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 int mnt_get_count(struct mount *mnt)
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)
1129 if (likely(READ_ONCE(mnt->mnt_ns))) {
1131 * Since we don't do lock_mount_hash() here,
1132 * ->mnt_ns can change under us. However, if it's
1133 * non-NULL, then there's a reference that won't
1134 * be dropped until after an RCU delay done after
1135 * turning ->mnt_ns NULL. So if we observe it
1136 * non-NULL under rcu_read_lock(), the reference
1137 * we are dropping is not the final one.
1139 mnt_add_count(mnt, -1);
1145 * make sure that if __legitimize_mnt() has not seen us grab
1146 * mount_lock, we'll see their refcount increment here.
1149 mnt_add_count(mnt, -1);
1150 count = mnt_get_count(mnt);
1154 unlock_mount_hash();
1157 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1159 unlock_mount_hash();
1162 mnt->mnt.mnt_flags |= MNT_DOOMED;
1165 list_del(&mnt->mnt_instance);
1167 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1168 struct mount *p, *tmp;
1169 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1170 __put_mountpoint(unhash_mnt(p), &list);
1171 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1174 unlock_mount_hash();
1175 shrink_dentry_list(&list);
1177 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1178 struct task_struct *task = current;
1179 if (likely(!(task->flags & PF_KTHREAD))) {
1180 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1181 if (!task_work_add(task, &mnt->mnt_rcu, true))
1184 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1185 schedule_delayed_work(&delayed_mntput_work, 1);
1191 void mntput(struct vfsmount *mnt)
1194 struct mount *m = real_mount(mnt);
1195 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1196 if (unlikely(m->mnt_expiry_mark))
1197 m->mnt_expiry_mark = 0;
1198 mntput_no_expire(m);
1201 EXPORT_SYMBOL(mntput);
1203 struct vfsmount *mntget(struct vfsmount *mnt)
1206 mnt_add_count(real_mount(mnt), 1);
1209 EXPORT_SYMBOL(mntget);
1211 /* path_is_mountpoint() - Check if path is a mount in the current
1214 * d_mountpoint() can only be used reliably to establish if a dentry is
1215 * not mounted in any namespace and that common case is handled inline.
1216 * d_mountpoint() isn't aware of the possibility there may be multiple
1217 * mounts using a given dentry in a different namespace. This function
1218 * checks if the passed in path is a mountpoint rather than the dentry
1221 bool path_is_mountpoint(const struct path *path)
1226 if (!d_mountpoint(path->dentry))
1231 seq = read_seqbegin(&mount_lock);
1232 res = __path_is_mountpoint(path);
1233 } while (read_seqretry(&mount_lock, seq));
1238 EXPORT_SYMBOL(path_is_mountpoint);
1240 struct vfsmount *mnt_clone_internal(const struct path *path)
1243 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1246 p->mnt.mnt_flags |= MNT_INTERNAL;
1250 #ifdef CONFIG_PROC_FS
1251 /* iterator; we want it to have access to namespace_sem, thus here... */
1252 static void *m_start(struct seq_file *m, loff_t *pos)
1254 struct proc_mounts *p = m->private;
1256 down_read(&namespace_sem);
1257 if (p->cached_event == p->ns->event) {
1258 void *v = p->cached_mount;
1259 if (*pos == p->cached_index)
1261 if (*pos == p->cached_index + 1) {
1262 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1263 return p->cached_mount = v;
1267 p->cached_event = p->ns->event;
1268 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1269 p->cached_index = *pos;
1270 return p->cached_mount;
1273 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1275 struct proc_mounts *p = m->private;
1277 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1278 p->cached_index = *pos;
1279 return p->cached_mount;
1282 static void m_stop(struct seq_file *m, void *v)
1284 up_read(&namespace_sem);
1287 static int m_show(struct seq_file *m, void *v)
1289 struct proc_mounts *p = m->private;
1290 struct mount *r = list_entry(v, struct mount, mnt_list);
1291 return p->show(m, &r->mnt);
1294 const struct seq_operations mounts_op = {
1300 #endif /* CONFIG_PROC_FS */
1303 * may_umount_tree - check if a mount tree is busy
1304 * @mnt: root of mount tree
1306 * This is called to check if a tree of mounts has any
1307 * open files, pwds, chroots or sub mounts that are
1310 int may_umount_tree(struct vfsmount *m)
1312 struct mount *mnt = real_mount(m);
1313 int actual_refs = 0;
1314 int minimum_refs = 0;
1318 /* write lock needed for mnt_get_count */
1320 for (p = mnt; p; p = next_mnt(p, mnt)) {
1321 actual_refs += mnt_get_count(p);
1324 unlock_mount_hash();
1326 if (actual_refs > minimum_refs)
1332 EXPORT_SYMBOL(may_umount_tree);
1335 * may_umount - check if a mount point is busy
1336 * @mnt: root of mount
1338 * This is called to check if a mount point has any
1339 * open files, pwds, chroots or sub mounts. If the
1340 * mount has sub mounts this will return busy
1341 * regardless of whether the sub mounts are busy.
1343 * Doesn't take quota and stuff into account. IOW, in some cases it will
1344 * give false negatives. The main reason why it's here is that we need
1345 * a non-destructive way to look for easily umountable filesystems.
1347 int may_umount(struct vfsmount *mnt)
1350 down_read(&namespace_sem);
1352 if (propagate_mount_busy(real_mount(mnt), 2))
1354 unlock_mount_hash();
1355 up_read(&namespace_sem);
1359 EXPORT_SYMBOL(may_umount);
1361 static void namespace_unlock(void)
1363 struct hlist_head head;
1364 struct hlist_node *p;
1368 hlist_move_list(&unmounted, &head);
1369 list_splice_init(&ex_mountpoints, &list);
1371 up_write(&namespace_sem);
1373 shrink_dentry_list(&list);
1375 if (likely(hlist_empty(&head)))
1378 synchronize_rcu_expedited();
1380 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1381 hlist_del(&m->mnt_umount);
1386 static inline void namespace_lock(void)
1388 down_write(&namespace_sem);
1391 enum umount_tree_flags {
1393 UMOUNT_PROPAGATE = 2,
1394 UMOUNT_CONNECTED = 4,
1397 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1399 /* Leaving mounts connected is only valid for lazy umounts */
1400 if (how & UMOUNT_SYNC)
1403 /* A mount without a parent has nothing to be connected to */
1404 if (!mnt_has_parent(mnt))
1407 /* Because the reference counting rules change when mounts are
1408 * unmounted and connected, umounted mounts may not be
1409 * connected to mounted mounts.
