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 if (!d_mountpoint(dentry))
675 down_read(&namespace_sem);
676 list_for_each_entry(mnt, &ns->list, mnt_list) {
677 is_covered = (mnt->mnt_mountpoint == dentry);
681 up_read(&namespace_sem);
686 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
688 struct hlist_head *chain = mp_hash(dentry);
689 struct mountpoint *mp;
691 hlist_for_each_entry(mp, chain, m_hash) {
692 if (mp->m_dentry == dentry) {
700 static struct mountpoint *get_mountpoint(struct dentry *dentry)
702 struct mountpoint *mp, *new = NULL;
705 if (d_mountpoint(dentry)) {
706 /* might be worth a WARN_ON() */
707 if (d_unlinked(dentry))
708 return ERR_PTR(-ENOENT);
710 read_seqlock_excl(&mount_lock);
711 mp = lookup_mountpoint(dentry);
712 read_sequnlock_excl(&mount_lock);
718 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
720 return ERR_PTR(-ENOMEM);
723 /* Exactly one processes may set d_mounted */
724 ret = d_set_mounted(dentry);
726 /* Someone else set d_mounted? */
730 /* The dentry is not available as a mountpoint? */
735 /* Add the new mountpoint to the hash table */
736 read_seqlock_excl(&mount_lock);
737 new->m_dentry = dget(dentry);
739 hlist_add_head(&new->m_hash, mp_hash(dentry));
740 INIT_HLIST_HEAD(&new->m_list);
741 read_sequnlock_excl(&mount_lock);
751 * vfsmount lock must be held. Additionally, the caller is responsible
752 * for serializing calls for given disposal list.
754 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
756 if (!--mp->m_count) {
757 struct dentry *dentry = mp->m_dentry;
758 BUG_ON(!hlist_empty(&mp->m_list));
759 spin_lock(&dentry->d_lock);
760 dentry->d_flags &= ~DCACHE_MOUNTED;
761 spin_unlock(&dentry->d_lock);
762 dput_to_list(dentry, list);
763 hlist_del(&mp->m_hash);
768 /* called with namespace_lock and vfsmount lock */
769 static void put_mountpoint(struct mountpoint *mp)
771 __put_mountpoint(mp, &ex_mountpoints);
774 static inline int check_mnt(struct mount *mnt)
776 return mnt->mnt_ns == current->nsproxy->mnt_ns;
780 * vfsmount lock must be held for write
782 static void touch_mnt_namespace(struct mnt_namespace *ns)
786 wake_up_interruptible(&ns->poll);
791 * vfsmount lock must be held for write
793 static void __touch_mnt_namespace(struct mnt_namespace *ns)
795 if (ns && ns->event != event) {
797 wake_up_interruptible(&ns->poll);
802 * vfsmount lock must be held for write
804 static struct mountpoint *unhash_mnt(struct mount *mnt)
806 struct mountpoint *mp;
807 mnt->mnt_parent = mnt;
808 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
809 list_del_init(&mnt->mnt_child);
810 hlist_del_init_rcu(&mnt->mnt_hash);
811 hlist_del_init(&mnt->mnt_mp_list);
818 * vfsmount lock must be held for write
820 static void umount_mnt(struct mount *mnt)
822 put_mountpoint(unhash_mnt(mnt));
826 * vfsmount lock must be held for write
828 void mnt_set_mountpoint(struct mount *mnt,
829 struct mountpoint *mp,
830 struct mount *child_mnt)
833 mnt_add_count(mnt, 1); /* essentially, that's mntget */
834 child_mnt->mnt_mountpoint = mp->m_dentry;
835 child_mnt->mnt_parent = mnt;
836 child_mnt->mnt_mp = mp;
837 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
840 static void __attach_mnt(struct mount *mnt, struct mount *parent)
842 hlist_add_head_rcu(&mnt->mnt_hash,
843 m_hash(&parent->mnt, mnt->mnt_mountpoint));
844 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
848 * vfsmount lock must be held for write
850 static void attach_mnt(struct mount *mnt,
851 struct mount *parent,
852 struct mountpoint *mp)
854 mnt_set_mountpoint(parent, mp, mnt);
855 __attach_mnt(mnt, parent);
858 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
860 struct mountpoint *old_mp = mnt->mnt_mp;
861 struct mount *old_parent = mnt->mnt_parent;
863 list_del_init(&mnt->mnt_child);
864 hlist_del_init(&mnt->mnt_mp_list);
865 hlist_del_init_rcu(&mnt->mnt_hash);
867 attach_mnt(mnt, parent, mp);
869 put_mountpoint(old_mp);
870 mnt_add_count(old_parent, -1);
874 * vfsmount lock must be held for write
876 static void commit_tree(struct mount *mnt)
878 struct mount *parent = mnt->mnt_parent;
881 struct mnt_namespace *n = parent->mnt_ns;
883 BUG_ON(parent == mnt);
885 list_add_tail(&head, &mnt->mnt_list);
886 list_for_each_entry(m, &head, mnt_list)
889 list_splice(&head, n->list.prev);
891 n->mounts += n->pending_mounts;
892 n->pending_mounts = 0;
894 __attach_mnt(mnt, parent);
895 touch_mnt_namespace(n);
898 static struct mount *next_mnt(struct mount *p, struct mount *root)
900 struct list_head *next = p->mnt_mounts.next;
901 if (next == &p->mnt_mounts) {
905 next = p->mnt_child.next;
906 if (next != &p->mnt_parent->mnt_mounts)
911 return list_entry(next, struct mount, mnt_child);
914 static struct mount *skip_mnt_tree(struct mount *p)
916 struct list_head *prev = p->mnt_mounts.prev;
917 while (prev != &p->mnt_mounts) {
918 p = list_entry(prev, struct mount, mnt_child);
919 prev = p->mnt_mounts.prev;
925 * vfs_create_mount - Create a mount for a configured superblock
926 * @fc: The configuration context with the superblock attached
928 * Create a mount to an already configured superblock. If necessary, the
929 * caller should invoke vfs_get_tree() before calling this.
931 * Note that this does not attach the mount to anything.
933 struct vfsmount *vfs_create_mount(struct fs_context *fc)
938 return ERR_PTR(-EINVAL);
940 mnt = alloc_vfsmnt(fc->source ?: "none");
942 return ERR_PTR(-ENOMEM);
944 if (fc->sb_flags & SB_KERNMOUNT)
945 mnt->mnt.mnt_flags = MNT_INTERNAL;
947 atomic_inc(&fc->root->d_sb->s_active);
948 mnt->mnt.mnt_sb = fc->root->d_sb;
949 mnt->mnt.mnt_root = dget(fc->root);
950 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
951 mnt->mnt_parent = mnt;
954 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
958 EXPORT_SYMBOL(vfs_create_mount);
960 struct vfsmount *fc_mount(struct fs_context *fc)
962 int err = vfs_get_tree(fc);
964 up_write(&fc->root->d_sb->s_umount);
965 return vfs_create_mount(fc);
969 EXPORT_SYMBOL(fc_mount);
971 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
972 int flags, const char *name,
975 struct fs_context *fc;
976 struct vfsmount *mnt;
980 return ERR_PTR(-EINVAL);
982 fc = fs_context_for_mount(type, flags);
987 ret = vfs_parse_fs_string(fc, "source",
990 ret = parse_monolithic_mount_data(fc, data);
999 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1002 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1003 const char *name, void *data)
1005 /* Until it is worked out how to pass the user namespace
1006 * through from the parent mount to the submount don't support
1007 * unprivileged mounts with submounts.
1009 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1010 return ERR_PTR(-EPERM);
1012 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1014 EXPORT_SYMBOL_GPL(vfs_submount);
1016 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1019 struct super_block *sb = old->mnt.mnt_sb;
1023 mnt = alloc_vfsmnt(old->mnt_devname);
1025 return ERR_PTR(-ENOMEM);
1027 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1028 mnt->mnt_group_id = 0; /* not a peer of original */
1030 mnt->mnt_group_id = old->mnt_group_id;
1032 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1033 err = mnt_alloc_group_id(mnt);
1038 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1039 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1041 atomic_inc(&sb->s_active);
1042 mnt->mnt.mnt_sb = sb;
1043 mnt->mnt.mnt_root = dget(root);
1044 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1045 mnt->mnt_parent = mnt;
1047 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1048 unlock_mount_hash();
1050 if ((flag & CL_SLAVE) ||
1051 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1052 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1053 mnt->mnt_master = old;
1054 CLEAR_MNT_SHARED(mnt);
1055 } else if (!(flag & CL_PRIVATE)) {
1056 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1057 list_add(&mnt->mnt_share, &old->mnt_share);
1058 if (IS_MNT_SLAVE(old))
1059 list_add(&mnt->mnt_slave, &old->mnt_slave);
1060 mnt->mnt_master = old->mnt_master;
1062 CLEAR_MNT_SHARED(mnt);
1064 if (flag & CL_MAKE_SHARED)
1065 set_mnt_shared(mnt);
1067 /* stick the duplicate mount on the same expiry list
1068 * as the original if that was on one */
1069 if (flag & CL_EXPIRE) {
1070 if (!list_empty(&old->mnt_expire))
1071 list_add(&mnt->mnt_expire, &old->mnt_expire);
1079 return ERR_PTR(err);
1082 static void cleanup_mnt(struct mount *mnt)
1084 struct hlist_node *p;
1087 * The warning here probably indicates that somebody messed
1088 * up a mnt_want/drop_write() pair. If this happens, the
1089 * filesystem was probably unable to make r/w->r/o transitions.
