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 /* Maximum number of mounts in a mount namespace */
31 unsigned int sysctl_mount_max __read_mostly = 100000;
33 static unsigned int m_hash_mask __read_mostly;
34 static unsigned int m_hash_shift __read_mostly;
35 static unsigned int mp_hash_mask __read_mostly;
36 static unsigned int mp_hash_shift __read_mostly;
38 static __initdata unsigned long mhash_entries;
39 static int __init set_mhash_entries(char *str)
43 mhash_entries = simple_strtoul(str, &str, 0);
46 __setup("mhash_entries=", set_mhash_entries);
48 static __initdata unsigned long mphash_entries;
49 static int __init set_mphash_entries(char *str)
53 mphash_entries = simple_strtoul(str, &str, 0);
56 __setup("mphash_entries=", set_mphash_entries);
59 static DEFINE_IDA(mnt_id_ida);
60 static DEFINE_IDA(mnt_group_ida);
61 static DEFINE_SPINLOCK(mnt_id_lock);
62 static int mnt_id_start = 0;
63 static int mnt_group_start = 1;
65 static struct hlist_head *mount_hashtable __read_mostly;
66 static struct hlist_head *mountpoint_hashtable __read_mostly;
67 static struct kmem_cache *mnt_cache __read_mostly;
68 static DECLARE_RWSEM(namespace_sem);
71 struct kobject *fs_kobj;
72 EXPORT_SYMBOL_GPL(fs_kobj);
75 * vfsmount lock may be taken for read to prevent changes to the
76 * vfsmount hash, ie. during mountpoint lookups or walking back
79 * It should be taken for write in all cases where the vfsmount
80 * tree or hash is modified or when a vfsmount structure is modified.
82 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
84 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
86 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
87 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
88 tmp = tmp + (tmp >> m_hash_shift);
89 return &mount_hashtable[tmp & m_hash_mask];
92 static inline struct hlist_head *mp_hash(struct dentry *dentry)
94 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
95 tmp = tmp + (tmp >> mp_hash_shift);
96 return &mountpoint_hashtable[tmp & mp_hash_mask];
100 * allocation is serialized by namespace_sem, but we need the spinlock to
101 * serialize with freeing.
103 static int mnt_alloc_id(struct mount *mnt)
108 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
109 spin_lock(&mnt_id_lock);
110 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
112 mnt_id_start = mnt->mnt_id + 1;
113 spin_unlock(&mnt_id_lock);
120 static void mnt_free_id(struct mount *mnt)
122 int id = mnt->mnt_id;
123 spin_lock(&mnt_id_lock);
124 ida_remove(&mnt_id_ida, id);
125 if (mnt_id_start > id)
127 spin_unlock(&mnt_id_lock);
131 * Allocate a new peer group ID
133 * mnt_group_ida is protected by namespace_sem
135 static int mnt_alloc_group_id(struct mount *mnt)
139 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
142 res = ida_get_new_above(&mnt_group_ida,
146 mnt_group_start = mnt->mnt_group_id + 1;
152 * Release a peer group ID
154 void mnt_release_group_id(struct mount *mnt)
156 int id = mnt->mnt_group_id;
157 ida_remove(&mnt_group_ida, id);
158 if (mnt_group_start > id)
159 mnt_group_start = id;
160 mnt->mnt_group_id = 0;
164 * vfsmount lock must be held for read
166 static inline void mnt_add_count(struct mount *mnt, int n)
169 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
178 * vfsmount lock must be held for write
180 unsigned int mnt_get_count(struct mount *mnt)
183 unsigned int count = 0;
186 for_each_possible_cpu(cpu) {
187 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
192 return mnt->mnt_count;
196 static void drop_mountpoint(struct fs_pin *p)
198 struct mount *m = container_of(p, struct mount, mnt_umount);
199 dput(m->mnt_ex_mountpoint);
204 static struct mount *alloc_vfsmnt(const char *name)
206 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
210 err = mnt_alloc_id(mnt);
215 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
216 if (!mnt->mnt_devname)
221 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
223 goto out_free_devname;
225 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
228 mnt->mnt_writers = 0;
230 mnt->mnt.data = NULL;
232 INIT_HLIST_NODE(&mnt->mnt_hash);
233 INIT_LIST_HEAD(&mnt->mnt_child);
234 INIT_LIST_HEAD(&mnt->mnt_mounts);
235 INIT_LIST_HEAD(&mnt->mnt_list);
236 INIT_LIST_HEAD(&mnt->mnt_expire);
237 INIT_LIST_HEAD(&mnt->mnt_share);
238 INIT_LIST_HEAD(&mnt->mnt_slave_list);
239 INIT_LIST_HEAD(&mnt->mnt_slave);
240 INIT_HLIST_NODE(&mnt->mnt_mp_list);
241 INIT_LIST_HEAD(&mnt->mnt_umounting);
242 #ifdef CONFIG_FSNOTIFY
243 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
245 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
251 kfree_const(mnt->mnt_devname);
256 kmem_cache_free(mnt_cache, mnt);
261 * Most r/o checks on a fs are for operations that take
262 * discrete amounts of time, like a write() or unlink().
263 * We must keep track of when those operations start
264 * (for permission checks) and when they end, so that
265 * we can determine when writes are able to occur to
269 * __mnt_is_readonly: check whether a mount is read-only
270 * @mnt: the mount to check for its write status
272 * This shouldn't be used directly ouside of the VFS.
273 * It does not guarantee that the filesystem will stay
274 * r/w, just that it is right *now*. This can not and
275 * should not be used in place of IS_RDONLY(inode).
276 * mnt_want/drop_write() will _keep_ the filesystem
279 int __mnt_is_readonly(struct vfsmount *mnt)
281 if (mnt->mnt_flags & MNT_READONLY)
283 if (mnt->mnt_sb->s_flags & MS_RDONLY)
287 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
289 static inline void mnt_inc_writers(struct mount *mnt)
292 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
298 static inline void mnt_dec_writers(struct mount *mnt)
301 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
307 static unsigned int mnt_get_writers(struct mount *mnt)
310 unsigned int count = 0;
313 for_each_possible_cpu(cpu) {
314 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
319 return mnt->mnt_writers;
323 static int mnt_is_readonly(struct vfsmount *mnt)
325 if (mnt->mnt_sb->s_readonly_remount)
327 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
329 return __mnt_is_readonly(mnt);
333 * Most r/o & frozen checks on a fs are for operations that take discrete
334 * amounts of time, like a write() or unlink(). We must keep track of when
335 * those operations start (for permission checks) and when they end, so that we
336 * can determine when writes are able to occur to a filesystem.
339 * __mnt_want_write - get write access to a mount without freeze protection
340 * @m: the mount on which to take a write
342 * This tells the low-level filesystem that a write is about to be performed to
343 * it, and makes sure that writes are allowed (mnt it read-write) before
344 * returning success. This operation does not protect against filesystem being
345 * frozen. When the write operation is finished, __mnt_drop_write() must be
346 * called. This is effectively a refcount.
348 int __mnt_want_write(struct vfsmount *m)
350 struct mount *mnt = real_mount(m);
354 mnt_inc_writers(mnt);
356 * The store to mnt_inc_writers must be visible before we pass
357 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
358 * incremented count after it has set MNT_WRITE_HOLD.
361 while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
364 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
365 * be set to match its requirements. So we must not load that until
366 * MNT_WRITE_HOLD is cleared.
369 if (mnt_is_readonly(m)) {
370 mnt_dec_writers(mnt);
379 * mnt_want_write - get write access to a mount
380 * @m: the mount on which to take a write
382 * This tells the low-level filesystem that a write is about to be performed to
383 * it, and makes sure that writes are allowed (mount is read-write, filesystem
384 * is not frozen) before returning success. When the write operation is
385 * finished, mnt_drop_write() must be called. This is effectively a refcount.
387 int mnt_want_write(struct vfsmount *m)
391 sb_start_write(m->mnt_sb);
392 ret = __mnt_want_write(m);
394 sb_end_write(m->mnt_sb);
397 EXPORT_SYMBOL_GPL(mnt_want_write);
400 * mnt_clone_write - get write access to a mount
401 * @mnt: the mount on which to take a write
403 * This is effectively like mnt_want_write, except
404 * it must only be used to take an extra write reference
405 * on a mountpoint that we already know has a write reference
406 * on it. This allows some optimisation.
408 * After finished, mnt_drop_write must be called as usual to
409 * drop the reference.
411 int mnt_clone_write(struct vfsmount *mnt)
413 /* superblock may be r/o */
414 if (__mnt_is_readonly(mnt))
417 mnt_inc_writers(real_mount(mnt));
421 EXPORT_SYMBOL_GPL(mnt_clone_write);
424 * __mnt_want_write_file - get write access to a file's mount
425 * @file: the file who's mount on which to take a write
427 * This is like __mnt_want_write, but it takes a file and can
428 * do some optimisations if the file is open for write already
430 int __mnt_want_write_file(struct file *file)
432 if (!(file->f_mode & FMODE_WRITER))
433 return __mnt_want_write(file->f_path.mnt);
435 return mnt_clone_write(file->f_path.mnt);
439 * mnt_want_write_file - get write access to a file's mount
440 * @file: the file who's mount on which to take a write
442 * This is like mnt_want_write, but it takes a file and can
443 * do some optimisations if the file is open for write already
445 int mnt_want_write_file(struct file *file)
449 sb_start_write(file->f_path.mnt->mnt_sb);
450 ret = __mnt_want_write_file(file);
452 sb_end_write(file->f_path.mnt->mnt_sb);
455 EXPORT_SYMBOL_GPL(mnt_want_write_file);
458 * __mnt_drop_write - give up write access to a mount
459 * @mnt: the mount on which to give up write access
461 * Tells the low-level filesystem that we are done
462 * performing writes to it. Must be matched with
463 * __mnt_want_write() call above.
