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 (sb_rdonly(mnt->mnt_sb))
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_path - 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 * Called by the vfs for cases when we have an open file at hand, but will do an
441 * inode operation on it (important distinction for files opened on overlayfs,
442 * since the file operations will come from the real underlying file, while
443 * inode operations come from the overlay).
445 int mnt_want_write_file_path(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);
456 static inline int may_write_real(struct file *file)
458 struct dentry *dentry = file->f_path.dentry;
459 struct dentry *upperdentry;
462 if (file->f_mode & FMODE_WRITER)
466 if (likely(!(dentry->d_flags & DCACHE_OP_REAL)))
469 /* File refers to upper, writable layer? */
470 upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
471 if (upperdentry && file_inode(file) == d_inode(upperdentry))
474 /* Lower layer: can't write to real file, sorry... */
479 * mnt_want_write_file - get write access to a file's mount
480 * @file: the file who's mount on which to take a write
482 * This is like mnt_want_write, but it takes a file and can
483 * do some optimisations if the file is open for write already
485 * Mostly called by filesystems from their ioctl operation before performing
486 * modification. On overlayfs this needs to check if the file is on a read-only
487 * lower layer and deny access in that case.
489 int mnt_want_write_file(struct file *file)
493 ret = may_write_real(file);
495 sb_start_write(file_inode(file)->i_sb);
496 ret = __mnt_want_write_file(file);
498 sb_end_write(file_inode(file)->i_sb);
502 EXPORT_SYMBOL_GPL(mnt_want_write_file);
505 * __mnt_drop_write - give up write access to a mount
506 * @mnt: the mount on which to give up write access
508 * Tells the low-level filesystem that we are done
509 * performing writes to it. Must be matched with
510 * __mnt_want_write() call above.
512 void __mnt_drop_write(struct vfsmount *mnt)
515 mnt_dec_writers(real_mount(mnt));
520 * mnt_drop_write - give up write access to a mount
521 * @mnt: the mount on which to give up write access
523 * Tells the low-level filesystem that we are done performing writes to it and
524 * also allows filesystem to be frozen again. Must be matched with
525 * mnt_want_write() call above.
527 void mnt_drop_write(struct vfsmount *mnt)
529 __mnt_drop_write(mnt);
530 sb_end_write(mnt->mnt_sb);
532 EXPORT_SYMBOL_GPL(mnt_drop_write);
534 void __mnt_drop_write_file(struct file *file)
536 __mnt_drop_write(file->f_path.mnt);
539 void mnt_drop_write_file_path(struct file *file)
541 mnt_drop_write(file->f_path.mnt);
544 void mnt_drop_write_file(struct file *file)
546 __mnt_drop_write(file->f_path.mnt);
547 sb_end_write(file_inode(file)->i_sb);
549 EXPORT_SYMBOL(mnt_drop_write_file);
551 static int mnt_make_readonly(struct mount *mnt)
556 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
558 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
559 * should be visible before we do.
564 * With writers on hold, if this value is zero, then there are
565 * definitely no active writers (although held writers may subsequently
566 * increment the count, they'll have to wait, and decrement it after
567 * seeing MNT_READONLY).
569 * It is OK to have counter incremented on one CPU and decremented on
570 * another: the sum will add up correctly. The danger would be when we
571 * sum up each counter, if we read a counter before it is incremented,
572 * but then read another CPU's count which it has been subsequently
573 * decremented from -- we would see more decrements than we should.
574 * MNT_WRITE_HOLD protects against this scenario, because
575 * mnt_want_write first increments count, then smp_mb, then spins on
576 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
577 * we're counting up here.
579 if (mnt_get_writers(mnt) > 0)
582 mnt->mnt.mnt_flags |= MNT_READONLY;
584 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
585 * that become unheld will see MNT_READONLY.
588 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
593 static void __mnt_unmake_readonly(struct mount *mnt)
596 mnt->mnt.mnt_flags &= ~MNT_READONLY;
600 int sb_prepare_remount_readonly(struct super_block *sb)
605 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
606 if (atomic_long_read(&sb->s_remove_count))
610 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
611 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
612 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
614 if (mnt_get_writers(mnt) > 0) {
620 if (!err && atomic_long_read(&sb->s_remove_count))
624 sb->s_readonly_remount = 1;
627 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
628 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
629 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
636 static void free_vfsmnt(struct mount *mnt)
638 kfree_const(mnt->mnt_devname);
640 free_percpu(mnt->mnt_pcp);
642 kmem_cache_free(mnt_cache, mnt);
645 static void delayed_free_vfsmnt(struct rcu_head *head)
647 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
650 /* call under rcu_read_lock */
651 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
654 if (read_seqretry(&mount_lock, seq))
658 mnt = real_mount(bastard);
659 mnt_add_count(mnt, 1);
660 if (likely(!read_seqretry(&mount_lock, seq)))
662 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
663 mnt_add_count(mnt, -1);
669 /* call under rcu_read_lock */
670 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
672 int res = __legitimize_mnt(bastard, seq);
675 if (unlikely(res < 0)) {
684 * find the first mount at @dentry on vfsmount @mnt.
685 * call under rcu_read_lock()
687 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
689 struct hlist_head *head = m_hash(mnt, dentry);
692 hlist_for_each_entry_rcu(p, head, mnt_hash)
693 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
699 * lookup_mnt - Return the first child mount mounted at path
701 * "First" means first mounted chronologically. If you create the
704 * mount /dev/sda1 /mnt
705 * mount /dev/sda2 /mnt
706 * mount /dev/sda3 /mnt
708 * Then lookup_mnt() on the base /mnt dentry in the root mount will
709 * return successively the root dentry and vfsmount of /dev/sda1, then
710 * /dev/sda2, then /dev/sda3, then NULL.
712 * lookup_mnt takes a reference to the found vfsmount.
714 struct vfsmount *lookup_mnt(const struct path *path)
716 struct mount *child_mnt;
722 seq = read_seqbegin(&mount_lock);
723 child_mnt = __lookup_mnt(path->mnt, path->dentry);
724 m = child_mnt ? &child_mnt->mnt : NULL;
725 } while (!legitimize_mnt(m, seq));
731 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
732 * current mount namespace.
