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 struct mount *alloc_vfsmnt(const char *name)
195 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
199 err = mnt_alloc_id(mnt);
204 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
205 if (!mnt->mnt_devname)
210 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
212 goto out_free_devname;
214 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
217 mnt->mnt_writers = 0;
220 INIT_HLIST_NODE(&mnt->mnt_hash);
221 INIT_LIST_HEAD(&mnt->mnt_child);
222 INIT_LIST_HEAD(&mnt->mnt_mounts);
223 INIT_LIST_HEAD(&mnt->mnt_list);
224 INIT_LIST_HEAD(&mnt->mnt_expire);
225 INIT_LIST_HEAD(&mnt->mnt_share);
226 INIT_LIST_HEAD(&mnt->mnt_slave_list);
227 INIT_LIST_HEAD(&mnt->mnt_slave);
228 INIT_HLIST_NODE(&mnt->mnt_mp_list);
229 #ifdef CONFIG_FSNOTIFY
230 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
237 kfree(mnt->mnt_devname);
242 kmem_cache_free(mnt_cache, mnt);
247 * Most r/o checks on a fs are for operations that take
248 * discrete amounts of time, like a write() or unlink().
249 * We must keep track of when those operations start
250 * (for permission checks) and when they end, so that
251 * we can determine when writes are able to occur to
255 * __mnt_is_readonly: check whether a mount is read-only
256 * @mnt: the mount to check for its write status
258 * This shouldn't be used directly ouside of the VFS.
259 * It does not guarantee that the filesystem will stay
260 * r/w, just that it is right *now*. This can not and
261 * should not be used in place of IS_RDONLY(inode).
262 * mnt_want/drop_write() will _keep_ the filesystem
265 int __mnt_is_readonly(struct vfsmount *mnt)
267 if (mnt->mnt_flags & MNT_READONLY)
269 if (mnt->mnt_sb->s_flags & MS_RDONLY)
273 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
275 static inline void mnt_inc_writers(struct mount *mnt)
278 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
284 static inline void mnt_dec_writers(struct mount *mnt)
287 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
293 static unsigned int mnt_get_writers(struct mount *mnt)
296 unsigned int count = 0;
299 for_each_possible_cpu(cpu) {
300 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
305 return mnt->mnt_writers;
309 static int mnt_is_readonly(struct vfsmount *mnt)
311 if (mnt->mnt_sb->s_readonly_remount)
313 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
315 return __mnt_is_readonly(mnt);
319 * Most r/o & frozen checks on a fs are for operations that take discrete
320 * amounts of time, like a write() or unlink(). We must keep track of when
321 * those operations start (for permission checks) and when they end, so that we
322 * can determine when writes are able to occur to a filesystem.
325 * __mnt_want_write - get write access to a mount without freeze protection
326 * @m: the mount on which to take a write
328 * This tells the low-level filesystem that a write is about to be performed to
329 * it, and makes sure that writes are allowed (mnt it read-write) before
330 * returning success. This operation does not protect against filesystem being
331 * frozen. When the write operation is finished, __mnt_drop_write() must be
332 * called. This is effectively a refcount.
334 int __mnt_want_write(struct vfsmount *m)
336 struct mount *mnt = real_mount(m);
340 mnt_inc_writers(mnt);
342 * The store to mnt_inc_writers must be visible before we pass
343 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
344 * incremented count after it has set MNT_WRITE_HOLD.
347 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
350 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
351 * be set to match its requirements. So we must not load that until
352 * MNT_WRITE_HOLD is cleared.
355 if (mnt_is_readonly(m)) {
356 mnt_dec_writers(mnt);
365 * mnt_want_write - get write access to a mount
366 * @m: the mount on which to take a write
368 * This tells the low-level filesystem that a write is about to be performed to
369 * it, and makes sure that writes are allowed (mount is read-write, filesystem
370 * is not frozen) before returning success. When the write operation is
371 * finished, mnt_drop_write() must be called. This is effectively a refcount.
373 int mnt_want_write(struct vfsmount *m)
377 sb_start_write(m->mnt_sb);
378 ret = __mnt_want_write(m);
380 sb_end_write(m->mnt_sb);
383 EXPORT_SYMBOL_GPL(mnt_want_write);
386 * mnt_clone_write - get write access to a mount
387 * @mnt: the mount on which to take a write
389 * This is effectively like mnt_want_write, except
390 * it must only be used to take an extra write reference
391 * on a mountpoint that we already know has a write reference
392 * on it. This allows some optimisation.
394 * After finished, mnt_drop_write must be called as usual to
395 * drop the reference.
397 int mnt_clone_write(struct vfsmount *mnt)
399 /* superblock may be r/o */
400 if (__mnt_is_readonly(mnt))
403 mnt_inc_writers(real_mount(mnt));
407 EXPORT_SYMBOL_GPL(mnt_clone_write);
410 * __mnt_want_write_file - get write access to a file's mount
411 * @file: the file who's mount on which to take a write
413 * This is like __mnt_want_write, but it takes a file and can
414 * do some optimisations if the file is open for write already
416 int __mnt_want_write_file(struct file *file)
418 if (!(file->f_mode & FMODE_WRITER))
419 return __mnt_want_write(file->f_path.mnt);
421 return mnt_clone_write(file->f_path.mnt);
425 * mnt_want_write_file - get write access to a file's mount
426 * @file: the file who's mount on which to take a write
428 * This is like mnt_want_write, but it takes a file and can
429 * do some optimisations if the file is open for write already
431 int mnt_want_write_file(struct file *file)
435 sb_start_write(file->f_path.mnt->mnt_sb);
436 ret = __mnt_want_write_file(file);
438 sb_end_write(file->f_path.mnt->mnt_sb);
441 EXPORT_SYMBOL_GPL(mnt_want_write_file);
444 * __mnt_drop_write - give up write access to a mount
445 * @mnt: the mount on which to give up write access
447 * Tells the low-level filesystem that we are done
448 * performing writes to it. Must be matched with
449 * __mnt_want_write() call above.
451 void __mnt_drop_write(struct vfsmount *mnt)
454 mnt_dec_writers(real_mount(mnt));
459 * mnt_drop_write - give up write access to a mount
460 * @mnt: the mount on which to give up write access
462 * Tells the low-level filesystem that we are done performing writes to it and
463 * also allows filesystem to be frozen again. Must be matched with
464 * mnt_want_write() call above.
466 void mnt_drop_write(struct vfsmount *mnt)
468 __mnt_drop_write(mnt);
469 sb_end_write(mnt->mnt_sb);
471 EXPORT_SYMBOL_GPL(mnt_drop_write);
473 void __mnt_drop_write_file(struct file *file)
475 __mnt_drop_write(file->f_path.mnt);
478 void mnt_drop_write_file(struct file *file)
480 mnt_drop_write(file->f_path.mnt);
482 EXPORT_SYMBOL(mnt_drop_write_file);
484 static int mnt_make_readonly(struct mount *mnt)
489 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
491 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
492 * should be visible before we do.
497 * With writers on hold, if this value is zero, then there are
498 * definitely no active writers (although held writers may subsequently
499 * increment the count, they'll have to wait, and decrement it after
500 * seeing MNT_READONLY).
502 * It is OK to have counter incremented on one CPU and decremented on
503 * another: the sum will add up correctly. The danger would be when we
504 * sum up each counter, if we read a counter before it is incremented,
505 * but then read another CPU's count which it has been subsequently
506 * decremented from -- we would see more decrements than we should.
507 * MNT_WRITE_HOLD protects against this scenario, because
508 * mnt_want_write first increments count, then smp_mb, then spins on
509 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
510 * we're counting up here.
512 if (mnt_get_writers(mnt) > 0)
515 mnt->mnt.mnt_flags |= MNT_READONLY;
517 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
518 * that become unheld will see MNT_READONLY.
