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_const(mnt->mnt_devname);
582 free_percpu(mnt->mnt_pcp);
584 kmem_cache_free(mnt_cache, mnt);
587 static void delayed_free_vfsmnt(struct rcu_head *head)
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
592 /* call under rcu_read_lock */
593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
596 if (read_seqretry(&mount_lock, seq))
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
615 * find the first mount at @dentry on vfsmount @mnt.
616 * call under rcu_read_lock()
618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
620 struct hlist_head *head = m_hash(mnt, dentry);
623 hlist_for_each_entry_rcu(p, head, mnt_hash)
624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
630 * find the last mount at @dentry on vfsmount @mnt.
631 * mount_lock must be held.
633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
635 struct mount *p, *res = NULL;
636 p = __lookup_mnt(mnt, dentry);
639 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
641 hlist_for_each_entry_continue(p, mnt_hash) {
642 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
644 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
652 * lookup_mnt - Return the first child mount mounted at path
654 * "First" means first mounted chronologically. If you create the
657 * mount /dev/sda1 /mnt
658 * mount /dev/sda2 /mnt
659 * mount /dev/sda3 /mnt
661 * Then lookup_mnt() on the base /mnt dentry in the root mount will
662 * return successively the root dentry and vfsmount of /dev/sda1, then
663 * /dev/sda2, then /dev/sda3, then NULL.
665 * lookup_mnt takes a reference to the found vfsmount.
667 struct vfsmount *lookup_mnt(struct path *path)
669 struct mount *child_mnt;
675 seq = read_seqbegin(&mount_lock);
676 child_mnt = __lookup_mnt(path->mnt, path->dentry);
677 m = child_mnt ? &child_mnt->mnt : NULL;
678 } while (!legitimize_mnt(m, seq));
684 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685 * current mount namespace.
687 * The common case is dentries are not mountpoints at all and that
688 * test is handled inline. For the slow case when we are actually
689 * dealing with a mountpoint of some kind, walk through all of the
690 * mounts in the current mount namespace and test to see if the dentry
693 * The mount_hashtable is not usable in the context because we
694 * need to identify all mounts that may be in the current mount
695 * namespace not just a mount that happens to have some specified
698 bool __is_local_mountpoint(struct dentry *dentry)
700 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
702 bool is_covered = false;
704 if (!d_mountpoint(dentry))
707 down_read(&namespace_sem);
708 list_for_each_entry(mnt, &ns->list, mnt_list) {
709 is_covered = (mnt->mnt_mountpoint == dentry);
713 up_read(&namespace_sem);
718 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
720 struct hlist_head *chain = mp_hash(dentry);
721 struct mountpoint *mp;
723 hlist_for_each_entry(mp, chain, m_hash) {
724 if (mp->m_dentry == dentry) {
725 /* might be worth a WARN_ON() */
726 if (d_unlinked(dentry))
727 return ERR_PTR(-ENOENT);
735 static struct mountpoint *new_mountpoint(struct dentry *dentry)
737 struct hlist_head *chain = mp_hash(dentry);
738 struct mountpoint *mp;
741 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
743 return ERR_PTR(-ENOMEM);
745 ret = d_set_mounted(dentry);
751 mp->m_dentry = dentry;
753 hlist_add_head(&mp->m_hash, chain);
754 INIT_HLIST_HEAD(&mp->m_list);
758 static void put_mountpoint(struct mountpoint *mp)
760 if (!--mp->m_count) {
761 struct dentry *dentry = mp->m_dentry;
762 BUG_ON(!hlist_empty(&mp->m_list));
763 spin_lock(&dentry->d_lock);
764 dentry->d_flags &= ~DCACHE_MOUNTED;
765 spin_unlock(&dentry->d_lock);
766 hlist_del(&mp->m_hash);
771 static inline int check_mnt(struct mount *mnt)
773 return mnt->mnt_ns == current->nsproxy->mnt_ns;
777 * vfsmount lock must be held for write
779 static void touch_mnt_namespace(struct mnt_namespace *ns)
783 wake_up_interruptible(&ns->poll);
788 * vfsmount lock must be held for write
790 static void __touch_mnt_namespace(struct mnt_namespace *ns)
792 if (ns && ns->event != event) {
794 wake_up_interruptible(&ns->poll);
799 * vfsmount lock must be held for write
801 static void unhash_mnt(struct mount *mnt)
803 mnt->mnt_parent = mnt;
804 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
805 list_del_init(&mnt->mnt_child);
806 hlist_del_init_rcu(&mnt->mnt_hash);
807 hlist_del_init(&mnt->mnt_mp_list);
808 put_mountpoint(mnt->mnt_mp);
813 * vfsmount lock must be held for write
815 static void detach_mnt(struct mount *mnt, struct path *old_path)
817 old_path->dentry = mnt->mnt_mountpoint;
818 old_path->mnt = &mnt->mnt_parent->mnt;
823 * vfsmount lock must be held for write
825 void mnt_set_mountpoint(struct mount *mnt,
826 struct mountpoint *mp,
827 struct mount *child_mnt)
830 mnt_add_count(mnt, 1); /* essentially, that's mntget */
831 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
832 child_mnt->mnt_parent = mnt;
833 child_mnt->mnt_mp = mp;
834 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
838 * vfsmount lock must be held for write
840 static void attach_mnt(struct mount *mnt,
841 struct mount *parent,
842 struct mountpoint *mp)
844 mnt_set_mountpoint(parent, mp, mnt);
845 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
846 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
849 static void attach_shadowed(struct mount *mnt,
850 struct mount *parent,
851 struct mount *shadows)
854 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
855 list_add(&mnt->mnt_child, &shadows->mnt_child);
857 hlist_add_head_rcu(&mnt->mnt_hash,
858 m_hash(&parent->mnt, mnt->mnt_mountpoint));
859 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
864 * vfsmount lock must be held for write
866 static void commit_tree(struct mount *mnt, struct mount *shadows)
868 struct mount *parent = mnt->mnt_parent;
871 struct mnt_namespace *n = parent->mnt_ns;
873 BUG_ON(parent == mnt);
875 list_add_tail(&head, &mnt->mnt_list);
876 list_for_each_entry(m, &head, mnt_list)
879 list_splice(&head, n->list.prev);
881 attach_shadowed(mnt, parent, shadows);
882 touch_mnt_namespace(n);
885 static struct mount *next_mnt(struct mount *p, struct mount *root)
