnt);
2411 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2417 int finish_automount(struct vfsmount *m, struct path *path)
2419 struct mount *mnt = real_mount(m);
2421 /* The new mount record should have at least 2 refs to prevent it being
2422 * expired before we get a chance to add it
2424 BUG_ON(mnt_get_count(mnt) < 2);
2426 if (m->mnt_sb == path->mnt->mnt_sb &&
2427 m->mnt_root == path->dentry) {
2432 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2436 /* remove m from any expiration list it may be on */
2437 if (!list_empty(&mnt->mnt_expire)) {
2439 list_del_init(&mnt->mnt_expire);
2448 * mnt_set_expiry - Put a mount on an expiration list
2449 * @mnt: The mount to list.
2450 * @expiry_list: The list to add the mount to.
2452 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2456 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2460 EXPORT_SYMBOL(mnt_set_expiry);
2463 * process a list of expirable mountpoints with the intent of discarding any
2464 * mountpoints that aren't in use and haven't been touched since last we came
2467 void mark_mounts_for_expiry(struct list_head *mounts)
2469 struct mount *mnt, *next;
2470 LIST_HEAD(graveyard);
2472 if (list_empty(mounts))
2478 /* extract from the expiration list every vfsmount that matches the
2479 * following criteria:
2480 * - only referenced by its parent vfsmount
2481 * - still marked for expiry (marked on the last call here; marks are
2482 * cleared by mntput())
2484 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2485 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2486 propagate_mount_busy(mnt, 1))
2488 list_move(&mnt->mnt_expire, &graveyard);
2490 while (!list_empty(&graveyard)) {
2491 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2492 touch_mnt_namespace(mnt->mnt_ns);
2493 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2495 unlock_mount_hash();
2499 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2502 * Ripoff of 'select_parent()'
2504 * search the list of submounts for a given mountpoint, and move any
2505 * shrinkable submounts to the 'graveyard' list.
2507 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2509 struct mount *this_parent = parent;
2510 struct list_head *next;
2514 next = this_parent->mnt_mounts.next;
2516 while (next != &this_parent->mnt_mounts) {
2517 struct list_head *tmp = next;
2518 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2521 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2524 * Descend a level if the d_mounts list is non-empty.
2526 if (!list_empty(&mnt->mnt_mounts)) {
2531 if (!propagate_mount_busy(mnt, 1)) {
2532 list_move_tail(&mnt->mnt_expire, graveyard);
2537 * All done at this level ... ascend and resume the search
2539 if (this_parent != parent) {
2540 next = this_parent->mnt_child.next;
2541 this_parent = this_parent->mnt_parent;
2548 * process a list of expirable mountpoints with the intent of discarding any
2549 * submounts of a specific parent mountpoint
2551 * mount_lock must be held for write
2553 static void shrink_submounts(struct mount *mnt)
2555 LIST_HEAD(graveyard);
2558 /* extract submounts of 'mountpoint' from the expiration list */
2559 while (select_submounts(mnt, &graveyard)) {
2560 while (!list_empty(&graveyard)) {
2561 m = list_first_entry(&graveyard, struct mount,
2563 touch_mnt_namespace(m->mnt_ns);
2564 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2570 * Some copy_from_user() implementations do not return the exact number of
2571 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2572 * Note that this function differs from copy_from_user() in that it will oops
2573 * on bad values of `to', rather than returning a short copy.
2575 static long exact_copy_from_user(void *to, const void __user * from,
2579 const char __user *f = from;
2582 if (!access_ok(VERIFY_READ, from, n))
2586 if (__get_user(c, f)) {
2597 int copy_mount_options(const void __user * data, unsigned long *where)
2607 if (!(page = __get_free_page(GFP_KERNEL)))
2610 /* We only care that *some* data at the address the user
2611 * gave us is valid. Just in case, we'll zero
2612 * the remainder of the page.
2614 /* copy_from_user cannot cross TASK_SIZE ! */
2615 size = TASK_SIZE - (unsigned long)data;
2616 if (size > PAGE_SIZE)
2619 i = size - exact_copy_from_user((void *)page, data, size);
2625 memset((char *)page + i, 0, PAGE_SIZE - i);
2630 char *copy_mount_string(const void __user *data)
2632 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2636 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2637 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2639 * data is a (void *) that can point to any structure up to
2640 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2641 * information (or be NULL).
2643 * Pre-0.97 versions of mount() didn't have a flags word.
2644 * When the flags word was introduced its top half was required
2645 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2646 * Therefore, if this magic number is present, it carries no information
2647 * and must be discarded.
