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_inode(file)->i_sb);
450 ret = __mnt_want_write_file(file);
452 sb_end_write(file_inode(file)->i_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);
472 (file_inode(file) == d_inode(upperdentry) ||
473 file_inode(file) == d_inode(dentry)))
476 /* Lower layer: can't write to real file, sorry... */
481 * mnt_want_write_file - get write access to a file's mount
482 * @file: the file who's mount on which to take a write
484 * This is like mnt_want_write, but it takes a file and can
485 * do some optimisations if the file is open for write already
487 * Mostly called by filesystems from their ioctl operation before performing
488 * modification. On overlayfs this needs to check if the file is on a read-only
489 * lower layer and deny access in that case.
491 int mnt_want_write_file(struct file *file)
495 ret = may_write_real(file);
497 sb_start_write(file_inode(file)->i_sb);
498 ret = __mnt_want_write_file(file);
500 sb_end_write(file_inode(file)->i_sb);
504 EXPORT_SYMBOL_GPL(mnt_want_write_file);
507 * __mnt_drop_write - give up write access to a mount
508 * @mnt: the mount on which to give up write access
510 * Tells the low-level filesystem that we are done
511 * performing writes to it. Must be matched with
512 * __mnt_want_write() call above.
514 void __mnt_drop_write(struct vfsmount *mnt)
517 mnt_dec_writers(real_mount(mnt));
522 * mnt_drop_write - give up write access to a mount
523 * @mnt: the mount on which to give up write access
525 * Tells the low-level filesystem that we are done performing writes to it and
526 * also allows filesystem to be frozen again. Must be matched with
527 * mnt_want_write() call above.
529 void mnt_drop_write(struct vfsmount *mnt)
531 __mnt_drop_write(mnt);
532 sb_end_write(mnt->mnt_sb);
534 EXPORT_SYMBOL_GPL(mnt_drop_write);
536 void __mnt_drop_write_file(struct file *file)
538 __mnt_drop_write(file->f_path.mnt);
541 void mnt_drop_write_file_path(struct file *file)
543 __mnt_drop_write_file(file);
544 sb_end_write(file_inode(file)->i_sb);
547 void mnt_drop_write_file(struct file *file)
549 __mnt_drop_write(file->f_path.mnt);
550 sb_end_write(file_inode(file)->i_sb);
552 EXPORT_SYMBOL(mnt_drop_write_file);
554 static int mnt_make_readonly(struct mount *mnt)
559 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
561 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
562 * should be visible before we do.
567 * With writers on hold, if this value is zero, then there are
568 * definitely no active writers (although held writers may subsequently
569 * increment the count, they'll have to wait, and decrement it after
570 * seeing MNT_READONLY).
572 * It is OK to have counter incremented on one CPU and decremented on
573 * another: the sum will add up correctly. The danger would be when we
574 * sum up each counter, if we read a counter before it is incremented,
575 * but then read another CPU's count which it has been subsequently
576 * decremented from -- we would see more decrements than we should.
577 * MNT_WRITE_HOLD protects against this scenario, because
578 * mnt_want_write first increments count, then smp_mb, then spins on
579 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
580 * we're counting up here.
582 if (mnt_get_writers(mnt) > 0)
585 mnt->mnt.mnt_flags |= MNT_READONLY;
587 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
588 * that become unheld will see MNT_READONLY.
591 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
596 static void __mnt_unmake_readonly(struct mount *mnt)
599 mnt->mnt.mnt_flags &= ~MNT_READONLY;
603 int sb_prepare_remount_readonly(struct super_block *sb)
608 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
609 if (atomic_long_read(&sb->s_remove_count))
613 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
614 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
615 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
617 if (mnt_get_writers(mnt) > 0) {
623 if (!err && atomic_long_read(&sb->s_remove_count))
627 sb->s_readonly_remount = 1;
630 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
631 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
632 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
639 static void free_vfsmnt(struct mount *mnt)
641 kfree_const(mnt->mnt_devname);
643 free_percpu(mnt->mnt_pcp);
645 kmem_cache_free(mnt_cache, mnt);
648 static void delayed_free_vfsmnt(struct rcu_head *head)
650 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
653 /* call under rcu_read_lock */
654 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
657 if (read_seqretry(&mount_lock, seq))
661 mnt = real_mount(bastard);
662 mnt_add_count(mnt, 1);
663 smp_mb(); // see mntput_no_expire()
664 if (likely(!read_seqretry(&mount_lock, seq)))
666 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
667 mnt_add_count(mnt, -1);
671 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
672 mnt_add_count(mnt, -1);
677 /* caller will mntput() */
681 /* call under rcu_read_lock */
682 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
684 int res = __legitimize_mnt(bastard, seq);
687 if (unlikely(res < 0)) {
696 * find the first mount at @dentry on vfsmount @mnt.
697 * call under rcu_read_lock()
699 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
701 struct hlist_head *head = m_hash(mnt, dentry);
704 hlist_for_each_entry_rcu(p, head, mnt_hash)
705 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
711 * lookup_mnt - Return the first child mount mounted at path
713 * "First" means first mounted chronologically. If you create the
716 * mount /dev/sda1 /mnt
717 * mount /dev/sda2 /mnt
718 * mount /dev/sda3 /mnt
720 * Then lookup_mnt() on the base /mnt dentry in the root mount will
721 * return successively the root dentry and vfsmount of /dev/sda1, then
722 * /dev/sda2, then /dev/sda3, then NULL.
724 * lookup_mnt takes a reference to the found vfsmount.
726 struct vfsmount *lookup_mnt(const struct path *path)
728 struct mount *child_mnt;
734 seq = read_seqbegin(&mount_lock);
735 child_mnt = __lookup_mnt(path->mnt, path->dentry);
736 m = child_mnt ? &child_mnt->mnt : NULL;
737 } while (!legitimize_mnt(m, seq));
743 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
744 * current mount namespace.
