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/cred.h>
19 #include <linux/idr.h>
20 #include <linux/init.h> /* init_rootfs */
21 #include <linux/fs_struct.h> /* get_fs_root et.al. */
22 #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
27 #include <linux/task_work.h>
28 #include <linux/sched/task.h>
33 /* Maximum number of mounts in a mount namespace */
34 unsigned int sysctl_mount_max __read_mostly = 100000;
36 static unsigned int m_hash_mask __read_mostly;
37 static unsigned int m_hash_shift __read_mostly;
38 static unsigned int mp_hash_mask __read_mostly;
39 static unsigned int mp_hash_shift __read_mostly;
41 static __initdata unsigned long mhash_entries;
42 static int __init set_mhash_entries(char *str)
46 mhash_entries = simple_strtoul(str, &str, 0);
49 __setup("mhash_entries=", set_mhash_entries);
51 static __initdata unsigned long mphash_entries;
52 static int __init set_mphash_entries(char *str)
56 mphash_entries = simple_strtoul(str, &str, 0);
59 __setup("mphash_entries=", set_mphash_entries);
62 static DEFINE_IDA(mnt_id_ida);
63 static DEFINE_IDA(mnt_group_ida);
64 static DEFINE_SPINLOCK(mnt_id_lock);
65 static int mnt_id_start = 0;
66 static int mnt_group_start = 1;
68 static struct hlist_head *mount_hashtable __read_mostly;
69 static struct hlist_head *mountpoint_hashtable __read_mostly;
70 static struct kmem_cache *mnt_cache __read_mostly;
71 static DECLARE_RWSEM(namespace_sem);
74 struct kobject *fs_kobj;
75 EXPORT_SYMBOL_GPL(fs_kobj);
78 * vfsmount lock may be taken for read to prevent changes to the
79 * vfsmount hash, ie. during mountpoint lookups or walking back
82 * It should be taken for write in all cases where the vfsmount
83 * tree or hash is modified or when a vfsmount structure is modified.
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
87 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
89 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
90 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
91 tmp = tmp + (tmp >> m_hash_shift);
92 return &mount_hashtable[tmp & m_hash_mask];
95 static inline struct hlist_head *mp_hash(struct dentry *dentry)
97 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
98 tmp = tmp + (tmp >> mp_hash_shift);
99 return &mountpoint_hashtable[tmp & mp_hash_mask];
102 static int mnt_alloc_id(struct mount *mnt)
107 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
108 spin_lock(&mnt_id_lock);
109 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
111 mnt_id_start = mnt->mnt_id + 1;
112 spin_unlock(&mnt_id_lock);
119 static void mnt_free_id(struct mount *mnt)
121 int id = mnt->mnt_id;
122 spin_lock(&mnt_id_lock);
123 ida_remove(&mnt_id_ida, id);
124 if (mnt_id_start > id)
126 spin_unlock(&mnt_id_lock);
130 * Allocate a new peer group ID
132 * mnt_group_ida is protected by namespace_sem
134 static int mnt_alloc_group_id(struct mount *mnt)
138 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
141 res = ida_get_new_above(&mnt_group_ida,
145 mnt_group_start = mnt->mnt_group_id + 1;
151 * Release a peer group ID
153 void mnt_release_group_id(struct mount *mnt)
155 int id = mnt->mnt_group_id;
156 ida_remove(&mnt_group_ida, id);
157 if (mnt_group_start > id)
158 mnt_group_start = id;
159 mnt->mnt_group_id = 0;
163 * vfsmount lock must be held for read
165 static inline void mnt_add_count(struct mount *mnt, int n)
168 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
177 * vfsmount lock must be held for write
179 unsigned int mnt_get_count(struct mount *mnt)
182 unsigned int count = 0;
185 for_each_possible_cpu(cpu) {
186 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
191 return mnt->mnt_count;
195 static void drop_mountpoint(struct fs_pin *p)
197 struct mount *m = container_of(p, struct mount, mnt_umount);
198 dput(m->mnt_ex_mountpoint);
203 static struct mount *alloc_vfsmnt(const char *name)
205 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
209 err = mnt_alloc_id(mnt);
214 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
215 if (!mnt->mnt_devname)
220 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
222 goto out_free_devname;
224 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
227 mnt->mnt_writers = 0;
230 INIT_HLIST_NODE(&mnt->mnt_hash);
231 INIT_LIST_HEAD(&mnt->mnt_child);
232 INIT_LIST_HEAD(&mnt->mnt_mounts);
233 INIT_LIST_HEAD(&mnt->mnt_list);
234 INIT_LIST_HEAD(&mnt->mnt_expire);
235 INIT_LIST_HEAD(&mnt->mnt_share);
236 INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 INIT_LIST_HEAD(&mnt->mnt_slave);
238 INIT_HLIST_NODE(&mnt->mnt_mp_list);
239 INIT_LIST_HEAD(&mnt->mnt_umounting);
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
246 kfree_const(mnt->mnt_devname);
251 kmem_cache_free(mnt_cache, mnt);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 int __mnt_is_readonly(struct vfsmount *mnt)
276 if (mnt->mnt_flags & MNT_READONLY)
278 if (mnt->mnt_sb->s_flags & MS_RDONLY)
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
284 static inline void mnt_inc_writers(struct mount *mnt)
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
293 static inline void mnt_dec_writers(struct mount *mnt)
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
302 static unsigned int mnt_get_writers(struct mount *mnt)
305 unsigned int count = 0;
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
314 return mnt->mnt_writers;
318 static int mnt_is_readonly(struct vfsmount *mnt)
320 if (mnt->mnt_sb->s_readonly_remount)
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
324 return __mnt_is_readonly(mnt);
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
343 int __mnt_want_write(struct vfsmount *m)
345 struct mount *mnt = real_mount(m);
349 mnt_inc_writers(mnt);
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
356 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
382 int mnt_want_write(struct vfsmount *m)
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
389 sb_end_write(m->mnt_sb);
392 EXPORT_SYMBOL_GPL(mnt_want_write);
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
406 int mnt_clone_write(struct vfsmount *mnt)
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
412 mnt_inc_writers(real_mount(mnt));
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
425 int __mnt_want_write_file(struct file *file)
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
430 return mnt_clone_write(file->f_path.mnt);
434 * mnt_want_write_file - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
440 int mnt_want_write_file(struct file *file)
444 sb_start_write(file->f_path.mnt->mnt_sb);
445 ret = __mnt_want_write_file(file);
447 sb_end_write(file->f_path.mnt->mnt_sb);
450 EXPORT_SYMBOL_GPL(mnt_want_write_file);
453 * __mnt_drop_write - give up write access to a mount
454 * @mnt: the mount on which to give up write access
456 * Tells the low-level filesystem that we are done
457 * performing writes to it. Must be matched with
458 * __mnt_want_write() call above.
460 void __mnt_drop_write(struct vfsmount *mnt)
463 mnt_dec_writers(real_mount(mnt));
468 * mnt_drop_write - give up write access to a mount
469 * @mnt: the mount on which to give up write access
471 * Tells the low-level filesystem that we are done performing writes to it and
472 * also allows filesystem to be frozen again. Must be matched with
473 * mnt_want_write() call above.
475 void mnt_drop_write(struct vfsmount *mnt)
477 __mnt_drop_write(mnt);
478 sb_end_write(mnt->mnt_sb);
480 EXPORT_SYMBOL_GPL(mnt_drop_write);
482 void __mnt_drop_write_file(struct file *file)
484 __mnt_drop_write(file->f_path.mnt);
487 void mnt_drop_write_file(struct file *file)
489 mnt_drop_write(file->f_path.mnt);
491 EXPORT_SYMBOL(mnt_drop_write_file);
493 static int mnt_make_readonly(struct mount *mnt)
498 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
500 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
501 * should be visible before we do.
506 * With writers on hold, if this value is zero, then there are
507 * definitely no active writers (although held writers may subsequently
508 * increment the count, they'll have to wait, and decrement it after
509 * seeing MNT_READONLY).
511 * It is OK to have counter incremented on one CPU and decremented on
512 * another: the sum will add up correctly. The danger would be when we
513 * sum up each counter, if we read a counter before it is incremented,
514 * but then read another CPU's count which it has been subsequently
515 * decremented from -- we would see more decrements than we should.
516 * MNT_WRITE_HOLD protects against this scenario, because
517 * mnt_want_write first increments count, then smp_mb, then spins on
518 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
519 * we're counting up here.
521 if (mnt_get_writers(mnt) > 0)
524 mnt->mnt.mnt_flags |= MNT_READONLY;
526 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
527 * that become unheld will see MNT_READONLY.