1411 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1414 /* Has it been requested that the mount remain connected? */
1415 if (how & UMOUNT_CONNECTED)
1418 /* Is the mount locked such that it needs to remain connected? */
1419 if (IS_MNT_LOCKED(mnt))
1422 /* By default disconnect the mount */
1427 * mount_lock must be held
1428 * namespace_sem must be held for write
1430 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1432 LIST_HEAD(tmp_list);
1435 if (how & UMOUNT_PROPAGATE)
1436 propagate_mount_unlock(mnt);
1438 /* Gather the mounts to umount */
1439 for (p = mnt; p; p = next_mnt(p, mnt)) {
1440 p->mnt.mnt_flags |= MNT_UMOUNT;
1441 list_move(&p->mnt_list, &tmp_list);
1444 /* Hide the mounts from mnt_mounts */
1445 list_for_each_entry(p, &tmp_list, mnt_list) {
1446 list_del_init(&p->mnt_child);
1449 /* Add propogated mounts to the tmp_list */
1450 if (how & UMOUNT_PROPAGATE)
1451 propagate_umount(&tmp_list);
1453 while (!list_empty(&tmp_list)) {
1454 struct mnt_namespace *ns;
1456 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1457 list_del_init(&p->mnt_expire);
1458 list_del_init(&p->mnt_list);
1462 __touch_mnt_namespace(ns);
1465 if (how & UMOUNT_SYNC)
1466 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1468 disconnect = disconnect_mount(p, how);
1469 if (mnt_has_parent(p)) {
1470 mnt_add_count(p->mnt_parent, -1);
1472 /* Don't forget about p */
1473 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1478 change_mnt_propagation(p, MS_PRIVATE);
1480 hlist_add_head(&p->mnt_umount, &unmounted);
1484 static void shrink_submounts(struct mount *mnt);
1486 static int do_umount_root(struct super_block *sb)
1490 down_write(&sb->s_umount);
1491 if (!sb_rdonly(sb)) {
1492 struct fs_context *fc;
1494 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1499 ret = parse_monolithic_mount_data(fc, NULL);
1501 ret = reconfigure_super(fc);
1505 up_write(&sb->s_umount);
1509 static int do_umount(struct mount *mnt, int flags)
1511 struct super_block *sb = mnt->mnt.mnt_sb;
1514 retval = security_sb_umount(&mnt->mnt, flags);
1519 * Allow userspace to request a mountpoint be expired rather than
1520 * unmounting unconditionally. Unmount only happens if:
1521 * (1) the mark is already set (the mark is cleared by mntput())
1522 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1524 if (flags & MNT_EXPIRE) {
1525 if (&mnt->mnt == current->fs->root.mnt ||
1526 flags & (MNT_FORCE | MNT_DETACH))
1530 * probably don't strictly need the lock here if we examined
1531 * all race cases, but it's a slowpath.
1534 if (mnt_get_count(mnt) != 2) {
1535 unlock_mount_hash();
1538 unlock_mount_hash();
1540 if (!xchg(&mnt->mnt_expiry_mark, 1))
1545 * If we may have to abort operations to get out of this
1546 * mount, and they will themselves hold resources we must
1547 * allow the fs to do things. In the Unix tradition of
1548 * 'Gee thats tricky lets do it in userspace' the umount_begin
1549 * might fail to complete on the first run through as other tasks
1550 * must return, and the like. Thats for the mount program to worry
1551 * about for the moment.
1554 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1555 sb->s_op->umount_begin(sb);
1559 * No sense to grab the lock for this test, but test itself looks
1560 * somewhat bogus. Suggestions for better replacement?
1561 * Ho-hum... In principle, we might treat that as umount + switch
1562 * to rootfs. GC would eventually take care of the old vfsmount.
1563 * Actually it makes sense, especially if rootfs would contain a
1564 * /reboot - static binary that would close all descriptors and
1565 * call reboot(9). Then init(8) could umount root and exec /reboot.
1567 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1569 * Special case for "unmounting" root ...
1570 * we just try to remount it readonly.
1572 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1574 return do_umount_root(sb);
1580 /* Recheck MNT_LOCKED with the locks held */
1582 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1586 if (flags & MNT_DETACH) {
1587 if (!list_empty(&mnt->mnt_list))
1588 umount_tree(mnt, UMOUNT_PROPAGATE);
1591 shrink_submounts(mnt);
1593 if (!propagate_mount_busy(mnt, 2)) {
1594 if (!list_empty(&mnt->mnt_list))
1595 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1600 unlock_mount_hash();
1606 * __detach_mounts - lazily unmount all mounts on the specified dentry
1608 * During unlink, rmdir, and d_drop it is possible to loose the path
1609 * to an existing mountpoint, and wind up leaking the mount.
1610 * detach_mounts allows lazily unmounting those mounts instead of
1613 * The caller may hold dentry->d_inode->i_mutex.
1615 void __detach_mounts(struct dentry *dentry)
1617 struct mountpoint *mp;
1622 mp = lookup_mountpoint(dentry);
1627 while (!hlist_empty(&mp->m_list)) {
1628 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1629 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1631 hlist_add_head(&mnt->mnt_umount, &unmounted);
1633 else umount_tree(mnt, UMOUNT_CONNECTED);
1637 unlock_mount_hash();
1642 * Is the caller allowed to modify his namespace?
1644 static inline bool may_mount(void)
1646 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1649 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1650 static bool may_mandlock(void)
1652 pr_warn_once("======================================================\n"
1653 "WARNING: the mand mount option is being deprecated and\n"
1654 " will be removed in v5.15!\n"
1655 "======================================================\n");
1656 return capable(CAP_SYS_ADMIN);
1659 static inline bool may_mandlock(void)
1661 pr_warn("VFS: \"mand\" mount option not supported");
1667 * Now umount can handle mount points as well as block devices.
1668 * This is important for filesystems which use unnamed block devices.
1670 * We now support a flag for forced unmount like the other 'big iron'
1671 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1674 int ksys_umount(char __user *name, int flags)
1679 int lookup_flags = 0;
1681 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1687 if (!(flags & UMOUNT_NOFOLLOW))
1688 lookup_flags |= LOOKUP_FOLLOW;
1690 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1693 mnt = real_mount(path.mnt);
1695 if (path.dentry != path.mnt->mnt_root)
1697 if (!check_mnt(mnt))
1699 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1702 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1705 retval = do_umount(mnt, flags);
1707 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1709 mntput_no_expire(mnt);
1714 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1716 return ksys_umount(name, flags);
1719 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1722 * The 2.0 compatible umount. No flags.
1724 SYSCALL_DEFINE1(oldumount, char __user *, name)
1726 return ksys_umount(name, 0);
1731 static bool is_mnt_ns_file(struct dentry *dentry)
1733 /* Is this a proxy for a mount namespace? */
1734 return dentry->d_op == &ns_dentry_operations &&
1735 dentry->d_fsdata == &mntns_operations;
1738 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1740 return container_of(ns, struct mnt_namespace, ns);
1743 static bool mnt_ns_loop(struct dentry *dentry)
1745 /* Could bind mounting the mount namespace inode cause a
1746 * mount namespace loop?
