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 /* Maximum number of mounts in a mount namespace */
31 unsigned int sysctl_mount_max __read_mostly = 100000;
33 static unsigned int m_hash_mask __read_mostly;
34 static unsigned int m_hash_shift __read_mostly;
35 static unsigned int mp_hash_mask __read_mostly;
36 static unsigned int mp_hash_shift __read_mostly;
38 static __initdata unsigned long mhash_entries;
39 static int __init set_mhash_entries(char *str)
43 mhash_entries = simple_strtoul(str, &str, 0);
46 __setup("mhash_entries=", set_mhash_entries);
48 static __initdata unsigned long mphash_entries;
49 static int __init set_mphash_entries(char *str)
53 mphash_entries = simple_strtoul(str, &str, 0);
56 __setup("mphash_entries=", set_mphash_entries);
59 static DEFINE_IDA(mnt_id_ida);
60 static DEFINE_IDA(mnt_group_ida);
61 static DEFINE_SPINLOCK(mnt_id_lock);
62 static int mnt_id_start = 0;
63 static int mnt_group_start = 1;
65 static struct hlist_head *mount_hashtable __read_mostly;
66 static struct hlist_head *mountpoint_hashtable __read_mostly;
67 static struct kmem_cache *mnt_cache __read_mostly;
68 static DECLARE_RWSEM(namespace_sem);
71 struct kobject *fs_kobj;
72 EXPORT_SYMBOL_GPL(fs_kobj);
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
84 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
86 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
88 tmp = tmp + (tmp >> m_hash_shift);
89 return &mount_hashtable[tmp & m_hash_mask];
92 static inline struct hlist_head *mp_hash(struct dentry *dentry)
94 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> mp_hash_shift);
96 return &mountpoint_hashtable[tmp & mp_hash_mask];
99 static int mnt_alloc_id(struct mount *mnt)
104 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
105 spin_lock(&mnt_id_lock);
106 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
108 mnt_id_start = mnt->mnt_id + 1;
109 spin_unlock(&mnt_id_lock);
116 static void mnt_free_id(struct mount *mnt)
118 int id = mnt->mnt_id;
119 spin_lock(&mnt_id_lock);
120 ida_remove(&mnt_id_ida, id);
121 if (mnt_id_start > id)
123 spin_unlock(&mnt_id_lock);
127 * Allocate a new peer group ID
129 * mnt_group_ida is protected by namespace_sem
131 static int mnt_alloc_group_id(struct mount *mnt)
135 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
138 res = ida_get_new_above(&mnt_group_ida,
142 mnt_group_start = mnt->mnt_group_id + 1;
148 * Release a peer group ID
150 void mnt_release_group_id(struct mount *mnt)
152 int id = mnt->mnt_group_id;
153 ida_remove(&mnt_group_ida, id);
154 if (mnt_group_start > id)
155 mnt_group_start = id;
156 mnt->mnt_group_id = 0;
160 * vfsmount lock must be held for read
162 static inline void mnt_add_count(struct mount *mnt, int n)
165 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
174 * vfsmount lock must be held for write
176 unsigned int mnt_get_count(struct mount *mnt)
179 unsigned int count = 0;
182 for_each_possible_cpu(cpu) {
183 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
188 return mnt->mnt_count;
192 static void drop_mountpoint(struct fs_pin *p)
194 struct mount *m = container_of(p, struct mount, mnt_umount);
195 dput(m->mnt_ex_mountpoint);
200 static struct mount *alloc_vfsmnt(const char *name)
202 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
206 err = mnt_alloc_id(mnt);
211 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
212 if (!mnt->mnt_devname)
217 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
219 goto out_free_devname;
221 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
224 mnt->mnt_writers = 0;
227 INIT_HLIST_NODE(&mnt->mnt_hash);
228 INIT_LIST_HEAD(&mnt->mnt_child);
229 INIT_LIST_HEAD(&mnt->mnt_mounts);
230 INIT_LIST_HEAD(&mnt->mnt_list);
231 INIT_LIST_HEAD(&mnt->mnt_expire);
232 INIT_LIST_HEAD(&mnt->mnt_share);
233 INIT_LIST_HEAD(&mnt->mnt_slave_list);
234 INIT_LIST_HEAD(&mnt->mnt_slave);
235 INIT_HLIST_NODE(&mnt->mnt_mp_list);
236 #ifdef CONFIG_FSNOTIFY
237 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
239 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
245 kfree_const(mnt->mnt_devname);
250 kmem_cache_free(mnt_cache, mnt);
255 * Most r/o checks on a fs are for operations that take
256 * discrete amounts of time, like a write() or unlink().
257 * We must keep track of when those operations start
258 * (for permission checks) and when they end, so that
259 * we can determine when writes are able to occur to
263 * __mnt_is_readonly: check whether a mount is read-only
264 * @mnt: the mount to check for its write status
266 * This shouldn't be used directly ouside of the VFS.
267 * It does not guarantee that the filesystem will stay
268 * r/w, just that it is right *now*. This can not and
269 * should not be used in place of IS_RDONLY(inode).
270 * mnt_want/drop_write() will _keep_ the filesystem
273 int __mnt_is_readonly(struct vfsmount *mnt)
275 if (mnt->mnt_flags & MNT_READONLY)
277 if (mnt->mnt_sb->s_flags & MS_RDONLY)
281 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283 static inline void mnt_inc_writers(struct mount *mnt)
286 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
292 static inline void mnt_dec_writers(struct mount *mnt)
295 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
301 static unsigned int mnt_get_writers(struct mount *mnt)
304 unsigned int count = 0;
307 for_each_possible_cpu(cpu) {
308 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
313 return mnt->mnt_writers;
317 static int mnt_is_readonly(struct vfsmount *mnt)
319 if (mnt->mnt_sb->s_readonly_remount)
321 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 return __mnt_is_readonly(mnt);
327 * Most r/o & frozen checks on a fs are for operations that take discrete
328 * amounts of time, like a write() or unlink(). We must keep track of when
329 * those operations start (for permission checks) and when they end, so that we
330 * can determine when writes are able to occur to a filesystem.
333 * __mnt_want_write - get write access to a mount without freeze protection
334 * @m: the mount on which to take a write
336 * This tells the low-level filesystem that a write is about to be performed to
337 * it, and makes sure that writes are allowed (mnt it read-write) before
338 * returning success. This operation does not protect against filesystem being
339 * frozen. When the write operation is finished, __mnt_drop_write() must be
340 * called. This is effectively a refcount.
342 int __mnt_want_write(struct vfsmount *m)
344 struct mount *mnt = real_mount(m);
348 mnt_inc_writers(mnt);
350 * The store to mnt_inc_writers must be visible before we pass
351 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 * incremented count after it has set MNT_WRITE_HOLD.
355 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
358 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
359 * be set to match its requirements. So we must not load that until
360 * MNT_WRITE_HOLD is cleared.
363 if (mnt_is_readonly(m)) {
364 mnt_dec_writers(mnt);
373 * mnt_want_write - get write access to a mount
374 * @m: the mount on which to take a write
376 * This tells the low-level filesystem that a write is about to be performed to
377 * it, and makes sure that writes are allowed (mount is read-write, filesystem
378 * is not frozen) before returning success. When the write operation is
379 * finished, mnt_drop_write() must be called. This is effectively a refcount.
381 int mnt_want_write(struct vfsmount *m)
385 sb_start_write(m->mnt_sb);
386 ret = __mnt_want_write(m);
388 sb_end_write(m->mnt_sb);
391 EXPORT_SYMBOL_GPL(mnt_want_write);
394 * mnt_clone_write - get write access to a mount
395 * @mnt: the mount on which to take a write
397 * This is effectively like mnt_want_write, except
398 * it must only be used to take an extra write reference
399 * on a mountpoint that we already know has a write reference
400 * on it. This allows some optimisation.
402 * After finished, mnt_drop_write must be called as usual to
403 * drop the reference.
405 int mnt_clone_write(struct vfsmount *mnt)
407 /* superblock may be r/o */
408 if (__mnt_is_readonly(mnt))
411 mnt_inc_writers(real_mount(mnt));
415 EXPORT_SYMBOL_GPL(mnt_clone_write);
418 * __mnt_want_write_file - get write access to a file's mount
419 * @file: the file who's mount on which to take a write
421 * This is like __mnt_want_write, but it takes a file and can
422 * do some optimisations if the file is open for write already
424 int __mnt_want_write_file(struct file *file)
426 if (!(file->f_mode & FMODE_WRITER))
427 return __mnt_want_write(file->f_path.mnt);
429 return mnt_clone_write(file->f_path.mnt);
433 * mnt_want_write_file - get write access to a file's mount
434 * @file: the file who's mount on which to take a write
436 * This is like mnt_want_write, but it takes a file and can
437 * do some optimisations if the file is open for write already
439 int mnt_want_write_file(struct file *file)
443 sb_start_write(file->f_path.mnt->mnt_sb);
444 ret = __mnt_want_write_file(file);
446 sb_end_write(file->f_path.mnt->mnt_sb);
449 EXPORT_SYMBOL_GPL(mnt_want_write_file);
452 * __mnt_drop_write - give up write access to a mount
453 * @mnt: the mount on which to give up write access
455 * Tells the low-level filesystem that we are done
456 * performing writes to it. Must be matched with
457 * __mnt_want_write() call above.
