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 int __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);
611 /* call under rcu_read_lock */
612 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
614 int res = __legitimize_mnt(bastard, seq);
617 if (unlikely(res < 0)) {
626 * find the first mount at @dentry on vfsmount @mnt.
627 * call under rcu_read_lock()
629 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
631 struct hlist_head *head = m_hash(mnt, dentry);
634 hlist_for_each_entry_rcu(p, head, mnt_hash)
635 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
641 * find the last mount at @dentry on vfsmount @mnt.
642 * mount_lock must be held.
644 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
646 struct mount *p, *res = NULL;
647 p = __lookup_mnt(mnt, dentry);
650 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
652 hlist_for_each_entry_continue(p, mnt_hash) {
653 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
655 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
663 * lookup_mnt - Return the first child mount mounted at path
665 * "First" means first mounted chronologically. If you create the
668 * mount /dev/sda1 /mnt
669 * mount /dev/sda2 /mnt
670 * mount /dev/sda3 /mnt
672 * Then lookup_mnt() on the base /mnt dentry in the root mount will
673 * return successively the root dentry and vfsmount of /dev/sda1, then
674 * /dev/sda2, then /dev/sda3, then NULL.
676 * lookup_mnt takes a reference to the found vfsmount.
678 struct vfsmount *lookup_mnt(struct path *path)
680 struct mount *child_mnt;
686 seq = read_seqbegin(&mount_lock);
687 child_mnt = __lookup_mnt(path->mnt, path->dentry);
688 m = child_mnt ? &child_mnt->mnt : NULL;
689 } while (!legitimize_mnt(m, seq));
695 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
696 * current mount namespace.
698 * The common case is dentries are not mountpoints at all and that
699 * test is handled inline. For the slow case when we are actually
700 * dealing with a mountpoint of some kind, walk through all of the
701 * mounts in the current mount namespace and test to see if the dentry
704 * The mount_hashtable is not usable in the context because we
705 * need to identify all mounts that may be in the current mount
706 * namespace not just a mount that happens to have some specified
709 bool __is_local_mountpoint(struct dentry *dentry)
711 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
713 bool is_covered = false;
715 if (!d_mountpoint(dentry))
718 down_read(&namespace_sem);
719 list_for_each_entry(mnt, &ns->list, mnt_list) {
720 is_covered = (mnt->mnt_mountpoint == dentry);
724 up_read(&namespace_sem);
729 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
731 struct hlist_head *chain = mp_hash(dentry);
732 struct mountpoint *mp;
734 hlist_for_each_entry(mp, chain, m_hash) {
735 if (mp->m_dentry == dentry) {
736 /* might be worth a WARN_ON() */
737 if (d_unlinked(dentry))
738 return ERR_PTR(-ENOENT);
746 static struct mountpoint *get_mountpoint(struct dentry *dentry)
748 struct mountpoint *mp, *new = NULL;
751 if (d_mountpoint(dentry)) {
753 read_seqlock_excl(&mount_lock);
754 mp = lookup_mountpoint(dentry);
755 read_sequnlock_excl(&mount_lock);
761 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
763 return ERR_PTR(-ENOMEM);
766 /* Exactly one processes may set d_mounted */
767 ret = d_set_mounted(dentry);
769 /* Someone else set d_mounted? */
773 /* The dentry is not available as a mountpoint? */
778 /* Add the new mountpoint to the hash table */
779 read_seqlock_excl(&mount_lock);
780 new->m_dentry = dentry;
782 hlist_add_head(&new->m_hash, mp_hash(dentry));
783 INIT_HLIST_HEAD(&new->m_list);
784 read_sequnlock_excl(&mount_lock);
793 static void put_mountpoint(struct mountpoint *mp)
795 if (!--mp->m_count) {
796 struct dentry *dentry = mp->m_dentry;
797 BUG_ON(!hlist_empty(&mp->m_list));
798 spin_lock(&dentry->d_lock);
799 dentry->d_flags &= ~DCACHE_MOUNTED;
800 spin_unlock(&dentry->d_lock);
801 hlist_del(&mp->m_hash);
806 static inline int check_mnt(struct mount *mnt)
808 return mnt->mnt_ns == current->nsproxy->mnt_ns;
812 * vfsmount lock must be held for write
814 static void touch_mnt_namespace(struct mnt_namespace *ns)
818 wake_up_interruptible(&ns->poll);
823 * vfsmount lock must be held for write
825 static void __touch_mnt_namespace(struct mnt_namespace *ns)
827 if (ns && ns->event != event) {
829 wake_up_interruptible(&ns->poll);
834 * vfsmount lock must be held for write
836 static void unhash_mnt(struct mount *mnt)
838 mnt->mnt_parent = mnt;
839 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
840 list_del_init(&mnt->mnt_child);
841 hlist_del_init_rcu(&mnt->mnt_hash);
842 hlist_del_init(&mnt->mnt_mp_list);
843 put_mountpoint(mnt->mnt_mp);
848 * vfsmount lock must be held for write
850 static void detach_mnt(struct mount *mnt, struct path *old_path)
852 old_path->dentry = mnt->mnt_mountpoint;
853 old_path->mnt = &mnt->mnt_parent->mnt;
858 * vfsmount lock must be held for write
860 static void umount_mnt(struct mount *mnt)
862 /* old mountpoint will be dropped when we can do that */
863 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
868 * vfsmount lock must be held for write
870 void mnt_set_mountpoint(struct mount *mnt,
871 struct mountpoint *mp,
872 struct mount *child_mnt)
875 mnt_add_count(mnt, 1); /* essentially, that's mntget */
876 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
877 child_mnt->mnt_parent = mnt;
878 child_mnt->mnt_mp = mp;
879 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
883 * vfsmount lock must be held for write
885 static void attach_mnt(struct mount *mnt,
886 struct mount *parent,
887 struct mountpoint *mp)
889 mnt_set_mountpoint(parent, mp, mnt);
890 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
891 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
894 static void attach_shadowed(struct mount *mnt,
895 struct mount *parent,
896 struct mount *shadows)
899 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
900 list_add(&mnt->mnt_child, &shadows->mnt_child);
902 hlist_add_head_rcu(&mnt->mnt_hash,
903 m_hash(&parent->mnt, mnt->mnt_mountpoint));
904 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
909 * vfsmount lock must be held for write
911 static void commit_tree(struct mount *mnt, struct mount *shadows)
913 struct mount *parent = mnt->mnt_parent;
916 struct mnt_namespace *n = parent->mnt_ns;
918 BUG_ON(parent == mnt);
920 list_add_tail(&head, &mnt->mnt_list);
921 list_for_each_entry(m, &head, mnt_list)
924 list_splice(&head, n->list.prev);
926 attach_shadowed(mnt, parent, shadows);
927 touch_mnt_namespace(n);
930 static struct mount *next_mnt(struct mount *p, struct mount *root)
932 struct list_head *next = p->mnt_mounts.next;
933 if (next == &p->mnt_mounts) {
937 next = p->mnt_child.next;
938 if (next != &p->mnt_parent->mnt_mounts)
943 return list_entry(next, struct mount, mnt_child);
946 static struct mount *skip_mnt_tree(struct mount *p)
948 struct list_head *prev = p->mnt_mounts.prev;
949 while (prev != &p->mnt_mounts) {
950 p = list_entry(prev, struct mount, mnt_child);
951 prev = p->mnt_mounts.prev;
957 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
963 return ERR_PTR(-ENODEV);
965 mnt = alloc_vfsmnt(name);
967 return ERR_PTR(-ENOMEM);
969 if (flags & MS_KERNMOUNT)
970 mnt->mnt.mnt_flags = MNT_INTERNAL;
972 root = mount_fs(type, flags, name, data);
976 return ERR_CAST(root);
979 mnt->mnt.mnt_root = root;
980 mnt->mnt.mnt_sb = root->d_sb;
981 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
982 mnt->mnt_parent = mnt;
984 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
988 EXPORT_SYMBOL_GPL(vfs_kern_mount);
990 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
993 struct super_block *sb = old->mnt.mnt_sb;
997 mnt = alloc_vfsmnt(old->mnt_devname);
999 return ERR_PTR(-ENOMEM);
1001 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1002 mnt->mnt_group_id = 0; /* not a peer of original */
