4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/idr.h>
19 #include <linux/init.h> /* init_rootfs */
20 #include <linux/fs_struct.h> /* get_fs_root et.al. */
21 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
22 #include <linux/uaccess.h>
23 #include <linux/proc_ns.h>
24 #include <linux/magic.h>
25 #include <linux/bootmem.h>
26 #include <linux/task_work.h>
30 static unsigned int m_hash_mask __read_mostly;
31 static unsigned int m_hash_shift __read_mostly;
32 static unsigned int mp_hash_mask __read_mostly;
33 static unsigned int mp_hash_shift __read_mostly;
35 static __initdata unsigned long mhash_entries;
36 static int __init set_mhash_entries(char *str)
40 mhash_entries = simple_strtoul(str, &str, 0);
43 __setup("mhash_entries=", set_mhash_entries);
45 static __initdata unsigned long mphash_entries;
46 static int __init set_mphash_entries(char *str)
50 mphash_entries = simple_strtoul(str, &str, 0);
53 __setup("mphash_entries=", set_mphash_entries);
56 static DEFINE_IDA(mnt_id_ida);
57 static DEFINE_IDA(mnt_group_ida);
58 static DEFINE_SPINLOCK(mnt_id_lock);
59 static int mnt_id_start = 0;
60 static int mnt_group_start = 1;
62 static struct hlist_head *mount_hashtable __read_mostly;
63 static struct hlist_head *mountpoint_hashtable __read_mostly;
64 static struct kmem_cache *mnt_cache __read_mostly;
65 static DECLARE_RWSEM(namespace_sem);
68 struct kobject *fs_kobj;
69 EXPORT_SYMBOL_GPL(fs_kobj);
72 * vfsmount lock may be taken for read to prevent changes to the
73 * vfsmount hash, ie. during mountpoint lookups or walking back
76 * It should be taken for write in all cases where the vfsmount
77 * tree or hash is modified or when a vfsmount structure is modified.
79 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
81 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
83 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
84 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
85 tmp = tmp + (tmp >> m_hash_shift);
86 return &mount_hashtable[tmp & m_hash_mask];
89 static inline struct hlist_head *mp_hash(struct dentry *dentry)
91 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
92 tmp = tmp + (tmp >> mp_hash_shift);
93 return &mountpoint_hashtable[tmp & mp_hash_mask];
97 * allocation is serialized by namespace_sem, but we need the spinlock to
98 * serialize with freeing.
100 static int mnt_alloc_id(struct mount *mnt)
105 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
106 spin_lock(&mnt_id_lock);
107 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
109 mnt_id_start = mnt->mnt_id + 1;
110 spin_unlock(&mnt_id_lock);
117 static void mnt_free_id(struct mount *mnt)
119 int id = mnt->mnt_id;
120 spin_lock(&mnt_id_lock);
121 ida_remove(&mnt_id_ida, id);
122 if (mnt_id_start > id)
124 spin_unlock(&mnt_id_lock);
128 * Allocate a new peer group ID
130 * mnt_group_ida is protected by namespace_sem
132 static int mnt_alloc_group_id(struct mount *mnt)
136 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 res = ida_get_new_above(&mnt_group_ida,
143 mnt_group_start = mnt->mnt_group_id + 1;
149 * Release a peer group ID
151 void mnt_release_group_id(struct mount *mnt)
153 int id = mnt->mnt_group_id;
154 ida_remove(&mnt_group_ida, id);
155 if (mnt_group_start > id)
156 mnt_group_start = id;
157 mnt->mnt_group_id = 0;
161 * vfsmount lock must be held for read
163 static inline void mnt_add_count(struct mount *mnt, int n)
166 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
175 * vfsmount lock must be held for write
177 unsigned int mnt_get_count(struct mount *mnt)
180 unsigned int count = 0;
183 for_each_possible_cpu(cpu) {
184 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
189 return mnt->mnt_count;
193 static void drop_mountpoint(struct fs_pin *p)
195 struct mount *m = container_of(p, struct mount, mnt_umount);
196 dput(m->mnt_ex_mountpoint);
201 static struct mount *alloc_vfsmnt(const char *name)
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 err = mnt_alloc_id(mnt);
212 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
213 if (!mnt->mnt_devname)
218 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
220 goto out_free_devname;
222 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225 mnt->mnt_writers = 0;
228 INIT_HLIST_NODE(&mnt->mnt_hash);
229 INIT_LIST_HEAD(&mnt->mnt_child);
230 INIT_LIST_HEAD(&mnt->mnt_mounts);
231 INIT_LIST_HEAD(&mnt->mnt_list);
232 INIT_LIST_HEAD(&mnt->mnt_expire);
233 INIT_LIST_HEAD(&mnt->mnt_share);
234 INIT_LIST_HEAD(&mnt->mnt_slave_list);
235 INIT_LIST_HEAD(&mnt->mnt_slave);
236 INIT_HLIST_NODE(&mnt->mnt_mp_list);
237 #ifdef CONFIG_FSNOTIFY
238 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
246 kfree_const(mnt->mnt_devname);
251 kmem_cache_free(mnt_cache, mnt);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount *mnt)
276 if (mnt->mnt_flags & MNT_READONLY)
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
284 static inline void mnt_inc_writers(struct mount *mnt)
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
293 static inline void mnt_dec_writers(struct mount *mnt)
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
302 static unsigned int mnt_get_writers(struct mount *mnt)
305 unsigned int count = 0;
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
314 return mnt->mnt_writers;
318 static int mnt_is_readonly(struct vfsmount *mnt)
320 if (mnt->mnt_sb->s_readonly_remount)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount *m)
345 struct mount *mnt = real_mount(m);
349 mnt_inc_writers(mnt);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount *m)
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
389 sb_end_write(m->mnt_sb);
392 EXPORT_SYMBOL_GPL(mnt_want_write);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount *mnt)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
412 mnt_inc_writers(real_mount(mnt));
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file *file)
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
430 return mnt_clone_write(file->f_path.mnt);
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 int mnt_want_write_file(struct file *file)
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
447 sb_end_write(file->f_path.mnt->mnt_sb);
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
460 void __mnt_drop_write(struct vfsmount *mnt)
463 mnt_dec_writers(real_mount(mnt));
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
475 void mnt_drop_write(struct vfsmount *mnt)
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
482 void __mnt_drop_write_file(struct file *file)
484 __mnt_drop_write(file->f_path.mnt);
487 void mnt_drop_write_file(struct file *file)
489 mnt_drop_write(file->f_path.mnt);
491 EXPORT_SYMBOL(mnt_drop_write_file);
493 static int mnt_make_readonly(struct mount *mnt)
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
521 if (mnt_get_writers(mnt) > 0)
524 mnt->mnt.mnt_flags |= MNT_READONLY;
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
535 static void __mnt_unmake_readonly(struct mount *mnt)
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
542 int sb_prepare_remount_readonly(struct super_block *sb)
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
556 if (mnt_get_writers(mnt) > 0) {
562 if (!err && atomic_long_read(&sb->s_remove_count))
566 sb->s_readonly_remount = 1;
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
578 static void free_vfsmnt(struct mount *mnt)
580 kfree_const(mnt->mnt_devname);
582 free_percpu(mnt->mnt_pcp);
584 kmem_cache_free(mnt_cache, mnt);
587 static void delayed_free_vfsmnt(struct rcu_head *head)
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
592 /* call under rcu_read_lock */
593 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
596 if (read_seqretry(&mount_lock, seq))
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
615 * find the first mount at @dentry on vfsmount @mnt.
616 * call under rcu_read_lock()
618 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
620 struct hlist_head *head = m_hash(mnt, dentry);
623 hlist_for_each_entry_rcu(p, head, mnt_hash)
624 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
630 * find the last mount at @dentry on vfsmount @mnt.
631 * mount_lock must be held.
