4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
40 mhash_entries = simple_strtoul(str, &str, 0);
43 __setup("mhash_entries=", set_mhash_entries);
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
50 mphash_entries = simple_strtoul(str, &str, 0);
53 __setup("mphash_entries=", set_mphash_entries);
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount *mnt)
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
117 static void mnt_free_id(struct mount *mnt)
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
124 spin_unlock(&mnt_id_lock);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount *mnt)
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start = mnt->mnt_group_id + 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount *mnt)
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount *mnt, int n)
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount *mnt)
180 unsigned int count = 0;
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
189 return mnt->mnt_count;
193 static void drop_mountpoint(struct fs_pin *p)
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
201 static struct mount *alloc_vfsmnt(const char *name)
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 err = mnt_alloc_id(mnt);
212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 if (!mnt->mnt_devname)
218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
220 goto out_free_devname;
222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225 mnt->mnt_writers = 0;
228 INIT_HLIST_NODE(&mnt->mnt_hash);
229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
246 kfree_const(mnt->mnt_devname);
251 kmem_cache_free(mnt_cache, mnt);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount *mnt)
276 if (mnt->mnt_flags & MNT_READONLY)
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
284 static inline void mnt_inc_writers(struct mount *mnt)
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
293 static inline void mnt_dec_writers(struct mount *mnt)
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
302 static unsigned int mnt_get_writers(struct mount *mnt)
305 unsigned int count = 0;
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
314 return mnt->mnt_writers;
318 static int mnt_is_readonly(struct vfsmount *mnt)
320 if (mnt->mnt_sb->s_readonly_remount)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount *m)
345 struct mount *mnt = real_mount(m);
349 mnt_inc_writers(mnt);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount *m)
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
389 sb_end_write(m->mnt_sb);
392 EXPORT_SYMBOL_GPL(mnt_want_write);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount *mnt)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
412 mnt_inc_writers(real_mount(mnt));
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file *file)
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
430 return mnt_clone_write(file->f_path.mnt);
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 int mnt_want_write_file(struct file *file)
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
447 sb_end_write(file->f_path.mnt->mnt_sb);
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
460 void __mnt_drop_write(struct vfsmount *mnt)
463 mnt_dec_writers(real_mount(mnt));
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
475 void mnt_drop_write(struct vfsmount *mnt)
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
482 void __mnt_drop_write_file(struct file *file)
484 __mnt_drop_write(file->f_path.mnt);
487 void mnt_drop_write_file(struct file *file)
489 mnt_drop_write(file->f_path.mnt);
491 EXPORT_SYMBOL(mnt_drop_write_file);
493 static int mnt_make_readonly(struct mount *mnt)
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
521 if (mnt_get_writers(mnt) > 0)
524 mnt->mnt.mnt_flags |= MNT_READONLY;
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
535 static void __mnt_unmake_readonly(struct mount *mnt)
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
542 int sb_prepare_remount_readonly(struct super_block *sb)
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
556 if (mnt_get_writers(mnt) > 0) {
562 if (!err && atomic_long_read(&sb->s_remove_count))
566 sb->s_readonly_remount = 1;
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
578 static void free_vfsmnt(struct mount *mnt)
580 kfree_const(mnt->mnt_devname);
582 free_percpu(mnt->mnt_pcp);
584 kmem_cache_free(mnt_cache, mnt);
587 static void delayed_free_vfsmnt(struct rcu_head *head)
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
592 /* call under rcu_read_lock */
593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
596 if (read_seqretry(&mount_lock, seq))
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
615 * find the first mount at @dentry on vfsmount @mnt.
616 * call under rcu_read_lock()
618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
620 struct hlist_head *head = m_hash(mnt, dentry);
623 hlist_for_each_entry_rcu(p, head, mnt_hash)
624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
630 * find the last mount at @dentry on vfsmount @mnt.
631 * mount_lock must be held.
633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
635 struct mount *p, *res;
636 res = p = __lookup_mnt(mnt, dentry);
639 hlist_for_each_entry_continue(p, mnt_hash) {
640 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
649 * lookup_mnt - Return the first child mount mounted at path
651 * "First" means first mounted chronologically. If you create the
654 * mount /dev/sda1 /mnt
655 * mount /dev/sda2 /mnt
656 * mount /dev/sda3 /mnt
658 * Then lookup_mnt() on the base /mnt dentry in the root mount will
659 * return successively the root dentry and vfsmount of /dev/sda1, then
660 * /dev/sda2, then /dev/sda3, then NULL.
662 * lookup_mnt takes a reference to the found vfsmount.
664 struct vfsmount *lookup_mnt(struct path *path)
666 struct mount *child_mnt;
672 seq = read_seqbegin(&mount_lock);
673 child_mnt = __lookup_mnt(path->mnt, path->dentry);
674 m = child_mnt ? &child_mnt->mnt : NULL;
675 } while (!legitimize_mnt(m, seq));
681 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
682 * current mount namespace.
684 * The common case is dentries are not mountpoints at all and that
685 * test is handled inline. For the slow case when we are actually
686 * dealing with a mountpoint of some kind, walk through all of the
687 * mounts in the current mount namespace and test to see if the dentry
690 * The mount_hashtable is not usable in the context because we
691 * need to identify all mounts that may be in the current mount
692 * namespace not just a mount that happens to have some specified
695 bool __is_local_mountpoint(struct dentry *dentry)
697 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
699 bool is_covered = false;
701 if (!d_mountpoint(dentry))
704 down_read(&namespace_sem);
705 list_for_each_entry(mnt, &ns->list, mnt_list) {
706 is_covered = (mnt->mnt_mountpoint == dentry);
710 up_read(&namespace_sem);
715 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
717 struct hlist_head *chain = mp_hash(dentry);
718 struct mountpoint *mp;
720 hlist_for_each_entry(mp, chain, m_hash) {
721 if (mp->m_dentry == dentry) {
722 /* might be worth a WARN_ON() */
723 if (d_unlinked(dentry))
724 return ERR_PTR(-ENOENT);
732 static struct mountpoint *new_mountpoint(struct dentry *dentry)
734 struct hlist_head *chain = mp_hash(dentry);
735 struct mountpoint *mp;
738 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
740 return ERR_PTR(-ENOMEM);
742 ret = d_set_mounted(dentry);
748 mp->m_dentry = dentry;
750 hlist_add_head(&mp->m_hash, chain);
751 INIT_HLIST_HEAD(&mp->m_list);
755 static void put_mountpoint(struct mountpoint *mp)
757 if (!--mp->m_count) {
758 struct dentry *dentry = mp->m_dentry;
759 BUG_ON(!hlist_empty(&mp->m_list));
760 spin_lock(&dentry->d_lock);
761 dentry->d_flags &= ~DCACHE_MOUNTED;
762 spin_unlock(&dentry->d_lock);
763 hlist_del(&mp->m_hash);
768 static inline int check_mnt(struct mount *mnt)
770 return mnt->mnt_ns == current->nsproxy->mnt_ns;
774 * vfsmount lock must be held for write
