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 *new_mountpoint(struct dentry *dentry)
708 struct hlist_head *chain = mp_hash(dentry);
709 struct mountpoint *mp;
712 hlist_for_each_entry(mp, chain, m_hash) {
713 if (mp->m_dentry == dentry) {
714 /* might be worth a WARN_ON() */
715 if (d_unlinked(dentry))
716 return ERR_PTR(-ENOENT);
722 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
724 return ERR_PTR(-ENOMEM);
726 ret = d_set_mounted(dentry);
732 mp->m_dentry = dentry;
734 hlist_add_head(&mp->m_hash, chain);
735 INIT_HLIST_HEAD(&mp->m_list);
739 static void put_mountpoint(struct mountpoint *mp)
741 if (!--mp->m_count) {
742 struct dentry *dentry = mp->m_dentry;
743 BUG_ON(!hlist_empty(&mp->m_list));
744 spin_lock(&dentry->d_lock);
745 dentry->d_flags &= ~DCACHE_MOUNTED;
746 spin_unlock(&dentry->d_lock);
747 hlist_del(&mp->m_hash);
752 static inline int check_mnt(struct mount *mnt)
754 return mnt->mnt_ns == current->nsproxy->mnt_ns;
758 * vfsmount lock must be held for write
760 static void touch_mnt_namespace(struct mnt_namespace *ns)
764 wake_up_interruptible(&ns->poll);
769 * vfsmount lock must be held for write
771 static void __touch_mnt_namespace(struct mnt_namespace *ns)
773 if (ns && ns->event != event) {
775 wake_up_interruptible(&ns->poll);
780 * vfsmount lock must be held for write
782 static void detach_mnt(struct mount *mnt, struct path *old_path)
784 old_path->dentry = mnt->mnt_mountpoint;
785 old_path->mnt = &mnt->mnt_parent->mnt;
786 mnt->mnt_parent = mnt;
787 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
788 list_del_init(&mnt->mnt_child);
789 hlist_del_init_rcu(&mnt->mnt_hash);
790 hlist_del_init(&mnt->mnt_mp_list);
791 put_mountpoint(mnt->mnt_mp);
796 * vfsmount lock must be held for write
798 void mnt_set_mountpoint(struct mount *mnt,
799 struct mountpoint *mp,
800 struct mount *child_mnt)
803 mnt_add_count(mnt, 1); /* essentially, that's mntget */
804 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
805 child_mnt->mnt_parent = mnt;
806 child_mnt->mnt_mp = mp;
807 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
811 * vfsmount lock must be held for write
813 static void attach_mnt(struct mount *mnt,
814 struct mount *parent,
815 struct mountpoint *mp)
817 mnt_set_mountpoint(parent, mp, mnt);
818 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
819 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
822 static void attach_shadowed(struct mount *mnt,
823 struct mount *parent,
824 struct mount *shadows)
827 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
828 list_add(&mnt->mnt_child, &shadows->mnt_child);
830 hlist_add_head_rcu(&mnt->mnt_hash,
831 m_hash(&parent->mnt, mnt->mnt_mountpoint));
832 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
837 * vfsmount lock must be held for write
839 static void commit_tree(struct mount *mnt, struct mount *shadows)
841 struct mount *parent = mnt->mnt_parent;
844 struct mnt_namespace *n = parent->mnt_ns;
846 BUG_ON(parent == mnt);
848 list_add_tail(&head, &mnt->mnt_list);
849 list_for_each_entry(m, &head, mnt_list)
852 list_splice(&head, n->list.prev);
854 attach_shadowed(mnt, parent, shadows);
855 touch_mnt_namespace(n);
858 static struct mount *next_mnt(struct mount *p, struct mount *root)
860 struct list_head *next = p->mnt_mounts.next;
861 if (next == &p->mnt_mounts) {
865 next = p->mnt_child.next;
866 if (next != &p->mnt_parent->mnt_mounts)
871 return list_entry(next, struct mount, mnt_child);
874 static struct mount *skip_mnt_tree(struct mount *p)
876 struct list_head *prev = p->mnt_mounts.prev;
877 while (prev != &p->mnt_mounts) {
878 p = list_entry(prev, struct mount, mnt_child);
879 prev = p->mnt_mounts.prev;
885 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
891 return ERR_PTR(-ENODEV);
893 mnt = alloc_vfsmnt(name);
895 return ERR_PTR(-ENOMEM);
897 if (flags & MS_KERNMOUNT)
898 mnt->mnt.mnt_flags = MNT_INTERNAL;
900 root = mount_fs(type, flags, name, data);
904 return ERR_CAST(root);
907 mnt->mnt.mnt_root = root;
908 mnt->mnt.mnt_sb = root->d_sb;
909 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
910 mnt->mnt_parent = mnt;
912 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
916 EXPORT_SYMBOL_GPL(vfs_kern_mount);
918 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
921 struct super_block *sb = old->mnt.mnt_sb;
925 mnt = alloc_vfsmnt(old->mnt_devname);
927 return ERR_PTR(-ENOMEM);
929 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
930 mnt->mnt_group_id = 0; /* not a peer of original */
932 mnt->mnt_group_id = old->mnt_group_id;
934 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
935 err = mnt_alloc_group_id(mnt);
940 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
941 /* Don't allow unprivileged users to change mount flags */
942 if (flag & CL_UNPRIVILEGED) {
943 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
945 if (mnt->mnt.mnt_flags & MNT_READONLY)
946 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
948 if (mnt->mnt.mnt_flags & MNT_NODEV)
949 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
951 if (mnt->mnt.mnt_flags & MNT_NOSUID)
952 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
954 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
955 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
958 /* Don't allow unprivileged users to reveal what is under a mount */
959 if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
960 mnt->mnt.mnt_flags |= MNT_LOCKED;
962 atomic_inc(&sb->s_active);
963 mnt->mnt.mnt_sb = sb;
964 mnt->mnt.mnt_root = dget(root);
965 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
966 mnt->mnt_parent = mnt;
968 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
971 if ((flag & CL_SLAVE) ||
972 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
973 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
974 mnt->mnt_master = old;
975 CLEAR_MNT_SHARED(mnt);
976 } else if (!(flag & CL_PRIVATE)) {
977 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
978 list_add(&mnt->mnt_share, &old->mnt_share);
979 if (IS_MNT_SLAVE(old))
980 list_add(&mnt->mnt_slave, &old->mnt_slave);
981 mnt->mnt_master = old->mnt_master;
983 if (flag & CL_MAKE_SHARED)
986 /* stick the duplicate mount on the same expiry list
987 * as the original if that was on one */
988 if (flag & CL_EXPIRE) {
989 if (!list_empty(&old->mnt_expire))
990 list_add(&mnt->mnt_expire, &old->mnt_expire);
1001 static void cleanup_mnt(struct mount *mnt)
1004 * This probably indicates that somebody messed
1005 * up a mnt_want/drop_write() pair. If this
1006 * happens, the filesystem was probably unable
1007 * to make r/w->r/o transitions.