1090 * The locking used to deal with mnt_count decrement provides barriers,
1091 * so mnt_get_writers() below is safe.
1093 WARN_ON(mnt_get_writers(mnt));
1094 if (unlikely(mnt->mnt_pins.first))
1096 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1097 hlist_del(&m->mnt_umount);
1100 fsnotify_vfsmount_delete(&mnt->mnt);
1101 dput(mnt->mnt.mnt_root);
1102 deactivate_super(mnt->mnt.mnt_sb);
1104 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1107 static void __cleanup_mnt(struct rcu_head *head)
1109 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1112 static LLIST_HEAD(delayed_mntput_list);
1113 static void delayed_mntput(struct work_struct *unused)
1115 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1116 struct mount *m, *t;
1118 llist_for_each_entry_safe(m, t, node, mnt_llist)
1121 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1123 static void mntput_no_expire(struct mount *mnt)
1128 if (likely(READ_ONCE(mnt->mnt_ns))) {
1130 * Since we don't do lock_mount_hash() here,
1131 * ->mnt_ns can change under us. However, if it's
1132 * non-NULL, then there's a reference that won't
1133 * be dropped until after an RCU delay done after
1134 * turning ->mnt_ns NULL. So if we observe it
1135 * non-NULL under rcu_read_lock(), the reference
1136 * we are dropping is not the final one.
1138 mnt_add_count(mnt, -1);
1144 * make sure that if __legitimize_mnt() has not seen us grab
1145 * mount_lock, we'll see their refcount increment here.
1148 mnt_add_count(mnt, -1);
1149 if (mnt_get_count(mnt)) {
1151 unlock_mount_hash();
1154 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1156 unlock_mount_hash();
1159 mnt->mnt.mnt_flags |= MNT_DOOMED;
1162 list_del(&mnt->mnt_instance);
1164 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1165 struct mount *p, *tmp;
1166 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1167 __put_mountpoint(unhash_mnt(p), &list);
1168 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1171 unlock_mount_hash();
1172 shrink_dentry_list(&list);
1174 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1175 struct task_struct *task = current;
1176 if (likely(!(task->flags & PF_KTHREAD))) {
1177 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1178 if (!task_work_add(task, &mnt->mnt_rcu, true))
1181 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1182 schedule_delayed_work(&delayed_mntput_work, 1);
1188 void mntput(struct vfsmount *mnt)
1191 struct mount *m = real_mount(mnt);
1192 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1193 if (unlikely(m->mnt_expiry_mark))
1194 m->mnt_expiry_mark = 0;
1195 mntput_no_expire(m);
1198 EXPORT_SYMBOL(mntput);
1200 struct vfsmount *mntget(struct vfsmount *mnt)
1203 mnt_add_count(real_mount(mnt), 1);
1206 EXPORT_SYMBOL(mntget);
1208 /* path_is_mountpoint() - Check if path is a mount in the current
1211 * d_mountpoint() can only be used reliably to establish if a dentry is
1212 * not mounted in any namespace and that common case is handled inline.
1213 * d_mountpoint() isn't aware of the possibility there may be multiple
1214 * mounts using a given dentry in a different namespace. This function
1215 * checks if the passed in path is a mountpoint rather than the dentry
1218 bool path_is_mountpoint(const struct path *path)
1223 if (!d_mountpoint(path->dentry))
1228 seq = read_seqbegin(&mount_lock);
1229 res = __path_is_mountpoint(path);
1230 } while (read_seqretry(&mount_lock, seq));
1235 EXPORT_SYMBOL(path_is_mountpoint);
1237 struct vfsmount *mnt_clone_internal(const struct path *path)
1240 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1243 p->mnt.mnt_flags |= MNT_INTERNAL;
1247 #ifdef CONFIG_PROC_FS
1248 /* iterator; we want it to have access to namespace_sem, thus here... */
1249 static void *m_start(struct seq_file *m, loff_t *pos)
1251 struct proc_mounts *p = m->private;
1253 down_read(&namespace_sem);
1254 if (p->cached_event == p->ns->event) {
1255 void *v = p->cached_mount;
1256 if (*pos == p->cached_index)
1258 if (*pos == p->cached_index + 1) {
1259 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1260 return p->cached_mount = v;
1264 p->cached_event = p->ns->event;
1265 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1266 p->cached_index = *pos;
1267 return p->cached_mount;
1270 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1272 struct proc_mounts *p = m->private;
1274 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1275 p->cached_index = *pos;
1276 return p->cached_mount;
1279 static void m_stop(struct seq_file *m, void *v)
1281 up_read(&namespace_sem);
1284 static int m_show(struct seq_file *m, void *v)
1286 struct proc_mounts *p = m->private;
1287 struct mount *r = list_entry(v, struct mount, mnt_list);
1288 return p->show(m, &r->mnt);
1291 const struct seq_operations mounts_op = {
1297 #endif /* CONFIG_PROC_FS */
1300 * may_umount_tree - check if a mount tree is busy
1301 * @mnt: root of mount tree
1303 * This is called to check if a tree of mounts has any
1304 * open files, pwds, chroots or sub mounts that are
1307 int may_umount_tree(struct vfsmount *m)
1309 struct mount *mnt = real_mount(m);
1310 int actual_refs = 0;
1311 int minimum_refs = 0;
1315 /* write lock needed for mnt_get_count */
1317 for (p = mnt; p; p = next_mnt(p, mnt)) {
1318 actual_refs += mnt_get_count(p);
1321 unlock_mount_hash();
1323 if (actual_refs > minimum_refs)
1329 EXPORT_SYMBOL(may_umount_tree);
1332 * may_umount - check if a mount point is busy
1333 * @mnt: root of mount
1335 * This is called to check if a mount point has any
1336 * open files, pwds, chroots or sub mounts. If the
1337 * mount has sub mounts this will return busy
1338 * regardless of whether the sub mounts are busy.
1340 * Doesn't take quota and stuff into account. IOW, in some cases it will
1341 * give false negatives. The main reason why it's here is that we need
1342 * a non-destructive way to look for easily umountable filesystems.
1344 int may_umount(struct vfsmount *mnt)
1347 down_read(&namespace_sem);
1349 if (propagate_mount_busy(real_mount(mnt), 2))
1351 unlock_mount_hash();
1352 up_read(&namespace_sem);
1356 EXPORT_SYMBOL(may_umount);
1358 static void namespace_unlock(void)
1360 struct hlist_head head;
1361 struct hlist_node *p;
1365 hlist_move_list(&unmounted, &head);
1366 list_splice_init(&ex_mountpoints, &list);
1368 up_write(&namespace_sem);
1370 shrink_dentry_list(&list);
1372 if (likely(hlist_empty(&head)))
1375 synchronize_rcu_expedited();
1377 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1378 hlist_del(&m->mnt_umount);
1383 static inline void namespace_lock(void)
1385 down_write(&namespace_sem);
1388 enum umount_tree_flags {
1390 UMOUNT_PROPAGATE = 2,
1391 UMOUNT_CONNECTED = 4,
1394 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1396 /* Leaving mounts connected is only valid for lazy umounts */
1397 if (how & UMOUNT_SYNC)
1400 /* A mount without a parent has nothing to be connected to */
1401 if (!mnt_has_parent(mnt))
1404 /* Because the reference counting rules change when mounts are
1405 * unmounted and connected, umounted mounts may not be
1406 * connected to mounted mounts.