459 void __mnt_drop_write(struct vfsmount *mnt)
462 mnt_dec_writers(real_mount(mnt));
467 * mnt_drop_write - give up write access to a mount
468 * @mnt: the mount on which to give up write access
470 * Tells the low-level filesystem that we are done performing writes to it and
471 * also allows filesystem to be frozen again. Must be matched with
472 * mnt_want_write() call above.
474 void mnt_drop_write(struct vfsmount *mnt)
476 __mnt_drop_write(mnt);
477 sb_end_write(mnt->mnt_sb);
479 EXPORT_SYMBOL_GPL(mnt_drop_write);
481 void __mnt_drop_write_file(struct file *file)
483 __mnt_drop_write(file->f_path.mnt);
486 void mnt_drop_write_file(struct file *file)
488 mnt_drop_write(file->f_path.mnt);
490 EXPORT_SYMBOL(mnt_drop_write_file);
492 static int mnt_make_readonly(struct mount *mnt)
497 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
499 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
500 * should be visible before we do.
505 * With writers on hold, if this value is zero, then there are
506 * definitely no active writers (although held writers may subsequently
507 * increment the count, they'll have to wait, and decrement it after
508 * seeing MNT_READONLY).
510 * It is OK to have counter incremented on one CPU and decremented on
511 * another: the sum will add up correctly. The danger would be when we
512 * sum up each counter, if we read a counter before it is incremented,
513 * but then read another CPU's count which it has been subsequently
514 * decremented from -- we would see more decrements than we should.
515 * MNT_WRITE_HOLD protects against this scenario, because
516 * mnt_want_write first increments count, then smp_mb, then spins on
517 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
518 * we're counting up here.
520 if (mnt_get_writers(mnt) > 0)
523 mnt->mnt.mnt_flags |= MNT_READONLY;
525 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
526 * that become unheld will see MNT_READONLY.
529 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
534 static void __mnt_unmake_readonly(struct mount *mnt)
537 mnt->mnt.mnt_flags &= ~MNT_READONLY;
541 int sb_prepare_remount_readonly(struct super_block *sb)
546 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
547 if (atomic_long_read(&sb->s_remove_count))
551 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
552 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
553 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
555 if (mnt_get_writers(mnt) > 0) {
561 if (!err && atomic_long_read(&sb->s_remove_count))
565 sb->s_readonly_remount = 1;
568 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
569 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
570 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
577 static void free_vfsmnt(struct mount *mnt)
579 kfree_const(mnt->mnt_devname);
581 free_percpu(mnt->mnt_pcp);
583 kmem_cache_free(mnt_cache, mnt);
586 static void delayed_free_vfsmnt(struct rcu_head *head)
588 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
591 /* call under rcu_read_lock */
592 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
595 if (read_seqretry(&mount_lock, seq))
599 mnt = real_mount(bastard);
600 mnt_add_count(mnt, 1);
601 if (likely(!read_seqretry(&mount_lock, seq)))
603 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
604 mnt_add_count(mnt, -1);
610 /* call under rcu_read_lock */
611 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
613 int res = __legitimize_mnt(bastard, seq);
616 if (unlikely(res < 0)) {
625 * find the first mount at @dentry on vfsmount @mnt.
626 * call under rcu_read_lock()
628 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
630 struct hlist_head *head = m_hash(mnt, dentry);
633 hlist_for_each_entry_rcu(p, head, mnt_hash)
634 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
640 * find the last mount at @dentry on vfsmount @mnt.
641 * mount_lock must be held.
643 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
645 struct mount *p, *res = NULL;
646 p = __lookup_mnt(mnt, dentry);
649 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
651 hlist_for_each_entry_continue(p, mnt_hash) {
652 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
654 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
662 * lookup_mnt - Return the first child mount mounted at path
664 * "First" means first mounted chronologically. If you create the
667 * mount /dev/sda1 /mnt
668 * mount /dev/sda2 /mnt
669 * mount /dev/sda3 /mnt
671 * Then lookup_mnt() on the base /mnt dentry in the root mount will
672 * return successively the root dentry and vfsmount of /dev/sda1, then
673 * /dev/sda2, then /dev/sda3, then NULL.
675 * lookup_mnt takes a reference to the found vfsmount.
677 struct vfsmount *lookup_mnt(struct path *path)
679 struct mount *child_mnt;
685 seq = read_seqbegin(&mount_lock);
686 child_mnt = __lookup_mnt(path->mnt, path->dentry);
687 m = child_mnt ? &child_mnt->mnt : NULL;
688 } while (!legitimize_mnt(m, seq));
694 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
695 * current mount namespace.
697 * The common case is dentries are not mountpoints at all and that
698 * test is handled inline. For the slow case when we are actually
699 * dealing with a mountpoint of some kind, walk through all of the
700 * mounts in the current mount namespace and test to see if the dentry
703 * The mount_hashtable is not usable in the context because we
704 * need to identify all mounts that may be in the current mount
705 * namespace not just a mount that happens to have some specified
708 bool __is_local_mountpoint(struct dentry *dentry)
710 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
712 bool is_covered = false;
714 if (!d_mountpoint(dentry))
717 down_read(&namespace_sem);
718 list_for_each_entry(mnt, &ns->list, mnt_list) {
719 is_covered = (mnt->mnt_mountpoint == dentry);
723 up_read(&namespace_sem);
728 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
730 struct hlist_head *chain = mp_hash(dentry);
731 struct mountpoint *mp;
733 hlist_for_each_entry(mp, chain, m_hash) {
734 if (mp->m_dentry == dentry) {
735 /* might be worth a WARN_ON() */
736 if (d_unlinked(dentry))
737 return ERR_PTR(-ENOENT);
745 static struct mountpoint *new_mountpoint(struct dentry *dentry)
747 struct hlist_head *chain = mp_hash(dentry);
748 struct mountpoint *mp;
751 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
753 return ERR_PTR(-ENOMEM);
755 ret = d_set_mounted(dentry);
761 mp->m_dentry = dentry;
763 hlist_add_head(&mp->m_hash, chain);
764 INIT_HLIST_HEAD(&mp->m_list);
768 static void put_mountpoint(struct mountpoint *mp)
770 if (!--mp->m_count) {
771 struct dentry *dentry = mp->m_dentry;
772 BUG_ON(!hlist_empty(&mp->m_list));
773 spin_lock(&dentry->d_lock);
774 dentry->d_flags &= ~DCACHE_MOUNTED;
775 spin_unlock(&dentry->d_lock);
776 hlist_del(&mp->m_hash);
781 static inline int check_mnt(struct mount *mnt)
783 return mnt->mnt_ns == current->nsproxy->mnt_ns;
787 * vfsmount lock must be held for write
789 static void touch_mnt_namespace(struct mnt_namespace *ns)
793 wake_up_interruptible(&ns->poll);
798 * vfsmount lock must be held for write
800 static void __touch_mnt_namespace(struct mnt_namespace *ns)
802 if (ns && ns->event != event) {
804 wake_up_interruptible(&ns->poll);
809 * vfsmount lock must be held for write
811 static void unhash_mnt(struct mount *mnt)
813 mnt->mnt_parent = mnt;
814 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
815 list_del_init(&mnt->mnt_child);
816 hlist_del_init_rcu(&mnt->mnt_hash);
817 hlist_del_init(&mnt->mnt_mp_list);
818 put_mountpoint(mnt->mnt_mp);
823 * vfsmount lock must be held for write
825 static void detach_mnt(struct mount *mnt, struct path *old_path)
827 old_path->dentry = mnt->mnt_mountpoint;
828 old_path->mnt = &mnt->mnt_parent->mnt;
833 * vfsmount lock must be held for write
835 static void umount_mnt(struct mount *mnt)
837 /* old mountpoint will be dropped when we can do that */
838 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
843 * vfsmount lock must be held for write
845 void mnt_set_mountpoint(struct mount *mnt,
846 struct mountpoint *mp,
847 struct mount *child_mnt)
850 mnt_add_count(mnt, 1); /* essentially, that's mntget */
851 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
852 child_mnt->mnt_parent = mnt;
853 child_mnt->mnt_mp = mp;
854 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
858 * vfsmount lock must be held for write
860 static void attach_mnt(struct mount *mnt,
861 struct mount *parent,
862 struct mountpoint *mp)
864 mnt_set_mountpoint(parent, mp, mnt);
865 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
866 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
869 static void attach_shadowed(struct mount *mnt,
870 struct mount *parent,
871 struct mount *shadows)
874 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
875 list_add(&mnt->mnt_child, &shadows->mnt_child);
877 hlist_add_head_rcu(&mnt->mnt_hash,
878 m_hash(&parent->mnt, mnt->mnt_mountpoint));
879 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
884 * vfsmount lock must be held for write
886 static void commit_tree(struct mount *mnt, struct mount *shadows)
888 struct mount *parent = mnt->mnt_parent;
891 struct mnt_namespace *n = parent->mnt_ns;
893 BUG_ON(parent == mnt);
895 list_add_tail(&head, &mnt->mnt_list);
896 list_for_each_entry(m, &head, mnt_list)
899 list_splice(&head, n->list.prev);
901 n->mounts += n->pending_mounts;
