1 // SPDX-License-Identifier: GPL-2.0-only
5 * (C) Copyright Al Viro 2000, 2001
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/file.h>
24 #include <linux/uaccess.h>
25 #include <linux/proc_ns.h>
26 #include <linux/magic.h>
27 #include <linux/memblock.h>
28 #include <linux/task_work.h>
29 #include <linux/sched/task.h>
30 #include <uapi/linux/mount.h>
31 #include <linux/fs_context.h>
32 #include <linux/shmem_fs.h>
37 /* Maximum number of mounts in a mount namespace */
38 unsigned int sysctl_mount_max __read_mostly = 100000;
40 static unsigned int m_hash_mask __read_mostly;
41 static unsigned int m_hash_shift __read_mostly;
42 static unsigned int mp_hash_mask __read_mostly;
43 static unsigned int mp_hash_shift __read_mostly;
45 static __initdata unsigned long mhash_entries;
46 static int __init set_mhash_entries(char *str)
50 mhash_entries = simple_strtoul(str, &str, 0);
53 __setup("mhash_entries=", set_mhash_entries);
55 static __initdata unsigned long mphash_entries;
56 static int __init set_mphash_entries(char *str)
60 mphash_entries = simple_strtoul(str, &str, 0);
63 __setup("mphash_entries=", set_mphash_entries);
66 static DEFINE_IDA(mnt_id_ida);
67 static DEFINE_IDA(mnt_group_ida);
69 static struct hlist_head *mount_hashtable __read_mostly;
70 static struct hlist_head *mountpoint_hashtable __read_mostly;
71 static struct kmem_cache *mnt_cache __read_mostly;
72 static DECLARE_RWSEM(namespace_sem);
73 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
74 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
77 struct kobject *fs_kobj;
78 EXPORT_SYMBOL_GPL(fs_kobj);
81 * vfsmount lock may be taken for read to prevent changes to the
82 * vfsmount hash, ie. during mountpoint lookups or walking back
85 * It should be taken for write in all cases where the vfsmount
86 * tree or hash is modified or when a vfsmount structure is modified.
88 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
90 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
92 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
93 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
94 tmp = tmp + (tmp >> m_hash_shift);
95 return &mount_hashtable[tmp & m_hash_mask];
98 static inline struct hlist_head *mp_hash(struct dentry *dentry)
100 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
101 tmp = tmp + (tmp >> mp_hash_shift);
102 return &mountpoint_hashtable[tmp & mp_hash_mask];
105 static int mnt_alloc_id(struct mount *mnt)
107 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
115 static void mnt_free_id(struct mount *mnt)
117 ida_free(&mnt_id_ida, mnt->mnt_id);
121 * Allocate a new peer group ID
123 static int mnt_alloc_group_id(struct mount *mnt)
125 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
129 mnt->mnt_group_id = res;
134 * Release a peer group ID
136 void mnt_release_group_id(struct mount *mnt)
138 ida_free(&mnt_group_ida, mnt->mnt_group_id);
139 mnt->mnt_group_id = 0;
143 * vfsmount lock must be held for read
145 static inline void mnt_add_count(struct mount *mnt, int n)
148 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
157 * vfsmount lock must be held for write
159 unsigned int mnt_get_count(struct mount *mnt)
162 unsigned int count = 0;
165 for_each_possible_cpu(cpu) {
166 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
171 return mnt->mnt_count;
175 static struct mount *alloc_vfsmnt(const char *name)
177 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
181 err = mnt_alloc_id(mnt);
186 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
192 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
194 goto out_free_devname;
196 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
199 mnt->mnt_writers = 0;
202 INIT_HLIST_NODE(&mnt->mnt_hash);
203 INIT_LIST_HEAD(&mnt->mnt_child);
204 INIT_LIST_HEAD(&mnt->mnt_mounts);
205 INIT_LIST_HEAD(&mnt->mnt_list);
206 INIT_LIST_HEAD(&mnt->mnt_expire);
207 INIT_LIST_HEAD(&mnt->mnt_share);
208 INIT_LIST_HEAD(&mnt->mnt_slave_list);
209 INIT_LIST_HEAD(&mnt->mnt_slave);
210 INIT_HLIST_NODE(&mnt->mnt_mp_list);
211 INIT_LIST_HEAD(&mnt->mnt_umounting);
212 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
218 kfree_const(mnt->mnt_devname);
223 kmem_cache_free(mnt_cache, mnt);
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
246 bool __mnt_is_readonly(struct vfsmount *mnt)
248 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
250 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
252 static inline void mnt_inc_writers(struct mount *mnt)
255 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
261 static inline void mnt_dec_writers(struct mount *mnt)
264 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
270 static unsigned int mnt_get_writers(struct mount *mnt)
273 unsigned int count = 0;
276 for_each_possible_cpu(cpu) {
277 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
282 return mnt->mnt_writers;
286 static int mnt_is_readonly(struct vfsmount *mnt)
288 if (mnt->mnt_sb->s_readonly_remount)
290 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
292 return __mnt_is_readonly(mnt);
296 * Most r/o & frozen checks on a fs are for operations that take discrete
297 * amounts of time, like a write() or unlink(). We must keep track of when
298 * those operations start (for permission checks) and when they end, so that we
299 * can determine when writes are able to occur to a filesystem.
302 * __mnt_want_write - get write access to a mount without freeze protection
303 * @m: the mount on which to take a write
305 * This tells the low-level filesystem that a write is about to be performed to
306 * it, and makes sure that writes are allowed (mnt it read-write) before
307 * returning success. This operation does not protect against filesystem being
308 * frozen. When the write operation is finished, __mnt_drop_write() must be
309 * called. This is effectively a refcount.
311 int __mnt_want_write(struct vfsmount *m)
313 struct mount *mnt = real_mount(m);
317 mnt_inc_writers(mnt);
319 * The store to mnt_inc_writers must be visible before we pass
320 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
321 * incremented count after it has set MNT_WRITE_HOLD.
324 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
327 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
328 * be set to match its requirements. So we must not load that until
329 * MNT_WRITE_HOLD is cleared.
332 if (mnt_is_readonly(m)) {
333 mnt_dec_writers(mnt);
342 * mnt_want_write - get write access to a mount
343 * @m: the mount on which to take a write
345 * This tells the low-level filesystem that a write is about to be performed to
346 * it, and makes sure that writes are allowed (mount is read-write, filesystem
347 * is not frozen) before returning success. When the write operation is
348 * finished, mnt_drop_write() must be called. This is effectively a refcount.
350 int mnt_want_write(struct vfsmount *m)
354 sb_start_write(m->mnt_sb);
355 ret = __mnt_want_write(m);
357 sb_end_write(m->mnt_sb);
360 EXPORT_SYMBOL_GPL(mnt_want_write);
363 * mnt_clone_write - get write access to a mount
364 * @mnt: the mount on which to take a write
366 * This is effectively like mnt_want_write, except
367 * it must only be used to take an extra write reference
368 * on a mountpoint that we already know has a write reference
369 * on it. This allows some optimisation.
371 * After finished, mnt_drop_write must be called as usual to
372 * drop the reference.
374 int mnt_clone_write(struct vfsmount *mnt)
376 /* superblock may be r/o */
377 if (__mnt_is_readonly(mnt))
380 mnt_inc_writers(real_mount(mnt));
384 EXPORT_SYMBOL_GPL(mnt_clone_write);
387 * __mnt_want_write_file - get write access to a file's mount
388 * @file: the file who's mount on which to take a write
390 * This is like __mnt_want_write, but it takes a file and can
391 * do some optimisations if the file is open for write already
393 int __mnt_want_write_file(struct file *file)
395 if (!(file->f_mode & FMODE_WRITER))
396 return __mnt_want_write(file->f_path.mnt);
398 return mnt_clone_write(file->f_path.mnt);
402 * mnt_want_write_file - get write access to a file's mount
403 * @file: the file who's mount on which to take a write
405 * This is like mnt_want_write, but it takes a file and can
406 * do some optimisations if the file is open for write already
408 int mnt_want_write_file(struct file *file)
412 sb_start_write(file_inode(file)->i_sb);
413 ret = __mnt_want_write_file(file);
415 sb_end_write(file_inode(file)->i_sb);
418 EXPORT_SYMBOL_GPL(mnt_want_write_file);
421 * __mnt_drop_write - give up write access to a mount
422 * @mnt: the mount on which to give up write access
424 * Tells the low-level filesystem that we are done
425 * performing writes to it. Must be matched with
426 * __mnt_want_write() call above.
428 void __mnt_drop_write(struct vfsmount *mnt)
431 mnt_dec_writers(real_mount(mnt));
436 * mnt_drop_write - give up write access to a mount
437 * @mnt: the mount on which to give up write access
439 * Tells the low-level filesystem that we are done performing writes to it and
440 * also allows filesystem to be frozen again. Must be matched with
441 * mnt_want_write() call above.
443 void mnt_drop_write(struct vfsmount *mnt)
445 __mnt_drop_write(mnt);
446 sb_end_write(mnt->mnt_sb);
448 EXPORT_SYMBOL_GPL(mnt_drop_write);
450 void __mnt_drop_write_file(struct file *file)
452 __mnt_drop_write(file->f_path.mnt);
455 void mnt_drop_write_file(struct file *file)
457 __mnt_drop_write_file(file);
458 sb_end_write(file_inode(file)->i_sb);
460 EXPORT_SYMBOL(mnt_drop_write_file);
462 static int mnt_make_readonly(struct mount *mnt)
467 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
469 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
470 * should be visible before we do.
475 * With writers on hold, if this value is zero, then there are
476 * definitely no active writers (although held writers may subsequently
477 * increment the count, they'll have to wait, and decrement it after
478 * seeing MNT_READONLY).
480 * It is OK to have counter incremented on one CPU and decremented on
481 * another: the sum will add up correctly. The danger would be when we
482 * sum up each counter, if we read a counter before it is incremented,
483 * but then read another CPU's count which it has been subsequently
484 * decremented from -- we would see more decrements than we should.
485 * MNT_WRITE_HOLD protects against this scenario, because
486 * mnt_want_write first increments count, then smp_mb, then spins on
487 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
488 * we're counting up here.
490 if (mnt_get_writers(mnt) > 0)
493 mnt->mnt.mnt_flags |= MNT_READONLY;
495 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
496 * that become unheld will see MNT_READONLY.