734 * The common case is dentries are not mountpoints at all and that
735 * test is handled inline. For the slow case when we are actually
736 * dealing with a mountpoint of some kind, walk through all of the
737 * mounts in the current mount namespace and test to see if the dentry
740 * The mount_hashtable is not usable in the context because we
741 * need to identify all mounts that may be in the current mount
742 * namespace not just a mount that happens to have some specified
745 bool __is_local_mountpoint(struct dentry *dentry)
747 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
749 bool is_covered = false;
751 if (!d_mountpoint(dentry))
754 down_read(&namespace_sem);
755 list_for_each_entry(mnt, &ns->list, mnt_list) {
756 is_covered = (mnt->mnt_mountpoint == dentry);
760 up_read(&namespace_sem);
765 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
767 struct hlist_head *chain = mp_hash(dentry);
768 struct mountpoint *mp;
770 hlist_for_each_entry(mp, chain, m_hash) {
771 if (mp->m_dentry == dentry) {
772 /* might be worth a WARN_ON() */
773 if (d_unlinked(dentry))
774 return ERR_PTR(-ENOENT);
782 static struct mountpoint *get_mountpoint(struct dentry *dentry)
784 struct mountpoint *mp, *new = NULL;
787 if (d_mountpoint(dentry)) {
789 read_seqlock_excl(&mount_lock);
790 mp = lookup_mountpoint(dentry);
791 read_sequnlock_excl(&mount_lock);
797 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
799 return ERR_PTR(-ENOMEM);
802 /* Exactly one processes may set d_mounted */
803 ret = d_set_mounted(dentry);
805 /* Someone else set d_mounted? */
809 /* The dentry is not available as a mountpoint? */
814 /* Add the new mountpoint to the hash table */
815 read_seqlock_excl(&mount_lock);
816 new->m_dentry = dentry;
818 hlist_add_head(&new->m_hash, mp_hash(dentry));
819 INIT_HLIST_HEAD(&new->m_list);
820 read_sequnlock_excl(&mount_lock);
829 static void put_mountpoint(struct mountpoint *mp)
831 if (!--mp->m_count) {
832 struct dentry *dentry = mp->m_dentry;
833 BUG_ON(!hlist_empty(&mp->m_list));
834 spin_lock(&dentry->d_lock);
835 dentry->d_flags &= ~DCACHE_MOUNTED;
836 spin_unlock(&dentry->d_lock);
837 hlist_del(&mp->m_hash);
842 static inline int check_mnt(struct mount *mnt)
844 return mnt->mnt_ns == current->nsproxy->mnt_ns;
848 * vfsmount lock must be held for write
850 static void touch_mnt_namespace(struct mnt_namespace *ns)
854 wake_up_interruptible(&ns->poll);
859 * vfsmount lock must be held for write
861 static void __touch_mnt_namespace(struct mnt_namespace *ns)
863 if (ns && ns->event != event) {
865 wake_up_interruptible(&ns->poll);
870 * vfsmount lock must be held for write
872 static void unhash_mnt(struct mount *mnt)
874 mnt->mnt_parent = mnt;
875 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
876 list_del_init(&mnt->mnt_child);
877 hlist_del_init_rcu(&mnt->mnt_hash);
878 hlist_del_init(&mnt->mnt_mp_list);
879 put_mountpoint(mnt->mnt_mp);
884 * vfsmount lock must be held for write
886 static void detach_mnt(struct mount *mnt, struct path *old_path)
888 old_path->dentry = mnt->mnt_mountpoint;
889 old_path->mnt = &mnt->mnt_parent->mnt;
894 * vfsmount lock must be held for write
896 static void umount_mnt(struct mount *mnt)
898 /* old mountpoint will be dropped when we can do that */
899 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
904 * vfsmount lock must be held for write
906 void mnt_set_mountpoint(struct mount *mnt,
907 struct mountpoint *mp,
908 struct mount *child_mnt)
911 mnt_add_count(mnt, 1); /* essentially, that's mntget */
912 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
913 child_mnt->mnt_parent = mnt;
914 child_mnt->mnt_mp = mp;
915 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
918 static void __attach_mnt(struct mount *mnt, struct mount *parent)
920 hlist_add_head_rcu(&mnt->mnt_hash,
921 m_hash(&parent->mnt, mnt->mnt_mountpoint));
922 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
926 * vfsmount lock must be held for write
928 static void attach_mnt(struct mount *mnt,
929 struct mount *parent,
930 struct mountpoint *mp)
932 mnt_set_mountpoint(parent, mp, mnt);
933 __attach_mnt(mnt, parent);
936 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
938 struct mountpoint *old_mp = mnt->mnt_mp;
939 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
940 struct mount *old_parent = mnt->mnt_parent;
942 list_del_init(&mnt->mnt_child);
943 hlist_del_init(&mnt->mnt_mp_list);
944 hlist_del_init_rcu(&mnt->mnt_hash);
946 attach_mnt(mnt, parent, mp);
948 put_mountpoint(old_mp);
951 * Safely avoid even the suggestion this code might sleep or
952 * lock the mount hash by taking advantage of the knowledge that
953 * mnt_change_mountpoint will not release the final reference
956 * During mounting, the mount passed in as the parent mount will
957 * continue to use the old mountpoint and during unmounting, the
958 * old mountpoint will continue to exist until namespace_unlock,
959 * which happens well after mnt_change_mountpoint.
961 spin_lock(&old_mountpoint->d_lock);
962 old_mountpoint->d_lockref.count--;
963 spin_unlock(&old_mountpoint->d_lock);
965 mnt_add_count(old_parent, -1);
969 * vfsmount lock must be held for write
971 static void commit_tree(struct mount *mnt)
973 struct mount *parent = mnt->mnt_parent;
976 struct mnt_namespace *n = parent->mnt_ns;
978 BUG_ON(parent == mnt);
980 list_add_tail(&head, &mnt->mnt_list);
981 list_for_each_entry(m, &head, mnt_list)
984 list_splice(&head, n->list.prev);
986 n->mounts += n->pending_mounts;
987 n->pending_mounts = 0;
989 __attach_mnt(mnt, parent);
990 touch_mnt_namespace(n);
993 static struct mount *next_mnt(struct mount *p, struct mount *root)
995 struct list_head *next = p->mnt_mounts.next;
996 if (next == &p->mnt_mounts) {
1000 next = p->mnt_child.next;
1001 if (next != &p->mnt_parent->mnt_mounts)
1006 return list_entry(next, struct mount, mnt_child);
1009 static struct mount *skip_mnt_tree(struct mount *p)
1011 struct list_head *prev = p->mnt_mounts.prev;
1012 while (prev != &p->mnt_mounts) {
1013 p = list_entry(prev, struct mount, mnt_child);
1014 prev = p->mnt_mounts.prev;
1020 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
1023 struct dentry *root;
1026 return ERR_PTR(-ENODEV);
1028 mnt = alloc_vfsmnt(name);
1030 return ERR_PTR(-ENOMEM);
1032 if (flags & SB_KERNMOUNT)
1033 mnt->mnt.mnt_flags = MNT_INTERNAL;
1035 root = mount_fs(type, flags, name, data);
1039 return ERR_CAST(root);
1042 mnt->mnt.mnt_root = root;
1043 mnt->mnt.mnt_sb = root->d_sb;
1044 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1045 mnt->mnt_parent = mnt;
1047 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1048 unlock_mount_hash();
1051 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1054 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1055 const char *name, void *data)
1057 /* Until it is worked out how to pass the user namespace
1058 * through from the parent mount to the submount don't support
1059 * unprivileged mounts with submounts.
1061 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1062 return ERR_PTR(-EPERM);
1064 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1066 EXPORT_SYMBOL_GPL(vfs_submount);
1068 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1071 struct super_block *sb = old->mnt.mnt_sb;
1075 mnt = alloc_vfsmnt(old->mnt_devname);
1077 return ERR_PTR(-ENOMEM);
1079 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1080 mnt->mnt_group_id = 0; /* not a peer of original */
1082 mnt->mnt_group_id = old->mnt_group_id;
1084 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1085 err = mnt_alloc_group_id(mnt);
1090 mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1091 /* Don't allow unprivileged users to change mount flags */
1092 if (flag & CL_UNPRIVILEGED) {
1093 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1095 if (mnt->mnt.mnt_flags & MNT_READONLY)
1096 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1098 if (mnt->mnt.mnt_flags & MNT_NODEV)
1099 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1101 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1102 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1104 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1105 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1108 /* Don't allow unprivileged users to reveal what is under a mount */
1109 if ((flag & CL_UNPRIVILEGED) &&
1110 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1111 mnt->mnt.mnt_flags |= MNT_LOCKED;
1113 atomic_inc(&sb->s_active);
1114 mnt->mnt.mnt_sb = sb;
1115 mnt->mnt.mnt_root = dget(root);
1116 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1117 mnt->mnt_parent = mnt;
1119 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1120 unlock_mount_hash();
1122 if ((flag & CL_SLAVE) ||
1123 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1124 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1125 mnt->mnt_master = old;
1126 CLEAR_MNT_SHARED(mnt);
1127 } else if (!(flag & CL_PRIVATE)) {
1128 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1129 list_add(&mnt->mnt_share, &old->mnt_share);
1130 if (IS_MNT_SLAVE(old))
1131 list_add(&mnt->mnt_slave, &old->mnt_slave);
1132 mnt->mnt_master = old->mnt_master;
1134 CLEAR_MNT_SHARED(mnt);
1136 if (flag & CL_MAKE_SHARED)
1137 set_mnt_shared(mnt);
1139 /* stick the duplicate mount on the same expiry list
1140 * as the original if that was on one */
1141 if (flag & CL_EXPIRE) {
1142 if (!list_empty(&old->mnt_expire))
1143 list_add(&mnt->mnt_expire, &old->mnt_expire);
1151 return ERR_PTR(err);
1154 static void cleanup_mnt(struct mount *mnt)
1157 * This probably indicates that somebody messed
1158 * up a mnt_want/drop_write() pair. If this
1159 * happens, the filesystem was probably unable
1160 * to make r/w->r/o transitions.