887 struct list_head *next = p->mnt_mounts.next;
888 if (next == &p->mnt_mounts) {
892 next = p->mnt_child.next;
893 if (next != &p->mnt_parent->mnt_mounts)
898 return list_entry(next, struct mount, mnt_child);
901 static struct mount *skip_mnt_tree(struct mount *p)
903 struct list_head *prev = p->mnt_mounts.prev;
904 while (prev != &p->mnt_mounts) {
905 p = list_entry(prev, struct mount, mnt_child);
906 prev = p->mnt_mounts.prev;
912 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
918 return ERR_PTR(-ENODEV);
920 mnt = alloc_vfsmnt(name);
922 return ERR_PTR(-ENOMEM);
924 if (flags & MS_KERNMOUNT)
925 mnt->mnt.mnt_flags = MNT_INTERNAL;
927 root = mount_fs(type, flags, name, data);
931 return ERR_CAST(root);
934 mnt->mnt.mnt_root = root;
935 mnt->mnt.mnt_sb = root->d_sb;
936 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
937 mnt->mnt_parent = mnt;
939 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
943 EXPORT_SYMBOL_GPL(vfs_kern_mount);
945 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
948 struct super_block *sb = old->mnt.mnt_sb;
952 mnt = alloc_vfsmnt(old->mnt_devname);
954 return ERR_PTR(-ENOMEM);
956 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
957 mnt->mnt_group_id = 0; /* not a peer of original */
959 mnt->mnt_group_id = old->mnt_group_id;
961 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
962 err = mnt_alloc_group_id(mnt);
967 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
968 /* Don't allow unprivileged users to change mount flags */
969 if (flag & CL_UNPRIVILEGED) {
970 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
972 if (mnt->mnt.mnt_flags & MNT_READONLY)
973 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
975 if (mnt->mnt.mnt_flags & MNT_NODEV)
976 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
978 if (mnt->mnt.mnt_flags & MNT_NOSUID)
979 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
981 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
982 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
985 /* Don't allow unprivileged users to reveal what is under a mount */
986 if ((flag & CL_UNPRIVILEGED) &&
987 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
988 mnt->mnt.mnt_flags |= MNT_LOCKED;
990 atomic_inc(&sb->s_active);
991 mnt->mnt.mnt_sb = sb;
992 mnt->mnt.mnt_root = dget(root);
993 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
994 mnt->mnt_parent = mnt;
996 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
999 if ((flag & CL_SLAVE) ||
1000 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1001 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1002 mnt->mnt_master = old;
1003 CLEAR_MNT_SHARED(mnt);
1004 } else if (!(flag & CL_PRIVATE)) {
1005 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1006 list_add(&mnt->mnt_share, &old->mnt_share);
1007 if (IS_MNT_SLAVE(old))
1008 list_add(&mnt->mnt_slave, &old->mnt_slave);
1009 mnt->mnt_master = old->mnt_master;
1011 if (flag & CL_MAKE_SHARED)
1012 set_mnt_shared(mnt);
1014 /* stick the duplicate mount on the same expiry list
1015 * as the original if that was on one */
1016 if (flag & CL_EXPIRE) {
1017 if (!list_empty(&old->mnt_expire))
1018 list_add(&mnt->mnt_expire, &old->mnt_expire);
1026 return ERR_PTR(err);
1029 static void cleanup_mnt(struct mount *mnt)
1032 * This probably indicates that somebody messed
1033 * up a mnt_want/drop_write() pair. If this
1034 * happens, the filesystem was probably unable
1035 * to make r/w->r/o transitions.
1038 * The locking used to deal with mnt_count decrement provides barriers,
1039 * so mnt_get_writers() below is safe.
1041 WARN_ON(mnt_get_writers(mnt));
1042 if (unlikely(mnt->mnt_pins.first))
1044 fsnotify_vfsmount_delete(&mnt->mnt);
1045 dput(mnt->mnt.mnt_root);
1046 deactivate_super(mnt->mnt.mnt_sb);
1048 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1051 static void __cleanup_mnt(struct rcu_head *head)
1053 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1056 static LLIST_HEAD(delayed_mntput_list);
1057 static void delayed_mntput(struct work_struct *unused)
1059 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1060 struct llist_node *next;
1062 for (; node; node = next) {
1063 next = llist_next(node);
1064 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1067 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1069 static void mntput_no_expire(struct mount *mnt)
1072 mnt_add_count(mnt, -1);
1073 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1078 if (mnt_get_count(mnt)) {
1080 unlock_mount_hash();
1083 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1085 unlock_mount_hash();
1088 mnt->mnt.mnt_flags |= MNT_DOOMED;
1091 list_del(&mnt->mnt_instance);
1092 unlock_mount_hash();
1094 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1095 struct task_struct *task = current;
1096 if (likely(!(task->flags & PF_KTHREAD))) {
1097 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1098 if (!task_work_add(task, &mnt->mnt_rcu, true))
1101 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1102 schedule_delayed_work(&delayed_mntput_work, 1);
1108 void mntput(struct vfsmount *mnt)
1111 struct mount *m = real_mount(mnt);
1112 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1113 if (unlikely(m->mnt_expiry_mark))
1114 m->mnt_expiry_mark = 0;
1115 mntput_no_expire(m);
1118 EXPORT_SYMBOL(mntput);
1120 struct vfsmount *mntget(struct vfsmount *mnt)
1123 mnt_add_count(real_mount(mnt), 1);
1126 EXPORT_SYMBOL(mntget);
1128 struct vfsmount *mnt_clone_internal(struct path *path)
1131 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1134 p->mnt.mnt_flags |= MNT_INTERNAL;
1138 static inline void mangle(struct seq_file *m, const char *s)
1140 seq_escape(m, s, " \t\n\\");
1144 * Simple .show_options callback for filesystems which don't want to
1145 * implement more complex mount option showing.
1147 * See also save_mount_options().
1149 int generic_show_options(struct seq_file *m, struct dentry *root)
1151 const char *options;
1154 options = rcu_dereference(root->d_sb->s_options);
1156 if (options != NULL && options[0]) {
1164 EXPORT_SYMBOL(generic_show_options);
1167 * If filesystem uses generic_show_options(), this function should be
1168 * called from the fill_super() callback.