633 struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
635 struct mount *p, *res = NULL;
636 p = __lookup_mnt(mnt, dentry);
639 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
641 hlist_for_each_entry_continue(p, mnt_hash) {
642 if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry)
644 if (!(p->mnt.mnt_flags & MNT_UMOUNT))
652 * lookup_mnt - Return the first child mount mounted at path
654 * "First" means first mounted chronologically. If you create the
657 * mount /dev/sda1 /mnt
658 * mount /dev/sda2 /mnt
659 * mount /dev/sda3 /mnt
661 * Then lookup_mnt() on the base /mnt dentry in the root mount will
662 * return successively the root dentry and vfsmount of /dev/sda1, then
663 * /dev/sda2, then /dev/sda3, then NULL.
665 * lookup_mnt takes a reference to the found vfsmount.
667 struct vfsmount *lookup_mnt(struct path *path)
669 struct mount *child_mnt;
675 seq = read_seqbegin(&mount_lock);
676 child_mnt = __lookup_mnt(path->mnt, path->dentry);
677 m = child_mnt ? &child_mnt->mnt : NULL;
678 } while (!legitimize_mnt(m, seq));
684 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
685 * current mount namespace.
687 * The common case is dentries are not mountpoints at all and that
688 * test is handled inline. For the slow case when we are actually
689 * dealing with a mountpoint of some kind, walk through all of the
690 * mounts in the current mount namespace and test to see if the dentry
693 * The mount_hashtable is not usable in the context because we
694 * need to identify all mounts that may be in the current mount
695 * namespace not just a mount that happens to have some specified
698 bool __is_local_mountpoint(struct dentry *dentry)
700 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
702 bool is_covered = false;
704 if (!d_mountpoint(dentry))
707 down_read(&namespace_sem);
708 list_for_each_entry(mnt, &ns->list, mnt_list) {
709 is_covered = (mnt->mnt_mountpoint == dentry);
713 up_read(&namespace_sem);
718 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
720 struct hlist_head *chain = mp_hash(dentry);
721 struct mountpoint *mp;
723 hlist_for_each_entry(mp, chain, m_hash) {
724 if (mp->m_dentry == dentry) {
725 /* might be worth a WARN_ON() */
726 if (d_unlinked(dentry))
727 return ERR_PTR(-ENOENT);
735 static struct mountpoint *new_mountpoint(struct dentry *dentry)
737 struct hlist_head *chain = mp_hash(dentry);
738 struct mountpoint *mp;
741 mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
743 return ERR_PTR(-ENOMEM);
745 ret = d_set_mounted(dentry);
751 mp->m_dentry = dentry;
753 hlist_add_head(&mp->m_hash, chain);
754 INIT_HLIST_HEAD(&mp->m_list);
758 static void put_mountpoint(struct mountpoint *mp)
760 if (!--mp->m_count) {
761 struct dentry *dentry = mp->m_dentry;
762 BUG_ON(!hlist_empty(&mp->m_list));
763 spin_lock(&dentry->d_lock);
764 dentry->d_flags &= ~DCACHE_MOUNTED;
765 spin_unlock(&dentry->d_lock);
766 hlist_del(&mp->m_hash);
771 static inline int check_mnt(struct mount *mnt)
773 return mnt->mnt_ns == current->nsproxy->mnt_ns;
777 * vfsmount lock must be held for write
779 static void touch_mnt_namespace(struct mnt_namespace *ns)
783 wake_up_interruptible(&ns->poll);
788 * vfsmount lock must be held for write
790 static void __touch_mnt_namespace(struct mnt_namespace *ns)
792 if (ns && ns->event != event) {
794 wake_up_interruptible(&ns->poll);
799 * vfsmount lock must be held for write
801 static void unhash_mnt(struct mount *mnt)
803 mnt->mnt_parent = mnt;
804 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
805 list_del_init(&mnt->mnt_child);
806 hlist_del_init_rcu(&mnt->mnt_hash);
807 hlist_del_init(&mnt->mnt_mp_list);
808 put_mountpoint(mnt->mnt_mp);
813 * vfsmount lock must be held for write
815 static void detach_mnt(struct mount *mnt, struct path *old_path)
817 old_path->dentry = mnt->mnt_mountpoint;
818 old_path->mnt = &mnt->mnt_parent->mnt;
823 * vfsmount lock must be held for write
825 static void umount_mnt(struct mount *mnt)
827 /* old mountpoint will be dropped when we can do that */
828 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
833 * vfsmount lock must be held for write
835 void mnt_set_mountpoint(struct mount *mnt,
836 struct mountpoint *mp,
837 struct mount *child_mnt)
840 mnt_add_count(mnt, 1); /* essentially, that's mntget */
841 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
842 child_mnt->mnt_parent = mnt;
843 child_mnt->mnt_mp = mp;
844 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
848 * vfsmount lock must be held for write
850 static void attach_mnt(struct mount *mnt,
851 struct mount *parent,
852 struct mountpoint *mp)
854 mnt_set_mountpoint(parent, mp, mnt);
855 hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry));
856 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
859 static void attach_shadowed(struct mount *mnt,
860 struct mount *parent,
861 struct mount *shadows)
864 hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash);
865 list_add(&mnt->mnt_child, &shadows->mnt_child);
867 hlist_add_head_rcu(&mnt->mnt_hash,
868 m_hash(&parent->mnt, mnt->mnt_mountpoint));
869 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
874 * vfsmount lock must be held for write
876 static void commit_tree(struct mount *mnt, struct mount *shadows)
878 struct mount *parent = mnt->mnt_parent;
881 struct mnt_namespace *n = parent->mnt_ns;
883 BUG_ON(parent == mnt);
885 list_add_tail(&head, &mnt->mnt_list);
886 list_for_each_entry(m, &head, mnt_list)
889 list_splice(&head, n->list.prev);
891 attach_shadowed(mnt, parent, shadows);
892 touch_mnt_namespace(n);
895 static struct mount *next_mnt(struct mount *p, struct mount *root)
897 struct list_head *next = p->mnt_mounts.next;
898 if (next == &p->mnt_mounts) {
902 next = p->mnt_child.next;
903 if (next != &p->mnt_parent->mnt_mounts)
908 return list_entry(next, struct mount, mnt_child);
911 static struct mount *skip_mnt_tree(struct mount *p)
913 struct list_head *prev = p->mnt_mounts.prev;
914 while (prev != &p->mnt_mounts) {
915 p = list_entry(prev, struct mount, mnt_child);
916 prev = p->mnt_mounts.prev;
922 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
928 return ERR_PTR(-ENODEV);
930 mnt = alloc_vfsmnt(name);
932 return ERR_PTR(-ENOMEM);
934 if (flags & MS_KERNMOUNT)
935 mnt->mnt.mnt_flags = MNT_INTERNAL;
937 root = mount_fs(type, flags, name, data);
941 return ERR_CAST(root);
944 mnt->mnt.mnt_root = root;
945 mnt->mnt.mnt_sb = root->d_sb;
946 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
947 mnt->mnt_parent = mnt;
949 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
953 EXPORT_SYMBOL_GPL(vfs_kern_mount);
955 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
958 struct super_block *sb = old->mnt.mnt_sb;
962 mnt = alloc_vfsmnt(old->mnt_devname);
964 return ERR_PTR(-ENOMEM);
966 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
967 mnt->mnt_group_id = 0; /* not a peer of original */
969 mnt->mnt_group_id = old->mnt_group_id;
971 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
972 err = mnt_alloc_group_id(mnt);
977 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
978 /* Don't allow unprivileged users to change mount flags */
979 if (flag & CL_UNPRIVILEGED) {
980 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
982 if (mnt->mnt.mnt_flags & MNT_READONLY)
983 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
985 if (mnt->mnt.mnt_flags & MNT_NODEV)
986 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
988 if (mnt->mnt.mnt_flags & MNT_NOSUID)
989 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
991 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
992 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
995 /* Don't allow unprivileged users to reveal what is under a mount */
996 if ((flag & CL_UNPRIVILEGED) &&
997 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
998 mnt->mnt.mnt_flags |= MNT_LOCKED;
1000 atomic_inc(&sb->s_active);
1001 mnt->mnt.mnt_sb = sb;
1002 mnt->mnt.mnt_root = dget(root);
1003 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1004 mnt->mnt_parent = mnt;
1006 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1007 unlock_mount_hash();
1009 if ((flag & CL_SLAVE) ||
1010 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1011 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1012 mnt->mnt_master = old;
1013 CLEAR_MNT_SHARED(mnt);
1014 } else if (!(flag & CL_PRIVATE)) {
1015 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1016 list_add(&mnt->mnt_share, &old->mnt_share);
1017 if (IS_MNT_SLAVE(old))
1018 list_add(&mnt->mnt_slave, &old->mnt_slave);
1019 mnt->mnt_master = old->mnt_master;
1021 if (flag & CL_MAKE_SHARED)
1022 set_mnt_shared(mnt);
1024 /* stick the duplicate mount on the same expiry list
1025 * as the original if that was on one */
1026 if (flag & CL_EXPIRE) {
1027 if (!list_empty(&old->mnt_expire))
1028 list_add(&mnt->mnt_expire, &old->mnt_expire);
1036 return ERR_PTR(err);
1039 static void cleanup_mnt(struct mount *mnt)
1042 * This probably indicates that somebody messed
1043 * up a mnt_want/drop_write() pair. If this
1044 * happens, the filesystem was probably unable
1045 * to make r/w->r/o transitions.