776 static void touch_mnt_namespace(struct mnt_namespace *ns)
780 wake_up_interruptible(&ns->poll);
785 * vfsmount lock must be held for write
787 static void __touch_mnt_namespace(struct mnt_namespace *ns)
789 if (ns && ns->event != event) {
791 wake_up_interruptible(&ns->poll);
796 * vfsmount lock must be held for write
798 static void detach_mnt(struct mount *mnt, struct path *old_path)
800 old_path->dentry = mnt->mnt_mountpoint;
801 old_path->mnt = &mnt->mnt_parent->mnt;
802 mnt->mnt_parent = mnt;
803 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
804 list_del_init(&mnt->mnt_child);
805 hlist_del_init_rcu(&mnt->mnt_hash);
806 hlist_del_init(&mnt->mnt_mp_list);
807 put_mountpoint(mnt->mnt_mp);
812 * vfsmount lock must be held for write
814 void mnt_set_mountpoint(struct mount *mnt,
815 struct mountpoint *mp,
816 struct mount *child_mnt)
819 mnt_add_count(mnt, 1); /* essentially, that's mntget */
820 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
821 child_mnt->mnt_parent = mnt;
822 child_mnt->mnt_mp = mp;
823 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
827 * vfsmount lock must be held for write
829 static void attach_mnt(struct mount *mnt,
830 struct mount *parent,
831 struct mountpoint *mp)
833 mnt_set_mountpoint(parent, mp, mnt);
834 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
835 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
838 static void attach_shadowed(struct mount *mnt,
839 struct mount *parent,
840 struct mount *shadows)
843 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
844 list_add(&mnt->mnt_child, &shadows->mnt_child);
846 hlist_add_head_rcu(&mnt->mnt_hash,
847 m_hash(&parent->mnt, mnt->mnt_mountpoint));
848 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
853 * vfsmount lock must be held for write
855 static void commit_tree(struct mount *mnt, struct mount *shadows)
857 struct mount *parent = mnt->mnt_parent;
860 struct mnt_namespace *n = parent->mnt_ns;
862 BUG_ON(parent == mnt);
864 list_add_tail(&head, &mnt->mnt_list);
865 list_for_each_entry(m, &head, mnt_list)
868 list_splice(&head, n->list.prev);
870 attach_shadowed(mnt, parent, shadows);
871 touch_mnt_namespace(n);
874 static struct mount *next_mnt(struct mount *p, struct mount *root)
876 struct list_head *next = p->mnt_mounts.next;
877 if (next == &p->mnt_mounts) {
881 next = p->mnt_child.next;
882 if (next != &p->mnt_parent->mnt_mounts)
887 return list_entry(next, struct mount, mnt_child);
890 static struct mount *skip_mnt_tree(struct mount *p)
892 struct list_head *prev = p->mnt_mounts.prev;
893 while (prev != &p->mnt_mounts) {
894 p = list_entry(prev, struct mount, mnt_child);
895 prev = p->mnt_mounts.prev;
901 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
907 return ERR_PTR(-ENODEV);
909 mnt = alloc_vfsmnt(name);
911 return ERR_PTR(-ENOMEM);
913 if (flags & MS_KERNMOUNT)
914 mnt->mnt.mnt_flags = MNT_INTERNAL;
916 root = mount_fs(type, flags, name, data);
920 return ERR_CAST(root);
923 mnt->mnt.mnt_root = root;
924 mnt->mnt.mnt_sb = root->d_sb;
925 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
926 mnt->mnt_parent = mnt;
928 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
932 EXPORT_SYMBOL_GPL(vfs_kern_mount);
934 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
937 struct super_block *sb = old->mnt.mnt_sb;
941 mnt = alloc_vfsmnt(old->mnt_devname);
943 return ERR_PTR(-ENOMEM);
945 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
946 mnt->mnt_group_id = 0; /* not a peer of original */
948 mnt->mnt_group_id = old->mnt_group_id;
950 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
951 err = mnt_alloc_group_id(mnt);
956 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
957 /* Don't allow unprivileged users to change mount flags */
958 if (flag & CL_UNPRIVILEGED) {
959 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
961 if (mnt->mnt.mnt_flags & MNT_READONLY)
962 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
964 if (mnt->mnt.mnt_flags & MNT_NODEV)
965 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
967 if (mnt->mnt.mnt_flags & MNT_NOSUID)
968 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
970 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
971 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
974 /* Don't allow unprivileged users to reveal what is under a mount */
975 if ((flag & CL_UNPRIVILEGED) &&
976 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
977 mnt->mnt.mnt_flags |= MNT_LOCKED;
979 atomic_inc(&sb->s_active);
980 mnt->mnt.mnt_sb = sb;
981 mnt->mnt.mnt_root = dget(root);
982 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
983 mnt->mnt_parent = mnt;
985 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
988 if ((flag & CL_SLAVE) ||
989 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
990 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
991 mnt->mnt_master = old;
992 CLEAR_MNT_SHARED(mnt);
993 } else if (!(flag & CL_PRIVATE)) {
994 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
995 list_add(&mnt->mnt_share, &old->mnt_share);
996 if (IS_MNT_SLAVE(old))
997 list_add(&mnt->mnt_slave, &old->mnt_slave);
998 mnt->mnt_master = old->mnt_master;
1000 if (flag & CL_MAKE_SHARED)
1001 set_mnt_shared(mnt);
1003 /* stick the duplicate mount on the same expiry list
1004 * as the original if that was on one */
1005 if (flag & CL_EXPIRE) {
1006 if (!list_empty(&old->mnt_expire))
1007 list_add(&mnt->mnt_expire, &old->mnt_expire);
1015 return ERR_PTR(err);
1018 static void cleanup_mnt(struct mount *mnt)
1021 * This probably indicates that somebody messed
1022 * up a mnt_want/drop_write() pair. If this
1023 * happens, the filesystem was probably unable
1024 * to make r/w->r/o transitions.
1027 * The locking used to deal with mnt_count decrement provides barriers,
1028 * so mnt_get_writers() below is safe.
1030 WARN_ON(mnt_get_writers(mnt));
1031 if (unlikely(mnt->mnt_pins.first))
1033 fsnotify_vfsmount_delete(&mnt->mnt);
1034 dput(mnt->mnt.mnt_root);
1035 deactivate_super(mnt->mnt.mnt_sb);
1037 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1040 static void __cleanup_mnt(struct rcu_head *head)
1042 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1045 static LLIST_HEAD(delayed_mntput_list);
1046 static void delayed_mntput(struct work_struct *unused)
1048 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1049 struct llist_node *next;
1051 for (; node; node = next) {
1052 next = llist_next(node);
1053 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1056 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1058 static void mntput_no_expire(struct mount *mnt)
1061 mnt_add_count(mnt, -1);
1062 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1067 if (mnt_get_count(mnt)) {
1069 unlock_mount_hash();
1072 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1074 unlock_mount_hash();
1077 mnt->mnt.mnt_flags |= MNT_DOOMED;
1080 list_del(&mnt->mnt_instance);
1081 unlock_mount_hash();
1083 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1084 struct task_struct *task = current;
1085 if (likely(!(task->flags & PF_KTHREAD))) {
1086 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1087 if (!task_work_add(task, &mnt->mnt_rcu, true))
1090 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1091 schedule_delayed_work(&delayed_mntput_work, 1);
1097 void mntput(struct vfsmount *mnt)
1100 struct mount *m = real_mount(mnt);
1101 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1102 if (unlikely(m->mnt_expiry_mark))
1103 m->mnt_expiry_mark = 0;
1104 mntput_no_expire(m);
1107 EXPORT_SYMBOL(mntput);
1109 struct vfsmount *mntget(struct vfsmount *mnt)
1112 mnt_add_count(real_mount(mnt), 1);
1115 EXPORT_SYMBOL(mntget);
1117 struct vfsmount *mnt_clone_internal(struct path *path)
1120 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1123 p->mnt.mnt_flags |= MNT_INTERNAL;
1127 static inline void mangle(struct seq_file *m, const char *s)
1129 seq_escape(m, s, " \t\n\\");
1133 * Simple .show_options callback for filesystems which don't want to
1134 * implement more complex mount option showing.