1170 * The .remount_fs callback usually needs to be handled in a special
1171 * way, to make sure, that previous options are not overwritten if the
1174 * Also note, that if the filesystem's .remount_fs function doesn't
1175 * reset all options to their default value, but changes only newly
1176 * given options, then the displayed options will not reflect reality
1179 void save_mount_options(struct super_block *sb, char *options)
1181 BUG_ON(sb->s_options);
1182 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1184 EXPORT_SYMBOL(save_mount_options);
1186 void replace_mount_options(struct super_block *sb, char *options)
1188 char *old = sb->s_options;
1189 rcu_assign_pointer(sb->s_options, options);
1195 EXPORT_SYMBOL(replace_mount_options);
1197 #ifdef CONFIG_PROC_FS
1198 /* iterator; we want it to have access to namespace_sem, thus here... */
1199 static void *m_start(struct seq_file *m, loff_t *pos)
1201 struct proc_mounts *p = proc_mounts(m);
1203 down_read(&namespace_sem);
1204 if (p->cached_event == p->ns->event) {
1205 void *v = p->cached_mount;
1206 if (*pos == p->cached_index)
1208 if (*pos == p->cached_index + 1) {
1209 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1210 return p->cached_mount = v;
1214 p->cached_event = p->ns->event;
1215 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1216 p->cached_index = *pos;
1217 return p->cached_mount;
1220 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1222 struct proc_mounts *p = proc_mounts(m);
1224 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1225 p->cached_index = *pos;
1226 return p->cached_mount;
1229 static void m_stop(struct seq_file *m, void *v)
1231 up_read(&namespace_sem);
1234 static int m_show(struct seq_file *m, void *v)
1236 struct proc_mounts *p = proc_mounts(m);
1237 struct mount *r = list_entry(v, struct mount, mnt_list);
1238 return p->show(m, &r->mnt);
1241 const struct seq_operations mounts_op = {
1247 #endif /* CONFIG_PROC_FS */
1250 * may_umount_tree - check if a mount tree is busy
1251 * @mnt: root of mount tree
1253 * This is called to check if a tree of mounts has any
1254 * open files, pwds, chroots or sub mounts that are
1257 int may_umount_tree(struct vfsmount *m)
1259 struct mount *mnt = real_mount(m);
1260 int actual_refs = 0;
1261 int minimum_refs = 0;
1265 /* write lock needed for mnt_get_count */
1267 for (p = mnt; p; p = next_mnt(p, mnt)) {
1268 actual_refs += mnt_get_count(p);
1271 unlock_mount_hash();
1273 if (actual_refs > minimum_refs)
1279 EXPORT_SYMBOL(may_umount_tree);
1282 * may_umount - check if a mount point is busy
1283 * @mnt: root of mount
1285 * This is called to check if a mount point has any
1286 * open files, pwds, chroots or sub mounts. If the
1287 * mount has sub mounts this will return busy
1288 * regardless of whether the sub mounts are busy.
1290 * Doesn't take quota and stuff into account. IOW, in some cases it will
1291 * give false negatives. The main reason why it's here is that we need
1292 * a non-destructive way to look for easily umountable filesystems.
1294 int may_umount(struct vfsmount *mnt)
1297 down_read(&namespace_sem);
1299 if (propagate_mount_busy(real_mount(mnt), 2))
1301 unlock_mount_hash();
1302 up_read(&namespace_sem);
1306 EXPORT_SYMBOL(may_umount);
1308 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1310 static void namespace_unlock(void)
1312 struct hlist_head head;
1314 hlist_move_list(&unmounted, &head);
1316 up_write(&namespace_sem);
1318 if (likely(hlist_empty(&head)))
1323 group_pin_kill(&head);
1326 static inline void namespace_lock(void)
1328 down_write(&namespace_sem);
1331 enum umount_tree_flags {
1333 UMOUNT_PROPAGATE = 2,
1336 * mount_lock must be held
1337 * namespace_sem must be held for write
1339 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1341 LIST_HEAD(tmp_list);
1344 if (how & UMOUNT_PROPAGATE)
1345 propagate_mount_unlock(mnt);
1347 /* Gather the mounts to umount */
1348 for (p = mnt; p; p = next_mnt(p, mnt)) {
1349 p->mnt.mnt_flags |= MNT_UMOUNT;
1350 list_move(&p->mnt_list, &tmp_list);
1353 /* Hide the mounts from mnt_mounts */
1354 list_for_each_entry(p, &tmp_list, mnt_list) {
1355 list_del_init(&p->mnt_child);
1358 /* Add propogated mounts to the tmp_list */
1359 if (how & UMOUNT_PROPAGATE)
1360 propagate_umount(&tmp_list);
1362 while (!list_empty(&tmp_list)) {
1363 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1364 list_del_init(&p->mnt_expire);
1365 list_del_init(&p->mnt_list);
1366 __touch_mnt_namespace(p->mnt_ns);
1368 if (how & UMOUNT_SYNC)
1369 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1371 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, &unmounted);
1372 if (mnt_has_parent(p)) {
1373 mnt_add_count(p->mnt_parent, -1);
1374 /* old mountpoint will be dropped when we can do that */
1375 p->mnt_ex_mountpoint = p->mnt_mountpoint;
1378 change_mnt_propagation(p, MS_PRIVATE);
1382 static void shrink_submounts(struct mount *mnt);
1384 static int do_umount(struct mount *mnt, int flags)
1386 struct super_block *sb = mnt->mnt.mnt_sb;
1389 retval = security_sb_umount(&mnt->mnt, flags);
1394 * Allow userspace to request a mountpoint be expired rather than
1395 * unmounting unconditionally. Unmount only happens if:
1396 * (1) the mark is already set (the mark is cleared by mntput())
1397 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1399 if (flags & MNT_EXPIRE) {
1400 if (&mnt->mnt == current->fs->root.mnt ||
1401 flags & (MNT_FORCE | MNT_DETACH))
1405 * probably don't strictly need the lock here if we examined
1406 * all race cases, but it's a slowpath.
1409 if (mnt_get_count(mnt) != 2) {
1410 unlock_mount_hash();
1413 unlock_mount_hash();
1415 if (!xchg(&mnt->mnt_expiry_mark, 1))
1420 * If we may have to abort operations to get out of this
1421 * mount, and they will themselves hold resources we must
1422 * allow the fs to do things. In the Unix tradition of
1423 * 'Gee thats tricky lets do it in userspace' the umount_begin
1424 * might fail to complete on the first run through as other tasks
1425 * must return, and the like. Thats for the mount program to worry
1426 * about for the moment.
1429 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1430 sb->s_op->umount_begin(sb);
1434 * No sense to grab the lock for this test, but test itself looks
1435 * somewhat bogus. Suggestions for better replacement?
1436 * Ho-hum... In principle, we might treat that as umount + switch
1437 * to rootfs. GC would eventually take care of the old vfsmount.
1438 * Actually it makes sense, especially if rootfs would contain a
1439 * /reboot - static binary that would close all descriptors and
1440 * call reboot(9). Then init(8) could umount root and exec /reboot.
1442 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1444 * Special case for "unmounting" root ...
1445 * we just try to remount it readonly.
1447 if (!capable(CAP_SYS_ADMIN))
1449 down_write(&sb->s_umount);
1450 if (!(sb->s_flags & MS_RDONLY))
1451 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1452 up_write(&sb->s_umount);
1460 if (flags & MNT_DETACH) {
1461 if (!list_empty(&mnt->mnt_list))
1462 umount_tree(mnt, UMOUNT_PROPAGATE);
1465 shrink_submounts(mnt);
1467 if (!propagate_mount_busy(mnt, 2)) {
1468 if (!list_empty(&mnt->mnt_list))
1469 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1473 unlock_mount_hash();
1479 * __detach_mounts - lazily unmount all mounts on the specified dentry
1481 * During unlink, rmdir, and d_drop it is possible to loose the path
1482 * to an existing mountpoint, and wind up leaking the mount.
1483 * detach_mounts allows lazily unmounting those mounts instead of
1486 * The caller may hold dentry->d_inode->i_mutex.
1488 void __detach_mounts(struct dentry *dentry)
1490 struct mountpoint *mp;
1494 mp = lookup_mountpoint(dentry);
1499 while (!hlist_empty(&mp->m_list)) {
1500 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1501 umount_tree(mnt, 0);
1503 unlock_mount_hash();
1510 * Is the caller allowed to modify his namespace?