465 void __mnt_drop_write(struct vfsmount *mnt)
468 mnt_dec_writers(real_mount(mnt));
473 * mnt_drop_write - give up write access to a mount
474 * @mnt: the mount on which to give up write access
476 * Tells the low-level filesystem that we are done performing writes to it and
477 * also allows filesystem to be frozen again. Must be matched with
478 * mnt_want_write() call above.
480 void mnt_drop_write(struct vfsmount *mnt)
482 __mnt_drop_write(mnt);
483 sb_end_write(mnt->mnt_sb);
485 EXPORT_SYMBOL_GPL(mnt_drop_write);
487 void __mnt_drop_write_file(struct file *file)
489 __mnt_drop_write(file->f_path.mnt);
492 void mnt_drop_write_file(struct file *file)
494 mnt_drop_write(file->f_path.mnt);
496 EXPORT_SYMBOL(mnt_drop_write_file);
498 static int mnt_make_readonly(struct mount *mnt)
503 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
505 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
506 * should be visible before we do.
511 * With writers on hold, if this value is zero, then there are
512 * definitely no active writers (although held writers may subsequently
513 * increment the count, they'll have to wait, and decrement it after
514 * seeing MNT_READONLY).
516 * It is OK to have counter incremented on one CPU and decremented on
517 * another: the sum will add up correctly. The danger would be when we
518 * sum up each counter, if we read a counter before it is incremented,
519 * but then read another CPU's count which it has been subsequently
520 * decremented from -- we would see more decrements than we should.
521 * MNT_WRITE_HOLD protects against this scenario, because
522 * mnt_want_write first increments count, then smp_mb, then spins on
523 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
524 * we're counting up here.
526 if (mnt_get_writers(mnt) > 0)
529 mnt->mnt.mnt_flags |= MNT_READONLY;
531 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
532 * that become unheld will see MNT_READONLY.
535 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
540 static void __mnt_unmake_readonly(struct mount *mnt)
543 mnt->mnt.mnt_flags &= ~MNT_READONLY;
547 int sb_prepare_remount_readonly(struct super_block *sb)
552 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
553 if (atomic_long_read(&sb->s_remove_count))
557 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
558 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
559 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
561 if (mnt_get_writers(mnt) > 0) {
567 if (!err && atomic_long_read(&sb->s_remove_count))
571 sb->s_readonly_remount = 1;
574 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
575 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
576 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
583 static void free_vfsmnt(struct mount *mnt)
585 kfree(mnt->mnt.data);
586 kfree_const(mnt->mnt_devname);
588 free_percpu(mnt->mnt_pcp);
590 kmem_cache_free(mnt_cache, mnt);
593 static void delayed_free_vfsmnt(struct rcu_head *head)
595 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
598 /* call under rcu_read_lock */
599 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
602 if (read_seqretry(&mount_lock, seq))
606 mnt = real_mount(bastard);
607 mnt_add_count(mnt, 1);
608 smp_mb(); // see mntput_no_expire()
609 if (likely(!read_seqretry(&mount_lock, seq)))
611 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
612 mnt_add_count(mnt, -1);
616 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
617 mnt_add_count(mnt, -1);
622 /* caller will mntput() */
626 /* call under rcu_read_lock */
627 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
629 int res = __legitimize_mnt(bastard, seq);
632 if (unlikely(res < 0)) {
641 * find the first mount at @dentry on vfsmount @mnt.
642 * call under rcu_read_lock()
644 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
646 struct hlist_head *head = m_hash(mnt, dentry);
649 hlist_for_each_entry_rcu(p, head, mnt_hash)
650 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
656 * lookup_mnt - Return the first child mount mounted at path
658 * "First" means first mounted chronologically. If you create the
661 * mount /dev/sda1 /mnt
662 * mount /dev/sda2 /mnt
663 * mount /dev/sda3 /mnt
665 * Then lookup_mnt() on the base /mnt dentry in the root mount will
666 * return successively the root dentry and vfsmount of /dev/sda1, then
667 * /dev/sda2, then /dev/sda3, then NULL.
669 * lookup_mnt takes a reference to the found vfsmount.
671 struct vfsmount *lookup_mnt(struct path *path)
673 struct mount *child_mnt;
679 seq = read_seqbegin(&mount_lock);
680 child_mnt = __lookup_mnt(path->mnt, path->dentry);
681 m = child_mnt ? &child_mnt->mnt : NULL;
682 } while (!legitimize_mnt(m, seq));
688 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
689 * current mount namespace.
691 * The common case is dentries are not mountpoints at all and that
692 * test is handled inline. For the slow case when we are actually
693 * dealing with a mountpoint of some kind, walk through all of the
694 * mounts in the current mount namespace and test to see if the dentry
697 * The mount_hashtable is not usable in the context because we
698 * need to identify all mounts that may be in the current mount
699 * namespace not just a mount that happens to have some specified
702 bool __is_local_mountpoint(struct dentry *dentry)
704 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
706 bool is_covered = false;
708 if (!d_mountpoint(dentry))
711 down_read(&namespace_sem);
712 list_for_each_entry(mnt, &ns->list, mnt_list) {
713 is_covered = (mnt->mnt_mountpoint == dentry);
717 up_read(&namespace_sem);
722 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
724 struct hlist_head *chain = mp_hash(dentry);
725 struct mountpoint *mp;
727 hlist_for_each_entry(mp, chain, m_hash) {
728 if (mp->m_dentry == dentry) {
729 /* might be worth a WARN_ON() */
730 if (d_unlinked(dentry))
731 return ERR_PTR(-ENOENT);
739 static struct mountpoint *get_mountpoint(struct dentry *dentry)
741 struct mountpoint *mp, *new = NULL;
744 if (d_mountpoint(dentry)) {
746 read_seqlock_excl(&mount_lock);
747 mp = lookup_mountpoint(dentry);
748 read_sequnlock_excl(&mount_lock);
754 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
756 return ERR_PTR(-ENOMEM);
759 /* Exactly one processes may set d_mounted */
760 ret = d_set_mounted(dentry);
762 /* Someone else set d_mounted? */
766 /* The dentry is not available as a mountpoint? */
771 /* Add the new mountpoint to the hash table */
772 read_seqlock_excl(&mount_lock);
773 new->m_dentry = dentry;
775 hlist_add_head(&new->m_hash, mp_hash(dentry));
776 INIT_HLIST_HEAD(&new->m_list);
777 read_sequnlock_excl(&mount_lock);
786 static void put_mountpoint(struct mountpoint *mp)
788 if (!--mp->m_count) {
789 struct dentry *dentry = mp->m_dentry;
790 BUG_ON(!hlist_empty(&mp->m_list));
791 spin_lock(&dentry->d_lock);
792 dentry->d_flags &= ~DCACHE_MOUNTED;
793 spin_unlock(&dentry->d_lock);
794 hlist_del(&mp->m_hash);
799 static inline int check_mnt(struct mount *mnt)
801 return mnt->mnt_ns == current->nsproxy->mnt_ns;
805 * vfsmount lock must be held for write
807 static void touch_mnt_namespace(struct mnt_namespace *ns)
811 wake_up_interruptible(&ns->poll);
816 * vfsmount lock must be held for write
818 static void __touch_mnt_namespace(struct mnt_namespace *ns)
820 if (ns && ns->event != event) {
822 wake_up_interruptible(&ns->poll);
827 * vfsmount lock must be held for write
829 static void unhash_mnt(struct mount *mnt)
831 mnt->mnt_parent = mnt;
832 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
833 list_del_init(&mnt->mnt_child);
834 hlist_del_init_rcu(&mnt->mnt_hash);
835 hlist_del_init(&mnt->mnt_mp_list);
836 put_mountpoint(mnt->mnt_mp);
841 * vfsmount lock must be held for write
843 static void detach_mnt(struct mount *mnt, struct path *old_path)
845 old_path->dentry = mnt->mnt_mountpoint;
846 old_path->mnt = &mnt->mnt_parent->mnt;
851 * vfsmount lock must be held for write
853 static void umount_mnt(struct mount *mnt)
855 /* old mountpoint will be dropped when we can do that */
856 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
861 * vfsmount lock must be held for write
863 void mnt_set_mountpoint(struct mount *mnt,
864 struct mountpoint *mp,
865 struct mount *child_mnt)
868 mnt_add_count(mnt, 1); /* essentially, that's mntget */
869 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
870 child_mnt->mnt_parent = mnt;
871 child_mnt->mnt_mp = mp;
872 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
875 static void __attach_mnt(struct mount *mnt, struct mount *parent)
877 hlist_add_head_rcu(&mnt->mnt_hash,
878 m_hash(&parent->mnt, mnt->mnt_mountpoint));
879 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
883 * vfsmount lock must be held for write
885 static void attach_mnt(struct mount *mnt,
886 struct mount *parent,
887 struct mountpoint *mp)
889 mnt_set_mountpoint(parent, mp, mnt);
890 __attach_mnt(mnt, parent);
893 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
895 struct mountpoint *old_mp = mnt->mnt_mp;
896 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
897 struct mount *old_parent = mnt->mnt_parent;
899 list_del_init(&mnt->mnt_child);
900 hlist_del_init(&mnt->mnt_mp_list);
901 hlist_del_init_rcu(&mnt->mnt_hash);
903 attach_mnt(mnt, parent, mp);
905 put_mountpoint(old_mp);
908 * Safely avoid even the suggestion this code might sleep or
909 * lock the mount hash by taking advantage of the knowledge that
910 * mnt_change_mountpoint will not release the final reference
913 * During mounting, the mount passed in as the parent mount will
914 * continue to use the old mountpoint and during unmounting, the
915 * old mountpoint will continue to exist until namespace_unlock,
916 * which happens well after mnt_change_mountpoint.