1004 mnt->mnt_group_id = old->mnt_group_id;
1006 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1007 err = mnt_alloc_group_id(mnt);
1012 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1013 /* Don't allow unprivileged users to change mount flags */
1014 if (flag & CL_UNPRIVILEGED) {
1015 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1017 if (mnt->mnt.mnt_flags & MNT_READONLY)
1018 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1020 if (mnt->mnt.mnt_flags & MNT_NODEV)
1021 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1023 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1024 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1026 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1027 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1030 /* Don't allow unprivileged users to reveal what is under a mount */
1031 if ((flag & CL_UNPRIVILEGED) &&
1032 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1033 mnt->mnt.mnt_flags |= MNT_LOCKED;
1035 atomic_inc(&sb->s_active);
1036 mnt->mnt.mnt_sb = sb;
1037 mnt->mnt.mnt_root = dget(root);
1038 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1039 mnt->mnt_parent = mnt;
1041 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1042 unlock_mount_hash();
1044 if ((flag & CL_SLAVE) ||
1045 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1046 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1047 mnt->mnt_master = old;
1048 CLEAR_MNT_SHARED(mnt);
1049 } else if (!(flag & CL_PRIVATE)) {
1050 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1051 list_add(&mnt->mnt_share, &old->mnt_share);
1052 if (IS_MNT_SLAVE(old))
1053 list_add(&mnt->mnt_slave, &old->mnt_slave);
1054 mnt->mnt_master = old->mnt_master;
1056 if (flag & CL_MAKE_SHARED)
1057 set_mnt_shared(mnt);
1059 /* stick the duplicate mount on the same expiry list
1060 * as the original if that was on one */
1061 if (flag & CL_EXPIRE) {
1062 if (!list_empty(&old->mnt_expire))
1063 list_add(&mnt->mnt_expire, &old->mnt_expire);
1071 return ERR_PTR(err);
1074 static void cleanup_mnt(struct mount *mnt)
1077 * This probably indicates that somebody messed
1078 * up a mnt_want/drop_write() pair. If this
1079 * happens, the filesystem was probably unable
1080 * to make r/w->r/o transitions.
1083 * The locking used to deal with mnt_count decrement provides barriers,
1084 * so mnt_get_writers() below is safe.
1086 WARN_ON(mnt_get_writers(mnt));
1087 if (unlikely(mnt->mnt_pins.first))
1089 fsnotify_vfsmount_delete(&mnt->mnt);
1090 dput(mnt->mnt.mnt_root);
1091 deactivate_super(mnt->mnt.mnt_sb);
1093 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1096 static void __cleanup_mnt(struct rcu_head *head)
1098 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1101 static LLIST_HEAD(delayed_mntput_list);
1102 static void delayed_mntput(struct work_struct *unused)
1104 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1105 struct llist_node *next;
1107 for (; node; node = next) {
1108 next = llist_next(node);
1109 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1112 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1114 static void mntput_no_expire(struct mount *mnt)
1117 mnt_add_count(mnt, -1);
1118 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1123 if (mnt_get_count(mnt)) {
1125 unlock_mount_hash();
1128 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1130 unlock_mount_hash();
1133 mnt->mnt.mnt_flags |= MNT_DOOMED;
1136 list_del(&mnt->mnt_instance);
1138 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1139 struct mount *p, *tmp;
1140 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1144 unlock_mount_hash();
1146 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1147 struct task_struct *task = current;
1148 if (likely(!(task->flags & PF_KTHREAD))) {
1149 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1150 if (!task_work_add(task, &mnt->mnt_rcu, true))
1153 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1154 schedule_delayed_work(&delayed_mntput_work, 1);
1160 void mntput(struct vfsmount *mnt)
1163 struct mount *m = real_mount(mnt);
1164 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1165 if (unlikely(m->mnt_expiry_mark))
1166 m->mnt_expiry_mark = 0;
1167 mntput_no_expire(m);
1170 EXPORT_SYMBOL(mntput);
1172 struct vfsmount *mntget(struct vfsmount *mnt)
1175 mnt_add_count(real_mount(mnt), 1);
1178 EXPORT_SYMBOL(mntget);
1180 struct vfsmount *mnt_clone_internal(struct path *path)
1183 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1186 p->mnt.mnt_flags |= MNT_INTERNAL;
1190 static inline void mangle(struct seq_file *m, const char *s)
1192 seq_escape(m, s, " \t\n\\");
1196 * Simple .show_options callback for filesystems which don't want to
1197 * implement more complex mount option showing.
1199 * See also save_mount_options().
1201 int generic_show_options(struct seq_file *m, struct dentry *root)
1203 const char *options;
1206 options = rcu_dereference(root->d_sb->s_options);
1208 if (options != NULL && options[0]) {
1216 EXPORT_SYMBOL(generic_show_options);
1219 * If filesystem uses generic_show_options(), this function should be
1220 * called from the fill_super() callback.
1222 * The .remount_fs callback usually needs to be handled in a special
1223 * way, to make sure, that previous options are not overwritten if the
1226 * Also note, that if the filesystem's .remount_fs function doesn't
1227 * reset all options to their default value, but changes only newly
1228 * given options, then the displayed options will not reflect reality
1231 void save_mount_options(struct super_block *sb, char *options)
1233 BUG_ON(sb->s_options);
1234 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1236 EXPORT_SYMBOL(save_mount_options);
1238 void replace_mount_options(struct super_block *sb, char *options)
1240 char *old = sb->s_options;
1241 rcu_assign_pointer(sb->s_options, options);
1247 EXPORT_SYMBOL(replace_mount_options);
1249 #ifdef CONFIG_PROC_FS
1250 /* iterator; we want it to have access to namespace_sem, thus here... */
1251 static void *m_start(struct seq_file *m, loff_t *pos)
1253 struct proc_mounts *p = m->private;
1255 down_read(&namespace_sem);
1256 if (p->cached_event == p->ns->event) {
1257 void *v = p->cached_mount;
1258 if (*pos == p->cached_index)
1260 if (*pos == p->cached_index + 1) {
1261 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1262 return p->cached_mount = v;
1266 p->cached_event = p->ns->event;
1267 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1268 p->cached_index = *pos;
1269 return p->cached_mount;
1272 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1274 struct proc_mounts *p = m->private;
1276 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1277 p->cached_index = *pos;
1278 return p->cached_mount;
1281 static void m_stop(struct seq_file *m, void *v)
1283 up_read(&namespace_sem);
1286 static int m_show(struct seq_file *m, void *v)
1288 struct proc_mounts *p = m->private;
1289 struct mount *r = list_entry(v, struct mount, mnt_list);
1290 return p->show(m, &r->mnt);
1293 const struct seq_operations mounts_op = {
1299 #endif /* CONFIG_PROC_FS */
1302 * may_umount_tree - check if a mount tree is busy
1303 * @mnt: root of mount tree
1305 * This is called to check if a tree of mounts has any
1306 * open files, pwds, chroots or sub mounts that are
1309 int may_umount_tree(struct vfsmount *m)
1311 struct mount *mnt = real_mount(m);
1312 int actual_refs = 0;
1313 int minimum_refs = 0;
1317 /* write lock needed for mnt_get_count */
1319 for (p = mnt; p; p = next_mnt(p, mnt)) {
1320 actual_refs += mnt_get_count(p);
1323 unlock_mount_hash();
1325 if (actual_refs > minimum_refs)
1331 EXPORT_SYMBOL(may_umount_tree);
1334 * may_umount - check if a mount point is busy
1335 * @mnt: root of mount
1337 * This is called to check if a mount point has any
1338 * open files, pwds, chroots or sub mounts. If the
1339 * mount has sub mounts this will return busy
1340 * regardless of whether the sub mounts are busy.