746 * The common case is dentries are not mountpoints at all and that
747 * test is handled inline. For the slow case when we are actually
748 * dealing with a mountpoint of some kind, walk through all of the
749 * mounts in the current mount namespace and test to see if the dentry
752 * The mount_hashtable is not usable in the context because we
753 * need to identify all mounts that may be in the current mount
754 * namespace not just a mount that happens to have some specified
757 bool __is_local_mountpoint(struct dentry *dentry)
759 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
761 bool is_covered = false;
763 if (!d_mountpoint(dentry))
766 down_read(&namespace_sem);
767 list_for_each_entry(mnt, &ns->list, mnt_list) {
768 is_covered = (mnt->mnt_mountpoint == dentry);
772 up_read(&namespace_sem);
777 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
779 struct hlist_head *chain = mp_hash(dentry);
780 struct mountpoint *mp;
782 hlist_for_each_entry(mp, chain, m_hash) {
783 if (mp->m_dentry == dentry) {
784 /* might be worth a WARN_ON() */
785 if (d_unlinked(dentry))
786 return ERR_PTR(-ENOENT);
794 static struct mountpoint *get_mountpoint(struct dentry *dentry)
796 struct mountpoint *mp, *new = NULL;
799 if (d_mountpoint(dentry)) {
801 read_seqlock_excl(&mount_lock);
802 mp = lookup_mountpoint(dentry);
803 read_sequnlock_excl(&mount_lock);
809 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
811 return ERR_PTR(-ENOMEM);
814 /* Exactly one processes may set d_mounted */
815 ret = d_set_mounted(dentry);
817 /* Someone else set d_mounted? */
821 /* The dentry is not available as a mountpoint? */
826 /* Add the new mountpoint to the hash table */
827 read_seqlock_excl(&mount_lock);
828 new->m_dentry = dentry;
830 hlist_add_head(&new->m_hash, mp_hash(dentry));
831 INIT_HLIST_HEAD(&new->m_list);
832 read_sequnlock_excl(&mount_lock);
841 static void put_mountpoint(struct mountpoint *mp)
843 if (!--mp->m_count) {
844 struct dentry *dentry = mp->m_dentry;
845 BUG_ON(!hlist_empty(&mp->m_list));
846 spin_lock(&dentry->d_lock);
847 dentry->d_flags &= ~DCACHE_MOUNTED;
848 spin_unlock(&dentry->d_lock);
849 hlist_del(&mp->m_hash);
854 static inline int check_mnt(struct mount *mnt)
856 return mnt->mnt_ns == current->nsproxy->mnt_ns;
860 * vfsmount lock must be held for write
862 static void touch_mnt_namespace(struct mnt_namespace *ns)
866 wake_up_interruptible(&ns->poll);
871 * vfsmount lock must be held for write
873 static void __touch_mnt_namespace(struct mnt_namespace *ns)
875 if (ns && ns->event != event) {
877 wake_up_interruptible(&ns->poll);
882 * vfsmount lock must be held for write
884 static void unhash_mnt(struct mount *mnt)
886 mnt->mnt_parent = mnt;
887 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
888 list_del_init(&mnt->mnt_child);
889 hlist_del_init_rcu(&mnt->mnt_hash);
890 hlist_del_init(&mnt->mnt_mp_list);
891 put_mountpoint(mnt->mnt_mp);
896 * vfsmount lock must be held for write
898 static void detach_mnt(struct mount *mnt, struct path *old_path)
900 old_path->dentry = mnt->mnt_mountpoint;
901 old_path->mnt = &mnt->mnt_parent->mnt;
906 * vfsmount lock must be held for write
908 static void umount_mnt(struct mount *mnt)
910 /* old mountpoint will be dropped when we can do that */
911 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
916 * vfsmount lock must be held for write
918 void mnt_set_mountpoint(struct mount *mnt,
919 struct mountpoint *mp,
920 struct mount *child_mnt)
923 mnt_add_count(mnt, 1); /* essentially, that's mntget */
924 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
925 child_mnt->mnt_parent = mnt;
926 child_mnt->mnt_mp = mp;
927 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
930 static void __attach_mnt(struct mount *mnt, struct mount *parent)
932 hlist_add_head_rcu(&mnt->mnt_hash,
933 m_hash(&parent->mnt, mnt->mnt_mountpoint));
934 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
938 * vfsmount lock must be held for write
940 static void attach_mnt(struct mount *mnt,
941 struct mount *parent,
942 struct mountpoint *mp)
944 mnt_set_mountpoint(parent, mp, mnt);
945 __attach_mnt(mnt, parent);
948 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
950 struct mountpoint *old_mp = mnt->mnt_mp;
951 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
952 struct mount *old_parent = mnt->mnt_parent;
954 list_del_init(&mnt->mnt_child);
955 hlist_del_init(&mnt->mnt_mp_list);
956 hlist_del_init_rcu(&mnt->mnt_hash);
958 attach_mnt(mnt, parent, mp);
960 put_mountpoint(old_mp);
963 * Safely avoid even the suggestion this code might sleep or
964 * lock the mount hash by taking advantage of the knowledge that
965 * mnt_change_mountpoint will not release the final reference
968 * During mounting, the mount passed in as the parent mount will
969 * continue to use the old mountpoint and during unmounting, the
970 * old mountpoint will continue to exist until namespace_unlock,
971 * which happens well after mnt_change_mountpoint.
973 spin_lock(&old_mountpoint->d_lock);
974 old_mountpoint->d_lockref.count--;
975 spin_unlock(&old_mountpoint->d_lock);
977 mnt_add_count(old_parent, -1);
981 * vfsmount lock must be held for write
983 static void commit_tree(struct mount *mnt)
985 struct mount *parent = mnt->mnt_parent;
988 struct mnt_namespace *n = parent->mnt_ns;
990 BUG_ON(parent == mnt);
992 list_add_tail(&head, &mnt->mnt_list);
993 list_for_each_entry(m, &head, mnt_list)
996 list_splice(&head, n->list.prev);
998 n->mounts += n->pending_mounts;
999 n->pending_mounts = 0;
1001 __attach_mnt(mnt, parent);
1002 touch_mnt_namespace(n);
1005 static struct mount *next_mnt(struct mount *p, struct mount *root)
1007 struct list_head *next = p->mnt_mounts.next;
1008 if (next == &p->mnt_mounts) {
1012 next = p->mnt_child.next;
1013 if (next != &p->mnt_parent->mnt_mounts)
1018 return list_entry(next, struct mount, mnt_child);
1021 static struct mount *skip_mnt_tree(struct mount *p)
1023 struct list_head *prev = p->mnt_mounts.prev;
1024 while (prev != &p->mnt_mounts) {
1025 p = list_entry(prev, struct mount, mnt_child);
1026 prev = p->mnt_mounts.prev;
1032 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
1035 struct dentry *root;
1038 return ERR_PTR(-ENODEV);
1040 mnt = alloc_vfsmnt(name);
1042 return ERR_PTR(-ENOMEM);
1044 if (flags & SB_KERNMOUNT)
1045 mnt->mnt.mnt_flags = MNT_INTERNAL;
1047 root = mount_fs(type, flags, name, data);
1051 return ERR_CAST(root);
1054 mnt->mnt.mnt_root = root;
1055 mnt->mnt.mnt_sb = root->d_sb;
1056 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1057 mnt->mnt_parent = mnt;
1059 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1060 unlock_mount_hash();
1063 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1066 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1067 const char *name, void *data)
1069 /* Until it is worked out how to pass the user namespace
1070 * through from the parent mount to the submount don't support
1071 * unprivileged mounts with submounts.
1073 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1074 return ERR_PTR(-EPERM);
1076 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1078 EXPORT_SYMBOL_GPL(vfs_submount);
1080 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1083 struct super_block *sb = old->mnt.mnt_sb;
1087 mnt = alloc_vfsmnt(old->mnt_devname);
1089 return ERR_PTR(-ENOMEM);
1091 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1092 mnt->mnt_group_id = 0; /* not a peer of original */
1094 mnt->mnt_group_id = old->mnt_group_id;
1096 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1097 err = mnt_alloc_group_id(mnt);
1102 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1103 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1104 /* Don't allow unprivileged users to change mount flags */
1105 if (flag & CL_UNPRIVILEGED) {
1106 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1108 if (mnt->mnt.mnt_flags & MNT_READONLY)
1109 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1111 if (mnt->mnt.mnt_flags & MNT_NODEV)
1112 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1114 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1115 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1117 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1118 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1121 /* Don't allow unprivileged users to reveal what is under a mount */
1122 if ((flag & CL_UNPRIVILEGED) &&
1123 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1124 mnt->mnt.mnt_flags |= MNT_LOCKED;
1126 atomic_inc(&sb->s_active);
1127 mnt->mnt.mnt_sb = sb;
1128 mnt->mnt.mnt_root = dget(root);
1129 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1130 mnt->mnt_parent = mnt;
1132 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1133 unlock_mount_hash();
1135 if ((flag & CL_SLAVE) ||
1136 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1137 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1138 mnt->mnt_master = old;
1139 CLEAR_MNT_SHARED(mnt);
1140 } else if (!(flag & CL_PRIVATE)) {
1141 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1142 list_add(&mnt->mnt_share, &old->mnt_share);
1143 if (IS_MNT_SLAVE(old))
1144 list_add(&mnt->mnt_slave, &old->mnt_slave);
1145 mnt->mnt_master = old->mnt_master;
1147 CLEAR_MNT_SHARED(mnt);
1149 if (flag & CL_MAKE_SHARED)
1150 set_mnt_shared(mnt);
1152 /* stick the duplicate mount on the same expiry list
1153 * as the original if that was on one */
1154 if (flag & CL_EXPIRE) {
1155 if (!list_empty(&old->mnt_expire))
1156 list_add(&mnt->mnt_expire, &old->mnt_expire);
1164 return ERR_PTR(err);
1167 static void cleanup_mnt(struct mount *mnt)
1170 * This probably indicates that somebody messed
1171 * up a mnt_want/drop_write() pair. If this
1172 * happens, the filesystem was probably unable
1173 * to make r/w->r/o transitions.
1176 * The locking used to deal with mnt_count decrement provides barriers,
1177 * so mnt_get_writers() below is safe.