1163 * The locking used to deal with mnt_count decrement provides barriers,
1164 * so mnt_get_writers() below is safe.
1166 WARN_ON(mnt_get_writers(mnt));
1167 if (unlikely(mnt->mnt_pins.first))
1169 fsnotify_vfsmount_delete(&mnt->mnt);
1170 dput(mnt->mnt.mnt_root);
1171 deactivate_super(mnt->mnt.mnt_sb);
1173 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1176 static void __cleanup_mnt(struct rcu_head *head)
1178 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1181 static LLIST_HEAD(delayed_mntput_list);
1182 static void delayed_mntput(struct work_struct *unused)
1184 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1185 struct mount *m, *t;
1187 llist_for_each_entry_safe(m, t, node, mnt_llist)
1190 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1192 static void mntput_no_expire(struct mount *mnt)
1195 mnt_add_count(mnt, -1);
1196 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1201 if (mnt_get_count(mnt)) {
1203 unlock_mount_hash();
1206 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1208 unlock_mount_hash();
1211 mnt->mnt.mnt_flags |= MNT_DOOMED;
1214 list_del(&mnt->mnt_instance);
1216 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1217 struct mount *p, *tmp;
1218 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1222 unlock_mount_hash();
1224 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1225 struct task_struct *task = current;
1226 if (likely(!(task->flags & PF_KTHREAD))) {
1227 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1228 if (!task_work_add(task, &mnt->mnt_rcu, true))
1231 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1232 schedule_delayed_work(&delayed_mntput_work, 1);
1238 void mntput(struct vfsmount *mnt)
1241 struct mount *m = real_mount(mnt);
1242 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1243 if (unlikely(m->mnt_expiry_mark))
1244 m->mnt_expiry_mark = 0;
1245 mntput_no_expire(m);
1248 EXPORT_SYMBOL(mntput);
1250 struct vfsmount *mntget(struct vfsmount *mnt)
1253 mnt_add_count(real_mount(mnt), 1);
1256 EXPORT_SYMBOL(mntget);
1258 /* path_is_mountpoint() - Check if path is a mount in the current
1261 * d_mountpoint() can only be used reliably to establish if a dentry is
1262 * not mounted in any namespace and that common case is handled inline.
1263 * d_mountpoint() isn't aware of the possibility there may be multiple
1264 * mounts using a given dentry in a different namespace. This function
1265 * checks if the passed in path is a mountpoint rather than the dentry
1268 bool path_is_mountpoint(const struct path *path)
1273 if (!d_mountpoint(path->dentry))
1278 seq = read_seqbegin(&mount_lock);
1279 res = __path_is_mountpoint(path);
1280 } while (read_seqretry(&mount_lock, seq));
1285 EXPORT_SYMBOL(path_is_mountpoint);
1287 struct vfsmount *mnt_clone_internal(const struct path *path)
1290 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1293 p->mnt.mnt_flags |= MNT_INTERNAL;
1297 #ifdef CONFIG_PROC_FS
1298 /* iterator; we want it to have access to namespace_sem, thus here... */
1299 static void *m_start(struct seq_file *m, loff_t *pos)
1301 struct proc_mounts *p = m->private;
1303 down_read(&namespace_sem);
1304 if (p->cached_event == p->ns->event) {
1305 void *v = p->cached_mount;
1306 if (*pos == p->cached_index)
1308 if (*pos == p->cached_index + 1) {
1309 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1310 return p->cached_mount = v;
1314 p->cached_event = p->ns->event;
1315 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1316 p->cached_index = *pos;
1317 return p->cached_mount;
1320 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1322 struct proc_mounts *p = m->private;
1324 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1325 p->cached_index = *pos;
1326 return p->cached_mount;
1329 static void m_stop(struct seq_file *m, void *v)
1331 up_read(&namespace_sem);
1334 static int m_show(struct seq_file *m, void *v)
1336 struct proc_mounts *p = m->private;
1337 struct mount *r = list_entry(v, struct mount, mnt_list);
1338 return p->show(m, &r->mnt);
1341 const struct seq_operations mounts_op = {
1347 #endif /* CONFIG_PROC_FS */
1350 * may_umount_tree - check if a mount tree is busy
1351 * @mnt: root of mount tree
1353 * This is called to check if a tree of mounts has any
1354 * open files, pwds, chroots or sub mounts that are
1357 int may_umount_tree(struct vfsmount *m)
1359 struct mount *mnt = real_mount(m);
1360 int actual_refs = 0;
1361 int minimum_refs = 0;
1365 /* write lock needed for mnt_get_count */
1367 for (p = mnt; p; p = next_mnt(p, mnt)) {
1368 actual_refs += mnt_get_count(p);
1371 unlock_mount_hash();
1373 if (actual_refs > minimum_refs)
1379 EXPORT_SYMBOL(may_umount_tree);
1382 * may_umount - check if a mount point is busy
1383 * @mnt: root of mount
1385 * This is called to check if a mount point has any
1386 * open files, pwds, chroots or sub mounts. If the
1387 * mount has sub mounts this will return busy
1388 * regardless of whether the sub mounts are busy.
1390 * Doesn't take quota and stuff into account. IOW, in some cases it will
1391 * give false negatives. The main reason why it's here is that we need
1392 * a non-destructive way to look for easily umountable filesystems.
1394 int may_umount(struct vfsmount *mnt)
1397 down_read(&namespace_sem);
1399 if (propagate_mount_busy(real_mount(mnt), 2))
1401 unlock_mount_hash();
1402 up_read(&namespace_sem);
1406 EXPORT_SYMBOL(may_umount);
1408 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1410 static void namespace_unlock(void)
1412 struct hlist_head head;
1414 hlist_move_list(&unmounted, &head);
1416 up_write(&namespace_sem);
1418 if (likely(hlist_empty(&head)))
1423 group_pin_kill(&head);
1426 static inline void namespace_lock(void)
1428 down_write(&namespace_sem);
1431 enum umount_tree_flags {
1433 UMOUNT_PROPAGATE = 2,
1434 UMOUNT_CONNECTED = 4,
1437 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1439 /* Leaving mounts connected is only valid for lazy umounts */
1440 if (how & UMOUNT_SYNC)
1443 /* A mount without a parent has nothing to be connected to */
1444 if (!mnt_has_parent(mnt))
1447 /* Because the reference counting rules change when mounts are
1448 * unmounted and connected, umounted mounts may not be
1449 * connected to mounted mounts.