1512 static inline bool may_mount(void)
1514 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1518 * Now umount can handle mount points as well as block devices.
1519 * This is important for filesystems which use unnamed block devices.
1521 * We now support a flag for forced unmount like the other 'big iron'
1522 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1525 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1530 int lookup_flags = 0;
1532 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1538 if (!(flags & UMOUNT_NOFOLLOW))
1539 lookup_flags |= LOOKUP_FOLLOW;
1541 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1544 mnt = real_mount(path.mnt);
1546 if (path.dentry != path.mnt->mnt_root)
1548 if (!check_mnt(mnt))
1550 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1553 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1556 retval = do_umount(mnt, flags);
1558 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1560 mntput_no_expire(mnt);
1565 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1568 * The 2.0 compatible umount. No flags.
1570 SYSCALL_DEFINE1(oldumount, char __user *, name)
1572 return sys_umount(name, 0);
1577 static bool is_mnt_ns_file(struct dentry *dentry)
1579 /* Is this a proxy for a mount namespace? */
1580 return dentry->d_op == &ns_dentry_operations &&
1581 dentry->d_fsdata == &mntns_operations;
1584 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1586 return container_of(ns, struct mnt_namespace, ns);
1589 static bool mnt_ns_loop(struct dentry *dentry)
1591 /* Could bind mounting the mount namespace inode cause a
1592 * mount namespace loop?
1594 struct mnt_namespace *mnt_ns;
1595 if (!is_mnt_ns_file(dentry))
1598 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1599 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1602 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1605 struct mount *res, *p, *q, *r, *parent;
1607 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1608 return ERR_PTR(-EINVAL);
1610 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1611 return ERR_PTR(-EINVAL);
1613 res = q = clone_mnt(mnt, dentry, flag);
1617 q->mnt_mountpoint = mnt->mnt_mountpoint;
1620 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1622 if (!is_subdir(r->mnt_mountpoint, dentry))
1625 for (s = r; s; s = next_mnt(s, r)) {
1626 struct mount *t = NULL;
1627 if (!(flag & CL_COPY_UNBINDABLE) &&
1628 IS_MNT_UNBINDABLE(s)) {
1629 s = skip_mnt_tree(s);
1632 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1633 is_mnt_ns_file(s->mnt.mnt_root)) {
1634 s = skip_mnt_tree(s);
1637 while (p != s->mnt_parent) {
1643 q = clone_mnt(p, p->mnt.mnt_root, flag);
1647 list_add_tail(&q->mnt_list, &res->mnt_list);
1648 mnt_set_mountpoint(parent, p->mnt_mp, q);
1649 if (!list_empty(&parent->mnt_mounts)) {
1650 t = list_last_entry(&parent->mnt_mounts,
1651 struct mount, mnt_child);
1652 if (t->mnt_mp != p->mnt_mp)
1655 attach_shadowed(q, parent, t);
1656 unlock_mount_hash();
1663 umount_tree(res, UMOUNT_SYNC);
1664 unlock_mount_hash();
1669 /* Caller should check returned pointer for errors */
1671 struct vfsmount *collect_mounts(struct path *path)
1675 if (!check_mnt(real_mount(path->mnt)))
1676 tree = ERR_PTR(-EINVAL);
1678 tree = copy_tree(real_mount(path->mnt), path->dentry,
1679 CL_COPY_ALL | CL_PRIVATE);
1682 return ERR_CAST(tree);
1686 void drop_collected_mounts(struct vfsmount *mnt)
1690 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1691 unlock_mount_hash();
1696 * clone_private_mount - create a private clone of a path
1698 * This creates a new vfsmount, which will be the clone of @path. The new will
1699 * not be attached anywhere in the namespace and will be private (i.e. changes
1700 * to the originating mount won't be propagated into this).
1702 * Release with mntput().
1704 struct vfsmount *clone_private_mount(struct path *path)
1706 struct mount *old_mnt = real_mount(path->mnt);
1707 struct mount *new_mnt;
1709 if (IS_MNT_UNBINDABLE(old_mnt))
1710 return ERR_PTR(-EINVAL);
1712 down_read(&namespace_sem);
1713 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1714 up_read(&namespace_sem);
1715 if (IS_ERR(new_mnt))
1716 return ERR_CAST(new_mnt);
1718 return &new_mnt->mnt;
1720 EXPORT_SYMBOL_GPL(clone_private_mount);
1722 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1723 struct vfsmount *root)
1726 int res = f(root, arg);
1729 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1730 res = f(&mnt->mnt, arg);
1737 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1741 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1742 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1743 mnt_release_group_id(p);
1747 static int invent_group_ids(struct mount *mnt, bool recurse)
1751 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1752 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1753 int err = mnt_alloc_group_id(p);
1755 cleanup_group_ids(mnt, p);
1765 * @source_mnt : mount tree to be attached
1766 * @nd : place the mount tree @source_mnt is attached
1767 * @parent_nd : if non-null, detach the source_mnt from its parent and
1768 * store the parent mount and mountpoint dentry.
1769 * (done when source_mnt is moved)
1771 * NOTE: in the table below explains the semantics when a source mount
1772 * of a given type is attached to a destination mount of a given type.
1773 * ---------------------------------------------------------------------------
1774 * | BIND MOUNT OPERATION |
1775 * |**************************************************************************
1776 * | source-->| shared | private | slave | unbindable |
1780 * |**************************************************************************
1781 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1783 * |non-shared| shared (+) | private | slave (*) | invalid |
1784 * ***************************************************************************
1785 * A bind operation clones the source mount and mounts the clone on the
1786 * destination mount.
1788 * (++) the cloned mount is propagated to all the mounts in the propagation
1789 * tree of the destination mount and the cloned mount is added to
1790 * the peer group of the source mount.
1791 * (+) the cloned mount is created under the destination mount and is marked
1792 * as shared. The cloned mount is added to the peer group of the source
1794 * (+++) the mount is propagated to all the mounts in the propagation tree
1795 * of the destination mount and the cloned mount is made slave
1796 * of the same master as that of the source mount. The cloned mount
1797 * is marked as 'shared and slave'.
1798 * (*) the cloned mount is made a slave of the same master as that of the
1801 * ---------------------------------------------------------------------------
1802 * | MOVE MOUNT OPERATION |
1803 * |**************************************************************************
1804 * | source-->| shared | private | slave | unbindable |
1808 * |**************************************************************************
1809 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1811 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1812 * ***************************************************************************
1814 * (+) the mount is moved to the destination. And is then propagated to
1815 * all the mounts in the propagation tree of the destination mount.
1816 * (+*) the mount is moved to the destination.
1817 * (+++) the mount is moved to the destination and is then propagated to
1818 * all the mounts belonging to the destination mount's propagation tree.
1819 * the mount is marked as 'shared and slave'.
1820 * (*) the mount continues to be a slave at the new location.