521 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
526 static void __mnt_unmake_readonly(struct mount *mnt)
529 mnt->mnt.mnt_flags &= ~MNT_READONLY;
533 int sb_prepare_remount_readonly(struct super_block *sb)
538 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
539 if (atomic_long_read(&sb->s_remove_count))
543 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
544 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
545 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
547 if (mnt_get_writers(mnt) > 0) {
553 if (!err && atomic_long_read(&sb->s_remove_count))
557 sb->s_readonly_remount = 1;
560 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
561 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
562 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
569 static void free_vfsmnt(struct mount *mnt)
571 kfree(mnt->mnt_devname);
573 free_percpu(mnt->mnt_pcp);
575 kmem_cache_free(mnt_cache, mnt);
578 static void delayed_free_vfsmnt(struct rcu_head *head)
580 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
583 /* call under rcu_read_lock */
584 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
587 if (read_seqretry(&mount_lock, seq))
591 mnt = real_mount(bastard);
592 mnt_add_count(mnt, 1);
593 if (likely(!read_seqretry(&mount_lock, seq)))
595 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
596 mnt_add_count(mnt, -1);
606 * find the first mount at @dentry on vfsmount @mnt.
607 * call under rcu_read_lock()
609 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
611 struct hlist_head *head = m_hash(mnt, dentry);
614 hlist_for_each_entry_rcu(p, head, mnt_hash)
615 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
621 * find the last mount at @dentry on vfsmount @mnt.
622 * mount_lock must be held.
624 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
626 struct mount *p, *res;
627 res = p = __lookup_mnt(mnt, dentry);
630 hlist_for_each_entry_continue(p, mnt_hash) {
631 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
640 * lookup_mnt - Return the first child mount mounted at path
642 * "First" means first mounted chronologically. If you create the
645 * mount /dev/sda1 /mnt
646 * mount /dev/sda2 /mnt
647 * mount /dev/sda3 /mnt
649 * Then lookup_mnt() on the base /mnt dentry in the root mount will
650 * return successively the root dentry and vfsmount of /dev/sda1, then
651 * /dev/sda2, then /dev/sda3, then NULL.
653 * lookup_mnt takes a reference to the found vfsmount.
655 struct vfsmount *lookup_mnt(struct path *path)
657 struct mount *child_mnt;
663 seq = read_seqbegin(&mount_lock);
664 child_mnt = __lookup_mnt(path->mnt, path->dentry);
665 m = child_mnt ? &child_mnt->mnt : NULL;
666 } while (!legitimize_mnt(m, seq));
672 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
673 * current mount namespace.
675 * The common case is dentries are not mountpoints at all and that
676 * test is handled inline. For the slow case when we are actually
677 * dealing with a mountpoint of some kind, walk through all of the
678 * mounts in the current mount namespace and test to see if the dentry
681 * The mount_hashtable is not usable in the context because we
682 * need to identify all mounts that may be in the current mount
683 * namespace not just a mount that happens to have some specified
686 bool __is_local_mountpoint(struct dentry *dentry)
688 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
690 bool is_covered = false;
692 if (!d_mountpoint(dentry))
695 down_read(&namespace_sem);
696 list_for_each_entry(mnt, &ns->list, mnt_list) {
697 is_covered = (mnt->mnt_mountpoint == dentry);
701 up_read(&namespace_sem);
706 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
708 struct hlist_head *chain = mp_hash(dentry);
709 struct mountpoint *mp;
711 hlist_for_each_entry(mp, chain, m_hash) {
712 if (mp->m_dentry == dentry) {
713 /* might be worth a WARN_ON() */
714 if (d_unlinked(dentry))
715 return ERR_PTR(-ENOENT);
723 static struct mountpoint *new_mountpoint(struct dentry *dentry)
725 struct hlist_head *chain = mp_hash(dentry);
726 struct mountpoint *mp;
729 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
731 return ERR_PTR(-ENOMEM);
733 ret = d_set_mounted(dentry);
739 mp->m_dentry = dentry;
741 hlist_add_head(&mp->m_hash, chain);
742 INIT_HLIST_HEAD(&mp->m_list);
746 static void put_mountpoint(struct mountpoint *mp)
748 if (!--mp->m_count) {
749 struct dentry *dentry = mp->m_dentry;
750 BUG_ON(!hlist_empty(&mp->m_list));
751 spin_lock(&dentry->d_lock);
752 dentry->d_flags &= ~DCACHE_MOUNTED;
753 spin_unlock(&dentry->d_lock);
754 hlist_del(&mp->m_hash);
759 static inline int check_mnt(struct mount *mnt)
761 return mnt->mnt_ns == current->nsproxy->mnt_ns;
765 * vfsmount lock must be held for write
767 static void touch_mnt_namespace(struct mnt_namespace *ns)
771 wake_up_interruptible(&ns->poll);
776 * vfsmount lock must be held for write
778 static void __touch_mnt_namespace(struct mnt_namespace *ns)
780 if (ns && ns->event != event) {
782 wake_up_interruptible(&ns->poll);
787 * vfsmount lock must be held for write
789 static void detach_mnt(struct mount *mnt, struct path *old_path)
791 old_path->dentry = mnt->mnt_mountpoint;
792 old_path->mnt = &mnt->mnt_parent->mnt;
793 mnt->mnt_parent = mnt;
794 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
795 list_del_init(&mnt->mnt_child);
796 hlist_del_init_rcu(&mnt->mnt_hash);
797 hlist_del_init(&mnt->mnt_mp_list);
798 put_mountpoint(mnt->mnt_mp);
803 * vfsmount lock must be held for write
805 void mnt_set_mountpoint(struct mount *mnt,
806 struct mountpoint *mp,
807 struct mount *child_mnt)
810 mnt_add_count(mnt, 1); /* essentially, that's mntget */
811 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
812 child_mnt->mnt_parent = mnt;
813 child_mnt->mnt_mp = mp;
814 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
818 * vfsmount lock must be held for write
820 static void attach_mnt(struct mount *mnt,
821 struct mount *parent,
822 struct mountpoint *mp)
824 mnt_set_mountpoint(parent, mp, mnt);
825 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
826 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
829 static void attach_shadowed(struct mount *mnt,
830 struct mount *parent,
831 struct mount *shadows)
834 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
835 list_add(&mnt->mnt_child, &shadows->mnt_child);
837 hlist_add_head_rcu(&mnt->mnt_hash,
838 m_hash(&parent->mnt, mnt->mnt_mountpoint));
839 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
844 * vfsmount lock must be held for write
846 static void commit_tree(struct mount *mnt, struct mount *shadows)
848 struct mount *parent = mnt->mnt_parent;
851 struct mnt_namespace *n = parent->mnt_ns;
853 BUG_ON(parent == mnt);
855 list_add_tail(&head, &mnt->mnt_list);
856 list_for_each_entry(m, &head, mnt_list)
859 list_splice(&head, n->list.prev);
861 attach_shadowed(mnt, parent, shadows);
862 touch_mnt_namespace(n);
865 static struct mount *next_mnt(struct mount *p, struct mount *root)
867 struct list_head *next = p->mnt_mounts.next;
868 if (next == &p->mnt_mounts) {
872 next = p->mnt_child.next;
873 if (next != &p->mnt_parent->mnt_mounts)
878 return list_entry(next, struct mount, mnt_child);
881 static struct mount *skip_mnt_tree(struct mount *p)
883 struct list_head *prev = p->mnt_mounts.prev;
884 while (prev != &p->mnt_mounts) {
885 p = list_entry(prev, struct mount, mnt_child);
886 prev = p->mnt_mounts.prev;
892 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
898 return ERR_PTR(-ENODEV);
900 mnt = alloc_vfsmnt(name);
902 return ERR_PTR(-ENOMEM);
904 if (flags & MS_KERNMOUNT)
905 mnt->mnt.mnt_flags = MNT_INTERNAL;
907 root = mount_fs(type, flags, name, data);
911 return ERR_CAST(root);
914 mnt->mnt.mnt_root = root;
915 mnt->mnt.mnt_sb = root->d_sb;
916 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
917 mnt->mnt_parent = mnt;
919 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
923 EXPORT_SYMBOL_GPL(vfs_kern_mount);
925 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
928 struct super_block *sb = old->mnt.mnt_sb;
932 mnt = alloc_vfsmnt(old->mnt_devname);
934 return ERR_PTR(-ENOMEM);
936 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
937 mnt->mnt_group_id = 0; /* not a peer of original */
939 mnt->mnt_group_id = old->mnt_group_id;
941 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
942 err = mnt_alloc_group_id(mnt);
947 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
948 /* Don't allow unprivileged users to change mount flags */
949 if (flag & CL_UNPRIVILEGED) {
950 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
952 if (mnt->mnt.mnt_flags & MNT_READONLY)
953 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
955 if (mnt->mnt.mnt_flags & MNT_NODEV)
956 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
958 if (mnt->mnt.mnt_flags & MNT_NOSUID)
959 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
961 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
962 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
965 /* Don't allow unprivileged users to reveal what is under a mount */
966 if ((flag & CL_UNPRIVILEGED) &&
967 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
968 mnt->mnt.mnt_flags |= MNT_LOCKED;
970 atomic_inc(&sb->s_active);
971 mnt->mnt.mnt_sb = sb;
972 mnt->mnt.mnt_root = dget(root);
973 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
974 mnt->mnt_parent = mnt;
976 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
979 if ((flag & CL_SLAVE) ||
980 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
981 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
982 mnt->mnt_master = old;
983 CLEAR_MNT_SHARED(mnt);
984 } else if (!(flag & CL_PRIVATE)) {
985 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
986 list_add(&mnt->mnt_share, &old->mnt_share);
987 if (IS_MNT_SLAVE(old))
988 list_add(&mnt->mnt_slave, &old->mnt_slave);
989 mnt->mnt_master = old->mnt_master;
991 if (flag & CL_MAKE_SHARED)
994 /* stick the duplicate mount on the same expiry list
995 * as the original if that was on one */
996 if (flag & CL_EXPIRE) {
997 if (!list_empty(&old->mnt_expire))
998 list_add(&mnt->mnt_expire, &old->mnt_expire);
1006 return ERR_PTR(err);
1009 static void cleanup_mnt(struct mount *mnt)
1012 * This probably indicates that somebody messed
1013 * up a mnt_want/drop_write() pair. If this
1014 * happens, the filesystem was probably unable
1015 * to make r/w->r/o transitions.