1048 * The locking used to deal with mnt_count decrement provides barriers,
1049 * so mnt_get_writers() below is safe.
1051 WARN_ON(mnt_get_writers(mnt));
1052 if (unlikely(mnt->mnt_pins.first))
1054 fsnotify_vfsmount_delete(&mnt->mnt);
1055 dput(mnt->mnt.mnt_root);
1056 deactivate_super(mnt->mnt.mnt_sb);
1058 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1061 static void __cleanup_mnt(struct rcu_head *head)
1063 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1066 static LLIST_HEAD(delayed_mntput_list);
1067 static void delayed_mntput(struct work_struct *unused)
1069 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1070 struct llist_node *next;
1072 for (; node; node = next) {
1073 next = llist_next(node);
1074 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1077 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1079 static void mntput_no_expire(struct mount *mnt)
1082 mnt_add_count(mnt, -1);
1083 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1088 if (mnt_get_count(mnt)) {
1090 unlock_mount_hash();
1093 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1095 unlock_mount_hash();
1098 mnt->mnt.mnt_flags |= MNT_DOOMED;
1101 list_del(&mnt->mnt_instance);
1103 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1104 struct mount *p, *tmp;
1105 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1109 unlock_mount_hash();
1111 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1112 struct task_struct *task = current;
1113 if (likely(!(task->flags & PF_KTHREAD))) {
1114 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1115 if (!task_work_add(task, &mnt->mnt_rcu, true))
1118 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1119 schedule_delayed_work(&delayed_mntput_work, 1);
1125 void mntput(struct vfsmount *mnt)
1128 struct mount *m = real_mount(mnt);
1129 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1130 if (unlikely(m->mnt_expiry_mark))
1131 m->mnt_expiry_mark = 0;
1132 mntput_no_expire(m);
1135 EXPORT_SYMBOL(mntput);
1137 struct vfsmount *mntget(struct vfsmount *mnt)
1140 mnt_add_count(real_mount(mnt), 1);
1143 EXPORT_SYMBOL(mntget);
1145 struct vfsmount *mnt_clone_internal(struct path *path)
1148 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1151 p->mnt.mnt_flags |= MNT_INTERNAL;
1155 static inline void mangle(struct seq_file *m, const char *s)
1157 seq_escape(m, s, " \t\n\\");
1161 * Simple .show_options callback for filesystems which don't want to
1162 * implement more complex mount option showing.
1164 * See also save_mount_options().
1166 int generic_show_options(struct seq_file *m, struct dentry *root)
1168 const char *options;
1171 options = rcu_dereference(root->d_sb->s_options);
1173 if (options != NULL && options[0]) {
1181 EXPORT_SYMBOL(generic_show_options);
1184 * If filesystem uses generic_show_options(), this function should be
1185 * called from the fill_super() callback.
1187 * The .remount_fs callback usually needs to be handled in a special
1188 * way, to make sure, that previous options are not overwritten if the
1191 * Also note, that if the filesystem's .remount_fs function doesn't
1192 * reset all options to their default value, but changes only newly
1193 * given options, then the displayed options will not reflect reality
1196 void save_mount_options(struct super_block *sb, char *options)
1198 BUG_ON(sb->s_options);
1199 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1201 EXPORT_SYMBOL(save_mount_options);
1203 void replace_mount_options(struct super_block *sb, char *options)
1205 char *old = sb->s_options;
1206 rcu_assign_pointer(sb->s_options, options);
1212 EXPORT_SYMBOL(replace_mount_options);
1214 #ifdef CONFIG_PROC_FS
1215 /* iterator; we want it to have access to namespace_sem, thus here... */
1216 static void *m_start(struct seq_file *m, loff_t *pos)
1218 struct proc_mounts *p = proc_mounts(m);
1220 down_read(&namespace_sem);
1221 if (p->cached_event == p->ns->event) {
1222 void *v = p->cached_mount;
1223 if (*pos == p->cached_index)
1225 if (*pos == p->cached_index + 1) {
1226 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1227 return p->cached_mount = v;
1231 p->cached_event = p->ns->event;
1232 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1233 p->cached_index = *pos;
1234 return p->cached_mount;
1237 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1239 struct proc_mounts *p = proc_mounts(m);
1241 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1242 p->cached_index = *pos;
1243 return p->cached_mount;
1246 static void m_stop(struct seq_file *m, void *v)
1248 up_read(&namespace_sem);
1251 static int m_show(struct seq_file *m, void *v)
1253 struct proc_mounts *p = proc_mounts(m);
1254 struct mount *r = list_entry(v, struct mount, mnt_list);
1255 return p->show(m, &r->mnt);
1258 const struct seq_operations mounts_op = {
1264 #endif /* CONFIG_PROC_FS */
1267 * may_umount_tree - check if a mount tree is busy
1268 * @mnt: root of mount tree
1270 * This is called to check if a tree of mounts has any
1271 * open files, pwds, chroots or sub mounts that are
1274 int may_umount_tree(struct vfsmount *m)
1276 struct mount *mnt = real_mount(m);
1277 int actual_refs = 0;
1278 int minimum_refs = 0;
1282 /* write lock needed for mnt_get_count */
1284 for (p = mnt; p; p = next_mnt(p, mnt)) {
1285 actual_refs += mnt_get_count(p);
1288 unlock_mount_hash();
1290 if (actual_refs > minimum_refs)
1296 EXPORT_SYMBOL(may_umount_tree);
1299 * may_umount - check if a mount point is busy
1300 * @mnt: root of mount
1302 * This is called to check if a mount point has any
1303 * open files, pwds, chroots or sub mounts. If the
1304 * mount has sub mounts this will return busy
1305 * regardless of whether the sub mounts are busy.
1307 * Doesn't take quota and stuff into account. IOW, in some cases it will
1308 * give false negatives. The main reason why it's here is that we need
1309 * a non-destructive way to look for easily umountable filesystems.