1408 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1411 /* Has it been requested that the mount remain connected? */
1412 if (how & UMOUNT_CONNECTED)
1415 /* Is the mount locked such that it needs to remain connected? */
1416 if (IS_MNT_LOCKED(mnt))
1419 /* By default disconnect the mount */
1424 * mount_lock must be held
1425 * namespace_sem must be held for write
1427 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1429 LIST_HEAD(tmp_list);
1432 if (how & UMOUNT_PROPAGATE)
1433 propagate_mount_unlock(mnt);
1435 /* Gather the mounts to umount */
1436 for (p = mnt; p; p = next_mnt(p, mnt)) {
1437 p->mnt.mnt_flags |= MNT_UMOUNT;
1438 list_move(&p->mnt_list, &tmp_list);
1441 /* Hide the mounts from mnt_mounts */
1442 list_for_each_entry(p, &tmp_list, mnt_list) {
1443 list_del_init(&p->mnt_child);
1446 /* Add propogated mounts to the tmp_list */
1447 if (how & UMOUNT_PROPAGATE)
1448 propagate_umount(&tmp_list);
1450 while (!list_empty(&tmp_list)) {
1451 struct mnt_namespace *ns;
1453 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1454 list_del_init(&p->mnt_expire);
1455 list_del_init(&p->mnt_list);
1459 __touch_mnt_namespace(ns);
1462 if (how & UMOUNT_SYNC)
1463 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1465 disconnect = disconnect_mount(p, how);
1466 if (mnt_has_parent(p)) {
1467 mnt_add_count(p->mnt_parent, -1);
1469 /* Don't forget about p */
1470 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1475 change_mnt_propagation(p, MS_PRIVATE);
1477 hlist_add_head(&p->mnt_umount, &unmounted);
1481 static void shrink_submounts(struct mount *mnt);
1483 static int do_umount_root(struct super_block *sb)
1487 down_write(&sb->s_umount);
1488 if (!sb_rdonly(sb)) {
1489 struct fs_context *fc;
1491 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1496 ret = parse_monolithic_mount_data(fc, NULL);
1498 ret = reconfigure_super(fc);
1502 up_write(&sb->s_umount);
1506 static int do_umount(struct mount *mnt, int flags)
1508 struct super_block *sb = mnt->mnt.mnt_sb;
1511 retval = security_sb_umount(&mnt->mnt, flags);
1516 * Allow userspace to request a mountpoint be expired rather than
1517 * unmounting unconditionally. Unmount only happens if:
1518 * (1) the mark is already set (the mark is cleared by mntput())
1519 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1521 if (flags & MNT_EXPIRE) {
1522 if (&mnt->mnt == current->fs->root.mnt ||
1523 flags & (MNT_FORCE | MNT_DETACH))
1527 * probably don't strictly need the lock here if we examined
1528 * all race cases, but it's a slowpath.
1531 if (mnt_get_count(mnt) != 2) {
1532 unlock_mount_hash();
1535 unlock_mount_hash();
1537 if (!xchg(&mnt->mnt_expiry_mark, 1))
1542 * If we may have to abort operations to get out of this
1543 * mount, and they will themselves hold resources we must
1544 * allow the fs to do things. In the Unix tradition of
1545 * 'Gee thats tricky lets do it in userspace' the umount_begin
1546 * might fail to complete on the first run through as other tasks
1547 * must return, and the like. Thats for the mount program to worry
1548 * about for the moment.
1551 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1552 sb->s_op->umount_begin(sb);
1556 * No sense to grab the lock for this test, but test itself looks
1557 * somewhat bogus. Suggestions for better replacement?
1558 * Ho-hum... In principle, we might treat that as umount + switch
1559 * to rootfs. GC would eventually take care of the old vfsmount.
1560 * Actually it makes sense, especially if rootfs would contain a
1561 * /reboot - static binary that would close all descriptors and
1562 * call reboot(9). Then init(8) could umount root and exec /reboot.
1564 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1566 * Special case for "unmounting" root ...
1567 * we just try to remount it readonly.
1569 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1571 return do_umount_root(sb);
1577 /* Recheck MNT_LOCKED with the locks held */
1579 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1583 if (flags & MNT_DETACH) {
1584 if (!list_empty(&mnt->mnt_list))
1585 umount_tree(mnt, UMOUNT_PROPAGATE);
1588 shrink_submounts(mnt);
1590 if (!propagate_mount_busy(mnt, 2)) {
1591 if (!list_empty(&mnt->mnt_list))
1592 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1597 unlock_mount_hash();
1603 * __detach_mounts - lazily unmount all mounts on the specified dentry
1605 * During unlink, rmdir, and d_drop it is possible to loose the path
1606 * to an existing mountpoint, and wind up leaking the mount.
1607 * detach_mounts allows lazily unmounting those mounts instead of
1610 * The caller may hold dentry->d_inode->i_mutex.
1612 void __detach_mounts(struct dentry *dentry)
1614 struct mountpoint *mp;
1619 mp = lookup_mountpoint(dentry);
1624 while (!hlist_empty(&mp->m_list)) {
1625 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1626 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1628 hlist_add_head(&mnt->mnt_umount, &unmounted);
1630 else umount_tree(mnt, UMOUNT_CONNECTED);
1634 unlock_mount_hash();
1639 * Is the caller allowed to modify his namespace?
1641 static inline bool may_mount(void)
1643 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1646 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1647 static inline bool may_mandlock(void)
1649 return capable(CAP_SYS_ADMIN);
1652 static inline bool may_mandlock(void)
1654 pr_warn("VFS: \"mand\" mount option not supported");
1660 * Now umount can handle mount points as well as block devices.
1661 * This is important for filesystems which use unnamed block devices.
1663 * We now support a flag for forced unmount like the other 'big iron'
1664 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1667 int ksys_umount(char __user *name, int flags)
1672 int lookup_flags = 0;
1674 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1680 if (!(flags & UMOUNT_NOFOLLOW))
1681 lookup_flags |= LOOKUP_FOLLOW;
1683 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1686 mnt = real_mount(path.mnt);
1688 if (path.dentry != path.mnt->mnt_root)
1690 if (!check_mnt(mnt))
1692 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1695 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1698 retval = do_umount(mnt, flags);
1700 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1702 mntput_no_expire(mnt);
1707 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1709 return ksys_umount(name, flags);
1712 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1715 * The 2.0 compatible umount. No flags.
1717 SYSCALL_DEFINE1(oldumount, char __user *, name)
1719 return ksys_umount(name, 0);
1724 static bool is_mnt_ns_file(struct dentry *dentry)
1726 /* Is this a proxy for a mount namespace? */
1727 return dentry->d_op == &ns_dentry_operations &&
1728 dentry->d_fsdata == &mntns_operations;
1731 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1733 return container_of(ns, struct mnt_namespace, ns);
1736 static bool mnt_ns_loop(struct dentry *dentry)
1738 /* Could bind mounting the mount namespace inode cause a
1739 * mount namespace loop?
1741 struct mnt_namespace *mnt_ns;
1742 if (!is_mnt_ns_file(dentry))
1745 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1746 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1749 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1752 struct mount *res, *p, *q, *r, *parent;
1754 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1755 return ERR_PTR(-EINVAL);
1757 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1758 return ERR_PTR(-EINVAL);
1760 res = q = clone_mnt(mnt, dentry, flag);
1764 q->mnt_mountpoint = mnt->mnt_mountpoint;
1767 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1769 if (!is_subdir(r->mnt_mountpoint, dentry))
1772 for (s = r; s; s = next_mnt(s, r)) {
1773 if (!(flag & CL_COPY_UNBINDABLE) &&
1774 IS_MNT_UNBINDABLE(s)) {
1775 if (s->mnt.mnt_flags & MNT_LOCKED) {
1776 /* Both unbindable and locked. */
1777 q = ERR_PTR(-EPERM);
1780 s = skip_mnt_tree(s);
1784 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1785 is_mnt_ns_file(s->mnt.mnt_root)) {
1786 s = skip_mnt_tree(s);
1789 while (p != s->mnt_parent) {
1795 q = clone_mnt(p, p->mnt.mnt_root, flag);
1799 list_add_tail(&q->mnt_list, &res->mnt_list);
1800 attach_mnt(q, parent, p->mnt_mp);
1801 unlock_mount_hash();
1808 umount_tree(res, UMOUNT_SYNC);
1809 unlock_mount_hash();
1814 /* Caller should check returned pointer for errors */
1816 struct vfsmount *collect_mounts(const struct path *path)
1820 if (!check_mnt(real_mount(path->mnt)))
1821 tree = ERR_PTR(-EINVAL);
1823 tree = copy_tree(real_mount(path->mnt), path->dentry,
1824 CL_COPY_ALL | CL_PRIVATE);
1827 return ERR_CAST(tree);
1831 static void free_mnt_ns(struct mnt_namespace *);
1832 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1834 void dissolve_on_fput(struct vfsmount *mnt)
1836 struct mnt_namespace *ns;
1839 ns = real_mount(mnt)->mnt_ns;
1842 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1846 unlock_mount_hash();
1852 void drop_collected_mounts(struct vfsmount *mnt)
1856 umount_tree(real_mount(mnt), 0);
1857 unlock_mount_hash();
1862 * clone_private_mount - create a private clone of a path
1864 * This creates a new vfsmount, which will be the clone of @path. The new will
1865 * not be attached anywhere in the namespace and will be private (i.e. changes
1866 * to the originating mount won't be propagated into this).