1136 * See also save_mount_options().
1138 int generic_show_options(struct seq_file *m, struct dentry *root)
1140 const char *options;
1143 options = rcu_dereference(root->d_sb->s_options);
1145 if (options != NULL && options[0]) {
1153 EXPORT_SYMBOL(generic_show_options);
1156 * If filesystem uses generic_show_options(), this function should be
1157 * called from the fill_super() callback.
1159 * The .remount_fs callback usually needs to be handled in a special
1160 * way, to make sure, that previous options are not overwritten if the
1163 * Also note, that if the filesystem's .remount_fs function doesn't
1164 * reset all options to their default value, but changes only newly
1165 * given options, then the displayed options will not reflect reality
1168 void save_mount_options(struct super_block *sb, char *options)
1170 BUG_ON(sb->s_options);
1171 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1173 EXPORT_SYMBOL(save_mount_options);
1175 void replace_mount_options(struct super_block *sb, char *options)
1177 char *old = sb->s_options;
1178 rcu_assign_pointer(sb->s_options, options);
1184 EXPORT_SYMBOL(replace_mount_options);
1186 #ifdef CONFIG_PROC_FS
1187 /* iterator; we want it to have access to namespace_sem, thus here... */
1188 static void *m_start(struct seq_file *m, loff_t *pos)
1190 struct proc_mounts *p = proc_mounts(m);
1192 down_read(&namespace_sem);
1193 if (p->cached_event == p->ns->event) {
1194 void *v = p->cached_mount;
1195 if (*pos == p->cached_index)
1197 if (*pos == p->cached_index + 1) {
1198 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1199 return p->cached_mount = v;
1203 p->cached_event = p->ns->event;
1204 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1205 p->cached_index = *pos;
1206 return p->cached_mount;
1209 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1211 struct proc_mounts *p = proc_mounts(m);
1213 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1214 p->cached_index = *pos;
1215 return p->cached_mount;
1218 static void m_stop(struct seq_file *m, void *v)
1220 up_read(&namespace_sem);
1223 static int m_show(struct seq_file *m, void *v)
1225 struct proc_mounts *p = proc_mounts(m);
1226 struct mount *r = list_entry(v, struct mount, mnt_list);
1227 return p->show(m, &r->mnt);
1230 const struct seq_operations mounts_op = {
1236 #endif /* CONFIG_PROC_FS */
1239 * may_umount_tree - check if a mount tree is busy
1240 * @mnt: root of mount tree
1242 * This is called to check if a tree of mounts has any
1243 * open files, pwds, chroots or sub mounts that are
1246 int may_umount_tree(struct vfsmount *m)
1248 struct mount *mnt = real_mount(m);
1249 int actual_refs = 0;
1250 int minimum_refs = 0;
1254 /* write lock needed for mnt_get_count */
1256 for (p = mnt; p; p = next_mnt(p, mnt)) {
1257 actual_refs += mnt_get_count(p);
1260 unlock_mount_hash();
1262 if (actual_refs > minimum_refs)
1268 EXPORT_SYMBOL(may_umount_tree);
1271 * may_umount - check if a mount point is busy
1272 * @mnt: root of mount
1274 * This is called to check if a mount point has any
1275 * open files, pwds, chroots or sub mounts. If the
1276 * mount has sub mounts this will return busy
1277 * regardless of whether the sub mounts are busy.
1279 * Doesn't take quota and stuff into account. IOW, in some cases it will
1280 * give false negatives. The main reason why it's here is that we need
1281 * a non-destructive way to look for easily umountable filesystems.
1283 int may_umount(struct vfsmount *mnt)
1286 down_read(&namespace_sem);
1288 if (propagate_mount_busy(real_mount(mnt), 2))
1290 unlock_mount_hash();
1291 up_read(&namespace_sem);
1295 EXPORT_SYMBOL(may_umount);
1297 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1299 static void namespace_unlock(void)
1301 struct hlist_head head;
1303 hlist_move_list(&unmounted, &head);
1305 up_write(&namespace_sem);
1307 if (likely(hlist_empty(&head)))
1312 group_pin_kill(&head);
1315 static inline void namespace_lock(void)
1317 down_write(&namespace_sem);
1320 enum umount_tree_flags {
1322 UMOUNT_PROPAGATE = 2,
1325 * mount_lock must be held
1326 * namespace_sem must be held for write
1328 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1330 LIST_HEAD(tmp_list);
1333 /* Gather the mounts to umount */
1334 for (p = mnt; p; p = next_mnt(p, mnt))
1335 list_move(&p->mnt_list, &tmp_list);
1337 /* Hide the mounts from lookup_mnt and mnt_mounts */
1338 list_for_each_entry(p, &tmp_list, mnt_list) {
1339 hlist_del_init_rcu(&p->mnt_hash);
1340 list_del_init(&p->mnt_child);
1343 /* Add propogated mounts to the tmp_list */
1344 if (how & UMOUNT_PROPAGATE)
1345 propagate_umount(&tmp_list);
1347 while (!list_empty(&tmp_list)) {
1348 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1349 list_del_init(&p->mnt_expire);
1350 list_del_init(&p->mnt_list);
1351 __touch_mnt_namespace(p->mnt_ns);
1353 if (how & UMOUNT_SYNC)
1354 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1356 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, &unmounted);
1357 if (mnt_has_parent(p)) {
1358 hlist_del_init(&p->mnt_mp_list);
1359 put_mountpoint(p->mnt_mp);
1360 mnt_add_count(p->mnt_parent, -1);
1361 /* old mountpoint will be dropped when we can do that */
1362 p->mnt_ex_mountpoint = p->mnt_mountpoint;
1363 p->mnt_mountpoint = p->mnt.mnt_root;
1367 change_mnt_propagation(p, MS_PRIVATE);
1371 static void shrink_submounts(struct mount *mnt);
1373 static int do_umount(struct mount *mnt, int flags)
1375 struct super_block *sb = mnt->mnt.mnt_sb;
1378 retval = security_sb_umount(&mnt->mnt, flags);
1383 * Allow userspace to request a mountpoint be expired rather than
1384 * unmounting unconditionally. Unmount only happens if:
1385 * (1) the mark is already set (the mark is cleared by mntput())
1386 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1388 if (flags & MNT_EXPIRE) {
1389 if (&mnt->mnt == current->fs->root.mnt ||
1390 flags & (MNT_FORCE | MNT_DETACH))
1394 * probably don't strictly need the lock here if we examined
1395 * all race cases, but it's a slowpath.
1398 if (mnt_get_count(mnt) != 2) {
1399 unlock_mount_hash();
1402 unlock_mount_hash();
1404 if (!xchg(&mnt->mnt_expiry_mark, 1))
1409 * If we may have to abort operations to get out of this
1410 * mount, and they will themselves hold resources we must
1411 * allow the fs to do things. In the Unix tradition of
1412 * 'Gee thats tricky lets do it in userspace' the umount_begin
1413 * might fail to complete on the first run through as other tasks
1414 * must return, and the like. Thats for the mount program to worry
1415 * about for the moment.