1010 * The locking used to deal with mnt_count decrement provides barriers,
1011 * so mnt_get_writers() below is safe.
1013 WARN_ON(mnt_get_writers(mnt));
1014 if (unlikely(mnt->mnt_pins.first))
1016 fsnotify_vfsmount_delete(&mnt->mnt);
1017 dput(mnt->mnt.mnt_root);
1018 deactivate_super(mnt->mnt.mnt_sb);
1020 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1023 static void __cleanup_mnt(struct rcu_head *head)
1025 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1028 static LLIST_HEAD(delayed_mntput_list);
1029 static void delayed_mntput(struct work_struct *unused)
1031 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1032 struct llist_node *next;
1034 for (; node; node = next) {
1035 next = llist_next(node);
1036 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1039 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1041 static void mntput_no_expire(struct mount *mnt)
1044 mnt_add_count(mnt, -1);
1045 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1050 if (mnt_get_count(mnt)) {
1052 unlock_mount_hash();
1055 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1057 unlock_mount_hash();
1060 mnt->mnt.mnt_flags |= MNT_DOOMED;
1063 list_del(&mnt->mnt_instance);
1064 unlock_mount_hash();
1066 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1067 struct task_struct *task = current;
1068 if (likely(!(task->flags & PF_KTHREAD))) {
1069 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1070 if (!task_work_add(task, &mnt->mnt_rcu, true))
1073 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1074 schedule_delayed_work(&delayed_mntput_work, 1);
1080 void mntput(struct vfsmount *mnt)
1083 struct mount *m = real_mount(mnt);
1084 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1085 if (unlikely(m->mnt_expiry_mark))
1086 m->mnt_expiry_mark = 0;
1087 mntput_no_expire(m);
1090 EXPORT_SYMBOL(mntput);
1092 struct vfsmount *mntget(struct vfsmount *mnt)
1095 mnt_add_count(real_mount(mnt), 1);
1098 EXPORT_SYMBOL(mntget);
1100 struct vfsmount *mnt_clone_internal(struct path *path)
1103 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1106 p->mnt.mnt_flags |= MNT_INTERNAL;
1110 static inline void mangle(struct seq_file *m, const char *s)
1112 seq_escape(m, s, " \t\n\\");
1116 * Simple .show_options callback for filesystems which don't want to
1117 * implement more complex mount option showing.
1119 * See also save_mount_options().
1121 int generic_show_options(struct seq_file *m, struct dentry *root)
1123 const char *options;
1126 options = rcu_dereference(root->d_sb->s_options);
1128 if (options != NULL && options[0]) {
1136 EXPORT_SYMBOL(generic_show_options);
1139 * If filesystem uses generic_show_options(), this function should be
1140 * called from the fill_super() callback.
1142 * The .remount_fs callback usually needs to be handled in a special
1143 * way, to make sure, that previous options are not overwritten if the
1146 * Also note, that if the filesystem's .remount_fs function doesn't
1147 * reset all options to their default value, but changes only newly
1148 * given options, then the displayed options will not reflect reality
1151 void save_mount_options(struct super_block *sb, char *options)
1153 BUG_ON(sb->s_options);
1154 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1156 EXPORT_SYMBOL(save_mount_options);
1158 void replace_mount_options(struct super_block *sb, char *options)
1160 char *old = sb->s_options;
1161 rcu_assign_pointer(sb->s_options, options);
1167 EXPORT_SYMBOL(replace_mount_options);
1169 #ifdef CONFIG_PROC_FS
1170 /* iterator; we want it to have access to namespace_sem, thus here... */
1171 static void *m_start(struct seq_file *m, loff_t *pos)
1173 struct proc_mounts *p = proc_mounts(m);
1175 down_read(&namespace_sem);
1176 if (p->cached_event == p->ns->event) {
1177 void *v = p->cached_mount;
1178 if (*pos == p->cached_index)
1180 if (*pos == p->cached_index + 1) {
1181 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1182 return p->cached_mount = v;
1186 p->cached_event = p->ns->event;
1187 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1188 p->cached_index = *pos;
1189 return p->cached_mount;
1192 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1194 struct proc_mounts *p = proc_mounts(m);
1196 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1197 p->cached_index = *pos;
1198 return p->cached_mount;
1201 static void m_stop(struct seq_file *m, void *v)
1203 up_read(&namespace_sem);
1206 static int m_show(struct seq_file *m, void *v)
1208 struct proc_mounts *p = proc_mounts(m);
1209 struct mount *r = list_entry(v, struct mount, mnt_list);
1210 return p->show(m, &r->mnt);
1213 const struct seq_operations mounts_op = {
1219 #endif /* CONFIG_PROC_FS */
1222 * may_umount_tree - check if a mount tree is busy
1223 * @mnt: root of mount tree
1225 * This is called to check if a tree of mounts has any
1226 * open files, pwds, chroots or sub mounts that are
1229 int may_umount_tree(struct vfsmount *m)
1231 struct mount *mnt = real_mount(m);
1232 int actual_refs = 0;
1233 int minimum_refs = 0;
1237 /* write lock needed for mnt_get_count */
1239 for (p = mnt; p; p = next_mnt(p, mnt)) {
1240 actual_refs += mnt_get_count(p);
1243 unlock_mount_hash();
1245 if (actual_refs > minimum_refs)
1251 EXPORT_SYMBOL(may_umount_tree);
1254 * may_umount - check if a mount point is busy
1255 * @mnt: root of mount
1257 * This is called to check if a mount point has any
1258 * open files, pwds, chroots or sub mounts. If the
1259 * mount has sub mounts this will return busy
1260 * regardless of whether the sub mounts are busy.
1262 * Doesn't take quota and stuff into account. IOW, in some cases it will
1263 * give false negatives. The main reason why it's here is that we need
1264 * a non-destructive way to look for easily umountable filesystems.