499 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
504 static int __mnt_unmake_readonly(struct mount *mnt)
507 mnt->mnt.mnt_flags &= ~MNT_READONLY;
512 int sb_prepare_remount_readonly(struct super_block *sb)
517 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
518 if (atomic_long_read(&sb->s_remove_count))
522 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
523 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
524 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
526 if (mnt_get_writers(mnt) > 0) {
532 if (!err && atomic_long_read(&sb->s_remove_count))
536 sb->s_readonly_remount = 1;
539 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
540 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
541 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
548 static void free_vfsmnt(struct mount *mnt)
550 kfree_const(mnt->mnt_devname);
552 free_percpu(mnt->mnt_pcp);
554 kmem_cache_free(mnt_cache, mnt);
557 static void delayed_free_vfsmnt(struct rcu_head *head)
559 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
562 /* call under rcu_read_lock */
563 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
566 if (read_seqretry(&mount_lock, seq))
570 mnt = real_mount(bastard);
571 mnt_add_count(mnt, 1);
572 smp_mb(); // see mntput_no_expire()
573 if (likely(!read_seqretry(&mount_lock, seq)))
575 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
576 mnt_add_count(mnt, -1);
580 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
581 mnt_add_count(mnt, -1);
586 /* caller will mntput() */
590 /* call under rcu_read_lock */
591 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
593 int res = __legitimize_mnt(bastard, seq);
596 if (unlikely(res < 0)) {
605 * find the first mount at @dentry on vfsmount @mnt.
606 * call under rcu_read_lock()
608 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
610 struct hlist_head *head = m_hash(mnt, dentry);
613 hlist_for_each_entry_rcu(p, head, mnt_hash)
614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
620 * lookup_mnt - Return the first child mount mounted at path
622 * "First" means first mounted chronologically. If you create the
625 * mount /dev/sda1 /mnt
626 * mount /dev/sda2 /mnt
627 * mount /dev/sda3 /mnt
629 * Then lookup_mnt() on the base /mnt dentry in the root mount will
630 * return successively the root dentry and vfsmount of /dev/sda1, then
631 * /dev/sda2, then /dev/sda3, then NULL.
633 * lookup_mnt takes a reference to the found vfsmount.
635 struct vfsmount *lookup_mnt(const struct path *path)
637 struct mount *child_mnt;
643 seq = read_seqbegin(&mount_lock);
644 child_mnt = __lookup_mnt(path->mnt, path->dentry);
645 m = child_mnt ? &child_mnt->mnt : NULL;
646 } while (!legitimize_mnt(m, seq));
652 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
653 * current mount namespace.
655 * The common case is dentries are not mountpoints at all and that
656 * test is handled inline. For the slow case when we are actually
657 * dealing with a mountpoint of some kind, walk through all of the
658 * mounts in the current mount namespace and test to see if the dentry
661 * The mount_hashtable is not usable in the context because we
662 * need to identify all mounts that may be in the current mount
663 * namespace not just a mount that happens to have some specified
666 bool __is_local_mountpoint(struct dentry *dentry)
668 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
670 bool is_covered = false;
672 down_read(&namespace_sem);
673 list_for_each_entry(mnt, &ns->list, mnt_list) {
674 is_covered = (mnt->mnt_mountpoint == dentry);
678 up_read(&namespace_sem);
683 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
685 struct hlist_head *chain = mp_hash(dentry);
686 struct mountpoint *mp;
688 hlist_for_each_entry(mp, chain, m_hash) {
689 if (mp->m_dentry == dentry) {
697 static struct mountpoint *get_mountpoint(struct dentry *dentry)
699 struct mountpoint *mp, *new = NULL;
702 if (d_mountpoint(dentry)) {
703 /* might be worth a WARN_ON() */
704 if (d_unlinked(dentry))
705 return ERR_PTR(-ENOENT);
707 read_seqlock_excl(&mount_lock);
708 mp = lookup_mountpoint(dentry);
709 read_sequnlock_excl(&mount_lock);
715 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
717 return ERR_PTR(-ENOMEM);
720 /* Exactly one processes may set d_mounted */
721 ret = d_set_mounted(dentry);
723 /* Someone else set d_mounted? */
727 /* The dentry is not available as a mountpoint? */
732 /* Add the new mountpoint to the hash table */
733 read_seqlock_excl(&mount_lock);
734 new->m_dentry = dget(dentry);
736 hlist_add_head(&new->m_hash, mp_hash(dentry));
737 INIT_HLIST_HEAD(&new->m_list);
738 read_sequnlock_excl(&mount_lock);
748 * vfsmount lock must be held. Additionally, the caller is responsible
749 * for serializing calls for given disposal list.
751 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
753 if (!--mp->m_count) {
754 struct dentry *dentry = mp->m_dentry;
755 BUG_ON(!hlist_empty(&mp->m_list));
756 spin_lock(&dentry->d_lock);
757 dentry->d_flags &= ~DCACHE_MOUNTED;
758 spin_unlock(&dentry->d_lock);
759 dput_to_list(dentry, list);
760 hlist_del(&mp->m_hash);
765 /* called with namespace_lock and vfsmount lock */
766 static void put_mountpoint(struct mountpoint *mp)
768 __put_mountpoint(mp, &ex_mountpoints);
771 static inline int check_mnt(struct mount *mnt)
773 return mnt->mnt_ns == current->nsproxy->mnt_ns;
777 * vfsmount lock must be held for write
779 static void touch_mnt_namespace(struct mnt_namespace *ns)
783 wake_up_interruptible(&ns->poll);
788 * vfsmount lock must be held for write
790 static void __touch_mnt_namespace(struct mnt_namespace *ns)
792 if (ns && ns->event != event) {
794 wake_up_interruptible(&ns->poll);
799 * vfsmount lock must be held for write
801 static struct mountpoint *unhash_mnt(struct mount *mnt)
803 struct mountpoint *mp;
804 mnt->mnt_parent = mnt;
805 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
806 list_del_init(&mnt->mnt_child);
807 hlist_del_init_rcu(&mnt->mnt_hash);
808 hlist_del_init(&mnt->mnt_mp_list);
815 * vfsmount lock must be held for write
817 static void umount_mnt(struct mount *mnt)
819 put_mountpoint(unhash_mnt(mnt));
823 * vfsmount lock must be held for write
825 void mnt_set_mountpoint(struct mount *mnt,
826 struct mountpoint *mp,
827 struct mount *child_mnt)
830 mnt_add_count(mnt, 1); /* essentially, that's mntget */
831 child_mnt->mnt_mountpoint = mp->m_dentry;
832 child_mnt->mnt_parent = mnt;
833 child_mnt->mnt_mp = mp;
834 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
837 static void __attach_mnt(struct mount *mnt, struct mount *parent)
839 hlist_add_head_rcu(&mnt->mnt_hash,
840 m_hash(&parent->mnt, mnt->mnt_mountpoint));
841 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
845 * vfsmount lock must be held for write
847 static void attach_mnt(struct mount *mnt,
848 struct mount *parent,
849 struct mountpoint *mp)
851 mnt_set_mountpoint(parent, mp, mnt);
852 __attach_mnt(mnt, parent);
855 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
857 struct mountpoint *old_mp = mnt->mnt_mp;
858 struct mount *old_parent = mnt->mnt_parent;
860 list_del_init(&mnt->mnt_child);
861 hlist_del_init(&mnt->mnt_mp_list);
862 hlist_del_init_rcu(&mnt->mnt_hash);
864 attach_mnt(mnt, parent, mp);
866 put_mountpoint(old_mp);
867 mnt_add_count(old_parent, -1);
871 * vfsmount lock must be held for write
873 static void commit_tree(struct mount *mnt)
875 struct mount *parent = mnt->mnt_parent;
878 struct mnt_namespace *n = parent->mnt_ns;
880 BUG_ON(parent == mnt);
882 list_add_tail(&head, &mnt->mnt_list);
883 list_for_each_entry(m, &head, mnt_list)
886 list_splice(&head, n->list.prev);
888 n->mounts += n->pending_mounts;
889 n->pending_mounts = 0;
891 __attach_mnt(mnt, parent);
892 touch_mnt_namespace(n);
895 static struct mount *next_mnt(struct mount *p, struct mount *root)
897 struct list_head *next = p->mnt_mounts.next;
898 if (next == &p->mnt_mounts) {
902 next = p->mnt_child.next;
903 if (next != &p->mnt_parent->mnt_mounts)
908 return list_entry(next, struct mount, mnt_child);
911 static struct mount *skip_mnt_tree(struct mount *p)
913 struct list_head *prev = p->mnt_mounts.prev;
914 while (prev != &p->mnt_mounts) {
915 p = list_entry(prev, struct mount, mnt_child);
916 prev = p->mnt_mounts.prev;
922 * vfs_create_mount - Create a mount for a configured superblock
923 * @fc: The configuration context with the superblock attached
925 * Create a mount to an already configured superblock. If necessary, the
926 * caller should invoke vfs_get_tree() before calling this.
928 * Note that this does not attach the mount to anything.
930 struct vfsmount *vfs_create_mount(struct fs_context *fc)
935 return ERR_PTR(-EINVAL);
937 mnt = alloc_vfsmnt(fc->source ?: "none");
939 return ERR_PTR(-ENOMEM);
941 if (fc->sb_flags & SB_KERNMOUNT)
942 mnt->mnt.mnt_flags = MNT_INTERNAL;
944 atomic_inc(&fc->root->d_sb->s_active);
945 mnt->mnt.mnt_sb = fc->root->d_sb;
946 mnt->mnt.mnt_root = dget(fc->root);
947 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
948 mnt->mnt_parent = mnt;
951 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
955 EXPORT_SYMBOL(vfs_create_mount);
957 struct vfsmount *fc_mount(struct fs_context *fc)
959 int err = vfs_get_tree(fc);
961 up_write(&fc->root->d_sb->s_umount);
962 return vfs_create_mount(fc);
966 EXPORT_SYMBOL(fc_mount);
968 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
969 int flags, const char *name,
972 struct fs_context *fc;
973 struct vfsmount *mnt;
977 return ERR_PTR(-EINVAL);
979 fc = fs_context_for_mount(type, flags);
984 ret = vfs_parse_fs_string(fc, "source",
987 ret = parse_monolithic_mount_data(fc, data);
996 EXPORT_SYMBOL_GPL(vfs_kern_mount);
999 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1000 const char *name, void *data)
1002 /* Until it is worked out how to pass the user namespace
1003 * through from the parent mount to the submount don't support
1004 * unprivileged mounts with submounts.