918 spin_lock(&old_mountpoint->d_lock);
919 old_mountpoint->d_lockref.count--;
920 spin_unlock(&old_mountpoint->d_lock);
922 mnt_add_count(old_parent, -1);
926 * vfsmount lock must be held for write
928 static void commit_tree(struct mount *mnt)
930 struct mount *parent = mnt->mnt_parent;
933 struct mnt_namespace *n = parent->mnt_ns;
935 BUG_ON(parent == mnt);
937 list_add_tail(&head, &mnt->mnt_list);
938 list_for_each_entry(m, &head, mnt_list)
941 list_splice(&head, n->list.prev);
943 n->mounts += n->pending_mounts;
944 n->pending_mounts = 0;
946 __attach_mnt(mnt, parent);
947 touch_mnt_namespace(n);
950 static struct mount *next_mnt(struct mount *p, struct mount *root)
952 struct list_head *next = p->mnt_mounts.next;
953 if (next == &p->mnt_mounts) {
957 next = p->mnt_child.next;
958 if (next != &p->mnt_parent->mnt_mounts)
963 return list_entry(next, struct mount, mnt_child);
966 static struct mount *skip_mnt_tree(struct mount *p)
968 struct list_head *prev = p->mnt_mounts.prev;
969 while (prev != &p->mnt_mounts) {
970 p = list_entry(prev, struct mount, mnt_child);
971 prev = p->mnt_mounts.prev;
977 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
983 return ERR_PTR(-ENODEV);
985 mnt = alloc_vfsmnt(name);
987 return ERR_PTR(-ENOMEM);
989 if (type->alloc_mnt_data) {
990 mnt->mnt.data = type->alloc_mnt_data();
991 if (!mnt->mnt.data) {
994 return ERR_PTR(-ENOMEM);
997 if (flags & MS_KERNMOUNT)
998 mnt->mnt.mnt_flags = MNT_INTERNAL;
1000 root = mount_fs(type, flags, name, &mnt->mnt, data);
1004 return ERR_CAST(root);
1007 mnt->mnt.mnt_root = root;
1008 mnt->mnt.mnt_sb = root->d_sb;
1009 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1010 mnt->mnt_parent = mnt;
1012 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1013 unlock_mount_hash();
1016 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1018 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1021 struct super_block *sb = old->mnt.mnt_sb;
1025 mnt = alloc_vfsmnt(old->mnt_devname);
1027 return ERR_PTR(-ENOMEM);
1029 if (sb->s_op->clone_mnt_data) {
1030 mnt->mnt.data = sb->s_op->clone_mnt_data(old->mnt.data);
1031 if (!mnt->mnt.data) {
1037 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1038 mnt->mnt_group_id = 0; /* not a peer of original */
1040 mnt->mnt_group_id = old->mnt_group_id;
1042 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1043 err = mnt_alloc_group_id(mnt);
1048 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1049 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1050 /* Don't allow unprivileged users to change mount flags */
1051 if (flag & CL_UNPRIVILEGED) {
1052 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1054 if (mnt->mnt.mnt_flags & MNT_READONLY)
1055 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1057 if (mnt->mnt.mnt_flags & MNT_NODEV)
1058 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1060 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1061 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1063 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1064 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1067 /* Don't allow unprivileged users to reveal what is under a mount */
1068 if ((flag & CL_UNPRIVILEGED) &&
1069 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1070 mnt->mnt.mnt_flags |= MNT_LOCKED;
1072 atomic_inc(&sb->s_active);
1073 mnt->mnt.mnt_sb = sb;
1074 mnt->mnt.mnt_root = dget(root);
1075 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1076 mnt->mnt_parent = mnt;
1078 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1079 unlock_mount_hash();
1081 if ((flag & CL_SLAVE) ||
1082 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1083 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1084 mnt->mnt_master = old;
1085 CLEAR_MNT_SHARED(mnt);
1086 } else if (!(flag & CL_PRIVATE)) {
1087 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1088 list_add(&mnt->mnt_share, &old->mnt_share);
1089 if (IS_MNT_SLAVE(old))
1090 list_add(&mnt->mnt_slave, &old->mnt_slave);
1091 mnt->mnt_master = old->mnt_master;
1093 if (flag & CL_MAKE_SHARED)
1094 set_mnt_shared(mnt);
1096 /* stick the duplicate mount on the same expiry list
1097 * as the original if that was on one */
1098 if (flag & CL_EXPIRE) {
1099 if (!list_empty(&old->mnt_expire))
1100 list_add(&mnt->mnt_expire, &old->mnt_expire);
1108 return ERR_PTR(err);
1111 static void cleanup_mnt(struct mount *mnt)
1114 * This probably indicates that somebody messed
1115 * up a mnt_want/drop_write() pair. If this
1116 * happens, the filesystem was probably unable
1117 * to make r/w->r/o transitions.
1120 * The locking used to deal with mnt_count decrement provides barriers,
1121 * so mnt_get_writers() below is safe.
1123 WARN_ON(mnt_get_writers(mnt));
1124 if (unlikely(mnt->mnt_pins.first))
1126 fsnotify_vfsmount_delete(&mnt->mnt);
1127 dput(mnt->mnt.mnt_root);
1128 deactivate_super(mnt->mnt.mnt_sb);
1130 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1133 static void __cleanup_mnt(struct rcu_head *head)
1135 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1138 static LLIST_HEAD(delayed_mntput_list);
1139 static void delayed_mntput(struct work_struct *unused)
1141 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1142 struct llist_node *next;
1144 for (; node; node = next) {
1145 next = llist_next(node);
1146 cleanup_mnt(llist_entry(node, struct mount, mnt_llist));
1149 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1151 static void mntput_no_expire(struct mount *mnt)
1154 if (likely(READ_ONCE(mnt->mnt_ns))) {
1156 * Since we don't do lock_mount_hash() here,
1157 * ->mnt_ns can change under us. However, if it's
1158 * non-NULL, then there's a reference that won't
1159 * be dropped until after an RCU delay done after
1160 * turning ->mnt_ns NULL. So if we observe it
1161 * non-NULL under rcu_read_lock(), the reference
1162 * we are dropping is not the final one.
1164 mnt_add_count(mnt, -1);
1170 * make sure that if __legitimize_mnt() has not seen us grab
1171 * mount_lock, we'll see their refcount increment here.
1174 mnt_add_count(mnt, -1);
1175 if (mnt_get_count(mnt)) {
1177 unlock_mount_hash();
1180 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1182 unlock_mount_hash();
1185 mnt->mnt.mnt_flags |= MNT_DOOMED;
1188 list_del(&mnt->mnt_instance);
1190 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1191 struct mount *p, *tmp;
1192 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1196 unlock_mount_hash();
1198 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1199 struct task_struct *task = current;
1200 if (likely(!(task->flags & PF_KTHREAD))) {
1201 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1202 if (!task_work_add(task, &mnt->mnt_rcu, true))
1205 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1206 schedule_delayed_work(&delayed_mntput_work, 1);
1212 void mntput(struct vfsmount *mnt)
1215 struct mount *m = real_mount(mnt);
1216 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1217 if (unlikely(m->mnt_expiry_mark))
1218 m->mnt_expiry_mark = 0;
1219 mntput_no_expire(m);
1222 EXPORT_SYMBOL(mntput);
1224 struct vfsmount *mntget(struct vfsmount *mnt)
1227 mnt_add_count(real_mount(mnt), 1);
1230 EXPORT_SYMBOL(mntget);
1232 struct vfsmount *mnt_clone_internal(struct path *path)
1235 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1238 p->mnt.mnt_flags |= MNT_INTERNAL;
1242 static inline void mangle(struct seq_file *m, const char *s)
1244 seq_escape(m, s, " \t\n\\");
1248 * Simple .show_options callback for filesystems which don't want to
1249 * implement more complex mount option showing.
1251 * See also save_mount_options().
1253 int generic_show_options(struct seq_file *m, struct dentry *root)
1255 const char *options;
1258 options = rcu_dereference(root->d_sb->s_options);
1260 if (options != NULL && options[0]) {
1268 EXPORT_SYMBOL(generic_show_options);
1271 * If filesystem uses generic_show_options(), this function should be
1272 * called from the fill_super() callback.