1266 int may_umount(struct vfsmount *mnt)
1269 down_read(&namespace_sem);
1271 if (propagate_mount_busy(real_mount(mnt), 2))
1273 unlock_mount_hash();
1274 up_read(&namespace_sem);
1278 EXPORT_SYMBOL(may_umount);
1280 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1282 static void namespace_unlock(void)
1285 struct hlist_head head = unmounted;
1287 if (likely(hlist_empty(&head))) {
1288 up_write(&namespace_sem);
1292 head.first->pprev = &head.first;
1293 INIT_HLIST_HEAD(&unmounted);
1295 /* undo decrements we'd done in umount_tree() */
1296 hlist_for_each_entry(mnt, &head, mnt_hash)
1297 if (mnt->mnt_ex_mountpoint.mnt)
1298 mntget(mnt->mnt_ex_mountpoint.mnt);
1300 up_write(&namespace_sem);
1304 while (!hlist_empty(&head)) {
1305 mnt = hlist_entry(head.first, struct mount, mnt_hash);
1306 hlist_del_init(&mnt->mnt_hash);
1307 if (mnt->mnt_ex_mountpoint.mnt)
1308 path_put(&mnt->mnt_ex_mountpoint);
1313 static inline void namespace_lock(void)
1315 down_write(&namespace_sem);
1319 * mount_lock must be held
1320 * namespace_sem must be held for write
1321 * how = 0 => just this tree, don't propagate
1322 * how = 1 => propagate; we know that nobody else has reference to any victims
1323 * how = 2 => lazy umount
1325 void umount_tree(struct mount *mnt, int how)
1327 HLIST_HEAD(tmp_list);
1329 struct mount *last = NULL;
1331 for (p = mnt; p; p = next_mnt(p, mnt)) {
1332 hlist_del_init_rcu(&p->mnt_hash);
1333 hlist_add_head(&p->mnt_hash, &tmp_list);
1336 hlist_for_each_entry(p, &tmp_list, mnt_hash)
1337 list_del_init(&p->mnt_child);
1340 propagate_umount(&tmp_list);
1342 hlist_for_each_entry(p, &tmp_list, mnt_hash) {
1343 list_del_init(&p->mnt_expire);
1344 list_del_init(&p->mnt_list);
1345 __touch_mnt_namespace(p->mnt_ns);
1348 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1349 if (mnt_has_parent(p)) {
1350 hlist_del_init(&p->mnt_mp_list);
1351 put_mountpoint(p->mnt_mp);
1352 mnt_add_count(p->mnt_parent, -1);
1353 /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1354 p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1355 p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1356 p->mnt_mountpoint = p->mnt.mnt_root;
1360 change_mnt_propagation(p, MS_PRIVATE);
1364 last->mnt_hash.next = unmounted.first;
1365 unmounted.first = tmp_list.first;
1366 unmounted.first->pprev = &unmounted.first;
1370 static void shrink_submounts(struct mount *mnt);
1372 static int do_umount(struct mount *mnt, int flags)
1374 struct super_block *sb = mnt->mnt.mnt_sb;
1377 retval = security_sb_umount(&mnt->mnt, flags);
1382 * Allow userspace to request a mountpoint be expired rather than
1383 * unmounting unconditionally. Unmount only happens if:
1384 * (1) the mark is already set (the mark is cleared by mntput())
1385 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1387 if (flags & MNT_EXPIRE) {
1388 if (&mnt->mnt == current->fs->root.mnt ||
1389 flags & (MNT_FORCE | MNT_DETACH))
1393 * probably don't strictly need the lock here if we examined
1394 * all race cases, but it's a slowpath.
1397 if (mnt_get_count(mnt) != 2) {
1398 unlock_mount_hash();
1401 unlock_mount_hash();
1403 if (!xchg(&mnt->mnt_expiry_mark, 1))
1408 * If we may have to abort operations to get out of this
1409 * mount, and they will themselves hold resources we must
1410 * allow the fs to do things. In the Unix tradition of
1411 * 'Gee thats tricky lets do it in userspace' the umount_begin
1412 * might fail to complete on the first run through as other tasks
1413 * must return, and the like. Thats for the mount program to worry
1414 * about for the moment.
1417 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1418 sb->s_op->umount_begin(sb);
1422 * No sense to grab the lock for this test, but test itself looks
1423 * somewhat bogus. Suggestions for better replacement?
1424 * Ho-hum... In principle, we might treat that as umount + switch
1425 * to rootfs. GC would eventually take care of the old vfsmount.
1426 * Actually it makes sense, especially if rootfs would contain a
1427 * /reboot - static binary that would close all descriptors and
1428 * call reboot(9). Then init(8) could umount root and exec /reboot.
1430 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1432 * Special case for "unmounting" root ...
1433 * we just try to remount it readonly.
1435 down_write(&sb->s_umount);
1436 if (!(sb->s_flags & MS_RDONLY))
1437 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1438 up_write(&sb->s_umount);
1446 if (flags & MNT_DETACH) {
1447 if (!list_empty(&mnt->mnt_list))
1448 umount_tree(mnt, 2);
1451 shrink_submounts(mnt);
1453 if (!propagate_mount_busy(mnt, 2)) {
1454 if (!list_empty(&mnt->mnt_list))
1455 umount_tree(mnt, 1);
1459 unlock_mount_hash();
1465 * Is the caller allowed to modify his namespace?
1467 static inline bool may_mount(void)
1469 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1473 * Now umount can handle mount points as well as block devices.
1474 * This is important for filesystems which use unnamed block devices.
1476 * We now support a flag for forced unmount like the other 'big iron'
1477 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1480 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1485 int lookup_flags = 0;
1487 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1493 if (!(flags & UMOUNT_NOFOLLOW))
1494 lookup_flags |= LOOKUP_FOLLOW;
1496 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1499 mnt = real_mount(path.mnt);
1501 if (path.dentry != path.mnt->mnt_root)
1503 if (!check_mnt(mnt))
1505 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1508 retval = do_umount(mnt, flags);
1510 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1512 mntput_no_expire(mnt);
1517 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1520 * The 2.0 compatible umount. No flags.
1522 SYSCALL_DEFINE1(oldumount, char __user *, name)
1524 return sys_umount(name, 0);
1529 static bool is_mnt_ns_file(struct dentry *dentry)
1531 /* Is this a proxy for a mount namespace? */
1532 struct inode *inode = dentry->d_inode;
1535 if (!proc_ns_inode(inode))
1538 ei = get_proc_ns(inode);
1539 if (ei->ns_ops != &mntns_operations)
1545 static bool mnt_ns_loop(struct dentry *dentry)
1547 /* Could bind mounting the mount namespace inode cause a
1548 * mount namespace loop?