1311 int may_umount(struct vfsmount *mnt)
1314 down_read(&namespace_sem);
1316 if (propagate_mount_busy(real_mount(mnt), 2))
1318 unlock_mount_hash();
1319 up_read(&namespace_sem);
1323 EXPORT_SYMBOL(may_umount);
1325 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1327 static void namespace_unlock(void)
1329 struct hlist_head head = unmounted;
1331 if (likely(hlist_empty(&head))) {
1332 up_write(&namespace_sem);
1336 head.first->pprev = &head.first;
1337 INIT_HLIST_HEAD(&unmounted);
1338 up_write(&namespace_sem);
1342 group_pin_kill(&head);
1345 static inline void namespace_lock(void)
1347 down_write(&namespace_sem);
1350 enum umount_tree_flags {
1352 UMOUNT_PROPAGATE = 2,
1353 UMOUNT_CONNECTED = 4,
1356 * mount_lock must be held
1357 * namespace_sem must be held for write
1359 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1361 LIST_HEAD(tmp_list);
1364 if (how & UMOUNT_PROPAGATE)
1365 propagate_mount_unlock(mnt);
1367 /* Gather the mounts to umount */
1368 for (p = mnt; p; p = next_mnt(p, mnt)) {
1369 p->mnt.mnt_flags |= MNT_UMOUNT;
1370 list_move(&p->mnt_list, &tmp_list);
1373 /* Hide the mounts from mnt_mounts */
1374 list_for_each_entry(p, &tmp_list, mnt_list) {
1375 list_del_init(&p->mnt_child);
1378 /* Add propogated mounts to the tmp_list */
1379 if (how & UMOUNT_PROPAGATE)
1380 propagate_umount(&tmp_list);
1382 while (!list_empty(&tmp_list)) {
1384 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1385 list_del_init(&p->mnt_expire);
1386 list_del_init(&p->mnt_list);
1387 __touch_mnt_namespace(p->mnt_ns);
1389 if (how & UMOUNT_SYNC)
1390 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1392 disconnect = !(((how & UMOUNT_CONNECTED) &&
1393 mnt_has_parent(p) &&
1394 (p->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) ||
1395 IS_MNT_LOCKED_AND_LAZY(p));
1397 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1398 disconnect ? &unmounted : NULL);
1399 if (mnt_has_parent(p)) {
1400 mnt_add_count(p->mnt_parent, -1);
1402 /* Don't forget about p */
1403 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1408 change_mnt_propagation(p, MS_PRIVATE);
1412 static void shrink_submounts(struct mount *mnt);
1414 static int do_umount(struct mount *mnt, int flags)
1416 struct super_block *sb = mnt->mnt.mnt_sb;
1419 retval = security_sb_umount(&mnt->mnt, flags);
1424 * Allow userspace to request a mountpoint be expired rather than
1425 * unmounting unconditionally. Unmount only happens if:
1426 * (1) the mark is already set (the mark is cleared by mntput())
1427 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1429 if (flags & MNT_EXPIRE) {
1430 if (&mnt->mnt == current->fs->root.mnt ||
1431 flags & (MNT_FORCE | MNT_DETACH))
1435 * probably don't strictly need the lock here if we examined
1436 * all race cases, but it's a slowpath.
1439 if (mnt_get_count(mnt) != 2) {
1440 unlock_mount_hash();
1443 unlock_mount_hash();
1445 if (!xchg(&mnt->mnt_expiry_mark, 1))
1450 * If we may have to abort operations to get out of this
1451 * mount, and they will themselves hold resources we must
1452 * allow the fs to do things. In the Unix tradition of
1453 * 'Gee thats tricky lets do it in userspace' the umount_begin
1454 * might fail to complete on the first run through as other tasks
1455 * must return, and the like. Thats for the mount program to worry
1456 * about for the moment.
1459 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1460 sb->s_op->umount_begin(sb);
1464 * No sense to grab the lock for this test, but test itself looks
1465 * somewhat bogus. Suggestions for better replacement?
1466 * Ho-hum... In principle, we might treat that as umount + switch
1467 * to rootfs. GC would eventually take care of the old vfsmount.
1468 * Actually it makes sense, especially if rootfs would contain a
1469 * /reboot - static binary that would close all descriptors and
1470 * call reboot(9). Then init(8) could umount root and exec /reboot.
1472 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1474 * Special case for "unmounting" root ...
1475 * we just try to remount it readonly.
1477 if (!capable(CAP_SYS_ADMIN))
1479 down_write(&sb->s_umount);
1480 if (!(sb->s_flags & MS_RDONLY))
1481 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1482 up_write(&sb->s_umount);
1490 if (flags & MNT_DETACH) {
1491 if (!list_empty(&mnt->mnt_list))
1492 umount_tree(mnt, UMOUNT_PROPAGATE);
1495 shrink_submounts(mnt);
1497 if (!propagate_mount_busy(mnt, 2)) {
1498 if (!list_empty(&mnt->mnt_list))
1499 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1503 unlock_mount_hash();
1509 * __detach_mounts - lazily unmount all mounts on the specified dentry
1511 * During unlink, rmdir, and d_drop it is possible to loose the path
1512 * to an existing mountpoint, and wind up leaking the mount.
1513 * detach_mounts allows lazily unmounting those mounts instead of
1516 * The caller may hold dentry->d_inode->i_mutex.
1518 void __detach_mounts(struct dentry *dentry)
1520 struct mountpoint *mp;
1524 mp = lookup_mountpoint(dentry);
1525 if (IS_ERR_OR_NULL(mp))
1529 while (!hlist_empty(&mp->m_list)) {
1530 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1531 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1532 struct mount *p, *tmp;
1533 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1534 hlist_add_head(&p->mnt_umount.s_list, &unmounted);
1538 else umount_tree(mnt, UMOUNT_CONNECTED);
1540 unlock_mount_hash();
1547 * Is the caller allowed to modify his namespace?
1549 static inline bool may_mount(void)
1551 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1555 * Now umount can handle mount points as well as block devices.
1556 * This is important for filesystems which use unnamed block devices.
1558 * We now support a flag for forced unmount like the other 'big iron'
1559 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1562 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1567 int lookup_flags = 0;
1569 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1575 if (!(flags & UMOUNT_NOFOLLOW))
1576 lookup_flags |= LOOKUP_FOLLOW;
1578 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1581 mnt = real_mount(path.mnt);
1583 if (path.dentry != path.mnt->mnt_root)
1585 if (!check_mnt(mnt))
1587 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1590 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1593 retval = do_umount(mnt, flags);
1595 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1597 mntput_no_expire(mnt);
1602 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1605 * The 2.0 compatible umount. No flags.
1607 SYSCALL_DEFINE1(oldumount, char __user *, name)
1609 return sys_umount(name, 0);
1614 static bool is_mnt_ns_file(struct dentry *dentry)
1616 /* Is this a proxy for a mount namespace? */
1617 return dentry->d_op == &ns_dentry_operations &&
1618 dentry->d_fsdata == &mntns_operations;
1621 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1623 return container_of(ns, struct mnt_namespace, ns);
1626 static bool mnt_ns_loop(struct dentry *dentry)
1628 /* Could bind mounting the mount namespace inode cause a
1629 * mount namespace loop?
1631 struct mnt_namespace *mnt_ns;
1632 if (!is_mnt_ns_file(dentry))
1635 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1636 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1639 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1642 struct mount *res, *p, *q, *r, *parent;
1644 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1645 return ERR_PTR(-EINVAL);
1647 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1648 return ERR_PTR(-EINVAL);
1650 res = q = clone_mnt(mnt, dentry, flag);
1654 q->mnt_mountpoint = mnt->mnt_mountpoint;
1657 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1659 if (!is_subdir(r->mnt_mountpoint, dentry))
1662 for (s = r; s; s = next_mnt(s, r)) {
1663 struct mount *t = NULL;
1664 if (!(flag & CL_COPY_UNBINDABLE) &&
1665 IS_MNT_UNBINDABLE(s)) {
1666 s = skip_mnt_tree(s);
1669 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1670 is_mnt_ns_file(s->mnt.mnt_root)) {
1671 s = skip_mnt_tree(s);
1674 while (p != s->mnt_parent) {
1680 q = clone_mnt(p, p->mnt.mnt_root, flag);
1684 list_add_tail(&q->mnt_list, &res->mnt_list);
1685 mnt_set_mountpoint(parent, p->mnt_mp, q);
1686 if (!list_empty(&parent->mnt_mounts)) {
1687 t = list_last_entry(&parent->mnt_mounts,
1688 struct mount, mnt_child);
1689 if (t->mnt_mp != p->mnt_mp)
1692 attach_shadowed(q, parent, t);
1693 unlock_mount_hash();
1700 umount_tree(res, UMOUNT_SYNC);
1701 unlock_mount_hash();
1706 /* Caller should check returned pointer for errors */
1708 struct vfsmount *collect_mounts(struct path *path)
1712 if (!check_mnt(real_mount(path->mnt)))
1713 tree = ERR_PTR(-EINVAL);
1715 tree = copy_tree(real_mount(path->mnt), path->dentry,
1716 CL_COPY_ALL | CL_PRIVATE);
1719 return ERR_CAST(tree);
1723 void drop_collected_mounts(struct vfsmount *mnt)
1727 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1728 unlock_mount_hash();
1733 * clone_private_mount - create a private clone of a path
1735 * This creates a new vfsmount, which will be the clone of @path. The new will
1736 * not be attached anywhere in the namespace and will be private (i.e. changes
1737 * to the originating mount won't be propagated into this).