2649 long do_mount(const char *dev_name, const char __user *dir_name,
2650 const char *type_page, unsigned long flags, void *data_page)
2657 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2658 flags &= ~MS_MGC_MSK;
2660 /* Basic sanity checks */
2662 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2664 /* ... and get the mountpoint */
2665 retval = user_path(dir_name, &path);
2669 retval = security_sb_mount(dev_name, &path,
2670 type_page, flags, data_page);
2671 if (!retval && !may_mount())
2676 /* Default to relatime unless overriden */
2677 if (!(flags & MS_NOATIME))
2678 mnt_flags |= MNT_RELATIME;
2680 /* Separate the per-mountpoint flags */
2681 if (flags & MS_NOSUID)
2682 mnt_flags |= MNT_NOSUID;
2683 if (flags & MS_NODEV)
2684 mnt_flags |= MNT_NODEV;
2685 if (flags & MS_NOEXEC)
2686 mnt_flags |= MNT_NOEXEC;
2687 if (flags & MS_NOATIME)
2688 mnt_flags |= MNT_NOATIME;
2689 if (flags & MS_NODIRATIME)
2690 mnt_flags |= MNT_NODIRATIME;
2691 if (flags & MS_STRICTATIME)
2692 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2693 if (flags & MS_RDONLY)
2694 mnt_flags |= MNT_READONLY;
2696 /* The default atime for remount is preservation */
2697 if ((flags & MS_REMOUNT) &&
2698 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2699 MS_STRICTATIME)) == 0)) {
2700 mnt_flags &= ~MNT_ATIME_MASK;
2701 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2704 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2705 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2708 if (flags & MS_REMOUNT)
2709 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2711 else if (flags & MS_BIND)
2712 retval = do_loopback(&path, dev_name, flags & MS_REC);
2713 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2714 retval = do_change_type(&path, flags);
2715 else if (flags & MS_MOVE)
2716 retval = do_move_mount(&path, dev_name);
2718 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2719 dev_name, data_page);
2725 static void free_mnt_ns(struct mnt_namespace *ns)
2727 ns_free_inum(&ns->ns);
2728 put_user_ns(ns->user_ns);
2733 * Assign a sequence number so we can detect when we attempt to bind
2734 * mount a reference to an older mount namespace into the current
2735 * mount namespace, preventing reference counting loops. A 64bit
2736 * number incrementing at 10Ghz will take 12,427 years to wrap which
2737 * is effectively never, so we can ignore the possibility.
2739 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2741 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2743 struct mnt_namespace *new_ns;
2746 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2748 return ERR_PTR(-ENOMEM);
2749 ret = ns_alloc_inum(&new_ns->ns);
2752 return ERR_PTR(ret);
2754 new_ns->ns.ops = &mntns_operations;
2755 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2756 atomic_set(&new_ns->count, 1);
2757 new_ns->root = NULL;
2758 INIT_LIST_HEAD(&new_ns->list);
2759 init_waitqueue_head(&new_ns->poll);
2761 new_ns->user_ns = get_user_ns(user_ns);
2765 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2766 struct user_namespace *user_ns, struct fs_struct *new_fs)
2768 struct mnt_namespace *new_ns;
2769 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2770 struct mount *p, *q;
2777 if (likely(!(flags & CLONE_NEWNS))) {
2784 new_ns = alloc_mnt_ns(user_ns);
2789 /* First pass: copy the tree topology */
2790 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2791 if (user_ns != ns->user_ns)
2792 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2793 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2796 free_mnt_ns(new_ns);
2797 return ERR_CAST(new);
2800 list_add_tail(&new_ns->list, &new->mnt_list);
2803 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2804 * as belonging to new namespace. We have already acquired a private
2805 * fs_struct, so tsk->fs->lock is not needed.