1342 * Doesn't take quota and stuff into account. IOW, in some cases it will
1343 * give false negatives. The main reason why it's here is that we need
1344 * a non-destructive way to look for easily umountable filesystems.
1346 int may_umount(struct vfsmount *mnt)
1349 down_read(&namespace_sem);
1351 if (propagate_mount_busy(real_mount(mnt), 2))
1353 unlock_mount_hash();
1354 up_read(&namespace_sem);
1358 EXPORT_SYMBOL(may_umount);
1360 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1362 static void namespace_unlock(void)
1364 struct hlist_head head;
1366 hlist_move_list(&unmounted, &head);
1368 up_write(&namespace_sem);
1370 if (likely(hlist_empty(&head)))
1375 group_pin_kill(&head);
1378 static inline void namespace_lock(void)
1380 down_write(&namespace_sem);
1383 enum umount_tree_flags {
1385 UMOUNT_PROPAGATE = 2,
1386 UMOUNT_CONNECTED = 4,
1389 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1391 /* Leaving mounts connected is only valid for lazy umounts */
1392 if (how & UMOUNT_SYNC)
1395 /* A mount without a parent has nothing to be connected to */
1396 if (!mnt_has_parent(mnt))
1399 /* Because the reference counting rules change when mounts are
1400 * unmounted and connected, umounted mounts may not be
1401 * connected to mounted mounts.
1403 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1406 /* Has it been requested that the mount remain connected? */
1407 if (how & UMOUNT_CONNECTED)
1410 /* Is the mount locked such that it needs to remain connected? */
1411 if (IS_MNT_LOCKED(mnt))
1414 /* By default disconnect the mount */
1419 * mount_lock must be held
1420 * namespace_sem must be held for write
1422 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1424 LIST_HEAD(tmp_list);
1427 if (how & UMOUNT_PROPAGATE)
1428 propagate_mount_unlock(mnt);
1430 /* Gather the mounts to umount */
1431 for (p = mnt; p; p = next_mnt(p, mnt)) {
1432 p->mnt.mnt_flags |= MNT_UMOUNT;
1433 list_move(&p->mnt_list, &tmp_list);
1436 /* Hide the mounts from mnt_mounts */
1437 list_for_each_entry(p, &tmp_list, mnt_list) {
1438 list_del_init(&p->mnt_child);
1441 /* Add propogated mounts to the tmp_list */
1442 if (how & UMOUNT_PROPAGATE)
1443 propagate_umount(&tmp_list);
1445 while (!list_empty(&tmp_list)) {
1447 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1448 list_del_init(&p->mnt_expire);
1449 list_del_init(&p->mnt_list);
1450 __touch_mnt_namespace(p->mnt_ns);
1452 if (how & UMOUNT_SYNC)
1453 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1455 disconnect = disconnect_mount(p, how);
1457 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1458 disconnect ? &unmounted : NULL);
1459 if (mnt_has_parent(p)) {
1460 mnt_add_count(p->mnt_parent, -1);
1462 /* Don't forget about p */
1463 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1468 change_mnt_propagation(p, MS_PRIVATE);
1472 static void shrink_submounts(struct mount *mnt);
1474 static int do_umount(struct mount *mnt, int flags)
1476 struct super_block *sb = mnt->mnt.mnt_sb;
1479 retval = security_sb_umount(&mnt->mnt, flags);
1484 * Allow userspace to request a mountpoint be expired rather than
1485 * unmounting unconditionally. Unmount only happens if:
1486 * (1) the mark is already set (the mark is cleared by mntput())
1487 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1489 if (flags & MNT_EXPIRE) {
1490 if (&mnt->mnt == current->fs->root.mnt ||
1491 flags & (MNT_FORCE | MNT_DETACH))
1495 * probably don't strictly need the lock here if we examined
1496 * all race cases, but it's a slowpath.
1499 if (mnt_get_count(mnt) != 2) {
1500 unlock_mount_hash();
1503 unlock_mount_hash();
1505 if (!xchg(&mnt->mnt_expiry_mark, 1))
1510 * If we may have to abort operations to get out of this
1511 * mount, and they will themselves hold resources we must
1512 * allow the fs to do things. In the Unix tradition of
1513 * 'Gee thats tricky lets do it in userspace' the umount_begin
1514 * might fail to complete on the first run through as other tasks
1515 * must return, and the like. Thats for the mount program to worry
1516 * about for the moment.
1519 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1520 sb->s_op->umount_begin(sb);
1524 * No sense to grab the lock for this test, but test itself looks
1525 * somewhat bogus. Suggestions for better replacement?
1526 * Ho-hum... In principle, we might treat that as umount + switch
1527 * to rootfs. GC would eventually take care of the old vfsmount.
1528 * Actually it makes sense, especially if rootfs would contain a
1529 * /reboot - static binary that would close all descriptors and
1530 * call reboot(9). Then init(8) could umount root and exec /reboot.
1532 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1534 * Special case for "unmounting" root ...
1535 * we just try to remount it readonly.
1537 if (!capable(CAP_SYS_ADMIN))
1539 down_write(&sb->s_umount);
1540 if (!(sb->s_flags & MS_RDONLY))
1541 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1542 up_write(&sb->s_umount);
1550 if (flags & MNT_DETACH) {
1551 if (!list_empty(&mnt->mnt_list))
1552 umount_tree(mnt, UMOUNT_PROPAGATE);
1555 shrink_submounts(mnt);
1557 if (!propagate_mount_busy(mnt, 2)) {
1558 if (!list_empty(&mnt->mnt_list))
1559 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1563 unlock_mount_hash();
1569 * __detach_mounts - lazily unmount all mounts on the specified dentry
1571 * During unlink, rmdir, and d_drop it is possible to loose the path
1572 * to an existing mountpoint, and wind up leaking the mount.
1573 * detach_mounts allows lazily unmounting those mounts instead of
1576 * The caller may hold dentry->d_inode->i_mutex.