902 n->pending_mounts = 0;
904 attach_shadowed(mnt, parent, shadows);
905 touch_mnt_namespace(n);
908 static struct mount *next_mnt(struct mount *p, struct mount *root)
910 struct list_head *next = p->mnt_mounts.next;
911 if (next == &p->mnt_mounts) {
915 next = p->mnt_child.next;
916 if (next != &p->mnt_parent->mnt_mounts)
921 return list_entry(next, struct mount, mnt_child);
924 static struct mount *skip_mnt_tree(struct mount *p)
926 struct list_head *prev = p->mnt_mounts.prev;
927 while (prev != &p->mnt_mounts) {
928 p = list_entry(prev, struct mount, mnt_child);
929 prev = p->mnt_mounts.prev;
935 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
941 return ERR_PTR(-ENODEV);
943 mnt = alloc_vfsmnt(name);
945 return ERR_PTR(-ENOMEM);
947 if (flags & MS_KERNMOUNT)
948 mnt->mnt.mnt_flags = MNT_INTERNAL;
950 root = mount_fs(type, flags, name, data);
954 return ERR_CAST(root);
957 mnt->mnt.mnt_root = root;
958 mnt->mnt.mnt_sb = root->d_sb;
959 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
960 mnt->mnt_parent = mnt;
962 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
966 EXPORT_SYMBOL_GPL(vfs_kern_mount);
968 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
971 struct super_block *sb = old->mnt.mnt_sb;
975 mnt = alloc_vfsmnt(old->mnt_devname);
977 return ERR_PTR(-ENOMEM);
979 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
980 mnt->mnt_group_id = 0; /* not a peer of original */
982 mnt->mnt_group_id = old->mnt_group_id;
984 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
985 err = mnt_alloc_group_id(mnt);
990 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
991 /* Don't allow unprivileged users to change mount flags */
992 if (flag & CL_UNPRIVILEGED) {
993 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
995 if (mnt->mnt.mnt_flags & MNT_READONLY)
996 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
998 if (mnt->mnt.mnt_flags & MNT_NODEV)
999 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1001 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1002 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1004 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1005 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1008 /* Don't allow unprivileged users to reveal what is under a mount */
1009 if ((flag & CL_UNPRIVILEGED) &&
1010 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1011 mnt->mnt.mnt_flags |= MNT_LOCKED;
1013 atomic_inc(&sb->s_active);
1014 mnt->mnt.mnt_sb = sb;
1015 mnt->mnt.mnt_root = dget(root);
1016 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1017 mnt->mnt_parent = mnt;
1019 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1020 unlock_mount_hash();
1022 if ((flag & CL_SLAVE) ||
1023 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1024 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1025 mnt->mnt_master = old;
1026 CLEAR_MNT_SHARED(mnt);
1027 } else if (!(flag & CL_PRIVATE)) {
1028 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1029 list_add(&mnt->mnt_share, &old->mnt_share);
1030 if (IS_MNT_SLAVE(old))
1031 list_add(&mnt->mnt_slave, &old->mnt_slave);
1032 mnt->mnt_master = old->mnt_master;
1034 if (flag & CL_MAKE_SHARED)
1035 set_mnt_shared(mnt);
1037 /* stick the duplicate mount on the same expiry list
1038 * as the original if that was on one */
1039 if (flag & CL_EXPIRE) {
1040 if (!list_empty(&old->mnt_expire))
1041 list_add(&mnt->mnt_expire, &old->mnt_expire);
1049 return ERR_PTR(err);
1052 static void cleanup_mnt(struct mount *mnt)
1055 * This probably indicates that somebody messed
1056 * up a mnt_want/drop_write() pair. If this
1057 * happens, the filesystem was probably unable
1058 * to make r/w->r/o transitions.
1061 * The locking used to deal with mnt_count decrement provides barriers,
1062 * so mnt_get_writers() below is safe.
1064 WARN_ON(mnt_get_writers(mnt));
1065 if (unlikely(mnt->mnt_pins.first))
1067 fsnotify_vfsmount_delete(&mnt->mnt);
1068 dput(mnt->mnt.mnt_root);
1069 deactivate_super(mnt->mnt.mnt_sb);
1071 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1074 static void __cleanup_mnt(struct rcu_head *head)
1076 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1079 static LLIST_HEAD(delayed_mntput_list);
1080 static void delayed_mntput(struct work_struct *unused)
1082 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1083 struct llist_node *next;
1085 for (; node; node = next) {
1086 next = llist_next(node);
1087 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1090 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1092 static void mntput_no_expire(struct mount *mnt)
1095 mnt_add_count(mnt, -1);
1096 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1101 if (mnt_get_count(mnt)) {
1103 unlock_mount_hash();
1106 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1108 unlock_mount_hash();
1111 mnt->mnt.mnt_flags |= MNT_DOOMED;
1114 list_del(&mnt->mnt_instance);
1116 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1117 struct mount *p, *tmp;
1118 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1122 unlock_mount_hash();
1124 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1125 struct task_struct *task = current;
1126 if (likely(!(task->flags & PF_KTHREAD))) {
1127 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1128 if (!task_work_add(task, &mnt->mnt_rcu, true))
1131 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1132 schedule_delayed_work(&delayed_mntput_work, 1);
1138 void mntput(struct vfsmount *mnt)
1141 struct mount *m = real_mount(mnt);
1142 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1143 if (unlikely(m->mnt_expiry_mark))
1144 m->mnt_expiry_mark = 0;
1145 mntput_no_expire(m);
1148 EXPORT_SYMBOL(mntput);
1150 struct vfsmount *mntget(struct vfsmount *mnt)
1153 mnt_add_count(real_mount(mnt), 1);
1156 EXPORT_SYMBOL(mntget);
1158 struct vfsmount *mnt_clone_internal(struct path *path)
1161 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1164 p->mnt.mnt_flags |= MNT_INTERNAL;
1168 static inline void mangle(struct seq_file *m, const char *s)
1170 seq_escape(m, s, " \t\n\\");
1174 * Simple .show_options callback for filesystems which don't want to
1175 * implement more complex mount option showing.
1177 * See also save_mount_options().
1179 int generic_show_options(struct seq_file *m, struct dentry *root)
1181 const char *options;
1184 options = rcu_dereference(root->d_sb->s_options);
1186 if (options != NULL && options[0]) {
1194 EXPORT_SYMBOL(generic_show_options);
1197 * If filesystem uses generic_show_options(), this function should be
1198 * called from the fill_super() callback.
1200 * The .remount_fs callback usually needs to be handled in a special
1201 * way, to make sure, that previous options are not overwritten if the
1204 * Also note, that if the filesystem's .remount_fs function doesn't
1205 * reset all options to their default value, but changes only newly
1206 * given options, then the displayed options will not reflect reality
1209 void save_mount_options(struct super_block *sb, char *options)
1211 BUG_ON(sb->s_options);
1212 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1214 EXPORT_SYMBOL(save_mount_options);
1216 void replace_mount_options(struct super_block *sb, char *options)
1218 char *old = sb->s_options;
1219 rcu_assign_pointer(sb->s_options, options);
1225 EXPORT_SYMBOL(replace_mount_options);
1227 #ifdef CONFIG_PROC_FS
1228 /* iterator; we want it to have access to namespace_sem, thus here... */
1229 static void *m_start(struct seq_file *m, loff_t *pos)
1231 struct proc_mounts *p = m->private;
1233 down_read(&namespace_sem);
1234 if (p->cached_event == p->ns->event) {
1235 void *v = p->cached_mount;
1236 if (*pos == p->cached_index)
1238 if (*pos == p->cached_index + 1) {
1239 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1240 return p->cached_mount = v;
1244 p->cached_event = p->ns->event;
1245 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1246 p->cached_index = *pos;
1247 return p->cached_mount;
1250 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1252 struct proc_mounts *p = m->private;
1254 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1255 p->cached_index = *pos;
1256 return p->cached_mount;
1259 static void m_stop(struct seq_file *m, void *v)
1261 up_read(&namespace_sem);
1264 static int m_show(struct seq_file *m, void *v)
1266 struct proc_mounts *p = m->private;
1267 struct mount *r = list_entry(v, struct mount, mnt_list);
1268 return p->show(m, &r->mnt);
1271 const struct seq_operations mounts_op = {
1277 #endif /* CONFIG_PROC_FS */
1280 * may_umount_tree - check if a mount tree is busy
1281 * @mnt: root of mount tree
1283 * This is called to check if a tree of mounts has any
1284 * open files, pwds, chroots or sub mounts that are
1287 int may_umount_tree(struct vfsmount *m)
1289 struct mount *mnt = real_mount(m);
1290 int actual_refs = 0;
1291 int minimum_refs = 0;
1295 /* write lock needed for mnt_get_count */
1297 for (p = mnt; p; p = next_mnt(p, mnt)) {
1298 actual_refs += mnt_get_count(p);
1301 unlock_mount_hash();
1303 if (actual_refs > minimum_refs)
1309 EXPORT_SYMBOL(may_umount_tree);
1312 * may_umount - check if a mount point is busy
1313 * @mnt: root of mount
1315 * This is called to check if a mount point has any
1316 * open files, pwds, chroots or sub mounts. If the
1317 * mount has sub mounts this will return busy
1318 * regardless of whether the sub mounts are busy.