1739 * Release with mntput().
1741 struct vfsmount *clone_private_mount(struct path *path)
1743 struct mount *old_mnt = real_mount(path->mnt);
1744 struct mount *new_mnt;
1746 if (IS_MNT_UNBINDABLE(old_mnt))
1747 return ERR_PTR(-EINVAL);
1749 down_read(&namespace_sem);
1750 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1751 up_read(&namespace_sem);
1752 if (IS_ERR(new_mnt))
1753 return ERR_CAST(new_mnt);
1755 return &new_mnt->mnt;
1757 EXPORT_SYMBOL_GPL(clone_private_mount);
1759 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1760 struct vfsmount *root)
1763 int res = f(root, arg);
1766 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1767 res = f(&mnt->mnt, arg);
1774 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1778 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1779 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1780 mnt_release_group_id(p);
1784 static int invent_group_ids(struct mount *mnt, bool recurse)
1788 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1789 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1790 int err = mnt_alloc_group_id(p);
1792 cleanup_group_ids(mnt, p);
1802 * @source_mnt : mount tree to be attached
1803 * @nd : place the mount tree @source_mnt is attached
1804 * @parent_nd : if non-null, detach the source_mnt from its parent and
1805 * store the parent mount and mountpoint dentry.
1806 * (done when source_mnt is moved)
1808 * NOTE: in the table below explains the semantics when a source mount
1809 * of a given type is attached to a destination mount of a given type.
1810 * ---------------------------------------------------------------------------
1811 * | BIND MOUNT OPERATION |
1812 * |**************************************************************************
1813 * | source-->| shared | private | slave | unbindable |
1817 * |**************************************************************************
1818 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1820 * |non-shared| shared (+) | private | slave (*) | invalid |
1821 * ***************************************************************************
1822 * A bind operation clones the source mount and mounts the clone on the
1823 * destination mount.
1825 * (++) the cloned mount is propagated to all the mounts in the propagation
1826 * tree of the destination mount and the cloned mount is added to
1827 * the peer group of the source mount.
1828 * (+) the cloned mount is created under the destination mount and is marked
1829 * as shared. The cloned mount is added to the peer group of the source
1831 * (+++) the mount is propagated to all the mounts in the propagation tree
1832 * of the destination mount and the cloned mount is made slave
1833 * of the same master as that of the source mount. The cloned mount
1834 * is marked as 'shared and slave'.
1835 * (*) the cloned mount is made a slave of the same master as that of the
1838 * ---------------------------------------------------------------------------
1839 * | MOVE MOUNT OPERATION |
1840 * |**************************************************************************
1841 * | source-->| shared | private | slave | unbindable |
1845 * |**************************************************************************
1846 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1848 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1849 * ***************************************************************************
1851 * (+) the mount is moved to the destination. And is then propagated to
1852 * all the mounts in the propagation tree of the destination mount.
1853 * (+*) the mount is moved to the destination.
1854 * (+++) the mount is moved to the destination and is then propagated to
1855 * all the mounts belonging to the destination mount's propagation tree.
1856 * the mount is marked as 'shared and slave'.
1857 * (*) the mount continues to be a slave at the new location.
1859 * if the source mount is a tree, the operations explained above is
1860 * applied to each mount in the tree.
1861 * Must be called without spinlocks held, since this function can sleep
1864 static int attach_recursive_mnt(struct mount *source_mnt,
1865 struct mount *dest_mnt,
1866 struct mountpoint *dest_mp,
1867 struct path *parent_path)
1869 HLIST_HEAD(tree_list);
1870 struct mount *child, *p;
1871 struct hlist_node *n;
1874 if (IS_MNT_SHARED(dest_mnt)) {
1875 err = invent_group_ids(source_mnt, true);
1878 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1881 goto out_cleanup_ids;
1882 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1888 detach_mnt(source_mnt, parent_path);
1889 attach_mnt(source_mnt, dest_mnt, dest_mp);
1890 touch_mnt_namespace(source_mnt->mnt_ns);
1892 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1893 commit_tree(source_mnt, NULL);
1896 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
1898 hlist_del_init(&child->mnt_hash);
1899 q = __lookup_mnt_last(&child->mnt_parent->mnt,
1900 child->mnt_mountpoint);
1901 commit_tree(child, q);
1903 unlock_mount_hash();
1908 while (!hlist_empty(&tree_list)) {
1909 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
1910 umount_tree(child, UMOUNT_SYNC);
1912 unlock_mount_hash();
1913 cleanup_group_ids(source_mnt, NULL);
1918 static struct mountpoint *lock_mount(struct path *path)
1920 struct vfsmount *mnt;
1921 struct dentry *dentry = path->dentry;
1923 mutex_lock(&dentry->d_inode->i_mutex);
1924 if (unlikely(cant_mount(dentry))) {
1925 mutex_unlock(&dentry->d_inode->i_mutex);
1926 return ERR_PTR(-ENOENT);
1929 mnt = lookup_mnt(path);
1931 struct mountpoint *mp = lookup_mountpoint(dentry);
1933 mp = new_mountpoint(dentry);
1936 mutex_unlock(&dentry->d_inode->i_mutex);
1942 mutex_unlock(&path->dentry->d_inode->i_mutex);
1945 dentry = path->dentry = dget(mnt->mnt_root);
1949 static void unlock_mount(struct mountpoint *where)
1951 struct dentry *dentry = where->m_dentry;
1952 put_mountpoint(where);
1954 mutex_unlock(&dentry->d_inode->i_mutex);
1957 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1959 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1962 if (d_is_dir(mp->m_dentry) !=
1963 d_is_dir(mnt->mnt.mnt_root))
1966 return attach_recursive_mnt(mnt, p, mp, NULL);
1970 * Sanity check the flags to change_mnt_propagation.
1973 static int flags_to_propagation_type(int flags)
1975 int type = flags & ~(MS_REC | MS_SILENT);
1977 /* Fail if any non-propagation flags are set */
1978 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1980 /* Only one propagation flag should be set */
1981 if (!is_power_of_2(type))
1987 * recursively change the type of the mountpoint.
1989 static int do_change_type(struct path *path, int flag)
1992 struct mount *mnt = real_mount(path->mnt);
1993 int recurse = flag & MS_REC;
1997 if (path->dentry != path->mnt->mnt_root)
2000 type = flags_to_propagation_type(flag);
2005 if (type == MS_SHARED) {
2006 err = invent_group_ids(mnt, recurse);
2012 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2013 change_mnt_propagation(m, type);
2014 unlock_mount_hash();
2021 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2023 struct mount *child;
2024 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2025 if (!is_subdir(child->mnt_mountpoint, dentry))
2028 if (child->mnt.mnt_flags & MNT_LOCKED)
2035 * do loopback mount.
2037 static int do_loopback(struct path *path, const char *old_name,
2040 struct path old_path;
2041 struct mount *mnt = NULL, *old, *parent;
2042 struct mountpoint *mp;
2044 if (!old_name || !*old_name)
2046 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2051 if (mnt_ns_loop(old_path.dentry))
2054 mp = lock_mount(path);
2059 old = real_mount(old_path.mnt);
2060 parent = real_mount(path->mnt);
2063 if (IS_MNT_UNBINDABLE(old))
2066 if (!check_mnt(parent))
2069 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2072 if (!recurse && has_locked_children(old, old_path.dentry))
2076 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2078 mnt = clone_mnt(old, old_path.dentry, 0);
2085 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2087 err = graft_tree(mnt, parent, mp);
2090 umount_tree(mnt, UMOUNT_SYNC);
2091 unlock_mount_hash();
2096 path_put(&old_path);
2100 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2103 int readonly_request = 0;
2105 if (ms_flags & MS_RDONLY)
2106 readonly_request = 1;
2107 if (readonly_request == __mnt_is_readonly(mnt))
2110 if (readonly_request)
2111 error = mnt_make_readonly(real_mount(mnt));
2113 __mnt_unmake_readonly(real_mount(mnt));
2118 * change filesystem flags. dir should be a physical root of filesystem.