1550 struct mnt_namespace *mnt_ns;
1551 if (!is_mnt_ns_file(dentry))
1554 mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1555 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1558 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1561 struct mount *res, *p, *q, *r, *parent;
1563 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1564 return ERR_PTR(-EINVAL);
1566 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1567 return ERR_PTR(-EINVAL);
1569 res = q = clone_mnt(mnt, dentry, flag);
1573 q->mnt.mnt_flags &= ~MNT_LOCKED;
1574 q->mnt_mountpoint = mnt->mnt_mountpoint;
1577 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1579 if (!is_subdir(r->mnt_mountpoint, dentry))
1582 for (s = r; s; s = next_mnt(s, r)) {
1583 struct mount *t = NULL;
1584 if (!(flag & CL_COPY_UNBINDABLE) &&
1585 IS_MNT_UNBINDABLE(s)) {
1586 s = skip_mnt_tree(s);
1589 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1590 is_mnt_ns_file(s->mnt.mnt_root)) {
1591 s = skip_mnt_tree(s);
1594 while (p != s->mnt_parent) {
1600 q = clone_mnt(p, p->mnt.mnt_root, flag);
1604 list_add_tail(&q->mnt_list, &res->mnt_list);
1605 mnt_set_mountpoint(parent, p->mnt_mp, q);
1606 if (!list_empty(&parent->mnt_mounts)) {
1607 t = list_last_entry(&parent->mnt_mounts,
1608 struct mount, mnt_child);
1609 if (t->mnt_mp != p->mnt_mp)
1612 attach_shadowed(q, parent, t);
1613 unlock_mount_hash();
1620 umount_tree(res, 0);
1621 unlock_mount_hash();
1626 /* Caller should check returned pointer for errors */
1628 struct vfsmount *collect_mounts(struct path *path)
1632 tree = copy_tree(real_mount(path->mnt), path->dentry,
1633 CL_COPY_ALL | CL_PRIVATE);
1636 return ERR_CAST(tree);
1640 void drop_collected_mounts(struct vfsmount *mnt)
1644 umount_tree(real_mount(mnt), 0);
1645 unlock_mount_hash();
1649 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1650 struct vfsmount *root)
1653 int res = f(root, arg);
1656 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1657 res = f(&mnt->mnt, arg);
1664 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1668 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1669 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1670 mnt_release_group_id(p);
1674 static int invent_group_ids(struct mount *mnt, bool recurse)
1678 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1679 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1680 int err = mnt_alloc_group_id(p);
1682 cleanup_group_ids(mnt, p);
1692 * @source_mnt : mount tree to be attached
1693 * @nd : place the mount tree @source_mnt is attached
1694 * @parent_nd : if non-null, detach the source_mnt from its parent and
1695 * store the parent mount and mountpoint dentry.
1696 * (done when source_mnt is moved)
1698 * NOTE: in the table below explains the semantics when a source mount
1699 * of a given type is attached to a destination mount of a given type.
1700 * ---------------------------------------------------------------------------
1701 * | BIND MOUNT OPERATION |
1702 * |**************************************************************************
1703 * | source-->| shared | private | slave | unbindable |
1707 * |**************************************************************************
1708 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1710 * |non-shared| shared (+) | private | slave (*) | invalid |
1711 * ***************************************************************************
1712 * A bind operation clones the source mount and mounts the clone on the
1713 * destination mount.
1715 * (++) the cloned mount is propagated to all the mounts in the propagation
1716 * tree of the destination mount and the cloned mount is added to
1717 * the peer group of the source mount.
1718 * (+) the cloned mount is created under the destination mount and is marked
1719 * as shared. The cloned mount is added to the peer group of the source
1721 * (+++) the mount is propagated to all the mounts in the propagation tree
1722 * of the destination mount and the cloned mount is made slave
1723 * of the same master as that of the source mount. The cloned mount
1724 * is marked as 'shared and slave'.
1725 * (*) the cloned mount is made a slave of the same master as that of the
1728 * ---------------------------------------------------------------------------
1729 * | MOVE MOUNT OPERATION |
1730 * |**************************************************************************
1731 * | source-->| shared | private | slave | unbindable |
1735 * |**************************************************************************
1736 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1738 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1739 * ***************************************************************************
1741 * (+) the mount is moved to the destination. And is then propagated to
1742 * all the mounts in the propagation tree of the destination mount.
1743 * (+*) the mount is moved to the destination.
1744 * (+++) the mount is moved to the destination and is then propagated to
1745 * all the mounts belonging to the destination mount's propagation tree.
1746 * the mount is marked as 'shared and slave'.
1747 * (*) the mount continues to be a slave at the new location.
1749 * if the source mount is a tree, the operations explained above is
1750 * applied to each mount in the tree.
1751 * Must be called without spinlocks held, since this function can sleep
1754 static int attach_recursive_mnt(struct mount *source_mnt,
1755 struct mount *dest_mnt,
1756 struct mountpoint *dest_mp,
1757 struct path *parent_path)
1759 HLIST_HEAD(tree_list);
1760 struct mount *child, *p;
1761 struct hlist_node *n;
1764 if (IS_MNT_SHARED(dest_mnt)) {
1765 err = invent_group_ids(source_mnt, true);
1768 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1771 goto out_cleanup_ids;
1772 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1778 detach_mnt(source_mnt, parent_path);
1779 attach_mnt(source_mnt, dest_mnt, dest_mp);
1780 touch_mnt_namespace(source_mnt->mnt_ns);
1782 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1783 commit_tree(source_mnt, NULL);
1786 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1788 hlist_del_init(&child->mnt_hash);
1789 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1790 child->mnt_mountpoint);
1791 commit_tree(child, q);
1793 unlock_mount_hash();
1798 while (!hlist_empty(&tree_list)) {
1799 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1800 umount_tree(child, 0);
1802 unlock_mount_hash();
1803 cleanup_group_ids(source_mnt, NULL);
1808 static struct mountpoint *lock_mount(struct path *path)
1810 struct vfsmount *mnt;
1811 struct dentry *dentry = path->dentry;
1813 mutex_lock(&dentry->d_inode->i_mutex);
1814 if (unlikely(cant_mount(dentry))) {
1815 mutex_unlock(&dentry->d_inode->i_mutex);
1816 return ERR_PTR(-ENOENT);
1819 mnt = lookup_mnt(path);
1821 struct mountpoint *mp = new_mountpoint(dentry);
1824 mutex_unlock(&dentry->d_inode->i_mutex);
1830 mutex_unlock(&path->dentry->d_inode->i_mutex);
1833 dentry = path->dentry = dget(mnt->mnt_root);
1837 static void unlock_mount(struct mountpoint *where)
1839 struct dentry *dentry = where->m_dentry;
1840 put_mountpoint(where);
1842 mutex_unlock(&dentry->d_inode->i_mutex);
1845 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1847 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1850 if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1851 S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1854 return attach_recursive_mnt(mnt, p, mp, NULL);
1858 * Sanity check the flags to change_mnt_propagation.