530 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
535 static void __mnt_unmake_readonly(struct mount *mnt)
538 mnt->mnt.mnt_flags &= ~MNT_READONLY;
542 int sb_prepare_remount_readonly(struct super_block *sb)
547 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
548 if (atomic_long_read(&sb->s_remove_count))
552 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
553 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
554 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
556 if (mnt_get_writers(mnt) > 0) {
562 if (!err && atomic_long_read(&sb->s_remove_count))
566 sb->s_readonly_remount = 1;
569 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
570 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
571 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
578 static void free_vfsmnt(struct mount *mnt)
580 kfree_const(mnt->mnt_devname);
582 free_percpu(mnt->mnt_pcp);
584 kmem_cache_free(mnt_cache, mnt);
587 static void delayed_free_vfsmnt(struct rcu_head *head)
589 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
592 /* call under rcu_read_lock */
593 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
596 if (read_seqretry(&mount_lock, seq))
600 mnt = real_mount(bastard);
601 mnt_add_count(mnt, 1);
602 if (likely(!read_seqretry(&mount_lock, seq)))
604 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
605 mnt_add_count(mnt, -1);
611 /* call under rcu_read_lock */
612 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
614 int res = __legitimize_mnt(bastard, seq);
617 if (unlikely(res < 0)) {
626 * find the first mount at @dentry on vfsmount @mnt.
627 * call under rcu_read_lock()
629 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
631 struct hlist_head *head = m_hash(mnt, dentry);
634 hlist_for_each_entry_rcu(p, head, mnt_hash)
635 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
641 * lookup_mnt - Return the first child mount mounted at path
643 * "First" means first mounted chronologically. If you create the
646 * mount /dev/sda1 /mnt
647 * mount /dev/sda2 /mnt
648 * mount /dev/sda3 /mnt
650 * Then lookup_mnt() on the base /mnt dentry in the root mount will
651 * return successively the root dentry and vfsmount of /dev/sda1, then
652 * /dev/sda2, then /dev/sda3, then NULL.
654 * lookup_mnt takes a reference to the found vfsmount.
656 struct vfsmount *lookup_mnt(const struct path *path)
658 struct mount *child_mnt;
664 seq = read_seqbegin(&mount_lock);
665 child_mnt = __lookup_mnt(path->mnt, path->dentry);
666 m = child_mnt ? &child_mnt->mnt : NULL;
667 } while (!legitimize_mnt(m, seq));
673 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
674 * current mount namespace.
676 * The common case is dentries are not mountpoints at all and that
677 * test is handled inline. For the slow case when we are actually
678 * dealing with a mountpoint of some kind, walk through all of the
679 * mounts in the current mount namespace and test to see if the dentry
682 * The mount_hashtable is not usable in the context because we
683 * need to identify all mounts that may be in the current mount
684 * namespace not just a mount that happens to have some specified
687 bool __is_local_mountpoint(struct dentry *dentry)
689 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
691 bool is_covered = false;
693 if (!d_mountpoint(dentry))
696 down_read(&namespace_sem);
697 list_for_each_entry(mnt, &ns->list, mnt_list) {
698 is_covered = (mnt->mnt_mountpoint == dentry);
702 up_read(&namespace_sem);
707 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
709 struct hlist_head *chain = mp_hash(dentry);
710 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);
724 static struct mountpoint *get_mountpoint(struct dentry *dentry)
726 struct mountpoint *mp, *new = NULL;
729 if (d_mountpoint(dentry)) {
731 read_seqlock_excl(&mount_lock);
732 mp = lookup_mountpoint(dentry);
733 read_sequnlock_excl(&mount_lock);
739 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
741 return ERR_PTR(-ENOMEM);
744 /* Exactly one processes may set d_mounted */
745 ret = d_set_mounted(dentry);
747 /* Someone else set d_mounted? */
751 /* The dentry is not available as a mountpoint? */
756 /* Add the new mountpoint to the hash table */
757 read_seqlock_excl(&mount_lock);
758 new->m_dentry = dentry;
760 hlist_add_head(&new->m_hash, mp_hash(dentry));
761 INIT_HLIST_HEAD(&new->m_list);
762 read_sequnlock_excl(&mount_lock);
771 static void put_mountpoint(struct mountpoint *mp)
773 if (!--mp->m_count) {
774 struct dentry *dentry = mp->m_dentry;
775 BUG_ON(!hlist_empty(&mp->m_list));
776 spin_lock(&dentry->d_lock);
777 dentry->d_flags &= ~DCACHE_MOUNTED;
778 spin_unlock(&dentry->d_lock);
779 hlist_del(&mp->m_hash);
784 static inline int check_mnt(struct mount *mnt)
786 return mnt->mnt_ns == current->nsproxy->mnt_ns;
790 * vfsmount lock must be held for write
792 static void touch_mnt_namespace(struct mnt_namespace *ns)
796 wake_up_interruptible(&ns->poll);
801 * vfsmount lock must be held for write
803 static void __touch_mnt_namespace(struct mnt_namespace *ns)
805 if (ns && ns->event != event) {
807 wake_up_interruptible(&ns->poll);
812 * vfsmount lock must be held for write
814 static void unhash_mnt(struct mount *mnt)
816 mnt->mnt_parent = mnt;
817 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
818 list_del_init(&mnt->mnt_child);
819 hlist_del_init_rcu(&mnt->mnt_hash);
820 hlist_del_init(&mnt->mnt_mp_list);
821 put_mountpoint(mnt->mnt_mp);
826 * vfsmount lock must be held for write
828 static void detach_mnt(struct mount *mnt, struct path *old_path)
830 old_path->dentry = mnt->mnt_mountpoint;
831 old_path->mnt = &mnt->mnt_parent->mnt;
836 * vfsmount lock must be held for write
838 static void umount_mnt(struct mount *mnt)
840 /* old mountpoint will be dropped when we can do that */
841 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
846 * vfsmount lock must be held for write
848 void mnt_set_mountpoint(struct mount *mnt,
849 struct mountpoint *mp,
850 struct mount *child_mnt)
853 mnt_add_count(mnt, 1); /* essentially, that's mntget */
854 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
855 child_mnt->mnt_parent = mnt;
856 child_mnt->mnt_mp = mp;
857 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
860 static void __attach_mnt(struct mount *mnt, struct mount *parent)
862 hlist_add_head_rcu(&mnt->mnt_hash,
863 m_hash(&parent->mnt, mnt->mnt_mountpoint));
864 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
868 * vfsmount lock must be held for write
870 static void attach_mnt(struct mount *mnt,
871 struct mount *parent,
872 struct mountpoint *mp)
874 mnt_set_mountpoint(parent, mp, mnt);
875 __attach_mnt(mnt, parent);
878 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
880 struct mountpoint *old_mp = mnt->mnt_mp;
881 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
882 struct mount *old_parent = mnt->mnt_parent;
884 list_del_init(&mnt->mnt_child);
885 hlist_del_init(&mnt->mnt_mp_list);
886 hlist_del_init_rcu(&mnt->mnt_hash);
888 attach_mnt(mnt, parent, mp);
890 put_mountpoint(old_mp);
893 * Safely avoid even the suggestion this code might sleep or
894 * lock the mount hash by taking advantage of the knowledge that
895 * mnt_change_mountpoint will not release the final reference
898 * During mounting, the mount passed in as the parent mount will
899 * continue to use the old mountpoint and during unmounting, the
900 * old mountpoint will continue to exist until namespace_unlock,
901 * which happens well after mnt_change_mountpoint.
903 spin_lock(&old_mountpoint->d_lock);
904 old_mountpoint->d_lockref.count--;
905 spin_unlock(&old_mountpoint->d_lock);
907 mnt_add_count(old_parent, -1);
911 * vfsmount lock must be held for write
913 static void commit_tree(struct mount *mnt)
915 struct mount *parent = mnt->mnt_parent;
918 struct mnt_namespace *n = parent->mnt_ns;
920 BUG_ON(parent == mnt);
922 list_add_tail(&head, &mnt->mnt_list);
923 list_for_each_entry(m, &head, mnt_list)
926 list_splice(&head, n->list.prev);
928 n->mounts += n->pending_mounts;
929 n->pending_mounts = 0;
931 __attach_mnt(mnt, parent);
932 touch_mnt_namespace(n);
935 static struct mount *next_mnt(struct mount *p, struct mount *root)
937 struct list_head *next = p->mnt_mounts.next;
938 if (next == &p->mnt_mounts) {
942 next = p->mnt_child.next;
943 if (next != &p->mnt_parent->mnt_mounts)
948 return list_entry(next, struct mount, mnt_child);
951 static struct mount *skip_mnt_tree(struct mount *p)
953 struct list_head *prev = p->mnt_mounts.prev;
954 while (prev != &p->mnt_mounts) {
955 p = list_entry(prev, struct mount, mnt_child);
956 prev = p->mnt_mounts.prev;
962 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
968 return ERR_PTR(-ENODEV);
970 mnt = alloc_vfsmnt(name);
972 return ERR_PTR(-ENOMEM);
974 if (flags & MS_KERNMOUNT)
975 mnt->mnt.mnt_flags = MNT_INTERNAL;
977 root = mount_fs(type, flags, name, data);
981 return ERR_CAST(root);
984 mnt->mnt.mnt_root = root;
985 mnt->mnt.mnt_sb = root->d_sb;
986 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
987 mnt->mnt_parent = mnt;
989 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
993 EXPORT_SYMBOL_GPL(vfs_kern_mount);
996 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
997 const char *name, void *data)
999 /* Until it is worked out how to pass the user namespace
1000 * through from the parent mount to the submount don't support
1001 * unprivileged mounts with submounts.