1748 struct mnt_namespace *mnt_ns;
1749 if (!is_mnt_ns_file(dentry))
1752 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1753 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1756 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1759 struct mount *res, *p, *q, *r, *parent;
1761 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1762 return ERR_PTR(-EINVAL);
1764 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1765 return ERR_PTR(-EINVAL);
1767 res = q = clone_mnt(mnt, dentry, flag);
1771 q->mnt_mountpoint = mnt->mnt_mountpoint;
1774 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1776 if (!is_subdir(r->mnt_mountpoint, dentry))
1779 for (s = r; s; s = next_mnt(s, r)) {
1780 if (!(flag & CL_COPY_UNBINDABLE) &&
1781 IS_MNT_UNBINDABLE(s)) {
1782 if (s->mnt.mnt_flags & MNT_LOCKED) {
1783 /* Both unbindable and locked. */
1784 q = ERR_PTR(-EPERM);
1787 s = skip_mnt_tree(s);
1791 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1792 is_mnt_ns_file(s->mnt.mnt_root)) {
1793 s = skip_mnt_tree(s);
1796 while (p != s->mnt_parent) {
1802 q = clone_mnt(p, p->mnt.mnt_root, flag);
1806 list_add_tail(&q->mnt_list, &res->mnt_list);
1807 attach_mnt(q, parent, p->mnt_mp);
1808 unlock_mount_hash();
1815 umount_tree(res, UMOUNT_SYNC);
1816 unlock_mount_hash();
1821 /* Caller should check returned pointer for errors */
1823 struct vfsmount *collect_mounts(const struct path *path)
1827 if (!check_mnt(real_mount(path->mnt)))
1828 tree = ERR_PTR(-EINVAL);
1830 tree = copy_tree(real_mount(path->mnt), path->dentry,
1831 CL_COPY_ALL | CL_PRIVATE);
1834 return ERR_CAST(tree);
1838 static void free_mnt_ns(struct mnt_namespace *);
1839 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1841 void dissolve_on_fput(struct vfsmount *mnt)
1843 struct mnt_namespace *ns;
1846 ns = real_mount(mnt)->mnt_ns;
1849 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1853 unlock_mount_hash();
1859 void drop_collected_mounts(struct vfsmount *mnt)
1863 umount_tree(real_mount(mnt), 0);
1864 unlock_mount_hash();
1868 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1870 struct mount *child;
1872 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1873 if (!is_subdir(child->mnt_mountpoint, dentry))
1876 if (child->mnt.mnt_flags & MNT_LOCKED)
1883 * clone_private_mount - create a private clone of a path
1885 * This creates a new vfsmount, which will be the clone of @path. The new will
1886 * not be attached anywhere in the namespace and will be private (i.e. changes
1887 * to the originating mount won't be propagated into this).
1889 * Release with mntput().
1891 struct vfsmount *clone_private_mount(const struct path *path)
1893 struct mount *old_mnt = real_mount(path->mnt);
1894 struct mount *new_mnt;
1896 down_read(&namespace_sem);
1897 if (IS_MNT_UNBINDABLE(old_mnt))
1900 if (!check_mnt(old_mnt))
1903 if (has_locked_children(old_mnt, path->dentry))
1906 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1907 up_read(&namespace_sem);
1909 if (IS_ERR(new_mnt))
1910 return ERR_CAST(new_mnt);
1912 return &new_mnt->mnt;
1915 up_read(&namespace_sem);
1916 return ERR_PTR(-EINVAL);
1918 EXPORT_SYMBOL_GPL(clone_private_mount);
1920 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1921 struct vfsmount *root)
1924 int res = f(root, arg);
1927 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1928 res = f(&mnt->mnt, arg);
1935 static void lock_mnt_tree(struct mount *mnt)
1939 for (p = mnt; p; p = next_mnt(p, mnt)) {
1940 int flags = p->mnt.mnt_flags;
1941 /* Don't allow unprivileged users to change mount flags */
1942 flags |= MNT_LOCK_ATIME;
1944 if (flags & MNT_READONLY)
1945 flags |= MNT_LOCK_READONLY;
1947 if (flags & MNT_NODEV)
1948 flags |= MNT_LOCK_NODEV;
1950 if (flags & MNT_NOSUID)
1951 flags |= MNT_LOCK_NOSUID;
1953 if (flags & MNT_NOEXEC)
1954 flags |= MNT_LOCK_NOEXEC;
1955 /* Don't allow unprivileged users to reveal what is under a mount */
1956 if (list_empty(&p->mnt_expire))
1957 flags |= MNT_LOCKED;
1958 p->mnt.mnt_flags = flags;
1962 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1966 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1967 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1968 mnt_release_group_id(p);
1972 static int invent_group_ids(struct mount *mnt, bool recurse)
1976 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1977 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1978 int err = mnt_alloc_group_id(p);
1980 cleanup_group_ids(mnt, p);
1989 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1991 unsigned int max = READ_ONCE(sysctl_mount_max);
1992 unsigned int mounts = 0, old, pending, sum;
1995 for (p = mnt; p; p = next_mnt(p, mnt))
1999 pending = ns->pending_mounts;
2000 sum = old + pending;
2004 (mounts > (max - sum)))
2007 ns->pending_mounts = pending + mounts;
2012 * @source_mnt : mount tree to be attached
2013 * @nd : place the mount tree @source_mnt is attached
2014 * @parent_nd : if non-null, detach the source_mnt from its parent and
2015 * store the parent mount and mountpoint dentry.
2016 * (done when source_mnt is moved)
2018 * NOTE: in the table below explains the semantics when a source mount
2019 * of a given type is attached to a destination mount of a given type.
2020 * ---------------------------------------------------------------------------
2021 * | BIND MOUNT OPERATION |
2022 * |**************************************************************************
2023 * | source-->| shared | private | slave | unbindable |
2027 * |**************************************************************************
2028 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2030 * |non-shared| shared (+) | private | slave (*) | invalid |
2031 * ***************************************************************************
2032 * A bind operation clones the source mount and mounts the clone on the
2033 * destination mount.
2035 * (++) the cloned mount is propagated to all the mounts in the propagation
2036 * tree of the destination mount and the cloned mount is added to
2037 * the peer group of the source mount.
2038 * (+) the cloned mount is created under the destination mount and is marked
2039 * as shared. The cloned mount is added to the peer group of the source
2041 * (+++) the mount is propagated to all the mounts in the propagation tree
2042 * of the destination mount and the cloned mount is made slave
2043 * of the same master as that of the source mount. The cloned mount
2044 * is marked as 'shared and slave'.
2045 * (*) the cloned mount is made a slave of the same master as that of the
2048 * ---------------------------------------------------------------------------
2049 * | MOVE MOUNT OPERATION |
2050 * |**************************************************************************
2051 * | source-->| shared | private | slave | unbindable |
2055 * |**************************************************************************
2056 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2058 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2059 * ***************************************************************************
2061 * (+) the mount is moved to the destination. And is then propagated to
2062 * all the mounts in the propagation tree of the destination mount.
2063 * (+*) the mount is moved to the destination.
2064 * (+++) the mount is moved to the destination and is then propagated to
2065 * all the mounts belonging to the destination mount's propagation tree.
2066 * the mount is marked as 'shared and slave'.
2067 * (*) the mount continues to be a slave at the new location.
2069 * if the source mount is a tree, the operations explained above is
2070 * applied to each mount in the tree.
2071 * Must be called without spinlocks held, since this function can sleep
2074 static int attach_recursive_mnt(struct mount *source_mnt,
2075 struct mount *dest_mnt,
2076 struct mountpoint *dest_mp,
2079 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2080 HLIST_HEAD(tree_list);
2081 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2082 struct mountpoint *smp;
2083 struct mount *child, *p;
2084 struct hlist_node *n;
2087 /* Preallocate a mountpoint in case the new mounts need
2088 * to be tucked under other mounts.