1578 void __detach_mounts(struct dentry *dentry)
1580 struct mountpoint *mp;
1585 mp = lookup_mountpoint(dentry);
1586 if (IS_ERR_OR_NULL(mp))
1590 while (!hlist_empty(&mp->m_list)) {
1591 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1592 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1593 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1596 else umount_tree(mnt, UMOUNT_CONNECTED);
1600 unlock_mount_hash();
1605 * Is the caller allowed to modify his namespace?
1607 static inline bool may_mount(void)
1609 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1613 * Now umount can handle mount points as well as block devices.
1614 * This is important for filesystems which use unnamed block devices.
1616 * We now support a flag for forced unmount like the other 'big iron'
1617 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1620 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1625 int lookup_flags = 0;
1627 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1633 if (!(flags & UMOUNT_NOFOLLOW))
1634 lookup_flags |= LOOKUP_FOLLOW;
1636 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1639 mnt = real_mount(path.mnt);
1641 if (path.dentry != path.mnt->mnt_root)
1643 if (!check_mnt(mnt))
1645 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1648 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1651 retval = do_umount(mnt, flags);
1653 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1655 mntput_no_expire(mnt);
1660 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1663 * The 2.0 compatible umount. No flags.
1665 SYSCALL_DEFINE1(oldumount, char __user *, name)
1667 return sys_umount(name, 0);
1672 static bool is_mnt_ns_file(struct dentry *dentry)
1674 /* Is this a proxy for a mount namespace? */
1675 return dentry->d_op == &ns_dentry_operations &&
1676 dentry->d_fsdata == &mntns_operations;
1679 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1681 return container_of(ns, struct mnt_namespace, ns);
1684 static bool mnt_ns_loop(struct dentry *dentry)
1686 /* Could bind mounting the mount namespace inode cause a
1687 * mount namespace loop?
1689 struct mnt_namespace *mnt_ns;
1690 if (!is_mnt_ns_file(dentry))
1693 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1694 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1697 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1700 struct mount *res, *p, *q, *r, *parent;
1702 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1703 return ERR_PTR(-EINVAL);
1705 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1706 return ERR_PTR(-EINVAL);
1708 res = q = clone_mnt(mnt, dentry, flag);
1712 q->mnt_mountpoint = mnt->mnt_mountpoint;
1715 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1717 if (!is_subdir(r->mnt_mountpoint, dentry))
1720 for (s = r; s; s = next_mnt(s, r)) {
1721 struct mount *t = NULL;
1722 if (!(flag & CL_COPY_UNBINDABLE) &&
1723 IS_MNT_UNBINDABLE(s)) {
1724 s = skip_mnt_tree(s);
1727 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1728 is_mnt_ns_file(s->mnt.mnt_root)) {
1729 s = skip_mnt_tree(s);
1732 while (p != s->mnt_parent) {
1738 q = clone_mnt(p, p->mnt.mnt_root, flag);
1742 list_add_tail(&q->mnt_list, &res->mnt_list);
1743 mnt_set_mountpoint(parent, p->mnt_mp, q);
1744 if (!list_empty(&parent->mnt_mounts)) {
1745 t = list_last_entry(&parent->mnt_mounts,
1746 struct mount, mnt_child);
1747 if (t->mnt_mp != p->mnt_mp)
1750 attach_shadowed(q, parent, t);
1751 unlock_mount_hash();
1758 umount_tree(res, UMOUNT_SYNC);
1759 unlock_mount_hash();
1764 /* Caller should check returned pointer for errors */
1766 struct vfsmount *collect_mounts(struct path *path)
1770 if (!check_mnt(real_mount(path->mnt)))
1771 tree = ERR_PTR(-EINVAL);
1773 tree = copy_tree(real_mount(path->mnt), path->dentry,
1774 CL_COPY_ALL | CL_PRIVATE);
1777 return ERR_CAST(tree);
1781 void drop_collected_mounts(struct vfsmount *mnt)
1785 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1786 unlock_mount_hash();
1791 * clone_private_mount - create a private clone of a path
1793 * This creates a new vfsmount, which will be the clone of @path. The new will
1794 * not be attached anywhere in the namespace and will be private (i.e. changes
1795 * to the originating mount won't be propagated into this).
1797 * Release with mntput().
1799 struct vfsmount *clone_private_mount(struct path *path)
1801 struct mount *old_mnt = real_mount(path->mnt);
1802 struct mount *new_mnt;
1804 if (IS_MNT_UNBINDABLE(old_mnt))
1805 return ERR_PTR(-EINVAL);
1807 down_read(&namespace_sem);
1808 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1809 up_read(&namespace_sem);
1810 if (IS_ERR(new_mnt))
1811 return ERR_CAST(new_mnt);
1813 return &new_mnt->mnt;
1815 EXPORT_SYMBOL_GPL(clone_private_mount);
1817 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1818 struct vfsmount *root)
1821 int res = f(root, arg);
1824 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1825 res = f(&mnt->mnt, arg);
1832 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1836 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1837 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1838 mnt_release_group_id(p);
1842 static int invent_group_ids(struct mount *mnt, bool recurse)
1846 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1847 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1848 int err = mnt_alloc_group_id(p);
1850 cleanup_group_ids(mnt, p);
1860 * @source_mnt : mount tree to be attached
1861 * @nd : place the mount tree @source_mnt is attached
1862 * @parent_nd : if non-null, detach the source_mnt from its parent and
1863 * store the parent mount and mountpoint dentry.
1864 * (done when source_mnt is moved)
1866 * NOTE: in the table below explains the semantics when a source mount
1867 * of a given type is attached to a destination mount of a given type.
1868 * ---------------------------------------------------------------------------
1869 * | BIND MOUNT OPERATION |
1870 * |**************************************************************************
1871 * | source-->| shared | private | slave | unbindable |
1875 * |**************************************************************************
1876 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1878 * |non-shared| shared (+) | private | slave (*) | invalid |
1879 * ***************************************************************************
1880 * A bind operation clones the source mount and mounts the clone on the
1881 * destination mount.
1883 * (++) the cloned mount is propagated to all the mounts in the propagation
1884 * tree of the destination mount and the cloned mount is added to
1885 * the peer group of the source mount.
1886 * (+) the cloned mount is created under the destination mount and is marked
1887 * as shared. The cloned mount is added to the peer group of the source
1889 * (+++) the mount is propagated to all the mounts in the propagation tree
1890 * of the destination mount and the cloned mount is made slave
1891 * of the same master as that of the source mount. The cloned mount
1892 * is marked as 'shared and slave'.
1893 * (*) the cloned mount is made a slave of the same master as that of the
1896 * ---------------------------------------------------------------------------
1897 * | MOVE MOUNT OPERATION |
1898 * |**************************************************************************
1899 * | source-->| shared | private | slave | unbindable |
1903 * |**************************************************************************
1904 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1906 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1907 * ***************************************************************************
1909 * (+) the mount is moved to the destination. And is then propagated to
1910 * all the mounts in the propagation tree of the destination mount.
1911 * (+*) the mount is moved to the destination.
1912 * (+++) the mount is moved to the destination and is then propagated to
1913 * all the mounts belonging to the destination mount's propagation tree.
1914 * the mount is marked as 'shared and slave'.
1915 * (*) the mount continues to be a slave at the new location.