1418 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1419 sb->s_op->umount_begin(sb);
1423 * No sense to grab the lock for this test, but test itself looks
1424 * somewhat bogus. Suggestions for better replacement?
1425 * Ho-hum... In principle, we might treat that as umount + switch
1426 * to rootfs. GC would eventually take care of the old vfsmount.
1427 * Actually it makes sense, especially if rootfs would contain a
1428 * /reboot - static binary that would close all descriptors and
1429 * call reboot(9). Then init(8) could umount root and exec /reboot.
1431 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1433 * Special case for "unmounting" root ...
1434 * we just try to remount it readonly.
1436 if (!capable(CAP_SYS_ADMIN))
1438 down_write(&sb->s_umount);
1439 if (!(sb->s_flags & MS_RDONLY))
1440 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1441 up_write(&sb->s_umount);
1449 if (flags & MNT_DETACH) {
1450 if (!list_empty(&mnt->mnt_list))
1451 umount_tree(mnt, UMOUNT_PROPAGATE);
1454 shrink_submounts(mnt);
1456 if (!propagate_mount_busy(mnt, 2)) {
1457 if (!list_empty(&mnt->mnt_list))
1458 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1462 unlock_mount_hash();
1468 * __detach_mounts - lazily unmount all mounts on the specified dentry
1470 * During unlink, rmdir, and d_drop it is possible to loose the path
1471 * to an existing mountpoint, and wind up leaking the mount.
1472 * detach_mounts allows lazily unmounting those mounts instead of
1475 * The caller may hold dentry->d_inode->i_mutex.
1477 void __detach_mounts(struct dentry *dentry)
1479 struct mountpoint *mp;
1483 mp = lookup_mountpoint(dentry);
1488 while (!hlist_empty(&mp->m_list)) {
1489 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1490 umount_tree(mnt, 0);
1492 unlock_mount_hash();
1499 * Is the caller allowed to modify his namespace?
1501 static inline bool may_mount(void)
1503 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1507 * Now umount can handle mount points as well as block devices.
1508 * This is important for filesystems which use unnamed block devices.
1510 * We now support a flag for forced unmount like the other 'big iron'
1511 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1514 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1519 int lookup_flags = 0;
1521 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1527 if (!(flags & UMOUNT_NOFOLLOW))
1528 lookup_flags |= LOOKUP_FOLLOW;
1530 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1533 mnt = real_mount(path.mnt);
1535 if (path.dentry != path.mnt->mnt_root)
1537 if (!check_mnt(mnt))
1539 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1542 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1545 retval = do_umount(mnt, flags);
1547 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1549 mntput_no_expire(mnt);
1554 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1557 * The 2.0 compatible umount. No flags.
1559 SYSCALL_DEFINE1(oldumount, char __user *, name)
1561 return sys_umount(name, 0);
1566 static bool is_mnt_ns_file(struct dentry *dentry)
1568 /* Is this a proxy for a mount namespace? */
1569 return dentry->d_op == &ns_dentry_operations &&
1570 dentry->d_fsdata == &mntns_operations;
1573 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1575 return container_of(ns, struct mnt_namespace, ns);
1578 static bool mnt_ns_loop(struct dentry *dentry)
1580 /* Could bind mounting the mount namespace inode cause a
1581 * mount namespace loop?
1583 struct mnt_namespace *mnt_ns;
1584 if (!is_mnt_ns_file(dentry))
1587 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1588 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1591 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1594 struct mount *res, *p, *q, *r, *parent;
1596 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1597 return ERR_PTR(-EINVAL);
1599 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1600 return ERR_PTR(-EINVAL);
1602 res = q = clone_mnt(mnt, dentry, flag);
1606 q->mnt_mountpoint = mnt->mnt_mountpoint;
1609 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1611 if (!is_subdir(r->mnt_mountpoint, dentry))
1614 for (s = r; s; s = next_mnt(s, r)) {
1615 struct mount *t = NULL;
1616 if (!(flag & CL_COPY_UNBINDABLE) &&
1617 IS_MNT_UNBINDABLE(s)) {
1618 s = skip_mnt_tree(s);
1621 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1622 is_mnt_ns_file(s->mnt.mnt_root)) {
1623 s = skip_mnt_tree(s);
1626 while (p != s->mnt_parent) {
1632 q = clone_mnt(p, p->mnt.mnt_root, flag);
1636 list_add_tail(&q->mnt_list, &res->mnt_list);
1637 mnt_set_mountpoint(parent, p->mnt_mp, q);
1638 if (!list_empty(&parent->mnt_mounts)) {
1639 t = list_last_entry(&parent->mnt_mounts,
1640 struct mount, mnt_child);
1641 if (t->mnt_mp != p->mnt_mp)
1644 attach_shadowed(q, parent, t);
1645 unlock_mount_hash();
1652 umount_tree(res, UMOUNT_SYNC);
1653 unlock_mount_hash();
1658 /* Caller should check returned pointer for errors */
1660 struct vfsmount *collect_mounts(struct path *path)
1664 tree = copy_tree(real_mount(path->mnt), path->dentry,
1665 CL_COPY_ALL | CL_PRIVATE);
1668 return ERR_CAST(tree);
1672 void drop_collected_mounts(struct vfsmount *mnt)
1676 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1677 unlock_mount_hash();
1682 * clone_private_mount - create a private clone of a path
1684 * This creates a new vfsmount, which will be the clone of @path. The new will
1685 * not be attached anywhere in the namespace and will be private (i.e. changes
1686 * to the originating mount won't be propagated into this).
1688 * Release with mntput().
1690 struct vfsmount *clone_private_mount(struct path *path)
1692 struct mount *old_mnt = real_mount(path->mnt);
1693 struct mount *new_mnt;
1695 if (IS_MNT_UNBINDABLE(old_mnt))
1696 return ERR_PTR(-EINVAL);
1698 down_read(&namespace_sem);
1699 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1700 up_read(&namespace_sem);
1701 if (IS_ERR(new_mnt))
1702 return ERR_CAST(new_mnt);
1704 return &new_mnt->mnt;
1706 EXPORT_SYMBOL_GPL(clone_private_mount);
1708 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1709 struct vfsmount *root)
1712 int res = f(root, arg);
1715 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1716 res = f(&mnt->mnt, arg);
1723 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1727 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1728 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1729 mnt_release_group_id(p);
1733 static int invent_group_ids(struct mount *mnt, bool recurse)
1737 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1738 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1739 int err = mnt_alloc_group_id(p);
1741 cleanup_group_ids(mnt, p);
1751 * @source_mnt : mount tree to be attached
1752 * @nd : place the mount tree @source_mnt is attached
1753 * @parent_nd : if non-null, detach the source_mnt from its parent and
1754 * store the parent mount and mountpoint dentry.
1755 * (done when source_mnt is moved)
1757 * NOTE: in the table below explains the semantics when a source mount
1758 * of a given type is attached to a destination mount of a given type.