1179 WARN_ON(mnt_get_writers(mnt));
1180 if (unlikely(mnt->mnt_pins.first))
1182 fsnotify_vfsmount_delete(&mnt->mnt);
1183 dput(mnt->mnt.mnt_root);
1184 deactivate_super(mnt->mnt.mnt_sb);
1186 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1189 static void __cleanup_mnt(struct rcu_head *head)
1191 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1194 static LLIST_HEAD(delayed_mntput_list);
1195 static void delayed_mntput(struct work_struct *unused)
1197 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1198 struct mount *m, *t;
1200 llist_for_each_entry_safe(m, t, node, mnt_llist)
1203 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1205 static void mntput_no_expire(struct mount *mnt)
1208 if (likely(READ_ONCE(mnt->mnt_ns))) {
1210 * Since we don't do lock_mount_hash() here,
1211 * ->mnt_ns can change under us. However, if it's
1212 * non-NULL, then there's a reference that won't
1213 * be dropped until after an RCU delay done after
1214 * turning ->mnt_ns NULL. So if we observe it
1215 * non-NULL under rcu_read_lock(), the reference
1216 * we are dropping is not the final one.
1218 mnt_add_count(mnt, -1);
1224 * make sure that if __legitimize_mnt() has not seen us grab
1225 * mount_lock, we'll see their refcount increment here.
1228 mnt_add_count(mnt, -1);
1229 if (mnt_get_count(mnt)) {
1231 unlock_mount_hash();
1234 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1236 unlock_mount_hash();
1239 mnt->mnt.mnt_flags |= MNT_DOOMED;
1242 list_del(&mnt->mnt_instance);
1244 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1245 struct mount *p, *tmp;
1246 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1250 unlock_mount_hash();
1252 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1253 struct task_struct *task = current;
1254 if (likely(!(task->flags & PF_KTHREAD))) {
1255 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1256 if (!task_work_add(task, &mnt->mnt_rcu, true))
1259 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1260 schedule_delayed_work(&delayed_mntput_work, 1);
1266 void mntput(struct vfsmount *mnt)
1269 struct mount *m = real_mount(mnt);
1270 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1271 if (unlikely(m->mnt_expiry_mark))
1272 m->mnt_expiry_mark = 0;
1273 mntput_no_expire(m);
1276 EXPORT_SYMBOL(mntput);
1278 struct vfsmount *mntget(struct vfsmount *mnt)
1281 mnt_add_count(real_mount(mnt), 1);
1284 EXPORT_SYMBOL(mntget);
1286 /* path_is_mountpoint() - Check if path is a mount in the current
1289 * d_mountpoint() can only be used reliably to establish if a dentry is
1290 * not mounted in any namespace and that common case is handled inline.
1291 * d_mountpoint() isn't aware of the possibility there may be multiple
1292 * mounts using a given dentry in a different namespace. This function
1293 * checks if the passed in path is a mountpoint rather than the dentry
1296 bool path_is_mountpoint(const struct path *path)
1301 if (!d_mountpoint(path->dentry))
1306 seq = read_seqbegin(&mount_lock);
1307 res = __path_is_mountpoint(path);
1308 } while (read_seqretry(&mount_lock, seq));
1313 EXPORT_SYMBOL(path_is_mountpoint);
1315 struct vfsmount *mnt_clone_internal(const struct path *path)
1318 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1321 p->mnt.mnt_flags |= MNT_INTERNAL;
1325 #ifdef CONFIG_PROC_FS
1326 /* iterator; we want it to have access to namespace_sem, thus here... */
1327 static void *m_start(struct seq_file *m, loff_t *pos)
1329 struct proc_mounts *p = m->private;
1331 down_read(&namespace_sem);
1332 if (p->cached_event == p->ns->event) {
1333 void *v = p->cached_mount;
1334 if (*pos == p->cached_index)
1336 if (*pos == p->cached_index + 1) {
1337 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1338 return p->cached_mount = v;
1342 p->cached_event = p->ns->event;
1343 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1344 p->cached_index = *pos;
1345 return p->cached_mount;
1348 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1350 struct proc_mounts *p = m->private;
1352 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1353 p->cached_index = *pos;
1354 return p->cached_mount;
1357 static void m_stop(struct seq_file *m, void *v)
1359 up_read(&namespace_sem);
1362 static int m_show(struct seq_file *m, void *v)
1364 struct proc_mounts *p = m->private;
1365 struct mount *r = list_entry(v, struct mount, mnt_list);
1366 return p->show(m, &r->mnt);
1369 const struct seq_operations mounts_op = {
1375 #endif /* CONFIG_PROC_FS */
1378 * may_umount_tree - check if a mount tree is busy
1379 * @mnt: root of mount tree
1381 * This is called to check if a tree of mounts has any
1382 * open files, pwds, chroots or sub mounts that are
1385 int may_umount_tree(struct vfsmount *m)
1387 struct mount *mnt = real_mount(m);
1388 int actual_refs = 0;
1389 int minimum_refs = 0;
1393 /* write lock needed for mnt_get_count */
1395 for (p = mnt; p; p = next_mnt(p, mnt)) {
1396 actual_refs += mnt_get_count(p);
1399 unlock_mount_hash();
1401 if (actual_refs > minimum_refs)
1407 EXPORT_SYMBOL(may_umount_tree);
1410 * may_umount - check if a mount point is busy
1411 * @mnt: root of mount
1413 * This is called to check if a mount point has any
1414 * open files, pwds, chroots or sub mounts. If the
1415 * mount has sub mounts this will return busy
1416 * regardless of whether the sub mounts are busy.
1418 * Doesn't take quota and stuff into account. IOW, in some cases it will
1419 * give false negatives. The main reason why it's here is that we need
1420 * a non-destructive way to look for easily umountable filesystems.
1422 int may_umount(struct vfsmount *mnt)
1425 down_read(&namespace_sem);
1427 if (propagate_mount_busy(real_mount(mnt), 2))
1429 unlock_mount_hash();
1430 up_read(&namespace_sem);
1434 EXPORT_SYMBOL(may_umount);
1436 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1438 static void namespace_unlock(void)
1440 struct hlist_head head;
1442 hlist_move_list(&unmounted, &head);
1444 up_write(&namespace_sem);
1446 if (likely(hlist_empty(&head)))
1451 group_pin_kill(&head);
1454 static inline void namespace_lock(void)
1456 down_write(&namespace_sem);
1459 enum umount_tree_flags {
1461 UMOUNT_PROPAGATE = 2,
1462 UMOUNT_CONNECTED = 4,
1465 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1467 /* Leaving mounts connected is only valid for lazy umounts */
1468 if (how & UMOUNT_SYNC)
1471 /* A mount without a parent has nothing to be connected to */
1472 if (!mnt_has_parent(mnt))
1475 /* Because the reference counting rules change when mounts are
1476 * unmounted and connected, umounted mounts may not be
1477 * connected to mounted mounts.
1479 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1482 /* Has it been requested that the mount remain connected? */
1483 if (how & UMOUNT_CONNECTED)
1486 /* Is the mount locked such that it needs to remain connected? */
1487 if (IS_MNT_LOCKED(mnt))
1490 /* By default disconnect the mount */
1495 * mount_lock must be held
1496 * namespace_sem must be held for write
1498 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1500 LIST_HEAD(tmp_list);
1503 if (how & UMOUNT_PROPAGATE)
1504 propagate_mount_unlock(mnt);
1506 /* Gather the mounts to umount */
1507 for (p = mnt; p; p = next_mnt(p, mnt)) {
1508 p->mnt.mnt_flags |= MNT_UMOUNT;
1509 list_move(&p->mnt_list, &tmp_list);
1512 /* Hide the mounts from mnt_mounts */
1513 list_for_each_entry(p, &tmp_list, mnt_list) {
1514 list_del_init(&p->mnt_child);
1517 /* Add propogated mounts to the tmp_list */
1518 if (how & UMOUNT_PROPAGATE)
1519 propagate_umount(&tmp_list);
1521 while (!list_empty(&tmp_list)) {
1522 struct mnt_namespace *ns;
1524 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1525 list_del_init(&p->mnt_expire);
1526 list_del_init(&p->mnt_list);
1530 __touch_mnt_namespace(ns);
1533 if (how & UMOUNT_SYNC)
1534 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1536 disconnect = disconnect_mount(p, how);
1538 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1539 disconnect ? &unmounted : NULL);
1540 if (mnt_has_parent(p)) {
1541 mnt_add_count(p->mnt_parent, -1);
1543 /* Don't forget about p */
1544 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1549 change_mnt_propagation(p, MS_PRIVATE);
1553 static void shrink_submounts(struct mount *mnt);
1555 static int do_umount(struct mount *mnt, int flags)
1557 struct super_block *sb = mnt->mnt.mnt_sb;
1560 retval = security_sb_umount(&mnt->mnt, flags);
1565 * Allow userspace to request a mountpoint be expired rather than
1566 * unmounting unconditionally. Unmount only happens if:
1567 * (1) the mark is already set (the mark is cleared by mntput())
1568 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1570 if (flags & MNT_EXPIRE) {
1571 if (&mnt->mnt == current->fs->root.mnt ||
1572 flags & (MNT_FORCE | MNT_DETACH))
1576 * probably don't strictly need the lock here if we examined
1577 * all race cases, but it's a slowpath.