1006 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1007 return ERR_PTR(-EPERM);
1009 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1011 EXPORT_SYMBOL_GPL(vfs_submount);
1013 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1016 struct super_block *sb = old->mnt.mnt_sb;
1020 mnt = alloc_vfsmnt(old->mnt_devname);
1022 return ERR_PTR(-ENOMEM);
1024 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1025 mnt->mnt_group_id = 0; /* not a peer of original */
1027 mnt->mnt_group_id = old->mnt_group_id;
1029 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1030 err = mnt_alloc_group_id(mnt);
1035 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1036 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1038 atomic_inc(&sb->s_active);
1039 mnt->mnt.mnt_sb = sb;
1040 mnt->mnt.mnt_root = dget(root);
1041 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1042 mnt->mnt_parent = mnt;
1044 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1045 unlock_mount_hash();
1047 if ((flag & CL_SLAVE) ||
1048 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1049 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1050 mnt->mnt_master = old;
1051 CLEAR_MNT_SHARED(mnt);
1052 } else if (!(flag & CL_PRIVATE)) {
1053 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1054 list_add(&mnt->mnt_share, &old->mnt_share);
1055 if (IS_MNT_SLAVE(old))
1056 list_add(&mnt->mnt_slave, &old->mnt_slave);
1057 mnt->mnt_master = old->mnt_master;
1059 CLEAR_MNT_SHARED(mnt);
1061 if (flag & CL_MAKE_SHARED)
1062 set_mnt_shared(mnt);
1064 /* stick the duplicate mount on the same expiry list
1065 * as the original if that was on one */
1066 if (flag & CL_EXPIRE) {
1067 if (!list_empty(&old->mnt_expire))
1068 list_add(&mnt->mnt_expire, &old->mnt_expire);
1076 return ERR_PTR(err);
1079 static void cleanup_mnt(struct mount *mnt)
1081 struct hlist_node *p;
1084 * The warning here probably indicates that somebody messed
1085 * up a mnt_want/drop_write() pair. If this happens, the
1086 * filesystem was probably unable to make r/w->r/o transitions.
1087 * The locking used to deal with mnt_count decrement provides barriers,
1088 * so mnt_get_writers() below is safe.
1090 WARN_ON(mnt_get_writers(mnt));
1091 if (unlikely(mnt->mnt_pins.first))
1093 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1094 hlist_del(&m->mnt_umount);
1097 fsnotify_vfsmount_delete(&mnt->mnt);
1098 dput(mnt->mnt.mnt_root);
1099 deactivate_super(mnt->mnt.mnt_sb);
1101 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1104 static void __cleanup_mnt(struct rcu_head *head)
1106 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1109 static LLIST_HEAD(delayed_mntput_list);
1110 static void delayed_mntput(struct work_struct *unused)
1112 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1113 struct mount *m, *t;
1115 llist_for_each_entry_safe(m, t, node, mnt_llist)
1118 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1120 static void mntput_no_expire(struct mount *mnt)
1125 if (likely(READ_ONCE(mnt->mnt_ns))) {
1127 * Since we don't do lock_mount_hash() here,
1128 * ->mnt_ns can change under us. However, if it's
1129 * non-NULL, then there's a reference that won't
1130 * be dropped until after an RCU delay done after
1131 * turning ->mnt_ns NULL. So if we observe it
1132 * non-NULL under rcu_read_lock(), the reference
1133 * we are dropping is not the final one.
1135 mnt_add_count(mnt, -1);
1141 * make sure that if __legitimize_mnt() has not seen us grab
1142 * mount_lock, we'll see their refcount increment here.
1145 mnt_add_count(mnt, -1);
1146 if (mnt_get_count(mnt)) {
1148 unlock_mount_hash();
1151 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1153 unlock_mount_hash();
1156 mnt->mnt.mnt_flags |= MNT_DOOMED;
1159 list_del(&mnt->mnt_instance);
1161 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1162 struct mount *p, *tmp;
1163 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1164 __put_mountpoint(unhash_mnt(p), &list);
1165 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1168 unlock_mount_hash();
1169 shrink_dentry_list(&list);
1171 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1172 struct task_struct *task = current;
1173 if (likely(!(task->flags & PF_KTHREAD))) {
1174 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1175 if (!task_work_add(task, &mnt->mnt_rcu, true))
1178 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1179 schedule_delayed_work(&delayed_mntput_work, 1);
1185 void mntput(struct vfsmount *mnt)
1188 struct mount *m = real_mount(mnt);
1189 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1190 if (unlikely(m->mnt_expiry_mark))
1191 m->mnt_expiry_mark = 0;
1192 mntput_no_expire(m);
1195 EXPORT_SYMBOL(mntput);
1197 struct vfsmount *mntget(struct vfsmount *mnt)
1200 mnt_add_count(real_mount(mnt), 1);
1203 EXPORT_SYMBOL(mntget);
1205 /* path_is_mountpoint() - Check if path is a mount in the current
1208 * d_mountpoint() can only be used reliably to establish if a dentry is
1209 * not mounted in any namespace and that common case is handled inline.
1210 * d_mountpoint() isn't aware of the possibility there may be multiple
1211 * mounts using a given dentry in a different namespace. This function
1212 * checks if the passed in path is a mountpoint rather than the dentry
1215 bool path_is_mountpoint(const struct path *path)
1220 if (!d_mountpoint(path->dentry))
1225 seq = read_seqbegin(&mount_lock);
1226 res = __path_is_mountpoint(path);
1227 } while (read_seqretry(&mount_lock, seq));
1232 EXPORT_SYMBOL(path_is_mountpoint);
1234 struct vfsmount *mnt_clone_internal(const struct path *path)
1237 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1240 p->mnt.mnt_flags |= MNT_INTERNAL;
1244 #ifdef CONFIG_PROC_FS
1245 /* iterator; we want it to have access to namespace_sem, thus here... */
1246 static void *m_start(struct seq_file *m, loff_t *pos)
1248 struct proc_mounts *p = m->private;
1250 down_read(&namespace_sem);
1251 if (p->cached_event == p->ns->event) {
1252 void *v = p->cached_mount;
1253 if (*pos == p->cached_index)
1255 if (*pos == p->cached_index + 1) {
1256 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1257 return p->cached_mount = v;
1261 p->cached_event = p->ns->event;
1262 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1263 p->cached_index = *pos;
1264 return p->cached_mount;
1267 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1269 struct proc_mounts *p = m->private;
1271 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1272 p->cached_index = *pos;
1273 return p->cached_mount;
1276 static void m_stop(struct seq_file *m, void *v)
1278 up_read(&namespace_sem);
1281 static int m_show(struct seq_file *m, void *v)
1283 struct proc_mounts *p = m->private;
1284 struct mount *r = list_entry(v, struct mount, mnt_list);
1285 return p->show(m, &r->mnt);
1288 const struct seq_operations mounts_op = {
1294 #endif /* CONFIG_PROC_FS */
1297 * may_umount_tree - check if a mount tree is busy
1298 * @mnt: root of mount tree
1300 * This is called to check if a tree of mounts has any
1301 * open files, pwds, chroots or sub mounts that are
1304 int may_umount_tree(struct vfsmount *m)
1306 struct mount *mnt = real_mount(m);
1307 int actual_refs = 0;
1308 int minimum_refs = 0;
1312 /* write lock needed for mnt_get_count */
1314 for (p = mnt; p; p = next_mnt(p, mnt)) {
1315 actual_refs += mnt_get_count(p);
1318 unlock_mount_hash();
1320 if (actual_refs > minimum_refs)
1326 EXPORT_SYMBOL(may_umount_tree);
1329 * may_umount - check if a mount point is busy
1330 * @mnt: root of mount
1332 * This is called to check if a mount point has any
1333 * open files, pwds, chroots or sub mounts. If the
1334 * mount has sub mounts this will return busy
1335 * regardless of whether the sub mounts are busy.
1337 * Doesn't take quota and stuff into account. IOW, in some cases it will
1338 * give false negatives. The main reason why it's here is that we need
1339 * a non-destructive way to look for easily umountable filesystems.
1341 int may_umount(struct vfsmount *mnt)
1344 down_read(&namespace_sem);
1346 if (propagate_mount_busy(real_mount(mnt), 2))
1348 unlock_mount_hash();
1349 up_read(&namespace_sem);
1353 EXPORT_SYMBOL(may_umount);
1355 static void namespace_unlock(void)
1357 struct hlist_head head;
1358 struct hlist_node *p;
1362 hlist_move_list(&unmounted, &head);
1363 list_splice_init(&ex_mountpoints, &list);
1365 up_write(&namespace_sem);
1367 shrink_dentry_list(&list);
1369 if (likely(hlist_empty(&head)))
1372 synchronize_rcu_expedited();
1374 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1375 hlist_del(&m->mnt_umount);
1380 static inline void namespace_lock(void)
1382 down_write(&namespace_sem);
1385 enum umount_tree_flags {
1387 UMOUNT_PROPAGATE = 2,
1388 UMOUNT_CONNECTED = 4,
1391 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1393 /* Leaving mounts connected is only valid for lazy umounts */
1394 if (how & UMOUNT_SYNC)
1397 /* A mount without a parent has nothing to be connected to */
1398 if (!mnt_has_parent(mnt))
1401 /* Because the reference counting rules change when mounts are
1402 * unmounted and connected, umounted mounts may not be
1403 * connected to mounted mounts.