1320 * Doesn't take quota and stuff into account. IOW, in some cases it will
1321 * give false negatives. The main reason why it's here is that we need
1322 * a non-destructive way to look for easily umountable filesystems.
1324 int may_umount(struct vfsmount *mnt)
1327 down_read(&namespace_sem);
1329 if (propagate_mount_busy(real_mount(mnt), 2))
1331 unlock_mount_hash();
1332 up_read(&namespace_sem);
1336 EXPORT_SYMBOL(may_umount);
1338 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1340 static void namespace_unlock(void)
1342 struct hlist_head head;
1344 hlist_move_list(&unmounted, &head);
1346 up_write(&namespace_sem);
1348 if (likely(hlist_empty(&head)))
1353 group_pin_kill(&head);
1356 static inline void namespace_lock(void)
1358 down_write(&namespace_sem);
1361 enum umount_tree_flags {
1363 UMOUNT_PROPAGATE = 2,
1364 UMOUNT_CONNECTED = 4,
1367 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1369 /* Leaving mounts connected is only valid for lazy umounts */
1370 if (how & UMOUNT_SYNC)
1373 /* A mount without a parent has nothing to be connected to */
1374 if (!mnt_has_parent(mnt))
1377 /* Because the reference counting rules change when mounts are
1378 * unmounted and connected, umounted mounts may not be
1379 * connected to mounted mounts.
1381 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1384 /* Has it been requested that the mount remain connected? */
1385 if (how & UMOUNT_CONNECTED)
1388 /* Is the mount locked such that it needs to remain connected? */
1389 if (IS_MNT_LOCKED(mnt))
1392 /* By default disconnect the mount */
1397 * mount_lock must be held
1398 * namespace_sem must be held for write
1400 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1402 LIST_HEAD(tmp_list);
1405 if (how & UMOUNT_PROPAGATE)
1406 propagate_mount_unlock(mnt);
1408 /* Gather the mounts to umount */
1409 for (p = mnt; p; p = next_mnt(p, mnt)) {
1410 p->mnt.mnt_flags |= MNT_UMOUNT;
1411 list_move(&p->mnt_list, &tmp_list);
1414 /* Hide the mounts from mnt_mounts */
1415 list_for_each_entry(p, &tmp_list, mnt_list) {
1416 list_del_init(&p->mnt_child);
1419 /* Add propogated mounts to the tmp_list */
1420 if (how & UMOUNT_PROPAGATE)
1421 propagate_umount(&tmp_list);
1423 while (!list_empty(&tmp_list)) {
1424 struct mnt_namespace *ns;
1426 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1427 list_del_init(&p->mnt_expire);
1428 list_del_init(&p->mnt_list);
1432 __touch_mnt_namespace(ns);
1435 if (how & UMOUNT_SYNC)
1436 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1438 disconnect = disconnect_mount(p, how);
1440 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1441 disconnect ? &unmounted : NULL);
1442 if (mnt_has_parent(p)) {
1443 mnt_add_count(p->mnt_parent, -1);
1445 /* Don't forget about p */
1446 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1451 change_mnt_propagation(p, MS_PRIVATE);
1455 static void shrink_submounts(struct mount *mnt);
1457 static int do_umount(struct mount *mnt, int flags)
1459 struct super_block *sb = mnt->mnt.mnt_sb;
1462 retval = security_sb_umount(&mnt->mnt, flags);
1467 * Allow userspace to request a mountpoint be expired rather than
1468 * unmounting unconditionally. Unmount only happens if:
1469 * (1) the mark is already set (the mark is cleared by mntput())
1470 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1472 if (flags & MNT_EXPIRE) {
1473 if (&mnt->mnt == current->fs->root.mnt ||
1474 flags & (MNT_FORCE | MNT_DETACH))
1478 * probably don't strictly need the lock here if we examined
1479 * all race cases, but it's a slowpath.
1482 if (mnt_get_count(mnt) != 2) {
1483 unlock_mount_hash();
1486 unlock_mount_hash();
1488 if (!xchg(&mnt->mnt_expiry_mark, 1))
1493 * If we may have to abort operations to get out of this
1494 * mount, and they will themselves hold resources we must
1495 * allow the fs to do things. In the Unix tradition of
1496 * 'Gee thats tricky lets do it in userspace' the umount_begin
1497 * might fail to complete on the first run through as other tasks
1498 * must return, and the like. Thats for the mount program to worry
1499 * about for the moment.
1502 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1503 sb->s_op->umount_begin(sb);
1507 * No sense to grab the lock for this test, but test itself looks
1508 * somewhat bogus. Suggestions for better replacement?
1509 * Ho-hum... In principle, we might treat that as umount + switch
1510 * to rootfs. GC would eventually take care of the old vfsmount.
1511 * Actually it makes sense, especially if rootfs would contain a
1512 * /reboot - static binary that would close all descriptors and
1513 * call reboot(9). Then init(8) could umount root and exec /reboot.
1515 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1517 * Special case for "unmounting" root ...
1518 * we just try to remount it readonly.
1520 if (!capable(CAP_SYS_ADMIN))
1522 down_write(&sb->s_umount);
1523 if (!(sb->s_flags & MS_RDONLY))
1524 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1525 up_write(&sb->s_umount);
1533 if (flags & MNT_DETACH) {
1534 if (!list_empty(&mnt->mnt_list))
1535 umount_tree(mnt, UMOUNT_PROPAGATE);
1538 shrink_submounts(mnt);
1540 if (!propagate_mount_busy(mnt, 2)) {
1541 if (!list_empty(&mnt->mnt_list))
1542 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1546 unlock_mount_hash();
1552 * __detach_mounts - lazily unmount all mounts on the specified dentry
1554 * During unlink, rmdir, and d_drop it is possible to loose the path
1555 * to an existing mountpoint, and wind up leaking the mount.
1556 * detach_mounts allows lazily unmounting those mounts instead of
1559 * The caller may hold dentry->d_inode->i_mutex.
1561 void __detach_mounts(struct dentry *dentry)
1563 struct mountpoint *mp;
1567 mp = lookup_mountpoint(dentry);
1568 if (IS_ERR_OR_NULL(mp))
1573 while (!hlist_empty(&mp->m_list)) {
1574 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1575 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1576 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1579 else umount_tree(mnt, UMOUNT_CONNECTED);
1581 unlock_mount_hash();
1588 * Is the caller allowed to modify his namespace?
1590 static inline bool may_mount(void)
1592 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1595 static inline bool may_mandlock(void)
1597 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1600 return capable(CAP_SYS_ADMIN);
1604 * Now umount can handle mount points as well as block devices.
1605 * This is important for filesystems which use unnamed block devices.
1607 * We now support a flag for forced unmount like the other 'big iron'
1608 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1611 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1616 int lookup_flags = 0;
1618 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1624 if (!(flags & UMOUNT_NOFOLLOW))
1625 lookup_flags |= LOOKUP_FOLLOW;
1627 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1630 mnt = real_mount(path.mnt);
1632 if (path.dentry != path.mnt->mnt_root)
1634 if (!check_mnt(mnt))
1636 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1639 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1642 retval = do_umount(mnt, flags);
1644 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1646 mntput_no_expire(mnt);
1651 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1654 * The 2.0 compatible umount. No flags.
1656 SYSCALL_DEFINE1(oldumount, char __user *, name)
1658 return sys_umount(name, 0);
1663 static bool is_mnt_ns_file(struct dentry *dentry)
1665 /* Is this a proxy for a mount namespace? */
1666 return dentry->d_op == &ns_dentry_operations &&
1667 dentry->d_fsdata == &mntns_operations;
1670 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1672 return container_of(ns, struct mnt_namespace, ns);
1675 static bool mnt_ns_loop(struct dentry *dentry)
1677 /* Could bind mounting the mount namespace inode cause a
1678 * mount namespace loop?