1917 * if the source mount is a tree, the operations explained above is
1918 * applied to each mount in the tree.
1919 * Must be called without spinlocks held, since this function can sleep
1922 static int attach_recursive_mnt(struct mount *source_mnt,
1923 struct mount *dest_mnt,
1924 struct mountpoint *dest_mp,
1925 struct path *parent_path)
1927 HLIST_HEAD(tree_list);
1928 struct mount *child, *p;
1929 struct hlist_node *n;
1932 if (IS_MNT_SHARED(dest_mnt)) {
1933 err = invent_group_ids(source_mnt, true);
1936 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1939 goto out_cleanup_ids;
1940 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1946 detach_mnt(source_mnt, parent_path);
1947 attach_mnt(source_mnt, dest_mnt, dest_mp);
1948 touch_mnt_namespace(source_mnt->mnt_ns);
1950 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1951 commit_tree(source_mnt, NULL);
1954 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1956 hlist_del_init(&child->mnt_hash);
1957 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1958 child->mnt_mountpoint);
1959 commit_tree(child, q);
1961 unlock_mount_hash();
1966 while (!hlist_empty(&tree_list)) {
1967 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1968 umount_tree(child, UMOUNT_SYNC);
1970 unlock_mount_hash();
1971 cleanup_group_ids(source_mnt, NULL);
1976 static struct mountpoint *lock_mount(struct path *path)
1978 struct vfsmount *mnt;
1979 struct dentry *dentry = path->dentry;
1981 mutex_lock(&dentry->d_inode->i_mutex);
1982 if (unlikely(cant_mount(dentry))) {
1983 mutex_unlock(&dentry->d_inode->i_mutex);
1984 return ERR_PTR(-ENOENT);
1987 mnt = lookup_mnt(path);
1989 struct mountpoint *mp = get_mountpoint(dentry);
1992 mutex_unlock(&dentry->d_inode->i_mutex);
1998 mutex_unlock(&path->dentry->d_inode->i_mutex);
2001 dentry = path->dentry = dget(mnt->mnt_root);
2005 static void unlock_mount(struct mountpoint *where)
2007 struct dentry *dentry = where->m_dentry;
2009 read_seqlock_excl(&mount_lock);
2010 put_mountpoint(where);
2011 read_sequnlock_excl(&mount_lock);
2014 mutex_unlock(&dentry->d_inode->i_mutex);
2017 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2019 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2022 if (d_is_dir(mp->m_dentry) !=
2023 d_is_dir(mnt->mnt.mnt_root))
2026 return attach_recursive_mnt(mnt, p, mp, NULL);
2030 * Sanity check the flags to change_mnt_propagation.
2033 static int flags_to_propagation_type(int flags)
2035 int type = flags & ~(MS_REC | MS_SILENT);
2037 /* Fail if any non-propagation flags are set */
2038 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2040 /* Only one propagation flag should be set */
2041 if (!is_power_of_2(type))
2047 * recursively change the type of the mountpoint.
2049 static int do_change_type(struct path *path, int flag)
2052 struct mount *mnt = real_mount(path->mnt);
2053 int recurse = flag & MS_REC;
2057 if (path->dentry != path->mnt->mnt_root)
2060 type = flags_to_propagation_type(flag);
2065 if (type == MS_SHARED) {
2066 err = invent_group_ids(mnt, recurse);
2072 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2073 change_mnt_propagation(m, type);
2074 unlock_mount_hash();
2081 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2083 struct mount *child;
2084 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2085 if (!is_subdir(child->mnt_mountpoint, dentry))
2088 if (child->mnt.mnt_flags & MNT_LOCKED)
2095 * do loopback mount.
2097 static int do_loopback(struct path *path, const char *old_name,
2100 struct path old_path;
2101 struct mount *mnt = NULL, *old, *parent;
2102 struct mountpoint *mp;
2104 if (!old_name || !*old_name)
2106 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2111 if (mnt_ns_loop(old_path.dentry))
2114 mp = lock_mount(path);
2119 old = real_mount(old_path.mnt);
2120 parent = real_mount(path->mnt);
2123 if (IS_MNT_UNBINDABLE(old))
2126 if (!check_mnt(parent))
2129 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2132 if (!recurse && has_locked_children(old, old_path.dentry))
2136 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2138 mnt = clone_mnt(old, old_path.dentry, 0);
2145 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2147 err = graft_tree(mnt, parent, mp);
2150 umount_tree(mnt, UMOUNT_SYNC);
2151 unlock_mount_hash();
2156 path_put(&old_path);
2160 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2163 int readonly_request = 0;
2165 if (ms_flags & MS_RDONLY)
2166 readonly_request = 1;
2167 if (readonly_request == __mnt_is_readonly(mnt))
2170 if (readonly_request)
2171 error = mnt_make_readonly(real_mount(mnt));
2173 __mnt_unmake_readonly(real_mount(mnt));
2178 * change filesystem flags. dir should be a physical root of filesystem.
2179 * If you've mounted a non-root directory somewhere and want to do remount
2180 * on it - tough luck.
2182 static int do_remount(struct path *path, int flags, int mnt_flags,
2186 struct super_block *sb = path->mnt->mnt_sb;
2187 struct mount *mnt = real_mount(path->mnt);
2189 if (!check_mnt(mnt))
2192 if (path->dentry != path->mnt->mnt_root)
2195 /* Don't allow changing of locked mnt flags.
2197 * No locks need to be held here while testing the various
2198 * MNT_LOCK flags because those flags can never be cleared
2199 * once they are set.
2201 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2202 !(mnt_flags & MNT_READONLY)) {
2205 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2206 !(mnt_flags & MNT_NODEV)) {
2207 /* Was the nodev implicitly added in mount? */
2208 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2209 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2210 mnt_flags |= MNT_NODEV;
2215 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2216 !(mnt_flags & MNT_NOSUID)) {
2219 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2220 !(mnt_flags & MNT_NOEXEC)) {
2223 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2224 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2228 err = security_sb_remount(sb, data);
2232 down_write(&sb->s_umount);
2233 if (flags & MS_BIND)
2234 err = change_mount_flags(path->mnt, flags);
2235 else if (!capable(CAP_SYS_ADMIN))
2238 err = do_remount_sb(sb, flags, data, 0);
2241 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2242 mnt->mnt.mnt_flags = mnt_flags;
2243 touch_mnt_namespace(mnt->mnt_ns);
2244 unlock_mount_hash();
2246 up_write(&sb->s_umount);
2250 static inline int tree_contains_unbindable(struct mount *mnt)
2253 for (p = mnt; p; p = next_mnt(p, mnt)) {
2254 if (IS_MNT_UNBINDABLE(p))
2260 static int do_move_mount(struct path *path, const char *old_name)
2262 struct path old_path, parent_path;
2265 struct mountpoint *mp;
2267 if (!old_name || !*old_name)
2269 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2273 mp = lock_mount(path);
2278 old = real_mount(old_path.mnt);
2279 p = real_mount(path->mnt);
2282 if (!check_mnt(p) || !check_mnt(old))
2285 if (old->mnt.mnt_flags & MNT_LOCKED)
2289 if (old_path.dentry != old_path.mnt->mnt_root)
2292 if (!mnt_has_parent(old))
2295 if (d_is_dir(path->dentry) !=
2296 d_is_dir(old_path.dentry))
2299 * Don't move a mount residing in a shared parent.