1405 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1408 /* Has it been requested that the mount remain connected? */
1409 if (how & UMOUNT_CONNECTED)
1412 /* Is the mount locked such that it needs to remain connected? */
1413 if (IS_MNT_LOCKED(mnt))
1416 /* By default disconnect the mount */
1421 * mount_lock must be held
1422 * namespace_sem must be held for write
1424 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1426 LIST_HEAD(tmp_list);
1429 if (how & UMOUNT_PROPAGATE)
1430 propagate_mount_unlock(mnt);
1432 /* Gather the mounts to umount */
1433 for (p = mnt; p; p = next_mnt(p, mnt)) {
1434 p->mnt.mnt_flags |= MNT_UMOUNT;
1435 list_move(&p->mnt_list, &tmp_list);
1438 /* Hide the mounts from mnt_mounts */
1439 list_for_each_entry(p, &tmp_list, mnt_list) {
1440 list_del_init(&p->mnt_child);
1443 /* Add propogated mounts to the tmp_list */
1444 if (how & UMOUNT_PROPAGATE)
1445 propagate_umount(&tmp_list);
1447 while (!list_empty(&tmp_list)) {
1448 struct mnt_namespace *ns;
1450 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1451 list_del_init(&p->mnt_expire);
1452 list_del_init(&p->mnt_list);
1456 __touch_mnt_namespace(ns);
1459 if (how & UMOUNT_SYNC)
1460 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1462 disconnect = disconnect_mount(p, how);
1463 if (mnt_has_parent(p)) {
1464 mnt_add_count(p->mnt_parent, -1);
1466 /* Don't forget about p */
1467 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1472 change_mnt_propagation(p, MS_PRIVATE);
1474 hlist_add_head(&p->mnt_umount, &unmounted);
1478 static void shrink_submounts(struct mount *mnt);
1480 static int do_umount_root(struct super_block *sb)
1484 down_write(&sb->s_umount);
1485 if (!sb_rdonly(sb)) {
1486 struct fs_context *fc;
1488 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1493 ret = parse_monolithic_mount_data(fc, NULL);
1495 ret = reconfigure_super(fc);
1499 up_write(&sb->s_umount);
1503 static int do_umount(struct mount *mnt, int flags)
1505 struct super_block *sb = mnt->mnt.mnt_sb;
1508 retval = security_sb_umount(&mnt->mnt, flags);
1513 * Allow userspace to request a mountpoint be expired rather than
1514 * unmounting unconditionally. Unmount only happens if:
1515 * (1) the mark is already set (the mark is cleared by mntput())
1516 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1518 if (flags & MNT_EXPIRE) {
1519 if (&mnt->mnt == current->fs->root.mnt ||
1520 flags & (MNT_FORCE | MNT_DETACH))
1524 * probably don't strictly need the lock here if we examined
1525 * all race cases, but it's a slowpath.
1528 if (mnt_get_count(mnt) != 2) {
1529 unlock_mount_hash();
1532 unlock_mount_hash();
1534 if (!xchg(&mnt->mnt_expiry_mark, 1))
1539 * If we may have to abort operations to get out of this
1540 * mount, and they will themselves hold resources we must
1541 * allow the fs to do things. In the Unix tradition of
1542 * 'Gee thats tricky lets do it in userspace' the umount_begin
1543 * might fail to complete on the first run through as other tasks
1544 * must return, and the like. Thats for the mount program to worry
1545 * about for the moment.
1548 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1549 sb->s_op->umount_begin(sb);
1553 * No sense to grab the lock for this test, but test itself looks
1554 * somewhat bogus. Suggestions for better replacement?
1555 * Ho-hum... In principle, we might treat that as umount + switch
1556 * to rootfs. GC would eventually take care of the old vfsmount.
1557 * Actually it makes sense, especially if rootfs would contain a
1558 * /reboot - static binary that would close all descriptors and
1559 * call reboot(9). Then init(8) could umount root and exec /reboot.
1561 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1563 * Special case for "unmounting" root ...
1564 * we just try to remount it readonly.
1566 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1568 return do_umount_root(sb);
1574 /* Recheck MNT_LOCKED with the locks held */
1576 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1580 if (flags & MNT_DETACH) {
1581 if (!list_empty(&mnt->mnt_list))
1582 umount_tree(mnt, UMOUNT_PROPAGATE);
1585 shrink_submounts(mnt);
1587 if (!propagate_mount_busy(mnt, 2)) {
1588 if (!list_empty(&mnt->mnt_list))
1589 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1594 unlock_mount_hash();
1600 * __detach_mounts - lazily unmount all mounts on the specified dentry
1602 * During unlink, rmdir, and d_drop it is possible to loose the path
1603 * to an existing mountpoint, and wind up leaking the mount.
1604 * detach_mounts allows lazily unmounting those mounts instead of
1607 * The caller may hold dentry->d_inode->i_mutex.
1609 void __detach_mounts(struct dentry *dentry)
1611 struct mountpoint *mp;
1616 mp = lookup_mountpoint(dentry);
1621 while (!hlist_empty(&mp->m_list)) {
1622 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1623 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1625 hlist_add_head(&mnt->mnt_umount, &unmounted);
1627 else umount_tree(mnt, UMOUNT_CONNECTED);
1631 unlock_mount_hash();
1636 * Is the caller allowed to modify his namespace?
1638 static inline bool may_mount(void)
1640 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1643 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1644 static inline bool may_mandlock(void)
1646 return capable(CAP_SYS_ADMIN);
1649 static inline bool may_mandlock(void)
1651 pr_warn("VFS: \"mand\" mount option not supported");
1657 * Now umount can handle mount points as well as block devices.
1658 * This is important for filesystems which use unnamed block devices.
1660 * We now support a flag for forced unmount like the other 'big iron'
1661 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1664 int ksys_umount(char __user *name, int flags)
1669 int lookup_flags = LOOKUP_MOUNTPOINT;
1671 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1677 if (!(flags & UMOUNT_NOFOLLOW))
1678 lookup_flags |= LOOKUP_FOLLOW;
1680 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1683 mnt = real_mount(path.mnt);
1685 if (path.dentry != path.mnt->mnt_root)
1687 if (!check_mnt(mnt))
1689 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1692 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1695 retval = do_umount(mnt, flags);
1697 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1699 mntput_no_expire(mnt);
1704 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1706 return ksys_umount(name, flags);
1709 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1712 * The 2.0 compatible umount. No flags.
1714 SYSCALL_DEFINE1(oldumount, char __user *, name)
1716 return ksys_umount(name, 0);
1721 static bool is_mnt_ns_file(struct dentry *dentry)
1723 /* Is this a proxy for a mount namespace? */
1724 return dentry->d_op == &ns_dentry_operations &&
1725 dentry->d_fsdata == &mntns_operations;
1728 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1730 return container_of(ns, struct mnt_namespace, ns);
1733 static bool mnt_ns_loop(struct dentry *dentry)
1735 /* Could bind mounting the mount namespace inode cause a
1736 * mount namespace loop?
1738 struct mnt_namespace *mnt_ns;
1739 if (!is_mnt_ns_file(dentry))
1742 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1743 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1746 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1749 struct mount *res, *p, *q, *r, *parent;
1751 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1752 return ERR_PTR(-EINVAL);
1754 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1755 return ERR_PTR(-EINVAL);
1757 res = q = clone_mnt(mnt, dentry, flag);
1761 q->mnt_mountpoint = mnt->mnt_mountpoint;
1764 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1766 if (!is_subdir(r->mnt_mountpoint, dentry))
1769 for (s = r; s; s = next_mnt(s, r)) {
1770 if (!(flag & CL_COPY_UNBINDABLE) &&
1771 IS_MNT_UNBINDABLE(s)) {
1772 if (s->mnt.mnt_flags & MNT_LOCKED) {
1773 /* Both unbindable and locked. */
1774 q = ERR_PTR(-EPERM);
1777 s = skip_mnt_tree(s);
1781 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1782 is_mnt_ns_file(s->mnt.mnt_root)) {
1783 s = skip_mnt_tree(s);
1786 while (p != s->mnt_parent) {
1792 q = clone_mnt(p, p->mnt.mnt_root, flag);
1796 list_add_tail(&q->mnt_list, &res->mnt_list);
1797 attach_mnt(q, parent, p->mnt_mp);
1798 unlock_mount_hash();
1805 umount_tree(res, UMOUNT_SYNC);
1806 unlock_mount_hash();
1811 /* Caller should check returned pointer for errors */
1813 struct vfsmount *collect_mounts(const struct path *path)
1817 if (!check_mnt(real_mount(path->mnt)))
1818 tree = ERR_PTR(-EINVAL);
1820 tree = copy_tree(real_mount(path->mnt), path->dentry,
1821 CL_COPY_ALL | CL_PRIVATE);
1824 return ERR_CAST(tree);
1828 static void free_mnt_ns(struct mnt_namespace *);
1829 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1831 void dissolve_on_fput(struct vfsmount *mnt)
1833 struct mnt_namespace *ns;
1836 ns = real_mount(mnt)->mnt_ns;
1839 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1843 unlock_mount_hash();
1849 void drop_collected_mounts(struct vfsmount *mnt)
1853 umount_tree(real_mount(mnt), 0);
1854 unlock_mount_hash();
1859 * clone_private_mount - create a private clone of a path
1861 * This creates a new vfsmount, which will be the clone of @path. The new will
1862 * not be attached anywhere in the namespace and will be private (i.e. changes
1863 * to the originating mount won't be propagated into this).
1865 * Release with mntput().
1867 struct vfsmount *clone_private_mount(const struct path *path)
1869 struct mount *old_mnt = real_mount(path->mnt);
1870 struct mount *new_mnt;
1872 if (IS_MNT_UNBINDABLE(old_mnt))
1873 return ERR_PTR(-EINVAL);
1875 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1876 if (IS_ERR(new_mnt))
1877 return ERR_CAST(new_mnt);
1879 return &new_mnt->mnt;
1881 EXPORT_SYMBOL_GPL(clone_private_mount);
1883 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1884 struct vfsmount *root)
1887 int res = f(root, arg);
1890 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1891 res = f(&mnt->mnt, arg);
1898 static void lock_mnt_tree(struct mount *mnt)
1902 for (p = mnt; p; p = next_mnt(p, mnt)) {
1903 int flags = p->mnt.mnt_flags;
1904 /* Don't allow unprivileged users to change mount flags */
1905 flags |= MNT_LOCK_ATIME;
1907 if (flags & MNT_READONLY)
1908 flags |= MNT_LOCK_READONLY;
1910 if (flags & MNT_NODEV)
1911 flags |= MNT_LOCK_NODEV;
1913 if (flags & MNT_NOSUID)
1914 flags |= MNT_LOCK_NOSUID;
1916 if (flags & MNT_NOEXEC)
1917 flags |= MNT_LOCK_NOEXEC;
1918 /* Don't allow unprivileged users to reveal what is under a mount */
1919 if (list_empty(&p->mnt_expire))
1920 flags |= MNT_LOCKED;
1921 p->mnt.mnt_flags = flags;
1925 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1929 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1930 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1931 mnt_release_group_id(p);
1935 static int invent_group_ids(struct mount *mnt, bool recurse)
1939 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1940 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1941 int err = mnt_alloc_group_id(p);
1943 cleanup_group_ids(mnt, p);
1952 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1954 unsigned int max = READ_ONCE(sysctl_mount_max);
1955 unsigned int mounts = 0, old, pending, sum;
1958 for (p = mnt; p; p = next_mnt(p, mnt))
1962 pending = ns->pending_mounts;
1963 sum = old + pending;
1967 (mounts > (max - sum)))
1970 ns->pending_mounts = pending + mounts;
1975 * @source_mnt : mount tree to be attached
1976 * @nd : place the mount tree @source_mnt is attached
1977 * @parent_nd : if non-null, detach the source_mnt from its parent and
1978 * store the parent mount and mountpoint dentry.