1759 * ---------------------------------------------------------------------------
1760 * | BIND MOUNT OPERATION |
1761 * |**************************************************************************
1762 * | source-->| shared | private | slave | unbindable |
1766 * |**************************************************************************
1767 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1769 * |non-shared| shared (+) | private | slave (*) | invalid |
1770 * ***************************************************************************
1771 * A bind operation clones the source mount and mounts the clone on the
1772 * destination mount.
1774 * (++) the cloned mount is propagated to all the mounts in the propagation
1775 * tree of the destination mount and the cloned mount is added to
1776 * the peer group of the source mount.
1777 * (+) the cloned mount is created under the destination mount and is marked
1778 * as shared. The cloned mount is added to the peer group of the source
1780 * (+++) the mount is propagated to all the mounts in the propagation tree
1781 * of the destination mount and the cloned mount is made slave
1782 * of the same master as that of the source mount. The cloned mount
1783 * is marked as 'shared and slave'.
1784 * (*) the cloned mount is made a slave of the same master as that of the
1787 * ---------------------------------------------------------------------------
1788 * | MOVE MOUNT OPERATION |
1789 * |**************************************************************************
1790 * | source-->| shared | private | slave | unbindable |
1794 * |**************************************************************************
1795 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1797 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1798 * ***************************************************************************
1800 * (+) the mount is moved to the destination. And is then propagated to
1801 * all the mounts in the propagation tree of the destination mount.
1802 * (+*) the mount is moved to the destination.
1803 * (+++) the mount is moved to the destination and is then propagated to
1804 * all the mounts belonging to the destination mount's propagation tree.
1805 * the mount is marked as 'shared and slave'.
1806 * (*) the mount continues to be a slave at the new location.
1808 * if the source mount is a tree, the operations explained above is
1809 * applied to each mount in the tree.
1810 * Must be called without spinlocks held, since this function can sleep
1813 static int attach_recursive_mnt(struct mount *source_mnt,
1814 struct mount *dest_mnt,
1815 struct mountpoint *dest_mp,
1816 struct path *parent_path)
1818 HLIST_HEAD(tree_list);
1819 struct mount *child, *p;
1820 struct hlist_node *n;
1823 if (IS_MNT_SHARED(dest_mnt)) {
1824 err = invent_group_ids(source_mnt, true);
1827 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1830 goto out_cleanup_ids;
1831 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1837 detach_mnt(source_mnt, parent_path);
1838 attach_mnt(source_mnt, dest_mnt, dest_mp);
1839 touch_mnt_namespace(source_mnt->mnt_ns);
1841 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1842 commit_tree(source_mnt, NULL);
1845 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1847 hlist_del_init(&child->mnt_hash);
1848 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1849 child->mnt_mountpoint);
1850 commit_tree(child, q);
1852 unlock_mount_hash();
1857 while (!hlist_empty(&tree_list)) {
1858 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1859 umount_tree(child, UMOUNT_SYNC);
1861 unlock_mount_hash();
1862 cleanup_group_ids(source_mnt, NULL);
1867 static struct mountpoint *lock_mount(struct path *path)
1869 struct vfsmount *mnt;
1870 struct dentry *dentry = path->dentry;
1872 mutex_lock(&dentry->d_inode->i_mutex);
1873 if (unlikely(cant_mount(dentry))) {
1874 mutex_unlock(&dentry->d_inode->i_mutex);
1875 return ERR_PTR(-ENOENT);
1878 mnt = lookup_mnt(path);
1880 struct mountpoint *mp = lookup_mountpoint(dentry);
1882 mp = new_mountpoint(dentry);
1885 mutex_unlock(&dentry->d_inode->i_mutex);
1891 mutex_unlock(&path->dentry->d_inode->i_mutex);
1894 dentry = path->dentry = dget(mnt->mnt_root);
1898 static void unlock_mount(struct mountpoint *where)
1900 struct dentry *dentry = where->m_dentry;
1901 put_mountpoint(where);
1903 mutex_unlock(&dentry->d_inode->i_mutex);
1906 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1908 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1911 if (d_is_dir(mp->m_dentry) !=
1912 d_is_dir(mnt->mnt.mnt_root))
1915 return attach_recursive_mnt(mnt, p, mp, NULL);
1919 * Sanity check the flags to change_mnt_propagation.
1922 static int flags_to_propagation_type(int flags)
1924 int type = flags & ~(MS_REC | MS_SILENT);
1926 /* Fail if any non-propagation flags are set */
1927 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1929 /* Only one propagation flag should be set */
1930 if (!is_power_of_2(type))
1936 * recursively change the type of the mountpoint.
1938 static int do_change_type(struct path *path, int flag)
1941 struct mount *mnt = real_mount(path->mnt);
1942 int recurse = flag & MS_REC;
1946 if (path->dentry != path->mnt->mnt_root)
1949 type = flags_to_propagation_type(flag);
1954 if (type == MS_SHARED) {
1955 err = invent_group_ids(mnt, recurse);
1961 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1962 change_mnt_propagation(m, type);
1963 unlock_mount_hash();
1970 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1972 struct mount *child;
1973 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1974 if (!is_subdir(child->mnt_mountpoint, dentry))
1977 if (child->mnt.mnt_flags & MNT_LOCKED)
1984 * do loopback mount.
1986 static int do_loopback(struct path *path, const char *old_name,
1989 struct path old_path;
1990 struct mount *mnt = NULL, *old, *parent;
1991 struct mountpoint *mp;
1993 if (!old_name || !*old_name)
1995 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2000 if (mnt_ns_loop(old_path.dentry))
2003 mp = lock_mount(path);
2008 old = real_mount(old_path.mnt);
2009 parent = real_mount(path->mnt);
2012 if (IS_MNT_UNBINDABLE(old))
2015 if (!check_mnt(parent))
2018 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2021 if (!recurse && has_locked_children(old, old_path.dentry))
2025 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2027 mnt = clone_mnt(old, old_path.dentry, 0);
2034 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2036 err = graft_tree(mnt, parent, mp);
2039 umount_tree(mnt, UMOUNT_SYNC);
2040 unlock_mount_hash();
2045 path_put(&old_path);
2049 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2052 int readonly_request = 0;
2054 if (ms_flags & MS_RDONLY)
2055 readonly_request = 1;
2056 if (readonly_request == __mnt_is_readonly(mnt))
2059 if (readonly_request)
2060 error = mnt_make_readonly(real_mount(mnt));
2062 __mnt_unmake_readonly(real_mount(mnt));
2067 * change filesystem flags. dir should be a physical root of filesystem.
2068 * If you've mounted a non-root directory somewhere and want to do remount
2069 * on it - tough luck.
2071 static int do_remount(struct path *path, int flags, int mnt_flags,
2075 struct super_block *sb = path->mnt->mnt_sb;
2076 struct mount *mnt = real_mount(path->mnt);
2078 if (!check_mnt(mnt))
2081 if (path->dentry != path->mnt->mnt_root)
2084 /* Don't allow changing of locked mnt flags.
2086 * No locks need to be held here while testing the various
2087 * MNT_LOCK flags because those flags can never be cleared
2088 * once they are set.