1580 if (mnt_get_count(mnt) != 2) {
1581 unlock_mount_hash();
1584 unlock_mount_hash();
1586 if (!xchg(&mnt->mnt_expiry_mark, 1))
1591 * If we may have to abort operations to get out of this
1592 * mount, and they will themselves hold resources we must
1593 * allow the fs to do things. In the Unix tradition of
1594 * 'Gee thats tricky lets do it in userspace' the umount_begin
1595 * might fail to complete on the first run through as other tasks
1596 * must return, and the like. Thats for the mount program to worry
1597 * about for the moment.
1600 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1601 sb->s_op->umount_begin(sb);
1605 * No sense to grab the lock for this test, but test itself looks
1606 * somewhat bogus. Suggestions for better replacement?
1607 * Ho-hum... In principle, we might treat that as umount + switch
1608 * to rootfs. GC would eventually take care of the old vfsmount.
1609 * Actually it makes sense, especially if rootfs would contain a
1610 * /reboot - static binary that would close all descriptors and
1611 * call reboot(9). Then init(8) could umount root and exec /reboot.
1613 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1615 * Special case for "unmounting" root ...
1616 * we just try to remount it readonly.
1618 if (!capable(CAP_SYS_ADMIN))
1620 down_write(&sb->s_umount);
1622 retval = do_remount_sb(sb, SB_RDONLY, NULL, 0);
1623 up_write(&sb->s_umount);
1631 if (flags & MNT_DETACH) {
1632 if (!list_empty(&mnt->mnt_list))
1633 umount_tree(mnt, UMOUNT_PROPAGATE);
1636 shrink_submounts(mnt);
1638 if (!propagate_mount_busy(mnt, 2)) {
1639 if (!list_empty(&mnt->mnt_list))
1640 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1644 unlock_mount_hash();
1650 * __detach_mounts - lazily unmount all mounts on the specified dentry
1652 * During unlink, rmdir, and d_drop it is possible to loose the path
1653 * to an existing mountpoint, and wind up leaking the mount.
1654 * detach_mounts allows lazily unmounting those mounts instead of
1657 * The caller may hold dentry->d_inode->i_mutex.
1659 void __detach_mounts(struct dentry *dentry)
1661 struct mountpoint *mp;
1666 mp = lookup_mountpoint(dentry);
1667 if (IS_ERR_OR_NULL(mp))
1671 while (!hlist_empty(&mp->m_list)) {
1672 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1673 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1674 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1677 else umount_tree(mnt, UMOUNT_CONNECTED);
1681 unlock_mount_hash();
1686 * Is the caller allowed to modify his namespace?
1688 static inline bool may_mount(void)
1690 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1693 static inline bool may_mandlock(void)
1695 #ifndef CONFIG_MANDATORY_FILE_LOCKING
1698 return capable(CAP_SYS_ADMIN);
1702 * Now umount can handle mount points as well as block devices.
1703 * This is important for filesystems which use unnamed block devices.
1705 * We now support a flag for forced unmount like the other 'big iron'
1706 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1709 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1714 int lookup_flags = 0;
1716 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1722 if (!(flags & UMOUNT_NOFOLLOW))
1723 lookup_flags |= LOOKUP_FOLLOW;
1725 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1728 mnt = real_mount(path.mnt);
1730 if (path.dentry != path.mnt->mnt_root)
1732 if (!check_mnt(mnt))
1734 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1737 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1740 retval = do_umount(mnt, flags);
1742 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1744 mntput_no_expire(mnt);
1749 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1752 * The 2.0 compatible umount. No flags.
1754 SYSCALL_DEFINE1(oldumount, char __user *, name)
1756 return sys_umount(name, 0);
1761 static bool is_mnt_ns_file(struct dentry *dentry)
1763 /* Is this a proxy for a mount namespace? */
1764 return dentry->d_op == &ns_dentry_operations &&
1765 dentry->d_fsdata == &mntns_operations;
1768 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1770 return container_of(ns, struct mnt_namespace, ns);
1773 static bool mnt_ns_loop(struct dentry *dentry)
1775 /* Could bind mounting the mount namespace inode cause a
1776 * mount namespace loop?
1778 struct mnt_namespace *mnt_ns;
1779 if (!is_mnt_ns_file(dentry))
1782 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1783 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1786 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1789 struct mount *res, *p, *q, *r, *parent;
1791 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1792 return ERR_PTR(-EINVAL);
1794 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1795 return ERR_PTR(-EINVAL);
1797 res = q = clone_mnt(mnt, dentry, flag);
1801 q->mnt_mountpoint = mnt->mnt_mountpoint;
1804 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1806 if (!is_subdir(r->mnt_mountpoint, dentry))
1809 for (s = r; s; s = next_mnt(s, r)) {
1810 if (!(flag & CL_COPY_UNBINDABLE) &&
1811 IS_MNT_UNBINDABLE(s)) {
1812 s = skip_mnt_tree(s);
1815 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1816 is_mnt_ns_file(s->mnt.mnt_root)) {
1817 s = skip_mnt_tree(s);
1820 while (p != s->mnt_parent) {
1826 q = clone_mnt(p, p->mnt.mnt_root, flag);
1830 list_add_tail(&q->mnt_list, &res->mnt_list);
1831 attach_mnt(q, parent, p->mnt_mp);
1832 unlock_mount_hash();
1839 umount_tree(res, UMOUNT_SYNC);
1840 unlock_mount_hash();
1845 /* Caller should check returned pointer for errors */
1847 struct vfsmount *collect_mounts(const struct path *path)
1851 if (!check_mnt(real_mount(path->mnt)))
1852 tree = ERR_PTR(-EINVAL);
1854 tree = copy_tree(real_mount(path->mnt), path->dentry,
1855 CL_COPY_ALL | CL_PRIVATE);
1858 return ERR_CAST(tree);
1862 void drop_collected_mounts(struct vfsmount *mnt)
1866 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1867 unlock_mount_hash();
1872 * clone_private_mount - create a private clone of a path
1874 * This creates a new vfsmount, which will be the clone of @path. The new will
1875 * not be attached anywhere in the namespace and will be private (i.e. changes
1876 * to the originating mount won't be propagated into this).
1878 * Release with mntput().
1880 struct vfsmount *clone_private_mount(const struct path *path)
1882 struct mount *old_mnt = real_mount(path->mnt);
1883 struct mount *new_mnt;
1885 if (IS_MNT_UNBINDABLE(old_mnt))
1886 return ERR_PTR(-EINVAL);
1888 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1889 if (IS_ERR(new_mnt))
1890 return ERR_CAST(new_mnt);
1892 return &new_mnt->mnt;
1894 EXPORT_SYMBOL_GPL(clone_private_mount);
1896 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1897 struct vfsmount *root)
1900 int res = f(root, arg);
1903 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1904 res = f(&mnt->mnt, arg);
1911 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1915 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1916 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1917 mnt_release_group_id(p);
1921 static int invent_group_ids(struct mount *mnt, bool recurse)
1925 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1926 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1927 int err = mnt_alloc_group_id(p);
1929 cleanup_group_ids(mnt, p);
1938 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1940 unsigned int max = READ_ONCE(sysctl_mount_max);
1941 unsigned int mounts = 0, old, pending, sum;
1944 for (p = mnt; p; p = next_mnt(p, mnt))
1948 pending = ns->pending_mounts;
1949 sum = old + pending;
1953 (mounts > (max - sum)))
1956 ns->pending_mounts = pending + mounts;
1961 * @source_mnt : mount tree to be attached
1962 * @nd : place the mount tree @source_mnt is attached
1963 * @parent_nd : if non-null, detach the source_mnt from its parent and
1964 * store the parent mount and mountpoint dentry.