2812 if (&p->mnt == new_fs->root.mnt) {
2813 new_fs->root.mnt = mntget(&q->mnt);
2816 if (&p->mnt == new_fs->pwd.mnt) {
2817 new_fs->pwd.mnt = mntget(&q->mnt);
2821 p = next_mnt(p, old);
2822 q = next_mnt(q, new);
2825 while (p->mnt.mnt_root != q->mnt.mnt_root)
2826 p = next_mnt(p, old);
2839 * create_mnt_ns - creates a private namespace and adds a root filesystem
2840 * @mnt: pointer to the new root filesystem mountpoint
2842 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2844 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2845 if (!IS_ERR(new_ns)) {
2846 struct mount *mnt = real_mount(m);
2847 mnt->mnt_ns = new_ns;
2849 list_add(&mnt->mnt_list, &new_ns->list);
2856 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2858 struct mnt_namespace *ns;
2859 struct super_block *s;
2863 ns = create_mnt_ns(mnt);
2865 return ERR_CAST(ns);
2867 err = vfs_path_lookup(mnt->mnt_root, mnt,
2868 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2873 return ERR_PTR(err);
2875 /* trade a vfsmount reference for active sb one */
2876 s = path.mnt->mnt_sb;
2877 atomic_inc(&s->s_active);
2879 /* lock the sucker */
2880 down_write(&s->s_umount);
2881 /* ... and return the root of (sub)tree on it */
2884 EXPORT_SYMBOL(mount_subtree);
2886 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2887 char __user *, type, unsigned long, flags, void __user *, data)
2892 unsigned long data_page;
2894 kernel_type = copy_mount_string(type);
2895 ret = PTR_ERR(kernel_type);
2896 if (IS_ERR(kernel_type))
2899 kernel_dev = copy_mount_string(dev_name);
2900 ret = PTR_ERR(kernel_dev);
2901 if (IS_ERR(kernel_dev))
2904 ret = copy_mount_options(data, &data_page);
2908 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2909 (void *) data_page);
2911 free_page(data_page);
2921 * Return true if path is reachable from root
2923 * namespace_sem or mount_lock is held
2925 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2926 const struct path *root)
2928 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2929 dentry = mnt->mnt_mountpoint;
2930 mnt = mnt->mnt_parent;
2932 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2935 int path_is_under(struct path *path1, struct path *path2)
2938 read_seqlock_excl(&mount_lock);
2939 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2940 read_sequnlock_excl(&mount_lock);
2943 EXPORT_SYMBOL(path_is_under);
2946 * pivot_root Semantics:
2947 * Moves the root file system of the current process to the directory put_old,
2948 * makes new_root as the new root file system of the current process, and sets
2949 * root/cwd of all processes which had them on the current root to new_root.
2952 * The new_root and put_old must be directories, and must not be on the
2953 * same file system as the current process root. The put_old must be
2954 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2955 * pointed to by put_old must yield the same directory as new_root. No other
2956 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2958 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2959 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2960 * in this situation.
2963 * - we don't move root/cwd if they are not at the root (reason: if something
2964 * cared enough to change them, it's probably wrong to force them elsewhere)
2965 * - it's okay to pick a root that isn't the root of a file system, e.g.
2966 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2967 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2970 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2971 const char __user *, put_old)
2973 struct path new, old, parent_path, root_parent, root;
2974 struct mount *new_mnt, *root_mnt, *old_mnt;
2975 struct mountpoint *old_mp, *root_mp;
2981 error = user_path_dir(new_root, &new);
2985 error = user_path_dir(put_old, &old);
2989 error = security_sb_pivotroot(&old, &new);
2993 get_fs_root(current->fs, &root);
2994 old_mp = lock_mount(&old);
2995 error = PTR_ERR(old_mp);
3000 new_mnt = real_mount(new.mnt);
3001 root_mnt = real_mount(root.mnt);
3002 old_mnt = real_mount(old.mnt);
3003 if (IS_MNT_SHARED(old_mnt) ||
3004 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3005 IS_MNT_SHARED(root_mnt->mnt_parent))
3007 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3009 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3012 if (d_unlinked(new.dentry))
3015 if (new_mnt == root_mnt || old_mnt == root_mnt)
3016 goto out4; /* loop, on the same file system */
3018 if (root.mnt->mnt_root != root.dentry)
3019 goto out4; /* not a mountpoint */
3020 if (!mnt_has_parent(root_mnt))
3021 goto out4; /* not attached */
3022 root_mp = root_mnt->mnt_mp;
3023 if (new.mnt->mnt_root != new.dentry)
3024 goto out4; /* not a mountpoint */
3025 if (!mnt_has_parent(new_mnt))
3026 goto out4; /* not attached */
3027 /* make sure we can reach put_old from new_root */
3028 if (!is_path_reachable(old_mnt, old.dentry, &new))
3030 /* make certain new is below the root */
3031 if (!is_path_reachable(new_mnt, new.dentry, &root))
3033 root_mp->m_count++; /* pin it so it won't go away */
3035 detach_mnt(new_mnt, &parent_path);
3036 detach_mnt(root_mnt, &root_parent);
3037 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3038 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3039 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3041 /* mount old root on put_old */
3042 attach_mnt(root_mnt, old_mnt, old_mp);
3043 /* mount new_root on / */
3044 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3045 touch_mnt_namespace(current->nsproxy->mnt_ns);