1822 * if the source mount is a tree, the operations explained above is
1823 * applied to each mount in the tree.
1824 * Must be called without spinlocks held, since this function can sleep
1827 static int attach_recursive_mnt(struct mount *source_mnt,
1828 struct mount *dest_mnt,
1829 struct mountpoint *dest_mp,
1830 struct path *parent_path)
1832 HLIST_HEAD(tree_list);
1833 struct mount *child, *p;
1834 struct hlist_node *n;
1837 if (IS_MNT_SHARED(dest_mnt)) {
1838 err = invent_group_ids(source_mnt, true);
1841 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1844 goto out_cleanup_ids;
1845 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1851 detach_mnt(source_mnt, parent_path);
1852 attach_mnt(source_mnt, dest_mnt, dest_mp);
1853 touch_mnt_namespace(source_mnt->mnt_ns);
1855 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1856 commit_tree(source_mnt, NULL);
1859 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1861 hlist_del_init(&child->mnt_hash);
1862 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1863 child->mnt_mountpoint);
1864 commit_tree(child, q);
1866 unlock_mount_hash();
1871 while (!hlist_empty(&tree_list)) {
1872 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1873 umount_tree(child, UMOUNT_SYNC);
1875 unlock_mount_hash();
1876 cleanup_group_ids(source_mnt, NULL);
1881 static struct mountpoint *lock_mount(struct path *path)
1883 struct vfsmount *mnt;
1884 struct dentry *dentry = path->dentry;
1886 mutex_lock(&dentry->d_inode->i_mutex);
1887 if (unlikely(cant_mount(dentry))) {
1888 mutex_unlock(&dentry->d_inode->i_mutex);
1889 return ERR_PTR(-ENOENT);
1892 mnt = lookup_mnt(path);
1894 struct mountpoint *mp = lookup_mountpoint(dentry);
1896 mp = new_mountpoint(dentry);
1899 mutex_unlock(&dentry->d_inode->i_mutex);
1905 mutex_unlock(&path->dentry->d_inode->i_mutex);
1908 dentry = path->dentry = dget(mnt->mnt_root);
1912 static void unlock_mount(struct mountpoint *where)
1914 struct dentry *dentry = where->m_dentry;
1915 put_mountpoint(where);
1917 mutex_unlock(&dentry->d_inode->i_mutex);
1920 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1922 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1925 if (d_is_dir(mp->m_dentry) !=
1926 d_is_dir(mnt->mnt.mnt_root))
1929 return attach_recursive_mnt(mnt, p, mp, NULL);
1933 * Sanity check the flags to change_mnt_propagation.
1936 static int flags_to_propagation_type(int flags)
1938 int type = flags & ~(MS_REC | MS_SILENT);
1940 /* Fail if any non-propagation flags are set */
1941 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1943 /* Only one propagation flag should be set */
1944 if (!is_power_of_2(type))
1950 * recursively change the type of the mountpoint.
1952 static int do_change_type(struct path *path, int flag)
1955 struct mount *mnt = real_mount(path->mnt);
1956 int recurse = flag & MS_REC;
1960 if (path->dentry != path->mnt->mnt_root)
1963 type = flags_to_propagation_type(flag);
1968 if (type == MS_SHARED) {
1969 err = invent_group_ids(mnt, recurse);
1975 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1976 change_mnt_propagation(m, type);
1977 unlock_mount_hash();
1984 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1986 struct mount *child;
1987 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1988 if (!is_subdir(child->mnt_mountpoint, dentry))
1991 if (child->mnt.mnt_flags & MNT_LOCKED)
1998 * do loopback mount.
2000 static int do_loopback(struct path *path, const char *old_name,
2003 struct path old_path;
2004 struct mount *mnt = NULL, *old, *parent;
2005 struct mountpoint *mp;
2007 if (!old_name || !*old_name)
2009 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2014 if (mnt_ns_loop(old_path.dentry))
2017 mp = lock_mount(path);
2022 old = real_mount(old_path.mnt);
2023 parent = real_mount(path->mnt);
2026 if (IS_MNT_UNBINDABLE(old))
2029 if (!check_mnt(parent))
2032 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2035 if (!recurse && has_locked_children(old, old_path.dentry))
2039 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2041 mnt = clone_mnt(old, old_path.dentry, 0);
2048 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2050 err = graft_tree(mnt, parent, mp);
2053 umount_tree(mnt, UMOUNT_SYNC);
2054 unlock_mount_hash();
2059 path_put(&old_path);
2063 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2066 int readonly_request = 0;
2068 if (ms_flags & MS_RDONLY)
2069 readonly_request = 1;
2070 if (readonly_request == __mnt_is_readonly(mnt))
2073 if (readonly_request)
2074 error = mnt_make_readonly(real_mount(mnt));
2076 __mnt_unmake_readonly(real_mount(mnt));
2081 * change filesystem flags. dir should be a physical root of filesystem.
2082 * If you've mounted a non-root directory somewhere and want to do remount
2083 * on it - tough luck.
2085 static int do_remount(struct path *path, int flags, int mnt_flags,
2089 struct super_block *sb = path->mnt->mnt_sb;
2090 struct mount *mnt = real_mount(path->mnt);
2092 if (!check_mnt(mnt))
2095 if (path->dentry != path->mnt->mnt_root)
2098 /* Don't allow changing of locked mnt flags.
2100 * No locks need to be held here while testing the various
2101 * MNT_LOCK flags because those flags can never be cleared
2102 * once they are set.
2104 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2105 !(mnt_flags & MNT_READONLY)) {
2108 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2109 !(mnt_flags & MNT_NODEV)) {
2110 /* Was the nodev implicitly added in mount? */
2111 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2112 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2113 mnt_flags |= MNT_NODEV;
2118 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2119 !(mnt_flags & MNT_NOSUID)) {
2122 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2123 !(mnt_flags & MNT_NOEXEC)) {
2126 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2127 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2131 err = security_sb_remount(sb, data);
2135 down_write(&sb->s_umount);
2136 if (flags & MS_BIND)
2137 err = change_mount_flags(path->mnt, flags);
2138 else if (!capable(CAP_SYS_ADMIN))
2141 err = do_remount_sb(sb, flags, data, 0);
2144 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2145 mnt->mnt.mnt_flags = mnt_flags;
2146 touch_mnt_namespace(mnt->mnt_ns);
2147 unlock_mount_hash();
2149 up_write(&sb->s_umount);
2153 static inline int tree_contains_unbindable(struct mount *mnt)
2156 for (p = mnt; p; p = next_mnt(p, mnt)) {
2157 if (IS_MNT_UNBINDABLE(p))
2163 static int do_move_mount(struct path *path, const char *old_name)
2165 struct path old_path, parent_path;
2168 struct mountpoint *mp;
2170 if (!old_name || !*old_name)
2172 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2176 mp = lock_mount(path);
2181 old = real_mount(old_path.mnt);
2182 p = real_mount(path->mnt);
2185 if (!check_mnt(p) || !check_mnt(old))
2188 if (old->mnt.mnt_flags & MNT_LOCKED)
2192 if (old_path.dentry != old_path.mnt->mnt_root)
2195 if (!mnt_has_parent(old))
2198 if (d_is_dir(path->dentry) !=
2199 d_is_dir(old_path.dentry))
2202 * Don't move a mount residing in a shared parent.