2119 * If you've mounted a non-root directory somewhere and want to do remount
2120 * on it - tough luck.
2122 static int do_remount(struct path *path, int flags, int mnt_flags,
2126 struct super_block *sb = path->mnt->mnt_sb;
2127 struct mount *mnt = real_mount(path->mnt);
2129 if (!check_mnt(mnt))
2132 if (path->dentry != path->mnt->mnt_root)
2135 /* Don't allow changing of locked mnt flags.
2137 * No locks need to be held here while testing the various
2138 * MNT_LOCK flags because those flags can never be cleared
2139 * once they are set.
2141 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2142 !(mnt_flags & MNT_READONLY)) {
2145 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2146 !(mnt_flags & MNT_NODEV)) {
2147 /* Was the nodev implicitly added in mount? */
2148 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2149 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2150 mnt_flags |= MNT_NODEV;
2155 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2156 !(mnt_flags & MNT_NOSUID)) {
2159 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2160 !(mnt_flags & MNT_NOEXEC)) {
2163 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2164 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2168 err = security_sb_remount(sb, data);
2172 down_write(&sb->s_umount);
2173 if (flags & MS_BIND)
2174 err = change_mount_flags(path->mnt, flags);
2175 else if (!capable(CAP_SYS_ADMIN))
2178 err = do_remount_sb(sb, flags, data, 0);
2181 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2182 mnt->mnt.mnt_flags = mnt_flags;
2183 touch_mnt_namespace(mnt->mnt_ns);
2184 unlock_mount_hash();
2186 up_write(&sb->s_umount);
2190 static inline int tree_contains_unbindable(struct mount *mnt)
2193 for (p = mnt; p; p = next_mnt(p, mnt)) {
2194 if (IS_MNT_UNBINDABLE(p))
2200 static int do_move_mount(struct path *path, const char *old_name)
2202 struct path old_path, parent_path;
2205 struct mountpoint *mp;
2207 if (!old_name || !*old_name)
2209 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2213 mp = lock_mount(path);
2218 old = real_mount(old_path.mnt);
2219 p = real_mount(path->mnt);
2222 if (!check_mnt(p) || !check_mnt(old))
2225 if (old->mnt.mnt_flags & MNT_LOCKED)
2229 if (old_path.dentry != old_path.mnt->mnt_root)
2232 if (!mnt_has_parent(old))
2235 if (d_is_dir(path->dentry) !=
2236 d_is_dir(old_path.dentry))
2239 * Don't move a mount residing in a shared parent.
2241 if (IS_MNT_SHARED(old->mnt_parent))
2244 * Don't move a mount tree containing unbindable mounts to a destination
2245 * mount which is shared.
2247 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2250 for (; mnt_has_parent(p); p = p->mnt_parent)
2254 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2258 /* if the mount is moved, it should no longer be expire
2260 list_del_init(&old->mnt_expire);
2265 path_put(&parent_path);
2266 path_put(&old_path);
2270 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2273 const char *subtype = strchr(fstype, '.');
2282 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2284 if (!mnt->mnt_sb->s_subtype)
2290 return ERR_PTR(err);
2294 * add a mount into a namespace's mount tree
2296 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2298 struct mountpoint *mp;
2299 struct mount *parent;
2302 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2304 mp = lock_mount(path);
2308 parent = real_mount(path->mnt);
2310 if (unlikely(!check_mnt(parent))) {
2311 /* that's acceptable only for automounts done in private ns */
2312 if (!(mnt_flags & MNT_SHRINKABLE))
2314 /* ... and for those we'd better have mountpoint still alive */
2315 if (!parent->mnt_ns)
2319 /* Refuse the same filesystem on the same mount point */
2321 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2322 path->mnt->mnt_root == path->dentry)
2326 if (d_is_symlink(newmnt->mnt.mnt_root))
2329 newmnt->mnt.mnt_flags = mnt_flags;
2330 err = graft_tree(newmnt, parent, mp);
2337 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2340 * create a new mount for userspace and request it to be added into the
2343 static int do_new_mount(struct path *path, const char *fstype, int flags,
2344 int mnt_flags, const char *name, void *data)
2346 struct file_system_type *type;
2347 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2348 struct vfsmount *mnt;
2354 type = get_fs_type(fstype);
2358 if (user_ns != &init_user_ns) {
2359 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2360 put_filesystem(type);
2363 /* Only in special cases allow devices from mounts
2364 * created outside the initial user namespace.
2366 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2368 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2370 if (type->fs_flags & FS_USERNS_VISIBLE) {
2371 if (!fs_fully_visible(type, &mnt_flags))
2376 mnt = vfs_kern_mount(type, flags, name, data);
2377 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2378 !mnt->mnt_sb->s_subtype)
2379 mnt = fs_set_subtype(mnt, fstype);
2381 put_filesystem(type);
2383 return PTR_ERR(mnt);
2385 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2391 int finish_automount(struct vfsmount *m, struct path *path)
2393 struct mount *mnt = real_mount(m);
2395 /* The new mount record should have at least 2 refs to prevent it being
2396 * expired before we get a chance to add it
2398 BUG_ON(mnt_get_count(mnt) < 2);
2400 if (m->mnt_sb == path->mnt->mnt_sb &&
2401 m->mnt_root == path->dentry) {
2406 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2410 /* remove m from any expiration list it may be on */
2411 if (!list_empty(&mnt->mnt_expire)) {
2413 list_del_init(&mnt->mnt_expire);
2422 * mnt_set_expiry - Put a mount on an expiration list
2423 * @mnt: The mount to list.
2424 * @expiry_list: The list to add the mount to.
2426 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2430 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2434 EXPORT_SYMBOL(mnt_set_expiry);
2437 * process a list of expirable mountpoints with the intent of discarding any
2438 * mountpoints that aren't in use and haven't been touched since last we came
2441 void mark_mounts_for_expiry(struct list_head *mounts)
2443 struct mount *mnt, *next;
2444 LIST_HEAD(graveyard);
2446 if (list_empty(mounts))
2452 /* extract from the expiration list every vfsmount that matches the
2453 * following criteria:
2454 * - only referenced by its parent vfsmount
2455 * - still marked for expiry (marked on the last call here; marks are
2456 * cleared by mntput())
2458 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2459 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2460 propagate_mount_busy(mnt, 1))
2462 list_move(&mnt->mnt_expire, &graveyard);
2464 while (!list_empty(&graveyard)) {
2465 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2466 touch_mnt_namespace(mnt->mnt_ns);
2467 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2469 unlock_mount_hash();
2473 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2476 * Ripoff of 'select_parent()'
2478 * search the list of submounts for a given mountpoint, and move any
2479 * shrinkable submounts to the 'graveyard' list.
2481 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2483 struct mount *this_parent = parent;
2484 struct list_head *next;
2488 next = this_parent->mnt_mounts.next;
2490 while (next != &this_parent->mnt_mounts) {
2491 struct list_head *tmp = next;
2492 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2495 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2498 * Descend a level if the d_mounts list is non-empty.
2500 if (!list_empty(&mnt->mnt_mounts)) {
2505 if (!propagate_mount_busy(mnt, 1)) {
2506 list_move_tail(&mnt->mnt_expire, graveyard);
2511 * All done at this level ... ascend and resume the search
2513 if (this_parent != parent) {
2514 next = this_parent->mnt_child.next;
2515 this_parent = this_parent->mnt_parent;
2522 * process a list of expirable mountpoints with the intent of discarding any
2523 * submounts of a specific parent mountpoint
2525 * mount_lock must be held for write
2527 static void shrink_submounts(struct mount *mnt)
2529 LIST_HEAD(graveyard);
2532 /* extract submounts of 'mountpoint' from the expiration list */
2533 while (select_submounts(mnt, &graveyard)) {
2534 while (!list_empty(&graveyard)) {
2535 m = list_first_entry(&graveyard, struct mount,
2537 touch_mnt_namespace(m->mnt_ns);
2538 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2544 * Some copy_from_user() implementations do not return the exact number of
2545 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2546 * Note that this function differs from copy_from_user() in that it will oops
2547 * on bad values of `to', rather than returning a short copy.