1680 struct mnt_namespace *mnt_ns;
1681 if (!is_mnt_ns_file(dentry))
1684 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1685 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1688 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1691 struct mount *res, *p, *q, *r, *parent;
1693 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1694 return ERR_PTR(-EINVAL);
1696 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1697 return ERR_PTR(-EINVAL);
1699 res = q = clone_mnt(mnt, dentry, flag);
1703 q->mnt_mountpoint = mnt->mnt_mountpoint;
1706 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1708 if (!is_subdir(r->mnt_mountpoint, dentry))
1711 for (s = r; s; s = next_mnt(s, r)) {
1712 struct mount *t = NULL;
1713 if (!(flag & CL_COPY_UNBINDABLE) &&
1714 IS_MNT_UNBINDABLE(s)) {
1715 s = skip_mnt_tree(s);
1718 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1719 is_mnt_ns_file(s->mnt.mnt_root)) {
1720 s = skip_mnt_tree(s);
1723 while (p != s->mnt_parent) {
1729 q = clone_mnt(p, p->mnt.mnt_root, flag);
1733 list_add_tail(&q->mnt_list, &res->mnt_list);
1734 mnt_set_mountpoint(parent, p->mnt_mp, q);
1735 if (!list_empty(&parent->mnt_mounts)) {
1736 t = list_last_entry(&parent->mnt_mounts,
1737 struct mount, mnt_child);
1738 if (t->mnt_mp != p->mnt_mp)
1741 attach_shadowed(q, parent, t);
1742 unlock_mount_hash();
1749 umount_tree(res, UMOUNT_SYNC);
1750 unlock_mount_hash();
1755 /* Caller should check returned pointer for errors */
1757 struct vfsmount *collect_mounts(struct path *path)
1761 if (!check_mnt(real_mount(path->mnt)))
1762 tree = ERR_PTR(-EINVAL);
1764 tree = copy_tree(real_mount(path->mnt), path->dentry,
1765 CL_COPY_ALL | CL_PRIVATE);
1768 return ERR_CAST(tree);
1772 void drop_collected_mounts(struct vfsmount *mnt)
1776 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1777 unlock_mount_hash();
1782 * clone_private_mount - create a private clone of a path
1784 * This creates a new vfsmount, which will be the clone of @path. The new will
1785 * not be attached anywhere in the namespace and will be private (i.e. changes
1786 * to the originating mount won't be propagated into this).
1788 * Release with mntput().
1790 struct vfsmount *clone_private_mount(struct path *path)
1792 struct mount *old_mnt = real_mount(path->mnt);
1793 struct mount *new_mnt;
1795 if (IS_MNT_UNBINDABLE(old_mnt))
1796 return ERR_PTR(-EINVAL);
1798 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1799 if (IS_ERR(new_mnt))
1800 return ERR_CAST(new_mnt);
1802 return &new_mnt->mnt;
1804 EXPORT_SYMBOL_GPL(clone_private_mount);
1806 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1807 struct vfsmount *root)
1810 int res = f(root, arg);
1813 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1814 res = f(&mnt->mnt, arg);
1821 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1825 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1826 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1827 mnt_release_group_id(p);
1831 static int invent_group_ids(struct mount *mnt, bool recurse)
1835 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1836 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1837 int err = mnt_alloc_group_id(p);
1839 cleanup_group_ids(mnt, p);
1848 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1850 unsigned int max = READ_ONCE(sysctl_mount_max);
1851 unsigned int mounts = 0, old, pending, sum;
1854 for (p = mnt; p; p = next_mnt(p, mnt))
1858 pending = ns->pending_mounts;
1859 sum = old + pending;
1863 (mounts > (max - sum)))
1866 ns->pending_mounts = pending + mounts;
1871 * @source_mnt : mount tree to be attached
1872 * @nd : place the mount tree @source_mnt is attached
1873 * @parent_nd : if non-null, detach the source_mnt from its parent and
1874 * store the parent mount and mountpoint dentry.
1875 * (done when source_mnt is moved)
1877 * NOTE: in the table below explains the semantics when a source mount
1878 * of a given type is attached to a destination mount of a given type.
1879 * ---------------------------------------------------------------------------
1880 * | BIND MOUNT OPERATION |
1881 * |**************************************************************************
1882 * | source-->| shared | private | slave | unbindable |
1886 * |**************************************************************************
1887 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1889 * |non-shared| shared (+) | private | slave (*) | invalid |
1890 * ***************************************************************************
1891 * A bind operation clones the source mount and mounts the clone on the
1892 * destination mount.
1894 * (++) the cloned mount is propagated to all the mounts in the propagation
1895 * tree of the destination mount and the cloned mount is added to
1896 * the peer group of the source mount.
1897 * (+) the cloned mount is created under the destination mount and is marked
1898 * as shared. The cloned mount is added to the peer group of the source
1900 * (+++) the mount is propagated to all the mounts in the propagation tree
1901 * of the destination mount and the cloned mount is made slave
1902 * of the same master as that of the source mount. The cloned mount
1903 * is marked as 'shared and slave'.
1904 * (*) the cloned mount is made a slave of the same master as that of the
1907 * ---------------------------------------------------------------------------
1908 * | MOVE MOUNT OPERATION |
1909 * |**************************************************************************
1910 * | source-->| shared | private | slave | unbindable |
1914 * |**************************************************************************
1915 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1917 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1918 * ***************************************************************************
1920 * (+) the mount is moved to the destination. And is then propagated to
1921 * all the mounts in the propagation tree of the destination mount.
1922 * (+*) the mount is moved to the destination.
1923 * (+++) the mount is moved to the destination and is then propagated to
1924 * all the mounts belonging to the destination mount's propagation tree.
1925 * the mount is marked as 'shared and slave'.
1926 * (*) the mount continues to be a slave at the new location.
1928 * if the source mount is a tree, the operations explained above is
1929 * applied to each mount in the tree.
1930 * Must be called without spinlocks held, since this function can sleep
1933 static int attach_recursive_mnt(struct mount *source_mnt,
1934 struct mount *dest_mnt,
1935 struct mountpoint *dest_mp,
1936 struct path *parent_path)
1938 HLIST_HEAD(tree_list);
1939 struct mnt_namespace *ns = dest_mnt->mnt_ns;
1940 struct mount *child, *p;
1941 struct hlist_node *n;
1944 /* Is there space to add these mounts to the mount namespace? */
1946 err = count_mounts(ns, source_mnt);
1951 if (IS_MNT_SHARED(dest_mnt)) {
1952 err = invent_group_ids(source_mnt, true);
1955 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1958 goto out_cleanup_ids;
1959 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1965 detach_mnt(source_mnt, parent_path);
1966 attach_mnt(source_mnt, dest_mnt, dest_mp);
1967 touch_mnt_namespace(source_mnt->mnt_ns);
1969 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1970 commit_tree(source_mnt, NULL);
1973 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1975 hlist_del_init(&child->mnt_hash);
1976 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1977 child->mnt_mountpoint);
1978 commit_tree(child, q);
1980 unlock_mount_hash();
1985 while (!hlist_empty(&tree_list)) {
1986 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1987 child->mnt_parent->mnt_ns->pending_mounts = 0;
1988 umount_tree(child, UMOUNT_SYNC);
1990 unlock_mount_hash();
1991 cleanup_group_ids(source_mnt, NULL);
1993 ns->pending_mounts = 0;
1997 static struct mountpoint *lock_mount(struct path *path)
1999 struct vfsmount *mnt;
2000 struct dentry *dentry = path->dentry;
2002 inode_lock(dentry->d_inode);
2003 if (unlikely(cant_mount(dentry))) {
2004 inode_unlock(dentry->d_inode);
2005 return ERR_PTR(-ENOENT);
2008 mnt = lookup_mnt(path);
2010 struct mountpoint *mp = lookup_mountpoint(dentry);
2012 mp = new_mountpoint(dentry);
2015 inode_unlock(dentry->d_inode);
2021 inode_unlock(path->dentry->d_inode);
2024 dentry = path->dentry = dget(mnt->mnt_root);
2028 static void unlock_mount(struct mountpoint *where)
2030 struct dentry *dentry = where->m_dentry;
2031 put_mountpoint(where);
2033 inode_unlock(dentry->d_inode);
2036 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2038 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2041 if (d_is_dir(mp->m_dentry) !=
2042 d_is_dir(mnt->mnt.mnt_root))
2045 return attach_recursive_mnt(mnt, p, mp, NULL);
2049 * Sanity check the flags to change_mnt_propagation.
2052 static int flags_to_propagation_type(int flags)
2054 int type = flags & ~(MS_REC | MS_SILENT);
2056 /* Fail if any non-propagation flags are set */
2057 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2059 /* Only one propagation flag should be set */
2060 if (!is_power_of_2(type))
2066 * recursively change the type of the mountpoint.