1003 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1004 return ERR_PTR(-EPERM);
1006 return vfs_kern_mount(type, MS_SUBMOUNT, name, data);
1008 EXPORT_SYMBOL_GPL(vfs_submount);
1010 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1013 struct super_block *sb = old->mnt.mnt_sb;
1017 mnt = alloc_vfsmnt(old->mnt_devname);
1019 return ERR_PTR(-ENOMEM);
1021 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1022 mnt->mnt_group_id = 0; /* not a peer of original */
1024 mnt->mnt_group_id = old->mnt_group_id;
1026 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1027 err = mnt_alloc_group_id(mnt);
1032 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1033 /* Don't allow unprivileged users to change mount flags */
1034 if (flag & CL_UNPRIVILEGED) {
1035 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1037 if (mnt->mnt.mnt_flags & MNT_READONLY)
1038 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1040 if (mnt->mnt.mnt_flags & MNT_NODEV)
1041 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1043 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1044 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1046 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1047 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1050 /* Don't allow unprivileged users to reveal what is under a mount */
1051 if ((flag & CL_UNPRIVILEGED) &&
1052 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1053 mnt->mnt.mnt_flags |= MNT_LOCKED;
1055 atomic_inc(&sb->s_active);
1056 mnt->mnt.mnt_sb = sb;
1057 mnt->mnt.mnt_root = dget(root);
1058 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1059 mnt->mnt_parent = mnt;
1061 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1062 unlock_mount_hash();
1064 if ((flag & CL_SLAVE) ||
1065 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1066 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1067 mnt->mnt_master = old;
1068 CLEAR_MNT_SHARED(mnt);
1069 } else if (!(flag & CL_PRIVATE)) {
1070 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1071 list_add(&mnt->mnt_share, &old->mnt_share);
1072 if (IS_MNT_SLAVE(old))
1073 list_add(&mnt->mnt_slave, &old->mnt_slave);
1074 mnt->mnt_master = old->mnt_master;
1076 CLEAR_MNT_SHARED(mnt);
1078 if (flag & CL_MAKE_SHARED)
1079 set_mnt_shared(mnt);
1081 /* stick the duplicate mount on the same expiry list
1082 * as the original if that was on one */
1083 if (flag & CL_EXPIRE) {
1084 if (!list_empty(&old->mnt_expire))
1085 list_add(&mnt->mnt_expire, &old->mnt_expire);
1093 return ERR_PTR(err);
1096 static void cleanup_mnt(struct mount *mnt)
1099 * This probably indicates that somebody messed
1100 * up a mnt_want/drop_write() pair. If this
1101 * happens, the filesystem was probably unable
1102 * to make r/w->r/o transitions.
1105 * The locking used to deal with mnt_count decrement provides barriers,
1106 * so mnt_get_writers() below is safe.
1108 WARN_ON(mnt_get_writers(mnt));
1109 if (unlikely(mnt->mnt_pins.first))
1111 fsnotify_vfsmount_delete(&mnt->mnt);
1112 dput(mnt->mnt.mnt_root);
1113 deactivate_super(mnt->mnt.mnt_sb);
1115 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1118 static void __cleanup_mnt(struct rcu_head *head)
1120 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1123 static LLIST_HEAD(delayed_mntput_list);
1124 static void delayed_mntput(struct work_struct *unused)
1126 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1127 struct llist_node *next;
1129 for (; node; node = next) {
1130 next = llist_next(node);
1131 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1134 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1136 static void mntput_no_expire(struct mount *mnt)
1139 mnt_add_count(mnt, -1);
1140 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1145 if (mnt_get_count(mnt)) {
1147 unlock_mount_hash();
1150 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1152 unlock_mount_hash();
1155 mnt->mnt.mnt_flags |= MNT_DOOMED;
1158 list_del(&mnt->mnt_instance);
1160 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1161 struct mount *p, *tmp;
1162 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1166 unlock_mount_hash();
1168 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1169 struct task_struct *task = current;
1170 if (likely(!(task->flags & PF_KTHREAD))) {
1171 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1172 if (!task_work_add(task, &mnt->mnt_rcu, true))
1175 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1176 schedule_delayed_work(&delayed_mntput_work, 1);
1182 void mntput(struct vfsmount *mnt)
1185 struct mount *m = real_mount(mnt);
1186 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1187 if (unlikely(m->mnt_expiry_mark))
1188 m->mnt_expiry_mark = 0;
1189 mntput_no_expire(m);
1192 EXPORT_SYMBOL(mntput);
1194 struct vfsmount *mntget(struct vfsmount *mnt)
1197 mnt_add_count(real_mount(mnt), 1);
1200 EXPORT_SYMBOL(mntget);
1202 /* path_is_mountpoint() - Check if path is a mount in the current
1205 * d_mountpoint() can only be used reliably to establish if a dentry is
1206 * not mounted in any namespace and that common case is handled inline.
1207 * d_mountpoint() isn't aware of the possibility there may be multiple
1208 * mounts using a given dentry in a different namespace. This function
1209 * checks if the passed in path is a mountpoint rather than the dentry
1212 bool path_is_mountpoint(const struct path *path)
1217 if (!d_mountpoint(path->dentry))
1222 seq = read_seqbegin(&mount_lock);
1223 res = __path_is_mountpoint(path);
1224 } while (read_seqretry(&mount_lock, seq));
1229 EXPORT_SYMBOL(path_is_mountpoint);
1231 struct vfsmount *mnt_clone_internal(const struct path *path)
1234 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1237 p->mnt.mnt_flags |= MNT_INTERNAL;
1241 static inline void mangle(struct seq_file *m, const char *s)
1243 seq_escape(m, s, " \t\n\\");
1247 * Simple .show_options callback for filesystems which don't want to
1248 * implement more complex mount option showing.
1250 * See also save_mount_options().
1252 int generic_show_options(struct seq_file *m, struct dentry *root)
1254 const char *options;
1257 options = rcu_dereference(root->d_sb->s_options);
1259 if (options != NULL && options[0]) {
1267 EXPORT_SYMBOL(generic_show_options);
1270 * If filesystem uses generic_show_options(), this function should be
1271 * called from the fill_super() callback.
1273 * The .remount_fs callback usually needs to be handled in a special
1274 * way, to make sure, that previous options are not overwritten if the
1277 * Also note, that if the filesystem's .remount_fs function doesn't
1278 * reset all options to their default value, but changes only newly
1279 * given options, then the displayed options will not reflect reality
1282 void save_mount_options(struct super_block *sb, char *options)
1284 BUG_ON(sb->s_options);
1285 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1287 EXPORT_SYMBOL(save_mount_options);
1289 void replace_mount_options(struct super_block *sb, char *options)
1291 char *old = sb->s_options;
1292 rcu_assign_pointer(sb->s_options, options);
1298 EXPORT_SYMBOL(replace_mount_options);
1300 #ifdef CONFIG_PROC_FS
1301 /* iterator; we want it to have access to namespace_sem, thus here... */
1302 static void *m_start(struct seq_file *m, loff_t *pos)
1304 struct proc_mounts *p = m->private;
1306 down_read(&namespace_sem);
1307 if (p->cached_event == p->ns->event) {
1308 void *v = p->cached_mount;
1309 if (*pos == p->cached_index)
1311 if (*pos == p->cached_index + 1) {
1312 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1313 return p->cached_mount = v;
1317 p->cached_event = p->ns->event;
1318 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1319 p->cached_index = *pos;
1320 return p->cached_mount;
1323 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1325 struct proc_mounts *p = m->private;
1327 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1328 p->cached_index = *pos;
1329 return p->cached_mount;
1332 static void m_stop(struct seq_file *m, void *v)
1334 up_read(&namespace_sem);
1337 static int m_show(struct seq_file *m, void *v)
1339 struct proc_mounts *p = m->private;
1340 struct mount *r = list_entry(v, struct mount, mnt_list);
1341 return p->show(m, &r->mnt);
1344 const struct seq_operations mounts_op = {
1350 #endif /* CONFIG_PROC_FS */
1353 * may_umount_tree - check if a mount tree is busy
1354 * @mnt: root of mount tree
1356 * This is called to check if a tree of mounts has any
1357 * open files, pwds, chroots or sub mounts that are
1360 int may_umount_tree(struct vfsmount *m)
1362 struct mount *mnt = real_mount(m);
1363 int actual_refs = 0;
1364 int minimum_refs = 0;
1368 /* write lock needed for mnt_get_count */
1370 for (p = mnt; p; p = next_mnt(p, mnt)) {
1371 actual_refs += mnt_get_count(p);
1374 unlock_mount_hash();
1376 if (actual_refs > minimum_refs)
1382 EXPORT_SYMBOL(may_umount_tree);
1385 * may_umount - check if a mount point is busy
1386 * @mnt: root of mount
1388 * This is called to check if a mount point has any
1389 * open files, pwds, chroots or sub mounts. If the
1390 * mount has sub mounts this will return busy
1391 * regardless of whether the sub mounts are busy.