2204 if (IS_MNT_SHARED(old->mnt_parent))
2207 * Don't move a mount tree containing unbindable mounts to a destination
2208 * mount which is shared.
2210 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2213 for (; mnt_has_parent(p); p = p->mnt_parent)
2217 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2221 /* if the mount is moved, it should no longer be expire
2223 list_del_init(&old->mnt_expire);
2228 path_put(&parent_path);
2229 path_put(&old_path);
2233 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2236 const char *subtype = strchr(fstype, '.');
2245 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2247 if (!mnt->mnt_sb->s_subtype)
2253 return ERR_PTR(err);
2257 * add a mount into a namespace's mount tree
2259 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2261 struct mountpoint *mp;
2262 struct mount *parent;
2265 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2267 mp = lock_mount(path);
2271 parent = real_mount(path->mnt);
2273 if (unlikely(!check_mnt(parent))) {
2274 /* that's acceptable only for automounts done in private ns */
2275 if (!(mnt_flags & MNT_SHRINKABLE))
2277 /* ... and for those we'd better have mountpoint still alive */
2278 if (!parent->mnt_ns)
2282 /* Refuse the same filesystem on the same mount point */
2284 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2285 path->mnt->mnt_root == path->dentry)
2289 if (d_is_symlink(newmnt->mnt.mnt_root))
2292 newmnt->mnt.mnt_flags = mnt_flags;
2293 err = graft_tree(newmnt, parent, mp);
2301 * create a new mount for userspace and request it to be added into the
2304 static int do_new_mount(struct path *path, const char *fstype, int flags,
2305 int mnt_flags, const char *name, void *data)
2307 struct file_system_type *type;
2308 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2309 struct vfsmount *mnt;
2315 type = get_fs_type(fstype);
2319 if (user_ns != &init_user_ns) {
2320 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2321 put_filesystem(type);
2324 /* Only in special cases allow devices from mounts
2325 * created outside the initial user namespace.
2327 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2329 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2333 mnt = vfs_kern_mount(type, flags, name, data);
2334 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2335 !mnt->mnt_sb->s_subtype)
2336 mnt = fs_set_subtype(mnt, fstype);
2338 put_filesystem(type);
2340 return PTR_ERR(mnt);
2342 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2348 int finish_automount(struct vfsmount *m, struct path *path)
2350 struct mount *mnt = real_mount(m);
2352 /* The new mount record should have at least 2 refs to prevent it being
2353 * expired before we get a chance to add it
2355 BUG_ON(mnt_get_count(mnt) < 2);
2357 if (m->mnt_sb == path->mnt->mnt_sb &&
2358 m->mnt_root == path->dentry) {
2363 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2367 /* remove m from any expiration list it may be on */
2368 if (!list_empty(&mnt->mnt_expire)) {
2370 list_del_init(&mnt->mnt_expire);
2379 * mnt_set_expiry - Put a mount on an expiration list
2380 * @mnt: The mount to list.
2381 * @expiry_list: The list to add the mount to.
2383 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2387 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2391 EXPORT_SYMBOL(mnt_set_expiry);
2394 * process a list of expirable mountpoints with the intent of discarding any
2395 * mountpoints that aren't in use and haven't been touched since last we came
2398 void mark_mounts_for_expiry(struct list_head *mounts)
2400 struct mount *mnt, *next;
2401 LIST_HEAD(graveyard);
2403 if (list_empty(mounts))
2409 /* extract from the expiration list every vfsmount that matches the
2410 * following criteria:
2411 * - only referenced by its parent vfsmount
2412 * - still marked for expiry (marked on the last call here; marks are
2413 * cleared by mntput())
2415 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2416 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2417 propagate_mount_busy(mnt, 1))
2419 list_move(&mnt->mnt_expire, &graveyard);
2421 while (!list_empty(&graveyard)) {
2422 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2423 touch_mnt_namespace(mnt->mnt_ns);
2424 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2426 unlock_mount_hash();
2430 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2433 * Ripoff of 'select_parent()'
2435 * search the list of submounts for a given mountpoint, and move any
2436 * shrinkable submounts to the 'graveyard' list.
2438 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2440 struct mount *this_parent = parent;
2441 struct list_head *next;
2445 next = this_parent->mnt_mounts.next;
2447 while (next != &this_parent->mnt_mounts) {
2448 struct list_head *tmp = next;
2449 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2452 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2455 * Descend a level if the d_mounts list is non-empty.
2457 if (!list_empty(&mnt->mnt_mounts)) {
2462 if (!propagate_mount_busy(mnt, 1)) {
2463 list_move_tail(&mnt->mnt_expire, graveyard);
2468 * All done at this level ... ascend and resume the search
2470 if (this_parent != parent) {
2471 next = this_parent->mnt_child.next;
2472 this_parent = this_parent->mnt_parent;
2479 * process a list of expirable mountpoints with the intent of discarding any
2480 * submounts of a specific parent mountpoint
2482 * mount_lock must be held for write
2484 static void shrink_submounts(struct mount *mnt)
2486 LIST_HEAD(graveyard);
2489 /* extract submounts of 'mountpoint' from the expiration list */
2490 while (select_submounts(mnt, &graveyard)) {
2491 while (!list_empty(&graveyard)) {
2492 m = list_first_entry(&graveyard, struct mount,
2494 touch_mnt_namespace(m->mnt_ns);
2495 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2501 * Some copy_from_user() implementations do not return the exact number of
2502 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2503 * Note that this function differs from copy_from_user() in that it will oops
2504 * on bad values of `to', rather than returning a short copy.
2506 static long exact_copy_from_user(void *to, const void __user * from,
2510 const char __user *f = from;
2513 if (!access_ok(VERIFY_READ, from, n))
2517 if (__get_user(c, f)) {
2528 int copy_mount_options(const void __user * data, unsigned long *where)
2538 if (!(page = __get_free_page(GFP_KERNEL)))
2541 /* We only care that *some* data at the address the user
2542 * gave us is valid. Just in case, we'll zero
2543 * the remainder of the page.
2545 /* copy_from_user cannot cross TASK_SIZE ! */
2546 size = TASK_SIZE - (unsigned long)data;
2547 if (size > PAGE_SIZE)
2550 i = size - exact_copy_from_user((void *)page, data, size);
2556 memset((char *)page + i, 0, PAGE_SIZE - i);
2561 char *copy_mount_string(const void __user *data)
2563 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2567 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2568 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2570 * data is a (void *) that can point to any structure up to
2571 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2572 * information (or be NULL).
2574 * Pre-0.97 versions of mount() didn't have a flags word.