2068 static int do_change_type(struct path *path, int flag)
2071 struct mount *mnt = real_mount(path->mnt);
2072 int recurse = flag & MS_REC;
2076 if (path->dentry != path->mnt->mnt_root)
2079 type = flags_to_propagation_type(flag);
2084 if (type == MS_SHARED) {
2085 err = invent_group_ids(mnt, recurse);
2091 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2092 change_mnt_propagation(m, type);
2093 unlock_mount_hash();
2100 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2102 struct mount *child;
2103 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2104 if (!is_subdir(child->mnt_mountpoint, dentry))
2107 if (child->mnt.mnt_flags & MNT_LOCKED)
2114 * do loopback mount.
2116 static int do_loopback(struct path *path, const char *old_name,
2119 struct path old_path;
2120 struct mount *mnt = NULL, *old, *parent;
2121 struct mountpoint *mp;
2123 if (!old_name || !*old_name)
2125 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2130 if (mnt_ns_loop(old_path.dentry))
2133 mp = lock_mount(path);
2138 old = real_mount(old_path.mnt);
2139 parent = real_mount(path->mnt);
2142 if (IS_MNT_UNBINDABLE(old))
2145 if (!check_mnt(parent))
2148 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2151 if (!recurse && has_locked_children(old, old_path.dentry))
2155 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2157 mnt = clone_mnt(old, old_path.dentry, 0);
2164 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2166 err = graft_tree(mnt, parent, mp);
2169 umount_tree(mnt, UMOUNT_SYNC);
2170 unlock_mount_hash();
2175 path_put(&old_path);
2179 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2182 int readonly_request = 0;
2184 if (ms_flags & MS_RDONLY)
2185 readonly_request = 1;
2186 if (readonly_request == __mnt_is_readonly(mnt))
2189 if (readonly_request)
2190 error = mnt_make_readonly(real_mount(mnt));
2192 __mnt_unmake_readonly(real_mount(mnt));
2197 * change filesystem flags. dir should be a physical root of filesystem.
2198 * If you've mounted a non-root directory somewhere and want to do remount
2199 * on it - tough luck.
2201 static int do_remount(struct path *path, int flags, int mnt_flags,
2205 struct super_block *sb = path->mnt->mnt_sb;
2206 struct mount *mnt = real_mount(path->mnt);
2208 if (!check_mnt(mnt))
2211 if (path->dentry != path->mnt->mnt_root)
2214 /* Don't allow changing of locked mnt flags.
2216 * No locks need to be held here while testing the various
2217 * MNT_LOCK flags because those flags can never be cleared
2218 * once they are set.
2220 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2221 !(mnt_flags & MNT_READONLY)) {
2224 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2225 !(mnt_flags & MNT_NODEV)) {
2228 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2229 !(mnt_flags & MNT_NOSUID)) {
2232 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2233 !(mnt_flags & MNT_NOEXEC)) {
2236 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2237 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2241 err = security_sb_remount(sb, data);
2245 down_write(&sb->s_umount);
2246 if (flags & MS_BIND)
2247 err = change_mount_flags(path->mnt, flags);
2248 else if (!capable(CAP_SYS_ADMIN))
2251 err = do_remount_sb(sb, flags, data, 0);
2254 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2255 mnt->mnt.mnt_flags = mnt_flags;
2256 touch_mnt_namespace(mnt->mnt_ns);
2257 unlock_mount_hash();
2259 up_write(&sb->s_umount);
2263 static inline int tree_contains_unbindable(struct mount *mnt)
2266 for (p = mnt; p; p = next_mnt(p, mnt)) {
2267 if (IS_MNT_UNBINDABLE(p))
2273 static int do_move_mount(struct path *path, const char *old_name)
2275 struct path old_path, parent_path;
2278 struct mountpoint *mp;
2280 if (!old_name || !*old_name)
2282 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2286 mp = lock_mount(path);
2291 old = real_mount(old_path.mnt);
2292 p = real_mount(path->mnt);
2295 if (!check_mnt(p) || !check_mnt(old))
2298 if (old->mnt.mnt_flags & MNT_LOCKED)
2302 if (old_path.dentry != old_path.mnt->mnt_root)
2305 if (!mnt_has_parent(old))
2308 if (d_is_dir(path->dentry) !=
2309 d_is_dir(old_path.dentry))
2312 * Don't move a mount residing in a shared parent.
2314 if (IS_MNT_SHARED(old->mnt_parent))
2317 * Don't move a mount tree containing unbindable mounts to a destination
2318 * mount which is shared.
2320 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2323 for (; mnt_has_parent(p); p = p->mnt_parent)
2327 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2331 /* if the mount is moved, it should no longer be expire
2333 list_del_init(&old->mnt_expire);
2338 path_put(&parent_path);
2339 path_put(&old_path);
2343 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2346 const char *subtype = strchr(fstype, '.');
2355 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2357 if (!mnt->mnt_sb->s_subtype)
2363 return ERR_PTR(err);
2367 * add a mount into a namespace's mount tree
2369 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2371 struct mountpoint *mp;
2372 struct mount *parent;
2375 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2377 mp = lock_mount(path);
2381 parent = real_mount(path->mnt);
2383 if (unlikely(!check_mnt(parent))) {
2384 /* that's acceptable only for automounts done in private ns */
2385 if (!(mnt_flags & MNT_SHRINKABLE))
2387 /* ... and for those we'd better have mountpoint still alive */
2388 if (!parent->mnt_ns)
2392 /* Refuse the same filesystem on the same mount point */
2394 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2395 path->mnt->mnt_root == path->dentry)
2399 if (d_is_symlink(newmnt->mnt.mnt_root))
2402 newmnt->mnt.mnt_flags = mnt_flags;
2403 err = graft_tree(newmnt, parent, mp);
2410 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2413 * create a new mount for userspace and request it to be added into the
2416 static int do_new_mount(struct path *path, const char *fstype, int flags,
2417 int mnt_flags, const char *name, void *data)
2419 struct file_system_type *type;
2420 struct vfsmount *mnt;
2426 type = get_fs_type(fstype);
2430 mnt = vfs_kern_mount(type, flags, name, data);
2431 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2432 !mnt->mnt_sb->s_subtype)
2433 mnt = fs_set_subtype(mnt, fstype);
2435 put_filesystem(type);
2437 return PTR_ERR(mnt);
2439 if (mount_too_revealing(mnt, &mnt_flags)) {
2444 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2450 int finish_automount(struct vfsmount *m, struct path *path)
2452 struct mount *mnt = real_mount(m);
2454 /* The new mount record should have at least 2 refs to prevent it being
2455 * expired before we get a chance to add it
2457 BUG_ON(mnt_get_count(mnt) < 2);
2459 if (m->mnt_sb == path->mnt->mnt_sb &&
2460 m->mnt_root == path->dentry) {
2465 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2469 /* remove m from any expiration list it may be on */
2470 if (!list_empty(&mnt->mnt_expire)) {
2472 list_del_init(&mnt->mnt_expire);
2481 * mnt_set_expiry - Put a mount on an expiration list
2482 * @mnt: The mount to list.
2483 * @expiry_list: The list to add the mount to.
2485 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2489 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2493 EXPORT_SYMBOL(mnt_set_expiry);
2496 * process a list of expirable mountpoints with the intent of discarding any
2497 * mountpoints that aren't in use and haven't been touched since last we came
2500 void mark_mounts_for_expiry(struct list_head *mounts)
2502 struct mount *mnt, *next;
2503 LIST_HEAD(graveyard);
2505 if (list_empty(mounts))
2511 /* extract from the expiration list every vfsmount that matches the
2512 * following criteria:
2513 * - only referenced by its parent vfsmount
2514 * - still marked for expiry (marked on the last call here; marks are
2515 * cleared by mntput())
2517 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2518 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2519 propagate_mount_busy(mnt, 1))
2521 list_move(&mnt->mnt_expire, &graveyard);
2523 while (!list_empty(&graveyard)) {
2524 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2525 touch_mnt_namespace(mnt->mnt_ns);
2526 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2528 unlock_mount_hash();
2532 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2535 * Ripoff of 'select_parent()'
2537 * search the list of submounts for a given mountpoint, and move any
2538 * shrinkable submounts to the 'graveyard' list.
2540 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2542 struct mount *this_parent = parent;
2543 struct list_head *next;
2547 next = this_parent->mnt_mounts.next;
2549 while (next != &this_parent->mnt_mounts) {
2550 struct list_head *tmp = next;
2551 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2554 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2557 * Descend a level if the d_mounts list is non-empty.