2090 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2091 !(mnt_flags & MNT_READONLY)) {
2094 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2095 !(mnt_flags & MNT_NODEV)) {
2096 /* Was the nodev implicitly added in mount? */
2097 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2098 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2099 mnt_flags |= MNT_NODEV;
2104 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2105 !(mnt_flags & MNT_NOSUID)) {
2108 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2109 !(mnt_flags & MNT_NOEXEC)) {
2112 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2113 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2117 err = security_sb_remount(sb, data);
2121 down_write(&sb->s_umount);
2122 if (flags & MS_BIND)
2123 err = change_mount_flags(path->mnt, flags);
2124 else if (!capable(CAP_SYS_ADMIN))
2127 err = do_remount_sb(sb, flags, data, 0);
2130 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2131 mnt->mnt.mnt_flags = mnt_flags;
2132 touch_mnt_namespace(mnt->mnt_ns);
2133 unlock_mount_hash();
2135 up_write(&sb->s_umount);
2139 static inline int tree_contains_unbindable(struct mount *mnt)
2142 for (p = mnt; p; p = next_mnt(p, mnt)) {
2143 if (IS_MNT_UNBINDABLE(p))
2149 static int do_move_mount(struct path *path, const char *old_name)
2151 struct path old_path, parent_path;
2154 struct mountpoint *mp;
2156 if (!old_name || !*old_name)
2158 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2162 mp = lock_mount(path);
2167 old = real_mount(old_path.mnt);
2168 p = real_mount(path->mnt);
2171 if (!check_mnt(p) || !check_mnt(old))
2174 if (old->mnt.mnt_flags & MNT_LOCKED)
2178 if (old_path.dentry != old_path.mnt->mnt_root)
2181 if (!mnt_has_parent(old))
2184 if (d_is_dir(path->dentry) !=
2185 d_is_dir(old_path.dentry))
2188 * Don't move a mount residing in a shared parent.
2190 if (IS_MNT_SHARED(old->mnt_parent))
2193 * Don't move a mount tree containing unbindable mounts to a destination
2194 * mount which is shared.
2196 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2199 for (; mnt_has_parent(p); p = p->mnt_parent)
2203 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2207 /* if the mount is moved, it should no longer be expire
2209 list_del_init(&old->mnt_expire);
2214 path_put(&parent_path);
2215 path_put(&old_path);
2219 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2222 const char *subtype = strchr(fstype, '.');
2231 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2233 if (!mnt->mnt_sb->s_subtype)
2239 return ERR_PTR(err);
2243 * add a mount into a namespace's mount tree
2245 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2247 struct mountpoint *mp;
2248 struct mount *parent;
2251 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2253 mp = lock_mount(path);
2257 parent = real_mount(path->mnt);
2259 if (unlikely(!check_mnt(parent))) {
2260 /* that's acceptable only for automounts done in private ns */
2261 if (!(mnt_flags & MNT_SHRINKABLE))
2263 /* ... and for those we'd better have mountpoint still alive */
2264 if (!parent->mnt_ns)
2268 /* Refuse the same filesystem on the same mount point */
2270 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2271 path->mnt->mnt_root == path->dentry)
2275 if (d_is_symlink(newmnt->mnt.mnt_root))
2278 newmnt->mnt.mnt_flags = mnt_flags;
2279 err = graft_tree(newmnt, parent, mp);
2287 * create a new mount for userspace and request it to be added into the
2290 static int do_new_mount(struct path *path, const char *fstype, int flags,
2291 int mnt_flags, const char *name, void *data)
2293 struct file_system_type *type;
2294 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2295 struct vfsmount *mnt;
2301 type = get_fs_type(fstype);
2305 if (user_ns != &init_user_ns) {
2306 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2307 put_filesystem(type);
2310 /* Only in special cases allow devices from mounts
2311 * created outside the initial user namespace.
2313 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2315 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2319 mnt = vfs_kern_mount(type, flags, name, data);
2320 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2321 !mnt->mnt_sb->s_subtype)
2322 mnt = fs_set_subtype(mnt, fstype);
2324 put_filesystem(type);
2326 return PTR_ERR(mnt);
2328 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2334 int finish_automount(struct vfsmount *m, struct path *path)
2336 struct mount *mnt = real_mount(m);
2338 /* The new mount record should have at least 2 refs to prevent it being
2339 * expired before we get a chance to add it
2341 BUG_ON(mnt_get_count(mnt) < 2);
2343 if (m->mnt_sb == path->mnt->mnt_sb &&
2344 m->mnt_root == path->dentry) {
2349 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2353 /* remove m from any expiration list it may be on */
2354 if (!list_empty(&mnt->mnt_expire)) {
2356 list_del_init(&mnt->mnt_expire);
2365 * mnt_set_expiry - Put a mount on an expiration list
2366 * @mnt: The mount to list.
2367 * @expiry_list: The list to add the mount to.
2369 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2373 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2377 EXPORT_SYMBOL(mnt_set_expiry);
2380 * process a list of expirable mountpoints with the intent of discarding any
2381 * mountpoints that aren't in use and haven't been touched since last we came
2384 void mark_mounts_for_expiry(struct list_head *mounts)
2386 struct mount *mnt, *next;
2387 LIST_HEAD(graveyard);
2389 if (list_empty(mounts))
2395 /* extract from the expiration list every vfsmount that matches the
2396 * following criteria:
2397 * - only referenced by its parent vfsmount
2398 * - still marked for expiry (marked on the last call here; marks are
2399 * cleared by mntput())
2401 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2402 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2403 propagate_mount_busy(mnt, 1))
2405 list_move(&mnt->mnt_expire, &graveyard);
2407 while (!list_empty(&graveyard)) {
2408 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2409 touch_mnt_namespace(mnt->mnt_ns);
2410 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2412 unlock_mount_hash();
2416 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2419 * Ripoff of 'select_parent()'
2421 * search the list of submounts for a given mountpoint, and move any
2422 * shrinkable submounts to the 'graveyard' list.
2424 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2426 struct mount *this_parent = parent;
2427 struct list_head *next;
2431 next = this_parent->mnt_mounts.next;
2433 while (next != &this_parent->mnt_mounts) {
2434 struct list_head *tmp = next;
2435 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2438 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2441 * Descend a level if the d_mounts list is non-empty.
2443 if (!list_empty(&mnt->mnt_mounts)) {
2448 if (!propagate_mount_busy(mnt, 1)) {
2449 list_move_tail(&mnt->mnt_expire, graveyard);
2454 * All done at this level ... ascend and resume the search
2456 if (this_parent != parent) {
2457 next = this_parent->mnt_child.next;
2458 this_parent = this_parent->mnt_parent;
2465 * process a list of expirable mountpoints with the intent of discarding any
2466 * submounts of a specific parent mountpoint
2468 * mount_lock must be held for write
2470 static void shrink_submounts(struct mount *mnt)
2472 LIST_HEAD(graveyard);
2475 /* extract submounts of 'mountpoint' from the expiration list */
2476 while (select_submounts(mnt, &graveyard)) {
2477 while (!list_empty(&graveyard)) {
2478 m = list_first_entry(&graveyard, struct mount,
2480 touch_mnt_namespace(m->mnt_ns);
2481 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2487 * Some copy_from_user() implementations do not return the exact number of
2488 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2489 * Note that this function differs from copy_from_user() in that it will oops
2490 * on bad values of `to', rather than returning a short copy.
2492 static long exact_copy_from_user(void *to, const void __user * from,
2496 const char __user *f = from;
2499 if (!access_ok(VERIFY_READ, from, n))
2503 if (__get_user(c, f)) {
2514 int copy_mount_options(const void __user * data, unsigned long *where)
2524 if (!(page = __get_free_page(GFP_KERNEL)))
2527 /* We only care that *some* data at the address the user
2528 * gave us is valid. Just in case, we'll zero
2529 * the remainder of the page.