1018 * The locking used to deal with mnt_count decrement provides barriers,
1019 * so mnt_get_writers() below is safe.
1021 WARN_ON(mnt_get_writers(mnt));
1022 if (unlikely(mnt->mnt_pins.first))
1024 fsnotify_vfsmount_delete(&mnt->mnt);
1025 dput(mnt->mnt.mnt_root);
1026 deactivate_super(mnt->mnt.mnt_sb);
1028 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1031 static void __cleanup_mnt(struct rcu_head *head)
1033 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1036 static LLIST_HEAD(delayed_mntput_list);
1037 static void delayed_mntput(struct work_struct *unused)
1039 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1040 struct llist_node *next;
1042 for (; node; node = next) {
1043 next = llist_next(node);
1044 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1047 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1049 static void mntput_no_expire(struct mount *mnt)
1052 mnt_add_count(mnt, -1);
1053 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1058 if (mnt_get_count(mnt)) {
1060 unlock_mount_hash();
1063 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1065 unlock_mount_hash();
1068 mnt->mnt.mnt_flags |= MNT_DOOMED;
1071 list_del(&mnt->mnt_instance);
1072 unlock_mount_hash();
1074 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1075 struct task_struct *task = current;
1076 if (likely(!(task->flags & PF_KTHREAD))) {
1077 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1078 if (!task_work_add(task, &mnt->mnt_rcu, true))
1081 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1082 schedule_delayed_work(&delayed_mntput_work, 1);
1088 void mntput(struct vfsmount *mnt)
1091 struct mount *m = real_mount(mnt);
1092 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1093 if (unlikely(m->mnt_expiry_mark))
1094 m->mnt_expiry_mark = 0;
1095 mntput_no_expire(m);
1098 EXPORT_SYMBOL(mntput);
1100 struct vfsmount *mntget(struct vfsmount *mnt)
1103 mnt_add_count(real_mount(mnt), 1);
1106 EXPORT_SYMBOL(mntget);
1108 struct vfsmount *mnt_clone_internal(struct path *path)
1111 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1114 p->mnt.mnt_flags |= MNT_INTERNAL;
1118 static inline void mangle(struct seq_file *m, const char *s)
1120 seq_escape(m, s, " \t\n\\");
1124 * Simple .show_options callback for filesystems which don't want to
1125 * implement more complex mount option showing.
1127 * See also save_mount_options().
1129 int generic_show_options(struct seq_file *m, struct dentry *root)
1131 const char *options;
1134 options = rcu_dereference(root->d_sb->s_options);
1136 if (options != NULL && options[0]) {
1144 EXPORT_SYMBOL(generic_show_options);
1147 * If filesystem uses generic_show_options(), this function should be
1148 * called from the fill_super() callback.
1150 * The .remount_fs callback usually needs to be handled in a special
1151 * way, to make sure, that previous options are not overwritten if the
1154 * Also note, that if the filesystem's .remount_fs function doesn't
1155 * reset all options to their default value, but changes only newly
1156 * given options, then the displayed options will not reflect reality
1159 void save_mount_options(struct super_block *sb, char *options)
1161 BUG_ON(sb->s_options);
1162 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1164 EXPORT_SYMBOL(save_mount_options);
1166 void replace_mount_options(struct super_block *sb, char *options)
1168 char *old = sb->s_options;
1169 rcu_assign_pointer(sb->s_options, options);
1175 EXPORT_SYMBOL(replace_mount_options);
1177 #ifdef CONFIG_PROC_FS
1178 /* iterator; we want it to have access to namespace_sem, thus here... */
1179 static void *m_start(struct seq_file *m, loff_t *pos)
1181 struct proc_mounts *p = proc_mounts(m);
1183 down_read(&namespace_sem);
1184 if (p->cached_event == p->ns->event) {
1185 void *v = p->cached_mount;
1186 if (*pos == p->cached_index)
1188 if (*pos == p->cached_index + 1) {
1189 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1190 return p->cached_mount = v;
1194 p->cached_event = p->ns->event;
1195 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1196 p->cached_index = *pos;
1197 return p->cached_mount;
1200 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1202 struct proc_mounts *p = proc_mounts(m);
1204 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1205 p->cached_index = *pos;
1206 return p->cached_mount;
1209 static void m_stop(struct seq_file *m, void *v)
1211 up_read(&namespace_sem);
1214 static int m_show(struct seq_file *m, void *v)
1216 struct proc_mounts *p = proc_mounts(m);
1217 struct mount *r = list_entry(v, struct mount, mnt_list);
1218 return p->show(m, &r->mnt);
1221 const struct seq_operations mounts_op = {
1227 #endif /* CONFIG_PROC_FS */
1230 * may_umount_tree - check if a mount tree is busy
1231 * @mnt: root of mount tree
1233 * This is called to check if a tree of mounts has any
1234 * open files, pwds, chroots or sub mounts that are
1237 int may_umount_tree(struct vfsmount *m)
1239 struct mount *mnt = real_mount(m);
1240 int actual_refs = 0;
1241 int minimum_refs = 0;
1245 /* write lock needed for mnt_get_count */
1247 for (p = mnt; p; p = next_mnt(p, mnt)) {
1248 actual_refs += mnt_get_count(p);
1251 unlock_mount_hash();
1253 if (actual_refs > minimum_refs)
1259 EXPORT_SYMBOL(may_umount_tree);
1262 * may_umount - check if a mount point is busy
1263 * @mnt: root of mount
1265 * This is called to check if a mount point has any
1266 * open files, pwds, chroots or sub mounts. If the
1267 * mount has sub mounts this will return busy
1268 * regardless of whether the sub mounts are busy.
1270 * Doesn't take quota and stuff into account. IOW, in some cases it will
1271 * give false negatives. The main reason why it's here is that we need
1272 * a non-destructive way to look for easily umountable filesystems.