1868 * Release with mntput().
1870 struct vfsmount *clone_private_mount(const struct path *path)
1872 struct mount *old_mnt = real_mount(path->mnt);
1873 struct mount *new_mnt;
1875 if (IS_MNT_UNBINDABLE(old_mnt))
1876 return ERR_PTR(-EINVAL);
1878 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1879 if (IS_ERR(new_mnt))
1880 return ERR_CAST(new_mnt);
1882 return &new_mnt->mnt;
1884 EXPORT_SYMBOL_GPL(clone_private_mount);
1886 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1887 struct vfsmount *root)
1890 int res = f(root, arg);
1893 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1894 res = f(&mnt->mnt, arg);
1901 static void lock_mnt_tree(struct mount *mnt)
1905 for (p = mnt; p; p = next_mnt(p, mnt)) {
1906 int flags = p->mnt.mnt_flags;
1907 /* Don't allow unprivileged users to change mount flags */
1908 flags |= MNT_LOCK_ATIME;
1910 if (flags & MNT_READONLY)
1911 flags |= MNT_LOCK_READONLY;
1913 if (flags & MNT_NODEV)
1914 flags |= MNT_LOCK_NODEV;
1916 if (flags & MNT_NOSUID)
1917 flags |= MNT_LOCK_NOSUID;
1919 if (flags & MNT_NOEXEC)
1920 flags |= MNT_LOCK_NOEXEC;
1921 /* Don't allow unprivileged users to reveal what is under a mount */
1922 if (list_empty(&p->mnt_expire))
1923 flags |= MNT_LOCKED;
1924 p->mnt.mnt_flags = flags;
1928 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1932 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1933 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1934 mnt_release_group_id(p);
1938 static int invent_group_ids(struct mount *mnt, bool recurse)
1942 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1943 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1944 int err = mnt_alloc_group_id(p);
1946 cleanup_group_ids(mnt, p);
1955 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1957 unsigned int max = READ_ONCE(sysctl_mount_max);
1958 unsigned int mounts = 0, old, pending, sum;
1961 for (p = mnt; p; p = next_mnt(p, mnt))
1965 pending = ns->pending_mounts;
1966 sum = old + pending;
1970 (mounts > (max - sum)))
1973 ns->pending_mounts = pending + mounts;
1978 * @source_mnt : mount tree to be attached
1979 * @nd : place the mount tree @source_mnt is attached
1980 * @parent_nd : if non-null, detach the source_mnt from its parent and
1981 * store the parent mount and mountpoint dentry.
1982 * (done when source_mnt is moved)
1984 * NOTE: in the table below explains the semantics when a source mount
1985 * of a given type is attached to a destination mount of a given type.
1986 * ---------------------------------------------------------------------------
1987 * | BIND MOUNT OPERATION |
1988 * |**************************************************************************
1989 * | source-->| shared | private | slave | unbindable |
1993 * |**************************************************************************
1994 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1996 * |non-shared| shared (+) | private | slave (*) | invalid |
1997 * ***************************************************************************
1998 * A bind operation clones the source mount and mounts the clone on the
1999 * destination mount.
2001 * (++) the cloned mount is propagated to all the mounts in the propagation
2002 * tree of the destination mount and the cloned mount is added to
2003 * the peer group of the source mount.
2004 * (+) the cloned mount is created under the destination mount and is marked
2005 * as shared. The cloned mount is added to the peer group of the source
2007 * (+++) the mount is propagated to all the mounts in the propagation tree
2008 * of the destination mount and the cloned mount is made slave
2009 * of the same master as that of the source mount. The cloned mount
2010 * is marked as 'shared and slave'.
2011 * (*) the cloned mount is made a slave of the same master as that of the
2014 * ---------------------------------------------------------------------------
2015 * | MOVE MOUNT OPERATION |
2016 * |**************************************************************************
2017 * | source-->| shared | private | slave | unbindable |
2021 * |**************************************************************************
2022 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2024 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2025 * ***************************************************************************
2027 * (+) the mount is moved to the destination. And is then propagated to
2028 * all the mounts in the propagation tree of the destination mount.
2029 * (+*) the mount is moved to the destination.
2030 * (+++) the mount is moved to the destination and is then propagated to
2031 * all the mounts belonging to the destination mount's propagation tree.
2032 * the mount is marked as 'shared and slave'.
2033 * (*) the mount continues to be a slave at the new location.
2035 * if the source mount is a tree, the operations explained above is
2036 * applied to each mount in the tree.
2037 * Must be called without spinlocks held, since this function can sleep
2040 static int attach_recursive_mnt(struct mount *source_mnt,
2041 struct mount *dest_mnt,
2042 struct mountpoint *dest_mp,
2045 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2046 HLIST_HEAD(tree_list);
2047 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2048 struct mountpoint *smp;
2049 struct mount *child, *p;
2050 struct hlist_node *n;
2053 /* Preallocate a mountpoint in case the new mounts need
2054 * to be tucked under other mounts.
2056 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2058 return PTR_ERR(smp);
2060 /* Is there space to add these mounts to the mount namespace? */
2062 err = count_mounts(ns, source_mnt);
2067 if (IS_MNT_SHARED(dest_mnt)) {
2068 err = invent_group_ids(source_mnt, true);
2071 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2074 goto out_cleanup_ids;
2075 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2081 unhash_mnt(source_mnt);
2082 attach_mnt(source_mnt, dest_mnt, dest_mp);
2083 touch_mnt_namespace(source_mnt->mnt_ns);
2085 if (source_mnt->mnt_ns) {
2086 /* move from anon - the caller will destroy */
2087 list_del_init(&source_mnt->mnt_ns->list);
2089 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2090 commit_tree(source_mnt);
2093 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2095 hlist_del_init(&child->mnt_hash);
2096 q = __lookup_mnt(&child->mnt_parent->mnt,
2097 child->mnt_mountpoint);
2099 mnt_change_mountpoint(child, smp, q);
2100 /* Notice when we are propagating across user namespaces */
2101 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2102 lock_mnt_tree(child);
2103 child->mnt.mnt_flags &= ~MNT_LOCKED;
2106 put_mountpoint(smp);
2107 unlock_mount_hash();
2112 while (!hlist_empty(&tree_list)) {
2113 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2114 child->mnt_parent->mnt_ns->pending_mounts = 0;
2115 umount_tree(child, UMOUNT_SYNC);
2117 unlock_mount_hash();
2118 cleanup_group_ids(source_mnt, NULL);
2120 ns->pending_mounts = 0;
2122 read_seqlock_excl(&mount_lock);
2123 put_mountpoint(smp);
2124 read_sequnlock_excl(&mount_lock);
2129 static struct mountpoint *lock_mount(struct path *path)
2131 struct vfsmount *mnt;
2132 struct dentry *dentry = path->dentry;
2134 inode_lock(dentry->d_inode);
2135 if (unlikely(cant_mount(dentry))) {
2136 inode_unlock(dentry->d_inode);
2137 return ERR_PTR(-ENOENT);
2140 mnt = lookup_mnt(path);
2142 struct mountpoint *mp = get_mountpoint(dentry);
2145 inode_unlock(dentry->d_inode);
2151 inode_unlock(path->dentry->d_inode);
2154 dentry = path->dentry = dget(mnt->mnt_root);
2158 static void unlock_mount(struct mountpoint *where)
2160 struct dentry *dentry = where->m_dentry;
2162 read_seqlock_excl(&mount_lock);
2163 put_mountpoint(where);
2164 read_sequnlock_excl(&mount_lock);
2167 inode_unlock(dentry->d_inode);
2170 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2172 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2175 if (d_is_dir(mp->m_dentry) !=
2176 d_is_dir(mnt->mnt.mnt_root))
2179 return attach_recursive_mnt(mnt, p, mp, false);
2183 * Sanity check the flags to change_mnt_propagation.
2186 static int flags_to_propagation_type(int ms_flags)
2188 int type = ms_flags & ~(MS_REC | MS_SILENT);
2190 /* Fail if any non-propagation flags are set */
2191 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2193 /* Only one propagation flag should be set */
2194 if (!is_power_of_2(type))
2200 * recursively change the type of the mountpoint.