2559 if (!list_empty(&mnt->mnt_mounts)) {
2564 if (!propagate_mount_busy(mnt, 1)) {
2565 list_move_tail(&mnt->mnt_expire, graveyard);
2570 * All done at this level ... ascend and resume the search
2572 if (this_parent != parent) {
2573 next = this_parent->mnt_child.next;
2574 this_parent = this_parent->mnt_parent;
2581 * process a list of expirable mountpoints with the intent of discarding any
2582 * submounts of a specific parent mountpoint
2584 * mount_lock must be held for write
2586 static void shrink_submounts(struct mount *mnt)
2588 LIST_HEAD(graveyard);
2591 /* extract submounts of 'mountpoint' from the expiration list */
2592 while (select_submounts(mnt, &graveyard)) {
2593 while (!list_empty(&graveyard)) {
2594 m = list_first_entry(&graveyard, struct mount,
2596 touch_mnt_namespace(m->mnt_ns);
2597 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2603 * Some copy_from_user() implementations do not return the exact number of
2604 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2605 * Note that this function differs from copy_from_user() in that it will oops
2606 * on bad values of `to', rather than returning a short copy.
2608 static long exact_copy_from_user(void *to, const void __user * from,
2612 const char __user *f = from;
2615 if (!access_ok(VERIFY_READ, from, n))
2619 if (__get_user(c, f)) {
2630 void *copy_mount_options(const void __user * data)
2639 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2641 return ERR_PTR(-ENOMEM);
2643 /* We only care that *some* data at the address the user
2644 * gave us is valid. Just in case, we'll zero
2645 * the remainder of the page.
2647 /* copy_from_user cannot cross TASK_SIZE ! */
2648 size = TASK_SIZE - (unsigned long)data;
2649 if (size > PAGE_SIZE)
2652 i = size - exact_copy_from_user(copy, data, size);
2655 return ERR_PTR(-EFAULT);
2658 memset(copy + i, 0, PAGE_SIZE - i);
2662 char *copy_mount_string(const void __user *data)
2664 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2668 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2669 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2671 * data is a (void *) that can point to any structure up to
2672 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2673 * information (or be NULL).
2675 * Pre-0.97 versions of mount() didn't have a flags word.
2676 * When the flags word was introduced its top half was required
2677 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2678 * Therefore, if this magic number is present, it carries no information
2679 * and must be discarded.
2681 long do_mount(const char *dev_name, const char __user *dir_name,
2682 const char *type_page, unsigned long flags, void *data_page)
2689 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2690 flags &= ~MS_MGC_MSK;
2692 /* Basic sanity checks */
2694 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2696 /* ... and get the mountpoint */
2697 retval = user_path(dir_name, &path);
2701 retval = security_sb_mount(dev_name, &path,
2702 type_page, flags, data_page);
2703 if (!retval && !may_mount())
2705 if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2710 /* Default to relatime unless overriden */
2711 if (!(flags & MS_NOATIME))
2712 mnt_flags |= MNT_RELATIME;
2714 /* Separate the per-mountpoint flags */
2715 if (flags & MS_NOSUID)
2716 mnt_flags |= MNT_NOSUID;
2717 if (flags & MS_NODEV)
2718 mnt_flags |= MNT_NODEV;
2719 if (flags & MS_NOEXEC)
2720 mnt_flags |= MNT_NOEXEC;
2721 if (flags & MS_NOATIME)
2722 mnt_flags |= MNT_NOATIME;
2723 if (flags & MS_NODIRATIME)
2724 mnt_flags |= MNT_NODIRATIME;
2725 if (flags & MS_STRICTATIME)
2726 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2727 if (flags & MS_RDONLY)
2728 mnt_flags |= MNT_READONLY;
2730 /* The default atime for remount is preservation */
2731 if ((flags & MS_REMOUNT) &&
2732 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2733 MS_STRICTATIME)) == 0)) {
2734 mnt_flags &= ~MNT_ATIME_MASK;
2735 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2738 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2739 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2740 MS_STRICTATIME | MS_NOREMOTELOCK);
2742 if (flags & MS_REMOUNT)
2743 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2745 else if (flags & MS_BIND)
2746 retval = do_loopback(&path, dev_name, flags & MS_REC);
2747 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2748 retval = do_change_type(&path, flags);
2749 else if (flags & MS_MOVE)
2750 retval = do_move_mount(&path, dev_name);
2752 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2753 dev_name, data_page);
2759 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2761 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2764 static void dec_mnt_namespaces(struct ucounts *ucounts)
2766 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2769 static void free_mnt_ns(struct mnt_namespace *ns)
2771 ns_free_inum(&ns->ns);
2772 dec_mnt_namespaces(ns->ucounts);
2773 put_user_ns(ns->user_ns);
2778 * Assign a sequence number so we can detect when we attempt to bind
2779 * mount a reference to an older mount namespace into the current
2780 * mount namespace, preventing reference counting loops. A 64bit
2781 * number incrementing at 10Ghz will take 12,427 years to wrap which
2782 * is effectively never, so we can ignore the possibility.
2784 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2786 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2788 struct mnt_namespace *new_ns;
2789 struct ucounts *ucounts;
2792 ucounts = inc_mnt_namespaces(user_ns);
2794 return ERR_PTR(-ENOSPC);
2796 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2798 dec_mnt_namespaces(ucounts);
2799 return ERR_PTR(-ENOMEM);
2801 ret = ns_alloc_inum(&new_ns->ns);
2804 dec_mnt_namespaces(ucounts);
2805 return ERR_PTR(ret);
2807 new_ns->ns.ops = &mntns_operations;
2808 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2809 atomic_set(&new_ns->count, 1);
2810 new_ns->root = NULL;
2811 INIT_LIST_HEAD(&new_ns->list);
2812 init_waitqueue_head(&new_ns->poll);
2814 new_ns->user_ns = get_user_ns(user_ns);
2815 new_ns->ucounts = ucounts;
2817 new_ns->pending_mounts = 0;
2822 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2823 struct user_namespace *user_ns, struct fs_struct *new_fs)
2825 struct mnt_namespace *new_ns;
2826 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2827 struct mount *p, *q;
2834 if (likely(!(flags & CLONE_NEWNS))) {
2841 new_ns = alloc_mnt_ns(user_ns);
2846 /* First pass: copy the tree topology */
2847 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2848 if (user_ns != ns->user_ns)
2849 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2850 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2853 free_mnt_ns(new_ns);
2854 return ERR_CAST(new);
2857 list_add_tail(&new_ns->list, &new->mnt_list);
2860 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2861 * as belonging to new namespace. We have already acquired a private
2862 * fs_struct, so tsk->fs->lock is not needed.
2870 if (&p->mnt == new_fs->root.mnt) {
2871 new_fs->root.mnt = mntget(&q->mnt);
2874 if (&p->mnt == new_fs->pwd.mnt) {
2875 new_fs->pwd.mnt = mntget(&q->mnt);
2879 p = next_mnt(p, old);
2880 q = next_mnt(q, new);
2883 while (p->mnt.mnt_root != q->mnt.mnt_root)
2884 p = next_mnt(p, old);
2897 * create_mnt_ns - creates a private namespace and adds a root filesystem
2898 * @mnt: pointer to the new root filesystem mountpoint
2900 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2902 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2903 if (!IS_ERR(new_ns)) {
2904 struct mount *mnt = real_mount(m);
2905 mnt->mnt_ns = new_ns;
2908 list_add(&mnt->mnt_list, &new_ns->list);
2915 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2917 struct mnt_namespace *ns;
2918 struct super_block *s;
2922 ns = create_mnt_ns(mnt);
2924 return ERR_CAST(ns);
2926 err = vfs_path_lookup(mnt->mnt_root, mnt,
2927 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2932 return ERR_PTR(err);
2934 /* trade a vfsmount reference for active sb one */
2935 s = path.mnt->mnt_sb;
2936 atomic_inc(&s->s_active);
2938 /* lock the sucker */
2939 down_write(&s->s_umount);
2940 /* ... and return the root of (sub)tree on it */
2943 EXPORT_SYMBOL(mount_subtree);
2945 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2946 char __user *, type, unsigned long, flags, void __user *, data)
2953 kernel_type = copy_mount_string(type);
2954 ret = PTR_ERR(kernel_type);
2955 if (IS_ERR(kernel_type))
2958 kernel_dev = copy_mount_string(dev_name);
2959 ret = PTR_ERR(kernel_dev);
2960 if (IS_ERR(kernel_dev))
2963 options = copy_mount_options(data);
2964 ret = PTR_ERR(options);
2965 if (IS_ERR(options))
2968 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
2980 * Return true if path is reachable from root
2982 * namespace_sem or mount_lock is held
2984 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2985 const struct path *root)
2987 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2988 dentry = mnt->mnt_mountpoint;
2989 mnt = mnt->mnt_parent;
2991 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2994 bool path_is_under(struct path *path1, struct path *path2)
2997 read_seqlock_excl(&mount_lock);
2998 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2999 read_sequnlock_excl(&mount_lock);
3002 EXPORT_SYMBOL(path_is_under);
3005 * pivot_root Semantics:
3006 * Moves the root file system of the current process to the directory put_old,
3007 * makes new_root as the new root file system of the current process, and sets
3008 * root/cwd of all processes which had them on the current root to new_root.