1451 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1454 /* Has it been requested that the mount remain connected? */
1455 if (how & UMOUNT_CONNECTED)
1458 /* Is the mount locked such that it needs to remain connected? */
1459 if (IS_MNT_LOCKED(mnt))
1462 /* By default disconnect the mount */
1467 * mount_lock must be held
1468 * namespace_sem must be held for write
1470 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1472 LIST_HEAD(tmp_list);
1475 if (how & UMOUNT_PROPAGATE)
1476 propagate_mount_unlock(mnt);
1478 /* Gather the mounts to umount */
1479 for (p = mnt; p; p = next_mnt(p, mnt)) {
1480 p->mnt.mnt_flags |= MNT_UMOUNT;
1481 list_move(&p->mnt_list, &tmp_list);
1484 /* Hide the mounts from mnt_mounts */
1485 list_for_each_entry(p, &tmp_list, mnt_list) {
1486 list_del_init(&p->mnt_child);
1489 /* Add propogated mounts to the tmp_list */
1490 if (how & UMOUNT_PROPAGATE)
1491 propagate_umount(&tmp_list);
1493 while (!list_empty(&tmp_list)) {
1494 struct mnt_namespace *ns;
1496 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1497 list_del_init(&p->mnt_expire);
1498 list_del_init(&p->mnt_list);
1502 __touch_mnt_namespace(ns);
1505 if (how & UMOUNT_SYNC)
1506 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1508 disconnect = disconnect_mount(p, how);
1510 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1511 disconnect ? &unmounted : NULL);
1512 if (mnt_has_parent(p)) {
1513 mnt_add_count(p->mnt_parent, -1);
1515 /* Don't forget about p */
1516 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1521 change_mnt_propagation(p, MS_PRIVATE);
1525 static void shrink_submounts(struct mount *mnt);
1527 static int do_umount(struct mount *mnt, int flags)
1529 struct super_block *sb = mnt->mnt.mnt_sb;
1532 retval = security_sb_umount(&mnt->mnt, flags);
1537 * Allow userspace to request a mountpoint be expired rather than
1538 * unmounting unconditionally. Unmount only happens if:
1539 * (1) the mark is already set (the mark is cleared by mntput())
1540 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1542 if (flags & MNT_EXPIRE) {
1543 if (&mnt->mnt == current->fs->root.mnt ||
1544 flags & (MNT_FORCE | MNT_DETACH))
1548 * probably don't strictly need the lock here if we examined
1549 * all race cases, but it's a slowpath.
1552 if (mnt_get_count(mnt) != 2) {
1553 unlock_mount_hash();
1556 unlock_mount_hash();
1558 if (!xchg(&mnt->mnt_expiry_mark, 1))
1563 * If we may have to abort operations to get out of this
1564 * mount, and they will themselves hold resources we must
1565 * allow the fs to do things. In the Unix tradition of
1566 * 'Gee thats tricky lets do it in userspace' the umount_begin
1567 * might fail to complete on the first run through as other tasks
1568 * must return, and the like. Thats for the mount program to worry
1569 * about for the moment.
1572 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1573 sb->s_op->umount_begin(sb);
1577 * No sense to grab the lock for this test, but test itself looks
1578 * somewhat bogus. Suggestions for better replacement?
1579 * Ho-hum... In principle, we might treat that as umount + switch
1580 * to rootfs. GC would eventually take care of the old vfsmount.
1581 * Actually it makes sense, especially if rootfs would contain a
1582 * /reboot - static binary that would close all descriptors and
1583 * call reboot(9). Then init(8) could umount root and exec /reboot.
1585 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1587 * Special case for "unmounting" root ...
1588 * we just try to remount it readonly.
1590 if (!capable(CAP_SYS_ADMIN))
1592 down_write(&sb->s_umount);
1594 retval = do_remount_sb(sb, SB_RDONLY, NULL, 0);
1595 up_write(&sb->s_umount);
1603 if (flags & MNT_DETACH) {
1604 if (!list_empty(&mnt->mnt_list))
1605 umount_tree(mnt, UMOUNT_PROPAGATE);
1608 shrink_submounts(mnt);
1610 if (!propagate_mount_busy(mnt, 2)) {
1611 if (!list_empty(&mnt->mnt_list))
1612 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1616 unlock_mount_hash();
1622 * __detach_mounts - lazily unmount all mounts on the specified dentry
1624 * During unlink, rmdir, and d_drop it is possible to loose the path
1625 * to an existing mountpoint, and wind up leaking the mount.
1626 * detach_mounts allows lazily unmounting those mounts instead of
1629 * The caller may hold dentry->d_inode->i_mutex.
1631 void __detach_mounts(struct dentry *dentry)
1633 struct mountpoint *mp;
1638 mp = lookup_mountpoint(dentry);
1639 if (IS_ERR_OR_NULL(mp))
1643 while (!hlist_empty(&mp->m_list)) {
1644 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1645 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1646 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1649 else umount_tree(mnt, UMOUNT_CONNECTED);
1653 unlock_mount_hash();
1658 * Is the caller allowed to modify his namespace?
1660 static inline bool may_mount(void)
1662 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1665 static inline bool may_mandlock(void)
1667 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1670 return capable(CAP_SYS_ADMIN);
1674 * Now umount can handle mount points as well as block devices.
1675 * This is important for filesystems which use unnamed block devices.
1677 * We now support a flag for forced unmount like the other 'big iron'
1678 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1681 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1686 int lookup_flags = 0;
1688 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1694 if (!(flags & UMOUNT_NOFOLLOW))
1695 lookup_flags |= LOOKUP_FOLLOW;
1697 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1700 mnt = real_mount(path.mnt);
1702 if (path.dentry != path.mnt->mnt_root)
1704 if (!check_mnt(mnt))
1706 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1709 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1712 retval = do_umount(mnt, flags);
1714 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1716 mntput_no_expire(mnt);
1721 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1724 * The 2.0 compatible umount. No flags.
1726 SYSCALL_DEFINE1(oldumount, char __user *, name)
1728 return sys_umount(name, 0);
1733 static bool is_mnt_ns_file(struct dentry *dentry)
1735 /* Is this a proxy for a mount namespace? */
1736 return dentry->d_op == &ns_dentry_operations &&
1737 dentry->d_fsdata == &mntns_operations;
1740 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1742 return container_of(ns, struct mnt_namespace, ns);
1745 static bool mnt_ns_loop(struct dentry *dentry)
1747 /* Could bind mounting the mount namespace inode cause a
1748 * mount namespace loop?
1750 struct mnt_namespace *mnt_ns;
1751 if (!is_mnt_ns_file(dentry))
1754 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1755 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1758 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1761 struct mount *res, *p, *q, *r, *parent;
1763 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1764 return ERR_PTR(-EINVAL);
1766 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1767 return ERR_PTR(-EINVAL);
1769 res = q = clone_mnt(mnt, dentry, flag);
1773 q->mnt_mountpoint = mnt->mnt_mountpoint;
1776 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1778 if (!is_subdir(r->mnt_mountpoint, dentry))
1781 for (s = r; s; s = next_mnt(s, r)) {
1782 if (!(flag & CL_COPY_UNBINDABLE) &&
1783 IS_MNT_UNBINDABLE(s)) {
1784 s = skip_mnt_tree(s);
1787 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1788 is_mnt_ns_file(s->mnt.mnt_root)) {
1789 s = skip_mnt_tree(s);
1792 while (p != s->mnt_parent) {
1798 q = clone_mnt(p, p->mnt.mnt_root, flag);
1802 list_add_tail(&q->mnt_list, &res->mnt_list);
1803 attach_mnt(q, parent, p->mnt_mp);
1804 unlock_mount_hash();
1811 umount_tree(res, UMOUNT_SYNC);
1812 unlock_mount_hash();
1817 /* Caller should check returned pointer for errors */
1819 struct vfsmount *collect_mounts(const struct path *path)
1823 if (!check_mnt(real_mount(path->mnt)))
1824 tree = ERR_PTR(-EINVAL);
1826 tree = copy_tree(real_mount(path->mnt), path->dentry,
1827 CL_COPY_ALL | CL_PRIVATE);
1830 return ERR_CAST(tree);
1834 void drop_collected_mounts(struct vfsmount *mnt)
1838 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1839 unlock_mount_hash();
1844 * clone_private_mount - create a private clone of a path
1846 * This creates a new vfsmount, which will be the clone of @path. The new will
1847 * not be attached anywhere in the namespace and will be private (i.e. changes
1848 * to the originating mount won't be propagated into this).
1850 * Release with mntput().