2090 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2092 return PTR_ERR(smp);
2094 /* Is there space to add these mounts to the mount namespace? */
2096 err = count_mounts(ns, source_mnt);
2101 if (IS_MNT_SHARED(dest_mnt)) {
2102 err = invent_group_ids(source_mnt, true);
2105 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2108 goto out_cleanup_ids;
2109 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2115 unhash_mnt(source_mnt);
2116 attach_mnt(source_mnt, dest_mnt, dest_mp);
2117 touch_mnt_namespace(source_mnt->mnt_ns);
2119 if (source_mnt->mnt_ns) {
2120 /* move from anon - the caller will destroy */
2121 list_del_init(&source_mnt->mnt_ns->list);
2123 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2124 commit_tree(source_mnt);
2127 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2129 hlist_del_init(&child->mnt_hash);
2130 q = __lookup_mnt(&child->mnt_parent->mnt,
2131 child->mnt_mountpoint);
2133 mnt_change_mountpoint(child, smp, q);
2134 /* Notice when we are propagating across user namespaces */
2135 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2136 lock_mnt_tree(child);
2137 child->mnt.mnt_flags &= ~MNT_LOCKED;
2140 put_mountpoint(smp);
2141 unlock_mount_hash();
2146 while (!hlist_empty(&tree_list)) {
2147 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2148 child->mnt_parent->mnt_ns->pending_mounts = 0;
2149 umount_tree(child, UMOUNT_SYNC);
2151 unlock_mount_hash();
2152 cleanup_group_ids(source_mnt, NULL);
2154 ns->pending_mounts = 0;
2156 read_seqlock_excl(&mount_lock);
2157 put_mountpoint(smp);
2158 read_sequnlock_excl(&mount_lock);
2163 static struct mountpoint *lock_mount(struct path *path)
2165 struct vfsmount *mnt;
2166 struct dentry *dentry = path->dentry;
2168 inode_lock(dentry->d_inode);
2169 if (unlikely(cant_mount(dentry))) {
2170 inode_unlock(dentry->d_inode);
2171 return ERR_PTR(-ENOENT);
2174 mnt = lookup_mnt(path);
2176 struct mountpoint *mp = get_mountpoint(dentry);
2179 inode_unlock(dentry->d_inode);
2185 inode_unlock(path->dentry->d_inode);
2188 dentry = path->dentry = dget(mnt->mnt_root);
2192 static void unlock_mount(struct mountpoint *where)
2194 struct dentry *dentry = where->m_dentry;
2196 read_seqlock_excl(&mount_lock);
2197 put_mountpoint(where);
2198 read_sequnlock_excl(&mount_lock);
2201 inode_unlock(dentry->d_inode);
2204 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2206 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2209 if (d_is_dir(mp->m_dentry) !=
2210 d_is_dir(mnt->mnt.mnt_root))
2213 return attach_recursive_mnt(mnt, p, mp, false);
2217 * Sanity check the flags to change_mnt_propagation.
2220 static int flags_to_propagation_type(int ms_flags)
2222 int type = ms_flags & ~(MS_REC | MS_SILENT);
2224 /* Fail if any non-propagation flags are set */
2225 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2227 /* Only one propagation flag should be set */
2228 if (!is_power_of_2(type))
2234 * recursively change the type of the mountpoint.
2236 static int do_change_type(struct path *path, int ms_flags)
2239 struct mount *mnt = real_mount(path->mnt);
2240 int recurse = ms_flags & MS_REC;
2244 if (path->dentry != path->mnt->mnt_root)
2247 type = flags_to_propagation_type(ms_flags);
2252 if (type == MS_SHARED) {
2253 err = invent_group_ids(mnt, recurse);
2259 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2260 change_mnt_propagation(m, type);
2261 unlock_mount_hash();
2268 static struct mount *__do_loopback(struct path *old_path, int recurse)
2270 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2272 if (IS_MNT_UNBINDABLE(old))
2275 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2278 if (!recurse && has_locked_children(old, old_path->dentry))
2282 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2284 mnt = clone_mnt(old, old_path->dentry, 0);
2287 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2293 * do loopback mount.
2295 static int do_loopback(struct path *path, const char *old_name,
2298 struct path old_path;
2299 struct mount *mnt = NULL, *parent;
2300 struct mountpoint *mp;
2302 if (!old_name || !*old_name)
2304 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2309 if (mnt_ns_loop(old_path.dentry))
2312 mp = lock_mount(path);
2318 parent = real_mount(path->mnt);
2319 if (!check_mnt(parent))
2322 mnt = __do_loopback(&old_path, recurse);
2328 err = graft_tree(mnt, parent, mp);
2331 umount_tree(mnt, UMOUNT_SYNC);
2332 unlock_mount_hash();
2337 path_put(&old_path);
2341 static struct file *open_detached_copy(struct path *path, bool recursive)
2343 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2344 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2345 struct mount *mnt, *p;
2349 return ERR_CAST(ns);
2352 mnt = __do_loopback(path, recursive);
2356 return ERR_CAST(mnt);
2360 for (p = mnt; p; p = next_mnt(p, mnt)) {
2365 list_add_tail(&ns->list, &mnt->mnt_list);
2367 unlock_mount_hash();
2371 path->mnt = &mnt->mnt;
2372 file = dentry_open(path, O_PATH, current_cred());
2374 dissolve_on_fput(path->mnt);
2376 file->f_mode |= FMODE_NEED_UNMOUNT;
2380 SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags)
2384 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2385 bool detached = flags & OPEN_TREE_CLONE;
2389 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2391 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2392 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2396 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2399 if (flags & AT_NO_AUTOMOUNT)
2400 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2401 if (flags & AT_SYMLINK_NOFOLLOW)
2402 lookup_flags &= ~LOOKUP_FOLLOW;
2403 if (flags & AT_EMPTY_PATH)
2404 lookup_flags |= LOOKUP_EMPTY;
2406 if (detached && !may_mount())
2409 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2413 error = user_path_at(dfd, filename, lookup_flags, &path);
2414 if (unlikely(error)) {
2415 file = ERR_PTR(error);
2418 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2420 file = dentry_open(&path, O_PATH, current_cred());
2425 return PTR_ERR(file);
2427 fd_install(fd, file);
2432 * Don't allow locked mount flags to be cleared.
2434 * No locks need to be held here while testing the various MNT_LOCK
2435 * flags because those flags can never be cleared once they are set.
2437 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2439 unsigned int fl = mnt->mnt.mnt_flags;
2441 if ((fl & MNT_LOCK_READONLY) &&
2442 !(mnt_flags & MNT_READONLY))
2445 if ((fl & MNT_LOCK_NODEV) &&
2446 !(mnt_flags & MNT_NODEV))
2449 if ((fl & MNT_LOCK_NOSUID) &&
2450 !(mnt_flags & MNT_NOSUID))
2453 if ((fl & MNT_LOCK_NOEXEC) &&
2454 !(mnt_flags & MNT_NOEXEC))
2457 if ((fl & MNT_LOCK_ATIME) &&
2458 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2464 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2466 bool readonly_request = (mnt_flags & MNT_READONLY);
2468 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2471 if (readonly_request)
2472 return mnt_make_readonly(mnt);
2474 return __mnt_unmake_readonly(mnt);
2478 * Update the user-settable attributes on a mount. The caller must hold
2479 * sb->s_umount for writing.
2481 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2484 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2485 mnt->mnt.mnt_flags = mnt_flags;
2486 touch_mnt_namespace(mnt->mnt_ns);
2487 unlock_mount_hash();
2490 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2492 struct super_block *sb = mnt->mnt_sb;
2494 if (!__mnt_is_readonly(mnt) &&
2495 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2496 char *buf = (char *)__get_free_page(GFP_KERNEL);
2497 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2500 time64_to_tm(sb->s_time_max, 0, &tm);
2502 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2504 is_mounted(mnt) ? "remounted" : "mounted",
2506 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2508 free_page((unsigned long)buf);
2513 * Handle reconfiguration of the mountpoint only without alteration of the
2514 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2517 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2519 struct super_block *sb = path->mnt->mnt_sb;
2520 struct mount *mnt = real_mount(path->mnt);
2523 if (!check_mnt(mnt))
2526 if (path->dentry != mnt->mnt.mnt_root)
2529 if (!can_change_locked_flags(mnt, mnt_flags))
2532 down_write(&sb->s_umount);
2533 ret = change_mount_ro_state(mnt, mnt_flags);
2535 set_mount_attributes(mnt, mnt_flags);
2536 up_write(&sb->s_umount);
2538 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2544 * change filesystem flags. dir should be a physical root of filesystem.