1274 int may_umount(struct vfsmount *mnt)
1277 down_read(&namespace_sem);
1279 if (propagate_mount_busy(real_mount(mnt), 2))
1281 unlock_mount_hash();
1282 up_read(&namespace_sem);
1286 EXPORT_SYMBOL(may_umount);
1288 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1290 static void namespace_unlock(void)
1293 struct hlist_head head = unmounted;
1295 if (likely(hlist_empty(&head))) {
1296 up_write(&namespace_sem);
1300 head.first->pprev = &head.first;
1301 INIT_HLIST_HEAD(&unmounted);
1303 /* undo decrements we'd done in umount_tree() */
1304 hlist_for_each_entry(mnt, &head, mnt_hash)
1305 if (mnt->mnt_ex_mountpoint.mnt)
1306 mntget(mnt->mnt_ex_mountpoint.mnt);
1308 up_write(&namespace_sem);
1312 while (!hlist_empty(&head)) {
1313 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1314 hlist_del_init(&mnt->mnt_hash);
1315 if (mnt->mnt_ex_mountpoint.mnt)
1316 path_put(&mnt->mnt_ex_mountpoint);
1321 static inline void namespace_lock(void)
1323 down_write(&namespace_sem);
1327 * mount_lock must be held
1328 * namespace_sem must be held for write
1329 * how = 0 => just this tree, don't propagate
1330 * how = 1 => propagate; we know that nobody else has reference to any victims
1331 * how = 2 => lazy umount
1333 void umount_tree(struct mount *mnt, int how)
1335 HLIST_HEAD(tmp_list);
1337 struct mount *last = NULL;
1339 for (p = mnt; p; p = next_mnt(p, mnt)) {
1340 hlist_del_init_rcu(&p->mnt_hash);
1341 hlist_add_head(&p->mnt_hash, &tmp_list);
1344 hlist_for_each_entry(p, &tmp_list, mnt_hash)
1345 list_del_init(&p->mnt_child);
1348 propagate_umount(&tmp_list);
1350 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
1351 list_del_init(&p->mnt_expire);
1352 list_del_init(&p->mnt_list);
1353 __touch_mnt_namespace(p->mnt_ns);
1356 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
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 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1362 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1363 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1364 p->mnt_mountpoint = p->mnt.mnt_root;
1368 change_mnt_propagation(p, MS_PRIVATE);
1372 last->mnt_hash.next = unmounted.first;
1373 unmounted.first = tmp_list.first;
1374 unmounted.first->pprev = &unmounted.first;
1378 static void shrink_submounts(struct mount *mnt);
1380 static int do_umount(struct mount *mnt, int flags)
1382 struct super_block *sb = mnt->mnt.mnt_sb;
1385 retval = security_sb_umount(&mnt->mnt, flags);
1390 * Allow userspace to request a mountpoint be expired rather than
1391 * unmounting unconditionally. Unmount only happens if:
1392 * (1) the mark is already set (the mark is cleared by mntput())
1393 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1395 if (flags & MNT_EXPIRE) {
1396 if (&mnt->mnt == current->fs->root.mnt ||
1397 flags & (MNT_FORCE | MNT_DETACH))
1401 * probably don't strictly need the lock here if we examined
1402 * all race cases, but it's a slowpath.
1405 if (mnt_get_count(mnt) != 2) {
1406 unlock_mount_hash();
1409 unlock_mount_hash();
1411 if (!xchg(&mnt->mnt_expiry_mark, 1))
1416 * If we may have to abort operations to get out of this
1417 * mount, and they will themselves hold resources we must
1418 * allow the fs to do things. In the Unix tradition of
1419 * 'Gee thats tricky lets do it in userspace' the umount_begin
1420 * might fail to complete on the first run through as other tasks
1421 * must return, and the like. Thats for the mount program to worry
1422 * about for the moment.
1425 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1426 sb->s_op->umount_begin(sb);
1430 * No sense to grab the lock for this test, but test itself looks
1431 * somewhat bogus. Suggestions for better replacement?
1432 * Ho-hum... In principle, we might treat that as umount + switch
1433 * to rootfs. GC would eventually take care of the old vfsmount.
1434 * Actually it makes sense, especially if rootfs would contain a
1435 * /reboot - static binary that would close all descriptors and
1436 * call reboot(9). Then init(8) could umount root and exec /reboot.
1438 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1440 * Special case for "unmounting" root ...
1441 * we just try to remount it readonly.
1443 if (!capable(CAP_SYS_ADMIN))
1445 down_write(&sb->s_umount);
1446 if (!(sb->s_flags & MS_RDONLY))
1447 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1448 up_write(&sb->s_umount);
1456 if (flags & MNT_DETACH) {
1457 if (!list_empty(&mnt->mnt_list))
1458 umount_tree(mnt, 2);
1461 shrink_submounts(mnt);
1463 if (!propagate_mount_busy(mnt, 2)) {
1464 if (!list_empty(&mnt->mnt_list))
1465 umount_tree(mnt, 1);
1469 unlock_mount_hash();
1475 * __detach_mounts - lazily unmount all mounts on the specified dentry
1477 * During unlink, rmdir, and d_drop it is possible to loose the path
1478 * to an existing mountpoint, and wind up leaking the mount.
1479 * detach_mounts allows lazily unmounting those mounts instead of
1482 * The caller may hold dentry->d_inode->i_mutex.
1484 void __detach_mounts(struct dentry *dentry)
1486 struct mountpoint *mp;
1490 mp = lookup_mountpoint(dentry);
1495 while (!hlist_empty(&mp->m_list)) {
1496 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1497 umount_tree(mnt, 2);
1499 unlock_mount_hash();
1506 * Is the caller allowed to modify his namespace?
1508 static inline bool may_mount(void)
1510 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1514 * Now umount can handle mount points as well as block devices.
1515 * This is important for filesystems which use unnamed block devices.
1517 * We now support a flag for forced unmount like the other 'big iron'
1518 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1521 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1526 int lookup_flags = 0;
1528 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1534 if (!(flags & UMOUNT_NOFOLLOW))
1535 lookup_flags |= LOOKUP_FOLLOW;
1537 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1540 mnt = real_mount(path.mnt);
1542 if (path.dentry != path.mnt->mnt_root)
1544 if (!check_mnt(mnt))
1546 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1549 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1552 retval = do_umount(mnt, flags);
1554 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1556 mntput_no_expire(mnt);
1561 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1564 * The 2.0 compatible umount. No flags.
1566 SYSCALL_DEFINE1(oldumount, char __user *, name)
1568 return sys_umount(name, 0);
1573 static bool is_mnt_ns_file(struct dentry *dentry)
1575 /* Is this a proxy for a mount namespace? */
1576 return dentry->d_op == &ns_dentry_operations &&
1577 dentry->d_fsdata == &mntns_operations;
1580 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1582 return container_of(ns, struct mnt_namespace, ns);
1585 static bool mnt_ns_loop(struct dentry *dentry)
1587 /* Could bind mounting the mount namespace inode cause a
1588 * mount namespace loop?