1979 * (done when source_mnt is moved)
1981 * NOTE: in the table below explains the semantics when a source mount
1982 * of a given type is attached to a destination mount of a given type.
1983 * ---------------------------------------------------------------------------
1984 * | BIND MOUNT OPERATION |
1985 * |**************************************************************************
1986 * | source-->| shared | private | slave | unbindable |
1990 * |**************************************************************************
1991 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1993 * |non-shared| shared (+) | private | slave (*) | invalid |
1994 * ***************************************************************************
1995 * A bind operation clones the source mount and mounts the clone on the
1996 * destination mount.
1998 * (++) the cloned mount is propagated to all the mounts in the propagation
1999 * tree of the destination mount and the cloned mount is added to
2000 * the peer group of the source mount.
2001 * (+) the cloned mount is created under the destination mount and is marked
2002 * as shared. The cloned mount is added to the peer group of the source
2004 * (+++) the mount is propagated to all the mounts in the propagation tree
2005 * of the destination mount and the cloned mount is made slave
2006 * of the same master as that of the source mount. The cloned mount
2007 * is marked as 'shared and slave'.
2008 * (*) the cloned mount is made a slave of the same master as that of the
2011 * ---------------------------------------------------------------------------
2012 * | MOVE MOUNT OPERATION |
2013 * |**************************************************************************
2014 * | source-->| shared | private | slave | unbindable |
2018 * |**************************************************************************
2019 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2021 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2022 * ***************************************************************************
2024 * (+) the mount is moved to the destination. And is then propagated to
2025 * all the mounts in the propagation tree of the destination mount.
2026 * (+*) the mount is moved to the destination.
2027 * (+++) the mount is moved to the destination and is then propagated to
2028 * all the mounts belonging to the destination mount's propagation tree.
2029 * the mount is marked as 'shared and slave'.
2030 * (*) the mount continues to be a slave at the new location.
2032 * if the source mount is a tree, the operations explained above is
2033 * applied to each mount in the tree.
2034 * Must be called without spinlocks held, since this function can sleep
2037 static int attach_recursive_mnt(struct mount *source_mnt,
2038 struct mount *dest_mnt,
2039 struct mountpoint *dest_mp,
2042 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2043 HLIST_HEAD(tree_list);
2044 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2045 struct mountpoint *smp;
2046 struct mount *child, *p;
2047 struct hlist_node *n;
2050 /* Preallocate a mountpoint in case the new mounts need
2051 * to be tucked under other mounts.
2053 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2055 return PTR_ERR(smp);
2057 /* Is there space to add these mounts to the mount namespace? */
2059 err = count_mounts(ns, source_mnt);
2064 if (IS_MNT_SHARED(dest_mnt)) {
2065 err = invent_group_ids(source_mnt, true);
2068 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2071 goto out_cleanup_ids;
2072 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2078 unhash_mnt(source_mnt);
2079 attach_mnt(source_mnt, dest_mnt, dest_mp);
2080 touch_mnt_namespace(source_mnt->mnt_ns);
2082 if (source_mnt->mnt_ns) {
2083 /* move from anon - the caller will destroy */
2084 list_del_init(&source_mnt->mnt_ns->list);
2086 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2087 commit_tree(source_mnt);
2090 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2092 hlist_del_init(&child->mnt_hash);
2093 q = __lookup_mnt(&child->mnt_parent->mnt,
2094 child->mnt_mountpoint);
2096 mnt_change_mountpoint(child, smp, q);
2097 /* Notice when we are propagating across user namespaces */
2098 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2099 lock_mnt_tree(child);
2100 child->mnt.mnt_flags &= ~MNT_LOCKED;
2103 put_mountpoint(smp);
2104 unlock_mount_hash();
2109 while (!hlist_empty(&tree_list)) {
2110 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2111 child->mnt_parent->mnt_ns->pending_mounts = 0;
2112 umount_tree(child, UMOUNT_SYNC);
2114 unlock_mount_hash();
2115 cleanup_group_ids(source_mnt, NULL);
2117 ns->pending_mounts = 0;
2119 read_seqlock_excl(&mount_lock);
2120 put_mountpoint(smp);
2121 read_sequnlock_excl(&mount_lock);
2126 static struct mountpoint *lock_mount(struct path *path)
2128 struct vfsmount *mnt;
2129 struct dentry *dentry = path->dentry;
2131 inode_lock(dentry->d_inode);
2132 if (unlikely(cant_mount(dentry))) {
2133 inode_unlock(dentry->d_inode);
2134 return ERR_PTR(-ENOENT);
2137 mnt = lookup_mnt(path);
2139 struct mountpoint *mp = get_mountpoint(dentry);
2142 inode_unlock(dentry->d_inode);
2148 inode_unlock(path->dentry->d_inode);
2151 dentry = path->dentry = dget(mnt->mnt_root);
2155 static void unlock_mount(struct mountpoint *where)
2157 struct dentry *dentry = where->m_dentry;
2159 read_seqlock_excl(&mount_lock);
2160 put_mountpoint(where);
2161 read_sequnlock_excl(&mount_lock);
2164 inode_unlock(dentry->d_inode);
2167 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2169 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2172 if (d_is_dir(mp->m_dentry) !=
2173 d_is_dir(mnt->mnt.mnt_root))
2176 return attach_recursive_mnt(mnt, p, mp, false);
2180 * Sanity check the flags to change_mnt_propagation.
2183 static int flags_to_propagation_type(int ms_flags)
2185 int type = ms_flags & ~(MS_REC | MS_SILENT);
2187 /* Fail if any non-propagation flags are set */
2188 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2190 /* Only one propagation flag should be set */
2191 if (!is_power_of_2(type))
2197 * recursively change the type of the mountpoint.
2199 static int do_change_type(struct path *path, int ms_flags)
2202 struct mount *mnt = real_mount(path->mnt);
2203 int recurse = ms_flags & MS_REC;
2207 if (path->dentry != path->mnt->mnt_root)
2210 type = flags_to_propagation_type(ms_flags);
2215 if (type == MS_SHARED) {
2216 err = invent_group_ids(mnt, recurse);
2222 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2223 change_mnt_propagation(m, type);
2224 unlock_mount_hash();
2231 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2233 struct mount *child;
2234 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2235 if (!is_subdir(child->mnt_mountpoint, dentry))
2238 if (child->mnt.mnt_flags & MNT_LOCKED)
2244 static struct mount *__do_loopback(struct path *old_path, int recurse)
2246 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2248 if (IS_MNT_UNBINDABLE(old))
2251 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2254 if (!recurse && has_locked_children(old, old_path->dentry))
2258 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2260 mnt = clone_mnt(old, old_path->dentry, 0);
2263 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2269 * do loopback mount.
2271 static int do_loopback(struct path *path, const char *old_name,
2274 struct path old_path;
2275 struct mount *mnt = NULL, *parent;
2276 struct mountpoint *mp;
2278 if (!old_name || !*old_name)
2280 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2285 if (mnt_ns_loop(old_path.dentry))
2288 mp = lock_mount(path);
2294 parent = real_mount(path->mnt);
2295 if (!check_mnt(parent))
2298 mnt = __do_loopback(&old_path, recurse);
2304 err = graft_tree(mnt, parent, mp);
2307 umount_tree(mnt, UMOUNT_SYNC);
2308 unlock_mount_hash();
2313 path_put(&old_path);
2317 static struct file *open_detached_copy(struct path *path, bool recursive)
2319 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2320 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2321 struct mount *mnt, *p;
2325 return ERR_CAST(ns);
2328 mnt = __do_loopback(path, recursive);
2332 return ERR_CAST(mnt);
2336 for (p = mnt; p; p = next_mnt(p, mnt)) {
2341 list_add_tail(&ns->list, &mnt->mnt_list);
2343 unlock_mount_hash();
2347 path->mnt = &mnt->mnt;
2348 file = dentry_open(path, O_PATH, current_cred());
2350 dissolve_on_fput(path->mnt);
2352 file->f_mode |= FMODE_NEED_UNMOUNT;
2356 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2360 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2361 bool detached = flags & OPEN_TREE_CLONE;
2365 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2367 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2368 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2372 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2375 if (flags & AT_NO_AUTOMOUNT)
2376 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2377 if (flags & AT_SYMLINK_NOFOLLOW)
2378 lookup_flags &= ~LOOKUP_FOLLOW;
2379 if (flags & AT_EMPTY_PATH)
2380 lookup_flags |= LOOKUP_EMPTY;
2382 if (detached && !may_mount())
2385 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2389 error = user_path_at(dfd, filename, lookup_flags, &path);
2390 if (unlikely(error)) {
2391 file = ERR_PTR(error);
2394 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2396 file = dentry_open(&path, O_PATH, current_cred());
2401 return PTR_ERR(file);
2403 fd_install(fd, file);
2408 * Don't allow locked mount flags to be cleared.
2410 * No locks need to be held here while testing the various MNT_LOCK
2411 * flags because those flags can never be cleared once they are set.
2413 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2415 unsigned int fl = mnt->mnt.mnt_flags;
2417 if ((fl & MNT_LOCK_READONLY) &&
2418 !(mnt_flags & MNT_READONLY))
2421 if ((fl & MNT_LOCK_NODEV) &&
2422 !(mnt_flags & MNT_NODEV))
2425 if ((fl & MNT_LOCK_NOSUID) &&
2426 !(mnt_flags & MNT_NOSUID))
2429 if ((fl & MNT_LOCK_NOEXEC) &&
2430 !(mnt_flags & MNT_NOEXEC))
2433 if ((fl & MNT_LOCK_ATIME) &&
2434 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2440 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2442 bool readonly_request = (mnt_flags & MNT_READONLY);
2444 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2447 if (readonly_request)
2448 return mnt_make_readonly(mnt);
2450 return __mnt_unmake_readonly(mnt);
2454 * Update the user-settable attributes on a mount. The caller must hold
2455 * sb->s_umount for writing.