1393 * Doesn't take quota and stuff into account. IOW, in some cases it will
1394 * give false negatives. The main reason why it's here is that we need
1395 * a non-destructive way to look for easily umountable filesystems.
1397 int may_umount(struct vfsmount *mnt)
1400 down_read(&namespace_sem);
1402 if (propagate_mount_busy(real_mount(mnt), 2))
1404 unlock_mount_hash();
1405 up_read(&namespace_sem);
1409 EXPORT_SYMBOL(may_umount);
1411 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1413 static void namespace_unlock(void)
1415 struct hlist_head head;
1417 hlist_move_list(&unmounted, &head);
1419 up_write(&namespace_sem);
1421 if (likely(hlist_empty(&head)))
1426 group_pin_kill(&head);
1429 static inline void namespace_lock(void)
1431 down_write(&namespace_sem);
1434 enum umount_tree_flags {
1436 UMOUNT_PROPAGATE = 2,
1437 UMOUNT_CONNECTED = 4,
1440 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1442 /* Leaving mounts connected is only valid for lazy umounts */
1443 if (how & UMOUNT_SYNC)
1446 /* A mount without a parent has nothing to be connected to */
1447 if (!mnt_has_parent(mnt))
1450 /* Because the reference counting rules change when mounts are
1451 * unmounted and connected, umounted mounts may not be
1452 * connected to mounted mounts.
1454 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1457 /* Has it been requested that the mount remain connected? */
1458 if (how & UMOUNT_CONNECTED)
1461 /* Is the mount locked such that it needs to remain connected? */
1462 if (IS_MNT_LOCKED(mnt))
1465 /* By default disconnect the mount */
1470 * mount_lock must be held
1471 * namespace_sem must be held for write
1473 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1475 LIST_HEAD(tmp_list);
1478 if (how & UMOUNT_PROPAGATE)
1479 propagate_mount_unlock(mnt);
1481 /* Gather the mounts to umount */
1482 for (p = mnt; p; p = next_mnt(p, mnt)) {
1483 p->mnt.mnt_flags |= MNT_UMOUNT;
1484 list_move(&p->mnt_list, &tmp_list);
1487 /* Hide the mounts from mnt_mounts */
1488 list_for_each_entry(p, &tmp_list, mnt_list) {
1489 list_del_init(&p->mnt_child);
1492 /* Add propogated mounts to the tmp_list */
1493 if (how & UMOUNT_PROPAGATE)
1494 propagate_umount(&tmp_list);
1496 while (!list_empty(&tmp_list)) {
1497 struct mnt_namespace *ns;
1499 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1500 list_del_init(&p->mnt_expire);
1501 list_del_init(&p->mnt_list);
1505 __touch_mnt_namespace(ns);
1508 if (how & UMOUNT_SYNC)
1509 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1511 disconnect = disconnect_mount(p, how);
1513 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1514 disconnect ? &unmounted : NULL);
1515 if (mnt_has_parent(p)) {
1516 mnt_add_count(p->mnt_parent, -1);
1518 /* Don't forget about p */
1519 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1524 change_mnt_propagation(p, MS_PRIVATE);
1528 static void shrink_submounts(struct mount *mnt);
1530 static int do_umount(struct mount *mnt, int flags)
1532 struct super_block *sb = mnt->mnt.mnt_sb;
1535 retval = security_sb_umount(&mnt->mnt, flags);
1540 * Allow userspace to request a mountpoint be expired rather than
1541 * unmounting unconditionally. Unmount only happens if:
1542 * (1) the mark is already set (the mark is cleared by mntput())
1543 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1545 if (flags & MNT_EXPIRE) {
1546 if (&mnt->mnt == current->fs->root.mnt ||
1547 flags & (MNT_FORCE | MNT_DETACH))
1551 * probably don't strictly need the lock here if we examined
1552 * all race cases, but it's a slowpath.
1555 if (mnt_get_count(mnt) != 2) {
1556 unlock_mount_hash();
1559 unlock_mount_hash();
1561 if (!xchg(&mnt->mnt_expiry_mark, 1))
1566 * If we may have to abort operations to get out of this
1567 * mount, and they will themselves hold resources we must
1568 * allow the fs to do things. In the Unix tradition of
1569 * 'Gee thats tricky lets do it in userspace' the umount_begin
1570 * might fail to complete on the first run through as other tasks
1571 * must return, and the like. Thats for the mount program to worry
1572 * about for the moment.
1575 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1576 sb->s_op->umount_begin(sb);
1580 * No sense to grab the lock for this test, but test itself looks
1581 * somewhat bogus. Suggestions for better replacement?
1582 * Ho-hum... In principle, we might treat that as umount + switch
1583 * to rootfs. GC would eventually take care of the old vfsmount.
1584 * Actually it makes sense, especially if rootfs would contain a
1585 * /reboot - static binary that would close all descriptors and
1586 * call reboot(9). Then init(8) could umount root and exec /reboot.
1588 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1590 * Special case for "unmounting" root ...
1591 * we just try to remount it readonly.
1593 if (!capable(CAP_SYS_ADMIN))
1595 down_write(&sb->s_umount);
1596 if (!(sb->s_flags & MS_RDONLY))
1597 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1598 up_write(&sb->s_umount);
1606 if (flags & MNT_DETACH) {
1607 if (!list_empty(&mnt->mnt_list))
1608 umount_tree(mnt, UMOUNT_PROPAGATE);
1611 shrink_submounts(mnt);
1613 if (!propagate_mount_busy(mnt, 2)) {
1614 if (!list_empty(&mnt->mnt_list))
1615 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1619 unlock_mount_hash();
1625 * __detach_mounts - lazily unmount all mounts on the specified dentry
1627 * During unlink, rmdir, and d_drop it is possible to loose the path
1628 * to an existing mountpoint, and wind up leaking the mount.
1629 * detach_mounts allows lazily unmounting those mounts instead of
1632 * The caller may hold dentry->d_inode->i_mutex.
1634 void __detach_mounts(struct dentry *dentry)
1636 struct mountpoint *mp;
1641 mp = lookup_mountpoint(dentry);
1642 if (IS_ERR_OR_NULL(mp))
1646 while (!hlist_empty(&mp->m_list)) {
1647 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1648 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1649 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1652 else umount_tree(mnt, UMOUNT_CONNECTED);
1656 unlock_mount_hash();
1661 * Is the caller allowed to modify his namespace?
1663 static inline bool may_mount(void)
1665 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1668 static inline bool may_mandlock(void)
1670 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1673 return capable(CAP_SYS_ADMIN);
1677 * Now umount can handle mount points as well as block devices.
1678 * This is important for filesystems which use unnamed block devices.
1680 * We now support a flag for forced unmount like the other 'big iron'
1681 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1684 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1689 int lookup_flags = 0;
1691 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1697 if (!(flags & UMOUNT_NOFOLLOW))
1698 lookup_flags |= LOOKUP_FOLLOW;
1700 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1703 mnt = real_mount(path.mnt);
1705 if (path.dentry != path.mnt->mnt_root)
1707 if (!check_mnt(mnt))
1709 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1712 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1715 retval = do_umount(mnt, flags);
1717 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1719 mntput_no_expire(mnt);
1724 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1727 * The 2.0 compatible umount. No flags.
1729 SYSCALL_DEFINE1(oldumount, char __user *, name)
1731 return sys_umount(name, 0);
1736 static bool is_mnt_ns_file(struct dentry *dentry)
1738 /* Is this a proxy for a mount namespace? */
1739 return dentry->d_op == &ns_dentry_operations &&
1740 dentry->d_fsdata == &mntns_operations;
1743 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1745 return container_of(ns, struct mnt_namespace, ns);
1748 static bool mnt_ns_loop(struct dentry *dentry)
1750 /* Could bind mounting the mount namespace inode cause a
1751 * mount namespace loop?