1861 static int flags_to_propagation_type(int flags)
1863 int type = flags & ~(MS_REC | MS_SILENT);
1865 /* Fail if any non-propagation flags are set */
1866 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1868 /* Only one propagation flag should be set */
1869 if (!is_power_of_2(type))
1875 * recursively change the type of the mountpoint.
1877 static int do_change_type(struct path *path, int flag)
1880 struct mount *mnt = real_mount(path->mnt);
1881 int recurse = flag & MS_REC;
1885 if (path->dentry != path->mnt->mnt_root)
1888 type = flags_to_propagation_type(flag);
1893 if (type == MS_SHARED) {
1894 err = invent_group_ids(mnt, recurse);
1900 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1901 change_mnt_propagation(m, type);
1902 unlock_mount_hash();
1909 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1911 struct mount *child;
1912 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1913 if (!is_subdir(child->mnt_mountpoint, dentry))
1916 if (child->mnt.mnt_flags & MNT_LOCKED)
1923 * do loopback mount.
1925 static int do_loopback(struct path *path, const char *old_name,
1928 struct path old_path;
1929 struct mount *mnt = NULL, *old, *parent;
1930 struct mountpoint *mp;
1932 if (!old_name || !*old_name)
1934 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1939 if (mnt_ns_loop(old_path.dentry))
1942 mp = lock_mount(path);
1947 old = real_mount(old_path.mnt);
1948 parent = real_mount(path->mnt);
1951 if (IS_MNT_UNBINDABLE(old))
1954 if (!check_mnt(parent) || !check_mnt(old))
1957 if (!recurse && has_locked_children(old, old_path.dentry))
1961 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1963 mnt = clone_mnt(old, old_path.dentry, 0);
1970 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1972 err = graft_tree(mnt, parent, mp);
1975 umount_tree(mnt, 0);
1976 unlock_mount_hash();
1981 path_put(&old_path);
1985 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1988 int readonly_request = 0;
1990 if (ms_flags & MS_RDONLY)
1991 readonly_request = 1;
1992 if (readonly_request == __mnt_is_readonly(mnt))
1995 if (readonly_request)
1996 error = mnt_make_readonly(real_mount(mnt));
1998 __mnt_unmake_readonly(real_mount(mnt));
2003 * change filesystem flags. dir should be a physical root of filesystem.
2004 * If you've mounted a non-root directory somewhere and want to do remount
2005 * on it - tough luck.
2007 static int do_remount(struct path *path, int flags, int mnt_flags,
2011 struct super_block *sb = path->mnt->mnt_sb;
2012 struct mount *mnt = real_mount(path->mnt);
2014 if (!check_mnt(mnt))
2017 if (path->dentry != path->mnt->mnt_root)
2020 /* Don't allow changing of locked mnt flags.
2022 * No locks need to be held here while testing the various
2023 * MNT_LOCK flags because those flags can never be cleared
2024 * once they are set.
2026 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2027 !(mnt_flags & MNT_READONLY)) {
2030 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2031 !(mnt_flags & MNT_NODEV)) {
2034 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2035 !(mnt_flags & MNT_NOSUID)) {
2038 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2039 !(mnt_flags & MNT_NOEXEC)) {
2042 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2043 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2047 err = security_sb_remount(sb, data);
2051 down_write(&sb->s_umount);
2052 if (flags & MS_BIND)
2053 err = change_mount_flags(path->mnt, flags);
2054 else if (!capable(CAP_SYS_ADMIN))
2057 err = do_remount_sb(sb, flags, data, 0);
2060 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2061 mnt->mnt.mnt_flags = mnt_flags;
2062 touch_mnt_namespace(mnt->mnt_ns);
2063 unlock_mount_hash();
2065 up_write(&sb->s_umount);
2069 static inline int tree_contains_unbindable(struct mount *mnt)
2072 for (p = mnt; p; p = next_mnt(p, mnt)) {
2073 if (IS_MNT_UNBINDABLE(p))
2079 static int do_move_mount(struct path *path, const char *old_name)
2081 struct path old_path, parent_path;
2084 struct mountpoint *mp;
2086 if (!old_name || !*old_name)
2088 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2092 mp = lock_mount(path);
2097 old = real_mount(old_path.mnt);
2098 p = real_mount(path->mnt);
2101 if (!check_mnt(p) || !check_mnt(old))
2104 if (old->mnt.mnt_flags & MNT_LOCKED)
2108 if (old_path.dentry != old_path.mnt->mnt_root)
2111 if (!mnt_has_parent(old))
2114 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
2115 S_ISDIR(old_path.dentry->d_inode->i_mode))
2118 * Don't move a mount residing in a shared parent.
2120 if (IS_MNT_SHARED(old->mnt_parent))
2123 * Don't move a mount tree containing unbindable mounts to a destination
2124 * mount which is shared.
2126 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2129 for (; mnt_has_parent(p); p = p->mnt_parent)
2133 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2137 /* if the mount is moved, it should no longer be expire
2139 list_del_init(&old->mnt_expire);
2144 path_put(&parent_path);
2145 path_put(&old_path);
2149 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2152 const char *subtype = strchr(fstype, '.');
2161 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2163 if (!mnt->mnt_sb->s_subtype)
2169 return ERR_PTR(err);
2173 * add a mount into a namespace's mount tree
2175 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2177 struct mountpoint *mp;
2178 struct mount *parent;
2181 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2183 mp = lock_mount(path);
2187 parent = real_mount(path->mnt);
2189 if (unlikely(!check_mnt(parent))) {
2190 /* that's acceptable only for automounts done in private ns */
2191 if (!(mnt_flags & MNT_SHRINKABLE))
2193 /* ... and for those we'd better have mountpoint still alive */
2194 if (!parent->mnt_ns)
2198 /* Refuse the same filesystem on the same mount point */
2200 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2201 path->mnt->mnt_root == path->dentry)
2205 if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2208 newmnt->mnt.mnt_flags = mnt_flags;
2209 err = graft_tree(newmnt, parent, mp);
2217 * create a new mount for userspace and request it to be added into the
2220 static int do_new_mount(struct path *path, const char *fstype, int flags,
2221 int mnt_flags, const char *name, void *data)
2223 struct file_system_type *type;
2224 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2225 struct vfsmount *mnt;
2231 type = get_fs_type(fstype);
2235 if (user_ns != &init_user_ns) {
2236 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2237 put_filesystem(type);
2240 /* Only in special cases allow devices from mounts
2241 * created outside the initial user namespace.