2549 static long exact_copy_from_user(void *to, const void __user * from,
2553 const char __user *f = from;
2556 if (!access_ok(VERIFY_READ, from, n))
2560 if (__get_user(c, f)) {
2571 int copy_mount_options(const void __user * data, unsigned long *where)
2581 if (!(page = __get_free_page(GFP_KERNEL)))
2584 /* We only care that *some* data at the address the user
2585 * gave us is valid. Just in case, we'll zero
2586 * the remainder of the page.
2588 /* copy_from_user cannot cross TASK_SIZE ! */
2589 size = TASK_SIZE - (unsigned long)data;
2590 if (size > PAGE_SIZE)
2593 i = size - exact_copy_from_user((void *)page, data, size);
2599 memset((char *)page + i, 0, PAGE_SIZE - i);
2604 char *copy_mount_string(const void __user *data)
2606 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2610 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2611 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2613 * data is a (void *) that can point to any structure up to
2614 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2615 * information (or be NULL).
2617 * Pre-0.97 versions of mount() didn't have a flags word.
2618 * When the flags word was introduced its top half was required
2619 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2620 * Therefore, if this magic number is present, it carries no information
2621 * and must be discarded.
2623 long do_mount(const char *dev_name, const char __user *dir_name,
2624 const char *type_page, unsigned long flags, void *data_page)
2631 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2632 flags &= ~MS_MGC_MSK;
2634 /* Basic sanity checks */
2636 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2638 /* ... and get the mountpoint */
2639 retval = user_path(dir_name, &path);
2643 retval = security_sb_mount(dev_name, &path,
2644 type_page, flags, data_page);
2645 if (!retval && !may_mount())
2650 /* Default to relatime unless overriden */
2651 if (!(flags & MS_NOATIME))
2652 mnt_flags |= MNT_RELATIME;
2654 /* Separate the per-mountpoint flags */
2655 if (flags & MS_NOSUID)
2656 mnt_flags |= MNT_NOSUID;
2657 if (flags & MS_NODEV)
2658 mnt_flags |= MNT_NODEV;
2659 if (flags & MS_NOEXEC)
2660 mnt_flags |= MNT_NOEXEC;
2661 if (flags & MS_NOATIME)
2662 mnt_flags |= MNT_NOATIME;
2663 if (flags & MS_NODIRATIME)
2664 mnt_flags |= MNT_NODIRATIME;
2665 if (flags & MS_STRICTATIME)
2666 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2667 if (flags & MS_RDONLY)
2668 mnt_flags |= MNT_READONLY;
2670 /* The default atime for remount is preservation */
2671 if ((flags & MS_REMOUNT) &&
2672 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2673 MS_STRICTATIME)) == 0)) {
2674 mnt_flags &= ~MNT_ATIME_MASK;
2675 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2678 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2679 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2682 if (flags & MS_REMOUNT)
2683 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2685 else if (flags & MS_BIND)
2686 retval = do_loopback(&path, dev_name, flags & MS_REC);
2687 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2688 retval = do_change_type(&path, flags);
2689 else if (flags & MS_MOVE)
2690 retval = do_move_mount(&path, dev_name);
2692 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2693 dev_name, data_page);
2699 static void free_mnt_ns(struct mnt_namespace *ns)
2701 ns_free_inum(&ns->ns);
2702 put_user_ns(ns->user_ns);
2707 * Assign a sequence number so we can detect when we attempt to bind
2708 * mount a reference to an older mount namespace into the current
2709 * mount namespace, preventing reference counting loops. A 64bit
2710 * number incrementing at 10Ghz will take 12,427 years to wrap which
2711 * is effectively never, so we can ignore the possibility.
2713 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2715 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2717 struct mnt_namespace *new_ns;
2720 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2722 return ERR_PTR(-ENOMEM);
2723 ret = ns_alloc_inum(&new_ns->ns);
2726 return ERR_PTR(ret);
2728 new_ns->ns.ops = &mntns_operations;
2729 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2730 atomic_set(&new_ns->count, 1);
2731 new_ns->root = NULL;
2732 INIT_LIST_HEAD(&new_ns->list);
2733 init_waitqueue_head(&new_ns->poll);
2735 new_ns->user_ns = get_user_ns(user_ns);
2739 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2740 struct user_namespace *user_ns, struct fs_struct *new_fs)
2742 struct mnt_namespace *new_ns;
2743 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2744 struct mount *p, *q;
2751 if (likely(!(flags & CLONE_NEWNS))) {
2758 new_ns = alloc_mnt_ns(user_ns);
2763 /* First pass: copy the tree topology */
2764 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2765 if (user_ns != ns->user_ns)
2766 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2767 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2770 free_mnt_ns(new_ns);
2771 return ERR_CAST(new);
2774 list_add_tail(&new_ns->list, &new->mnt_list);
2777 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2778 * as belonging to new namespace. We have already acquired a private
2779 * fs_struct, so tsk->fs->lock is not needed.
2786 if (&p->mnt == new_fs->root.mnt) {
2787 new_fs->root.mnt = mntget(&q->mnt);
2790 if (&p->mnt == new_fs->pwd.mnt) {
2791 new_fs->pwd.mnt = mntget(&q->mnt);
2795 p = next_mnt(p, old);
2796 q = next_mnt(q, new);
2799 while (p->mnt.mnt_root != q->mnt.mnt_root)
2800 p = next_mnt(p, old);
2813 * create_mnt_ns - creates a private namespace and adds a root filesystem
2814 * @mnt: pointer to the new root filesystem mountpoint
2816 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2818 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2819 if (!IS_ERR(new_ns)) {
2820 struct mount *mnt = real_mount(m);
2821 mnt->mnt_ns = new_ns;
2823 list_add(&mnt->mnt_list, &new_ns->list);
2830 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2832 struct mnt_namespace *ns;
2833 struct super_block *s;
2837 ns = create_mnt_ns(mnt);
2839 return ERR_CAST(ns);
2841 err = vfs_path_lookup(mnt->mnt_root, mnt,
2842 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2847 return ERR_PTR(err);
2849 /* trade a vfsmount reference for active sb one */
2850 s = path.mnt->mnt_sb;
2851 atomic_inc(&s->s_active);
2853 /* lock the sucker */
2854 down_write(&s->s_umount);
2855 /* ... and return the root of (sub)tree on it */
2858 EXPORT_SYMBOL(mount_subtree);
2860 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2861 char __user *, type, unsigned long, flags, void __user *, data)
2866 unsigned long data_page;
2868 kernel_type = copy_mount_string(type);
2869 ret = PTR_ERR(kernel_type);
2870 if (IS_ERR(kernel_type))
2873 kernel_dev = copy_mount_string(dev_name);
2874 ret = PTR_ERR(kernel_dev);
2875 if (IS_ERR(kernel_dev))
2878 ret = copy_mount_options(data, &data_page);
2882 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
2883 (void *) data_page);
2885 free_page(data_page);
2895 * Return true if path is reachable from root
2897 * namespace_sem or mount_lock is held
2899 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2900 const struct path *root)
2902 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2903 dentry = mnt->mnt_mountpoint;
2904 mnt = mnt->mnt_parent;
2906 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2909 int path_is_under(struct path *path1, struct path *path2)
2912 read_seqlock_excl(&mount_lock);
2913 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2914 read_sequnlock_excl(&mount_lock);
2917 EXPORT_SYMBOL(path_is_under);
2920 * pivot_root Semantics:
2921 * Moves the root file system of the current process to the directory put_old,
2922 * makes new_root as the new root file system of the current process, and sets
2923 * root/cwd of all processes which had them on the current root to new_root.