1965 * (done when source_mnt is moved)
1967 * NOTE: in the table below explains the semantics when a source mount
1968 * of a given type is attached to a destination mount of a given type.
1969 * ---------------------------------------------------------------------------
1970 * | BIND MOUNT OPERATION |
1971 * |**************************************************************************
1972 * | source-->| shared | private | slave | unbindable |
1976 * |**************************************************************************
1977 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1979 * |non-shared| shared (+) | private | slave (*) | invalid |
1980 * ***************************************************************************
1981 * A bind operation clones the source mount and mounts the clone on the
1982 * destination mount.
1984 * (++) the cloned mount is propagated to all the mounts in the propagation
1985 * tree of the destination mount and the cloned mount is added to
1986 * the peer group of the source mount.
1987 * (+) the cloned mount is created under the destination mount and is marked
1988 * as shared. The cloned mount is added to the peer group of the source
1990 * (+++) the mount is propagated to all the mounts in the propagation tree
1991 * of the destination mount and the cloned mount is made slave
1992 * of the same master as that of the source mount. The cloned mount
1993 * is marked as 'shared and slave'.
1994 * (*) the cloned mount is made a slave of the same master as that of the
1997 * ---------------------------------------------------------------------------
1998 * | MOVE MOUNT OPERATION |
1999 * |**************************************************************************
2000 * | source-->| shared | private | slave | unbindable |
2004 * |**************************************************************************
2005 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2007 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2008 * ***************************************************************************
2010 * (+) the mount is moved to the destination. And is then propagated to
2011 * all the mounts in the propagation tree of the destination mount.
2012 * (+*) the mount is moved to the destination.
2013 * (+++) the mount is moved to the destination and is then propagated to
2014 * all the mounts belonging to the destination mount's propagation tree.
2015 * the mount is marked as 'shared and slave'.
2016 * (*) the mount continues to be a slave at the new location.
2018 * if the source mount is a tree, the operations explained above is
2019 * applied to each mount in the tree.
2020 * Must be called without spinlocks held, since this function can sleep
2023 static int attach_recursive_mnt(struct mount *source_mnt,
2024 struct mount *dest_mnt,
2025 struct mountpoint *dest_mp,
2026 struct path *parent_path)
2028 HLIST_HEAD(tree_list);
2029 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2030 struct mountpoint *smp;
2031 struct mount *child, *p;
2032 struct hlist_node *n;
2035 /* Preallocate a mountpoint in case the new mounts need
2036 * to be tucked under other mounts.
2038 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2040 return PTR_ERR(smp);
2042 /* Is there space to add these mounts to the mount namespace? */
2044 err = count_mounts(ns, source_mnt);
2049 if (IS_MNT_SHARED(dest_mnt)) {
2050 err = invent_group_ids(source_mnt, true);
2053 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2056 goto out_cleanup_ids;
2057 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2063 detach_mnt(source_mnt, parent_path);
2064 attach_mnt(source_mnt, dest_mnt, dest_mp);
2065 touch_mnt_namespace(source_mnt->mnt_ns);
2067 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2068 commit_tree(source_mnt);
2071 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2073 hlist_del_init(&child->mnt_hash);
2074 q = __lookup_mnt(&child->mnt_parent->mnt,
2075 child->mnt_mountpoint);
2077 mnt_change_mountpoint(child, smp, q);
2080 put_mountpoint(smp);
2081 unlock_mount_hash();
2086 while (!hlist_empty(&tree_list)) {
2087 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2088 child->mnt_parent->mnt_ns->pending_mounts = 0;
2089 umount_tree(child, UMOUNT_SYNC);
2091 unlock_mount_hash();
2092 cleanup_group_ids(source_mnt, NULL);
2094 ns->pending_mounts = 0;
2096 read_seqlock_excl(&mount_lock);
2097 put_mountpoint(smp);
2098 read_sequnlock_excl(&mount_lock);
2103 static struct mountpoint *lock_mount(struct path *path)
2105 struct vfsmount *mnt;
2106 struct dentry *dentry = path->dentry;
2108 inode_lock(dentry->d_inode);
2109 if (unlikely(cant_mount(dentry))) {
2110 inode_unlock(dentry->d_inode);
2111 return ERR_PTR(-ENOENT);
2114 mnt = lookup_mnt(path);
2116 struct mountpoint *mp = get_mountpoint(dentry);
2119 inode_unlock(dentry->d_inode);
2125 inode_unlock(path->dentry->d_inode);
2128 dentry = path->dentry = dget(mnt->mnt_root);
2132 static void unlock_mount(struct mountpoint *where)
2134 struct dentry *dentry = where->m_dentry;
2136 read_seqlock_excl(&mount_lock);
2137 put_mountpoint(where);
2138 read_sequnlock_excl(&mount_lock);
2141 inode_unlock(dentry->d_inode);
2144 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2146 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2149 if (d_is_dir(mp->m_dentry) !=
2150 d_is_dir(mnt->mnt.mnt_root))
2153 return attach_recursive_mnt(mnt, p, mp, NULL);
2157 * Sanity check the flags to change_mnt_propagation.
2160 static int flags_to_propagation_type(int ms_flags)
2162 int type = ms_flags & ~(MS_REC | MS_SILENT);
2164 /* Fail if any non-propagation flags are set */
2165 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2167 /* Only one propagation flag should be set */
2168 if (!is_power_of_2(type))
2174 * recursively change the type of the mountpoint.
2176 static int do_change_type(struct path *path, int ms_flags)
2179 struct mount *mnt = real_mount(path->mnt);
2180 int recurse = ms_flags & MS_REC;
2184 if (path->dentry != path->mnt->mnt_root)
2187 type = flags_to_propagation_type(ms_flags);
2192 if (type == MS_SHARED) {
2193 err = invent_group_ids(mnt, recurse);
2199 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2200 change_mnt_propagation(m, type);
2201 unlock_mount_hash();
2208 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2210 struct mount *child;
2211 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2212 if (!is_subdir(child->mnt_mountpoint, dentry))
2215 if (child->mnt.mnt_flags & MNT_LOCKED)
2222 * do loopback mount.
2224 static int do_loopback(struct path *path, const char *old_name,
2227 struct path old_path;
2228 struct mount *mnt = NULL, *old, *parent;
2229 struct mountpoint *mp;
2231 if (!old_name || !*old_name)
2233 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2238 if (mnt_ns_loop(old_path.dentry))
2241 mp = lock_mount(path);
2246 old = real_mount(old_path.mnt);
2247 parent = real_mount(path->mnt);
2250 if (IS_MNT_UNBINDABLE(old))
2253 if (!check_mnt(parent))
2256 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2259 if (!recurse && has_locked_children(old, old_path.dentry))
2263 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2265 mnt = clone_mnt(old, old_path.dentry, 0);
2272 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2274 err = graft_tree(mnt, parent, mp);
2277 umount_tree(mnt, UMOUNT_SYNC);
2278 unlock_mount_hash();
2283 path_put(&old_path);
2287 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2290 int readonly_request = 0;
2292 if (ms_flags & MS_RDONLY)
2293 readonly_request = 1;
2294 if (readonly_request == __mnt_is_readonly(mnt))
2297 if (readonly_request)
2298 error = mnt_make_readonly(real_mount(mnt));
2300 __mnt_unmake_readonly(real_mount(mnt));
2305 * change filesystem flags. dir should be a physical root of filesystem.
2306 * If you've mounted a non-root directory somewhere and want to do remount
2307 * on it - tough luck.
2309 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2310 int mnt_flags, void *data)
2313 struct super_block *sb = path->mnt->mnt_sb;
2314 struct mount *mnt = real_mount(path->mnt);
2316 if (!check_mnt(mnt))
2319 if (path->dentry != path->mnt->mnt_root)
2322 /* Don't allow changing of locked mnt flags.