1852 struct vfsmount *clone_private_mount(const struct path *path)
1854 struct mount *old_mnt = real_mount(path->mnt);
1855 struct mount *new_mnt;
1857 if (IS_MNT_UNBINDABLE(old_mnt))
1858 return ERR_PTR(-EINVAL);
1860 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1861 if (IS_ERR(new_mnt))
1862 return ERR_CAST(new_mnt);
1864 return &new_mnt->mnt;
1866 EXPORT_SYMBOL_GPL(clone_private_mount);
1868 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1869 struct vfsmount *root)
1872 int res = f(root, arg);
1875 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1876 res = f(&mnt->mnt, arg);
1883 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1887 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1888 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1889 mnt_release_group_id(p);
1893 static int invent_group_ids(struct mount *mnt, bool recurse)
1897 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1898 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1899 int err = mnt_alloc_group_id(p);
1901 cleanup_group_ids(mnt, p);
1910 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1912 unsigned int max = READ_ONCE(sysctl_mount_max);
1913 unsigned int mounts = 0, old, pending, sum;
1916 for (p = mnt; p; p = next_mnt(p, mnt))
1920 pending = ns->pending_mounts;
1921 sum = old + pending;
1925 (mounts > (max - sum)))
1928 ns->pending_mounts = pending + mounts;
1933 * @source_mnt : mount tree to be attached
1934 * @nd : place the mount tree @source_mnt is attached
1935 * @parent_nd : if non-null, detach the source_mnt from its parent and
1936 * store the parent mount and mountpoint dentry.
1937 * (done when source_mnt is moved)
1939 * NOTE: in the table below explains the semantics when a source mount
1940 * of a given type is attached to a destination mount of a given type.
1941 * ---------------------------------------------------------------------------
1942 * | BIND MOUNT OPERATION |
1943 * |**************************************************************************
1944 * | source-->| shared | private | slave | unbindable |
1948 * |**************************************************************************
1949 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1951 * |non-shared| shared (+) | private | slave (*) | invalid |
1952 * ***************************************************************************
1953 * A bind operation clones the source mount and mounts the clone on the
1954 * destination mount.
1956 * (++) the cloned mount is propagated to all the mounts in the propagation
1957 * tree of the destination mount and the cloned mount is added to
1958 * the peer group of the source mount.
1959 * (+) the cloned mount is created under the destination mount and is marked
1960 * as shared. The cloned mount is added to the peer group of the source
1962 * (+++) the mount is propagated to all the mounts in the propagation tree
1963 * of the destination mount and the cloned mount is made slave
1964 * of the same master as that of the source mount. The cloned mount
1965 * is marked as 'shared and slave'.
1966 * (*) the cloned mount is made a slave of the same master as that of the
1969 * ---------------------------------------------------------------------------
1970 * | MOVE MOUNT OPERATION |
1971 * |**************************************************************************
1972 * | source-->| shared | private | slave | unbindable |
1976 * |**************************************************************************
1977 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1979 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1980 * ***************************************************************************
1982 * (+) the mount is moved to the destination. And is then propagated to
1983 * all the mounts in the propagation tree of the destination mount.
1984 * (+*) the mount is moved to the destination.
1985 * (+++) the mount is moved to the destination and is then propagated to
1986 * all the mounts belonging to the destination mount's propagation tree.
1987 * the mount is marked as 'shared and slave'.
1988 * (*) the mount continues to be a slave at the new location.
1990 * if the source mount is a tree, the operations explained above is
1991 * applied to each mount in the tree.
1992 * Must be called without spinlocks held, since this function can sleep
1995 static int attach_recursive_mnt(struct mount *source_mnt,
1996 struct mount *dest_mnt,
1997 struct mountpoint *dest_mp,
1998 struct path *parent_path)
2000 HLIST_HEAD(tree_list);
2001 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2002 struct mountpoint *smp;
2003 struct mount *child, *p;
2004 struct hlist_node *n;
2007 /* Preallocate a mountpoint in case the new mounts need
2008 * to be tucked under other mounts.
2010 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2012 return PTR_ERR(smp);
2014 /* Is there space to add these mounts to the mount namespace? */
2016 err = count_mounts(ns, source_mnt);
2021 if (IS_MNT_SHARED(dest_mnt)) {
2022 err = invent_group_ids(source_mnt, true);
2025 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2028 goto out_cleanup_ids;
2029 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2035 detach_mnt(source_mnt, parent_path);
2036 attach_mnt(source_mnt, dest_mnt, dest_mp);
2037 touch_mnt_namespace(source_mnt->mnt_ns);
2039 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2040 commit_tree(source_mnt);
2043 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2045 hlist_del_init(&child->mnt_hash);
2046 q = __lookup_mnt(&child->mnt_parent->mnt,
2047 child->mnt_mountpoint);
2049 mnt_change_mountpoint(child, smp, q);
2052 put_mountpoint(smp);
2053 unlock_mount_hash();
2058 while (!hlist_empty(&tree_list)) {
2059 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2060 child->mnt_parent->mnt_ns->pending_mounts = 0;
2061 umount_tree(child, UMOUNT_SYNC);
2063 unlock_mount_hash();
2064 cleanup_group_ids(source_mnt, NULL);
2066 ns->pending_mounts = 0;
2068 read_seqlock_excl(&mount_lock);
2069 put_mountpoint(smp);
2070 read_sequnlock_excl(&mount_lock);
2075 static struct mountpoint *lock_mount(struct path *path)
2077 struct vfsmount *mnt;
2078 struct dentry *dentry = path->dentry;
2080 inode_lock(dentry->d_inode);
2081 if (unlikely(cant_mount(dentry))) {
2082 inode_unlock(dentry->d_inode);
2083 return ERR_PTR(-ENOENT);
2086 mnt = lookup_mnt(path);
2088 struct mountpoint *mp = get_mountpoint(dentry);
2091 inode_unlock(dentry->d_inode);
2097 inode_unlock(path->dentry->d_inode);
2100 dentry = path->dentry = dget(mnt->mnt_root);
2104 static void unlock_mount(struct mountpoint *where)
2106 struct dentry *dentry = where->m_dentry;
2108 read_seqlock_excl(&mount_lock);
2109 put_mountpoint(where);
2110 read_sequnlock_excl(&mount_lock);
2113 inode_unlock(dentry->d_inode);
2116 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2118 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2121 if (d_is_dir(mp->m_dentry) !=
2122 d_is_dir(mnt->mnt.mnt_root))
2125 return attach_recursive_mnt(mnt, p, mp, NULL);
2129 * Sanity check the flags to change_mnt_propagation.
2132 static int flags_to_propagation_type(int ms_flags)
2134 int type = ms_flags & ~(MS_REC | MS_SILENT);
2136 /* Fail if any non-propagation flags are set */
2137 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2139 /* Only one propagation flag should be set */
2140 if (!is_power_of_2(type))
2146 * recursively change the type of the mountpoint.
2148 static int do_change_type(struct path *path, int ms_flags)
2151 struct mount *mnt = real_mount(path->mnt);
2152 int recurse = ms_flags & MS_REC;
2156 if (path->dentry != path->mnt->mnt_root)
2159 type = flags_to_propagation_type(ms_flags);
2164 if (type == MS_SHARED) {
2165 err = invent_group_ids(mnt, recurse);
2171 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2172 change_mnt_propagation(m, type);
2173 unlock_mount_hash();
2180 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2182 struct mount *child;
2183 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2184 if (!is_subdir(child->mnt_mountpoint, dentry))
2187 if (child->mnt.mnt_flags & MNT_LOCKED)
2194 * do loopback mount.
2196 static int do_loopback(struct path *path, const char *old_name,
2199 struct path old_path;
2200 struct mount *mnt = NULL, *old, *parent;
2201 struct mountpoint *mp;
2203 if (!old_name || !*old_name)
2205 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2210 if (mnt_ns_loop(old_path.dentry))
2213 mp = lock_mount(path);
2218 old = real_mount(old_path.mnt);
2219 parent = real_mount(path->mnt);
2222 if (IS_MNT_UNBINDABLE(old))
2225 if (!check_mnt(parent))
2228 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2231 if (!recurse && has_locked_children(old, old_path.dentry))
2235 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2237 mnt = clone_mnt(old, old_path.dentry, 0);
2244 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2246 err = graft_tree(mnt, parent, mp);
2249 umount_tree(mnt, UMOUNT_SYNC);
2250 unlock_mount_hash();
2255 path_put(&old_path);
2259 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2262 int readonly_request = 0;
2264 if (ms_flags & MS_RDONLY)
2265 readonly_request = 1;
2266 if (readonly_request == __mnt_is_readonly(mnt))
2269 if (readonly_request)
2270 error = mnt_make_readonly(real_mount(mnt));
2272 __mnt_unmake_readonly(real_mount(mnt));
2277 * change filesystem flags. dir should be a physical root of filesystem.