1274 * The .remount_fs callback usually needs to be handled in a special
1275 * way, to make sure, that previous options are not overwritten if the
1278 * Also note, that if the filesystem's .remount_fs function doesn't
1279 * reset all options to their default value, but changes only newly
1280 * given options, then the displayed options will not reflect reality
1283 void save_mount_options(struct super_block *sb, char *options)
1285 BUG_ON(sb->s_options);
1286 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1288 EXPORT_SYMBOL(save_mount_options);
1290 void replace_mount_options(struct super_block *sb, char *options)
1292 char *old = sb->s_options;
1293 rcu_assign_pointer(sb->s_options, options);
1299 EXPORT_SYMBOL(replace_mount_options);
1301 #ifdef CONFIG_PROC_FS
1302 /* iterator; we want it to have access to namespace_sem, thus here... */
1303 static void *m_start(struct seq_file *m, loff_t *pos)
1305 struct proc_mounts *p = m->private;
1307 down_read(&namespace_sem);
1308 if (p->cached_event == p->ns->event) {
1309 void *v = p->cached_mount;
1310 if (*pos == p->cached_index)
1312 if (*pos == p->cached_index + 1) {
1313 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1314 return p->cached_mount = v;
1318 p->cached_event = p->ns->event;
1319 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1320 p->cached_index = *pos;
1321 return p->cached_mount;
1324 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1326 struct proc_mounts *p = m->private;
1328 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1329 p->cached_index = *pos;
1330 return p->cached_mount;
1333 static void m_stop(struct seq_file *m, void *v)
1335 up_read(&namespace_sem);
1338 static int m_show(struct seq_file *m, void *v)
1340 struct proc_mounts *p = m->private;
1341 struct mount *r = list_entry(v, struct mount, mnt_list);
1342 return p->show(m, &r->mnt);
1345 const struct seq_operations mounts_op = {
1351 #endif /* CONFIG_PROC_FS */
1354 * may_umount_tree - check if a mount tree is busy
1355 * @mnt: root of mount tree
1357 * This is called to check if a tree of mounts has any
1358 * open files, pwds, chroots or sub mounts that are
1361 int may_umount_tree(struct vfsmount *m)
1363 struct mount *mnt = real_mount(m);
1364 int actual_refs = 0;
1365 int minimum_refs = 0;
1369 /* write lock needed for mnt_get_count */
1371 for (p = mnt; p; p = next_mnt(p, mnt)) {
1372 actual_refs += mnt_get_count(p);
1375 unlock_mount_hash();
1377 if (actual_refs > minimum_refs)
1383 EXPORT_SYMBOL(may_umount_tree);
1386 * may_umount - check if a mount point is busy
1387 * @mnt: root of mount
1389 * This is called to check if a mount point has any
1390 * open files, pwds, chroots or sub mounts. If the
1391 * mount has sub mounts this will return busy
1392 * regardless of whether the sub mounts are busy.
1394 * Doesn't take quota and stuff into account. IOW, in some cases it will
1395 * give false negatives. The main reason why it's here is that we need
1396 * a non-destructive way to look for easily umountable filesystems.
1398 int may_umount(struct vfsmount *mnt)
1401 down_read(&namespace_sem);
1403 if (propagate_mount_busy(real_mount(mnt), 2))
1405 unlock_mount_hash();
1406 up_read(&namespace_sem);
1410 EXPORT_SYMBOL(may_umount);
1412 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1414 static void namespace_unlock(void)
1416 struct hlist_head head;
1418 hlist_move_list(&unmounted, &head);
1420 up_write(&namespace_sem);
1422 if (likely(hlist_empty(&head)))
1427 group_pin_kill(&head);
1430 static inline void namespace_lock(void)
1432 down_write(&namespace_sem);
1435 enum umount_tree_flags {
1437 UMOUNT_PROPAGATE = 2,
1438 UMOUNT_CONNECTED = 4,
1441 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1443 /* Leaving mounts connected is only valid for lazy umounts */
1444 if (how & UMOUNT_SYNC)
1447 /* A mount without a parent has nothing to be connected to */
1448 if (!mnt_has_parent(mnt))
1451 /* Because the reference counting rules change when mounts are
1452 * unmounted and connected, umounted mounts may not be
1453 * connected to mounted mounts.
1455 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1458 /* Has it been requested that the mount remain connected? */
1459 if (how & UMOUNT_CONNECTED)
1462 /* Is the mount locked such that it needs to remain connected? */
1463 if (IS_MNT_LOCKED(mnt))
1466 /* By default disconnect the mount */
1471 * mount_lock must be held
1472 * namespace_sem must be held for write
1474 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1476 LIST_HEAD(tmp_list);
1479 if (how & UMOUNT_PROPAGATE)
1480 propagate_mount_unlock(mnt);
1482 /* Gather the mounts to umount */
1483 for (p = mnt; p; p = next_mnt(p, mnt)) {
1484 p->mnt.mnt_flags |= MNT_UMOUNT;
1485 list_move(&p->mnt_list, &tmp_list);
1488 /* Hide the mounts from mnt_mounts */
1489 list_for_each_entry(p, &tmp_list, mnt_list) {
1490 list_del_init(&p->mnt_child);
1493 /* Add propogated mounts to the tmp_list */
1494 if (how & UMOUNT_PROPAGATE)
1495 propagate_umount(&tmp_list);
1497 while (!list_empty(&tmp_list)) {
1498 struct mnt_namespace *ns;
1500 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1501 list_del_init(&p->mnt_expire);
1502 list_del_init(&p->mnt_list);
1506 __touch_mnt_namespace(ns);
1509 if (how & UMOUNT_SYNC)
1510 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1512 disconnect = disconnect_mount(p, how);
1514 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1515 disconnect ? &unmounted : NULL);
1516 if (mnt_has_parent(p)) {
1517 mnt_add_count(p->mnt_parent, -1);
1519 /* Don't forget about p */
1520 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1525 change_mnt_propagation(p, MS_PRIVATE);
1529 static void shrink_submounts(struct mount *mnt);
1531 static int do_umount(struct mount *mnt, int flags)
1533 struct super_block *sb = mnt->mnt.mnt_sb;
1536 retval = security_sb_umount(&mnt->mnt, flags);
1541 * Allow userspace to request a mountpoint be expired rather than
1542 * unmounting unconditionally. Unmount only happens if:
1543 * (1) the mark is already set (the mark is cleared by mntput())
1544 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1546 if (flags & MNT_EXPIRE) {
1547 if (&mnt->mnt == current->fs->root.mnt ||
1548 flags & (MNT_FORCE | MNT_DETACH))
1552 * probably don't strictly need the lock here if we examined
1553 * all race cases, but it's a slowpath.
1556 if (mnt_get_count(mnt) != 2) {
1557 unlock_mount_hash();
1560 unlock_mount_hash();
1562 if (!xchg(&mnt->mnt_expiry_mark, 1))
1567 * If we may have to abort operations to get out of this
1568 * mount, and they will themselves hold resources we must
1569 * allow the fs to do things. In the Unix tradition of
1570 * 'Gee thats tricky lets do it in userspace' the umount_begin
1571 * might fail to complete on the first run through as other tasks
1572 * must return, and the like. Thats for the mount program to worry
1573 * about for the moment.
1576 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1577 sb->s_op->umount_begin(sb);
1581 * No sense to grab the lock for this test, but test itself looks
1582 * somewhat bogus. Suggestions for better replacement?
1583 * Ho-hum... In principle, we might treat that as umount + switch
1584 * to rootfs. GC would eventually take care of the old vfsmount.
1585 * Actually it makes sense, especially if rootfs would contain a
1586 * /reboot - static binary that would close all descriptors and
1587 * call reboot(9). Then init(8) could umount root and exec /reboot.
1589 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1591 * Special case for "unmounting" root ...
1592 * we just try to remount it readonly.
1594 if (!capable(CAP_SYS_ADMIN))
1596 down_write(&sb->s_umount);
1597 if (!(sb->s_flags & MS_RDONLY))
1598 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1599 up_write(&sb->s_umount);
1606 /* Recheck MNT_LOCKED with the locks held */
1608 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1612 if (flags & MNT_DETACH) {
1613 if (!list_empty(&mnt->mnt_list))
1614 umount_tree(mnt, UMOUNT_PROPAGATE);
1617 shrink_submounts(mnt);
1619 if (!propagate_mount_busy(mnt, 2)) {
1620 if (!list_empty(&mnt->mnt_list))
1621 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1626 unlock_mount_hash();
1632 * __detach_mounts - lazily unmount all mounts on the specified dentry
1634 * During unlink, rmdir, and d_drop it is possible to loose the path
1635 * to an existing mountpoint, and wind up leaking the mount.
1636 * detach_mounts allows lazily unmounting those mounts instead of
1639 * The caller may hold dentry->d_inode->i_mutex.
1641 void __detach_mounts(struct dentry *dentry)
1643 struct mountpoint *mp;
1648 mp = lookup_mountpoint(dentry);
1649 if (IS_ERR_OR_NULL(mp))
1653 while (!hlist_empty(&mp->m_list)) {
1654 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1655 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1656 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1659 else umount_tree(mnt, UMOUNT_CONNECTED);
1663 unlock_mount_hash();
1668 * Is the caller allowed to modify his namespace?
1670 static inline bool may_mount(void)
1672 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1676 * Now umount can handle mount points as well as block devices.
1677 * This is important for filesystems which use unnamed block devices.