2202 static int do_change_type(struct path *path, int ms_flags)
2205 struct mount *mnt = real_mount(path->mnt);
2206 int recurse = ms_flags & MS_REC;
2210 if (path->dentry != path->mnt->mnt_root)
2213 type = flags_to_propagation_type(ms_flags);
2218 if (type == MS_SHARED) {
2219 err = invent_group_ids(mnt, recurse);
2225 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2226 change_mnt_propagation(m, type);
2227 unlock_mount_hash();
2234 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2236 struct mount *child;
2237 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2238 if (!is_subdir(child->mnt_mountpoint, dentry))
2241 if (child->mnt.mnt_flags & MNT_LOCKED)
2247 static struct mount *__do_loopback(struct path *old_path, int recurse)
2249 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2251 if (IS_MNT_UNBINDABLE(old))
2254 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2257 if (!recurse && has_locked_children(old, old_path->dentry))
2261 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2263 mnt = clone_mnt(old, old_path->dentry, 0);
2266 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2272 * do loopback mount.
2274 static int do_loopback(struct path *path, const char *old_name,
2277 struct path old_path;
2278 struct mount *mnt = NULL, *parent;
2279 struct mountpoint *mp;
2281 if (!old_name || !*old_name)
2283 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2288 if (mnt_ns_loop(old_path.dentry))
2291 mp = lock_mount(path);
2297 parent = real_mount(path->mnt);
2298 if (!check_mnt(parent))
2301 mnt = __do_loopback(&old_path, recurse);
2307 err = graft_tree(mnt, parent, mp);
2310 umount_tree(mnt, UMOUNT_SYNC);
2311 unlock_mount_hash();
2316 path_put(&old_path);
2320 static struct file *open_detached_copy(struct path *path, bool recursive)
2322 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2323 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2324 struct mount *mnt, *p;
2328 return ERR_CAST(ns);
2331 mnt = __do_loopback(path, recursive);
2335 return ERR_CAST(mnt);
2339 for (p = mnt; p; p = next_mnt(p, mnt)) {
2344 list_add_tail(&ns->list, &mnt->mnt_list);
2346 unlock_mount_hash();
2350 path->mnt = &mnt->mnt;
2351 file = dentry_open(path, O_PATH, current_cred());
2353 dissolve_on_fput(path->mnt);
2355 file->f_mode |= FMODE_NEED_UNMOUNT;
2359 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2363 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2364 bool detached = flags & OPEN_TREE_CLONE;
2368 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2370 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2371 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2375 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2378 if (flags & AT_NO_AUTOMOUNT)
2379 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2380 if (flags & AT_SYMLINK_NOFOLLOW)
2381 lookup_flags &= ~LOOKUP_FOLLOW;
2382 if (flags & AT_EMPTY_PATH)
2383 lookup_flags |= LOOKUP_EMPTY;
2385 if (detached && !may_mount())
2388 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2392 error = user_path_at(dfd, filename, lookup_flags, &path);
2393 if (unlikely(error)) {
2394 file = ERR_PTR(error);
2397 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2399 file = dentry_open(&path, O_PATH, current_cred());
2404 return PTR_ERR(file);
2406 fd_install(fd, file);
2411 * Don't allow locked mount flags to be cleared.
2413 * No locks need to be held here while testing the various MNT_LOCK
2414 * flags because those flags can never be cleared once they are set.
2416 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2418 unsigned int fl = mnt->mnt.mnt_flags;
2420 if ((fl & MNT_LOCK_READONLY) &&
2421 !(mnt_flags & MNT_READONLY))
2424 if ((fl & MNT_LOCK_NODEV) &&
2425 !(mnt_flags & MNT_NODEV))
2428 if ((fl & MNT_LOCK_NOSUID) &&
2429 !(mnt_flags & MNT_NOSUID))
2432 if ((fl & MNT_LOCK_NOEXEC) &&
2433 !(mnt_flags & MNT_NOEXEC))
2436 if ((fl & MNT_LOCK_ATIME) &&
2437 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2443 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2445 bool readonly_request = (mnt_flags & MNT_READONLY);
2447 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2450 if (readonly_request)
2451 return mnt_make_readonly(mnt);
2453 return __mnt_unmake_readonly(mnt);
2457 * Update the user-settable attributes on a mount. The caller must hold
2458 * sb->s_umount for writing.
2460 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2463 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2464 mnt->mnt.mnt_flags = mnt_flags;
2465 touch_mnt_namespace(mnt->mnt_ns);
2466 unlock_mount_hash();
2469 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2471 struct super_block *sb = mnt->mnt_sb;
2473 if (!__mnt_is_readonly(mnt) &&
2474 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2475 char *buf = (char *)__get_free_page(GFP_KERNEL);
2476 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2479 time64_to_tm(sb->s_time_max, 0, &tm);
2481 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2483 is_mounted(mnt) ? "remounted" : "mounted",
2485 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2487 free_page((unsigned long)buf);
2492 * Handle reconfiguration of the mountpoint only without alteration of the
2493 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2496 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2498 struct super_block *sb = path->mnt->mnt_sb;
2499 struct mount *mnt = real_mount(path->mnt);
2502 if (!check_mnt(mnt))
2505 if (path->dentry != mnt->mnt.mnt_root)
2508 if (!can_change_locked_flags(mnt, mnt_flags))
2511 down_write(&sb->s_umount);
2512 ret = change_mount_ro_state(mnt, mnt_flags);
2514 set_mount_attributes(mnt, mnt_flags);
2515 up_write(&sb->s_umount);
2517 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2523 * change filesystem flags. dir should be a physical root of filesystem.
2524 * If you've mounted a non-root directory somewhere and want to do remount
2525 * on it - tough luck.
2527 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2528 int mnt_flags, void *data)
2531 struct super_block *sb = path->mnt->mnt_sb;
2532 struct mount *mnt = real_mount(path->mnt);
2533 struct fs_context *fc;
2535 if (!check_mnt(mnt))
2538 if (path->dentry != path->mnt->mnt_root)
2541 if (!can_change_locked_flags(mnt, mnt_flags))
2544 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2548 err = parse_monolithic_mount_data(fc, data);
2550 down_write(&sb->s_umount);
2552 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2553 err = reconfigure_super(fc);
2555 set_mount_attributes(mnt, mnt_flags);
2557 up_write(&sb->s_umount);
2560 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2566 static inline int tree_contains_unbindable(struct mount *mnt)
2569 for (p = mnt; p; p = next_mnt(p, mnt)) {
2570 if (IS_MNT_UNBINDABLE(p))
2577 * Check that there aren't references to earlier/same mount namespaces in the
2578 * specified subtree. Such references can act as pins for mount namespaces
2579 * that aren't checked by the mount-cycle checking code, thereby allowing
2580 * cycles to be made.
2582 static bool check_for_nsfs_mounts(struct mount *subtree)
2588 for (p = subtree; p; p = next_mnt(p, subtree))
2589 if (mnt_ns_loop(p->mnt.mnt_root))
2594 unlock_mount_hash();
2598 static int do_move_mount(struct path *old_path, struct path *new_path)
2600 struct mnt_namespace *ns;
2603 struct mount *parent;
2604 struct mountpoint *mp, *old_mp;
2608 mp = lock_mount(new_path);
2612 old = real_mount(old_path->mnt);
2613 p = real_mount(new_path->mnt);
2614 parent = old->mnt_parent;
2615 attached = mnt_has_parent(old);
2616 old_mp = old->mnt_mp;
2620 /* The mountpoint must be in our namespace. */
2624 /* The thing moved must be mounted... */
2625 if (!is_mounted(&old->mnt))
2628 /* ... and either ours or the root of anon namespace */
2629 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2632 if (old->mnt.mnt_flags & MNT_LOCKED)
2635 if (old_path->dentry != old_path->mnt->mnt_root)
2638 if (d_is_dir(new_path->dentry) !=
2639 d_is_dir(old_path->dentry))
2642 * Don't move a mount residing in a shared parent.
2644 if (attached && IS_MNT_SHARED(parent))
2647 * Don't move a mount tree containing unbindable mounts to a destination
2648 * mount which is shared.
2650 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2653 if (!check_for_nsfs_mounts(old))
2655 for (; mnt_has_parent(p); p = p->mnt_parent)
2659 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2664 /* if the mount is moved, it should no longer be expire
2666 list_del_init(&old->mnt_expire);
2668 put_mountpoint(old_mp);
2673 mntput_no_expire(parent);
2680 static int do_move_mount_old(struct path *path, const char *old_name)
2682 struct path old_path;
2685 if (!old_name || !*old_name)
2688 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2692 err = do_move_mount(&old_path, path);
2693 path_put(&old_path);
2698 * add a mount into a namespace's mount tree
2700 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2702 struct mountpoint *mp;
2703 struct mount *parent;
2706 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2708 mp = lock_mount(path);
2712 parent = real_mount(path->mnt);
2714 if (unlikely(!check_mnt(parent))) {
2715 /* that's acceptable only for automounts done in private ns */
2716 if (!(mnt_flags & MNT_SHRINKABLE))
2718 /* ... and for those we'd better have mountpoint still alive */
2719 if (!parent->mnt_ns)
2723 /* Refuse the same filesystem on the same mount point */
2725 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2726 path->mnt->mnt_root == path->dentry)
2730 if (d_is_symlink(newmnt->mnt.mnt_root))
2733 newmnt->mnt.mnt_flags = mnt_flags;
2734 err = graft_tree(newmnt, parent, mp);
2741 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2744 * Create a new mount using a superblock configuration and request it
2745 * be added to the namespace tree.