2278 * If you've mounted a non-root directory somewhere and want to do remount
2279 * on it - tough luck.
2281 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2282 int mnt_flags, void *data)
2285 struct super_block *sb = path->mnt->mnt_sb;
2286 struct mount *mnt = real_mount(path->mnt);
2288 if (!check_mnt(mnt))
2291 if (path->dentry != path->mnt->mnt_root)
2294 /* Don't allow changing of locked mnt flags.
2296 * No locks need to be held here while testing the various
2297 * MNT_LOCK flags because those flags can never be cleared
2298 * once they are set.
2300 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2301 !(mnt_flags & MNT_READONLY)) {
2304 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2305 !(mnt_flags & MNT_NODEV)) {
2308 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2309 !(mnt_flags & MNT_NOSUID)) {
2312 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2313 !(mnt_flags & MNT_NOEXEC)) {
2316 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2317 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2321 err = security_sb_remount(sb, data);
2325 down_write(&sb->s_umount);
2326 if (ms_flags & MS_BIND)
2327 err = change_mount_flags(path->mnt, ms_flags);
2328 else if (!capable(CAP_SYS_ADMIN))
2331 err = do_remount_sb(sb, sb_flags, data, 0);
2334 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2335 mnt->mnt.mnt_flags = mnt_flags;
2336 touch_mnt_namespace(mnt->mnt_ns);
2337 unlock_mount_hash();
2339 up_write(&sb->s_umount);
2343 static inline int tree_contains_unbindable(struct mount *mnt)
2346 for (p = mnt; p; p = next_mnt(p, mnt)) {
2347 if (IS_MNT_UNBINDABLE(p))
2353 static int do_move_mount(struct path *path, const char *old_name)
2355 struct path old_path, parent_path;
2358 struct mountpoint *mp;
2360 if (!old_name || !*old_name)
2362 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2366 mp = lock_mount(path);
2371 old = real_mount(old_path.mnt);
2372 p = real_mount(path->mnt);
2375 if (!check_mnt(p) || !check_mnt(old))
2378 if (old->mnt.mnt_flags & MNT_LOCKED)
2382 if (old_path.dentry != old_path.mnt->mnt_root)
2385 if (!mnt_has_parent(old))
2388 if (d_is_dir(path->dentry) !=
2389 d_is_dir(old_path.dentry))
2392 * Don't move a mount residing in a shared parent.
2394 if (IS_MNT_SHARED(old->mnt_parent))
2397 * Don't move a mount tree containing unbindable mounts to a destination
2398 * mount which is shared.
2400 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2403 for (; mnt_has_parent(p); p = p->mnt_parent)
2407 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2411 /* if the mount is moved, it should no longer be expire
2413 list_del_init(&old->mnt_expire);
2418 path_put(&parent_path);
2419 path_put(&old_path);
2423 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2426 const char *subtype = strchr(fstype, '.');
2435 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2437 if (!mnt->mnt_sb->s_subtype)
2443 return ERR_PTR(err);
2447 * add a mount into a namespace's mount tree
2449 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2451 struct mountpoint *mp;
2452 struct mount *parent;
2455 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2457 mp = lock_mount(path);
2461 parent = real_mount(path->mnt);
2463 if (unlikely(!check_mnt(parent))) {
2464 /* that's acceptable only for automounts done in private ns */
2465 if (!(mnt_flags & MNT_SHRINKABLE))
2467 /* ... and for those we'd better have mountpoint still alive */
2468 if (!parent->mnt_ns)
2472 /* Refuse the same filesystem on the same mount point */
2474 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2475 path->mnt->mnt_root == path->dentry)
2479 if (d_is_symlink(newmnt->mnt.mnt_root))
2482 newmnt->mnt.mnt_flags = mnt_flags;
2483 err = graft_tree(newmnt, parent, mp);
2490 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2493 * create a new mount for userspace and request it to be added into the
2496 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2497 int mnt_flags, const char *name, void *data)
2499 struct file_system_type *type;
2500 struct vfsmount *mnt;
2506 type = get_fs_type(fstype);
2510 mnt = vfs_kern_mount(type, sb_flags, name, data);
2511 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2512 !mnt->mnt_sb->s_subtype)
2513 mnt = fs_set_subtype(mnt, fstype);
2515 put_filesystem(type);
2517 return PTR_ERR(mnt);
2519 if (mount_too_revealing(mnt, &mnt_flags)) {
2524 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2530 int finish_automount(struct vfsmount *m, struct path *path)
2532 struct mount *mnt = real_mount(m);
2534 /* The new mount record should have at least 2 refs to prevent it being
2535 * expired before we get a chance to add it
2537 BUG_ON(mnt_get_count(mnt) < 2);
2539 if (m->mnt_sb == path->mnt->mnt_sb &&
2540 m->mnt_root == path->dentry) {
2545 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2549 /* remove m from any expiration list it may be on */
2550 if (!list_empty(&mnt->mnt_expire)) {
2552 list_del_init(&mnt->mnt_expire);
2561 * mnt_set_expiry - Put a mount on an expiration list
2562 * @mnt: The mount to list.
2563 * @expiry_list: The list to add the mount to.
2565 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2569 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2573 EXPORT_SYMBOL(mnt_set_expiry);
2576 * process a list of expirable mountpoints with the intent of discarding any
2577 * mountpoints that aren't in use and haven't been touched since last we came
2580 void mark_mounts_for_expiry(struct list_head *mounts)
2582 struct mount *mnt, *next;
2583 LIST_HEAD(graveyard);
2585 if (list_empty(mounts))
2591 /* extract from the expiration list every vfsmount that matches the
2592 * following criteria:
2593 * - only referenced by its parent vfsmount
2594 * - still marked for expiry (marked on the last call here; marks are
2595 * cleared by mntput())
2597 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2598 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2599 propagate_mount_busy(mnt, 1))
2601 list_move(&mnt->mnt_expire, &graveyard);
2603 while (!list_empty(&graveyard)) {
2604 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2605 touch_mnt_namespace(mnt->mnt_ns);
2606 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2608 unlock_mount_hash();
2612 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2615 * Ripoff of 'select_parent()'
2617 * search the list of submounts for a given mountpoint, and move any
2618 * shrinkable submounts to the 'graveyard' list.
2620 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2622 struct mount *this_parent = parent;
2623 struct list_head *next;
2627 next = this_parent->mnt_mounts.next;
2629 while (next != &this_parent->mnt_mounts) {
2630 struct list_head *tmp = next;
2631 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2634 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2637 * Descend a level if the d_mounts list is non-empty.
2639 if (!list_empty(&mnt->mnt_mounts)) {
2644 if (!propagate_mount_busy(mnt, 1)) {
2645 list_move_tail(&mnt->mnt_expire, graveyard);
2650 * All done at this level ... ascend and resume the search
2652 if (this_parent != parent) {
2653 next = this_parent->mnt_child.next;
2654 this_parent = this_parent->mnt_parent;
2661 * process a list of expirable mountpoints with the intent of discarding any
2662 * submounts of a specific parent mountpoint
2664 * mount_lock must be held for write
2666 static void shrink_submounts(struct mount *mnt)
2668 LIST_HEAD(graveyard);
2671 /* extract submounts of 'mountpoint' from the expiration list */
2672 while (select_submounts(mnt, &graveyard)) {
2673 while (!list_empty(&graveyard)) {
2674 m = list_first_entry(&graveyard, struct mount,
2676 touch_mnt_namespace(m->mnt_ns);
2677 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2683 * Some copy_from_user() implementations do not return the exact number of
2684 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2685 * Note that this function differs from copy_from_user() in that it will oops
2686 * on bad values of `to', rather than returning a short copy.