1679 * We now support a flag for forced unmount like the other 'big iron'
1680 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1683 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1688 int lookup_flags = 0;
1690 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1696 if (!(flags & UMOUNT_NOFOLLOW))
1697 lookup_flags |= LOOKUP_FOLLOW;
1699 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1702 mnt = real_mount(path.mnt);
1704 if (path.dentry != path.mnt->mnt_root)
1706 if (!check_mnt(mnt))
1708 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1711 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1714 retval = do_umount(mnt, flags);
1716 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1718 mntput_no_expire(mnt);
1723 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1726 * The 2.0 compatible umount. No flags.
1728 SYSCALL_DEFINE1(oldumount, char __user *, name)
1730 return sys_umount(name, 0);
1735 static bool is_mnt_ns_file(struct dentry *dentry)
1737 /* Is this a proxy for a mount namespace? */
1738 return dentry->d_op == &ns_dentry_operations &&
1739 dentry->d_fsdata == &mntns_operations;
1742 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1744 return container_of(ns, struct mnt_namespace, ns);
1747 static bool mnt_ns_loop(struct dentry *dentry)
1749 /* Could bind mounting the mount namespace inode cause a
1750 * mount namespace loop?
1752 struct mnt_namespace *mnt_ns;
1753 if (!is_mnt_ns_file(dentry))
1756 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1757 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1760 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1763 struct mount *res, *p, *q, *r, *parent;
1765 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1766 return ERR_PTR(-EINVAL);
1768 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1769 return ERR_PTR(-EINVAL);
1771 res = q = clone_mnt(mnt, dentry, flag);
1775 q->mnt_mountpoint = mnt->mnt_mountpoint;
1778 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1780 if (!is_subdir(r->mnt_mountpoint, dentry))
1783 for (s = r; s; s = next_mnt(s, r)) {
1784 if (!(flag & CL_COPY_UNBINDABLE) &&
1785 IS_MNT_UNBINDABLE(s)) {
1786 if (s->mnt.mnt_flags & MNT_LOCKED) {
1787 /* Both unbindable and locked. */
1788 q = ERR_PTR(-EPERM);
1791 s = skip_mnt_tree(s);
1795 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1796 is_mnt_ns_file(s->mnt.mnt_root)) {
1797 s = skip_mnt_tree(s);
1800 while (p != s->mnt_parent) {
1806 q = clone_mnt(p, p->mnt.mnt_root, flag);
1810 list_add_tail(&q->mnt_list, &res->mnt_list);
1811 attach_mnt(q, parent, p->mnt_mp);
1812 unlock_mount_hash();
1819 umount_tree(res, UMOUNT_SYNC);
1820 unlock_mount_hash();
1825 /* Caller should check returned pointer for errors */
1827 struct vfsmount *collect_mounts(struct path *path)
1831 if (!check_mnt(real_mount(path->mnt)))
1832 tree = ERR_PTR(-EINVAL);
1834 tree = copy_tree(real_mount(path->mnt), path->dentry,
1835 CL_COPY_ALL | CL_PRIVATE);
1838 return ERR_CAST(tree);
1842 void drop_collected_mounts(struct vfsmount *mnt)
1846 umount_tree(real_mount(mnt), 0);
1847 unlock_mount_hash();
1852 * clone_private_mount - create a private clone of a path
1854 * This creates a new vfsmount, which will be the clone of @path. The new will
1855 * not be attached anywhere in the namespace and will be private (i.e. changes
1856 * to the originating mount won't be propagated into this).
1858 * Release with mntput().
1860 struct vfsmount *clone_private_mount(struct path *path)
1862 struct mount *old_mnt = real_mount(path->mnt);
1863 struct mount *new_mnt;
1865 if (IS_MNT_UNBINDABLE(old_mnt))
1866 return ERR_PTR(-EINVAL);
1868 down_read(&namespace_sem);
1869 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1870 up_read(&namespace_sem);
1871 if (IS_ERR(new_mnt))
1872 return ERR_CAST(new_mnt);
1874 return &new_mnt->mnt;
1876 EXPORT_SYMBOL_GPL(clone_private_mount);
1878 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1879 struct vfsmount *root)
1882 int res = f(root, arg);
1885 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1886 res = f(&mnt->mnt, arg);
1893 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1897 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1898 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1899 mnt_release_group_id(p);
1903 static int invent_group_ids(struct mount *mnt, bool recurse)
1907 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1908 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1909 int err = mnt_alloc_group_id(p);
1911 cleanup_group_ids(mnt, p);
1920 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1922 unsigned int max = READ_ONCE(sysctl_mount_max);
1923 unsigned int mounts = 0, old, pending, sum;
1926 for (p = mnt; p; p = next_mnt(p, mnt))
1930 pending = ns->pending_mounts;
1931 sum = old + pending;
1935 (mounts > (max - sum)))
1938 ns->pending_mounts = pending + mounts;
1943 * @source_mnt : mount tree to be attached
1944 * @nd : place the mount tree @source_mnt is attached
1945 * @parent_nd : if non-null, detach the source_mnt from its parent and
1946 * store the parent mount and mountpoint dentry.
1947 * (done when source_mnt is moved)
1949 * NOTE: in the table below explains the semantics when a source mount
1950 * of a given type is attached to a destination mount of a given type.
1951 * ---------------------------------------------------------------------------
1952 * | BIND MOUNT OPERATION |
1953 * |**************************************************************************
1954 * | source-->| shared | private | slave | unbindable |
1958 * |**************************************************************************
1959 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1961 * |non-shared| shared (+) | private | slave (*) | invalid |
1962 * ***************************************************************************
1963 * A bind operation clones the source mount and mounts the clone on the
1964 * destination mount.
1966 * (++) the cloned mount is propagated to all the mounts in the propagation
1967 * tree of the destination mount and the cloned mount is added to
1968 * the peer group of the source mount.
1969 * (+) the cloned mount is created under the destination mount and is marked
1970 * as shared. The cloned mount is added to the peer group of the source
1972 * (+++) the mount is propagated to all the mounts in the propagation tree
1973 * of the destination mount and the cloned mount is made slave
1974 * of the same master as that of the source mount. The cloned mount
1975 * is marked as 'shared and slave'.
1976 * (*) the cloned mount is made a slave of the same master as that of the
1979 * ---------------------------------------------------------------------------
1980 * | MOVE MOUNT OPERATION |
1981 * |**************************************************************************
1982 * | source-->| shared | private | slave | unbindable |
1986 * |**************************************************************************
1987 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1989 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1990 * ***************************************************************************
1992 * (+) the mount is moved to the destination. And is then propagated to
1993 * all the mounts in the propagation tree of the destination mount.
1994 * (+*) the mount is moved to the destination.
1995 * (+++) the mount is moved to the destination and is then propagated to
1996 * all the mounts belonging to the destination mount's propagation tree.
1997 * the mount is marked as 'shared and slave'.
1998 * (*) the mount continues to be a slave at the new location.
2000 * if the source mount is a tree, the operations explained above is
2001 * applied to each mount in the tree.
2002 * Must be called without spinlocks held, since this function can sleep
2005 static int attach_recursive_mnt(struct mount *source_mnt,
2006 struct mount *dest_mnt,
2007 struct mountpoint *dest_mp,
2008 struct path *parent_path)
2010 HLIST_HEAD(tree_list);
2011 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2012 struct mountpoint *smp;
2013 struct mount *child, *p;
2014 struct hlist_node *n;
2017 /* Preallocate a mountpoint in case the new mounts need
2018 * to be tucked under other mounts.
2020 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2022 return PTR_ERR(smp);
2024 /* Is there space to add these mounts to the mount namespace? */
2026 err = count_mounts(ns, source_mnt);
2031 if (IS_MNT_SHARED(dest_mnt)) {
2032 err = invent_group_ids(source_mnt, true);
2035 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2038 goto out_cleanup_ids;
2039 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2045 detach_mnt(source_mnt, parent_path);
2046 attach_mnt(source_mnt, dest_mnt, dest_mp);
2047 touch_mnt_namespace(source_mnt->mnt_ns);
2049 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2050 commit_tree(source_mnt);
2053 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2055 hlist_del_init(&child->mnt_hash);
2056 q = __lookup_mnt(&child->mnt_parent->mnt,
2057 child->mnt_mountpoint);
2059 mnt_change_mountpoint(child, smp, q);
2062 put_mountpoint(smp);
2063 unlock_mount_hash();
2068 while (!hlist_empty(&tree_list)) {
2069 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2070 child->mnt_parent->mnt_ns->pending_mounts = 0;
2071 umount_tree(child, UMOUNT_SYNC);
2073 unlock_mount_hash();
2074 cleanup_group_ids(source_mnt, NULL);
2076 ns->pending_mounts = 0;
2078 read_seqlock_excl(&mount_lock);
2079 put_mountpoint(smp);
2080 read_sequnlock_excl(&mount_lock);
2085 static struct mountpoint *lock_mount(struct path *path)
2087 struct vfsmount *mnt;
2088 struct dentry *dentry = path->dentry;
2090 mutex_lock(&dentry->d_inode->i_mutex);
2091 if (unlikely(cant_mount(dentry))) {
2092 mutex_unlock(&dentry->d_inode->i_mutex);
2093 return ERR_PTR(-ENOENT);
2096 mnt = lookup_mnt(path);
2098 struct mountpoint *mp = get_mountpoint(dentry);
2101 mutex_unlock(&dentry->d_inode->i_mutex);
2107 mutex_unlock(&path->dentry->d_inode->i_mutex);
2110 dentry = path->dentry = dget(mnt->mnt_root);
2114 static void unlock_mount(struct mountpoint *where)
2116 struct dentry *dentry = where->m_dentry;
2118 read_seqlock_excl(&mount_lock);
2119 put_mountpoint(where);
2120 read_sequnlock_excl(&mount_lock);
2123 mutex_unlock(&dentry->d_inode->i_mutex);
2126 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2128 if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
2131 if (d_is_dir(mp->m_dentry) !=
2132 d_is_dir(mnt->mnt.mnt_root))
2135 return attach_recursive_mnt(mnt, p, mp, NULL);
2139 * Sanity check the flags to change_mnt_propagation.