2926 * The new_root and put_old must be directories, and must not be on the
2927 * same file system as the current process root. The put_old must be
2928 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2929 * pointed to by put_old must yield the same directory as new_root. No other
2930 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2932 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2933 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2934 * in this situation.
2937 * - we don't move root/cwd if they are not at the root (reason: if something
2938 * cared enough to change them, it's probably wrong to force them elsewhere)
2939 * - it's okay to pick a root that isn't the root of a file system, e.g.
2940 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2941 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2944 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2945 const char __user *, put_old)
2947 struct path new, old, parent_path, root_parent, root;
2948 struct mount *new_mnt, *root_mnt, *old_mnt;
2949 struct mountpoint *old_mp, *root_mp;
2955 error = user_path_dir(new_root, &new);
2959 error = user_path_dir(put_old, &old);
2963 error = security_sb_pivotroot(&old, &new);
2967 get_fs_root(current->fs, &root);
2968 old_mp = lock_mount(&old);
2969 error = PTR_ERR(old_mp);
2974 new_mnt = real_mount(new.mnt);
2975 root_mnt = real_mount(root.mnt);
2976 old_mnt = real_mount(old.mnt);
2977 if (IS_MNT_SHARED(old_mnt) ||
2978 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2979 IS_MNT_SHARED(root_mnt->mnt_parent))
2981 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2983 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2986 if (d_unlinked(new.dentry))
2989 if (new_mnt == root_mnt || old_mnt == root_mnt)
2990 goto out4; /* loop, on the same file system */
2992 if (root.mnt->mnt_root != root.dentry)
2993 goto out4; /* not a mountpoint */
2994 if (!mnt_has_parent(root_mnt))
2995 goto out4; /* not attached */
2996 root_mp = root_mnt->mnt_mp;
2997 if (new.mnt->mnt_root != new.dentry)
2998 goto out4; /* not a mountpoint */
2999 if (!mnt_has_parent(new_mnt))
3000 goto out4; /* not attached */
3001 /* make sure we can reach put_old from new_root */
3002 if (!is_path_reachable(old_mnt, old.dentry, &new))
3004 /* make certain new is below the root */
3005 if (!is_path_reachable(new_mnt, new.dentry, &root))
3007 root_mp->m_count++; /* pin it so it won't go away */
3009 detach_mnt(new_mnt, &parent_path);
3010 detach_mnt(root_mnt, &root_parent);
3011 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3012 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3013 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3015 /* mount old root on put_old */
3016 attach_mnt(root_mnt, old_mnt, old_mp);
3017 /* mount new_root on / */
3018 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3019 touch_mnt_namespace(current->nsproxy->mnt_ns);
3020 /* A moved mount should not expire automatically */
3021 list_del_init(&new_mnt->mnt_expire);
3022 unlock_mount_hash();
3023 chroot_fs_refs(&root, &new);
3024 put_mountpoint(root_mp);
3027 unlock_mount(old_mp);
3029 path_put(&root_parent);
3030 path_put(&parent_path);
3042 static void __init init_mount_tree(void)
3044 struct vfsmount *mnt;
3045 struct mnt_namespace *ns;
3047 struct file_system_type *type;
3049 type = get_fs_type("rootfs");
3051 panic("Can't find rootfs type");
3052 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3053 put_filesystem(type);
3055 panic("Can't create rootfs");
3057 ns = create_mnt_ns(mnt);
3059 panic("Can't allocate initial namespace");
3061 init_task.nsproxy->mnt_ns = ns;
3065 root.dentry = mnt->mnt_root;
3066 mnt->mnt_flags |= MNT_LOCKED;
3068 set_fs_pwd(current->fs, &root);
3069 set_fs_root(current->fs, &root);
3072 void __init mnt_init(void)
3077 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3078 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3080 mount_hashtable = alloc_large_system_hash("Mount-cache",
3081 sizeof(struct hlist_head),
3084 &m_hash_shift, &m_hash_mask, 0, 0);
3085 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3086 sizeof(struct hlist_head),
3089 &mp_hash_shift, &mp_hash_mask, 0, 0);
3091 if (!mount_hashtable || !mountpoint_hashtable)
3092 panic("Failed to allocate mount hash table\n");
3094 for (u = 0; u <= m_hash_mask; u++)
3095 INIT_HLIST_HEAD(&mount_hashtable[u]);
3096 for (u = 0; u <= mp_hash_mask; u++)
3097 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3103 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3105 fs_kobj = kobject_create_and_add("fs", NULL);
3107 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3112 void put_mnt_ns(struct mnt_namespace *ns)
3114 if (!atomic_dec_and_test(&ns->count))
3116 drop_collected_mounts(&ns->root->mnt);
3120 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3122 struct vfsmount *mnt;
3123 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3126 * it is a longterm mount, don't release mnt until
3127 * we unmount before file sys is unregistered
3129 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3133 EXPORT_SYMBOL_GPL(kern_mount_data);
3135 void kern_unmount(struct vfsmount *mnt)
3137 /* release long term mount so mount point can be released */
3138 if (!IS_ERR_OR_NULL(mnt)) {
3139 real_mount(mnt)->mnt_ns = NULL;
3140 synchronize_rcu(); /* yecchhh... */
3144 EXPORT_SYMBOL(kern_unmount);
3146 bool our_mnt(struct vfsmount *mnt)
3148 return check_mnt(real_mount(mnt));
3151 bool current_chrooted(void)
3153 /* Does the current process have a non-standard root */
3154 struct path ns_root;
3155 struct path fs_root;
3158 /* Find the namespace root */
3159 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3160 ns_root.dentry = ns_root.mnt->mnt_root;
3162 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3165 get_fs_root(current->fs, &fs_root);
3167 chrooted = !path_equal(&fs_root, &ns_root);
3175 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3177 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3178 int new_flags = *new_mnt_flags;
3180 bool visible = false;
3185 down_read(&namespace_sem);
3186 list_for_each_entry(mnt, &ns->list, mnt_list) {
3187 struct mount *child;
3188 if (mnt->mnt.mnt_sb->s_type != type)
3191 /* This mount is not fully visible if it's root directory
3192 * is not the root directory of the filesystem.
3194 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3197 /* Verify the mount flags are equal to or more permissive
3198 * than the proposed new mount.
3200 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
3201 !(new_flags & MNT_READONLY))
3203 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
3204 !(new_flags & MNT_NODEV))
3206 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
3207 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3210 /* This mount is not fully visible if there are any
3211 * locked child mounts that cover anything except for
3212 * empty directories.
3214 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3215 struct inode *inode = child->mnt_mountpoint->d_inode;
3216 /* Only worry about locked mounts */
3217 if (!(mnt->mnt.mnt_flags & MNT_LOCKED))
3219 /* Is the directory permanetly empty? */
3220 if (!is_empty_dir_inode(inode))
3223 /* Preserve the locked attributes */
3224 *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \
3232 up_read(&namespace_sem);
3236 static struct ns_common *mntns_get(struct task_struct *task)
3238 struct ns_common *ns = NULL;
3239 struct nsproxy *nsproxy;
3242 nsproxy = task->nsproxy;
3244 ns = &nsproxy->mnt_ns->ns;
3245 get_mnt_ns(to_mnt_ns(ns));
3252 static void mntns_put(struct ns_common *ns)
3254 put_mnt_ns(to_mnt_ns(ns));
3257 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3259 struct fs_struct *fs = current->fs;
3260 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3263 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3264 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3265 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3272 put_mnt_ns(nsproxy->mnt_ns);
3273 nsproxy->mnt_ns = mnt_ns;
3276 root.mnt = &mnt_ns->root->mnt;
3277 root.dentry = mnt_ns->root->mnt.mnt_root;
3279 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3282 /* Update the pwd and root */
3283 set_fs_pwd(fs, &root);
3284 set_fs_root(fs, &root);
3290 const struct proc_ns_operations mntns_operations = {
3292 .type = CLONE_NEWNS,
3295 .install = mntns_install,