2688 static long exact_copy_from_user(void *to, const void __user * from,
2692 const char __user *f = from;
2695 if (!access_ok(VERIFY_READ, from, n))
2699 if (__get_user(c, f)) {
2710 void *copy_mount_options(const void __user * data)
2719 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2721 return ERR_PTR(-ENOMEM);
2723 /* We only care that *some* data at the address the user
2724 * gave us is valid. Just in case, we'll zero
2725 * the remainder of the page.
2727 /* copy_from_user cannot cross TASK_SIZE ! */
2728 size = TASK_SIZE - (unsigned long)data;
2729 if (size > PAGE_SIZE)
2732 i = size - exact_copy_from_user(copy, data, size);
2735 return ERR_PTR(-EFAULT);
2738 memset(copy + i, 0, PAGE_SIZE - i);
2742 char *copy_mount_string(const void __user *data)
2744 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2748 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2749 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2751 * data is a (void *) that can point to any structure up to
2752 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2753 * information (or be NULL).
2755 * Pre-0.97 versions of mount() didn't have a flags word.
2756 * When the flags word was introduced its top half was required
2757 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2758 * Therefore, if this magic number is present, it carries no information
2759 * and must be discarded.
2761 long do_mount(const char *dev_name, const char __user *dir_name,
2762 const char *type_page, unsigned long flags, void *data_page)
2765 unsigned int mnt_flags = 0, sb_flags;
2769 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2770 flags &= ~MS_MGC_MSK;
2772 /* Basic sanity checks */
2774 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2776 if (flags & MS_NOUSER)
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 & SB_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 & SB_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 sb_flags = flags & (SB_RDONLY |
2828 if (flags & MS_REMOUNT)
2829 retval = do_remount(&path, flags, sb_flags, mnt_flags,
2831 else if (flags & MS_BIND)
2832 retval = do_loopback(&path, dev_name, flags & MS_REC);
2833 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2834 retval = do_change_type(&path, flags);
2835 else if (flags & MS_MOVE)
2836 retval = do_move_mount(&path, dev_name);
2838 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2839 dev_name, data_page);
2845 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2847 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2850 static void dec_mnt_namespaces(struct ucounts *ucounts)
2852 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2855 static void free_mnt_ns(struct mnt_namespace *ns)
2857 ns_free_inum(&ns->ns);
2858 dec_mnt_namespaces(ns->ucounts);
2859 put_user_ns(ns->user_ns);
2864 * Assign a sequence number so we can detect when we attempt to bind
2865 * mount a reference to an older mount namespace into the current
2866 * mount namespace, preventing reference counting loops. A 64bit
2867 * number incrementing at 10Ghz will take 12,427 years to wrap which
2868 * is effectively never, so we can ignore the possibility.
2870 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2872 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2874 struct mnt_namespace *new_ns;
2875 struct ucounts *ucounts;
2878 ucounts = inc_mnt_namespaces(user_ns);
2880 return ERR_PTR(-ENOSPC);
2882 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2884 dec_mnt_namespaces(ucounts);
2885 return ERR_PTR(-ENOMEM);
2887 ret = ns_alloc_inum(&new_ns->ns);
2890 dec_mnt_namespaces(ucounts);
2891 return ERR_PTR(ret);
2893 new_ns->ns.ops = &mntns_operations;
2894 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2895 atomic_set(&new_ns->count, 1);
2896 new_ns->root = NULL;
2897 INIT_LIST_HEAD(&new_ns->list);
2898 init_waitqueue_head(&new_ns->poll);
2900 new_ns->user_ns = get_user_ns(user_ns);
2901 new_ns->ucounts = ucounts;
2903 new_ns->pending_mounts = 0;
2908 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2909 struct user_namespace *user_ns, struct fs_struct *new_fs)
2911 struct mnt_namespace *new_ns;
2912 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2913 struct mount *p, *q;
2920 if (likely(!(flags & CLONE_NEWNS))) {
2927 new_ns = alloc_mnt_ns(user_ns);
2932 /* First pass: copy the tree topology */
2933 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2934 if (user_ns != ns->user_ns)
2935 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2936 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2939 free_mnt_ns(new_ns);
2940 return ERR_CAST(new);
2943 list_add_tail(&new_ns->list, &new->mnt_list);
2946 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2947 * as belonging to new namespace. We have already acquired a private
2948 * fs_struct, so tsk->fs->lock is not needed.
2956 if (&p->mnt == new_fs->root.mnt) {
2957 new_fs->root.mnt = mntget(&q->mnt);
2960 if (&p->mnt == new_fs->pwd.mnt) {
2961 new_fs->pwd.mnt = mntget(&q->mnt);
2965 p = next_mnt(p, old);
2966 q = next_mnt(q, new);
2969 while (p->mnt.mnt_root != q->mnt.mnt_root)
2970 p = next_mnt(p, old);
2983 * create_mnt_ns - creates a private namespace and adds a root filesystem
2984 * @mnt: pointer to the new root filesystem mountpoint
2986 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2988 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2989 if (!IS_ERR(new_ns)) {
2990 struct mount *mnt = real_mount(m);
2991 mnt->mnt_ns = new_ns;
2994 list_add(&mnt->mnt_list, &new_ns->list);
3001 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3003 struct mnt_namespace *ns;
3004 struct super_block *s;
3008 ns = create_mnt_ns(mnt);
3010 return ERR_CAST(ns);
3012 err = vfs_path_lookup(mnt->mnt_root, mnt,
3013 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3018 return ERR_PTR(err);
3020 /* trade a vfsmount reference for active sb one */
3021 s = path.mnt->mnt_sb;
3022 atomic_inc(&s->s_active);
3024 /* lock the sucker */
3025 down_write(&s->s_umount);
3026 /* ... and return the root of (sub)tree on it */
3029 EXPORT_SYMBOL(mount_subtree);
3031 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3032 char __user *, type, unsigned long, flags, void __user *, data)
3039 kernel_type = copy_mount_string(type);
3040 ret = PTR_ERR(kernel_type);
3041 if (IS_ERR(kernel_type))
3044 kernel_dev = copy_mount_string(dev_name);
3045 ret = PTR_ERR(kernel_dev);
3046 if (IS_ERR(kernel_dev))
3049 options = copy_mount_options(data);
3050 ret = PTR_ERR(options);
3051 if (IS_ERR(options))
3054 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3066 * Return true if path is reachable from root
3068 * namespace_sem or mount_lock is held
3070 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3071 const struct path *root)
3073 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3074 dentry = mnt->mnt_mountpoint;
3075 mnt = mnt->mnt_parent;
3077 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3080 bool path_is_under(const struct path *path1, const struct path *path2)
3083 read_seqlock_excl(&mount_lock);
3084 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3085 read_sequnlock_excl(&mount_lock);
3088 EXPORT_SYMBOL(path_is_under);
3091 * pivot_root Semantics:
3092 * Moves the root file system of the current process to the directory put_old,
3093 * makes new_root as the new root file system of the current process, and sets
3094 * root/cwd of all processes which had them on the current root to new_root.
3097 * The new_root and put_old must be directories, and must not be on the
3098 * same file system as the current process root. The put_old must be
3099 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3100 * pointed to by put_old must yield the same directory as new_root. No other
3101 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3103 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3104 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3105 * in this situation.
3108 * - we don't move root/cwd if they are not at the root (reason: if something
3109 * cared enough to change them, it's probably wrong to force them elsewhere)
3110 * - it's okay to pick a root that isn't the root of a file system, e.g.