2301 if (IS_MNT_SHARED(old->mnt_parent))
2304 * Don't move a mount tree containing unbindable mounts to a destination
2305 * mount which is shared.
2307 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2310 for (; mnt_has_parent(p); p = p->mnt_parent)
2314 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2318 /* if the mount is moved, it should no longer be expire
2320 list_del_init(&old->mnt_expire);
2325 path_put(&parent_path);
2326 path_put(&old_path);
2330 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2333 const char *subtype = strchr(fstype, '.');
2342 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2344 if (!mnt->mnt_sb->s_subtype)
2350 return ERR_PTR(err);
2354 * add a mount into a namespace's mount tree
2356 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2358 struct mountpoint *mp;
2359 struct mount *parent;
2362 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2364 mp = lock_mount(path);
2368 parent = real_mount(path->mnt);
2370 if (unlikely(!check_mnt(parent))) {
2371 /* that's acceptable only for automounts done in private ns */
2372 if (!(mnt_flags & MNT_SHRINKABLE))
2374 /* ... and for those we'd better have mountpoint still alive */
2375 if (!parent->mnt_ns)
2379 /* Refuse the same filesystem on the same mount point */
2381 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2382 path->mnt->mnt_root == path->dentry)
2386 if (d_is_symlink(newmnt->mnt.mnt_root))
2389 newmnt->mnt.mnt_flags = mnt_flags;
2390 err = graft_tree(newmnt, parent, mp);
2397 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2400 * create a new mount for userspace and request it to be added into the
2403 static int do_new_mount(struct path *path, const char *fstype, int flags,
2404 int mnt_flags, const char *name, void *data)
2406 struct file_system_type *type;
2407 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2408 struct vfsmount *mnt;
2414 type = get_fs_type(fstype);
2418 if (user_ns != &init_user_ns) {
2419 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2420 put_filesystem(type);
2423 /* Only in special cases allow devices from mounts
2424 * created outside the initial user namespace.
2426 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2428 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2430 if (type->fs_flags & FS_USERNS_VISIBLE) {
2431 if (!fs_fully_visible(type, &mnt_flags)) {
2432 put_filesystem(type);
2438 mnt = vfs_kern_mount(type, flags, name, data);
2439 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2440 !mnt->mnt_sb->s_subtype)
2441 mnt = fs_set_subtype(mnt, fstype);
2443 put_filesystem(type);
2445 return PTR_ERR(mnt);
2447 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2453 int finish_automount(struct vfsmount *m, struct path *path)
2455 struct mount *mnt = real_mount(m);
2457 /* The new mount record should have at least 2 refs to prevent it being
2458 * expired before we get a chance to add it
2460 BUG_ON(mnt_get_count(mnt) < 2);
2462 if (m->mnt_sb == path->mnt->mnt_sb &&
2463 m->mnt_root == path->dentry) {
2468 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2472 /* remove m from any expiration list it may be on */
2473 if (!list_empty(&mnt->mnt_expire)) {
2475 list_del_init(&mnt->mnt_expire);
2484 * mnt_set_expiry - Put a mount on an expiration list
2485 * @mnt: The mount to list.
2486 * @expiry_list: The list to add the mount to.
2488 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2492 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2496 EXPORT_SYMBOL(mnt_set_expiry);
2499 * process a list of expirable mountpoints with the intent of discarding any
2500 * mountpoints that aren't in use and haven't been touched since last we came
2503 void mark_mounts_for_expiry(struct list_head *mounts)
2505 struct mount *mnt, *next;
2506 LIST_HEAD(graveyard);
2508 if (list_empty(mounts))
2514 /* extract from the expiration list every vfsmount that matches the
2515 * following criteria:
2516 * - only referenced by its parent vfsmount
2517 * - still marked for expiry (marked on the last call here; marks are
2518 * cleared by mntput())
2520 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2521 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2522 propagate_mount_busy(mnt, 1))
2524 list_move(&mnt->mnt_expire, &graveyard);
2526 while (!list_empty(&graveyard)) {
2527 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2528 touch_mnt_namespace(mnt->mnt_ns);
2529 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2531 unlock_mount_hash();
2535 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2538 * Ripoff of 'select_parent()'
2540 * search the list of submounts for a given mountpoint, and move any
2541 * shrinkable submounts to the 'graveyard' list.
2543 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2545 struct mount *this_parent = parent;
2546 struct list_head *next;
2550 next = this_parent->mnt_mounts.next;
2552 while (next != &this_parent->mnt_mounts) {
2553 struct list_head *tmp = next;
2554 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2557 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2560 * Descend a level if the d_mounts list is non-empty.
2562 if (!list_empty(&mnt->mnt_mounts)) {
2567 if (!propagate_mount_busy(mnt, 1)) {
2568 list_move_tail(&mnt->mnt_expire, graveyard);
2573 * All done at this level ... ascend and resume the search
2575 if (this_parent != parent) {
2576 next = this_parent->mnt_child.next;
2577 this_parent = this_parent->mnt_parent;
2584 * process a list of expirable mountpoints with the intent of discarding any
2585 * submounts of a specific parent mountpoint
2587 * mount_lock must be held for write
2589 static void shrink_submounts(struct mount *mnt)
2591 LIST_HEAD(graveyard);
2594 /* extract submounts of 'mountpoint' from the expiration list */
2595 while (select_submounts(mnt, &graveyard)) {
2596 while (!list_empty(&graveyard)) {
2597 m = list_first_entry(&graveyard, struct mount,
2599 touch_mnt_namespace(m->mnt_ns);
2600 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2606 * Some copy_from_user() implementations do not return the exact number of
2607 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2608 * Note that this function differs from copy_from_user() in that it will oops
2609 * on bad values of `to', rather than returning a short copy.
2611 static long exact_copy_from_user(void *to, const void __user * from,
2615 const char __user *f = from;
2618 if (!access_ok(VERIFY_READ, from, n))
2622 if (__get_user(c, f)) {
2633 int copy_mount_options(const void __user * data, unsigned long *where)
2643 if (!(page = __get_free_page(GFP_KERNEL)))
2646 /* We only care that *some* data at the address the user
2647 * gave us is valid. Just in case, we'll zero
2648 * the remainder of the page.
2650 /* copy_from_user cannot cross TASK_SIZE ! */
2651 size = TASK_SIZE - (unsigned long)data;
2652 if (size > PAGE_SIZE)
2655 i = size - exact_copy_from_user((void *)page, data, size);
2661 memset((char *)page + i, 0, PAGE_SIZE - i);
2666 char *copy_mount_string(const void __user *data)
2668 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2672 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2673 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2675 * data is a (void *) that can point to any structure up to
2676 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2677 * information (or be NULL).
2679 * Pre-0.97 versions of mount() didn't have a flags word.
2680 * When the flags word was introduced its top half was required
2681 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2682 * Therefore, if this magic number is present, it carries no information
2683 * and must be discarded.