1590 struct mnt_namespace *mnt_ns;
1591 if (!is_mnt_ns_file(dentry))
1594 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1595 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1598 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1601 struct mount *res, *p, *q, *r, *parent;
1603 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1604 return ERR_PTR(-EINVAL);
1606 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1607 return ERR_PTR(-EINVAL);
1609 res = q = clone_mnt(mnt, dentry, flag);
1613 q->mnt_mountpoint = mnt->mnt_mountpoint;
1616 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1618 if (!is_subdir(r->mnt_mountpoint, dentry))
1621 for (s = r; s; s = next_mnt(s, r)) {
1622 struct mount *t = NULL;
1623 if (!(flag & CL_COPY_UNBINDABLE) &&
1624 IS_MNT_UNBINDABLE(s)) {
1625 s = skip_mnt_tree(s);
1628 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1629 is_mnt_ns_file(s->mnt.mnt_root)) {
1630 s = skip_mnt_tree(s);
1633 while (p != s->mnt_parent) {
1639 q = clone_mnt(p, p->mnt.mnt_root, flag);
1643 list_add_tail(&q->mnt_list, &res->mnt_list);
1644 mnt_set_mountpoint(parent, p->mnt_mp, q);
1645 if (!list_empty(&parent->mnt_mounts)) {
1646 t = list_last_entry(&parent->mnt_mounts,
1647 struct mount, mnt_child);
1648 if (t->mnt_mp != p->mnt_mp)
1651 attach_shadowed(q, parent, t);
1652 unlock_mount_hash();
1659 umount_tree(res, 0);
1660 unlock_mount_hash();
1665 /* Caller should check returned pointer for errors */
1667 struct vfsmount *collect_mounts(struct path *path)
1671 tree = copy_tree(real_mount(path->mnt), path->dentry,
1672 CL_COPY_ALL | CL_PRIVATE);
1675 return ERR_CAST(tree);
1679 void drop_collected_mounts(struct vfsmount *mnt)
1683 umount_tree(real_mount(mnt), 0);
1684 unlock_mount_hash();
1689 * clone_private_mount - create a private clone of a path
1691 * This creates a new vfsmount, which will be the clone of @path. The new will
1692 * not be attached anywhere in the namespace and will be private (i.e. changes
1693 * to the originating mount won't be propagated into this).
1695 * Release with mntput().
1697 struct vfsmount *clone_private_mount(struct path *path)
1699 struct mount *old_mnt = real_mount(path->mnt);
1700 struct mount *new_mnt;
1702 if (IS_MNT_UNBINDABLE(old_mnt))
1703 return ERR_PTR(-EINVAL);
1705 down_read(&namespace_sem);
1706 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1707 up_read(&namespace_sem);
1708 if (IS_ERR(new_mnt))
1709 return ERR_CAST(new_mnt);
1711 return &new_mnt->mnt;
1713 EXPORT_SYMBOL_GPL(clone_private_mount);
1715 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1716 struct vfsmount *root)
1719 int res = f(root, arg);
1722 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1723 res = f(&mnt->mnt, arg);
1730 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1734 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1735 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1736 mnt_release_group_id(p);
1740 static int invent_group_ids(struct mount *mnt, bool recurse)
1744 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1745 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1746 int err = mnt_alloc_group_id(p);
1748 cleanup_group_ids(mnt, p);
1758 * @source_mnt : mount tree to be attached
1759 * @nd : place the mount tree @source_mnt is attached
1760 * @parent_nd : if non-null, detach the source_mnt from its parent and
1761 * store the parent mount and mountpoint dentry.
1762 * (done when source_mnt is moved)
1764 * NOTE: in the table below explains the semantics when a source mount
1765 * of a given type is attached to a destination mount of a given type.
1766 * ---------------------------------------------------------------------------
1767 * | BIND MOUNT OPERATION |
1768 * |**************************************************************************
1769 * | source-->| shared | private | slave | unbindable |
1773 * |**************************************************************************
1774 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1776 * |non-shared| shared (+) | private | slave (*) | invalid |
1777 * ***************************************************************************
1778 * A bind operation clones the source mount and mounts the clone on the
1779 * destination mount.
1781 * (++) the cloned mount is propagated to all the mounts in the propagation
1782 * tree of the destination mount and the cloned mount is added to
1783 * the peer group of the source mount.
1784 * (+) the cloned mount is created under the destination mount and is marked
1785 * as shared. The cloned mount is added to the peer group of the source
1787 * (+++) the mount is propagated to all the mounts in the propagation tree
1788 * of the destination mount and the cloned mount is made slave
1789 * of the same master as that of the source mount. The cloned mount
1790 * is marked as 'shared and slave'.
1791 * (*) the cloned mount is made a slave of the same master as that of the
1794 * ---------------------------------------------------------------------------
1795 * | MOVE MOUNT OPERATION |
1796 * |**************************************************************************
1797 * | source-->| shared | private | slave | unbindable |
1801 * |**************************************************************************
1802 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1804 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1805 * ***************************************************************************
1807 * (+) the mount is moved to the destination. And is then propagated to
1808 * all the mounts in the propagation tree of the destination mount.
1809 * (+*) the mount is moved to the destination.
1810 * (+++) the mount is moved to the destination and is then propagated to
1811 * all the mounts belonging to the destination mount's propagation tree.
1812 * the mount is marked as 'shared and slave'.
1813 * (*) the mount continues to be a slave at the new location.
1815 * if the source mount is a tree, the operations explained above is
1816 * applied to each mount in the tree.
1817 * Must be called without spinlocks held, since this function can sleep
1820 static int attach_recursive_mnt(struct mount *source_mnt,
1821 struct mount *dest_mnt,
1822 struct mountpoint *dest_mp,
1823 struct path *parent_path)
1825 HLIST_HEAD(tree_list);
1826 struct mount *child, *p;
1827 struct hlist_node *n;
1830 if (IS_MNT_SHARED(dest_mnt)) {
1831 err = invent_group_ids(source_mnt, true);
1834 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1837 goto out_cleanup_ids;
1838 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1844 detach_mnt(source_mnt, parent_path);
1845 attach_mnt(source_mnt, dest_mnt, dest_mp);
1846 touch_mnt_namespace(source_mnt->mnt_ns);
1848 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1849 commit_tree(source_mnt, NULL);
1852 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1854 hlist_del_init(&child->mnt_hash);
1855 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1856 child->mnt_mountpoint);
1857 commit_tree(child, q);
1859 unlock_mount_hash();
1864 while (!hlist_empty(&tree_list)) {
1865 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1866 umount_tree(child, 0);
1868 unlock_mount_hash();
1869 cleanup_group_ids(source_mnt, NULL);
1874 static struct mountpoint *lock_mount(struct path *path)
1876 struct vfsmount *mnt;
1877 struct dentry *dentry = path->dentry;
1879 mutex_lock(&dentry->d_inode->i_mutex);
1880 if (unlikely(cant_mount(dentry))) {
1881 mutex_unlock(&dentry->d_inode->i_mutex);
1882 return ERR_PTR(-ENOENT);
1885 mnt = lookup_mnt(path);
1887 struct mountpoint *mp = lookup_mountpoint(dentry);
1889 mp = new_mountpoint(dentry);
1892 mutex_unlock(&dentry->d_inode->i_mutex);
1898 mutex_unlock(&path->dentry->d_inode->i_mutex);
1901 dentry = path->dentry = dget(mnt->mnt_root);
1905 static void unlock_mount(struct mountpoint *where)
1907 struct dentry *dentry = where->m_dentry;
1908 put_mountpoint(where);
1910 mutex_unlock(&dentry->d_inode->i_mutex);
1913 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1915 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1918 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1919 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1922 return attach_recursive_mnt(mnt, p, mp, NULL);
1926 * Sanity check the flags to change_mnt_propagation.
1929 static int flags_to_propagation_type(int flags)
1931 int type = flags & ~(MS_REC | MS_SILENT);
1933 /* Fail if any non-propagation flags are set */
1934 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1936 /* Only one propagation flag should be set */
1937 if (!is_power_of_2(type))
1943 * recursively change the type of the mountpoint.
1945 static int do_change_type(struct path *path, int flag)
1948 struct mount *mnt = real_mount(path->mnt);
1949 int recurse = flag & MS_REC;
1953 if (path->dentry != path->mnt->mnt_root)
1956 type = flags_to_propagation_type(flag);
1961 if (type == MS_SHARED) {
1962 err = invent_group_ids(mnt, recurse);
1968 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1969 change_mnt_propagation(m, type);
1970 unlock_mount_hash();
1977 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1979 struct mount *child;
1980 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1981 if (!is_subdir(child->mnt_mountpoint, dentry))
1984 if (child->mnt.mnt_flags & MNT_LOCKED)
1991 * do loopback mount.