2243 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2245 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2249 mnt = vfs_kern_mount(type, flags, name, data);
2250 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2251 !mnt->mnt_sb->s_subtype)
2252 mnt = fs_set_subtype(mnt, fstype);
2254 put_filesystem(type);
2256 return PTR_ERR(mnt);
2258 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2264 int finish_automount(struct vfsmount *m, struct path *path)
2266 struct mount *mnt = real_mount(m);
2268 /* The new mount record should have at least 2 refs to prevent it being
2269 * expired before we get a chance to add it
2271 BUG_ON(mnt_get_count(mnt) < 2);
2273 if (m->mnt_sb == path->mnt->mnt_sb &&
2274 m->mnt_root == path->dentry) {
2279 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2283 /* remove m from any expiration list it may be on */
2284 if (!list_empty(&mnt->mnt_expire)) {
2286 list_del_init(&mnt->mnt_expire);
2295 * mnt_set_expiry - Put a mount on an expiration list
2296 * @mnt: The mount to list.
2297 * @expiry_list: The list to add the mount to.
2299 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2303 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2307 EXPORT_SYMBOL(mnt_set_expiry);
2310 * process a list of expirable mountpoints with the intent of discarding any
2311 * mountpoints that aren't in use and haven't been touched since last we came
2314 void mark_mounts_for_expiry(struct list_head *mounts)
2316 struct mount *mnt, *next;
2317 LIST_HEAD(graveyard);
2319 if (list_empty(mounts))
2325 /* extract from the expiration list every vfsmount that matches the
2326 * following criteria:
2327 * - only referenced by its parent vfsmount
2328 * - still marked for expiry (marked on the last call here; marks are
2329 * cleared by mntput())
2331 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2332 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2333 propagate_mount_busy(mnt, 1))
2335 list_move(&mnt->mnt_expire, &graveyard);
2337 while (!list_empty(&graveyard)) {
2338 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2339 touch_mnt_namespace(mnt->mnt_ns);
2340 umount_tree(mnt, 1);
2342 unlock_mount_hash();
2346 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2349 * Ripoff of 'select_parent()'
2351 * search the list of submounts for a given mountpoint, and move any
2352 * shrinkable submounts to the 'graveyard' list.
2354 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2356 struct mount *this_parent = parent;
2357 struct list_head *next;
2361 next = this_parent->mnt_mounts.next;
2363 while (next != &this_parent->mnt_mounts) {
2364 struct list_head *tmp = next;
2365 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2368 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2371 * Descend a level if the d_mounts list is non-empty.
2373 if (!list_empty(&mnt->mnt_mounts)) {
2378 if (!propagate_mount_busy(mnt, 1)) {
2379 list_move_tail(&mnt->mnt_expire, graveyard);
2384 * All done at this level ... ascend and resume the search
2386 if (this_parent != parent) {
2387 next = this_parent->mnt_child.next;
2388 this_parent = this_parent->mnt_parent;
2395 * process a list of expirable mountpoints with the intent of discarding any
2396 * submounts of a specific parent mountpoint
2398 * mount_lock must be held for write
2400 static void shrink_submounts(struct mount *mnt)
2402 LIST_HEAD(graveyard);
2405 /* extract submounts of 'mountpoint' from the expiration list */
2406 while (select_submounts(mnt, &graveyard)) {
2407 while (!list_empty(&graveyard)) {
2408 m = list_first_entry(&graveyard, struct mount,
2410 touch_mnt_namespace(m->mnt_ns);
2417 * Some copy_from_user() implementations do not return the exact number of
2418 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2419 * Note that this function differs from copy_from_user() in that it will oops
2420 * on bad values of `to', rather than returning a short copy.
2422 static long exact_copy_from_user(void *to, const void __user * from,
2426 const char __user *f = from;
2429 if (!access_ok(VERIFY_READ, from, n))
2433 if (__get_user(c, f)) {
2444 int copy_mount_options(const void __user * data, unsigned long *where)
2454 if (!(page = __get_free_page(GFP_KERNEL)))
2457 /* We only care that *some* data at the address the user
2458 * gave us is valid. Just in case, we'll zero
2459 * the remainder of the page.
2461 /* copy_from_user cannot cross TASK_SIZE ! */
2462 size = TASK_SIZE - (unsigned long)data;
2463 if (size > PAGE_SIZE)
2466 i = size - exact_copy_from_user((void *)page, data, size);
2472 memset((char *)page + i, 0, PAGE_SIZE - i);
2477 int copy_mount_string(const void __user *data, char **where)
2486 tmp = strndup_user(data, PAGE_SIZE);
2488 return PTR_ERR(tmp);
2495 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2496 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2498 * data is a (void *) that can point to any structure up to
2499 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2500 * information (or be NULL).
2502 * Pre-0.97 versions of mount() didn't have a flags word.
2503 * When the flags word was introduced its top half was required
2504 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2505 * Therefore, if this magic number is present, it carries no information
2506 * and must be discarded.