2545 * If you've mounted a non-root directory somewhere and want to do remount
2546 * on it - tough luck.
2548 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2549 int mnt_flags, void *data)
2552 struct super_block *sb = path->mnt->mnt_sb;
2553 struct mount *mnt = real_mount(path->mnt);
2554 struct fs_context *fc;
2556 if (!check_mnt(mnt))
2559 if (path->dentry != path->mnt->mnt_root)
2562 if (!can_change_locked_flags(mnt, mnt_flags))
2565 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2569 err = parse_monolithic_mount_data(fc, data);
2571 down_write(&sb->s_umount);
2573 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2574 err = reconfigure_super(fc);
2576 set_mount_attributes(mnt, mnt_flags);
2578 up_write(&sb->s_umount);
2581 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2587 static inline int tree_contains_unbindable(struct mount *mnt)
2590 for (p = mnt; p; p = next_mnt(p, mnt)) {
2591 if (IS_MNT_UNBINDABLE(p))
2598 * Check that there aren't references to earlier/same mount namespaces in the
2599 * specified subtree. Such references can act as pins for mount namespaces
2600 * that aren't checked by the mount-cycle checking code, thereby allowing
2601 * cycles to be made.
2603 static bool check_for_nsfs_mounts(struct mount *subtree)
2609 for (p = subtree; p; p = next_mnt(p, subtree))
2610 if (mnt_ns_loop(p->mnt.mnt_root))
2615 unlock_mount_hash();
2619 static int do_move_mount(struct path *old_path, struct path *new_path)
2621 struct mnt_namespace *ns;
2624 struct mount *parent;
2625 struct mountpoint *mp, *old_mp;
2629 mp = lock_mount(new_path);
2633 old = real_mount(old_path->mnt);
2634 p = real_mount(new_path->mnt);
2635 parent = old->mnt_parent;
2636 attached = mnt_has_parent(old);
2637 old_mp = old->mnt_mp;
2641 /* The mountpoint must be in our namespace. */
2645 /* The thing moved must be mounted... */
2646 if (!is_mounted(&old->mnt))
2649 /* ... and either ours or the root of anon namespace */
2650 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2653 if (old->mnt.mnt_flags & MNT_LOCKED)
2656 if (old_path->dentry != old_path->mnt->mnt_root)
2659 if (d_is_dir(new_path->dentry) !=
2660 d_is_dir(old_path->dentry))
2663 * Don't move a mount residing in a shared parent.
2665 if (attached && IS_MNT_SHARED(parent))
2668 * Don't move a mount tree containing unbindable mounts to a destination
2669 * mount which is shared.
2671 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2674 if (!check_for_nsfs_mounts(old))
2676 for (; mnt_has_parent(p); p = p->mnt_parent)
2680 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2685 /* if the mount is moved, it should no longer be expire
2687 list_del_init(&old->mnt_expire);
2689 put_mountpoint(old_mp);
2694 mntput_no_expire(parent);
2701 static int do_move_mount_old(struct path *path, const char *old_name)
2703 struct path old_path;
2706 if (!old_name || !*old_name)
2709 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2713 err = do_move_mount(&old_path, path);
2714 path_put(&old_path);
2719 * add a mount into a namespace's mount tree
2721 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2723 struct mountpoint *mp;
2724 struct mount *parent;
2727 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2729 mp = lock_mount(path);
2733 parent = real_mount(path->mnt);
2735 if (unlikely(!check_mnt(parent))) {
2736 /* that's acceptable only for automounts done in private ns */
2737 if (!(mnt_flags & MNT_SHRINKABLE))
2739 /* ... and for those we'd better have mountpoint still alive */
2740 if (!parent->mnt_ns)
2744 /* Refuse the same filesystem on the same mount point */
2746 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2747 path->mnt->mnt_root == path->dentry)
2751 if (d_is_symlink(newmnt->mnt.mnt_root))
2754 newmnt->mnt.mnt_flags = mnt_flags;
2755 err = graft_tree(newmnt, parent, mp);
2762 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2765 * Create a new mount using a superblock configuration and request it
2766 * be added to the namespace tree.
2768 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2769 unsigned int mnt_flags)
2771 struct vfsmount *mnt;
2772 struct super_block *sb = fc->root->d_sb;
2775 error = security_sb_kern_mount(sb);
2776 if (!error && mount_too_revealing(sb, &mnt_flags))
2779 if (unlikely(error)) {
2784 up_write(&sb->s_umount);
2786 mnt = vfs_create_mount(fc);
2788 return PTR_ERR(mnt);
2790 mnt_warn_timestamp_expiry(mountpoint, mnt);
2792 error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags);
2799 * create a new mount for userspace and request it to be added into the
2802 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2803 int mnt_flags, const char *name, void *data)
2805 struct file_system_type *type;
2806 struct fs_context *fc;
2807 const char *subtype = NULL;
2813 type = get_fs_type(fstype);
2817 if (type->fs_flags & FS_HAS_SUBTYPE) {
2818 subtype = strchr(fstype, '.');
2822 put_filesystem(type);
2828 fc = fs_context_for_mount(type, sb_flags);
2829 put_filesystem(type);
2834 err = vfs_parse_fs_string(fc, "subtype",
2835 subtype, strlen(subtype));
2837 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2839 err = parse_monolithic_mount_data(fc, data);
2840 if (!err && !mount_capable(fc))
2843 err = vfs_get_tree(fc);
2845 err = do_new_mount_fc(fc, path, mnt_flags);
2851 int finish_automount(struct vfsmount *m, struct path *path)
2853 struct mount *mnt = real_mount(m);
2855 /* The new mount record should have at least 2 refs to prevent it being
2856 * expired before we get a chance to add it
2858 BUG_ON(mnt_get_count(mnt) < 2);
2860 if (m->mnt_sb == path->mnt->mnt_sb &&
2861 m->mnt_root == path->dentry) {
2866 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2870 /* remove m from any expiration list it may be on */
2871 if (!list_empty(&mnt->mnt_expire)) {
2873 list_del_init(&mnt->mnt_expire);
2882 * mnt_set_expiry - Put a mount on an expiration list
2883 * @mnt: The mount to list.
2884 * @expiry_list: The list to add the mount to.
2886 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2890 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2894 EXPORT_SYMBOL(mnt_set_expiry);
2897 * process a list of expirable mountpoints with the intent of discarding any
2898 * mountpoints that aren't in use and haven't been touched since last we came
2901 void mark_mounts_for_expiry(struct list_head *mounts)
2903 struct mount *mnt, *next;
2904 LIST_HEAD(graveyard);
2906 if (list_empty(mounts))
2912 /* extract from the expiration list every vfsmount that matches the
2913 * following criteria:
2914 * - only referenced by its parent vfsmount
2915 * - still marked for expiry (marked on the last call here; marks are
2916 * cleared by mntput())
2918 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2919 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2920 propagate_mount_busy(mnt, 1))
2922 list_move(&mnt->mnt_expire, &graveyard);
2924 while (!list_empty(&graveyard)) {
2925 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2926 touch_mnt_namespace(mnt->mnt_ns);
2927 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2929 unlock_mount_hash();
2933 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2936 * Ripoff of 'select_parent()'
2938 * search the list of submounts for a given mountpoint, and move any
2939 * shrinkable submounts to the 'graveyard' list.