2531 /* copy_from_user cannot cross TASK_SIZE ! */
2532 size = TASK_SIZE - (unsigned long)data;
2533 if (size > PAGE_SIZE)
2536 i = size - exact_copy_from_user((void *)page, data, size);
2542 memset((char *)page + i, 0, PAGE_SIZE - i);
2547 char *copy_mount_string(const void __user *data)
2549 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2553 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2554 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2556 * data is a (void *) that can point to any structure up to
2557 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2558 * information (or be NULL).
2560 * Pre-0.97 versions of mount() didn't have a flags word.
2561 * When the flags word was introduced its top half was required
2562 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2563 * Therefore, if this magic number is present, it carries no information
2564 * and must be discarded.
2566 long do_mount(const char *dev_name, const char __user *dir_name,
2567 const char *type_page, unsigned long flags, void *data_page)
2574 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2575 flags &= ~MS_MGC_MSK;
2577 /* Basic sanity checks */
2579 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2581 /* ... and get the mountpoint */
2582 retval = user_path(dir_name, &path);
2586 retval = security_sb_mount(dev_name, &path,
2587 type_page, flags, data_page);
2588 if (!retval && !may_mount())
2593 /* Default to relatime unless overriden */
2594 if (!(flags & MS_NOATIME))
2595 mnt_flags |= MNT_RELATIME;
2597 /* Separate the per-mountpoint flags */
2598 if (flags & MS_NOSUID)
2599 mnt_flags |= MNT_NOSUID;
2600 if (flags & MS_NODEV)
2601 mnt_flags |= MNT_NODEV;
2602 if (flags & MS_NOEXEC)
2603 mnt_flags |= MNT_NOEXEC;
2604 if (flags & MS_NOATIME)
2605 mnt_flags |= MNT_NOATIME;
2606 if (flags & MS_NODIRATIME)
2607 mnt_flags |= MNT_NODIRATIME;
2608 if (flags & MS_STRICTATIME)
2609 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2610 if (flags & MS_RDONLY)
2611 mnt_flags |= MNT_READONLY;
2613 /* The default atime for remount is preservation */
2614 if ((flags & MS_REMOUNT) &&
2615 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2616 MS_STRICTATIME)) == 0)) {
2617 mnt_flags &= ~MNT_ATIME_MASK;
2618 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2621 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2622 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2625 if (flags & MS_REMOUNT)
2626 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2628 else if (flags & MS_BIND)
2629 retval = do_loopback(&path, dev_name, flags & MS_REC);
2630 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2631 retval = do_change_type(&path, flags);
2632 else if (flags & MS_MOVE)
2633 retval = do_move_mount(&path, dev_name);
2635 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2636 dev_name, data_page);
2642 static void free_mnt_ns(struct mnt_namespace *ns)
2644 ns_free_inum(&ns->ns);
2645 put_user_ns(ns->user_ns);
2650 * Assign a sequence number so we can detect when we attempt to bind
2651 * mount a reference to an older mount namespace into the current
2652 * mount namespace, preventing reference counting loops. A 64bit
2653 * number incrementing at 10Ghz will take 12,427 years to wrap which
2654 * is effectively never, so we can ignore the possibility.
2656 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2658 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2660 struct mnt_namespace *new_ns;
2663 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2665 return ERR_PTR(-ENOMEM);
2666 ret = ns_alloc_inum(&new_ns->ns);
2669 return ERR_PTR(ret);
2671 new_ns->ns.ops = &mntns_operations;
2672 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2673 atomic_set(&new_ns->count, 1);
2674 new_ns->root = NULL;
2675 INIT_LIST_HEAD(&new_ns->list);
2676 init_waitqueue_head(&new_ns->poll);
2678 new_ns->user_ns = get_user_ns(user_ns);
2682 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2683 struct user_namespace *user_ns, struct fs_struct *new_fs)
2685 struct mnt_namespace *new_ns;
2686 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2687 struct mount *p, *q;
2694 if (likely(!(flags & CLONE_NEWNS))) {
2701 new_ns = alloc_mnt_ns(user_ns);
2706 /* First pass: copy the tree topology */
2707 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2708 if (user_ns != ns->user_ns)
2709 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2710 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2713 free_mnt_ns(new_ns);
2714 return ERR_CAST(new);
2717 list_add_tail(&new_ns->list, &new->mnt_list);
2720 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2721 * as belonging to new namespace. We have already acquired a private
2722 * fs_struct, so tsk->fs->lock is not needed.
2729 if (&p->mnt == new_fs->root.mnt) {
2730 new_fs->root.mnt = mntget(&q->mnt);
2733 if (&p->mnt == new_fs->pwd.mnt) {
2734 new_fs->pwd.mnt = mntget(&q->mnt);
2738 p = next_mnt(p, old);
2739 q = next_mnt(q, new);
2742 while (p->mnt.mnt_root != q->mnt.mnt_root)
2743 p = next_mnt(p, old);
2756 * create_mnt_ns - creates a private namespace and adds a root filesystem
2757 * @mnt: pointer to the new root filesystem mountpoint
2759 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2761 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2762 if (!IS_ERR(new_ns)) {
2763 struct mount *mnt = real_mount(m);
2764 mnt->mnt_ns = new_ns;
2766 list_add(&mnt->mnt_list, &new_ns->list);
2773 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2775 struct mnt_namespace *ns;
2776 struct super_block *s;
2780 ns = create_mnt_ns(mnt);
2782 return ERR_CAST(ns);
2784 err = vfs_path_lookup(mnt->mnt_root, mnt,
2785 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2790 return ERR_PTR(err);
2792 /* trade a vfsmount reference for active sb one */
2793 s = path.mnt->mnt_sb;
2794 atomic_inc(&s->s_active);
2796 /* lock the sucker */
2797 down_write(&s->s_umount);
2798 /* ... and return the root of (sub)tree on it */
2801 EXPORT_SYMBOL(mount_subtree);
2803 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2804 char __user *, type, unsigned long, flags, void __user *, data)
2809 unsigned long data_page;
2811 kernel_type = copy_mount_string(type);
2812 ret = PTR_ERR(kernel_type);
2813 if (IS_ERR(kernel_type))
2816 kernel_dev = copy_mount_string(dev_name);
2817 ret = PTR_ERR(kernel_dev);
2818 if (IS_ERR(kernel_dev))
2821 ret = copy_mount_options(data, &data_page);
2825 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2826 (void *) data_page);
2828 free_page(data_page);
2838 * Return true if path is reachable from root
2840 * namespace_sem or mount_lock is held
2842 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2843 const struct path *root)
2845 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2846 dentry = mnt->mnt_mountpoint;
2847 mnt = mnt->mnt_parent;
2849 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2852 int path_is_under(struct path *path1, struct path *path2)
2855 read_seqlock_excl(&mount_lock);
2856 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2857 read_sequnlock_excl(&mount_lock);
2860 EXPORT_SYMBOL(path_is_under);
2863 * pivot_root Semantics:
2864 * Moves the root file system of the current process to the directory put_old,
2865 * makes new_root as the new root file system of the current process, and sets
2866 * root/cwd of all processes which had them on the current root to new_root.