2508 long do_mount(const char *dev_name, const char *dir_name,
2509 const char *type_page, unsigned long flags, void *data_page)
2516 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2517 flags &= ~MS_MGC_MSK;
2519 /* Basic sanity checks */
2521 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2525 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2527 /* ... and get the mountpoint */
2528 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2532 retval = security_sb_mount(dev_name, &path,
2533 type_page, flags, data_page);
2534 if (!retval && !may_mount())
2539 /* Default to relatime unless overriden */
2540 if (!(flags & MS_NOATIME))
2541 mnt_flags |= MNT_RELATIME;
2543 /* Separate the per-mountpoint flags */
2544 if (flags & MS_NOSUID)
2545 mnt_flags |= MNT_NOSUID;
2546 if (flags & MS_NODEV)
2547 mnt_flags |= MNT_NODEV;
2548 if (flags & MS_NOEXEC)
2549 mnt_flags |= MNT_NOEXEC;
2550 if (flags & MS_NOATIME)
2551 mnt_flags |= MNT_NOATIME;
2552 if (flags & MS_NODIRATIME)
2553 mnt_flags |= MNT_NODIRATIME;
2554 if (flags & MS_STRICTATIME)
2555 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2556 if (flags & MS_RDONLY)
2557 mnt_flags |= MNT_READONLY;
2559 /* The default atime for remount is preservation */
2560 if ((flags & MS_REMOUNT) &&
2561 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2562 MS_STRICTATIME)) == 0)) {
2563 mnt_flags &= ~MNT_ATIME_MASK;
2564 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2567 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2568 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2571 if (flags & MS_REMOUNT)
2572 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2574 else if (flags & MS_BIND)
2575 retval = do_loopback(&path, dev_name, flags & MS_REC);
2576 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2577 retval = do_change_type(&path, flags);
2578 else if (flags & MS_MOVE)
2579 retval = do_move_mount(&path, dev_name);
2581 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2582 dev_name, data_page);
2588 static void free_mnt_ns(struct mnt_namespace *ns)
2590 proc_free_inum(ns->proc_inum);
2591 put_user_ns(ns->user_ns);
2596 * Assign a sequence number so we can detect when we attempt to bind
2597 * mount a reference to an older mount namespace into the current
2598 * mount namespace, preventing reference counting loops. A 64bit
2599 * number incrementing at 10Ghz will take 12,427 years to wrap which
2600 * is effectively never, so we can ignore the possibility.
2602 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2604 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2606 struct mnt_namespace *new_ns;
2609 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2611 return ERR_PTR(-ENOMEM);
2612 ret = proc_alloc_inum(&new_ns->proc_inum);
2615 return ERR_PTR(ret);
2617 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2618 atomic_set(&new_ns->count, 1);
2619 new_ns->root = NULL;
2620 INIT_LIST_HEAD(&new_ns->list);
2621 init_waitqueue_head(&new_ns->poll);
2623 new_ns->user_ns = get_user_ns(user_ns);
2627 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2628 struct user_namespace *user_ns, struct fs_struct *new_fs)
2630 struct mnt_namespace *new_ns;
2631 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2632 struct mount *p, *q;
2639 if (likely(!(flags & CLONE_NEWNS))) {
2646 new_ns = alloc_mnt_ns(user_ns);
2651 /* First pass: copy the tree topology */
2652 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2653 if (user_ns != ns->user_ns)
2654 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2655 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2658 free_mnt_ns(new_ns);
2659 return ERR_CAST(new);
2662 list_add_tail(&new_ns->list, &new->mnt_list);
2665 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2666 * as belonging to new namespace. We have already acquired a private
2667 * fs_struct, so tsk->fs->lock is not needed.
2674 if (&p->mnt == new_fs->root.mnt) {
2675 new_fs->root.mnt = mntget(&q->mnt);
2678 if (&p->mnt == new_fs->pwd.mnt) {
2679 new_fs->pwd.mnt = mntget(&q->mnt);
2683 p = next_mnt(p, old);
2684 q = next_mnt(q, new);
2687 while (p->mnt.mnt_root != q->mnt.mnt_root)
2688 p = next_mnt(p, old);
2701 * create_mnt_ns - creates a private namespace and adds a root filesystem
2702 * @mnt: pointer to the new root filesystem mountpoint
2704 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2706 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2707 if (!IS_ERR(new_ns)) {
2708 struct mount *mnt = real_mount(m);
2709 mnt->mnt_ns = new_ns;
2711 list_add(&mnt->mnt_list, &new_ns->list);
2718 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2720 struct mnt_namespace *ns;
2721 struct super_block *s;
2725 ns = create_mnt_ns(mnt);
2727 return ERR_CAST(ns);
2729 err = vfs_path_lookup(mnt->mnt_root, mnt,
2730 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2735 return ERR_PTR(err);
2737 /* trade a vfsmount reference for active sb one */
2738 s = path.mnt->mnt_sb;
2739 atomic_inc(&s->s_active);
2741 /* lock the sucker */
2742 down_write(&s->s_umount);
2743 /* ... and return the root of (sub)tree on it */
2746 EXPORT_SYMBOL(mount_subtree);
2748 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2749 char __user *, type, unsigned long, flags, void __user *, data)
2753 struct filename *kernel_dir;
2755 unsigned long data_page;
2757 ret = copy_mount_string(type, &kernel_type);
2761 kernel_dir = getname(dir_name);
2762 if (IS_ERR(kernel_dir)) {
2763 ret = PTR_ERR(kernel_dir);
2767 ret = copy_mount_string(dev_name, &kernel_dev);
2771 ret = copy_mount_options(data, &data_page);
2775 ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2776 (void *) data_page);
2778 free_page(data_page);
2782 putname(kernel_dir);
2790 * Return true if path is reachable from root
2792 * namespace_sem or mount_lock is held
2794 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2795 const struct path *root)
2797 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2798 dentry = mnt->mnt_mountpoint;
2799 mnt = mnt->mnt_parent;
2801 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2804 int path_is_under(struct path *path1, struct path *path2)
2807 read_seqlock_excl(&mount_lock);
2808 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2809 read_sequnlock_excl(&mount_lock);
2812 EXPORT_SYMBOL(path_is_under);
2815 * pivot_root Semantics:
2816 * Moves the root file system of the current process to the directory put_old,
2817 * makes new_root as the new root file system of the current process, and sets
2818 * root/cwd of all processes which had them on the current root to new_root.
2821 * The new_root and put_old must be directories, and must not be on the
2822 * same file system as the current process root. The put_old must be
2823 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2824 * pointed to by put_old must yield the same directory as new_root. No other
2825 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2827 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2828 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2829 * in this situation.
2832 * - we don't move root/cwd if they are not at the root (reason: if something
2833 * cared enough to change them, it's probably wrong to force them elsewhere)
2834 * - it's okay to pick a root that isn't the root of a file system, e.g.