2747 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2748 unsigned int mnt_flags)
2750 struct vfsmount *mnt;
2751 struct super_block *sb = fc->root->d_sb;
2754 error = security_sb_kern_mount(sb);
2755 if (!error && mount_too_revealing(sb, &mnt_flags))
2758 if (unlikely(error)) {
2763 up_write(&sb->s_umount);
2765 mnt = vfs_create_mount(fc);
2767 return PTR_ERR(mnt);
2769 mnt_warn_timestamp_expiry(mountpoint, mnt);
2771 error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags);
2778 * create a new mount for userspace and request it to be added into the
2781 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2782 int mnt_flags, const char *name, void *data)
2784 struct file_system_type *type;
2785 struct fs_context *fc;
2786 const char *subtype = NULL;
2792 type = get_fs_type(fstype);
2796 if (type->fs_flags & FS_HAS_SUBTYPE) {
2797 subtype = strchr(fstype, '.');
2801 put_filesystem(type);
2807 fc = fs_context_for_mount(type, sb_flags);
2808 put_filesystem(type);
2813 err = vfs_parse_fs_string(fc, "subtype",
2814 subtype, strlen(subtype));
2816 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2818 err = parse_monolithic_mount_data(fc, data);
2819 if (!err && !mount_capable(fc))
2822 err = vfs_get_tree(fc);
2824 err = do_new_mount_fc(fc, path, mnt_flags);
2830 int finish_automount(struct vfsmount *m, struct path *path)
2832 struct mount *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 == path->dentry) {
2845 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2849 /* remove m from any expiration list it may be on */
2850 if (!list_empty(&mnt->mnt_expire)) {
2852 list_del_init(&mnt->mnt_expire);
2861 * mnt_set_expiry - Put a mount on an expiration list
2862 * @mnt: The mount to list.
2863 * @expiry_list: The list to add the mount to.
2865 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2869 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2873 EXPORT_SYMBOL(mnt_set_expiry);
2876 * process a list of expirable mountpoints with the intent of discarding any
2877 * mountpoints that aren't in use and haven't been touched since last we came
2880 void mark_mounts_for_expiry(struct list_head *mounts)
2882 struct mount *mnt, *next;
2883 LIST_HEAD(graveyard);
2885 if (list_empty(mounts))
2891 /* extract from the expiration list every vfsmount that matches the
2892 * following criteria:
2893 * - only referenced by its parent vfsmount
2894 * - still marked for expiry (marked on the last call here; marks are
2895 * cleared by mntput())
2897 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2898 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2899 propagate_mount_busy(mnt, 1))
2901 list_move(&mnt->mnt_expire, &graveyard);
2903 while (!list_empty(&graveyard)) {
2904 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2905 touch_mnt_namespace(mnt->mnt_ns);
2906 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2908 unlock_mount_hash();
2912 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2915 * Ripoff of 'select_parent()'
2917 * search the list of submounts for a given mountpoint, and move any
2918 * shrinkable submounts to the 'graveyard' list.
2920 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2922 struct mount *this_parent = parent;
2923 struct list_head *next;
2927 next = this_parent->mnt_mounts.next;
2929 while (next != &this_parent->mnt_mounts) {
2930 struct list_head *tmp = next;
2931 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2934 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2937 * Descend a level if the d_mounts list is non-empty.
2939 if (!list_empty(&mnt->mnt_mounts)) {
2944 if (!propagate_mount_busy(mnt, 1)) {
2945 list_move_tail(&mnt->mnt_expire, graveyard);
2950 * All done at this level ... ascend and resume the search
2952 if (this_parent != parent) {
2953 next = this_parent->mnt_child.next;
2954 this_parent = this_parent->mnt_parent;
2961 * process a list of expirable mountpoints with the intent of discarding any
2962 * submounts of a specific parent mountpoint
2964 * mount_lock must be held for write
2966 static void shrink_submounts(struct mount *mnt)
2968 LIST_HEAD(graveyard);
2971 /* extract submounts of 'mountpoint' from the expiration list */
2972 while (select_submounts(mnt, &graveyard)) {
2973 while (!list_empty(&graveyard)) {
2974 m = list_first_entry(&graveyard, struct mount,
2976 touch_mnt_namespace(m->mnt_ns);
2977 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2983 * Some copy_from_user() implementations do not return the exact number of
2984 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2985 * Note that this function differs from copy_from_user() in that it will oops
2986 * on bad values of `to', rather than returning a short copy.
2988 static long exact_copy_from_user(void *to, const void __user * from,
2992 const char __user *f = from;
2995 if (!access_ok(from, n))
2999 if (__get_user(c, f)) {
3010 void *copy_mount_options(const void __user * data)
3019 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3021 return ERR_PTR(-ENOMEM);
3023 /* We only care that *some* data at the address the user
3024 * gave us is valid. Just in case, we'll zero
3025 * the remainder of the page.
3027 /* copy_from_user cannot cross TASK_SIZE ! */
3028 size = TASK_SIZE - (unsigned long)untagged_addr(data);
3029 if (size > PAGE_SIZE)
3032 i = size - exact_copy_from_user(copy, data, size);
3035 return ERR_PTR(-EFAULT);
3038 memset(copy + i, 0, PAGE_SIZE - i);
3042 char *copy_mount_string(const void __user *data)
3044 return data ? strndup_user(data, PATH_MAX) : NULL;
3048 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3049 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3051 * data is a (void *) that can point to any structure up to
3052 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3053 * information (or be NULL).
3055 * Pre-0.97 versions of mount() didn't have a flags word.
3056 * When the flags word was introduced its top half was required
3057 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3058 * Therefore, if this magic number is present, it carries no information
3059 * and must be discarded.
3061 long do_mount(const char *dev_name, const char __user *dir_name,
3062 const char *type_page, unsigned long flags, void *data_page)
3065 unsigned int mnt_flags = 0, sb_flags;
3069 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3070 flags &= ~MS_MGC_MSK;
3072 /* Basic sanity checks */
3074 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3076 if (flags & MS_NOUSER)
3079 /* ... and get the mountpoint */
3080 retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3084 retval = security_sb_mount(dev_name, &path,
3085 type_page, flags, data_page);
3086 if (!retval && !may_mount())
3088 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
3093 /* Default to relatime unless overriden */
3094 if (!(flags & MS_NOATIME))
3095 mnt_flags |= MNT_RELATIME;
3097 /* Separate the per-mountpoint flags */
3098 if (flags & MS_NOSUID)
3099 mnt_flags |= MNT_NOSUID;
3100 if (flags & MS_NODEV)
3101 mnt_flags |= MNT_NODEV;
3102 if (flags & MS_NOEXEC)
3103 mnt_flags |= MNT_NOEXEC;
3104 if (flags & MS_NOATIME)
3105 mnt_flags |= MNT_NOATIME;
3106 if (flags & MS_NODIRATIME)
3107 mnt_flags |= MNT_NODIRATIME;
3108 if (flags & MS_STRICTATIME)
3109 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3110 if (flags & MS_RDONLY)
3111 mnt_flags |= MNT_READONLY;
3113 /* The default atime for remount is preservation */
3114 if ((flags & MS_REMOUNT) &&
3115 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3116 MS_STRICTATIME)) == 0)) {
3117 mnt_flags &= ~MNT_ATIME_MASK;
3118 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
3121 sb_flags = flags & (SB_RDONLY |
3130 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3131 retval = do_reconfigure_mnt(&path, mnt_flags);
3132 else if (flags & MS_REMOUNT)
3133 retval = do_remount(&path, flags, sb_flags, mnt_flags,
3135 else if (flags & MS_BIND)
3136 retval = do_loopback(&path, dev_name, flags & MS_REC);
3137 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3138 retval = do_change_type(&path, flags);
3139 else if (flags & MS_MOVE)
3140 retval = do_move_mount_old(&path, dev_name);
3142 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
3143 dev_name, data_page);
3149 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3151 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3154 static void dec_mnt_namespaces(struct ucounts *ucounts)
3156 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3159 static void free_mnt_ns(struct mnt_namespace *ns)
3161 if (!is_anon_ns(ns))
3162 ns_free_inum(&ns->ns);
3163 dec_mnt_namespaces(ns->ucounts);
3164 put_user_ns(ns->user_ns);
3169 * Assign a sequence number so we can detect when we attempt to bind
3170 * mount a reference to an older mount namespace into the current
3171 * mount namespace, preventing reference counting loops. A 64bit
3172 * number incrementing at 10Ghz will take 12,427 years to wrap which
3173 * is effectively never, so we can ignore the possibility.