2142 static int flags_to_propagation_type(int flags)
2144 int type = flags & ~(MS_REC | MS_SILENT);
2146 /* Fail if any non-propagation flags are set */
2147 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2149 /* Only one propagation flag should be set */
2150 if (!is_power_of_2(type))
2156 * recursively change the type of the mountpoint.
2158 static int do_change_type(struct path *path, int flag)
2161 struct mount *mnt = real_mount(path->mnt);
2162 int recurse = flag & MS_REC;
2166 if (path->dentry != path->mnt->mnt_root)
2169 type = flags_to_propagation_type(flag);
2174 if (type == MS_SHARED) {
2175 err = invent_group_ids(mnt, recurse);
2181 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2182 change_mnt_propagation(m, type);
2183 unlock_mount_hash();
2190 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2192 struct mount *child;
2193 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2194 if (!is_subdir(child->mnt_mountpoint, dentry))
2197 if (child->mnt.mnt_flags & MNT_LOCKED)
2204 * do loopback mount.
2206 static int do_loopback(struct path *path, const char *old_name,
2209 struct path old_path;
2210 struct mount *mnt = NULL, *old, *parent;
2211 struct mountpoint *mp;
2213 if (!old_name || !*old_name)
2215 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2220 if (mnt_ns_loop(old_path.dentry))
2223 mp = lock_mount(path);
2228 old = real_mount(old_path.mnt);
2229 parent = real_mount(path->mnt);
2232 if (IS_MNT_UNBINDABLE(old))
2235 if (!check_mnt(parent))
2238 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2241 if (!recurse && has_locked_children(old, old_path.dentry))
2245 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2247 mnt = clone_mnt(old, old_path.dentry, 0);
2254 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2256 err = graft_tree(mnt, parent, mp);
2259 umount_tree(mnt, UMOUNT_SYNC);
2260 unlock_mount_hash();
2265 path_put(&old_path);
2269 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2272 int readonly_request = 0;
2274 if (ms_flags & MS_RDONLY)
2275 readonly_request = 1;
2276 if (readonly_request == __mnt_is_readonly(mnt))
2279 if (readonly_request)
2280 error = mnt_make_readonly(real_mount(mnt));
2282 __mnt_unmake_readonly(real_mount(mnt));
2287 * change filesystem flags. dir should be a physical root of filesystem.
2288 * If you've mounted a non-root directory somewhere and want to do remount
2289 * on it - tough luck.
2291 static int do_remount(struct path *path, int flags, int mnt_flags,
2295 struct super_block *sb = path->mnt->mnt_sb;
2296 struct mount *mnt = real_mount(path->mnt);
2298 if (!check_mnt(mnt))
2301 if (path->dentry != path->mnt->mnt_root)
2304 /* Don't allow changing of locked mnt flags.
2306 * No locks need to be held here while testing the various
2307 * MNT_LOCK flags because those flags can never be cleared
2308 * once they are set.
2310 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2311 !(mnt_flags & MNT_READONLY)) {
2314 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2315 !(mnt_flags & MNT_NODEV)) {
2316 /* Was the nodev implicitly added in mount? */
2317 if ((mnt->mnt_ns->user_ns != &init_user_ns) &&
2318 !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2319 mnt_flags |= MNT_NODEV;
2324 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2325 !(mnt_flags & MNT_NOSUID)) {
2328 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2329 !(mnt_flags & MNT_NOEXEC)) {
2332 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2333 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2337 err = security_sb_remount(sb, data);
2341 down_write(&sb->s_umount);
2342 if (flags & MS_BIND)
2343 err = change_mount_flags(path->mnt, flags);
2344 else if (!capable(CAP_SYS_ADMIN))
2347 err = do_remount_sb2(path->mnt, sb, flags, data, 0);
2350 propagate_remount(mnt);
2351 unlock_mount_hash();
2356 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2357 mnt->mnt.mnt_flags = mnt_flags;
2358 touch_mnt_namespace(mnt->mnt_ns);
2359 unlock_mount_hash();
2361 up_write(&sb->s_umount);
2365 static inline int tree_contains_unbindable(struct mount *mnt)
2368 for (p = mnt; p; p = next_mnt(p, mnt)) {
2369 if (IS_MNT_UNBINDABLE(p))
2375 static int do_move_mount(struct path *path, const char *old_name)
2377 struct path old_path, parent_path;
2380 struct mountpoint *mp;
2382 if (!old_name || !*old_name)
2384 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2388 mp = lock_mount(path);
2393 old = real_mount(old_path.mnt);
2394 p = real_mount(path->mnt);
2397 if (!check_mnt(p) || !check_mnt(old))
2400 if (old->mnt.mnt_flags & MNT_LOCKED)
2404 if (old_path.dentry != old_path.mnt->mnt_root)
2407 if (!mnt_has_parent(old))
2410 if (d_is_dir(path->dentry) !=
2411 d_is_dir(old_path.dentry))
2414 * Don't move a mount residing in a shared parent.
2416 if (IS_MNT_SHARED(old->mnt_parent))
2419 * Don't move a mount tree containing unbindable mounts to a destination
2420 * mount which is shared.
2422 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2425 for (; mnt_has_parent(p); p = p->mnt_parent)
2429 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2433 /* if the mount is moved, it should no longer be expire
2435 list_del_init(&old->mnt_expire);
2440 path_put(&parent_path);
2441 path_put(&old_path);
2445 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2448 const char *subtype = strchr(fstype, '.');
2457 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2459 if (!mnt->mnt_sb->s_subtype)
2465 return ERR_PTR(err);
2469 * add a mount into a namespace's mount tree
2471 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2473 struct mountpoint *mp;
2474 struct mount *parent;
2477 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2479 mp = lock_mount(path);
2483 parent = real_mount(path->mnt);
2485 if (unlikely(!check_mnt(parent))) {
2486 /* that's acceptable only for automounts done in private ns */
2487 if (!(mnt_flags & MNT_SHRINKABLE))
2489 /* ... and for those we'd better have mountpoint still alive */
2490 if (!parent->mnt_ns)
2494 /* Refuse the same filesystem on the same mount point */
2496 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2497 path->mnt->mnt_root == path->dentry)
2501 if (d_is_symlink(newmnt->mnt.mnt_root))
2504 newmnt->mnt.mnt_flags = mnt_flags;
2505 err = graft_tree(newmnt, parent, mp);
2512 static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags);
2515 * create a new mount for userspace and request it to be added into the
2518 static int do_new_mount(struct path *path, const char *fstype, int flags,
2519 int mnt_flags, const char *name, void *data)
2521 struct file_system_type *type;
2522 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2523 struct vfsmount *mnt;
2529 type = get_fs_type(fstype);
2533 if (user_ns != &init_user_ns) {
2534 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2535 put_filesystem(type);
2538 /* Only in special cases allow devices from mounts
2539 * created outside the initial user namespace.
2541 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2543 mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV;
2545 if (type->fs_flags & FS_USERNS_VISIBLE) {
2546 if (!fs_fully_visible(type, &mnt_flags)) {
2547 put_filesystem(type);
2553 mnt = vfs_kern_mount(type, flags, name, data);
2554 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2555 !mnt->mnt_sb->s_subtype)
2556 mnt = fs_set_subtype(mnt, fstype);
2558 put_filesystem(type);
2560 return PTR_ERR(mnt);
2562 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2568 int finish_automount(struct vfsmount *m, struct path *path)
2570 struct mount *mnt = real_mount(m);
2572 /* The new mount record should have at least 2 refs to prevent it being
2573 * expired before we get a chance to add it
2575 BUG_ON(mnt_get_count(mnt) < 2);
2577 if (m->mnt_sb == path->mnt->mnt_sb &&
2578 m->mnt_root == path->dentry) {
2583 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2587 /* remove m from any expiration list it may be on */
2588 if (!list_empty(&mnt->mnt_expire)) {
2590 list_del_init(&mnt->mnt_expire);
2599 * mnt_set_expiry - Put a mount on an expiration list
2600 * @mnt: The mount to list.
2601 * @expiry_list: The list to add the mount to.