1753 struct mnt_namespace *mnt_ns;
1754 if (!is_mnt_ns_file(dentry))
1757 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1758 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1761 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1764 struct mount *res, *p, *q, *r, *parent;
1766 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1767 return ERR_PTR(-EINVAL);
1769 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1770 return ERR_PTR(-EINVAL);
1772 res = q = clone_mnt(mnt, dentry, flag);
1776 q->mnt_mountpoint = mnt->mnt_mountpoint;
1779 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1781 if (!is_subdir(r->mnt_mountpoint, dentry))
1784 for (s = r; s; s = next_mnt(s, r)) {
1785 if (!(flag & CL_COPY_UNBINDABLE) &&
1786 IS_MNT_UNBINDABLE(s)) {
1787 s = skip_mnt_tree(s);
1790 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1791 is_mnt_ns_file(s->mnt.mnt_root)) {
1792 s = skip_mnt_tree(s);
1795 while (p != s->mnt_parent) {
1801 q = clone_mnt(p, p->mnt.mnt_root, flag);
1805 list_add_tail(&q->mnt_list, &res->mnt_list);
1806 attach_mnt(q, parent, p->mnt_mp);
1807 unlock_mount_hash();
1814 umount_tree(res, UMOUNT_SYNC);
1815 unlock_mount_hash();
1820 /* Caller should check returned pointer for errors */
1822 struct vfsmount *collect_mounts(const struct path *path)
1826 if (!check_mnt(real_mount(path->mnt)))
1827 tree = ERR_PTR(-EINVAL);
1829 tree = copy_tree(real_mount(path->mnt), path->dentry,
1830 CL_COPY_ALL | CL_PRIVATE);
1833 return ERR_CAST(tree);
1837 void drop_collected_mounts(struct vfsmount *mnt)
1841 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1842 unlock_mount_hash();
1847 * clone_private_mount - create a private clone of a path
1849 * This creates a new vfsmount, which will be the clone of @path. The new will
1850 * not be attached anywhere in the namespace and will be private (i.e. changes
1851 * to the originating mount won't be propagated into this).
1853 * Release with mntput().
1855 struct vfsmount *clone_private_mount(const struct path *path)
1857 struct mount *old_mnt = real_mount(path->mnt);
1858 struct mount *new_mnt;
1860 if (IS_MNT_UNBINDABLE(old_mnt))
1861 return ERR_PTR(-EINVAL);
1863 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1864 if (IS_ERR(new_mnt))
1865 return ERR_CAST(new_mnt);
1867 return &new_mnt->mnt;
1869 EXPORT_SYMBOL_GPL(clone_private_mount);
1871 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1872 struct vfsmount *root)
1875 int res = f(root, arg);
1878 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1879 res = f(&mnt->mnt, arg);
1886 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1890 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1891 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1892 mnt_release_group_id(p);
1896 static int invent_group_ids(struct mount *mnt, bool recurse)
1900 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1901 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1902 int err = mnt_alloc_group_id(p);
1904 cleanup_group_ids(mnt, p);
1913 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1915 unsigned int max = READ_ONCE(sysctl_mount_max);
1916 unsigned int mounts = 0, old, pending, sum;
1919 for (p = mnt; p; p = next_mnt(p, mnt))
1923 pending = ns->pending_mounts;
1924 sum = old + pending;
1928 (mounts > (max - sum)))
1931 ns->pending_mounts = pending + mounts;
1936 * @source_mnt : mount tree to be attached
1937 * @nd : place the mount tree @source_mnt is attached
1938 * @parent_nd : if non-null, detach the source_mnt from its parent and
1939 * store the parent mount and mountpoint dentry.
1940 * (done when source_mnt is moved)
1942 * NOTE: in the table below explains the semantics when a source mount
1943 * of a given type is attached to a destination mount of a given type.
1944 * ---------------------------------------------------------------------------
1945 * | BIND MOUNT OPERATION |
1946 * |**************************************************************************
1947 * | source-->| shared | private | slave | unbindable |
1951 * |**************************************************************************
1952 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1954 * |non-shared| shared (+) | private | slave (*) | invalid |
1955 * ***************************************************************************
1956 * A bind operation clones the source mount and mounts the clone on the
1957 * destination mount.
1959 * (++) the cloned mount is propagated to all the mounts in the propagation
1960 * tree of the destination mount and the cloned mount is added to
1961 * the peer group of the source mount.
1962 * (+) the cloned mount is created under the destination mount and is marked
1963 * as shared. The cloned mount is added to the peer group of the source
1965 * (+++) the mount is propagated to all the mounts in the propagation tree
1966 * of the destination mount and the cloned mount is made slave
1967 * of the same master as that of the source mount. The cloned mount
1968 * is marked as 'shared and slave'.
1969 * (*) the cloned mount is made a slave of the same master as that of the
1972 * ---------------------------------------------------------------------------
1973 * | MOVE MOUNT OPERATION |
1974 * |**************************************************************************
1975 * | source-->| shared | private | slave | unbindable |
1979 * |**************************************************************************
1980 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1982 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1983 * ***************************************************************************
1985 * (+) the mount is moved to the destination. And is then propagated to
1986 * all the mounts in the propagation tree of the destination mount.
1987 * (+*) the mount is moved to the destination.
1988 * (+++) the mount is moved to the destination and is then propagated to
1989 * all the mounts belonging to the destination mount's propagation tree.
1990 * the mount is marked as 'shared and slave'.
1991 * (*) the mount continues to be a slave at the new location.
1993 * if the source mount is a tree, the operations explained above is
1994 * applied to each mount in the tree.
1995 * Must be called without spinlocks held, since this function can sleep
1998 static int attach_recursive_mnt(struct mount *source_mnt,
1999 struct mount *dest_mnt,
2000 struct mountpoint *dest_mp,
2001 struct path *parent_path)
2003 HLIST_HEAD(tree_list);
2004 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2005 struct mountpoint *smp;
2006 struct mount *child, *p;
2007 struct hlist_node *n;
2010 /* Preallocate a mountpoint in case the new mounts need
2011 * to be tucked under other mounts.
2013 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2015 return PTR_ERR(smp);
2017 /* Is there space to add these mounts to the mount namespace? */
2019 err = count_mounts(ns, source_mnt);
2024 if (IS_MNT_SHARED(dest_mnt)) {
2025 err = invent_group_ids(source_mnt, true);
2028 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2031 goto out_cleanup_ids;
2032 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2038 detach_mnt(source_mnt, parent_path);
2039 attach_mnt(source_mnt, dest_mnt, dest_mp);
2040 touch_mnt_namespace(source_mnt->mnt_ns);
2042 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2043 commit_tree(source_mnt);
2046 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2048 hlist_del_init(&child->mnt_hash);
2049 q = __lookup_mnt(&child->mnt_parent->mnt,
2050 child->mnt_mountpoint);
2052 mnt_change_mountpoint(child, smp, q);
2055 put_mountpoint(smp);
2056 unlock_mount_hash();
2061 while (!hlist_empty(&tree_list)) {
2062 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2063 child->mnt_parent->mnt_ns->pending_mounts = 0;
2064 umount_tree(child, UMOUNT_SYNC);
2066 unlock_mount_hash();
2067 cleanup_group_ids(source_mnt, NULL);
2069 ns->pending_mounts = 0;
2071 read_seqlock_excl(&mount_lock);
2072 put_mountpoint(smp);
2073 read_sequnlock_excl(&mount_lock);
2078 static struct mountpoint *lock_mount(struct path *path)
2080 struct vfsmount *mnt;
2081 struct dentry *dentry = path->dentry;
2083 inode_lock(dentry->d_inode);
2084 if (unlikely(cant_mount(dentry))) {
2085 inode_unlock(dentry->d_inode);
2086 return ERR_PTR(-ENOENT);
2089 mnt = lookup_mnt(path);
2091 struct mountpoint *mp = get_mountpoint(dentry);
2094 inode_unlock(dentry->d_inode);
2100 inode_unlock(path->dentry->d_inode);
2103 dentry = path->dentry = dget(mnt->mnt_root);
2107 static void unlock_mount(struct mountpoint *where)
2109 struct dentry *dentry = where->m_dentry;
2111 read_seqlock_excl(&mount_lock);
2112 put_mountpoint(where);
2113 read_sequnlock_excl(&mount_lock);
2116 inode_unlock(dentry->d_inode);
2119 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2121 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2124 if (d_is_dir(mp->m_dentry) !=
2125 d_is_dir(mnt->mnt.mnt_root))
2128 return attach_recursive_mnt(mnt, p, mp, NULL);
2132 * Sanity check the flags to change_mnt_propagation.
2135 static int flags_to_propagation_type(int flags)
2137 int type = flags & ~(MS_REC | MS_SILENT);
2139 /* Fail if any non-propagation flags are set */
2140 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2142 /* Only one propagation flag should be set */
2143 if (!is_power_of_2(type))
2149 * recursively change the type of the mountpoint.