2835 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2836 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2839 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2840 const char __user *, put_old)
2842 struct path new, old, parent_path, root_parent, root;
2843 struct mount *new_mnt, *root_mnt, *old_mnt;
2844 struct mountpoint *old_mp, *root_mp;
2850 error = user_path_dir(new_root, &new);
2854 error = user_path_dir(put_old, &old);
2858 error = security_sb_pivotroot(&old, &new);
2862 get_fs_root(current->fs, &root);
2863 old_mp = lock_mount(&old);
2864 error = PTR_ERR(old_mp);
2869 new_mnt = real_mount(new.mnt);
2870 root_mnt = real_mount(root.mnt);
2871 old_mnt = real_mount(old.mnt);
2872 if (IS_MNT_SHARED(old_mnt) ||
2873 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2874 IS_MNT_SHARED(root_mnt->mnt_parent))
2876 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2878 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2881 if (d_unlinked(new.dentry))
2884 if (new_mnt == root_mnt || old_mnt == root_mnt)
2885 goto out4; /* loop, on the same file system */
2887 if (root.mnt->mnt_root != root.dentry)
2888 goto out4; /* not a mountpoint */
2889 if (!mnt_has_parent(root_mnt))
2890 goto out4; /* not attached */
2891 root_mp = root_mnt->mnt_mp;
2892 if (new.mnt->mnt_root != new.dentry)
2893 goto out4; /* not a mountpoint */
2894 if (!mnt_has_parent(new_mnt))
2895 goto out4; /* not attached */
2896 /* make sure we can reach put_old from new_root */
2897 if (!is_path_reachable(old_mnt, old.dentry, &new))
2899 root_mp->m_count++; /* pin it so it won't go away */
2901 detach_mnt(new_mnt, &parent_path);
2902 detach_mnt(root_mnt, &root_parent);
2903 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2904 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2905 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2907 /* mount old root on put_old */
2908 attach_mnt(root_mnt, old_mnt, old_mp);
2909 /* mount new_root on / */
2910 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2911 touch_mnt_namespace(current->nsproxy->mnt_ns);
2912 unlock_mount_hash();
2913 chroot_fs_refs(&root, &new);
2914 put_mountpoint(root_mp);
2917 unlock_mount(old_mp);
2919 path_put(&root_parent);
2920 path_put(&parent_path);
2932 static void __init init_mount_tree(void)
2934 struct vfsmount *mnt;
2935 struct mnt_namespace *ns;
2937 struct file_system_type *type;
2939 type = get_fs_type("rootfs");
2941 panic("Can't find rootfs type");
2942 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2943 put_filesystem(type);
2945 panic("Can't create rootfs");
2947 ns = create_mnt_ns(mnt);
2949 panic("Can't allocate initial namespace");
2951 init_task.nsproxy->mnt_ns = ns;
2955 root.dentry = mnt->mnt_root;
2957 set_fs_pwd(current->fs, &root);
2958 set_fs_root(current->fs, &root);
2961 void __init mnt_init(void)
2966 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2967 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2969 mount_hashtable = alloc_large_system_hash("Mount-cache",
2970 sizeof(struct hlist_head),
2973 &m_hash_shift, &m_hash_mask, 0, 0);
2974 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
2975 sizeof(struct hlist_head),
2978 &mp_hash_shift, &mp_hash_mask, 0, 0);
2980 if (!mount_hashtable || !mountpoint_hashtable)
2981 panic("Failed to allocate mount hash table\n");
2983 for (u = 0; u <= m_hash_mask; u++)
2984 INIT_HLIST_HEAD(&mount_hashtable[u]);
2985 for (u = 0; u <= mp_hash_mask; u++)
2986 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
2992 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2994 fs_kobj = kobject_create_and_add("fs", NULL);
2996 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3001 void put_mnt_ns(struct mnt_namespace *ns)
3003 if (!atomic_dec_and_test(&ns->count))
3005 drop_collected_mounts(&ns->root->mnt);
3009 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3011 struct vfsmount *mnt;
3012 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3015 * it is a longterm mount, don't release mnt until
3016 * we unmount before file sys is unregistered
3018 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3022 EXPORT_SYMBOL_GPL(kern_mount_data);
3024 void kern_unmount(struct vfsmount *mnt)
3026 /* release long term mount so mount point can be released */
3027 if (!IS_ERR_OR_NULL(mnt)) {
3028 real_mount(mnt)->mnt_ns = NULL;
3029 synchronize_rcu(); /* yecchhh... */
3033 EXPORT_SYMBOL(kern_unmount);
3035 bool our_mnt(struct vfsmount *mnt)
3037 return check_mnt(real_mount(mnt));
3040 bool current_chrooted(void)
3042 /* Does the current process have a non-standard root */
3043 struct path ns_root;
3044 struct path fs_root;
3047 /* Find the namespace root */
3048 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3049 ns_root.dentry = ns_root.mnt->mnt_root;
3051 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3054 get_fs_root(current->fs, &fs_root);
3056 chrooted = !path_equal(&fs_root, &ns_root);
3064 bool fs_fully_visible(struct file_system_type *type)
3066 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3068 bool visible = false;
3073 down_read(&namespace_sem);
3074 list_for_each_entry(mnt, &ns->list, mnt_list) {
3075 struct mount *child;
3076 if (mnt->mnt.mnt_sb->s_type != type)
3079 /* This mount is not fully visible if there are any child mounts
3080 * that cover anything except for empty directories.
3082 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3083 struct inode *inode = child->mnt_mountpoint->d_inode;
3084 if (!S_ISDIR(inode->i_mode))
3086 if (inode->i_nlink > 2)
3094 up_read(&namespace_sem);
3098 static void *mntns_get(struct task_struct *task)
3100 struct mnt_namespace *ns = NULL;
3101 struct nsproxy *nsproxy;
3104 nsproxy = task->nsproxy;
3106 ns = nsproxy->mnt_ns;
3114 static void mntns_put(void *ns)
3119 static int mntns_install(struct nsproxy *nsproxy, void *ns)
3121 struct fs_struct *fs = current->fs;
3122 struct mnt_namespace *mnt_ns = ns;
3125 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3126 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3127 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3134 put_mnt_ns(nsproxy->mnt_ns);
3135 nsproxy->mnt_ns = mnt_ns;
3138 root.mnt = &mnt_ns->root->mnt;
3139 root.dentry = mnt_ns->root->mnt.mnt_root;
3141 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3144 /* Update the pwd and root */
3145 set_fs_pwd(fs, &root);
3146 set_fs_root(fs, &root);
3152 static unsigned int mntns_inum(void *ns)
3154 struct mnt_namespace *mnt_ns = ns;
3155 return mnt_ns->proc_inum;
3158 const struct proc_ns_operations mntns_operations = {
3160 .type = CLONE_NEWNS,
3163 .install = mntns_install,