2603 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2607 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2611 EXPORT_SYMBOL(mnt_set_expiry);
2614 * process a list of expirable mountpoints with the intent of discarding any
2615 * mountpoints that aren't in use and haven't been touched since last we came
2618 void mark_mounts_for_expiry(struct list_head *mounts)
2620 struct mount *mnt, *next;
2621 LIST_HEAD(graveyard);
2623 if (list_empty(mounts))
2629 /* extract from the expiration list every vfsmount that matches the
2630 * following criteria:
2631 * - only referenced by its parent vfsmount
2632 * - still marked for expiry (marked on the last call here; marks are
2633 * cleared by mntput())
2635 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2636 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2637 propagate_mount_busy(mnt, 1))
2639 list_move(&mnt->mnt_expire, &graveyard);
2641 while (!list_empty(&graveyard)) {
2642 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2643 touch_mnt_namespace(mnt->mnt_ns);
2644 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2646 unlock_mount_hash();
2650 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2653 * Ripoff of 'select_parent()'
2655 * search the list of submounts for a given mountpoint, and move any
2656 * shrinkable submounts to the 'graveyard' list.
2658 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2660 struct mount *this_parent = parent;
2661 struct list_head *next;
2665 next = this_parent->mnt_mounts.next;
2667 while (next != &this_parent->mnt_mounts) {
2668 struct list_head *tmp = next;
2669 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2672 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2675 * Descend a level if the d_mounts list is non-empty.
2677 if (!list_empty(&mnt->mnt_mounts)) {
2682 if (!propagate_mount_busy(mnt, 1)) {
2683 list_move_tail(&mnt->mnt_expire, graveyard);
2688 * All done at this level ... ascend and resume the search
2690 if (this_parent != parent) {
2691 next = this_parent->mnt_child.next;
2692 this_parent = this_parent->mnt_parent;
2699 * process a list of expirable mountpoints with the intent of discarding any
2700 * submounts of a specific parent mountpoint
2702 * mount_lock must be held for write
2704 static void shrink_submounts(struct mount *mnt)
2706 LIST_HEAD(graveyard);
2709 /* extract submounts of 'mountpoint' from the expiration list */
2710 while (select_submounts(mnt, &graveyard)) {
2711 while (!list_empty(&graveyard)) {
2712 m = list_first_entry(&graveyard, struct mount,
2714 touch_mnt_namespace(m->mnt_ns);
2715 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2721 * Some copy_from_user() implementations do not return the exact number of
2722 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2723 * Note that this function differs from copy_from_user() in that it will oops
2724 * on bad values of `to', rather than returning a short copy.
2726 static long exact_copy_from_user(void *to, const void __user * from,
2730 const char __user *f = from;
2733 if (!access_ok(VERIFY_READ, from, n))
2737 if (__get_user(c, f)) {
2748 int copy_mount_options(const void __user * data, unsigned long *where)
2758 if (!(page = __get_free_page(GFP_KERNEL)))
2761 /* We only care that *some* data at the address the user
2762 * gave us is valid. Just in case, we'll zero
2763 * the remainder of the page.
2765 /* copy_from_user cannot cross TASK_SIZE ! */
2766 size = TASK_SIZE - (unsigned long)data;
2767 if (size > PAGE_SIZE)
2770 i = size - exact_copy_from_user((void *)page, data, size);
2776 memset((char *)page + i, 0, PAGE_SIZE - i);
2781 char *copy_mount_string(const void __user *data)
2783 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2787 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2788 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2790 * data is a (void *) that can point to any structure up to
2791 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2792 * information (or be NULL).
2794 * Pre-0.97 versions of mount() didn't have a flags word.
2795 * When the flags word was introduced its top half was required
2796 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2797 * Therefore, if this magic number is present, it carries no information
2798 * and must be discarded.
2800 long do_mount(const char *dev_name, const char __user *dir_name,
2801 const char *type_page, unsigned long flags, void *data_page)
2808 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2809 flags &= ~MS_MGC_MSK;
2811 /* Basic sanity checks */
2813 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2815 /* ... and get the mountpoint */
2816 retval = user_path(dir_name, &path);
2820 retval = security_sb_mount(dev_name, &path,
2821 type_page, flags, data_page);
2822 if (!retval && !may_mount())
2827 /* Default to relatime unless overriden */
2828 if (!(flags & MS_NOATIME))
2829 mnt_flags |= MNT_RELATIME;
2831 /* Separate the per-mountpoint flags */
2832 if (flags & MS_NOSUID)
2833 mnt_flags |= MNT_NOSUID;
2834 if (flags & MS_NODEV)
2835 mnt_flags |= MNT_NODEV;
2836 if (flags & MS_NOEXEC)
2837 mnt_flags |= MNT_NOEXEC;
2838 if (flags & MS_NOATIME)
2839 mnt_flags |= MNT_NOATIME;
2840 if (flags & MS_NODIRATIME)
2841 mnt_flags |= MNT_NODIRATIME;
2842 if (flags & MS_STRICTATIME)
2843 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2844 if (flags & MS_RDONLY)
2845 mnt_flags |= MNT_READONLY;
2847 /* The default atime for remount is preservation */
2848 if ((flags & MS_REMOUNT) &&
2849 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2850 MS_STRICTATIME)) == 0)) {
2851 mnt_flags &= ~MNT_ATIME_MASK;
2852 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2855 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2856 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2859 if (flags & MS_REMOUNT)
2860 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2862 else if (flags & MS_BIND)
2863 retval = do_loopback(&path, dev_name, flags & MS_REC);
2864 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2865 retval = do_change_type(&path, flags);
2866 else if (flags & MS_MOVE)
2867 retval = do_move_mount(&path, dev_name);
2869 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2870 dev_name, data_page);
2876 static void free_mnt_ns(struct mnt_namespace *ns)
2878 ns_free_inum(&ns->ns);
2879 put_user_ns(ns->user_ns);
2884 * Assign a sequence number so we can detect when we attempt to bind
2885 * mount a reference to an older mount namespace into the current
2886 * mount namespace, preventing reference counting loops. A 64bit
2887 * number incrementing at 10Ghz will take 12,427 years to wrap which
2888 * is effectively never, so we can ignore the possibility.
2890 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2892 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2894 struct mnt_namespace *new_ns;
2897 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2899 return ERR_PTR(-ENOMEM);
2900 ret = ns_alloc_inum(&new_ns->ns);
2903 return ERR_PTR(ret);
2905 new_ns->ns.ops = &mntns_operations;
2906 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2907 atomic_set(&new_ns->count, 1);
2908 new_ns->root = NULL;
2909 INIT_LIST_HEAD(&new_ns->list);
2910 init_waitqueue_head(&new_ns->poll);
2912 new_ns->user_ns = get_user_ns(user_ns);
2914 new_ns->pending_mounts = 0;
2918 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2919 struct user_namespace *user_ns, struct fs_struct *new_fs)
2921 struct mnt_namespace *new_ns;
2922 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2923 struct mount *p, *q;
2930 if (likely(!(flags & CLONE_NEWNS))) {
2937 new_ns = alloc_mnt_ns(user_ns);
2942 /* First pass: copy the tree topology */
2943 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2944 if (user_ns != ns->user_ns)
2945 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2946 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2949 free_mnt_ns(new_ns);
2950 return ERR_CAST(new);
2953 list_add_tail(&new_ns->list, &new->mnt_list);
2956 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2957 * as belonging to new namespace. We have already acquired a private
2958 * fs_struct, so tsk->fs->lock is not needed.
2966 if (&p->mnt == new_fs->root.mnt) {
2967 new_fs->root.mnt = mntget(&q->mnt);
2970 if (&p->mnt == new_fs->pwd.mnt) {
2971 new_fs->pwd.mnt = mntget(&q->mnt);
2975 p = next_mnt(p, old);
2976 q = next_mnt(q, new);
2979 while (p->mnt.mnt_root != q->mnt.mnt_root)
2980 p = next_mnt(p, old);
2993 * create_mnt_ns - creates a private namespace and adds a root filesystem
2994 * @mnt: pointer to the new root filesystem mountpoint
2996 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2998 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2999 if (!IS_ERR(new_ns)) {
3000 struct mount *mnt = real_mount(m);
3001 mnt->mnt_ns = new_ns;
3004 list_add(&mnt->mnt_list, &new_ns->list);
3011 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3013 struct mnt_namespace *ns;
3014 struct super_block *s;
3018 ns = create_mnt_ns(mnt);
3020 return ERR_CAST(ns);
3022 err = vfs_path_lookup(mnt->mnt_root, mnt,
3023 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3028 return ERR_PTR(err);
3030 /* trade a vfsmount reference for active sb one */
3031 s = path.mnt->mnt_sb;
3032 atomic_inc(&s->s_active);
3034 /* lock the sucker */
3035 down_write(&s->s_umount);
3036 /* ... and return the root of (sub)tree on it */
3039 EXPORT_SYMBOL(mount_subtree);
3041 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3042 char __user *, type, unsigned long, flags, void __user *, data)
3047 unsigned long data_page;
3049 kernel_type = copy_mount_string(type);
3050 ret = PTR_ERR(kernel_type);
3051 if (IS_ERR(kernel_type))
3054 kernel_dev = copy_mount_string(dev_name);
3055 ret = PTR_ERR(kernel_dev);
3056 if (IS_ERR(kernel_dev))
3059 ret = copy_mount_options(data, &data_page);
3063 ret = do_mount(kernel_dev, dir_name, kernel_type, flags,
3064 (void *) data_page);
3066 free_page(data_page);
3076 * Return true if path is reachable from root
3078 * namespace_sem or mount_lock is held
3080 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3081 const struct path *root)
3083 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3084 dentry = mnt->mnt_mountpoint;
3085 mnt = mnt->mnt_parent;
3087 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3090 int path_is_under(struct path *path1, struct path *path2)
3093 read_seqlock_excl(&mount_lock);
3094 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3095 read_sequnlock_excl(&mount_lock);
3098 EXPORT_SYMBOL(path_is_under);
3101 * pivot_root Semantics:
3102 * Moves the root file system of the current process to the directory put_old,
3103 * makes new_root as the new root file system of the current process, and sets
3104 * root/cwd of all processes which had them on the current root to new_root.