2575 * When the flags word was introduced its top half was required
2576 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2577 * Therefore, if this magic number is present, it carries no information
2578 * and must be discarded.
2580 long do_mount(const char *dev_name, const char __user *dir_name,
2581 const char *type_page, unsigned long flags, void *data_page)
2588 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2589 flags &= ~MS_MGC_MSK;
2591 /* Basic sanity checks */
2593 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2595 /* ... and get the mountpoint */
2596 retval = user_path(dir_name, &path);
2600 retval = security_sb_mount(dev_name, &path,
2601 type_page, flags, data_page);
2602 if (!retval && !may_mount())
2607 /* Default to relatime unless overriden */
2608 if (!(flags & MS_NOATIME))
2609 mnt_flags |= MNT_RELATIME;
2611 /* Separate the per-mountpoint flags */
2612 if (flags & MS_NOSUID)
2613 mnt_flags |= MNT_NOSUID;
2614 if (flags & MS_NODEV)
2615 mnt_flags |= MNT_NODEV;
2616 if (flags & MS_NOEXEC)
2617 mnt_flags |= MNT_NOEXEC;
2618 if (flags & MS_NOATIME)
2619 mnt_flags |= MNT_NOATIME;
2620 if (flags & MS_NODIRATIME)
2621 mnt_flags |= MNT_NODIRATIME;
2622 if (flags & MS_STRICTATIME)
2623 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2624 if (flags & MS_RDONLY)
2625 mnt_flags |= MNT_READONLY;
2627 /* The default atime for remount is preservation */
2628 if ((flags & MS_REMOUNT) &&
2629 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2630 MS_STRICTATIME)) == 0)) {
2631 mnt_flags &= ~MNT_ATIME_MASK;
2632 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2635 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2636 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2639 if (flags & MS_REMOUNT)
2640 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2642 else if (flags & MS_BIND)
2643 retval = do_loopback(&path, dev_name, flags & MS_REC);
2644 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2645 retval = do_change_type(&path, flags);
2646 else if (flags & MS_MOVE)
2647 retval = do_move_mount(&path, dev_name);
2649 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2650 dev_name, data_page);
2656 static void free_mnt_ns(struct mnt_namespace *ns)
2658 ns_free_inum(&ns->ns);
2659 put_user_ns(ns->user_ns);
2664 * Assign a sequence number so we can detect when we attempt to bind
2665 * mount a reference to an older mount namespace into the current
2666 * mount namespace, preventing reference counting loops. A 64bit
2667 * number incrementing at 10Ghz will take 12,427 years to wrap which
2668 * is effectively never, so we can ignore the possibility.
2670 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2672 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2674 struct mnt_namespace *new_ns;
2677 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2679 return ERR_PTR(-ENOMEM);
2680 ret = ns_alloc_inum(&new_ns->ns);
2683 return ERR_PTR(ret);
2685 new_ns->ns.ops = &mntns_operations;
2686 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2687 atomic_set(&new_ns->count, 1);
2688 new_ns->root = NULL;
2689 INIT_LIST_HEAD(&new_ns->list);
2690 init_waitqueue_head(&new_ns->poll);
2692 new_ns->user_ns = get_user_ns(user_ns);
2696 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2697 struct user_namespace *user_ns, struct fs_struct *new_fs)
2699 struct mnt_namespace *new_ns;
2700 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2701 struct mount *p, *q;
2708 if (likely(!(flags & CLONE_NEWNS))) {
2715 new_ns = alloc_mnt_ns(user_ns);
2720 /* First pass: copy the tree topology */
2721 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2722 if (user_ns != ns->user_ns)
2723 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2724 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2727 free_mnt_ns(new_ns);
2728 return ERR_CAST(new);
2731 list_add_tail(&new_ns->list, &new->mnt_list);
2734 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2735 * as belonging to new namespace. We have already acquired a private
2736 * fs_struct, so tsk->fs->lock is not needed.
2743 if (&p->mnt == new_fs->root.mnt) {
2744 new_fs->root.mnt = mntget(&q->mnt);
2747 if (&p->mnt == new_fs->pwd.mnt) {
2748 new_fs->pwd.mnt = mntget(&q->mnt);
2752 p = next_mnt(p, old);
2753 q = next_mnt(q, new);
2756 while (p->mnt.mnt_root != q->mnt.mnt_root)
2757 p = next_mnt(p, old);
2770 * create_mnt_ns - creates a private namespace and adds a root filesystem
2771 * @mnt: pointer to the new root filesystem mountpoint
2773 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2775 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2776 if (!IS_ERR(new_ns)) {
2777 struct mount *mnt = real_mount(m);
2778 mnt->mnt_ns = new_ns;
2780 list_add(&mnt->mnt_list, &new_ns->list);
2787 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2789 struct mnt_namespace *ns;
2790 struct super_block *s;
2794 ns = create_mnt_ns(mnt);
2796 return ERR_CAST(ns);
2798 err = vfs_path_lookup(mnt->mnt_root, mnt,
2799 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2804 return ERR_PTR(err);
2806 /* trade a vfsmount reference for active sb one */
2807 s = path.mnt->mnt_sb;
2808 atomic_inc(&s->s_active);
2810 /* lock the sucker */
2811 down_write(&s->s_umount);
2812 /* ... and return the root of (sub)tree on it */
2815 EXPORT_SYMBOL(mount_subtree);
2817 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2818 char __user *, type, unsigned long, flags, void __user *, data)
2823 unsigned long data_page;
2825 kernel_type = copy_mount_string(type);
2826 ret = PTR_ERR(kernel_type);
2827 if (IS_ERR(kernel_type))
2830 kernel_dev = copy_mount_string(dev_name);
2831 ret = PTR_ERR(kernel_dev);
2832 if (IS_ERR(kernel_dev))
2835 ret = copy_mount_options(data, &data_page);
2839 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2840 (void *) data_page);
2842 free_page(data_page);
2852 * Return true if path is reachable from root
2854 * namespace_sem or mount_lock is held
2856 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2857 const struct path *root)
2859 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2860 dentry = mnt->mnt_mountpoint;
2861 mnt = mnt->mnt_parent;
2863 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2866 int path_is_under(struct path *path1, struct path *path2)
2869 read_seqlock_excl(&mount_lock);
2870 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2871 read_sequnlock_excl(&mount_lock);
2874 EXPORT_SYMBOL(path_is_under);
2877 * pivot_root Semantics:
2878 * Moves the root file system of the current process to the directory put_old,
2879 * makes new_root as the new root file system of the current process, and sets
2880 * root/cwd of all processes which had them on the current root to new_root.
2883 * The new_root and put_old must be directories, and must not be on the
2884 * same file system as the current process root. The put_old must be
2885 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2886 * pointed to by put_old must yield the same directory as new_root. No other
2887 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2889 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2890 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2891 * in this situation.
2894 * - we don't move root/cwd if they are not at the root (reason: if something
2895 * cared enough to change them, it's probably wrong to force them elsewhere)
2896 * - it's okay to pick a root that isn't the root of a file system, e.g.