1993 static int do_loopback(struct path *path, const char *old_name,
1996 struct path old_path;
1997 struct mount *mnt = NULL, *old, *parent;
1998 struct mountpoint *mp;
2000 if (!old_name || !*old_name)
2002 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2007 if (mnt_ns_loop(old_path.dentry))
2010 mp = lock_mount(path);
2015 old = real_mount(old_path.mnt);
2016 parent = real_mount(path->mnt);
2019 if (IS_MNT_UNBINDABLE(old))
2022 if (!check_mnt(parent))
2025 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2028 if (!recurse && has_locked_children(old, old_path.dentry))
2032 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2034 mnt = clone_mnt(old, old_path.dentry, 0);
2041 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2043 err = graft_tree(mnt, parent, mp);
2046 umount_tree(mnt, 0);
2047 unlock_mount_hash();
2052 path_put(&old_path);
2056 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2059 int readonly_request = 0;
2061 if (ms_flags & MS_RDONLY)
2062 readonly_request = 1;
2063 if (readonly_request == __mnt_is_readonly(mnt))
2066 if (readonly_request)
2067 error = mnt_make_readonly(real_mount(mnt));
2069 __mnt_unmake_readonly(real_mount(mnt));
2074 * change filesystem flags. dir should be a physical root of filesystem.
2075 * If you've mounted a non-root directory somewhere and want to do remount
2076 * on it - tough luck.
2078 static int do_remount(struct path *path, int flags, int mnt_flags,
2082 struct super_block *sb = path->mnt->mnt_sb;
2083 struct mount *mnt = real_mount(path->mnt);
2085 if (!check_mnt(mnt))
2088 if (path->dentry != path->mnt->mnt_root)
2091 /* Don't allow changing of locked mnt flags.
2093 * No locks need to be held here while testing the various
2094 * MNT_LOCK flags because those flags can never be cleared
2095 * once they are set.
2097 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2098 !(mnt_flags & MNT_READONLY)) {
2101 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2102 !(mnt_flags & MNT_NODEV)) {
2103 /* Was the nodev implicitly added in mount? */
2104 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2105 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2106 mnt_flags |= MNT_NODEV;
2111 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2112 !(mnt_flags & MNT_NOSUID)) {
2115 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2116 !(mnt_flags & MNT_NOEXEC)) {
2119 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2120 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2124 err = security_sb_remount(sb, data);
2128 down_write(&sb->s_umount);
2129 if (flags & MS_BIND)
2130 err = change_mount_flags(path->mnt, flags);
2131 else if (!capable(CAP_SYS_ADMIN))
2134 err = do_remount_sb(sb, flags, data, 0);
2137 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2138 mnt->mnt.mnt_flags = mnt_flags;
2139 touch_mnt_namespace(mnt->mnt_ns);
2140 unlock_mount_hash();
2142 up_write(&sb->s_umount);
2146 static inline int tree_contains_unbindable(struct mount *mnt)
2149 for (p = mnt; p; p = next_mnt(p, mnt)) {
2150 if (IS_MNT_UNBINDABLE(p))
2156 static int do_move_mount(struct path *path, const char *old_name)
2158 struct path old_path, parent_path;
2161 struct mountpoint *mp;
2163 if (!old_name || !*old_name)
2165 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2169 mp = lock_mount(path);
2174 old = real_mount(old_path.mnt);
2175 p = real_mount(path->mnt);
2178 if (!check_mnt(p) || !check_mnt(old))
2181 if (old->mnt.mnt_flags & MNT_LOCKED)
2185 if (old_path.dentry != old_path.mnt->mnt_root)
2188 if (!mnt_has_parent(old))
2191 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2192 S_ISDIR(old_path.dentry->d_inode->i_mode))
2195 * Don't move a mount residing in a shared parent.
2197 if (IS_MNT_SHARED(old->mnt_parent))
2200 * Don't move a mount tree containing unbindable mounts to a destination
2201 * mount which is shared.
2203 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2206 for (; mnt_has_parent(p); p = p->mnt_parent)
2210 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2214 /* if the mount is moved, it should no longer be expire
2216 list_del_init(&old->mnt_expire);
2221 path_put(&parent_path);
2222 path_put(&old_path);
2226 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2229 const char *subtype = strchr(fstype, '.');
2238 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2240 if (!mnt->mnt_sb->s_subtype)
2246 return ERR_PTR(err);
2250 * add a mount into a namespace's mount tree
2252 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2254 struct mountpoint *mp;
2255 struct mount *parent;
2258 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2260 mp = lock_mount(path);
2264 parent = real_mount(path->mnt);
2266 if (unlikely(!check_mnt(parent))) {
2267 /* that's acceptable only for automounts done in private ns */
2268 if (!(mnt_flags & MNT_SHRINKABLE))
2270 /* ... and for those we'd better have mountpoint still alive */
2271 if (!parent->mnt_ns)
2275 /* Refuse the same filesystem on the same mount point */
2277 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2278 path->mnt->mnt_root == path->dentry)
2282 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2285 newmnt->mnt.mnt_flags = mnt_flags;
2286 err = graft_tree(newmnt, parent, mp);
2294 * create a new mount for userspace and request it to be added into the
2297 static int do_new_mount(struct path *path, const char *fstype, int flags,
2298 int mnt_flags, const char *name, void *data)
2300 struct file_system_type *type;
2301 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2302 struct vfsmount *mnt;
2308 type = get_fs_type(fstype);
2312 if (user_ns != &init_user_ns) {
2313 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2314 put_filesystem(type);
2317 /* Only in special cases allow devices from mounts
2318 * created outside the initial user namespace.
2320 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2322 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2326 mnt = vfs_kern_mount(type, flags, name, data);
2327 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2328 !mnt->mnt_sb->s_subtype)
2329 mnt = fs_set_subtype(mnt, fstype);
2331 put_filesystem(type);
2333 return PTR_ERR(mnt);
2335 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2341 int finish_automount(struct vfsmount *m, struct path *path)
2343 struct mount *mnt = real_mount(m);
2345 /* The new mount record should have at least 2 refs to prevent it being
2346 * expired before we get a chance to add it
2348 BUG_ON(mnt_get_count(mnt) < 2);
2350 if (m->mnt_sb == path->mnt->mnt_sb &&
2351 m->mnt_root == path->dentry) {
2356 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2360 /* remove m from any expiration list it may be on */
2361 if (!list_empty(&mnt->mnt_expire)) {
2363 list_del_init(&mnt->mnt_expire);
2372 * mnt_set_expiry - Put a mount on an expiration list
2373 * @mnt: The mount to list.
2374 * @expiry_list: The list to add the mount to.
2376 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2380 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2384 EXPORT_SYMBOL(mnt_set_expiry);
2387 * process a list of expirable mountpoints with the intent of discarding any
2388 * mountpoints that aren't in use and haven't been touched since last we came
2391 void mark_mounts_for_expiry(struct list_head *mounts)
2393 struct mount *mnt, *next;
2394 LIST_HEAD(graveyard);
2396 if (list_empty(mounts))
2402 /* extract from the expiration list every vfsmount that matches the
2403 * following criteria:
2404 * - only referenced by its parent vfsmount
2405 * - still marked for expiry (marked on the last call here; marks are
2406 * cleared by mntput())
2408 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2409 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2410 propagate_mount_busy(mnt, 1))
2412 list_move(&mnt->mnt_expire, &graveyard);
2414 while (!list_empty(&graveyard)) {
2415 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2416 touch_mnt_namespace(mnt->mnt_ns);
2417 umount_tree(mnt, 1);
2419 unlock_mount_hash();
2423 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2426 * Ripoff of 'select_parent()'
2428 * search the list of submounts for a given mountpoint, and move any
2429 * shrinkable submounts to the 'graveyard' list.
2431 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2433 struct mount *this_parent = parent;
2434 struct list_head *next;
2438 next = this_parent->mnt_mounts.next;
2440 while (next != &this_parent->mnt_mounts) {
2441 struct list_head *tmp = next;
2442 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2445 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2448 * Descend a level if the d_mounts list is non-empty.