2324 * No locks need to be held here while testing the various
2325 * MNT_LOCK flags because those flags can never be cleared
2326 * once they are set.
2328 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2329 !(mnt_flags & MNT_READONLY)) {
2332 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2333 !(mnt_flags & MNT_NODEV)) {
2336 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2337 !(mnt_flags & MNT_NOSUID)) {
2340 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2341 !(mnt_flags & MNT_NOEXEC)) {
2344 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2345 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2349 err = security_sb_remount(sb, data);
2353 down_write(&sb->s_umount);
2354 if (ms_flags & MS_BIND)
2355 err = change_mount_flags(path->mnt, ms_flags);
2356 else if (!capable(CAP_SYS_ADMIN))
2359 err = do_remount_sb(sb, sb_flags, data, 0);
2362 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2363 mnt->mnt.mnt_flags = mnt_flags;
2364 touch_mnt_namespace(mnt->mnt_ns);
2365 unlock_mount_hash();
2367 up_write(&sb->s_umount);
2371 static inline int tree_contains_unbindable(struct mount *mnt)
2374 for (p = mnt; p; p = next_mnt(p, mnt)) {
2375 if (IS_MNT_UNBINDABLE(p))
2381 static int do_move_mount(struct path *path, const char *old_name)
2383 struct path old_path, parent_path;
2386 struct mountpoint *mp;
2388 if (!old_name || !*old_name)
2390 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2394 mp = lock_mount(path);
2399 old = real_mount(old_path.mnt);
2400 p = real_mount(path->mnt);
2403 if (!check_mnt(p) || !check_mnt(old))
2406 if (old->mnt.mnt_flags & MNT_LOCKED)
2410 if (old_path.dentry != old_path.mnt->mnt_root)
2413 if (!mnt_has_parent(old))
2416 if (d_is_dir(path->dentry) !=
2417 d_is_dir(old_path.dentry))
2420 * Don't move a mount residing in a shared parent.
2422 if (IS_MNT_SHARED(old->mnt_parent))
2425 * Don't move a mount tree containing unbindable mounts to a destination
2426 * mount which is shared.
2428 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2431 for (; mnt_has_parent(p); p = p->mnt_parent)
2435 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2439 /* if the mount is moved, it should no longer be expire
2441 list_del_init(&old->mnt_expire);
2446 path_put(&parent_path);
2447 path_put(&old_path);
2451 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2454 const char *subtype = strchr(fstype, '.');
2463 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2465 if (!mnt->mnt_sb->s_subtype)
2471 return ERR_PTR(err);
2475 * add a mount into a namespace's mount tree
2477 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2479 struct mountpoint *mp;
2480 struct mount *parent;
2483 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2485 mp = lock_mount(path);
2489 parent = real_mount(path->mnt);
2491 if (unlikely(!check_mnt(parent))) {
2492 /* that's acceptable only for automounts done in private ns */
2493 if (!(mnt_flags & MNT_SHRINKABLE))
2495 /* ... and for those we'd better have mountpoint still alive */
2496 if (!parent->mnt_ns)
2500 /* Refuse the same filesystem on the same mount point */
2502 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2503 path->mnt->mnt_root == path->dentry)
2507 if (d_is_symlink(newmnt->mnt.mnt_root))
2510 newmnt->mnt.mnt_flags = mnt_flags;
2511 err = graft_tree(newmnt, parent, mp);
2518 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2521 * create a new mount for userspace and request it to be added into the
2524 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2525 int mnt_flags, const char *name, void *data)
2527 struct file_system_type *type;
2528 struct vfsmount *mnt;
2534 type = get_fs_type(fstype);
2538 mnt = vfs_kern_mount(type, sb_flags, name, data);
2539 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2540 !mnt->mnt_sb->s_subtype)
2541 mnt = fs_set_subtype(mnt, fstype);
2543 put_filesystem(type);
2545 return PTR_ERR(mnt);
2547 if (mount_too_revealing(mnt, &mnt_flags)) {
2552 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2558 int finish_automount(struct vfsmount *m, struct path *path)
2560 struct mount *mnt = real_mount(m);
2562 /* The new mount record should have at least 2 refs to prevent it being
2563 * expired before we get a chance to add it
2565 BUG_ON(mnt_get_count(mnt) < 2);
2567 if (m->mnt_sb == path->mnt->mnt_sb &&
2568 m->mnt_root == path->dentry) {
2573 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2577 /* remove m from any expiration list it may be on */
2578 if (!list_empty(&mnt->mnt_expire)) {
2580 list_del_init(&mnt->mnt_expire);
2589 * mnt_set_expiry - Put a mount on an expiration list
2590 * @mnt: The mount to list.
2591 * @expiry_list: The list to add the mount to.
2593 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2597 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2601 EXPORT_SYMBOL(mnt_set_expiry);
2604 * process a list of expirable mountpoints with the intent of discarding any
2605 * mountpoints that aren't in use and haven't been touched since last we came
2608 void mark_mounts_for_expiry(struct list_head *mounts)
2610 struct mount *mnt, *next;
2611 LIST_HEAD(graveyard);
2613 if (list_empty(mounts))
2619 /* extract from the expiration list every vfsmount that matches the
2620 * following criteria:
2621 * - only referenced by its parent vfsmount
2622 * - still marked for expiry (marked on the last call here; marks are
2623 * cleared by mntput())
2625 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2626 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2627 propagate_mount_busy(mnt, 1))
2629 list_move(&mnt->mnt_expire, &graveyard);
2631 while (!list_empty(&graveyard)) {
2632 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2633 touch_mnt_namespace(mnt->mnt_ns);
2634 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2636 unlock_mount_hash();
2640 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2643 * Ripoff of 'select_parent()'
2645 * search the list of submounts for a given mountpoint, and move any
2646 * shrinkable submounts to the 'graveyard' list.
2648 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2650 struct mount *this_parent = parent;
2651 struct list_head *next;
2655 next = this_parent->mnt_mounts.next;
2657 while (next != &this_parent->mnt_mounts) {
2658 struct list_head *tmp = next;
2659 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2662 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2665 * Descend a level if the d_mounts list is non-empty.
2667 if (!list_empty(&mnt->mnt_mounts)) {
2672 if (!propagate_mount_busy(mnt, 1)) {
2673 list_move_tail(&mnt->mnt_expire, graveyard);
2678 * All done at this level ... ascend and resume the search
2680 if (this_parent != parent) {
2681 next = this_parent->mnt_child.next;
2682 this_parent = this_parent->mnt_parent;
2689 * process a list of expirable mountpoints with the intent of discarding any
2690 * submounts of a specific parent mountpoint
2692 * mount_lock must be held for write
2694 static void shrink_submounts(struct mount *mnt)
2696 LIST_HEAD(graveyard);
2699 /* extract submounts of 'mountpoint' from the expiration list */
2700 while (select_submounts(mnt, &graveyard)) {
2701 while (!list_empty(&graveyard)) {
2702 m = list_first_entry(&graveyard, struct mount,
2704 touch_mnt_namespace(m->mnt_ns);
2705 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2711 * Some copy_from_user() implementations do not return the exact number of
2712 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2713 * Note that this function differs from copy_from_user() in that it will oops
2714 * on bad values of `to', rather than returning a short copy.
2716 static long exact_copy_from_user(void *to, const void __user * from,
2720 const char __user *f = from;
2723 if (!access_ok(VERIFY_READ, from, n))
2727 if (__get_user(c, f)) {
2738 void *copy_mount_options(const void __user * data)
2747 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2749 return ERR_PTR(-ENOMEM);
2751 /* We only care that *some* data at the address the user
2752 * gave us is valid. Just in case, we'll zero
2753 * the remainder of the page.
2755 /* copy_from_user cannot cross TASK_SIZE ! */
2756 size = TASK_SIZE - (unsigned long)data;
2757 if (size > PAGE_SIZE)
2760 i = size - exact_copy_from_user(copy, data, size);
2763 return ERR_PTR(-EFAULT);
2766 memset(copy + i, 0, PAGE_SIZE - i);
2770 char *copy_mount_string(const void __user *data)
2772 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2776 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2777 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2779 * data is a (void *) that can point to any structure up to
2780 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2781 * information (or be NULL).