2941 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2943 struct mount *this_parent = parent;
2944 struct list_head *next;
2948 next = this_parent->mnt_mounts.next;
2950 while (next != &this_parent->mnt_mounts) {
2951 struct list_head *tmp = next;
2952 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2955 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2958 * Descend a level if the d_mounts list is non-empty.
2960 if (!list_empty(&mnt->mnt_mounts)) {
2965 if (!propagate_mount_busy(mnt, 1)) {
2966 list_move_tail(&mnt->mnt_expire, graveyard);
2971 * All done at this level ... ascend and resume the search
2973 if (this_parent != parent) {
2974 next = this_parent->mnt_child.next;
2975 this_parent = this_parent->mnt_parent;
2982 * process a list of expirable mountpoints with the intent of discarding any
2983 * submounts of a specific parent mountpoint
2985 * mount_lock must be held for write
2987 static void shrink_submounts(struct mount *mnt)
2989 LIST_HEAD(graveyard);
2992 /* extract submounts of 'mountpoint' from the expiration list */
2993 while (select_submounts(mnt, &graveyard)) {
2994 while (!list_empty(&graveyard)) {
2995 m = list_first_entry(&graveyard, struct mount,
2997 touch_mnt_namespace(m->mnt_ns);
2998 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3004 * Some copy_from_user() implementations do not return the exact number of
3005 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
3006 * Note that this function differs from copy_from_user() in that it will oops
3007 * on bad values of `to', rather than returning a short copy.
3009 static long exact_copy_from_user(void *to, const void __user * from,
3013 const char __user *f = from;
3016 if (!access_ok(from, n))
3020 if (__get_user(c, f)) {
3031 void *copy_mount_options(const void __user * data)
3040 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3042 return ERR_PTR(-ENOMEM);
3044 /* We only care that *some* data at the address the user
3045 * gave us is valid. Just in case, we'll zero
3046 * the remainder of the page.
3048 /* copy_from_user cannot cross TASK_SIZE ! */
3049 size = TASK_SIZE - (unsigned long)untagged_addr(data);
3050 if (size > PAGE_SIZE)
3053 i = size - exact_copy_from_user(copy, data, size);
3056 return ERR_PTR(-EFAULT);
3059 memset(copy + i, 0, PAGE_SIZE - i);
3063 char *copy_mount_string(const void __user *data)
3065 return data ? strndup_user(data, PATH_MAX) : NULL;
3069 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3070 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3072 * data is a (void *) that can point to any structure up to
3073 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3074 * information (or be NULL).
3076 * Pre-0.97 versions of mount() didn't have a flags word.
3077 * When the flags word was introduced its top half was required
3078 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3079 * Therefore, if this magic number is present, it carries no information
3080 * and must be discarded.
3082 long do_mount(const char *dev_name, const char __user *dir_name,
3083 const char *type_page, unsigned long flags, void *data_page)
3086 unsigned int mnt_flags = 0, sb_flags;
3090 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3091 flags &= ~MS_MGC_MSK;
3093 /* Basic sanity checks */
3095 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3097 if (flags & MS_NOUSER)
3100 /* ... and get the mountpoint */
3101 retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3105 retval = security_sb_mount(dev_name, &path,
3106 type_page, flags, data_page);
3107 if (!retval && !may_mount())
3109 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
3114 /* Default to relatime unless overriden */
3115 if (!(flags & MS_NOATIME))
3116 mnt_flags |= MNT_RELATIME;
3118 /* Separate the per-mountpoint flags */
3119 if (flags & MS_NOSUID)
3120 mnt_flags |= MNT_NOSUID;
3121 if (flags & MS_NODEV)
3122 mnt_flags |= MNT_NODEV;
3123 if (flags & MS_NOEXEC)
3124 mnt_flags |= MNT_NOEXEC;
3125 if (flags & MS_NOATIME)
3126 mnt_flags |= MNT_NOATIME;
3127 if (flags & MS_NODIRATIME)
3128 mnt_flags |= MNT_NODIRATIME;
3129 if (flags & MS_STRICTATIME)
3130 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3131 if (flags & MS_RDONLY)
3132 mnt_flags |= MNT_READONLY;
3134 /* The default atime for remount is preservation */
3135 if ((flags & MS_REMOUNT) &&
3136 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3137 MS_STRICTATIME)) == 0)) {
3138 mnt_flags &= ~MNT_ATIME_MASK;
3139 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
3142 sb_flags = flags & (SB_RDONLY |
3151 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3152 retval = do_reconfigure_mnt(&path, mnt_flags);
3153 else if (flags & MS_REMOUNT)
3154 retval = do_remount(&path, flags, sb_flags, mnt_flags,
3156 else if (flags & MS_BIND)
3157 retval = do_loopback(&path, dev_name, flags & MS_REC);
3158 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3159 retval = do_change_type(&path, flags);
3160 else if (flags & MS_MOVE)
3161 retval = do_move_mount_old(&path, dev_name);
3163 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
3164 dev_name, data_page);
3170 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3172 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3175 static void dec_mnt_namespaces(struct ucounts *ucounts)
3177 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3180 static void free_mnt_ns(struct mnt_namespace *ns)
3182 if (!is_anon_ns(ns))
3183 ns_free_inum(&ns->ns);
3184 dec_mnt_namespaces(ns->ucounts);
3185 put_user_ns(ns->user_ns);
3190 * Assign a sequence number so we can detect when we attempt to bind
3191 * mount a reference to an older mount namespace into the current
3192 * mount namespace, preventing reference counting loops. A 64bit
3193 * number incrementing at 10Ghz will take 12,427 years to wrap which
3194 * is effectively never, so we can ignore the possibility.
3196 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3198 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3200 struct mnt_namespace *new_ns;
3201 struct ucounts *ucounts;
3204 ucounts = inc_mnt_namespaces(user_ns);
3206 return ERR_PTR(-ENOSPC);
3208 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3210 dec_mnt_namespaces(ucounts);
3211 return ERR_PTR(-ENOMEM);
3214 ret = ns_alloc_inum(&new_ns->ns);
3217 dec_mnt_namespaces(ucounts);
3218 return ERR_PTR(ret);
3221 new_ns->ns.ops = &mntns_operations;
3223 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3224 atomic_set(&new_ns->count, 1);
3225 INIT_LIST_HEAD(&new_ns->list);
3226 init_waitqueue_head(&new_ns->poll);
3227 new_ns->user_ns = get_user_ns(user_ns);
3228 new_ns->ucounts = ucounts;
3233 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3234 struct user_namespace *user_ns, struct fs_struct *new_fs)
3236 struct mnt_namespace *new_ns;
3237 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3238 struct mount *p, *q;
3245 if (likely(!(flags & CLONE_NEWNS))) {
3252 new_ns = alloc_mnt_ns(user_ns, false);
3257 /* First pass: copy the tree topology */
3258 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3259 if (user_ns != ns->user_ns)
3260 copy_flags |= CL_SHARED_TO_SLAVE;
3261 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3264 free_mnt_ns(new_ns);
3265 return ERR_CAST(new);
3267 if (user_ns != ns->user_ns) {
3270 unlock_mount_hash();
3273 list_add_tail(&new_ns->list, &new->mnt_list);
3276 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3277 * as belonging to new namespace. We have already acquired a private
3278 * fs_struct, so tsk->fs->lock is not needed.