2897 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2898 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2901 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2902 const char __user *, put_old)
2904 struct path new, old, parent_path, root_parent, root;
2905 struct mount *new_mnt, *root_mnt, *old_mnt;
2906 struct mountpoint *old_mp, *root_mp;
2912 error = user_path_dir(new_root, &new);
2916 error = user_path_dir(put_old, &old);
2920 error = security_sb_pivotroot(&old, &new);
2924 get_fs_root(current->fs, &root);
2925 old_mp = lock_mount(&old);
2926 error = PTR_ERR(old_mp);
2931 new_mnt = real_mount(new.mnt);
2932 root_mnt = real_mount(root.mnt);
2933 old_mnt = real_mount(old.mnt);
2934 if (IS_MNT_SHARED(old_mnt) ||
2935 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2936 IS_MNT_SHARED(root_mnt->mnt_parent))
2938 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2940 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2943 if (d_unlinked(new.dentry))
2946 if (new_mnt == root_mnt || old_mnt == root_mnt)
2947 goto out4; /* loop, on the same file system */
2949 if (root.mnt->mnt_root != root.dentry)
2950 goto out4; /* not a mountpoint */
2951 if (!mnt_has_parent(root_mnt))
2952 goto out4; /* not attached */
2953 root_mp = root_mnt->mnt_mp;
2954 if (new.mnt->mnt_root != new.dentry)
2955 goto out4; /* not a mountpoint */
2956 if (!mnt_has_parent(new_mnt))
2957 goto out4; /* not attached */
2958 /* make sure we can reach put_old from new_root */
2959 if (!is_path_reachable(old_mnt, old.dentry, &new))
2961 /* make certain new is below the root */
2962 if (!is_path_reachable(new_mnt, new.dentry, &root))
2964 root_mp->m_count++; /* pin it so it won't go away */
2966 detach_mnt(new_mnt, &parent_path);
2967 detach_mnt(root_mnt, &root_parent);
2968 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2969 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2970 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2972 /* mount old root on put_old */
2973 attach_mnt(root_mnt, old_mnt, old_mp);
2974 /* mount new_root on / */
2975 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2976 touch_mnt_namespace(current->nsproxy->mnt_ns);
2977 /* A moved mount should not expire automatically */
2978 list_del_init(&new_mnt->mnt_expire);
2979 unlock_mount_hash();
2980 chroot_fs_refs(&root, &new);
2981 put_mountpoint(root_mp);
2984 unlock_mount(old_mp);
2986 path_put(&root_parent);
2987 path_put(&parent_path);
2999 static void __init init_mount_tree(void)
3001 struct vfsmount *mnt;
3002 struct mnt_namespace *ns;
3004 struct file_system_type *type;
3006 type = get_fs_type("rootfs");
3008 panic("Can't find rootfs type");
3009 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3010 put_filesystem(type);
3012 panic("Can't create rootfs");
3014 ns = create_mnt_ns(mnt);
3016 panic("Can't allocate initial namespace");
3018 init_task.nsproxy->mnt_ns = ns;
3022 root.dentry = mnt->mnt_root;
3023 mnt->mnt_flags |= MNT_LOCKED;
3025 set_fs_pwd(current->fs, &root);
3026 set_fs_root(current->fs, &root);
3029 void __init mnt_init(void)
3034 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3035 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3037 mount_hashtable = alloc_large_system_hash("Mount-cache",
3038 sizeof(struct hlist_head),
3041 &m_hash_shift, &m_hash_mask, 0, 0);
3042 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3043 sizeof(struct hlist_head),
3046 &mp_hash_shift, &mp_hash_mask, 0, 0);
3048 if (!mount_hashtable || !mountpoint_hashtable)
3049 panic("Failed to allocate mount hash table\n");
3051 for (u = 0; u <= m_hash_mask; u++)
3052 INIT_HLIST_HEAD(&mount_hashtable[u]);
3053 for (u = 0; u <= mp_hash_mask; u++)
3054 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3060 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3062 fs_kobj = kobject_create_and_add("fs", NULL);
3064 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3069 void put_mnt_ns(struct mnt_namespace *ns)
3071 if (!atomic_dec_and_test(&ns->count))
3073 drop_collected_mounts(&ns->root->mnt);
3077 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3079 struct vfsmount *mnt;
3080 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3083 * it is a longterm mount, don't release mnt until
3084 * we unmount before file sys is unregistered
3086 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3090 EXPORT_SYMBOL_GPL(kern_mount_data);
3092 void kern_unmount(struct vfsmount *mnt)
3094 /* release long term mount so mount point can be released */
3095 if (!IS_ERR_OR_NULL(mnt)) {
3096 real_mount(mnt)->mnt_ns = NULL;
3097 synchronize_rcu(); /* yecchhh... */
3101 EXPORT_SYMBOL(kern_unmount);
3103 bool our_mnt(struct vfsmount *mnt)
3105 return check_mnt(real_mount(mnt));
3108 bool current_chrooted(void)
3110 /* Does the current process have a non-standard root */
3111 struct path ns_root;
3112 struct path fs_root;
3115 /* Find the namespace root */
3116 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3117 ns_root.dentry = ns_root.mnt->mnt_root;
3119 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3122 get_fs_root(current->fs, &fs_root);
3124 chrooted = !path_equal(&fs_root, &ns_root);
3132 bool fs_fully_visible(struct file_system_type *type)
3134 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3136 bool visible = false;
3141 down_read(&namespace_sem);
3142 list_for_each_entry(mnt, &ns->list, mnt_list) {
3143 struct mount *child;
3144 if (mnt->mnt.mnt_sb->s_type != type)
3147 /* This mount is not fully visible if there are any child mounts
3148 * that cover anything except for empty directories.
3150 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3151 struct inode *inode = child->mnt_mountpoint->d_inode;
3152 if (!S_ISDIR(inode->i_mode))
3154 if (inode->i_nlink > 2)
3162 up_read(&namespace_sem);
3166 static struct ns_common *mntns_get(struct task_struct *task)
3168 struct ns_common *ns = NULL;
3169 struct nsproxy *nsproxy;
3172 nsproxy = task->nsproxy;
3174 ns = &nsproxy->mnt_ns->ns;
3175 get_mnt_ns(to_mnt_ns(ns));
3182 static void mntns_put(struct ns_common *ns)
3184 put_mnt_ns(to_mnt_ns(ns));
3187 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3189 struct fs_struct *fs = current->fs;
3190 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3193 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3194 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3195 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3202 put_mnt_ns(nsproxy->mnt_ns);
3203 nsproxy->mnt_ns = mnt_ns;
3206 root.mnt = &mnt_ns->root->mnt;
3207 root.dentry = mnt_ns->root->mnt.mnt_root;
3209 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3212 /* Update the pwd and root */
3213 set_fs_pwd(fs, &root);
3214 set_fs_root(fs, &root);
3220 const struct proc_ns_operations mntns_operations = {
3222 .type = CLONE_NEWNS,
3225 .install = mntns_install,