2685 long do_mount(const char *dev_name, const char __user *dir_name,
2686 const char *type_page, unsigned long flags, void *data_page)
2693 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2694 flags &= ~MS_MGC_MSK;
2696 /* Basic sanity checks */
2698 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2700 /* ... and get the mountpoint */
2701 retval = user_path(dir_name, &path);
2705 retval = security_sb_mount(dev_name, &path,
2706 type_page, flags, data_page);
2707 if (!retval && !may_mount())
2712 /* Default to relatime unless overriden */
2713 if (!(flags & MS_NOATIME))
2714 mnt_flags |= MNT_RELATIME;
2716 /* Separate the per-mountpoint flags */
2717 if (flags & MS_NOSUID)
2718 mnt_flags |= MNT_NOSUID;
2719 if (flags & MS_NODEV)
2720 mnt_flags |= MNT_NODEV;
2721 if (flags & MS_NOEXEC)
2722 mnt_flags |= MNT_NOEXEC;
2723 if (flags & MS_NOATIME)
2724 mnt_flags |= MNT_NOATIME;
2725 if (flags & MS_NODIRATIME)
2726 mnt_flags |= MNT_NODIRATIME;
2727 if (flags & MS_STRICTATIME)
2728 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2729 if (flags & MS_RDONLY)
2730 mnt_flags |= MNT_READONLY;
2732 /* The default atime for remount is preservation */
2733 if ((flags & MS_REMOUNT) &&
2734 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2735 MS_STRICTATIME)) == 0)) {
2736 mnt_flags &= ~MNT_ATIME_MASK;
2737 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2740 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2741 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2744 if (flags & MS_REMOUNT)
2745 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2747 else if (flags & MS_BIND)
2748 retval = do_loopback(&path, dev_name, flags & MS_REC);
2749 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2750 retval = do_change_type(&path, flags);
2751 else if (flags & MS_MOVE)
2752 retval = do_move_mount(&path, dev_name);
2754 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2755 dev_name, data_page);
2761 static void free_mnt_ns(struct mnt_namespace *ns)
2763 ns_free_inum(&ns->ns);
2764 put_user_ns(ns->user_ns);
2769 * Assign a sequence number so we can detect when we attempt to bind
2770 * mount a reference to an older mount namespace into the current
2771 * mount namespace, preventing reference counting loops. A 64bit
2772 * number incrementing at 10Ghz will take 12,427 years to wrap which
2773 * is effectively never, so we can ignore the possibility.
2775 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2777 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2779 struct mnt_namespace *new_ns;
2782 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2784 return ERR_PTR(-ENOMEM);
2785 ret = ns_alloc_inum(&new_ns->ns);
2788 return ERR_PTR(ret);
2790 new_ns->ns.ops = &mntns_operations;
2791 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2792 atomic_set(&new_ns->count, 1);
2793 new_ns->root = NULL;
2794 INIT_LIST_HEAD(&new_ns->list);
2795 init_waitqueue_head(&new_ns->poll);
2797 new_ns->user_ns = get_user_ns(user_ns);
2801 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2802 struct user_namespace *user_ns, struct fs_struct *new_fs)
2804 struct mnt_namespace *new_ns;
2805 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2806 struct mount *p, *q;
2813 if (likely(!(flags & CLONE_NEWNS))) {
2820 new_ns = alloc_mnt_ns(user_ns);
2825 /* First pass: copy the tree topology */
2826 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2827 if (user_ns != ns->user_ns)
2828 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2829 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2832 free_mnt_ns(new_ns);
2833 return ERR_CAST(new);
2836 list_add_tail(&new_ns->list, &new->mnt_list);
2839 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2840 * as belonging to new namespace. We have already acquired a private
2841 * fs_struct, so tsk->fs->lock is not needed.
2848 if (&p->mnt == new_fs->root.mnt) {
2849 new_fs->root.mnt = mntget(&q->mnt);
2852 if (&p->mnt == new_fs->pwd.mnt) {
2853 new_fs->pwd.mnt = mntget(&q->mnt);
2857 p = next_mnt(p, old);
2858 q = next_mnt(q, new);
2861 while (p->mnt.mnt_root != q->mnt.mnt_root)
2862 p = next_mnt(p, old);
2875 * create_mnt_ns - creates a private namespace and adds a root filesystem
2876 * @mnt: pointer to the new root filesystem mountpoint
2878 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2880 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2881 if (!IS_ERR(new_ns)) {
2882 struct mount *mnt = real_mount(m);
2883 mnt->mnt_ns = new_ns;
2885 list_add(&mnt->mnt_list, &new_ns->list);
2892 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2894 struct mnt_namespace *ns;
2895 struct super_block *s;
2899 ns = create_mnt_ns(mnt);
2901 return ERR_CAST(ns);
2903 err = vfs_path_lookup(mnt->mnt_root, mnt,
2904 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2909 return ERR_PTR(err);
2911 /* trade a vfsmount reference for active sb one */
2912 s = path.mnt->mnt_sb;
2913 atomic_inc(&s->s_active);
2915 /* lock the sucker */
2916 down_write(&s->s_umount);
2917 /* ... and return the root of (sub)tree on it */
2920 EXPORT_SYMBOL(mount_subtree);
2922 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2923 char __user *, type, unsigned long, flags, void __user *, data)
2928 unsigned long data_page;
2930 kernel_type = copy_mount_string(type);
2931 ret = PTR_ERR(kernel_type);
2932 if (IS_ERR(kernel_type))
2935 kernel_dev = copy_mount_string(dev_name);
2936 ret = PTR_ERR(kernel_dev);
2937 if (IS_ERR(kernel_dev))
2940 ret = copy_mount_options(data, &data_page);
2944 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2945 (void *) data_page);
2947 free_page(data_page);
2957 * Return true if path is reachable from root
2959 * namespace_sem or mount_lock is held
2961 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2962 const struct path *root)
2964 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2965 dentry = mnt->mnt_mountpoint;
2966 mnt = mnt->mnt_parent;
2968 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2971 int path_is_under(struct path *path1, struct path *path2)
2974 read_seqlock_excl(&mount_lock);
2975 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2976 read_sequnlock_excl(&mount_lock);
2979 EXPORT_SYMBOL(path_is_under);
2982 * pivot_root Semantics:
2983 * Moves the root file system of the current process to the directory put_old,
2984 * makes new_root as the new root file system of the current process, and sets
2985 * root/cwd of all processes which had them on the current root to new_root.
2988 * The new_root and put_old must be directories, and must not be on the
2989 * same file system as the current process root. The put_old must be
2990 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2991 * pointed to by put_old must yield the same directory as new_root. No other
2992 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2994 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2995 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2996 * in this situation.
2999 * - we don't move root/cwd if they are not at the root (reason: if something
3000 * cared enough to change them, it's probably wrong to force them elsewhere)
3001 * - it's okay to pick a root that isn't the root of a file system, e.g.