2457 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2460 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2461 mnt->mnt.mnt_flags = mnt_flags;
2462 touch_mnt_namespace(mnt->mnt_ns);
2463 unlock_mount_hash();
2466 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2468 struct super_block *sb = mnt->mnt_sb;
2470 if (!__mnt_is_readonly(mnt) &&
2471 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2472 char *buf = (char *)__get_free_page(GFP_KERNEL);
2473 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2476 time64_to_tm(sb->s_time_max, 0, &tm);
2478 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2480 is_mounted(mnt) ? "remounted" : "mounted",
2482 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2484 free_page((unsigned long)buf);
2489 * Handle reconfiguration of the mountpoint only without alteration of the
2490 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2493 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2495 struct super_block *sb = path->mnt->mnt_sb;
2496 struct mount *mnt = real_mount(path->mnt);
2499 if (!check_mnt(mnt))
2502 if (path->dentry != mnt->mnt.mnt_root)
2505 if (!can_change_locked_flags(mnt, mnt_flags))
2508 down_write(&sb->s_umount);
2509 ret = change_mount_ro_state(mnt, mnt_flags);
2511 set_mount_attributes(mnt, mnt_flags);
2512 up_write(&sb->s_umount);
2514 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2520 * change filesystem flags. dir should be a physical root of filesystem.
2521 * If you've mounted a non-root directory somewhere and want to do remount
2522 * on it - tough luck.
2524 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2525 int mnt_flags, void *data)
2528 struct super_block *sb = path->mnt->mnt_sb;
2529 struct mount *mnt = real_mount(path->mnt);
2530 struct fs_context *fc;
2532 if (!check_mnt(mnt))
2535 if (path->dentry != path->mnt->mnt_root)
2538 if (!can_change_locked_flags(mnt, mnt_flags))
2541 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2545 err = parse_monolithic_mount_data(fc, data);
2547 down_write(&sb->s_umount);
2549 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2550 err = reconfigure_super(fc);
2552 set_mount_attributes(mnt, mnt_flags);
2554 up_write(&sb->s_umount);
2557 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2563 static inline int tree_contains_unbindable(struct mount *mnt)
2566 for (p = mnt; p; p = next_mnt(p, mnt)) {
2567 if (IS_MNT_UNBINDABLE(p))
2574 * Check that there aren't references to earlier/same mount namespaces in the
2575 * specified subtree. Such references can act as pins for mount namespaces
2576 * that aren't checked by the mount-cycle checking code, thereby allowing
2577 * cycles to be made.
2579 static bool check_for_nsfs_mounts(struct mount *subtree)
2585 for (p = subtree; p; p = next_mnt(p, subtree))
2586 if (mnt_ns_loop(p->mnt.mnt_root))
2591 unlock_mount_hash();
2595 static int do_move_mount(struct path *old_path, struct path *new_path)
2597 struct mnt_namespace *ns;
2600 struct mount *parent;
2601 struct mountpoint *mp, *old_mp;
2605 mp = lock_mount(new_path);
2609 old = real_mount(old_path->mnt);
2610 p = real_mount(new_path->mnt);
2611 parent = old->mnt_parent;
2612 attached = mnt_has_parent(old);
2613 old_mp = old->mnt_mp;
2617 /* The mountpoint must be in our namespace. */
2621 /* The thing moved must be mounted... */
2622 if (!is_mounted(&old->mnt))
2625 /* ... and either ours or the root of anon namespace */
2626 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2629 if (old->mnt.mnt_flags & MNT_LOCKED)
2632 if (old_path->dentry != old_path->mnt->mnt_root)
2635 if (d_is_dir(new_path->dentry) !=
2636 d_is_dir(old_path->dentry))
2639 * Don't move a mount residing in a shared parent.
2641 if (attached && IS_MNT_SHARED(parent))
2644 * Don't move a mount tree containing unbindable mounts to a destination
2645 * mount which is shared.
2647 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2650 if (!check_for_nsfs_mounts(old))
2652 for (; mnt_has_parent(p); p = p->mnt_parent)
2656 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2661 /* if the mount is moved, it should no longer be expire
2663 list_del_init(&old->mnt_expire);
2665 put_mountpoint(old_mp);
2670 mntput_no_expire(parent);
2677 static int do_move_mount_old(struct path *path, const char *old_name)
2679 struct path old_path;
2682 if (!old_name || !*old_name)
2685 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2689 err = do_move_mount(&old_path, path);
2690 path_put(&old_path);
2695 * add a mount into a namespace's mount tree
2697 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2698 struct path *path, int mnt_flags)
2700 struct mount *parent = real_mount(path->mnt);
2702 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2704 if (unlikely(!check_mnt(parent))) {
2705 /* that's acceptable only for automounts done in private ns */
2706 if (!(mnt_flags & MNT_SHRINKABLE))
2708 /* ... and for those we'd better have mountpoint still alive */
2709 if (!parent->mnt_ns)
2713 /* Refuse the same filesystem on the same mount point */
2714 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2715 path->mnt->mnt_root == path->dentry)
2718 if (d_is_symlink(newmnt->mnt.mnt_root))
2721 newmnt->mnt.mnt_flags = mnt_flags;
2722 return graft_tree(newmnt, parent, mp);
2725 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2728 * Create a new mount using a superblock configuration and request it
2729 * be added to the namespace tree.
2731 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2732 unsigned int mnt_flags)
2734 struct vfsmount *mnt;
2735 struct mountpoint *mp;
2736 struct super_block *sb = fc->root->d_sb;
2739 error = security_sb_kern_mount(sb);
2740 if (!error && mount_too_revealing(sb, &mnt_flags))
2743 if (unlikely(error)) {
2748 up_write(&sb->s_umount);
2750 mnt = vfs_create_mount(fc);
2752 return PTR_ERR(mnt);
2754 mnt_warn_timestamp_expiry(mountpoint, mnt);
2756 mp = lock_mount(mountpoint);
2761 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2769 * create a new mount for userspace and request it to be added into the
2772 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2773 int mnt_flags, const char *name, void *data)
2775 struct file_system_type *type;
2776 struct fs_context *fc;
2777 const char *subtype = NULL;
2783 type = get_fs_type(fstype);
2787 if (type->fs_flags & FS_HAS_SUBTYPE) {
2788 subtype = strchr(fstype, '.');
2792 put_filesystem(type);
2798 fc = fs_context_for_mount(type, sb_flags);
2799 put_filesystem(type);
2804 err = vfs_parse_fs_string(fc, "subtype",
2805 subtype, strlen(subtype));
2807 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2809 err = parse_monolithic_mount_data(fc, data);
2810 if (!err && !mount_capable(fc))
2813 err = vfs_get_tree(fc);
2815 err = do_new_mount_fc(fc, path, mnt_flags);
2821 int finish_automount(struct vfsmount *m, struct path *path)
2823 struct dentry *dentry = path->dentry;
2824 struct mountpoint *mp;
2833 mnt = real_mount(m);
2834 /* The new mount record should have at least 2 refs to prevent it being
2835 * expired before we get a chance to add it
2837 BUG_ON(mnt_get_count(mnt) < 2);
2839 if (m->mnt_sb == path->mnt->mnt_sb &&
2840 m->mnt_root == dentry) {
2846 * we don't want to use lock_mount() - in this case finding something
2847 * that overmounts our mountpoint to be means "quitely drop what we've
2848 * got", not "try to mount it on top".
2850 inode_lock(dentry->d_inode);
2852 if (unlikely(cant_mount(dentry))) {
2854 goto discard_locked;
2857 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2860 goto discard_locked;
2863 mp = get_mountpoint(dentry);
2866 goto discard_locked;
2869 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2878 inode_unlock(dentry->d_inode);
2880 /* remove m from any expiration list it may be on */
2881 if (!list_empty(&mnt->mnt_expire)) {
2883 list_del_init(&mnt->mnt_expire);
2892 * mnt_set_expiry - Put a mount on an expiration list
2893 * @mnt: The mount to list.
2894 * @expiry_list: The list to add the mount to.
2896 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2900 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2904 EXPORT_SYMBOL(mnt_set_expiry);
2907 * process a list of expirable mountpoints with the intent of discarding any
2908 * mountpoints that aren't in use and haven't been touched since last we came
2911 void mark_mounts_for_expiry(struct list_head *mounts)
2913 struct mount *mnt, *next;
2914 LIST_HEAD(graveyard);
2916 if (list_empty(mounts))
2922 /* extract from the expiration list every vfsmount that matches the
2923 * following criteria:
2924 * - only referenced by its parent vfsmount
2925 * - still marked for expiry (marked on the last call here; marks are
2926 * cleared by mntput())
2928 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2929 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2930 propagate_mount_busy(mnt, 1))
2932 list_move(&mnt->mnt_expire, &graveyard);
2934 while (!list_empty(&graveyard)) {
2935 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2936 touch_mnt_namespace(mnt->mnt_ns);
2937 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2939 unlock_mount_hash();
2943 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2946 * Ripoff of 'select_parent()'
2948 * search the list of submounts for a given mountpoint, and move any
2949 * shrinkable submounts to the 'graveyard' list.
2951 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2953 struct mount *this_parent = parent;
2954 struct list_head *next;
2958 next = this_parent->mnt_mounts.next;
2960 while (next != &this_parent->mnt_mounts) {
2961 struct list_head *tmp = next;
2962 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2965 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2968 * Descend a level if the d_mounts list is non-empty.