2783 * Pre-0.97 versions of mount() didn't have a flags word.
2784 * When the flags word was introduced its top half was required
2785 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2786 * Therefore, if this magic number is present, it carries no information
2787 * and must be discarded.
2789 long do_mount(const char *dev_name, const char __user *dir_name,
2790 const char *type_page, unsigned long flags, void *data_page)
2793 unsigned int mnt_flags = 0, sb_flags;
2797 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2798 flags &= ~MS_MGC_MSK;
2800 /* Basic sanity checks */
2802 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2804 if (flags & MS_NOUSER)
2807 /* ... and get the mountpoint */
2808 retval = user_path(dir_name, &path);
2812 retval = security_sb_mount(dev_name, &path,
2813 type_page, flags, data_page);
2814 if (!retval && !may_mount())
2816 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
2821 /* Default to relatime unless overriden */
2822 if (!(flags & MS_NOATIME))
2823 mnt_flags |= MNT_RELATIME;
2825 /* Separate the per-mountpoint flags */
2826 if (flags & MS_NOSUID)
2827 mnt_flags |= MNT_NOSUID;
2828 if (flags & MS_NODEV)
2829 mnt_flags |= MNT_NODEV;
2830 if (flags & MS_NOEXEC)
2831 mnt_flags |= MNT_NOEXEC;
2832 if (flags & MS_NOATIME)
2833 mnt_flags |= MNT_NOATIME;
2834 if (flags & MS_NODIRATIME)
2835 mnt_flags |= MNT_NODIRATIME;
2836 if (flags & MS_STRICTATIME)
2837 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2838 if (flags & MS_RDONLY)
2839 mnt_flags |= MNT_READONLY;
2841 /* The default atime for remount is preservation */
2842 if ((flags & MS_REMOUNT) &&
2843 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2844 MS_STRICTATIME)) == 0)) {
2845 mnt_flags &= ~MNT_ATIME_MASK;
2846 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2849 sb_flags = flags & (SB_RDONLY |
2858 if (flags & MS_REMOUNT)
2859 retval = do_remount(&path, flags, sb_flags, mnt_flags,
2861 else if (flags & MS_BIND)
2862 retval = do_loopback(&path, dev_name, flags & MS_REC);
2863 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2864 retval = do_change_type(&path, flags);
2865 else if (flags & MS_MOVE)
2866 retval = do_move_mount(&path, dev_name);
2868 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2869 dev_name, data_page);
2875 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2877 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2880 static void dec_mnt_namespaces(struct ucounts *ucounts)
2882 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2885 static void free_mnt_ns(struct mnt_namespace *ns)
2887 ns_free_inum(&ns->ns);
2888 dec_mnt_namespaces(ns->ucounts);
2889 put_user_ns(ns->user_ns);
2894 * Assign a sequence number so we can detect when we attempt to bind
2895 * mount a reference to an older mount namespace into the current
2896 * mount namespace, preventing reference counting loops. A 64bit
2897 * number incrementing at 10Ghz will take 12,427 years to wrap which
2898 * is effectively never, so we can ignore the possibility.
2900 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2902 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2904 struct mnt_namespace *new_ns;
2905 struct ucounts *ucounts;
2908 ucounts = inc_mnt_namespaces(user_ns);
2910 return ERR_PTR(-ENOSPC);
2912 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2914 dec_mnt_namespaces(ucounts);
2915 return ERR_PTR(-ENOMEM);
2917 ret = ns_alloc_inum(&new_ns->ns);
2920 dec_mnt_namespaces(ucounts);
2921 return ERR_PTR(ret);
2923 new_ns->ns.ops = &mntns_operations;
2924 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2925 atomic_set(&new_ns->count, 1);
2926 new_ns->root = NULL;
2927 INIT_LIST_HEAD(&new_ns->list);
2928 init_waitqueue_head(&new_ns->poll);
2930 new_ns->user_ns = get_user_ns(user_ns);
2931 new_ns->ucounts = ucounts;
2933 new_ns->pending_mounts = 0;
2938 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2939 struct user_namespace *user_ns, struct fs_struct *new_fs)
2941 struct mnt_namespace *new_ns;
2942 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2943 struct mount *p, *q;
2950 if (likely(!(flags & CLONE_NEWNS))) {
2957 new_ns = alloc_mnt_ns(user_ns);
2962 /* First pass: copy the tree topology */
2963 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2964 if (user_ns != ns->user_ns)
2965 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2966 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2969 free_mnt_ns(new_ns);
2970 return ERR_CAST(new);
2973 list_add_tail(&new_ns->list, &new->mnt_list);
2976 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2977 * as belonging to new namespace. We have already acquired a private
2978 * fs_struct, so tsk->fs->lock is not needed.
2986 if (&p->mnt == new_fs->root.mnt) {
2987 new_fs->root.mnt = mntget(&q->mnt);
2990 if (&p->mnt == new_fs->pwd.mnt) {
2991 new_fs->pwd.mnt = mntget(&q->mnt);
2995 p = next_mnt(p, old);
2996 q = next_mnt(q, new);
2999 while (p->mnt.mnt_root != q->mnt.mnt_root)
3000 p = next_mnt(p, old);
3013 * create_mnt_ns - creates a private namespace and adds a root filesystem
3014 * @mnt: pointer to the new root filesystem mountpoint
3016 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
3018 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
3019 if (!IS_ERR(new_ns)) {
3020 struct mount *mnt = real_mount(m);
3021 mnt->mnt_ns = new_ns;
3024 list_add(&mnt->mnt_list, &new_ns->list);
3031 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3033 struct mnt_namespace *ns;
3034 struct super_block *s;
3038 ns = create_mnt_ns(mnt);
3040 return ERR_CAST(ns);
3042 err = vfs_path_lookup(mnt->mnt_root, mnt,
3043 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3048 return ERR_PTR(err);
3050 /* trade a vfsmount reference for active sb one */
3051 s = path.mnt->mnt_sb;
3052 atomic_inc(&s->s_active);
3054 /* lock the sucker */
3055 down_write(&s->s_umount);
3056 /* ... and return the root of (sub)tree on it */
3059 EXPORT_SYMBOL(mount_subtree);
3061 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3062 char __user *, type, unsigned long, flags, void __user *, data)
3069 kernel_type = copy_mount_string(type);
3070 ret = PTR_ERR(kernel_type);
3071 if (IS_ERR(kernel_type))
3074 kernel_dev = copy_mount_string(dev_name);
3075 ret = PTR_ERR(kernel_dev);
3076 if (IS_ERR(kernel_dev))
3079 options = copy_mount_options(data);
3080 ret = PTR_ERR(options);
3081 if (IS_ERR(options))
3084 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3096 * Return true if path is reachable from root
3098 * namespace_sem or mount_lock is held
3100 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3101 const struct path *root)
3103 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3104 dentry = mnt->mnt_mountpoint;
3105 mnt = mnt->mnt_parent;
3107 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3110 bool path_is_under(const struct path *path1, const struct path *path2)
3113 read_seqlock_excl(&mount_lock);
3114 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3115 read_sequnlock_excl(&mount_lock);
3118 EXPORT_SYMBOL(path_is_under);
3121 * pivot_root Semantics:
3122 * Moves the root file system of the current process to the directory put_old,
3123 * makes new_root as the new root file system of the current process, and sets
3124 * root/cwd of all processes which had them on the current root to new_root.
3127 * The new_root and put_old must be directories, and must not be on the
3128 * same file system as the current process root. The put_old must be
3129 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3130 * pointed to by put_old must yield the same directory as new_root. No other
3131 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3133 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3134 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3135 * in this situation.
3138 * - we don't move root/cwd if they are not at the root (reason: if something
3139 * cared enough to change them, it's probably wrong to force them elsewhere)
3140 * - it's okay to pick a root that isn't the root of a file system, e.g.