3286 if (&p->mnt == new_fs->root.mnt) {
3287 new_fs->root.mnt = mntget(&q->mnt);
3290 if (&p->mnt == new_fs->pwd.mnt) {
3291 new_fs->pwd.mnt = mntget(&q->mnt);
3295 p = next_mnt(p, old);
3296 q = next_mnt(q, new);
3299 while (p->mnt.mnt_root != q->mnt.mnt_root)
3300 p = next_mnt(p, old);
3312 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3314 struct mount *mnt = real_mount(m);
3315 struct mnt_namespace *ns;
3316 struct super_block *s;
3320 ns = alloc_mnt_ns(&init_user_ns, true);
3323 return ERR_CAST(ns);
3328 list_add(&mnt->mnt_list, &ns->list);
3330 err = vfs_path_lookup(m->mnt_root, m,
3331 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3336 return ERR_PTR(err);
3338 /* trade a vfsmount reference for active sb one */
3339 s = path.mnt->mnt_sb;
3340 atomic_inc(&s->s_active);
3342 /* lock the sucker */
3343 down_write(&s->s_umount);
3344 /* ... and return the root of (sub)tree on it */
3347 EXPORT_SYMBOL(mount_subtree);
3349 int ksys_mount(const char __user *dev_name, const char __user *dir_name,
3350 const char __user *type, unsigned long flags, void __user *data)
3357 kernel_type = copy_mount_string(type);
3358 ret = PTR_ERR(kernel_type);
3359 if (IS_ERR(kernel_type))
3362 kernel_dev = copy_mount_string(dev_name);
3363 ret = PTR_ERR(kernel_dev);
3364 if (IS_ERR(kernel_dev))
3367 options = copy_mount_options(data);
3368 ret = PTR_ERR(options);
3369 if (IS_ERR(options))
3372 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3383 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3384 char __user *, type, unsigned long, flags, void __user *, data)
3386 return ksys_mount(dev_name, dir_name, type, flags, data);
3390 * Create a kernel mount representation for a new, prepared superblock
3391 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3393 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3394 unsigned int, attr_flags)
3396 struct mnt_namespace *ns;
3397 struct fs_context *fc;
3399 struct path newmount;
3402 unsigned int mnt_flags = 0;
3408 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3411 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3416 MOUNT_ATTR_NODIRATIME))
3419 if (attr_flags & MOUNT_ATTR_RDONLY)
3420 mnt_flags |= MNT_READONLY;
3421 if (attr_flags & MOUNT_ATTR_NOSUID)
3422 mnt_flags |= MNT_NOSUID;
3423 if (attr_flags & MOUNT_ATTR_NODEV)
3424 mnt_flags |= MNT_NODEV;
3425 if (attr_flags & MOUNT_ATTR_NOEXEC)
3426 mnt_flags |= MNT_NOEXEC;
3427 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3428 mnt_flags |= MNT_NODIRATIME;
3430 switch (attr_flags & MOUNT_ATTR__ATIME) {
3431 case MOUNT_ATTR_STRICTATIME:
3433 case MOUNT_ATTR_NOATIME:
3434 mnt_flags |= MNT_NOATIME;
3436 case MOUNT_ATTR_RELATIME:
3437 mnt_flags |= MNT_RELATIME;
3448 if (f.file->f_op != &fscontext_fops)
3451 fc = f.file->private_data;
3453 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3457 /* There must be a valid superblock or we can't mount it */
3463 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3464 pr_warn("VFS: Mount too revealing\n");
3469 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3473 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3476 newmount.mnt = vfs_create_mount(fc);
3477 if (IS_ERR(newmount.mnt)) {
3478 ret = PTR_ERR(newmount.mnt);
3481 newmount.dentry = dget(fc->root);
3482 newmount.mnt->mnt_flags = mnt_flags;
3484 /* We've done the mount bit - now move the file context into more or
3485 * less the same state as if we'd done an fspick(). We don't want to
3486 * do any memory allocation or anything like that at this point as we
3487 * don't want to have to handle any errors incurred.
3489 vfs_clean_context(fc);
3491 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3496 mnt = real_mount(newmount.mnt);
3500 list_add(&mnt->mnt_list, &ns->list);
3501 mntget(newmount.mnt);
3503 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3504 * it, not just simply put it.
3506 file = dentry_open(&newmount, O_PATH, fc->cred);
3508 dissolve_on_fput(newmount.mnt);
3509 ret = PTR_ERR(file);
3512 file->f_mode |= FMODE_NEED_UNMOUNT;
3514 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3516 fd_install(ret, file);
3521 path_put(&newmount);
3523 mutex_unlock(&fc->uapi_mutex);
3530 * Move a mount from one place to another. In combination with
3531 * fsopen()/fsmount() this is used to install a new mount and in combination
3532 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3535 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3537 SYSCALL_DEFINE5(move_mount,
3538 int, from_dfd, const char *, from_pathname,
3539 int, to_dfd, const char *, to_pathname,
3540 unsigned int, flags)
3542 struct path from_path, to_path;
3543 unsigned int lflags;
3549 if (flags & ~MOVE_MOUNT__MASK)
3552 /* If someone gives a pathname, they aren't permitted to move
3553 * from an fd that requires unmount as we can't get at the flag
3554 * to clear it afterwards.
3557 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3558 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3559 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3561 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3566 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3567 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3568 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3570 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3574 ret = security_move_mount(&from_path, &to_path);
3578 ret = do_move_mount(&from_path, &to_path);
3583 path_put(&from_path);
3588 * Return true if path is reachable from root
3590 * namespace_sem or mount_lock is held
3592 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3593 const struct path *root)
3595 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3596 dentry = mnt->mnt_mountpoint;
3597 mnt = mnt->mnt_parent;
3599 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3602 bool path_is_under(const struct path *path1, const struct path *path2)
3605 read_seqlock_excl(&mount_lock);
3606 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3607 read_sequnlock_excl(&mount_lock);
3610 EXPORT_SYMBOL(path_is_under);
3613 * pivot_root Semantics:
3614 * Moves the root file system of the current process to the directory put_old,
3615 * makes new_root as the new root file system of the current process, and sets
3616 * root/cwd of all processes which had them on the current root to new_root.
3619 * The new_root and put_old must be directories, and must not be on the
3620 * same file system as the current process root. The put_old must be
3621 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3622 * pointed to by put_old must yield the same directory as new_root. No other
3623 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3625 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3626 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3627 * in this situation.
3630 * - we don't move root/cwd if they are not at the root (reason: if something
3631 * cared enough to change them, it's probably wrong to force them elsewhere)
3632 * - it's okay to pick a root that isn't the root of a file system, e.g.