2151 static int do_change_type(struct path *path, int flag)
2154 struct mount *mnt = real_mount(path->mnt);
2155 int recurse = flag & MS_REC;
2159 if (path->dentry != path->mnt->mnt_root)
2162 type = flags_to_propagation_type(flag);
2167 if (type == MS_SHARED) {
2168 err = invent_group_ids(mnt, recurse);
2174 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2175 change_mnt_propagation(m, type);
2176 unlock_mount_hash();
2183 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2185 struct mount *child;
2186 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2187 if (!is_subdir(child->mnt_mountpoint, dentry))
2190 if (child->mnt.mnt_flags & MNT_LOCKED)
2197 * do loopback mount.
2199 static int do_loopback(struct path *path, const char *old_name,
2202 struct path old_path;
2203 struct mount *mnt = NULL, *old, *parent;
2204 struct mountpoint *mp;
2206 if (!old_name || !*old_name)
2208 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2213 if (mnt_ns_loop(old_path.dentry))
2216 mp = lock_mount(path);
2221 old = real_mount(old_path.mnt);
2222 parent = real_mount(path->mnt);
2225 if (IS_MNT_UNBINDABLE(old))
2228 if (!check_mnt(parent))
2231 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2234 if (!recurse && has_locked_children(old, old_path.dentry))
2238 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2240 mnt = clone_mnt(old, old_path.dentry, 0);
2247 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2249 err = graft_tree(mnt, parent, mp);
2252 umount_tree(mnt, UMOUNT_SYNC);
2253 unlock_mount_hash();
2258 path_put(&old_path);
2262 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2265 int readonly_request = 0;
2267 if (ms_flags & MS_RDONLY)
2268 readonly_request = 1;
2269 if (readonly_request == __mnt_is_readonly(mnt))
2272 if (readonly_request)
2273 error = mnt_make_readonly(real_mount(mnt));
2275 __mnt_unmake_readonly(real_mount(mnt));
2280 * change filesystem flags. dir should be a physical root of filesystem.
2281 * If you've mounted a non-root directory somewhere and want to do remount
2282 * on it - tough luck.
2284 static int do_remount(struct path *path, int flags, int mnt_flags,
2288 struct super_block *sb = path->mnt->mnt_sb;
2289 struct mount *mnt = real_mount(path->mnt);
2291 if (!check_mnt(mnt))
2294 if (path->dentry != path->mnt->mnt_root)
2297 /* Don't allow changing of locked mnt flags.
2299 * No locks need to be held here while testing the various
2300 * MNT_LOCK flags because those flags can never be cleared
2301 * once they are set.
2303 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2304 !(mnt_flags & MNT_READONLY)) {
2307 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2308 !(mnt_flags & MNT_NODEV)) {
2311 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2312 !(mnt_flags & MNT_NOSUID)) {
2315 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2316 !(mnt_flags & MNT_NOEXEC)) {
2319 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2320 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2324 err = security_sb_remount(sb, data);
2328 down_write(&sb->s_umount);
2329 if (flags & MS_BIND)
2330 err = change_mount_flags(path->mnt, flags);
2331 else if (!capable(CAP_SYS_ADMIN))
2334 err = do_remount_sb(sb, flags, data, 0);
2337 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2338 mnt->mnt.mnt_flags = mnt_flags;
2339 touch_mnt_namespace(mnt->mnt_ns);
2340 unlock_mount_hash();
2342 up_write(&sb->s_umount);
2346 static inline int tree_contains_unbindable(struct mount *mnt)
2349 for (p = mnt; p; p = next_mnt(p, mnt)) {
2350 if (IS_MNT_UNBINDABLE(p))
2356 static int do_move_mount(struct path *path, const char *old_name)
2358 struct path old_path, parent_path;
2361 struct mountpoint *mp;
2363 if (!old_name || !*old_name)
2365 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2369 mp = lock_mount(path);
2374 old = real_mount(old_path.mnt);
2375 p = real_mount(path->mnt);
2378 if (!check_mnt(p) || !check_mnt(old))
2381 if (old->mnt.mnt_flags & MNT_LOCKED)
2385 if (old_path.dentry != old_path.mnt->mnt_root)
2388 if (!mnt_has_parent(old))
2391 if (d_is_dir(path->dentry) !=
2392 d_is_dir(old_path.dentry))
2395 * Don't move a mount residing in a shared parent.
2397 if (IS_MNT_SHARED(old->mnt_parent))
2400 * Don't move a mount tree containing unbindable mounts to a destination
2401 * mount which is shared.
2403 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2406 for (; mnt_has_parent(p); p = p->mnt_parent)
2410 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2414 /* if the mount is moved, it should no longer be expire
2416 list_del_init(&old->mnt_expire);
2421 path_put(&parent_path);
2422 path_put(&old_path);
2426 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2429 const char *subtype = strchr(fstype, '.');
2438 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2440 if (!mnt->mnt_sb->s_subtype)
2446 return ERR_PTR(err);
2450 * add a mount into a namespace's mount tree
2452 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2454 struct mountpoint *mp;
2455 struct mount *parent;
2458 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2460 mp = lock_mount(path);
2464 parent = real_mount(path->mnt);
2466 if (unlikely(!check_mnt(parent))) {
2467 /* that's acceptable only for automounts done in private ns */
2468 if (!(mnt_flags & MNT_SHRINKABLE))
2470 /* ... and for those we'd better have mountpoint still alive */
2471 if (!parent->mnt_ns)
2475 /* Refuse the same filesystem on the same mount point */
2477 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2478 path->mnt->mnt_root == path->dentry)
2482 if (d_is_symlink(newmnt->mnt.mnt_root))
2485 newmnt->mnt.mnt_flags = mnt_flags;
2486 err = graft_tree(newmnt, parent, mp);
2493 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2496 * create a new mount for userspace and request it to be added into the
2499 static int do_new_mount(struct path *path, const char *fstype, int flags,
2500 int mnt_flags, const char *name, void *data)
2502 struct file_system_type *type;
2503 struct vfsmount *mnt;
2509 type = get_fs_type(fstype);
2513 mnt = vfs_kern_mount(type, flags, name, data);
2514 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2515 !mnt->mnt_sb->s_subtype)
2516 mnt = fs_set_subtype(mnt, fstype);
2518 put_filesystem(type);
2520 return PTR_ERR(mnt);
2522 if (mount_too_revealing(mnt, &mnt_flags)) {
2527 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2533 int finish_automount(struct vfsmount *m, struct path *path)
2535 struct mount *mnt = real_mount(m);
2537 /* The new mount record should have at least 2 refs to prevent it being
2538 * expired before we get a chance to add it
2540 BUG_ON(mnt_get_count(mnt) < 2);
2542 if (m->mnt_sb == path->mnt->mnt_sb &&
2543 m->mnt_root == path->dentry) {
2548 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2552 /* remove m from any expiration list it may be on */
2553 if (!list_empty(&mnt->mnt_expire)) {
2555 list_del_init(&mnt->mnt_expire);
2564 * mnt_set_expiry - Put a mount on an expiration list
2565 * @mnt: The mount to list.
2566 * @expiry_list: The list to add the mount to.
2568 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2572 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2576 EXPORT_SYMBOL(mnt_set_expiry);
2579 * process a list of expirable mountpoints with the intent of discarding any
2580 * mountpoints that aren't in use and haven't been touched since last we came
2583 void mark_mounts_for_expiry(struct list_head *mounts)
2585 struct mount *mnt, *next;
2586 LIST_HEAD(graveyard);
2588 if (list_empty(mounts))
2594 /* extract from the expiration list every vfsmount that matches the
2595 * following criteria:
2596 * - only referenced by its parent vfsmount
2597 * - still marked for expiry (marked on the last call here; marks are
2598 * cleared by mntput())
2600 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2601 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2602 propagate_mount_busy(mnt, 1))
2604 list_move(&mnt->mnt_expire, &graveyard);
2606 while (!list_empty(&graveyard)) {
2607 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2608 touch_mnt_namespace(mnt->mnt_ns);
2609 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2611 unlock_mount_hash();
2615 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2618 * Ripoff of 'select_parent()'
2620 * search the list of submounts for a given mountpoint, and move any
2621 * shrinkable submounts to the 'graveyard' list.
2623 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2625 struct mount *this_parent = parent;
2626 struct list_head *next;
2630 next = this_parent->mnt_mounts.next;
2632 while (next != &this_parent->mnt_mounts) {
2633 struct list_head *tmp = next;
2634 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2637 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2640 * Descend a level if the d_mounts list is non-empty.
2642 if (!list_empty(&mnt->mnt_mounts)) {
2647 if (!propagate_mount_busy(mnt, 1)) {
2648 list_move_tail(&mnt->mnt_expire, graveyard);
2653 * All done at this level ... ascend and resume the search
2655 if (this_parent != parent) {
2656 next = this_parent->mnt_child.next;
2657 this_parent = this_parent->mnt_parent;
2664 * process a list of expirable mountpoints with the intent of discarding any
2665 * submounts of a specific parent mountpoint
2667 * mount_lock must be held for write
2669 static void shrink_submounts(struct mount *mnt)
2671 LIST_HEAD(graveyard);
2674 /* extract submounts of 'mountpoint' from the expiration list */
2675 while (select_submounts(mnt, &graveyard)) {
2676 while (!list_empty(&graveyard)) {
2677 m = list_first_entry(&graveyard, struct mount,
2679 touch_mnt_namespace(m->mnt_ns);
2680 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2686 * Some copy_from_user() implementations do not return the exact number of
2687 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2688 * Note that this function differs from copy_from_user() in that it will oops
2689 * on bad values of `to', rather than returning a short copy.