3002 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3003 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3006 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3007 const char __user *, put_old)
3009 struct path new, old, parent_path, root_parent, root;
3010 struct mount *new_mnt, *root_mnt, *old_mnt;
3011 struct mountpoint *old_mp, *root_mp;
3017 error = user_path_dir(new_root, &new);
3021 error = user_path_dir(put_old, &old);
3025 error = security_sb_pivotroot(&old, &new);
3029 get_fs_root(current->fs, &root);
3030 old_mp = lock_mount(&old);
3031 error = PTR_ERR(old_mp);
3036 new_mnt = real_mount(new.mnt);
3037 root_mnt = real_mount(root.mnt);
3038 old_mnt = real_mount(old.mnt);
3039 if (IS_MNT_SHARED(old_mnt) ||
3040 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3041 IS_MNT_SHARED(root_mnt->mnt_parent))
3043 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3045 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3048 if (d_unlinked(new.dentry))
3051 if (new_mnt == root_mnt || old_mnt == root_mnt)
3052 goto out4; /* loop, on the same file system */
3054 if (root.mnt->mnt_root != root.dentry)
3055 goto out4; /* not a mountpoint */
3056 if (!mnt_has_parent(root_mnt))
3057 goto out4; /* not attached */
3058 root_mp = root_mnt->mnt_mp;
3059 if (new.mnt->mnt_root != new.dentry)
3060 goto out4; /* not a mountpoint */
3061 if (!mnt_has_parent(new_mnt))
3062 goto out4; /* not attached */
3063 /* make sure we can reach put_old from new_root */
3064 if (!is_path_reachable(old_mnt, old.dentry, &new))
3066 /* make certain new is below the root */
3067 if (!is_path_reachable(new_mnt, new.dentry, &root))
3069 root_mp->m_count++; /* pin it so it won't go away */
3071 detach_mnt(new_mnt, &parent_path);
3072 detach_mnt(root_mnt, &root_parent);
3073 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3074 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3075 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3077 /* mount old root on put_old */
3078 attach_mnt(root_mnt, old_mnt, old_mp);
3079 /* mount new_root on / */
3080 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3081 touch_mnt_namespace(current->nsproxy->mnt_ns);
3082 /* A moved mount should not expire automatically */
3083 list_del_init(&new_mnt->mnt_expire);
3084 put_mountpoint(root_mp);
3085 unlock_mount_hash();
3086 chroot_fs_refs(&root, &new);
3089 unlock_mount(old_mp);
3091 path_put(&root_parent);
3092 path_put(&parent_path);
3104 static void __init init_mount_tree(void)
3106 struct vfsmount *mnt;
3107 struct mnt_namespace *ns;
3109 struct file_system_type *type;
3111 type = get_fs_type("rootfs");
3113 panic("Can't find rootfs type");
3114 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3115 put_filesystem(type);
3117 panic("Can't create rootfs");
3119 ns = create_mnt_ns(mnt);
3121 panic("Can't allocate initial namespace");
3123 init_task.nsproxy->mnt_ns = ns;
3127 root.dentry = mnt->mnt_root;
3128 mnt->mnt_flags |= MNT_LOCKED;
3130 set_fs_pwd(current->fs, &root);
3131 set_fs_root(current->fs, &root);
3134 void __init mnt_init(void)
3139 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3140 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3142 mount_hashtable = alloc_large_system_hash("Mount-cache",
3143 sizeof(struct hlist_head),
3146 &m_hash_shift, &m_hash_mask, 0, 0);
3147 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3148 sizeof(struct hlist_head),
3151 &mp_hash_shift, &mp_hash_mask, 0, 0);
3153 if (!mount_hashtable || !mountpoint_hashtable)
3154 panic("Failed to allocate mount hash table\n");
3156 for (u = 0; u <= m_hash_mask; u++)
3157 INIT_HLIST_HEAD(&mount_hashtable[u]);
3158 for (u = 0; u <= mp_hash_mask; u++)
3159 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3165 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3167 fs_kobj = kobject_create_and_add("fs", NULL);
3169 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3174 void put_mnt_ns(struct mnt_namespace *ns)
3176 if (!atomic_dec_and_test(&ns->count))
3178 drop_collected_mounts(&ns->root->mnt);
3182 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3184 struct vfsmount *mnt;
3185 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3188 * it is a longterm mount, don't release mnt until
3189 * we unmount before file sys is unregistered
3191 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3195 EXPORT_SYMBOL_GPL(kern_mount_data);
3197 void kern_unmount(struct vfsmount *mnt)
3199 /* release long term mount so mount point can be released */
3200 if (!IS_ERR_OR_NULL(mnt)) {
3201 real_mount(mnt)->mnt_ns = NULL;
3202 synchronize_rcu(); /* yecchhh... */
3206 EXPORT_SYMBOL(kern_unmount);
3208 bool our_mnt(struct vfsmount *mnt)
3210 return check_mnt(real_mount(mnt));
3213 bool current_chrooted(void)
3215 /* Does the current process have a non-standard root */
3216 struct path ns_root;
3217 struct path fs_root;
3220 /* Find the namespace root */
3221 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3222 ns_root.dentry = ns_root.mnt->mnt_root;
3224 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3227 get_fs_root(current->fs, &fs_root);
3229 chrooted = !path_equal(&fs_root, &ns_root);
3237 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3239 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3240 int new_flags = *new_mnt_flags;
3242 bool visible = false;
3247 down_read(&namespace_sem);
3248 list_for_each_entry(mnt, &ns->list, mnt_list) {
3249 struct mount *child;
3252 if (mnt->mnt.mnt_sb->s_type != type)
3255 /* This mount is not fully visible if it's root directory
3256 * is not the root directory of the filesystem.
3258 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3261 /* Read the mount flags and filter out flags that
3262 * may safely be ignored.
3264 mnt_flags = mnt->mnt.mnt_flags;
3265 if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3266 mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3268 /* Don't miss readonly hidden in the superblock flags */
3269 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3270 mnt_flags |= MNT_LOCK_READONLY;
3272 /* Verify the mount flags are equal to or more permissive
3273 * than the proposed new mount.
3275 if ((mnt_flags & MNT_LOCK_READONLY) &&
3276 !(new_flags & MNT_READONLY))
3278 if ((mnt_flags & MNT_LOCK_NODEV) &&
3279 !(new_flags & MNT_NODEV))
3281 if ((mnt_flags & MNT_LOCK_NOSUID) &&
3282 !(new_flags & MNT_NOSUID))
3284 if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3285 !(new_flags & MNT_NOEXEC))
3287 if ((mnt_flags & MNT_LOCK_ATIME) &&
3288 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3291 /* This mount is not fully visible if there are any
3292 * locked child mounts that cover anything except for
3293 * empty directories.
3295 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3296 struct inode *inode = child->mnt_mountpoint->d_inode;
3297 /* Only worry about locked mounts */
3298 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3300 /* Is the directory permanetly empty? */
3301 if (!is_empty_dir_inode(inode))
3304 /* Preserve the locked attributes */
3305 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3315 up_read(&namespace_sem);
3319 static struct ns_common *mntns_get(struct task_struct *task)
3321 struct ns_common *ns = NULL;
3322 struct nsproxy *nsproxy;
3325 nsproxy = task->nsproxy;
3327 ns = &nsproxy->mnt_ns->ns;
3328 get_mnt_ns(to_mnt_ns(ns));
3335 static void mntns_put(struct ns_common *ns)
3337 put_mnt_ns(to_mnt_ns(ns));
3340 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3342 struct fs_struct *fs = current->fs;
3343 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3346 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3347 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3348 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3355 put_mnt_ns(nsproxy->mnt_ns);
3356 nsproxy->mnt_ns = mnt_ns;
3359 root.mnt = &mnt_ns->root->mnt;
3360 root.dentry = mnt_ns->root->mnt.mnt_root;
3362 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3365 /* Update the pwd and root */
3366 set_fs_pwd(fs, &root);
3367 set_fs_root(fs, &root);
3373 const struct proc_ns_operations mntns_operations = {
3375 .type = CLONE_NEWNS,
3378 .install = mntns_install,