3175 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3177 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3179 struct mnt_namespace *new_ns;
3180 struct ucounts *ucounts;
3183 ucounts = inc_mnt_namespaces(user_ns);
3185 return ERR_PTR(-ENOSPC);
3187 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3189 dec_mnt_namespaces(ucounts);
3190 return ERR_PTR(-ENOMEM);
3193 ret = ns_alloc_inum(&new_ns->ns);
3196 dec_mnt_namespaces(ucounts);
3197 return ERR_PTR(ret);
3200 new_ns->ns.ops = &mntns_operations;
3202 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3203 atomic_set(&new_ns->count, 1);
3204 INIT_LIST_HEAD(&new_ns->list);
3205 init_waitqueue_head(&new_ns->poll);
3206 new_ns->user_ns = get_user_ns(user_ns);
3207 new_ns->ucounts = ucounts;
3212 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3213 struct user_namespace *user_ns, struct fs_struct *new_fs)
3215 struct mnt_namespace *new_ns;
3216 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3217 struct mount *p, *q;
3224 if (likely(!(flags & CLONE_NEWNS))) {
3231 new_ns = alloc_mnt_ns(user_ns, false);
3236 /* First pass: copy the tree topology */
3237 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3238 if (user_ns != ns->user_ns)
3239 copy_flags |= CL_SHARED_TO_SLAVE;
3240 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3243 free_mnt_ns(new_ns);
3244 return ERR_CAST(new);
3246 if (user_ns != ns->user_ns) {
3249 unlock_mount_hash();
3252 list_add_tail(&new_ns->list, &new->mnt_list);
3255 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3256 * as belonging to new namespace. We have already acquired a private
3257 * fs_struct, so tsk->fs->lock is not needed.
3265 if (&p->mnt == new_fs->root.mnt) {
3266 new_fs->root.mnt = mntget(&q->mnt);
3269 if (&p->mnt == new_fs->pwd.mnt) {
3270 new_fs->pwd.mnt = mntget(&q->mnt);
3274 p = next_mnt(p, old);
3275 q = next_mnt(q, new);
3278 while (p->mnt.mnt_root != q->mnt.mnt_root)
3279 p = next_mnt(p, old);
3291 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3293 struct mount *mnt = real_mount(m);
3294 struct mnt_namespace *ns;
3295 struct super_block *s;
3299 ns = alloc_mnt_ns(&init_user_ns, true);
3302 return ERR_CAST(ns);
3307 list_add(&mnt->mnt_list, &ns->list);
3309 err = vfs_path_lookup(m->mnt_root, m,
3310 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3315 return ERR_PTR(err);
3317 /* trade a vfsmount reference for active sb one */
3318 s = path.mnt->mnt_sb;
3319 atomic_inc(&s->s_active);
3321 /* lock the sucker */
3322 down_write(&s->s_umount);
3323 /* ... and return the root of (sub)tree on it */
3326 EXPORT_SYMBOL(mount_subtree);
3328 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3329 char __user *, type, unsigned long, flags, void __user *, data)
3336 kernel_type = copy_mount_string(type);
3337 ret = PTR_ERR(kernel_type);
3338 if (IS_ERR(kernel_type))
3341 kernel_dev = copy_mount_string(dev_name);
3342 ret = PTR_ERR(kernel_dev);
3343 if (IS_ERR(kernel_dev))
3346 options = copy_mount_options(data);
3347 ret = PTR_ERR(options);
3348 if (IS_ERR(options))
3351 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3363 * Create a kernel mount representation for a new, prepared superblock
3364 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3366 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3367 unsigned int, attr_flags)
3369 struct mnt_namespace *ns;
3370 struct fs_context *fc;
3372 struct path newmount;
3375 unsigned int mnt_flags = 0;
3381 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3384 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3389 MOUNT_ATTR_NODIRATIME))
3392 if (attr_flags & MOUNT_ATTR_RDONLY)
3393 mnt_flags |= MNT_READONLY;
3394 if (attr_flags & MOUNT_ATTR_NOSUID)
3395 mnt_flags |= MNT_NOSUID;
3396 if (attr_flags & MOUNT_ATTR_NODEV)
3397 mnt_flags |= MNT_NODEV;
3398 if (attr_flags & MOUNT_ATTR_NOEXEC)
3399 mnt_flags |= MNT_NOEXEC;
3400 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3401 mnt_flags |= MNT_NODIRATIME;
3403 switch (attr_flags & MOUNT_ATTR__ATIME) {
3404 case MOUNT_ATTR_STRICTATIME:
3406 case MOUNT_ATTR_NOATIME:
3407 mnt_flags |= MNT_NOATIME;
3409 case MOUNT_ATTR_RELATIME:
3410 mnt_flags |= MNT_RELATIME;
3421 if (f.file->f_op != &fscontext_fops)
3424 fc = f.file->private_data;
3426 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3430 /* There must be a valid superblock or we can't mount it */
3436 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3437 pr_warn("VFS: Mount too revealing\n");
3442 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3446 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3449 newmount.mnt = vfs_create_mount(fc);
3450 if (IS_ERR(newmount.mnt)) {
3451 ret = PTR_ERR(newmount.mnt);
3454 newmount.dentry = dget(fc->root);
3455 newmount.mnt->mnt_flags = mnt_flags;
3457 /* We've done the mount bit - now move the file context into more or
3458 * less the same state as if we'd done an fspick(). We don't want to
3459 * do any memory allocation or anything like that at this point as we
3460 * don't want to have to handle any errors incurred.
3462 vfs_clean_context(fc);
3464 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3469 mnt = real_mount(newmount.mnt);
3473 list_add(&mnt->mnt_list, &ns->list);
3474 mntget(newmount.mnt);
3476 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3477 * it, not just simply put it.
3479 file = dentry_open(&newmount, O_PATH, fc->cred);
3481 dissolve_on_fput(newmount.mnt);
3482 ret = PTR_ERR(file);
3485 file->f_mode |= FMODE_NEED_UNMOUNT;
3487 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3489 fd_install(ret, file);
3494 path_put(&newmount);
3496 mutex_unlock(&fc->uapi_mutex);
3503 * Move a mount from one place to another. In combination with
3504 * fsopen()/fsmount() this is used to install a new mount and in combination
3505 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3508 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3510 SYSCALL_DEFINE5(move_mount,
3511 int, from_dfd, const char __user *, from_pathname,
3512 int, to_dfd, const char __user *, to_pathname,
3513 unsigned int, flags)
3515 struct path from_path, to_path;
3516 unsigned int lflags;
3522 if (flags & ~MOVE_MOUNT__MASK)
3525 /* If someone gives a pathname, they aren't permitted to move
3526 * from an fd that requires unmount as we can't get at the flag
3527 * to clear it afterwards.
3530 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3531 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3532 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3534 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3539 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3540 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3541 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3543 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3547 ret = security_move_mount(&from_path, &to_path);
3551 ret = do_move_mount(&from_path, &to_path);
3556 path_put(&from_path);
3561 * Return true if path is reachable from root
3563 * namespace_sem or mount_lock is held
3565 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3566 const struct path *root)
3568 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3569 dentry = mnt->mnt_mountpoint;
3570 mnt = mnt->mnt_parent;
3572 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3575 bool path_is_under(const struct path *path1, const struct path *path2)
3578 read_seqlock_excl(&mount_lock);
3579 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3580 read_sequnlock_excl(&mount_lock);
3583 EXPORT_SYMBOL(path_is_under);
3586 * pivot_root Semantics:
3587 * Moves the root file system of the current process to the directory put_old,
3588 * makes new_root as the new root file system of the current process, and sets
3589 * root/cwd of all processes which had them on the current root to new_root.
3592 * The new_root and put_old must be directories, and must not be on the
3593 * same file system as the current process root. The put_old must be
3594 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3595 * pointed to by put_old must yield the same directory as new_root. No other
3596 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3598 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3599 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3600 * in this situation.
3603 * - we don't move root/cwd if they are not at the root (reason: if something
3604 * cared enough to change them, it's probably wrong to force them elsewhere)
3605 * - it's okay to pick a root that isn't the root of a file system, e.g.