2691 static long exact_copy_from_user(void *to, const void __user * from,
2695 const char __user *f = from;
2698 if (!access_ok(VERIFY_READ, from, n))
2702 if (__get_user(c, f)) {
2713 void *copy_mount_options(const void __user * data)
2722 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2724 return ERR_PTR(-ENOMEM);
2726 /* We only care that *some* data at the address the user
2727 * gave us is valid. Just in case, we'll zero
2728 * the remainder of the page.
2730 /* copy_from_user cannot cross TASK_SIZE ! */
2731 size = TASK_SIZE - (unsigned long)data;
2732 if (size > PAGE_SIZE)
2735 i = size - exact_copy_from_user(copy, data, size);
2738 return ERR_PTR(-EFAULT);
2741 memset(copy + i, 0, PAGE_SIZE - i);
2745 char *copy_mount_string(const void __user *data)
2747 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2751 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2752 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2754 * data is a (void *) that can point to any structure up to
2755 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2756 * information (or be NULL).
2758 * Pre-0.97 versions of mount() didn't have a flags word.
2759 * When the flags word was introduced its top half was required
2760 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2761 * Therefore, if this magic number is present, it carries no information
2762 * and must be discarded.
2764 long do_mount(const char *dev_name, const char __user *dir_name,
2765 const char *type_page, unsigned long flags, void *data_page)
2772 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2773 flags &= ~MS_MGC_MSK;
2775 /* Basic sanity checks */
2777 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2779 /* ... and get the mountpoint */
2780 retval = user_path(dir_name, &path);
2784 retval = security_sb_mount(dev_name, &path,
2785 type_page, flags, data_page);
2786 if (!retval && !may_mount())
2788 if (!retval && (flags & MS_MANDLOCK) && !may_mandlock())
2793 /* Default to relatime unless overriden */
2794 if (!(flags & MS_NOATIME))
2795 mnt_flags |= MNT_RELATIME;
2797 /* Separate the per-mountpoint flags */
2798 if (flags & MS_NOSUID)
2799 mnt_flags |= MNT_NOSUID;
2800 if (flags & MS_NODEV)
2801 mnt_flags |= MNT_NODEV;
2802 if (flags & MS_NOEXEC)
2803 mnt_flags |= MNT_NOEXEC;
2804 if (flags & MS_NOATIME)
2805 mnt_flags |= MNT_NOATIME;
2806 if (flags & MS_NODIRATIME)
2807 mnt_flags |= MNT_NODIRATIME;
2808 if (flags & MS_STRICTATIME)
2809 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2810 if (flags & MS_RDONLY)
2811 mnt_flags |= MNT_READONLY;
2813 /* The default atime for remount is preservation */
2814 if ((flags & MS_REMOUNT) &&
2815 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2816 MS_STRICTATIME)) == 0)) {
2817 mnt_flags &= ~MNT_ATIME_MASK;
2818 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2821 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2822 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2823 MS_STRICTATIME | MS_NOREMOTELOCK | MS_SUBMOUNT);
2825 if (flags & MS_REMOUNT)
2826 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2828 else if (flags & MS_BIND)
2829 retval = do_loopback(&path, dev_name, flags & MS_REC);
2830 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2831 retval = do_change_type(&path, flags);
2832 else if (flags & MS_MOVE)
2833 retval = do_move_mount(&path, dev_name);
2835 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2836 dev_name, data_page);
2842 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2844 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2847 static void dec_mnt_namespaces(struct ucounts *ucounts)
2849 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2852 static void free_mnt_ns(struct mnt_namespace *ns)
2854 ns_free_inum(&ns->ns);
2855 dec_mnt_namespaces(ns->ucounts);
2856 put_user_ns(ns->user_ns);
2861 * Assign a sequence number so we can detect when we attempt to bind
2862 * mount a reference to an older mount namespace into the current
2863 * mount namespace, preventing reference counting loops. A 64bit
2864 * number incrementing at 10Ghz will take 12,427 years to wrap which
2865 * is effectively never, so we can ignore the possibility.
2867 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2869 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2871 struct mnt_namespace *new_ns;
2872 struct ucounts *ucounts;
2875 ucounts = inc_mnt_namespaces(user_ns);
2877 return ERR_PTR(-ENOSPC);
2879 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2881 dec_mnt_namespaces(ucounts);
2882 return ERR_PTR(-ENOMEM);
2884 ret = ns_alloc_inum(&new_ns->ns);
2887 dec_mnt_namespaces(ucounts);
2888 return ERR_PTR(ret);
2890 new_ns->ns.ops = &mntns_operations;
2891 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2892 atomic_set(&new_ns->count, 1);
2893 new_ns->root = NULL;
2894 INIT_LIST_HEAD(&new_ns->list);
2895 init_waitqueue_head(&new_ns->poll);
2897 new_ns->user_ns = get_user_ns(user_ns);
2898 new_ns->ucounts = ucounts;
2900 new_ns->pending_mounts = 0;
2905 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2906 struct user_namespace *user_ns, struct fs_struct *new_fs)
2908 struct mnt_namespace *new_ns;
2909 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2910 struct mount *p, *q;
2917 if (likely(!(flags & CLONE_NEWNS))) {
2924 new_ns = alloc_mnt_ns(user_ns);
2929 /* First pass: copy the tree topology */
2930 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2931 if (user_ns != ns->user_ns)
2932 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2933 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2936 free_mnt_ns(new_ns);
2937 return ERR_CAST(new);
2940 list_add_tail(&new_ns->list, &new->mnt_list);
2943 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2944 * as belonging to new namespace. We have already acquired a private
2945 * fs_struct, so tsk->fs->lock is not needed.
2953 if (&p->mnt == new_fs->root.mnt) {
2954 new_fs->root.mnt = mntget(&q->mnt);
2957 if (&p->mnt == new_fs->pwd.mnt) {
2958 new_fs->pwd.mnt = mntget(&q->mnt);
2962 p = next_mnt(p, old);
2963 q = next_mnt(q, new);
2966 while (p->mnt.mnt_root != q->mnt.mnt_root)
2967 p = next_mnt(p, old);
2980 * create_mnt_ns - creates a private namespace and adds a root filesystem
2981 * @mnt: pointer to the new root filesystem mountpoint
2983 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2985 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2986 if (!IS_ERR(new_ns)) {
2987 struct mount *mnt = real_mount(m);
2988 mnt->mnt_ns = new_ns;
2991 list_add(&mnt->mnt_list, &new_ns->list);
2998 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3000 struct mnt_namespace *ns;
3001 struct super_block *s;
3005 ns = create_mnt_ns(mnt);
3007 return ERR_CAST(ns);
3009 err = vfs_path_lookup(mnt->mnt_root, mnt,
3010 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3015 return ERR_PTR(err);
3017 /* trade a vfsmount reference for active sb one */
3018 s = path.mnt->mnt_sb;
3019 atomic_inc(&s->s_active);
3021 /* lock the sucker */
3022 down_write(&s->s_umount);
3023 /* ... and return the root of (sub)tree on it */
3026 EXPORT_SYMBOL(mount_subtree);
3028 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3029 char __user *, type, unsigned long, flags, void __user *, data)
3036 kernel_type = copy_mount_string(type);
3037 ret = PTR_ERR(kernel_type);
3038 if (IS_ERR(kernel_type))
3041 kernel_dev = copy_mount_string(dev_name);
3042 ret = PTR_ERR(kernel_dev);
3043 if (IS_ERR(kernel_dev))
3046 options = copy_mount_options(data);
3047 ret = PTR_ERR(options);
3048 if (IS_ERR(options))
3051 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3063 * Return true if path is reachable from root
3065 * namespace_sem or mount_lock is held
3067 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3068 const struct path *root)
3070 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3071 dentry = mnt->mnt_mountpoint;
3072 mnt = mnt->mnt_parent;
3074 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3077 bool path_is_under(const struct path *path1, const struct path *path2)
3080 read_seqlock_excl(&mount_lock);
3081 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3082 read_sequnlock_excl(&mount_lock);
3085 EXPORT_SYMBOL(path_is_under);
3088 * pivot_root Semantics:
3089 * Moves the root file system of the current process to the directory put_old,
3090 * makes new_root as the new root file system of the current process, and sets
3091 * root/cwd of all processes which had them on the current root to new_root.