3107 * The new_root and put_old must be directories, and must not be on the
3108 * same file system as the current process root. The put_old must be
3109 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3110 * pointed to by put_old must yield the same directory as new_root. No other
3111 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3113 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3114 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3115 * in this situation.
3118 * - we don't move root/cwd if they are not at the root (reason: if something
3119 * cared enough to change them, it's probably wrong to force them elsewhere)
3120 * - it's okay to pick a root that isn't the root of a file system, e.g.
3121 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3122 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3125 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3126 const char __user *, put_old)
3128 struct path new, old, parent_path, root_parent, root;
3129 struct mount *new_mnt, *root_mnt, *old_mnt;
3130 struct mountpoint *old_mp, *root_mp;
3136 error = user_path_dir(new_root, &new);
3140 error = user_path_dir(put_old, &old);
3144 error = security_sb_pivotroot(&old, &new);
3148 get_fs_root(current->fs, &root);
3149 old_mp = lock_mount(&old);
3150 error = PTR_ERR(old_mp);
3155 new_mnt = real_mount(new.mnt);
3156 root_mnt = real_mount(root.mnt);
3157 old_mnt = real_mount(old.mnt);
3158 if (IS_MNT_SHARED(old_mnt) ||
3159 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3160 IS_MNT_SHARED(root_mnt->mnt_parent))
3162 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3164 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3167 if (d_unlinked(new.dentry))
3170 if (new_mnt == root_mnt || old_mnt == root_mnt)
3171 goto out4; /* loop, on the same file system */
3173 if (root.mnt->mnt_root != root.dentry)
3174 goto out4; /* not a mountpoint */
3175 if (!mnt_has_parent(root_mnt))
3176 goto out4; /* not attached */
3177 root_mp = root_mnt->mnt_mp;
3178 if (new.mnt->mnt_root != new.dentry)
3179 goto out4; /* not a mountpoint */
3180 if (!mnt_has_parent(new_mnt))
3181 goto out4; /* not attached */
3182 /* make sure we can reach put_old from new_root */
3183 if (!is_path_reachable(old_mnt, old.dentry, &new))
3185 /* make certain new is below the root */
3186 if (!is_path_reachable(new_mnt, new.dentry, &root))
3188 root_mp->m_count++; /* pin it so it won't go away */
3190 detach_mnt(new_mnt, &parent_path);
3191 detach_mnt(root_mnt, &root_parent);
3192 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3193 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3194 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3196 /* mount old root on put_old */
3197 attach_mnt(root_mnt, old_mnt, old_mp);
3198 /* mount new_root on / */
3199 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3200 touch_mnt_namespace(current->nsproxy->mnt_ns);
3201 /* A moved mount should not expire automatically */
3202 list_del_init(&new_mnt->mnt_expire);
3203 put_mountpoint(root_mp);
3204 unlock_mount_hash();
3205 chroot_fs_refs(&root, &new);
3208 unlock_mount(old_mp);
3210 path_put(&root_parent);
3211 path_put(&parent_path);
3223 static void __init init_mount_tree(void)
3225 struct vfsmount *mnt;
3226 struct mnt_namespace *ns;
3228 struct file_system_type *type;
3230 type = get_fs_type("rootfs");
3232 panic("Can't find rootfs type");
3233 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3234 put_filesystem(type);
3236 panic("Can't create rootfs");
3238 ns = create_mnt_ns(mnt);
3240 panic("Can't allocate initial namespace");
3242 init_task.nsproxy->mnt_ns = ns;
3246 root.dentry = mnt->mnt_root;
3247 mnt->mnt_flags |= MNT_LOCKED;
3249 set_fs_pwd(current->fs, &root);
3250 set_fs_root(current->fs, &root);
3253 void __init mnt_init(void)
3258 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3259 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3261 mount_hashtable = alloc_large_system_hash("Mount-cache",
3262 sizeof(struct hlist_head),
3265 &m_hash_shift, &m_hash_mask, 0, 0);
3266 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3267 sizeof(struct hlist_head),
3270 &mp_hash_shift, &mp_hash_mask, 0, 0);
3272 if (!mount_hashtable || !mountpoint_hashtable)
3273 panic("Failed to allocate mount hash table\n");
3275 for (u = 0; u <= m_hash_mask; u++)
3276 INIT_HLIST_HEAD(&mount_hashtable[u]);
3277 for (u = 0; u <= mp_hash_mask; u++)
3278 INIT_HLIST_HEAD(&mountpoint_hashtable[u]);
3284 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3286 fs_kobj = kobject_create_and_add("fs", NULL);
3288 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3293 void put_mnt_ns(struct mnt_namespace *ns)
3295 if (!atomic_dec_and_test(&ns->count))
3297 drop_collected_mounts(&ns->root->mnt);
3301 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3303 struct vfsmount *mnt;
3304 mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
3307 * it is a longterm mount, don't release mnt until
3308 * we unmount before file sys is unregistered
3310 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3314 EXPORT_SYMBOL_GPL(kern_mount_data);
3316 void kern_unmount(struct vfsmount *mnt)
3318 /* release long term mount so mount point can be released */
3319 if (!IS_ERR_OR_NULL(mnt)) {
3320 real_mount(mnt)->mnt_ns = NULL;
3321 synchronize_rcu(); /* yecchhh... */
3325 EXPORT_SYMBOL(kern_unmount);
3327 bool our_mnt(struct vfsmount *mnt)
3329 return check_mnt(real_mount(mnt));
3332 bool current_chrooted(void)
3334 /* Does the current process have a non-standard root */
3335 struct path ns_root;
3336 struct path fs_root;
3339 /* Find the namespace root */
3340 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3341 ns_root.dentry = ns_root.mnt->mnt_root;
3343 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3346 get_fs_root(current->fs, &fs_root);
3348 chrooted = !path_equal(&fs_root, &ns_root);
3356 static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags)
3358 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3359 int new_flags = *new_mnt_flags;
3361 bool visible = false;
3366 down_read(&namespace_sem);
3367 list_for_each_entry(mnt, &ns->list, mnt_list) {
3368 struct mount *child;
3371 if (mnt->mnt.mnt_sb->s_type != type)
3374 /* This mount is not fully visible if it's root directory
3375 * is not the root directory of the filesystem.
3377 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3380 /* Read the mount flags and filter out flags that
3381 * may safely be ignored.
3383 mnt_flags = mnt->mnt.mnt_flags;
3384 if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC)
3385 mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC);
3387 /* Don't miss readonly hidden in the superblock flags */
3388 if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY)
3389 mnt_flags |= MNT_LOCK_READONLY;
3391 /* Verify the mount flags are equal to or more permissive
3392 * than the proposed new mount.
3394 if ((mnt_flags & MNT_LOCK_READONLY) &&
3395 !(new_flags & MNT_READONLY))
3397 if ((mnt_flags & MNT_LOCK_NODEV) &&
3398 !(new_flags & MNT_NODEV))
3400 if ((mnt_flags & MNT_LOCK_NOSUID) &&
3401 !(new_flags & MNT_NOSUID))
3403 if ((mnt_flags & MNT_LOCK_NOEXEC) &&
3404 !(new_flags & MNT_NOEXEC))
3406 if ((mnt_flags & MNT_LOCK_ATIME) &&
3407 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3410 /* This mount is not fully visible if there are any
3411 * locked child mounts that cover anything except for
3412 * empty directories.
3414 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3415 struct inode *inode = child->mnt_mountpoint->d_inode;
3416 /* Only worry about locked mounts */
3417 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3419 /* Is the directory permanetly empty? */
3420 if (!is_empty_dir_inode(inode))
3423 /* Preserve the locked attributes */
3424 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3434 up_read(&namespace_sem);
3438 static struct ns_common *mntns_get(struct task_struct *task)
3440 struct ns_common *ns = NULL;
3441 struct nsproxy *nsproxy;
3444 nsproxy = task->nsproxy;
3446 ns = &nsproxy->mnt_ns->ns;
3447 get_mnt_ns(to_mnt_ns(ns));
3454 static void mntns_put(struct ns_common *ns)
3456 put_mnt_ns(to_mnt_ns(ns));
3459 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3461 struct fs_struct *fs = current->fs;
3462 struct mnt_namespace *mnt_ns = to_mnt_ns(ns);
3465 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3466 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3467 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3474 put_mnt_ns(nsproxy->mnt_ns);
3475 nsproxy->mnt_ns = mnt_ns;
3478 root.mnt = &mnt_ns->root->mnt;
3479 root.dentry = mnt_ns->root->mnt.mnt_root;
3481 while(d_mountpoint(root.dentry) && follow_down_one(&root))
3484 /* Update the pwd and root */
3485 set_fs_pwd(fs, &root);
3486 set_fs_root(fs, &root);
3492 const struct proc_ns_operations mntns_operations = {
3494 .type = CLONE_NEWNS,
3497 .install = mntns_install,