2970 if (!list_empty(&mnt->mnt_mounts)) {
2975 if (!propagate_mount_busy(mnt, 1)) {
2976 list_move_tail(&mnt->mnt_expire, graveyard);
2981 * All done at this level ... ascend and resume the search
2983 if (this_parent != parent) {
2984 next = this_parent->mnt_child.next;
2985 this_parent = this_parent->mnt_parent;
2992 * process a list of expirable mountpoints with the intent of discarding any
2993 * submounts of a specific parent mountpoint
2995 * mount_lock must be held for write
2997 static void shrink_submounts(struct mount *mnt)
2999 LIST_HEAD(graveyard);
3002 /* extract submounts of 'mountpoint' from the expiration list */
3003 while (select_submounts(mnt, &graveyard)) {
3004 while (!list_empty(&graveyard)) {
3005 m = list_first_entry(&graveyard, struct mount,
3007 touch_mnt_namespace(m->mnt_ns);
3008 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3013 void *copy_mount_options(const void __user * data)
3021 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3023 return ERR_PTR(-ENOMEM);
3025 size = PAGE_SIZE - offset_in_page(data);
3027 if (copy_from_user(copy, data, size)) {
3029 return ERR_PTR(-EFAULT);
3031 if (size != PAGE_SIZE) {
3032 if (copy_from_user(copy + size, data + size, PAGE_SIZE - size))
3033 memset(copy + size, 0, PAGE_SIZE - size);
3038 char *copy_mount_string(const void __user *data)
3040 return data ? strndup_user(data, PATH_MAX) : NULL;
3044 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3045 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3047 * data is a (void *) that can point to any structure up to
3048 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3049 * information (or be NULL).
3051 * Pre-0.97 versions of mount() didn't have a flags word.
3052 * When the flags word was introduced its top half was required
3053 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3054 * Therefore, if this magic number is present, it carries no information
3055 * and must be discarded.
3057 long do_mount(const char *dev_name, const char __user *dir_name,
3058 const char *type_page, unsigned long flags, void *data_page)
3061 unsigned int mnt_flags = 0, sb_flags;
3065 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3066 flags &= ~MS_MGC_MSK;
3068 /* Basic sanity checks */
3070 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3072 if (flags & MS_NOUSER)
3075 /* ... and get the mountpoint */
3076 retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3080 retval = security_sb_mount(dev_name, &path,
3081 type_page, flags, data_page);
3082 if (!retval && !may_mount())
3084 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
3089 /* Default to relatime unless overriden */
3090 if (!(flags & MS_NOATIME))
3091 mnt_flags |= MNT_RELATIME;
3093 /* Separate the per-mountpoint flags */
3094 if (flags & MS_NOSUID)
3095 mnt_flags |= MNT_NOSUID;
3096 if (flags & MS_NODEV)
3097 mnt_flags |= MNT_NODEV;
3098 if (flags & MS_NOEXEC)
3099 mnt_flags |= MNT_NOEXEC;
3100 if (flags & MS_NOATIME)
3101 mnt_flags |= MNT_NOATIME;
3102 if (flags & MS_NODIRATIME)
3103 mnt_flags |= MNT_NODIRATIME;
3104 if (flags & MS_STRICTATIME)
3105 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3106 if (flags & MS_RDONLY)
3107 mnt_flags |= MNT_READONLY;
3109 /* The default atime for remount is preservation */
3110 if ((flags & MS_REMOUNT) &&
3111 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3112 MS_STRICTATIME)) == 0)) {
3113 mnt_flags &= ~MNT_ATIME_MASK;
3114 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
3117 sb_flags = flags & (SB_RDONLY |
3126 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3127 retval = do_reconfigure_mnt(&path, mnt_flags);
3128 else if (flags & MS_REMOUNT)
3129 retval = do_remount(&path, flags, sb_flags, mnt_flags,
3131 else if (flags & MS_BIND)
3132 retval = do_loopback(&path, dev_name, flags & MS_REC);
3133 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3134 retval = do_change_type(&path, flags);
3135 else if (flags & MS_MOVE)
3136 retval = do_move_mount_old(&path, dev_name);
3138 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
3139 dev_name, data_page);
3145 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3147 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3150 static void dec_mnt_namespaces(struct ucounts *ucounts)
3152 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3155 static void free_mnt_ns(struct mnt_namespace *ns)
3157 if (!is_anon_ns(ns))
3158 ns_free_inum(&ns->ns);
3159 dec_mnt_namespaces(ns->ucounts);
3160 put_user_ns(ns->user_ns);
3165 * Assign a sequence number so we can detect when we attempt to bind
3166 * mount a reference to an older mount namespace into the current
3167 * mount namespace, preventing reference counting loops. A 64bit
3168 * number incrementing at 10Ghz will take 12,427 years to wrap which
3169 * is effectively never, so we can ignore the possibility.
3171 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3173 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3175 struct mnt_namespace *new_ns;
3176 struct ucounts *ucounts;
3179 ucounts = inc_mnt_namespaces(user_ns);
3181 return ERR_PTR(-ENOSPC);
3183 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3185 dec_mnt_namespaces(ucounts);
3186 return ERR_PTR(-ENOMEM);
3189 ret = ns_alloc_inum(&new_ns->ns);
3192 dec_mnt_namespaces(ucounts);
3193 return ERR_PTR(ret);
3196 new_ns->ns.ops = &mntns_operations;
3198 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3199 atomic_set(&new_ns->count, 1);
3200 INIT_LIST_HEAD(&new_ns->list);
3201 init_waitqueue_head(&new_ns->poll);
3202 new_ns->user_ns = get_user_ns(user_ns);
3203 new_ns->ucounts = ucounts;
3208 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3209 struct user_namespace *user_ns, struct fs_struct *new_fs)
3211 struct mnt_namespace *new_ns;
3212 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3213 struct mount *p, *q;
3220 if (likely(!(flags & CLONE_NEWNS))) {
3227 new_ns = alloc_mnt_ns(user_ns, false);
3232 /* First pass: copy the tree topology */
3233 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3234 if (user_ns != ns->user_ns)
3235 copy_flags |= CL_SHARED_TO_SLAVE;
3236 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3239 free_mnt_ns(new_ns);
3240 return ERR_CAST(new);
3242 if (user_ns != ns->user_ns) {
3245 unlock_mount_hash();
3248 list_add_tail(&new_ns->list, &new->mnt_list);
3251 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3252 * as belonging to new namespace. We have already acquired a private
3253 * fs_struct, so tsk->fs->lock is not needed.
3261 if (&p->mnt == new_fs->root.mnt) {
3262 new_fs->root.mnt = mntget(&q->mnt);
3265 if (&p->mnt == new_fs->pwd.mnt) {
3266 new_fs->pwd.mnt = mntget(&q->mnt);
3270 p = next_mnt(p, old);
3271 q = next_mnt(q, new);
3274 while (p->mnt.mnt_root != q->mnt.mnt_root)
3275 p = next_mnt(p, old);
3287 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3289 struct mount *mnt = real_mount(m);
3290 struct mnt_namespace *ns;
3291 struct super_block *s;
3295 ns = alloc_mnt_ns(&init_user_ns, true);
3298 return ERR_CAST(ns);
3303 list_add(&mnt->mnt_list, &ns->list);
3305 err = vfs_path_lookup(m->mnt_root, m,
3306 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3311 return ERR_PTR(err);
3313 /* trade a vfsmount reference for active sb one */
3314 s = path.mnt->mnt_sb;
3315 atomic_inc(&s->s_active);
3317 /* lock the sucker */
3318 down_write(&s->s_umount);
3319 /* ... and return the root of (sub)tree on it */
3322 EXPORT_SYMBOL(mount_subtree);
3324 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3325 char __user *, type, unsigned long, flags, void __user *, data)
3332 kernel_type = copy_mount_string(type);
3333 ret = PTR_ERR(kernel_type);
3334 if (IS_ERR(kernel_type))
3337 kernel_dev = copy_mount_string(dev_name);
3338 ret = PTR_ERR(kernel_dev);
3339 if (IS_ERR(kernel_dev))
3342 options = copy_mount_options(data);
3343 ret = PTR_ERR(options);
3344 if (IS_ERR(options))
3347 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3359 * Create a kernel mount representation for a new, prepared superblock
3360 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3362 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3363 unsigned int, attr_flags)
3365 struct mnt_namespace *ns;
3366 struct fs_context *fc;
3368 struct path newmount;
3371 unsigned int mnt_flags = 0;
3377 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3380 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3385 MOUNT_ATTR_NODIRATIME))
3388 if (attr_flags & MOUNT_ATTR_RDONLY)
3389 mnt_flags |= MNT_READONLY;
3390 if (attr_flags & MOUNT_ATTR_NOSUID)
3391 mnt_flags |= MNT_NOSUID;
3392 if (attr_flags & MOUNT_ATTR_NODEV)
3393 mnt_flags |= MNT_NODEV;
3394 if (attr_flags & MOUNT_ATTR_NOEXEC)
3395 mnt_flags |= MNT_NOEXEC;
3396 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3397 mnt_flags |= MNT_NODIRATIME;
3399 switch (attr_flags & MOUNT_ATTR__ATIME) {
3400 case MOUNT_ATTR_STRICTATIME:
3402 case MOUNT_ATTR_NOATIME:
3403 mnt_flags |= MNT_NOATIME;
3405 case MOUNT_ATTR_RELATIME:
3406 mnt_flags |= MNT_RELATIME;
3417 if (f.file->f_op != &fscontext_fops)
3420 fc = f.file->private_data;
3422 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3426 /* There must be a valid superblock or we can't mount it */
3432 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3433 pr_warn("VFS: Mount too revealing\n");
3438 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3442 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3445 newmount.mnt = vfs_create_mount(fc);
3446 if (IS_ERR(newmount.mnt)) {
3447 ret = PTR_ERR(newmount.mnt);
3450 newmount.dentry = dget(fc->root);
3451 newmount.mnt->mnt_flags = mnt_flags;
3453 /* We've done the mount bit - now move the file context into more or
3454 * less the same state as if we'd done an fspick(). We don't want to
3455 * do any memory allocation or anything like that at this point as we
3456 * don't want to have to handle any errors incurred.
3458 vfs_clean_context(fc);
3460 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3465 mnt = real_mount(newmount.mnt);
3469 list_add(&mnt->mnt_list, &ns->list);
3470 mntget(newmount.mnt);
3472 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3473 * it, not just simply put it.
3475 file = dentry_open(&newmount, O_PATH, fc->cred);
3477 dissolve_on_fput(newmount.mnt);
3478 ret = PTR_ERR(file);
3481 file->f_mode |= FMODE_NEED_UNMOUNT;
3483 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3485 fd_install(ret, file);
3490 path_put(&newmount);
3492 mutex_unlock(&fc->uapi_mutex);
3499 * Move a mount from one place to another. In combination with
3500 * fsopen()/fsmount() this is used to install a new mount and in combination
3501 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3504 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3506 SYSCALL_DEFINE5(move_mount,
3507 int, from_dfd, const char __user *, from_pathname,
3508 int, to_dfd, const char __user *, to_pathname,
3509 unsigned int, flags)
3511 struct path from_path, to_path;
3512 unsigned int lflags;
3518 if (flags & ~MOVE_MOUNT__MASK)
3521 /* If someone gives a pathname, they aren't permitted to move
3522 * from an fd that requires unmount as we can't get at the flag
3523 * to clear it afterwards.