3094 * The new_root and put_old must be directories, and must not be on the
3095 * same file system as the current process root. The put_old must be
3096 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3097 * pointed to by put_old must yield the same directory as new_root. No other
3098 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3100 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3101 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3102 * in this situation.
3105 * - we don't move root/cwd if they are not at the root (reason: if something
3106 * cared enough to change them, it's probably wrong to force them elsewhere)
3107 * - it's okay to pick a root that isn't the root of a file system, e.g.
3108 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3109 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3112 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3113 const char __user *, put_old)
3115 struct path new, old, parent_path, root_parent, root;
3116 struct mount *new_mnt, *root_mnt, *old_mnt;
3117 struct mountpoint *old_mp, *root_mp;
3123 error = user_path_dir(new_root, &new);
3127 error = user_path_dir(put_old, &old);
3131 error = security_sb_pivotroot(&old, &new);
3135 get_fs_root(current->fs, &root);
3136 old_mp = lock_mount(&old);
3137 error = PTR_ERR(old_mp);
3142 new_mnt = real_mount(new.mnt);
3143 root_mnt = real_mount(root.mnt);
3144 old_mnt = real_mount(old.mnt);
3145 if (IS_MNT_SHARED(old_mnt) ||
3146 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3147 IS_MNT_SHARED(root_mnt->mnt_parent))
3149 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3151 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3154 if (d_unlinked(new.dentry))
3157 if (new_mnt == root_mnt || old_mnt == root_mnt)
3158 goto out4; /* loop, on the same file system */
3160 if (root.mnt->mnt_root != root.dentry)
3161 goto out4; /* not a mountpoint */
3162 if (!mnt_has_parent(root_mnt))
3163 goto out4; /* not attached */
3164 root_mp = root_mnt->mnt_mp;
3165 if (new.mnt->mnt_root != new.dentry)
3166 goto out4; /* not a mountpoint */
3167 if (!mnt_has_parent(new_mnt))
3168 goto out4; /* not attached */
3169 /* make sure we can reach put_old from new_root */
3170 if (!is_path_reachable(old_mnt, old.dentry, &new))
3172 /* make certain new is below the root */
3173 if (!is_path_reachable(new_mnt, new.dentry, &root))
3175 root_mp->m_count++; /* pin it so it won't go away */
3177 detach_mnt(new_mnt, &parent_path);
3178 detach_mnt(root_mnt, &root_parent);
3179 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3180 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3181 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3183 /* mount old root on put_old */
3184 attach_mnt(root_mnt, old_mnt, old_mp);
3185 /* mount new_root on / */
3186 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3187 touch_mnt_namespace(current->nsproxy->mnt_ns);
3188 /* A moved mount should not expire automatically */
3189 list_del_init(&new_mnt->mnt_expire);
3190 put_mountpoint(root_mp);
3191 unlock_mount_hash();
3192 chroot_fs_refs(&root, &new);
3195 unlock_mount(old_mp);
3197 path_put(&root_parent);
3198 path_put(&parent_path);
3210 static void __init init_mount_tree(void)
3212 struct vfsmount *mnt;
3213 struct mnt_namespace *ns;
3215 struct file_system_type *type;
3217 type = get_fs_type("rootfs");
3219 panic("Can't find rootfs type");
3220 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3221 put_filesystem(type);
3223 panic("Can't create rootfs");
3225 ns = create_mnt_ns(mnt);
3227 panic("Can't allocate initial namespace");
3229 init_task.nsproxy->mnt_ns = ns;
3233 root.dentry = mnt->mnt_root;
3234 mnt->mnt_flags |= MNT_LOCKED;
3236 set_fs_pwd(current->fs, &root);
3237 set_fs_root(current->fs, &root);
3240 void __init mnt_init(void)
3244 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3245 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3247 mount_hashtable = alloc_large_system_hash("Mount-cache",
3248 sizeof(struct hlist_head),
3251 &m_hash_shift, &m_hash_mask, 0, 0);
3252 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3253 sizeof(struct hlist_head),
3256 &mp_hash_shift, &mp_hash_mask, 0, 0);
3258 if (!mount_hashtable || !mountpoint_hashtable)
3259 panic("Failed to allocate mount hash table\n");
3265 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3267 fs_kobj = kobject_create_and_add("fs", NULL);
3269 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3274 void put_mnt_ns(struct mnt_namespace *ns)
3276 if (!atomic_dec_and_test(&ns->count))
3278 drop_collected_mounts(&ns->root->mnt);
3282 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3284 struct vfsmount *mnt;
3285 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3288 * it is a longterm mount, don't release mnt until
3289 * we unmount before file sys is unregistered
3291 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3295 EXPORT_SYMBOL_GPL(kern_mount_data);
3297 void kern_unmount(struct vfsmount *mnt)
3299 /* release long term mount so mount point can be released */
3300 if (!IS_ERR_OR_NULL(mnt)) {
3301 real_mount(mnt)->mnt_ns = NULL;
3302 synchronize_rcu(); /* yecchhh... */
3306 EXPORT_SYMBOL(kern_unmount);
3308 bool our_mnt(struct vfsmount *mnt)
3310 return check_mnt(real_mount(mnt));
3313 bool current_chrooted(void)
3315 /* Does the current process have a non-standard root */
3316 struct path ns_root;
3317 struct path fs_root;
3320 /* Find the namespace root */
3321 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3322 ns_root.dentry = ns_root.mnt->mnt_root;
3324 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3327 get_fs_root(current->fs, &fs_root);
3329 chrooted = !path_equal(&fs_root, &ns_root);
3337 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3340 int new_flags = *new_mnt_flags;
3342 bool visible = false;
3344 down_read(&namespace_sem);
3345 list_for_each_entry(mnt, &ns->list, mnt_list) {
3346 struct mount *child;
3349 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3352 /* This mount is not fully visible if it's root directory
3353 * is not the root directory of the filesystem.
3355 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3358 /* A local view of the mount flags */
3359 mnt_flags = mnt->mnt.mnt_flags;
3361 /* Don't miss readonly hidden in the superblock flags */
3362 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3363 mnt_flags |= MNT_LOCK_READONLY;
3365 /* Verify the mount flags are equal to or more permissive
3366 * than the proposed new mount.
3368 if ((mnt_flags & MNT_LOCK_READONLY) &&
3369 !(new_flags & MNT_READONLY))
3371 if ((mnt_flags & MNT_LOCK_ATIME) &&
3372 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3375 /* This mount is not fully visible if there are any
3376 * locked child mounts that cover anything except for
3377 * empty directories.
3379 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3380 struct inode *inode = child->mnt_mountpoint->d_inode;
3381 /* Only worry about locked mounts */
3382 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3384 /* Is the directory permanetly empty? */
3385 if (!is_empty_dir_inode(inode))
3388 /* Preserve the locked attributes */
3389 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3396 up_read(&namespace_sem);
3400 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3402 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3403 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3404 unsigned long s_iflags;
3406 if (ns->user_ns == &init_user_ns)
3409 /* Can this filesystem be too revealing? */
3410 s_iflags = mnt->mnt_sb->s_iflags;
3411 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3414 if ((s_iflags & required_iflags) != required_iflags) {
3415 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3420 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3423 bool mnt_may_suid(struct vfsmount *mnt)
3426 * Foreign mounts (accessed via fchdir or through /proc
3427 * symlinks) are always treated as if they are nosuid. This
3428 * prevents namespaces from trusting potentially unsafe
3429 * suid/sgid bits, file caps, or security labels that originate
3430 * in other namespaces.
3432 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3433 current_in_userns(mnt->mnt_sb->s_user_ns);
3436 static struct ns_common *mntns_get(struct task_struct *task)
3438 struct ns_common *ns = NULL;
3439 struct nsproxy *nsproxy;
3442 nsproxy = task->nsproxy;
3444 ns = &nsproxy->mnt_ns->ns;
3445 get_mnt_ns(to_mnt_ns(ns));
3452 static void mntns_put(struct ns_common *ns)
3454 put_mnt_ns(to_mnt_ns(ns));
3457 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3459 struct fs_struct *fs = current->fs;
3460 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3464 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3465 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3466 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3473 old_mnt_ns = nsproxy->mnt_ns;
3474 nsproxy->mnt_ns = mnt_ns;
3477 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3478 "/", LOOKUP_DOWN, &root);
3480 /* revert to old namespace */
3481 nsproxy->mnt_ns = old_mnt_ns;
3486 put_mnt_ns(old_mnt_ns);
3488 /* Update the pwd and root */
3489 set_fs_pwd(fs, &root);
3490 set_fs_root(fs, &root);
3496 static struct user_namespace *mntns_owner(struct ns_common *ns)
3498 return to_mnt_ns(ns)->user_ns;
3501 const struct proc_ns_operations mntns_operations = {
3503 .type = CLONE_NEWNS,
3506 .install = mntns_install,
3507 .owner = mntns_owner,