3046 /* A moved mount should not expire automatically */
3047 list_del_init(&new_mnt->mnt_expire);
3048 unlock_mount_hash();
3049 chroot_fs_refs(&root, &new);
3050 put_mountpoint(root_mp);
3053 unlock_mount(old_mp);
3055 path_put(&root_parent);
3056 path_put(&parent_path);
3068 static void __init init_mount_tree(void)
3070 struct vfsmount *mnt;
3071 struct mnt_namespace *ns;
3073 struct file_system_type *type;
3075 type = get_fs_type("rootfs");
3077 panic("Can't find rootfs type");
3078 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3079 put_filesystem(type);
3081 panic("Can't create rootfs");
3083 ns = create_mnt_ns(mnt);
3085 panic("Can't allocate initial namespace");
3087 init_task.nsproxy->mnt_ns = ns;
3091 root.dentry = mnt->mnt_root;
3092 mnt->mnt_flags |= MNT_LOCKED;
3094 set_fs_pwd(current->fs, &root);
3095 set_fs_root(current->fs, &root);
3098 void __init mnt_init(void)
3103 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3104 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3106 mount_hashtable = alloc_large_system_hash("Mount-cache",
3107 sizeof(struct hlist_head),
3110 &m_hash_shift, &m_hash_mask, 0, 0);
3111 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3112 sizeof(struct hlist_head),
3115 &mp_hash_shift, &mp_hash_mask, 0, 0);
3117 if (!mount_hashtable || !mountpoint_hashtable)
3118 panic("Failed to allocate mount hash table\n");
3120 for (u = 0; u <= m_hash_mask; u++)
3121 INIT_HLIST_HEAD(&mount_hashtable[u]);
3122 for (u = 0; u <= mp_hash_mask; u++)
3123 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3129 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3131 fs_kobj = kobject_create_and_add("fs", NULL);
3133 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3138 void put_mnt_ns(struct mnt_namespace *ns)
3140 if (!atomic_dec_and_test(&ns->count))
3142 drop_collected_mounts(&ns->root->mnt);
3146 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3148 struct vfsmount *mnt;
3149 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3152 * it is a longterm mount, don't release mnt until
3153 * we unmount before file sys is unregistered
3155 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3159 EXPORT_SYMBOL_GPL(kern_mount_data);
3161 void kern_unmount(struct vfsmount *mnt)
3163 /* release long term mount so mount point can be released */
3164 if (!IS_ERR_OR_NULL(mnt)) {
3165 real_mount(mnt)->mnt_ns = NULL;
3166 synchronize_rcu(); /* yecchhh... */
3170 EXPORT_SYMBOL(kern_unmount);
3172 bool our_mnt(struct vfsmount *mnt)
3174 return check_mnt(real_mount(mnt));
3177 bool current_chrooted(void)
3179 /* Does the current process have a non-standard root */
3180 struct path ns_root;
3181 struct path fs_root;
3184 /* Find the namespace root */
3185 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3186 ns_root.dentry = ns_root.mnt->mnt_root;
3188 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3191 get_fs_root(current->fs, &fs_root);
3193 chrooted = !path_equal(&fs_root, &ns_root);
3201 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3203 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3204 int new_flags = *new_mnt_flags;
3206 bool visible = false;
3211 down_read(&namespace_sem);
3212 list_for_each_entry(mnt, &ns->list, mnt_list) {
3213 struct mount *child;
3214 if (mnt->mnt.mnt_sb->s_type != type)
3217 /* This mount is not fully visible if it's root directory
3218 * is not the root directory of the filesystem.
3220 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3223 /* Verify the mount flags are equal to or more permissive
3224 * than the proposed new mount.
3226 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
3227 !(new_flags & MNT_READONLY))
3229 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
3230 !(new_flags & MNT_NODEV))
3232 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
3233 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3236 /* This mount is not fully visible if there are any
3237 * locked child mounts that cover anything except for
3238 * empty directories.
3240 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3241 struct inode *inode = child->mnt_mountpoint->d_inode;
3242 /* Only worry about locked mounts */
3243 if (!(mnt->mnt.mnt_flags & MNT_LOCKED))
3245 /* Is the directory permanetly empty? */
3246 if (!is_empty_dir_inode(inode))
3249 /* Preserve the locked attributes */
3250 *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
3258 up_read(&namespace_sem);
3262 static struct ns_common *mntns_get(struct task_struct *task)
3264 struct ns_common *ns = NULL;
3265 struct nsproxy *nsproxy;
3268 nsproxy = task->nsproxy;
3270 ns = &nsproxy->mnt_ns->ns;
3271 get_mnt_ns(to_mnt_ns(ns));
3278 static void mntns_put(struct ns_common *ns)
3280 put_mnt_ns(to_mnt_ns(ns));
3283 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3285 struct fs_struct *fs = current->fs;
3286 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3289 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3290 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3291 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3298 put_mnt_ns(nsproxy->mnt_ns);
3299 nsproxy->mnt_ns = mnt_ns;
3302 root.mnt = &mnt_ns->root->mnt;
3303 root.dentry = mnt_ns->root->mnt.mnt_root;
3305 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3308 /* Update the pwd and root */
3309 set_fs_pwd(fs, &root);
3310 set_fs_root(fs, &root);
3316 const struct proc_ns_operations mntns_operations = {
3318 .type = CLONE_NEWNS,
3321 .install = mntns_install,