3526 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3527 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3528 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3530 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3535 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3536 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3537 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3539 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3543 ret = security_move_mount(&from_path, &to_path);
3547 ret = do_move_mount(&from_path, &to_path);
3552 path_put(&from_path);
3557 * Return true if path is reachable from root
3559 * namespace_sem or mount_lock is held
3561 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3562 const struct path *root)
3564 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3565 dentry = mnt->mnt_mountpoint;
3566 mnt = mnt->mnt_parent;
3568 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3571 bool path_is_under(const struct path *path1, const struct path *path2)
3574 read_seqlock_excl(&mount_lock);
3575 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3576 read_sequnlock_excl(&mount_lock);
3579 EXPORT_SYMBOL(path_is_under);
3582 * pivot_root Semantics:
3583 * Moves the root file system of the current process to the directory put_old,
3584 * makes new_root as the new root file system of the current process, and sets
3585 * root/cwd of all processes which had them on the current root to new_root.
3588 * The new_root and put_old must be directories, and must not be on the
3589 * same file system as the current process root. The put_old must be
3590 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3591 * pointed to by put_old must yield the same directory as new_root. No other
3592 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3594 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3595 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3596 * in this situation.
3599 * - we don't move root/cwd if they are not at the root (reason: if something
3600 * cared enough to change them, it's probably wrong to force them elsewhere)
3601 * - it's okay to pick a root that isn't the root of a file system, e.g.
3602 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3603 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3606 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3607 const char __user *, put_old)
3609 struct path new, old, root;
3610 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3611 struct mountpoint *old_mp, *root_mp;
3617 error = user_path_at(AT_FDCWD, new_root,
3618 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3622 error = user_path_at(AT_FDCWD, put_old,
3623 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3627 error = security_sb_pivotroot(&old, &new);
3631 get_fs_root(current->fs, &root);
3632 old_mp = lock_mount(&old);
3633 error = PTR_ERR(old_mp);
3638 new_mnt = real_mount(new.mnt);
3639 root_mnt = real_mount(root.mnt);
3640 old_mnt = real_mount(old.mnt);
3641 ex_parent = new_mnt->mnt_parent;
3642 root_parent = root_mnt->mnt_parent;
3643 if (IS_MNT_SHARED(old_mnt) ||
3644 IS_MNT_SHARED(ex_parent) ||
3645 IS_MNT_SHARED(root_parent))
3647 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3649 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3652 if (d_unlinked(new.dentry))
3655 if (new_mnt == root_mnt || old_mnt == root_mnt)
3656 goto out4; /* loop, on the same file system */
3658 if (root.mnt->mnt_root != root.dentry)
3659 goto out4; /* not a mountpoint */
3660 if (!mnt_has_parent(root_mnt))
3661 goto out4; /* not attached */
3662 if (new.mnt->mnt_root != new.dentry)
3663 goto out4; /* not a mountpoint */
3664 if (!mnt_has_parent(new_mnt))
3665 goto out4; /* not attached */
3666 /* make sure we can reach put_old from new_root */
3667 if (!is_path_reachable(old_mnt, old.dentry, &new))
3669 /* make certain new is below the root */
3670 if (!is_path_reachable(new_mnt, new.dentry, &root))
3673 umount_mnt(new_mnt);
3674 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3675 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3676 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3677 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3679 /* mount old root on put_old */
3680 attach_mnt(root_mnt, old_mnt, old_mp);
3681 /* mount new_root on / */
3682 attach_mnt(new_mnt, root_parent, root_mp);
3683 mnt_add_count(root_parent, -1);
3684 touch_mnt_namespace(current->nsproxy->mnt_ns);
3685 /* A moved mount should not expire automatically */
3686 list_del_init(&new_mnt->mnt_expire);
3687 put_mountpoint(root_mp);
3688 unlock_mount_hash();
3689 chroot_fs_refs(&root, &new);
3692 unlock_mount(old_mp);
3694 mntput_no_expire(ex_parent);
3705 static void __init init_mount_tree(void)
3707 struct vfsmount *mnt;
3709 struct mnt_namespace *ns;
3712 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3714 panic("Can't create rootfs");
3716 ns = alloc_mnt_ns(&init_user_ns, false);
3718 panic("Can't allocate initial namespace");
3719 m = real_mount(mnt);
3723 list_add(&m->mnt_list, &ns->list);
3724 init_task.nsproxy->mnt_ns = ns;
3728 root.dentry = mnt->mnt_root;
3729 mnt->mnt_flags |= MNT_LOCKED;
3731 set_fs_pwd(current->fs, &root);
3732 set_fs_root(current->fs, &root);
3735 void __init mnt_init(void)
3739 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3740 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3742 mount_hashtable = alloc_large_system_hash("Mount-cache",
3743 sizeof(struct hlist_head),
3746 &m_hash_shift, &m_hash_mask, 0, 0);
3747 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3748 sizeof(struct hlist_head),
3751 &mp_hash_shift, &mp_hash_mask, 0, 0);
3753 if (!mount_hashtable || !mountpoint_hashtable)
3754 panic("Failed to allocate mount hash table\n");
3760 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3762 fs_kobj = kobject_create_and_add("fs", NULL);
3764 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3770 void put_mnt_ns(struct mnt_namespace *ns)
3772 if (!atomic_dec_and_test(&ns->count))
3774 drop_collected_mounts(&ns->root->mnt);
3778 struct vfsmount *kern_mount(struct file_system_type *type)
3780 struct vfsmount *mnt;
3781 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3784 * it is a longterm mount, don't release mnt until
3785 * we unmount before file sys is unregistered
3787 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3791 EXPORT_SYMBOL_GPL(kern_mount);
3793 void kern_unmount(struct vfsmount *mnt)
3795 /* release long term mount so mount point can be released */
3796 if (!IS_ERR_OR_NULL(mnt)) {
3797 real_mount(mnt)->mnt_ns = NULL;
3798 synchronize_rcu(); /* yecchhh... */
3802 EXPORT_SYMBOL(kern_unmount);
3804 bool our_mnt(struct vfsmount *mnt)
3806 return check_mnt(real_mount(mnt));
3809 bool current_chrooted(void)
3811 /* Does the current process have a non-standard root */
3812 struct path ns_root;
3813 struct path fs_root;
3816 /* Find the namespace root */
3817 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3818 ns_root.dentry = ns_root.mnt->mnt_root;
3820 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3823 get_fs_root(current->fs, &fs_root);
3825 chrooted = !path_equal(&fs_root, &ns_root);
3833 static bool mnt_already_visible(struct mnt_namespace *ns,
3834 const struct super_block *sb,
3837 int new_flags = *new_mnt_flags;
3839 bool visible = false;
3841 down_read(&namespace_sem);
3842 list_for_each_entry(mnt, &ns->list, mnt_list) {
3843 struct mount *child;
3846 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3849 /* This mount is not fully visible if it's root directory
3850 * is not the root directory of the filesystem.
3852 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3855 /* A local view of the mount flags */
3856 mnt_flags = mnt->mnt.mnt_flags;
3858 /* Don't miss readonly hidden in the superblock flags */
3859 if (sb_rdonly(mnt->mnt.mnt_sb))
3860 mnt_flags |= MNT_LOCK_READONLY;
3862 /* Verify the mount flags are equal to or more permissive
3863 * than the proposed new mount.
3865 if ((mnt_flags & MNT_LOCK_READONLY) &&
3866 !(new_flags & MNT_READONLY))
3868 if ((mnt_flags & MNT_LOCK_ATIME) &&
3869 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3872 /* This mount is not fully visible if there are any
3873 * locked child mounts that cover anything except for
3874 * empty directories.
3876 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3877 struct inode *inode = child->mnt_mountpoint->d_inode;
3878 /* Only worry about locked mounts */
3879 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3881 /* Is the directory permanetly empty? */
3882 if (!is_empty_dir_inode(inode))
3885 /* Preserve the locked attributes */
3886 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3893 up_read(&namespace_sem);
3897 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3899 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3900 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3901 unsigned long s_iflags;
3903 if (ns->user_ns == &init_user_ns)
3906 /* Can this filesystem be too revealing? */
3907 s_iflags = sb->s_iflags;
3908 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3911 if ((s_iflags & required_iflags) != required_iflags) {
3912 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3917 return !mnt_already_visible(ns, sb, new_mnt_flags);
3920 bool mnt_may_suid(struct vfsmount *mnt)
3923 * Foreign mounts (accessed via fchdir or through /proc
3924 * symlinks) are always treated as if they are nosuid. This
3925 * prevents namespaces from trusting potentially unsafe
3926 * suid/sgid bits, file caps, or security labels that originate
3927 * in other namespaces.
3929 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3930 current_in_userns(mnt->mnt_sb->s_user_ns);
3933 static struct ns_common *mntns_get(struct task_struct *task)
3935 struct ns_common *ns = NULL;
3936 struct nsproxy *nsproxy;
3939 nsproxy = task->nsproxy;
3941 ns = &nsproxy->mnt_ns->ns;
3942 get_mnt_ns(to_mnt_ns(ns));
3949 static void mntns_put(struct ns_common *ns)
3951 put_mnt_ns(to_mnt_ns(ns));
3954 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3956 struct fs_struct *fs = current->fs;
3957 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3961 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3962 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3963 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3966 if (is_anon_ns(mnt_ns))
3973 old_mnt_ns = nsproxy->mnt_ns;
3974 nsproxy->mnt_ns = mnt_ns;
3977 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3978 "/", LOOKUP_DOWN, &root);
3980 /* revert to old namespace */
3981 nsproxy->mnt_ns = old_mnt_ns;
3986 put_mnt_ns(old_mnt_ns);
3988 /* Update the pwd and root */
3989 set_fs_pwd(fs, &root);
3990 set_fs_root(fs, &root);
3996 static struct user_namespace *mntns_owner(struct ns_common *ns)
3998 return to_mnt_ns(ns)->user_ns;
4001 const struct proc_ns_operations mntns_operations = {
4003 .type = CLONE_NEWNS,
4006 .install = mntns_install,
4007 .owner = mntns_owner,