3606 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3607 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3610 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3611 const char __user *, put_old)
3613 struct path new, old, root;
3614 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3615 struct mountpoint *old_mp, *root_mp;
3621 error = user_path_at(AT_FDCWD, new_root,
3622 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3626 error = user_path_at(AT_FDCWD, put_old,
3627 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3631 error = security_sb_pivotroot(&old, &new);
3635 get_fs_root(current->fs, &root);
3636 old_mp = lock_mount(&old);
3637 error = PTR_ERR(old_mp);
3642 new_mnt = real_mount(new.mnt);
3643 root_mnt = real_mount(root.mnt);
3644 old_mnt = real_mount(old.mnt);
3645 ex_parent = new_mnt->mnt_parent;
3646 root_parent = root_mnt->mnt_parent;
3647 if (IS_MNT_SHARED(old_mnt) ||
3648 IS_MNT_SHARED(ex_parent) ||
3649 IS_MNT_SHARED(root_parent))
3651 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3653 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3656 if (d_unlinked(new.dentry))
3659 if (new_mnt == root_mnt || old_mnt == root_mnt)
3660 goto out4; /* loop, on the same file system */
3662 if (root.mnt->mnt_root != root.dentry)
3663 goto out4; /* not a mountpoint */
3664 if (!mnt_has_parent(root_mnt))
3665 goto out4; /* not attached */
3666 if (new.mnt->mnt_root != new.dentry)
3667 goto out4; /* not a mountpoint */
3668 if (!mnt_has_parent(new_mnt))
3669 goto out4; /* not attached */
3670 /* make sure we can reach put_old from new_root */
3671 if (!is_path_reachable(old_mnt, old.dentry, &new))
3673 /* make certain new is below the root */
3674 if (!is_path_reachable(new_mnt, new.dentry, &root))
3677 umount_mnt(new_mnt);
3678 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3679 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3680 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3681 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3683 /* mount old root on put_old */
3684 attach_mnt(root_mnt, old_mnt, old_mp);
3685 /* mount new_root on / */
3686 attach_mnt(new_mnt, root_parent, root_mp);
3687 mnt_add_count(root_parent, -1);
3688 touch_mnt_namespace(current->nsproxy->mnt_ns);
3689 /* A moved mount should not expire automatically */
3690 list_del_init(&new_mnt->mnt_expire);
3691 put_mountpoint(root_mp);
3692 unlock_mount_hash();
3693 chroot_fs_refs(&root, &new);
3696 unlock_mount(old_mp);
3698 mntput_no_expire(ex_parent);
3709 static void __init init_mount_tree(void)
3711 struct vfsmount *mnt;
3713 struct mnt_namespace *ns;
3716 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3718 panic("Can't create rootfs");
3720 ns = alloc_mnt_ns(&init_user_ns, false);
3722 panic("Can't allocate initial namespace");
3723 m = real_mount(mnt);
3727 list_add(&m->mnt_list, &ns->list);
3728 init_task.nsproxy->mnt_ns = ns;
3732 root.dentry = mnt->mnt_root;
3733 mnt->mnt_flags |= MNT_LOCKED;
3735 set_fs_pwd(current->fs, &root);
3736 set_fs_root(current->fs, &root);
3739 void __init mnt_init(void)
3743 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3744 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3746 mount_hashtable = alloc_large_system_hash("Mount-cache",
3747 sizeof(struct hlist_head),
3750 &m_hash_shift, &m_hash_mask, 0, 0);
3751 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3752 sizeof(struct hlist_head),
3755 &mp_hash_shift, &mp_hash_mask, 0, 0);
3757 if (!mount_hashtable || !mountpoint_hashtable)
3758 panic("Failed to allocate mount hash table\n");
3764 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3766 fs_kobj = kobject_create_and_add("fs", NULL);
3768 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3774 void put_mnt_ns(struct mnt_namespace *ns)
3776 if (!atomic_dec_and_test(&ns->count))
3778 drop_collected_mounts(&ns->root->mnt);
3782 struct vfsmount *kern_mount(struct file_system_type *type)
3784 struct vfsmount *mnt;
3785 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3788 * it is a longterm mount, don't release mnt until
3789 * we unmount before file sys is unregistered
3791 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3795 EXPORT_SYMBOL_GPL(kern_mount);
3797 void kern_unmount(struct vfsmount *mnt)
3799 /* release long term mount so mount point can be released */
3800 if (!IS_ERR_OR_NULL(mnt)) {
3801 real_mount(mnt)->mnt_ns = NULL;
3802 synchronize_rcu(); /* yecchhh... */
3806 EXPORT_SYMBOL(kern_unmount);
3808 bool our_mnt(struct vfsmount *mnt)
3810 return check_mnt(real_mount(mnt));
3813 bool current_chrooted(void)
3815 /* Does the current process have a non-standard root */
3816 struct path ns_root;
3817 struct path fs_root;
3820 /* Find the namespace root */
3821 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3822 ns_root.dentry = ns_root.mnt->mnt_root;
3824 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3827 get_fs_root(current->fs, &fs_root);
3829 chrooted = !path_equal(&fs_root, &ns_root);
3837 static bool mnt_already_visible(struct mnt_namespace *ns,
3838 const struct super_block *sb,
3841 int new_flags = *new_mnt_flags;
3843 bool visible = false;
3845 down_read(&namespace_sem);
3846 list_for_each_entry(mnt, &ns->list, mnt_list) {
3847 struct mount *child;
3850 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3853 /* This mount is not fully visible if it's root directory
3854 * is not the root directory of the filesystem.
3856 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3859 /* A local view of the mount flags */
3860 mnt_flags = mnt->mnt.mnt_flags;
3862 /* Don't miss readonly hidden in the superblock flags */
3863 if (sb_rdonly(mnt->mnt.mnt_sb))
3864 mnt_flags |= MNT_LOCK_READONLY;
3866 /* Verify the mount flags are equal to or more permissive
3867 * than the proposed new mount.
3869 if ((mnt_flags & MNT_LOCK_READONLY) &&
3870 !(new_flags & MNT_READONLY))
3872 if ((mnt_flags & MNT_LOCK_ATIME) &&
3873 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3876 /* This mount is not fully visible if there are any
3877 * locked child mounts that cover anything except for
3878 * empty directories.
3880 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3881 struct inode *inode = child->mnt_mountpoint->d_inode;
3882 /* Only worry about locked mounts */
3883 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3885 /* Is the directory permanetly empty? */
3886 if (!is_empty_dir_inode(inode))
3889 /* Preserve the locked attributes */
3890 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3897 up_read(&namespace_sem);
3901 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3903 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3904 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3905 unsigned long s_iflags;
3907 if (ns->user_ns == &init_user_ns)
3910 /* Can this filesystem be too revealing? */
3911 s_iflags = sb->s_iflags;
3912 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3915 if ((s_iflags & required_iflags) != required_iflags) {
3916 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3921 return !mnt_already_visible(ns, sb, new_mnt_flags);
3924 bool mnt_may_suid(struct vfsmount *mnt)
3927 * Foreign mounts (accessed via fchdir or through /proc
3928 * symlinks) are always treated as if they are nosuid. This
3929 * prevents namespaces from trusting potentially unsafe
3930 * suid/sgid bits, file caps, or security labels that originate
3931 * in other namespaces.
3933 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3934 current_in_userns(mnt->mnt_sb->s_user_ns);
3937 static struct ns_common *mntns_get(struct task_struct *task)
3939 struct ns_common *ns = NULL;
3940 struct nsproxy *nsproxy;
3943 nsproxy = task->nsproxy;
3945 ns = &nsproxy->mnt_ns->ns;
3946 get_mnt_ns(to_mnt_ns(ns));
3953 static void mntns_put(struct ns_common *ns)
3955 put_mnt_ns(to_mnt_ns(ns));
3958 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3960 struct fs_struct *fs = current->fs;
3961 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3965 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3966 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3967 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3970 if (is_anon_ns(mnt_ns))
3977 old_mnt_ns = nsproxy->mnt_ns;
3978 nsproxy->mnt_ns = mnt_ns;
3981 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3982 "/", LOOKUP_DOWN, &root);
3984 /* revert to old namespace */
3985 nsproxy->mnt_ns = old_mnt_ns;
3990 put_mnt_ns(old_mnt_ns);
3992 /* Update the pwd and root */
3993 set_fs_pwd(fs, &root);
3994 set_fs_root(fs, &root);
4000 static struct user_namespace *mntns_owner(struct ns_common *ns)
4002 return to_mnt_ns(ns)->user_ns;
4005 const struct proc_ns_operations mntns_operations = {
4007 .type = CLONE_NEWNS,
4010 .install = mntns_install,
4011 .owner = mntns_owner,