2450 if (!list_empty(&mnt->mnt_mounts)) {
2455 if (!propagate_mount_busy(mnt, 1)) {
2456 list_move_tail(&mnt->mnt_expire, graveyard);
2461 * All done at this level ... ascend and resume the search
2463 if (this_parent != parent) {
2464 next = this_parent->mnt_child.next;
2465 this_parent = this_parent->mnt_parent;
2472 * process a list of expirable mountpoints with the intent of discarding any
2473 * submounts of a specific parent mountpoint
2475 * mount_lock must be held for write
2477 static void shrink_submounts(struct mount *mnt)
2479 LIST_HEAD(graveyard);
2482 /* extract submounts of 'mountpoint' from the expiration list */
2483 while (select_submounts(mnt, &graveyard)) {
2484 while (!list_empty(&graveyard)) {
2485 m = list_first_entry(&graveyard, struct mount,
2487 touch_mnt_namespace(m->mnt_ns);
2494 * Some copy_from_user() implementations do not return the exact number of
2495 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2496 * Note that this function differs from copy_from_user() in that it will oops
2497 * on bad values of `to', rather than returning a short copy.
2499 static long exact_copy_from_user(void *to, const void __user * from,
2503 const char __user *f = from;
2506 if (!access_ok(VERIFY_READ, from, n))
2510 if (__get_user(c, f)) {
2521 int copy_mount_options(const void __user * data, unsigned long *where)
2531 if (!(page = __get_free_page(GFP_KERNEL)))
2534 /* We only care that *some* data at the address the user
2535 * gave us is valid. Just in case, we'll zero
2536 * the remainder of the page.
2538 /* copy_from_user cannot cross TASK_SIZE ! */
2539 size = TASK_SIZE - (unsigned long)data;
2540 if (size > PAGE_SIZE)
2543 i = size - exact_copy_from_user((void *)page, data, size);
2549 memset((char *)page + i, 0, PAGE_SIZE - i);
2554 char *copy_mount_string(const void __user *data)
2556 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2560 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2561 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2563 * data is a (void *) that can point to any structure up to
2564 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2565 * information (or be NULL).
2567 * Pre-0.97 versions of mount() didn't have a flags word.
2568 * When the flags word was introduced its top half was required
2569 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2570 * Therefore, if this magic number is present, it carries no information
2571 * and must be discarded.
2573 long do_mount(const char *dev_name, const char __user *dir_name,
2574 const char *type_page, unsigned long flags, void *data_page)
2581 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2582 flags &= ~MS_MGC_MSK;
2584 /* Basic sanity checks */
2586 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2588 /* ... and get the mountpoint */
2589 retval = user_path(dir_name, &path);
2593 retval = security_sb_mount(dev_name, &path,
2594 type_page, flags, data_page);
2595 if (!retval && !may_mount())
2600 /* Default to relatime unless overriden */
2601 if (!(flags & MS_NOATIME))
2602 mnt_flags |= MNT_RELATIME;
2604 /* Separate the per-mountpoint flags */
2605 if (flags & MS_NOSUID)
2606 mnt_flags |= MNT_NOSUID;
2607 if (flags & MS_NODEV)
2608 mnt_flags |= MNT_NODEV;
2609 if (flags & MS_NOEXEC)
2610 mnt_flags |= MNT_NOEXEC;
2611 if (flags & MS_NOATIME)
2612 mnt_flags |= MNT_NOATIME;
2613 if (flags & MS_NODIRATIME)
2614 mnt_flags |= MNT_NODIRATIME;
2615 if (flags & MS_STRICTATIME)
2616 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2617 if (flags & MS_RDONLY)
2618 mnt_flags |= MNT_READONLY;
2620 /* The default atime for remount is preservation */
2621 if ((flags & MS_REMOUNT) &&
2622 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2623 MS_STRICTATIME)) == 0)) {
2624 mnt_flags &= ~MNT_ATIME_MASK;
2625 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2628 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2629 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2632 if (flags & MS_REMOUNT)
2633 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2635 else if (flags & MS_BIND)
2636 retval = do_loopback(&path, dev_name, flags & MS_REC);
2637 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2638 retval = do_change_type(&path, flags);
2639 else if (flags & MS_MOVE)
2640 retval = do_move_mount(&path, dev_name);
2642 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2643 dev_name, data_page);
2649 static void free_mnt_ns(struct mnt_namespace *ns)
2651 ns_free_inum(&ns->ns);
2652 put_user_ns(ns->user_ns);
2657 * Assign a sequence number so we can detect when we attempt to bind
2658 * mount a reference to an older mount namespace into the current
2659 * mount namespace, preventing reference counting loops. A 64bit
2660 * number incrementing at 10Ghz will take 12,427 years to wrap which
2661 * is effectively never, so we can ignore the possibility.
2663 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2665 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2667 struct mnt_namespace *new_ns;
2670 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2672 return ERR_PTR(-ENOMEM);
2673 ret = ns_alloc_inum(&new_ns->ns);
2676 return ERR_PTR(ret);
2678 new_ns->ns.ops = &mntns_operations;
2679 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2680 atomic_set(&new_ns->count, 1);
2681 new_ns->root = NULL;
2682 INIT_LIST_HEAD(&new_ns->list);
2683 init_waitqueue_head(&new_ns->poll);
2685 new_ns->user_ns = get_user_ns(user_ns);
2689 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2690 struct user_namespace *user_ns, struct fs_struct *new_fs)
2692 struct mnt_namespace *new_ns;
2693 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2694 struct mount *p, *q;
2701 if (likely(!(flags & CLONE_NEWNS))) {
2708 new_ns = alloc_mnt_ns(user_ns);
2713 /* First pass: copy the tree topology */
2714 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2715 if (user_ns != ns->user_ns)
2716 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2717 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2720 free_mnt_ns(new_ns);
2721 return ERR_CAST(new);
2724 list_add_tail(&new_ns->list, &new->mnt_list);
2727 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2728 * as belonging to new namespace. We have already acquired a private
2729 * fs_struct, so tsk->fs->lock is not needed.
2736 if (&p->mnt == new_fs->root.mnt) {
2737 new_fs->root.mnt = mntget(&q->mnt);
2740 if (&p->mnt == new_fs->pwd.mnt) {
2741 new_fs->pwd.mnt = mntget(&q->mnt);
2745 p = next_mnt(p, old);
2746 q = next_mnt(q, new);
2749 while (p->mnt.mnt_root != q->mnt.mnt_root)
2750 p = next_mnt(p, old);
2763 * create_mnt_ns - creates a private namespace and adds a root filesystem
2764 * @mnt: pointer to the new root filesystem mountpoint
2766 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2768 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2769 if (!IS_ERR(new_ns)) {
2770 struct mount *mnt = real_mount(m);
2771 mnt->mnt_ns = new_ns;
2773 list_add(&mnt->mnt_list, &new_ns->list);
2780 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2782 struct mnt_namespace *ns;
2783 struct super_block *s;
2787 ns = create_mnt_ns(mnt);
2789 return ERR_CAST(ns);
2791 err = vfs_path_lookup(mnt->mnt_root, mnt,
2792 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2797 return ERR_PTR(err);
2799 /* trade a vfsmount reference for active sb one */
2800 s = path.mnt->mnt_sb;
2801 atomic_inc(&s->s_active);
2803 /* lock the sucker */
2804 down_write(&s->s_umount);
2805 /* ... and return the root of (sub)tree on it */
2808 EXPORT_SYMBOL(mount_subtree);
2810 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2811 char __user *, type, unsigned long, flags, void __user *, data)
2816 unsigned long data_page;
2818 kernel_type = copy_mount_string(type);
2819 ret = PTR_ERR(kernel_type);
2820 if (IS_ERR(kernel_type))
2823 kernel_dev = copy_mount_string(dev_name);
2824 ret = PTR_ERR(kernel_dev);
2825 if (IS_ERR(kernel_dev))
2828 ret = copy_mount_options(data, &data_page);
2832 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2833 (void *) data_page);
2835 free_page(data_page);
2845 * Return true if path is reachable from root
2847 * namespace_sem or mount_lock is held
2849 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2850 const struct path *root)
2852 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2853 dentry = mnt->mnt_mountpoint;
2854 mnt = mnt->mnt_parent;
2856 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2859 int path_is_under(struct path *path1, struct path *path2)
2862 read_seqlock_excl(&mount_lock);
2863 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2864 read_sequnlock_excl(&mount_lock);
2867 EXPORT_SYMBOL(path_is_under);
2870 * pivot_root Semantics:
2871 * Moves the root file system of the current process to the directory put_old,
2872 * makes new_root as the new root file system of the current process, and sets
2873 * root/cwd of all processes which had them on the current root to new_root.