3011 * The new_root and put_old must be directories, and must not be on the
3012 * same file system as the current process root. The put_old must be
3013 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3014 * pointed to by put_old must yield the same directory as new_root. No other
3015 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3017 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3018 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3019 * in this situation.
3022 * - we don't move root/cwd if they are not at the root (reason: if something
3023 * cared enough to change them, it's probably wrong to force them elsewhere)
3024 * - it's okay to pick a root that isn't the root of a file system, e.g.
3025 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3026 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3029 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3030 const char __user *, put_old)
3032 struct path new, old, parent_path, root_parent, root;
3033 struct mount *new_mnt, *root_mnt, *old_mnt;
3034 struct mountpoint *old_mp, *root_mp;
3040 error = user_path_dir(new_root, &new);
3044 error = user_path_dir(put_old, &old);
3048 error = security_sb_pivotroot(&old, &new);
3052 get_fs_root(current->fs, &root);
3053 old_mp = lock_mount(&old);
3054 error = PTR_ERR(old_mp);
3059 new_mnt = real_mount(new.mnt);
3060 root_mnt = real_mount(root.mnt);
3061 old_mnt = real_mount(old.mnt);
3062 if (IS_MNT_SHARED(old_mnt) ||
3063 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3064 IS_MNT_SHARED(root_mnt->mnt_parent))
3066 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3068 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3071 if (d_unlinked(new.dentry))
3074 if (new_mnt == root_mnt || old_mnt == root_mnt)
3075 goto out4; /* loop, on the same file system */
3077 if (root.mnt->mnt_root != root.dentry)
3078 goto out4; /* not a mountpoint */
3079 if (!mnt_has_parent(root_mnt))
3080 goto out4; /* not attached */
3081 root_mp = root_mnt->mnt_mp;
3082 if (new.mnt->mnt_root != new.dentry)
3083 goto out4; /* not a mountpoint */
3084 if (!mnt_has_parent(new_mnt))
3085 goto out4; /* not attached */
3086 /* make sure we can reach put_old from new_root */
3087 if (!is_path_reachable(old_mnt, old.dentry, &new))
3089 /* make certain new is below the root */
3090 if (!is_path_reachable(new_mnt, new.dentry, &root))
3092 root_mp->m_count++; /* pin it so it won't go away */
3094 detach_mnt(new_mnt, &parent_path);
3095 detach_mnt(root_mnt, &root_parent);
3096 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3097 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3098 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3100 /* mount old root on put_old */
3101 attach_mnt(root_mnt, old_mnt, old_mp);
3102 /* mount new_root on / */
3103 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3104 touch_mnt_namespace(current->nsproxy->mnt_ns);
3105 /* A moved mount should not expire automatically */
3106 list_del_init(&new_mnt->mnt_expire);
3107 unlock_mount_hash();
3108 chroot_fs_refs(&root, &new);
3109 put_mountpoint(root_mp);
3112 unlock_mount(old_mp);
3114 path_put(&root_parent);
3115 path_put(&parent_path);
3127 static void __init init_mount_tree(void)
3129 struct vfsmount *mnt;
3130 struct mnt_namespace *ns;
3132 struct file_system_type *type;
3134 type = get_fs_type("rootfs");
3136 panic("Can't find rootfs type");
3137 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3138 put_filesystem(type);
3140 panic("Can't create rootfs");
3142 ns = create_mnt_ns(mnt);
3144 panic("Can't allocate initial namespace");
3146 init_task.nsproxy->mnt_ns = ns;
3150 root.dentry = mnt->mnt_root;
3151 mnt->mnt_flags |= MNT_LOCKED;
3153 set_fs_pwd(current->fs, &root);
3154 set_fs_root(current->fs, &root);
3157 void __init mnt_init(void)
3162 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3163 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3165 mount_hashtable = alloc_large_system_hash("Mount-cache",
3166 sizeof(struct hlist_head),
3169 &m_hash_shift, &m_hash_mask, 0, 0);
3170 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3171 sizeof(struct hlist_head),
3174 &mp_hash_shift, &mp_hash_mask, 0, 0);
3176 if (!mount_hashtable || !mountpoint_hashtable)
3177 panic("Failed to allocate mount hash table\n");
3179 for (u = 0; u <= m_hash_mask; u++)
3180 INIT_HLIST_HEAD(&mount_hashtable[u]);
3181 for (u = 0; u <= mp_hash_mask; u++)
3182 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3188 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3190 fs_kobj = kobject_create_and_add("fs", NULL);
3192 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3197 void put_mnt_ns(struct mnt_namespace *ns)
3199 if (!atomic_dec_and_test(&ns->count))
3201 drop_collected_mounts(&ns->root->mnt);
3205 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3207 struct vfsmount *mnt;
3208 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3211 * it is a longterm mount, don't release mnt until
3212 * we unmount before file sys is unregistered
3214 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3218 EXPORT_SYMBOL_GPL(kern_mount_data);
3220 void kern_unmount(struct vfsmount *mnt)
3222 /* release long term mount so mount point can be released */
3223 if (!IS_ERR_OR_NULL(mnt)) {
3224 real_mount(mnt)->mnt_ns = NULL;
3225 synchronize_rcu(); /* yecchhh... */
3229 EXPORT_SYMBOL(kern_unmount);
3231 bool our_mnt(struct vfsmount *mnt)
3233 return check_mnt(real_mount(mnt));
3236 bool current_chrooted(void)
3238 /* Does the current process have a non-standard root */
3239 struct path ns_root;
3240 struct path fs_root;
3243 /* Find the namespace root */
3244 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3245 ns_root.dentry = ns_root.mnt->mnt_root;
3247 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3250 get_fs_root(current->fs, &fs_root);
3252 chrooted = !path_equal(&fs_root, &ns_root);
3260 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3263 int new_flags = *new_mnt_flags;
3265 bool visible = false;
3267 down_read(&namespace_sem);
3268 list_for_each_entry(mnt, &ns->list, mnt_list) {
3269 struct mount *child;
3272 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3275 /* This mount is not fully visible if it's root directory
3276 * is not the root directory of the filesystem.
3278 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3281 /* A local view of the mount flags */
3282 mnt_flags = mnt->mnt.mnt_flags;
3284 /* Don't miss readonly hidden in the superblock flags */
3285 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3286 mnt_flags |= MNT_LOCK_READONLY;
3288 /* Verify the mount flags are equal to or more permissive
3289 * than the proposed new mount.
3291 if ((mnt_flags & MNT_LOCK_READONLY) &&
3292 !(new_flags & MNT_READONLY))
3294 if ((mnt_flags & MNT_LOCK_ATIME) &&
3295 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3298 /* This mount is not fully visible if there are any
3299 * locked child mounts that cover anything except for
3300 * empty directories.
3302 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3303 struct inode *inode = child->mnt_mountpoint->d_inode;
3304 /* Only worry about locked mounts */
3305 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3307 /* Is the directory permanetly empty? */
3308 if (!is_empty_dir_inode(inode))
3311 /* Preserve the locked attributes */
3312 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3319 up_read(&namespace_sem);
3323 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3325 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3326 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3327 unsigned long s_iflags;
3329 if (ns->user_ns == &init_user_ns)
3332 /* Can this filesystem be too revealing? */
3333 s_iflags = mnt->mnt_sb->s_iflags;
3334 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3337 if ((s_iflags & required_iflags) != required_iflags) {
3338 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3343 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3346 bool mnt_may_suid(struct vfsmount *mnt)
3349 * Foreign mounts (accessed via fchdir or through /proc
3350 * symlinks) are always treated as if they are nosuid. This
3351 * prevents namespaces from trusting potentially unsafe
3352 * suid/sgid bits, file caps, or security labels that originate
3353 * in other namespaces.
3355 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3356 current_in_userns(mnt->mnt_sb->s_user_ns);
3359 static struct ns_common *mntns_get(struct task_struct *task)
3361 struct ns_common *ns = NULL;
3362 struct nsproxy *nsproxy;
3365 nsproxy = task->nsproxy;
3367 ns = &nsproxy->mnt_ns->ns;
3368 get_mnt_ns(to_mnt_ns(ns));
3375 static void mntns_put(struct ns_common *ns)
3377 put_mnt_ns(to_mnt_ns(ns));
3380 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3382 struct fs_struct *fs = current->fs;
3383 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3386 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3387 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3388 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3395 put_mnt_ns(nsproxy->mnt_ns);
3396 nsproxy->mnt_ns = mnt_ns;
3399 root.mnt = &mnt_ns->root->mnt;
3400 root.dentry = mnt_ns->root->mnt.mnt_root;
3402 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3405 /* Update the pwd and root */
3406 set_fs_pwd(fs, &root);
3407 set_fs_root(fs, &root);
3413 static struct user_namespace *mntns_owner(struct ns_common *ns)
3415 return to_mnt_ns(ns)->user_ns;
3418 const struct proc_ns_operations mntns_operations = {
3420 .type = CLONE_NEWNS,
3423 .install = mntns_install,
3424 .owner = mntns_owner,