2869 * The new_root and put_old must be directories, and must not be on the
2870 * same file system as the current process root. The put_old must be
2871 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2872 * pointed to by put_old must yield the same directory as new_root. No other
2873 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2875 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2876 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2877 * in this situation.
2880 * - we don't move root/cwd if they are not at the root (reason: if something
2881 * cared enough to change them, it's probably wrong to force them elsewhere)
2882 * - it's okay to pick a root that isn't the root of a file system, e.g.
2883 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2884 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2887 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2888 const char __user *, put_old)
2890 struct path new, old, parent_path, root_parent, root;
2891 struct mount *new_mnt, *root_mnt, *old_mnt;
2892 struct mountpoint *old_mp, *root_mp;
2898 error = user_path_dir(new_root, &new);
2902 error = user_path_dir(put_old, &old);
2906 error = security_sb_pivotroot(&old, &new);
2910 get_fs_root(current->fs, &root);
2911 old_mp = lock_mount(&old);
2912 error = PTR_ERR(old_mp);
2917 new_mnt = real_mount(new.mnt);
2918 root_mnt = real_mount(root.mnt);
2919 old_mnt = real_mount(old.mnt);
2920 if (IS_MNT_SHARED(old_mnt) ||
2921 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2922 IS_MNT_SHARED(root_mnt->mnt_parent))
2924 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2926 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2929 if (d_unlinked(new.dentry))
2932 if (new_mnt == root_mnt || old_mnt == root_mnt)
2933 goto out4; /* loop, on the same file system */
2935 if (root.mnt->mnt_root != root.dentry)
2936 goto out4; /* not a mountpoint */
2937 if (!mnt_has_parent(root_mnt))
2938 goto out4; /* not attached */
2939 root_mp = root_mnt->mnt_mp;
2940 if (new.mnt->mnt_root != new.dentry)
2941 goto out4; /* not a mountpoint */
2942 if (!mnt_has_parent(new_mnt))
2943 goto out4; /* not attached */
2944 /* make sure we can reach put_old from new_root */
2945 if (!is_path_reachable(old_mnt, old.dentry, &new))
2947 /* make certain new is below the root */
2948 if (!is_path_reachable(new_mnt, new.dentry, &root))
2950 root_mp->m_count++; /* pin it so it won't go away */
2952 detach_mnt(new_mnt, &parent_path);
2953 detach_mnt(root_mnt, &root_parent);
2954 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2955 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2956 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2958 /* mount old root on put_old */
2959 attach_mnt(root_mnt, old_mnt, old_mp);
2960 /* mount new_root on / */
2961 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2962 touch_mnt_namespace(current->nsproxy->mnt_ns);
2963 /* A moved mount should not expire automatically */
2964 list_del_init(&new_mnt->mnt_expire);
2965 unlock_mount_hash();
2966 chroot_fs_refs(&root, &new);
2967 put_mountpoint(root_mp);
2970 unlock_mount(old_mp);
2972 path_put(&root_parent);
2973 path_put(&parent_path);
2985 static void __init init_mount_tree(void)
2987 struct vfsmount *mnt;
2988 struct mnt_namespace *ns;
2990 struct file_system_type *type;
2992 type = get_fs_type("rootfs");
2994 panic("Can't find rootfs type");
2995 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2996 put_filesystem(type);
2998 panic("Can't create rootfs");
3000 ns = create_mnt_ns(mnt);
3002 panic("Can't allocate initial namespace");
3004 init_task.nsproxy->mnt_ns = ns;
3008 root.dentry = mnt->mnt_root;
3009 mnt->mnt_flags |= MNT_LOCKED;
3011 set_fs_pwd(current->fs, &root);
3012 set_fs_root(current->fs, &root);
3015 void __init mnt_init(void)
3020 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3021 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3023 mount_hashtable = alloc_large_system_hash("Mount-cache",
3024 sizeof(struct hlist_head),
3027 &m_hash_shift, &m_hash_mask, 0, 0);
3028 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3029 sizeof(struct hlist_head),
3032 &mp_hash_shift, &mp_hash_mask, 0, 0);
3034 if (!mount_hashtable || !mountpoint_hashtable)
3035 panic("Failed to allocate mount hash table\n");
3037 for (u = 0; u <= m_hash_mask; u++)
3038 INIT_HLIST_HEAD(&mount_hashtable[u]);
3039 for (u = 0; u <= mp_hash_mask; u++)
3040 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3046 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3048 fs_kobj = kobject_create_and_add("fs", NULL);
3050 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3055 void put_mnt_ns(struct mnt_namespace *ns)
3057 if (!atomic_dec_and_test(&ns->count))
3059 drop_collected_mounts(&ns->root->mnt);
3063 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3065 struct vfsmount *mnt;
3066 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3069 * it is a longterm mount, don't release mnt until
3070 * we unmount before file sys is unregistered
3072 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3076 EXPORT_SYMBOL_GPL(kern_mount_data);
3078 void kern_unmount(struct vfsmount *mnt)
3080 /* release long term mount so mount point can be released */
3081 if (!IS_ERR_OR_NULL(mnt)) {
3082 real_mount(mnt)->mnt_ns = NULL;
3083 synchronize_rcu(); /* yecchhh... */
3087 EXPORT_SYMBOL(kern_unmount);
3089 bool our_mnt(struct vfsmount *mnt)
3091 return check_mnt(real_mount(mnt));
3094 bool current_chrooted(void)
3096 /* Does the current process have a non-standard root */
3097 struct path ns_root;
3098 struct path fs_root;
3101 /* Find the namespace root */
3102 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3103 ns_root.dentry = ns_root.mnt->mnt_root;
3105 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3108 get_fs_root(current->fs, &fs_root);
3110 chrooted = !path_equal(&fs_root, &ns_root);
3118 bool fs_fully_visible(struct file_system_type *type)
3120 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3122 bool visible = false;
3127 down_read(&namespace_sem);
3128 list_for_each_entry(mnt, &ns->list, mnt_list) {
3129 struct mount *child;
3130 if (mnt->mnt.mnt_sb->s_type != type)
3133 /* This mount is not fully visible if there are any child mounts
3134 * that cover anything except for empty directories.
3136 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3137 struct inode *inode = child->mnt_mountpoint->d_inode;
3138 if (!S_ISDIR(inode->i_mode))
3140 if (inode->i_nlink > 2)
3148 up_read(&namespace_sem);
3152 static struct ns_common *mntns_get(struct task_struct *task)
3154 struct ns_common *ns = NULL;
3155 struct nsproxy *nsproxy;
3158 nsproxy = task->nsproxy;
3160 ns = &nsproxy->mnt_ns->ns;
3161 get_mnt_ns(to_mnt_ns(ns));
3168 static void mntns_put(struct ns_common *ns)
3170 put_mnt_ns(to_mnt_ns(ns));
3173 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3175 struct fs_struct *fs = current->fs;
3176 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3179 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3180 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3181 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3188 put_mnt_ns(nsproxy->mnt_ns);
3189 nsproxy->mnt_ns = mnt_ns;
3192 root.mnt = &mnt_ns->root->mnt;
3193 root.dentry = mnt_ns->root->mnt.mnt_root;
3195 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3198 /* Update the pwd and root */
3199 set_fs_pwd(fs, &root);
3200 set_fs_root(fs, &root);
3206 const struct proc_ns_operations mntns_operations = {
3208 .type = CLONE_NEWNS,
3211 .install = mntns_install,