2876 * The new_root and put_old must be directories, and must not be on the
2877 * same file system as the current process root. The put_old must be
2878 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2879 * pointed to by put_old must yield the same directory as new_root. No other
2880 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2882 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2883 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2884 * in this situation.
2887 * - we don't move root/cwd if they are not at the root (reason: if something
2888 * cared enough to change them, it's probably wrong to force them elsewhere)
2889 * - it's okay to pick a root that isn't the root of a file system, e.g.
2890 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2891 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2894 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2895 const char __user *, put_old)
2897 struct path new, old, parent_path, root_parent, root;
2898 struct mount *new_mnt, *root_mnt, *old_mnt;
2899 struct mountpoint *old_mp, *root_mp;
2905 error = user_path_dir(new_root, &new);
2909 error = user_path_dir(put_old, &old);
2913 error = security_sb_pivotroot(&old, &new);
2917 get_fs_root(current->fs, &root);
2918 old_mp = lock_mount(&old);
2919 error = PTR_ERR(old_mp);
2924 new_mnt = real_mount(new.mnt);
2925 root_mnt = real_mount(root.mnt);
2926 old_mnt = real_mount(old.mnt);
2927 if (IS_MNT_SHARED(old_mnt) ||
2928 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2929 IS_MNT_SHARED(root_mnt->mnt_parent))
2931 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2933 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2936 if (d_unlinked(new.dentry))
2939 if (new_mnt == root_mnt || old_mnt == root_mnt)
2940 goto out4; /* loop, on the same file system */
2942 if (root.mnt->mnt_root != root.dentry)
2943 goto out4; /* not a mountpoint */
2944 if (!mnt_has_parent(root_mnt))
2945 goto out4; /* not attached */
2946 root_mp = root_mnt->mnt_mp;
2947 if (new.mnt->mnt_root != new.dentry)
2948 goto out4; /* not a mountpoint */
2949 if (!mnt_has_parent(new_mnt))
2950 goto out4; /* not attached */
2951 /* make sure we can reach put_old from new_root */
2952 if (!is_path_reachable(old_mnt, old.dentry, &new))
2954 /* make certain new is below the root */
2955 if (!is_path_reachable(new_mnt, new.dentry, &root))
2957 root_mp->m_count++; /* pin it so it won't go away */
2959 detach_mnt(new_mnt, &parent_path);
2960 detach_mnt(root_mnt, &root_parent);
2961 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2962 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2963 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2965 /* mount old root on put_old */
2966 attach_mnt(root_mnt, old_mnt, old_mp);
2967 /* mount new_root on / */
2968 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2969 touch_mnt_namespace(current->nsproxy->mnt_ns);
2970 /* A moved mount should not expire automatically */
2971 list_del_init(&new_mnt->mnt_expire);
2972 unlock_mount_hash();
2973 chroot_fs_refs(&root, &new);
2974 put_mountpoint(root_mp);
2977 unlock_mount(old_mp);
2979 path_put(&root_parent);
2980 path_put(&parent_path);
2992 static void __init init_mount_tree(void)
2994 struct vfsmount *mnt;
2995 struct mnt_namespace *ns;
2997 struct file_system_type *type;
2999 type = get_fs_type("rootfs");
3001 panic("Can't find rootfs type");
3002 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3003 put_filesystem(type);
3005 panic("Can't create rootfs");
3007 ns = create_mnt_ns(mnt);
3009 panic("Can't allocate initial namespace");
3011 init_task.nsproxy->mnt_ns = ns;
3015 root.dentry = mnt->mnt_root;
3016 mnt->mnt_flags |= MNT_LOCKED;
3018 set_fs_pwd(current->fs, &root);
3019 set_fs_root(current->fs, &root);
3022 void __init mnt_init(void)
3027 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3028 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3030 mount_hashtable = alloc_large_system_hash("Mount-cache",
3031 sizeof(struct hlist_head),
3034 &m_hash_shift, &m_hash_mask, 0, 0);
3035 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3036 sizeof(struct hlist_head),
3039 &mp_hash_shift, &mp_hash_mask, 0, 0);
3041 if (!mount_hashtable || !mountpoint_hashtable)
3042 panic("Failed to allocate mount hash table\n");
3044 for (u = 0; u <= m_hash_mask; u++)
3045 INIT_HLIST_HEAD(&mount_hashtable[u]);
3046 for (u = 0; u <= mp_hash_mask; u++)
3047 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3053 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3055 fs_kobj = kobject_create_and_add("fs", NULL);
3057 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3062 void put_mnt_ns(struct mnt_namespace *ns)
3064 if (!atomic_dec_and_test(&ns->count))
3066 drop_collected_mounts(&ns->root->mnt);
3070 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3072 struct vfsmount *mnt;
3073 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3076 * it is a longterm mount, don't release mnt until
3077 * we unmount before file sys is unregistered
3079 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3083 EXPORT_SYMBOL_GPL(kern_mount_data);
3085 void kern_unmount(struct vfsmount *mnt)
3087 /* release long term mount so mount point can be released */
3088 if (!IS_ERR_OR_NULL(mnt)) {
3089 real_mount(mnt)->mnt_ns = NULL;
3090 synchronize_rcu(); /* yecchhh... */
3094 EXPORT_SYMBOL(kern_unmount);
3096 bool our_mnt(struct vfsmount *mnt)
3098 return check_mnt(real_mount(mnt));
3101 bool current_chrooted(void)
3103 /* Does the current process have a non-standard root */
3104 struct path ns_root;
3105 struct path fs_root;
3108 /* Find the namespace root */
3109 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3110 ns_root.dentry = ns_root.mnt->mnt_root;
3112 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3115 get_fs_root(current->fs, &fs_root);
3117 chrooted = !path_equal(&fs_root, &ns_root);
3125 bool fs_fully_visible(struct file_system_type *type)
3127 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3129 bool visible = false;
3134 down_read(&namespace_sem);
3135 list_for_each_entry(mnt, &ns->list, mnt_list) {
3136 struct mount *child;
3137 if (mnt->mnt.mnt_sb->s_type != type)
3140 /* This mount is not fully visible if there are any child mounts
3141 * that cover anything except for empty directories.
3143 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3144 struct inode *inode = child->mnt_mountpoint->d_inode;
3145 if (!S_ISDIR(inode->i_mode))
3147 if (inode->i_nlink > 2)
3155 up_read(&namespace_sem);
3159 static struct ns_common *mntns_get(struct task_struct *task)
3161 struct ns_common *ns = NULL;
3162 struct nsproxy *nsproxy;
3165 nsproxy = task->nsproxy;
3167 ns = &nsproxy->mnt_ns->ns;
3168 get_mnt_ns(to_mnt_ns(ns));
3175 static void mntns_put(struct ns_common *ns)
3177 put_mnt_ns(to_mnt_ns(ns));
3180 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3182 struct fs_struct *fs = current->fs;
3183 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3186 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3187 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3188 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3195 put_mnt_ns(nsproxy->mnt_ns);
3196 nsproxy->mnt_ns = mnt_ns;
3199 root.mnt = &mnt_ns->root->mnt;
3200 root.dentry = mnt_ns->root->mnt.mnt_root;
3202 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3205 /* Update the pwd and root */
3206 set_fs_pwd(fs, &root);
3207 set_fs_root(fs, &root);
3213 const struct proc_ns_operations mntns_operations = {
3215 .type = CLONE_NEWNS,
3218 .install = mntns_install,