3111 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3112 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3115 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3116 const char __user *, put_old)
3118 struct path new, old, parent_path, root_parent, root;
3119 struct mount *new_mnt, *root_mnt, *old_mnt;
3120 struct mountpoint *old_mp, *root_mp;
3126 error = user_path_dir(new_root, &new);
3130 error = user_path_dir(put_old, &old);
3134 error = security_sb_pivotroot(&old, &new);
3138 get_fs_root(current->fs, &root);
3139 old_mp = lock_mount(&old);
3140 error = PTR_ERR(old_mp);
3145 new_mnt = real_mount(new.mnt);
3146 root_mnt = real_mount(root.mnt);
3147 old_mnt = real_mount(old.mnt);
3148 if (IS_MNT_SHARED(old_mnt) ||
3149 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3150 IS_MNT_SHARED(root_mnt->mnt_parent))
3152 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3154 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3157 if (d_unlinked(new.dentry))
3160 if (new_mnt == root_mnt || old_mnt == root_mnt)
3161 goto out4; /* loop, on the same file system */
3163 if (root.mnt->mnt_root != root.dentry)
3164 goto out4; /* not a mountpoint */
3165 if (!mnt_has_parent(root_mnt))
3166 goto out4; /* not attached */
3167 root_mp = root_mnt->mnt_mp;
3168 if (new.mnt->mnt_root != new.dentry)
3169 goto out4; /* not a mountpoint */
3170 if (!mnt_has_parent(new_mnt))
3171 goto out4; /* not attached */
3172 /* make sure we can reach put_old from new_root */
3173 if (!is_path_reachable(old_mnt, old.dentry, &new))
3175 /* make certain new is below the root */
3176 if (!is_path_reachable(new_mnt, new.dentry, &root))
3178 root_mp->m_count++; /* pin it so it won't go away */
3180 detach_mnt(new_mnt, &parent_path);
3181 detach_mnt(root_mnt, &root_parent);
3182 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3183 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3184 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3186 /* mount old root on put_old */
3187 attach_mnt(root_mnt, old_mnt, old_mp);
3188 /* mount new_root on / */
3189 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3190 touch_mnt_namespace(current->nsproxy->mnt_ns);
3191 /* A moved mount should not expire automatically */
3192 list_del_init(&new_mnt->mnt_expire);
3193 put_mountpoint(root_mp);
3194 unlock_mount_hash();
3195 chroot_fs_refs(&root, &new);
3198 unlock_mount(old_mp);
3200 path_put(&root_parent);
3201 path_put(&parent_path);
3213 static void __init init_mount_tree(void)
3215 struct vfsmount *mnt;
3216 struct mnt_namespace *ns;
3218 struct file_system_type *type;
3220 type = get_fs_type("rootfs");
3222 panic("Can't find rootfs type");
3223 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3224 put_filesystem(type);
3226 panic("Can't create rootfs");
3228 ns = create_mnt_ns(mnt);
3230 panic("Can't allocate initial namespace");
3232 init_task.nsproxy->mnt_ns = ns;
3236 root.dentry = mnt->mnt_root;
3237 mnt->mnt_flags |= MNT_LOCKED;
3239 set_fs_pwd(current->fs, &root);
3240 set_fs_root(current->fs, &root);
3243 void __init mnt_init(void)
3247 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3248 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3250 mount_hashtable = alloc_large_system_hash("Mount-cache",
3251 sizeof(struct hlist_head),
3254 &m_hash_shift, &m_hash_mask, 0, 0);
3255 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3256 sizeof(struct hlist_head),
3259 &mp_hash_shift, &mp_hash_mask, 0, 0);
3261 if (!mount_hashtable || !mountpoint_hashtable)
3262 panic("Failed to allocate mount hash table\n");
3268 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3270 fs_kobj = kobject_create_and_add("fs", NULL);
3272 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3277 void put_mnt_ns(struct mnt_namespace *ns)
3279 if (!atomic_dec_and_test(&ns->count))
3281 drop_collected_mounts(&ns->root->mnt);
3285 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3287 struct vfsmount *mnt;
3288 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data);
3291 * it is a longterm mount, don't release mnt until
3292 * we unmount before file sys is unregistered
3294 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3298 EXPORT_SYMBOL_GPL(kern_mount_data);
3300 void kern_unmount(struct vfsmount *mnt)
3302 /* release long term mount so mount point can be released */
3303 if (!IS_ERR_OR_NULL(mnt)) {
3304 real_mount(mnt)->mnt_ns = NULL;
3305 synchronize_rcu(); /* yecchhh... */
3309 EXPORT_SYMBOL(kern_unmount);
3311 bool our_mnt(struct vfsmount *mnt)
3313 return check_mnt(real_mount(mnt));
3316 bool current_chrooted(void)
3318 /* Does the current process have a non-standard root */
3319 struct path ns_root;
3320 struct path fs_root;
3323 /* Find the namespace root */
3324 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3325 ns_root.dentry = ns_root.mnt->mnt_root;
3327 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3330 get_fs_root(current->fs, &fs_root);
3332 chrooted = !path_equal(&fs_root, &ns_root);
3340 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3343 int new_flags = *new_mnt_flags;
3345 bool visible = false;
3347 down_read(&namespace_sem);
3348 list_for_each_entry(mnt, &ns->list, mnt_list) {
3349 struct mount *child;
3352 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3355 /* This mount is not fully visible if it's root directory
3356 * is not the root directory of the filesystem.
3358 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3361 /* A local view of the mount flags */
3362 mnt_flags = mnt->mnt.mnt_flags;
3364 /* Don't miss readonly hidden in the superblock flags */
3365 if (sb_rdonly(mnt->mnt.mnt_sb))
3366 mnt_flags |= MNT_LOCK_READONLY;
3368 /* Verify the mount flags are equal to or more permissive
3369 * than the proposed new mount.
3371 if ((mnt_flags & MNT_LOCK_READONLY) &&
3372 !(new_flags & MNT_READONLY))
3374 if ((mnt_flags & MNT_LOCK_ATIME) &&
3375 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3378 /* This mount is not fully visible if there are any
3379 * locked child mounts that cover anything except for
3380 * empty directories.
3382 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3383 struct inode *inode = child->mnt_mountpoint->d_inode;
3384 /* Only worry about locked mounts */
3385 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3387 /* Is the directory permanetly empty? */
3388 if (!is_empty_dir_inode(inode))
3391 /* Preserve the locked attributes */
3392 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3399 up_read(&namespace_sem);
3403 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3405 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3406 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3407 unsigned long s_iflags;
3409 if (ns->user_ns == &init_user_ns)
3412 /* Can this filesystem be too revealing? */
3413 s_iflags = mnt->mnt_sb->s_iflags;
3414 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3417 if ((s_iflags & required_iflags) != required_iflags) {
3418 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3423 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3426 bool mnt_may_suid(struct vfsmount *mnt)
3429 * Foreign mounts (accessed via fchdir or through /proc
3430 * symlinks) are always treated as if they are nosuid. This
3431 * prevents namespaces from trusting potentially unsafe
3432 * suid/sgid bits, file caps, or security labels that originate
3433 * in other namespaces.
3435 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3436 current_in_userns(mnt->mnt_sb->s_user_ns);
3439 static struct ns_common *mntns_get(struct task_struct *task)
3441 struct ns_common *ns = NULL;
3442 struct nsproxy *nsproxy;
3445 nsproxy = task->nsproxy;
3447 ns = &nsproxy->mnt_ns->ns;
3448 get_mnt_ns(to_mnt_ns(ns));
3455 static void mntns_put(struct ns_common *ns)
3457 put_mnt_ns(to_mnt_ns(ns));
3460 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3462 struct fs_struct *fs = current->fs;
3463 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3467 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3468 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3469 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3476 old_mnt_ns = nsproxy->mnt_ns;
3477 nsproxy->mnt_ns = mnt_ns;
3480 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3481 "/", LOOKUP_DOWN, &root);
3483 /* revert to old namespace */
3484 nsproxy->mnt_ns = old_mnt_ns;
3489 put_mnt_ns(old_mnt_ns);
3491 /* Update the pwd and root */
3492 set_fs_pwd(fs, &root);
3493 set_fs_root(fs, &root);
3499 static struct user_namespace *mntns_owner(struct ns_common *ns)
3501 return to_mnt_ns(ns)->user_ns;
3504 const struct proc_ns_operations mntns_operations = {
3506 .type = CLONE_NEWNS,
3509 .install = mntns_install,
3510 .owner = mntns_owner,