3141 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3142 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3145 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3146 const char __user *, put_old)
3148 struct path new, old, parent_path, root_parent, root;
3149 struct mount *new_mnt, *root_mnt, *old_mnt;
3150 struct mountpoint *old_mp, *root_mp;
3156 error = user_path_dir(new_root, &new);
3160 error = user_path_dir(put_old, &old);
3164 error = security_sb_pivotroot(&old, &new);
3168 get_fs_root(current->fs, &root);
3169 old_mp = lock_mount(&old);
3170 error = PTR_ERR(old_mp);
3175 new_mnt = real_mount(new.mnt);
3176 root_mnt = real_mount(root.mnt);
3177 old_mnt = real_mount(old.mnt);
3178 if (IS_MNT_SHARED(old_mnt) ||
3179 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3180 IS_MNT_SHARED(root_mnt->mnt_parent))
3182 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3184 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3187 if (d_unlinked(new.dentry))
3190 if (new_mnt == root_mnt || old_mnt == root_mnt)
3191 goto out4; /* loop, on the same file system */
3193 if (root.mnt->mnt_root != root.dentry)
3194 goto out4; /* not a mountpoint */
3195 if (!mnt_has_parent(root_mnt))
3196 goto out4; /* not attached */
3197 root_mp = root_mnt->mnt_mp;
3198 if (new.mnt->mnt_root != new.dentry)
3199 goto out4; /* not a mountpoint */
3200 if (!mnt_has_parent(new_mnt))
3201 goto out4; /* not attached */
3202 /* make sure we can reach put_old from new_root */
3203 if (!is_path_reachable(old_mnt, old.dentry, &new))
3205 /* make certain new is below the root */
3206 if (!is_path_reachable(new_mnt, new.dentry, &root))
3208 root_mp->m_count++; /* pin it so it won't go away */
3210 detach_mnt(new_mnt, &parent_path);
3211 detach_mnt(root_mnt, &root_parent);
3212 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3213 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3214 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3216 /* mount old root on put_old */
3217 attach_mnt(root_mnt, old_mnt, old_mp);
3218 /* mount new_root on / */
3219 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3220 touch_mnt_namespace(current->nsproxy->mnt_ns);
3221 /* A moved mount should not expire automatically */
3222 list_del_init(&new_mnt->mnt_expire);
3223 put_mountpoint(root_mp);
3224 unlock_mount_hash();
3225 chroot_fs_refs(&root, &new);
3228 unlock_mount(old_mp);
3230 path_put(&root_parent);
3231 path_put(&parent_path);
3243 static void __init init_mount_tree(void)
3245 struct vfsmount *mnt;
3246 struct mnt_namespace *ns;
3248 struct file_system_type *type;
3250 type = get_fs_type("rootfs");
3252 panic("Can't find rootfs type");
3253 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3254 put_filesystem(type);
3256 panic("Can't create rootfs");
3258 ns = create_mnt_ns(mnt);
3260 panic("Can't allocate initial namespace");
3262 init_task.nsproxy->mnt_ns = ns;
3266 root.dentry = mnt->mnt_root;
3267 mnt->mnt_flags |= MNT_LOCKED;
3269 set_fs_pwd(current->fs, &root);
3270 set_fs_root(current->fs, &root);
3273 void __init mnt_init(void)
3277 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3278 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3280 mount_hashtable = alloc_large_system_hash("Mount-cache",
3281 sizeof(struct hlist_head),
3284 &m_hash_shift, &m_hash_mask, 0, 0);
3285 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3286 sizeof(struct hlist_head),
3289 &mp_hash_shift, &mp_hash_mask, 0, 0);
3291 if (!mount_hashtable || !mountpoint_hashtable)
3292 panic("Failed to allocate mount hash table\n");
3298 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3300 fs_kobj = kobject_create_and_add("fs", NULL);
3302 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3307 void put_mnt_ns(struct mnt_namespace *ns)
3309 if (!atomic_dec_and_test(&ns->count))
3311 drop_collected_mounts(&ns->root->mnt);
3315 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3317 struct vfsmount *mnt;
3318 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data);
3321 * it is a longterm mount, don't release mnt until
3322 * we unmount before file sys is unregistered
3324 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3328 EXPORT_SYMBOL_GPL(kern_mount_data);
3330 void kern_unmount(struct vfsmount *mnt)
3332 /* release long term mount so mount point can be released */
3333 if (!IS_ERR_OR_NULL(mnt)) {
3334 real_mount(mnt)->mnt_ns = NULL;
3335 synchronize_rcu(); /* yecchhh... */
3339 EXPORT_SYMBOL(kern_unmount);
3341 bool our_mnt(struct vfsmount *mnt)
3343 return check_mnt(real_mount(mnt));
3346 bool current_chrooted(void)
3348 /* Does the current process have a non-standard root */
3349 struct path ns_root;
3350 struct path fs_root;
3353 /* Find the namespace root */
3354 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3355 ns_root.dentry = ns_root.mnt->mnt_root;
3357 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3360 get_fs_root(current->fs, &fs_root);
3362 chrooted = !path_equal(&fs_root, &ns_root);
3370 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3373 int new_flags = *new_mnt_flags;
3375 bool visible = false;
3377 down_read(&namespace_sem);
3378 list_for_each_entry(mnt, &ns->list, mnt_list) {
3379 struct mount *child;
3382 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3385 /* This mount is not fully visible if it's root directory
3386 * is not the root directory of the filesystem.
3388 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3391 /* A local view of the mount flags */
3392 mnt_flags = mnt->mnt.mnt_flags;
3394 /* Don't miss readonly hidden in the superblock flags */
3395 if (sb_rdonly(mnt->mnt.mnt_sb))
3396 mnt_flags |= MNT_LOCK_READONLY;
3398 /* Verify the mount flags are equal to or more permissive
3399 * than the proposed new mount.
3401 if ((mnt_flags & MNT_LOCK_READONLY) &&
3402 !(new_flags & MNT_READONLY))
3404 if ((mnt_flags & MNT_LOCK_ATIME) &&
3405 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3408 /* This mount is not fully visible if there are any
3409 * locked child mounts that cover anything except for
3410 * empty directories.
3412 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3413 struct inode *inode = child->mnt_mountpoint->d_inode;
3414 /* Only worry about locked mounts */
3415 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3417 /* Is the directory permanetly empty? */
3418 if (!is_empty_dir_inode(inode))
3421 /* Preserve the locked attributes */
3422 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3429 up_read(&namespace_sem);
3433 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3435 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3436 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3437 unsigned long s_iflags;
3439 if (ns->user_ns == &init_user_ns)
3442 /* Can this filesystem be too revealing? */
3443 s_iflags = mnt->mnt_sb->s_iflags;
3444 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3447 if ((s_iflags & required_iflags) != required_iflags) {
3448 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3453 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3456 bool mnt_may_suid(struct vfsmount *mnt)
3459 * Foreign mounts (accessed via fchdir or through /proc
3460 * symlinks) are always treated as if they are nosuid. This
3461 * prevents namespaces from trusting potentially unsafe
3462 * suid/sgid bits, file caps, or security labels that originate
3463 * in other namespaces.
3465 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3466 current_in_userns(mnt->mnt_sb->s_user_ns);
3469 static struct ns_common *mntns_get(struct task_struct *task)
3471 struct ns_common *ns = NULL;
3472 struct nsproxy *nsproxy;
3475 nsproxy = task->nsproxy;
3477 ns = &nsproxy->mnt_ns->ns;
3478 get_mnt_ns(to_mnt_ns(ns));
3485 static void mntns_put(struct ns_common *ns)
3487 put_mnt_ns(to_mnt_ns(ns));
3490 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3492 struct fs_struct *fs = current->fs;
3493 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3497 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3498 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3499 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3506 old_mnt_ns = nsproxy->mnt_ns;
3507 nsproxy->mnt_ns = mnt_ns;
3510 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3511 "/", LOOKUP_DOWN, &root);
3513 /* revert to old namespace */
3514 nsproxy->mnt_ns = old_mnt_ns;
3519 put_mnt_ns(old_mnt_ns);
3521 /* Update the pwd and root */
3522 set_fs_pwd(fs, &root);
3523 set_fs_root(fs, &root);
3529 static struct user_namespace *mntns_owner(struct ns_common *ns)
3531 return to_mnt_ns(ns)->user_ns;
3534 const struct proc_ns_operations mntns_operations = {
3536 .type = CLONE_NEWNS,
3539 .install = mntns_install,
3540 .owner = mntns_owner,