3633 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3634 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3637 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3638 const char __user *, put_old)
3640 struct path new, old, root;
3641 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3642 struct mountpoint *old_mp, *root_mp;
3648 error = user_path_at(AT_FDCWD, new_root,
3649 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3653 error = user_path_at(AT_FDCWD, put_old,
3654 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3658 error = security_sb_pivotroot(&old, &new);
3662 get_fs_root(current->fs, &root);
3663 old_mp = lock_mount(&old);
3664 error = PTR_ERR(old_mp);
3669 new_mnt = real_mount(new.mnt);
3670 root_mnt = real_mount(root.mnt);
3671 old_mnt = real_mount(old.mnt);
3672 ex_parent = new_mnt->mnt_parent;
3673 root_parent = root_mnt->mnt_parent;
3674 if (IS_MNT_SHARED(old_mnt) ||
3675 IS_MNT_SHARED(ex_parent) ||
3676 IS_MNT_SHARED(root_parent))
3678 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3680 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3683 if (d_unlinked(new.dentry))
3686 if (new_mnt == root_mnt || old_mnt == root_mnt)
3687 goto out4; /* loop, on the same file system */
3689 if (root.mnt->mnt_root != root.dentry)
3690 goto out4; /* not a mountpoint */
3691 if (!mnt_has_parent(root_mnt))
3692 goto out4; /* not attached */
3693 if (new.mnt->mnt_root != new.dentry)
3694 goto out4; /* not a mountpoint */
3695 if (!mnt_has_parent(new_mnt))
3696 goto out4; /* not attached */
3697 /* make sure we can reach put_old from new_root */
3698 if (!is_path_reachable(old_mnt, old.dentry, &new))
3700 /* make certain new is below the root */
3701 if (!is_path_reachable(new_mnt, new.dentry, &root))
3704 umount_mnt(new_mnt);
3705 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3706 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3707 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3708 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3710 /* mount old root on put_old */
3711 attach_mnt(root_mnt, old_mnt, old_mp);
3712 /* mount new_root on / */
3713 attach_mnt(new_mnt, root_parent, root_mp);
3714 mnt_add_count(root_parent, -1);
3715 touch_mnt_namespace(current->nsproxy->mnt_ns);
3716 /* A moved mount should not expire automatically */
3717 list_del_init(&new_mnt->mnt_expire);
3718 put_mountpoint(root_mp);
3719 unlock_mount_hash();
3720 chroot_fs_refs(&root, &new);
3723 unlock_mount(old_mp);
3725 mntput_no_expire(ex_parent);
3736 static void __init init_mount_tree(void)
3738 struct vfsmount *mnt;
3740 struct mnt_namespace *ns;
3743 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3745 panic("Can't create rootfs");
3747 ns = alloc_mnt_ns(&init_user_ns, false);
3749 panic("Can't allocate initial namespace");
3750 m = real_mount(mnt);
3754 list_add(&m->mnt_list, &ns->list);
3755 init_task.nsproxy->mnt_ns = ns;
3759 root.dentry = mnt->mnt_root;
3760 mnt->mnt_flags |= MNT_LOCKED;
3762 set_fs_pwd(current->fs, &root);
3763 set_fs_root(current->fs, &root);
3766 void __init mnt_init(void)
3770 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3771 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3773 mount_hashtable = alloc_large_system_hash("Mount-cache",
3774 sizeof(struct hlist_head),
3777 &m_hash_shift, &m_hash_mask, 0, 0);
3778 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3779 sizeof(struct hlist_head),
3782 &mp_hash_shift, &mp_hash_mask, 0, 0);
3784 if (!mount_hashtable || !mountpoint_hashtable)
3785 panic("Failed to allocate mount hash table\n");
3791 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3793 fs_kobj = kobject_create_and_add("fs", NULL);
3795 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3801 void put_mnt_ns(struct mnt_namespace *ns)
3803 if (!atomic_dec_and_test(&ns->count))
3805 drop_collected_mounts(&ns->root->mnt);
3809 struct vfsmount *kern_mount(struct file_system_type *type)
3811 struct vfsmount *mnt;
3812 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3815 * it is a longterm mount, don't release mnt until
3816 * we unmount before file sys is unregistered
3818 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3822 EXPORT_SYMBOL_GPL(kern_mount);
3824 void kern_unmount(struct vfsmount *mnt)
3826 /* release long term mount so mount point can be released */
3827 if (!IS_ERR_OR_NULL(mnt)) {
3828 real_mount(mnt)->mnt_ns = NULL;
3829 synchronize_rcu(); /* yecchhh... */
3833 EXPORT_SYMBOL(kern_unmount);
3835 bool our_mnt(struct vfsmount *mnt)
3837 return check_mnt(real_mount(mnt));
3840 bool current_chrooted(void)
3842 /* Does the current process have a non-standard root */
3843 struct path ns_root;
3844 struct path fs_root;
3847 /* Find the namespace root */
3848 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3849 ns_root.dentry = ns_root.mnt->mnt_root;
3851 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3854 get_fs_root(current->fs, &fs_root);
3856 chrooted = !path_equal(&fs_root, &ns_root);
3864 static bool mnt_already_visible(struct mnt_namespace *ns,
3865 const struct super_block *sb,
3868 int new_flags = *new_mnt_flags;
3870 bool visible = false;
3872 down_read(&namespace_sem);
3873 list_for_each_entry(mnt, &ns->list, mnt_list) {
3874 struct mount *child;
3877 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3880 /* This mount is not fully visible if it's root directory
3881 * is not the root directory of the filesystem.
3883 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3886 /* A local view of the mount flags */
3887 mnt_flags = mnt->mnt.mnt_flags;
3889 /* Don't miss readonly hidden in the superblock flags */
3890 if (sb_rdonly(mnt->mnt.mnt_sb))
3891 mnt_flags |= MNT_LOCK_READONLY;
3893 /* Verify the mount flags are equal to or more permissive
3894 * than the proposed new mount.
3896 if ((mnt_flags & MNT_LOCK_READONLY) &&
3897 !(new_flags & MNT_READONLY))
3899 if ((mnt_flags & MNT_LOCK_ATIME) &&
3900 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3903 /* This mount is not fully visible if there are any
3904 * locked child mounts that cover anything except for
3905 * empty directories.
3907 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3908 struct inode *inode = child->mnt_mountpoint->d_inode;
3909 /* Only worry about locked mounts */
3910 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3912 /* Is the directory permanetly empty? */
3913 if (!is_empty_dir_inode(inode))
3916 /* Preserve the locked attributes */
3917 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3924 up_read(&namespace_sem);
3928 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3930 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3931 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3932 unsigned long s_iflags;
3934 if (ns->user_ns == &init_user_ns)
3937 /* Can this filesystem be too revealing? */
3938 s_iflags = sb->s_iflags;
3939 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3942 if ((s_iflags & required_iflags) != required_iflags) {
3943 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3948 return !mnt_already_visible(ns, sb, new_mnt_flags);
3951 bool mnt_may_suid(struct vfsmount *mnt)
3954 * Foreign mounts (accessed via fchdir or through /proc
3955 * symlinks) are always treated as if they are nosuid. This
3956 * prevents namespaces from trusting potentially unsafe
3957 * suid/sgid bits, file caps, or security labels that originate
3958 * in other namespaces.
3960 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3961 current_in_userns(mnt->mnt_sb->s_user_ns);
3964 static struct ns_common *mntns_get(struct task_struct *task)
3966 struct ns_common *ns = NULL;
3967 struct nsproxy *nsproxy;
3970 nsproxy = task->nsproxy;
3972 ns = &nsproxy->mnt_ns->ns;
3973 get_mnt_ns(to_mnt_ns(ns));
3980 static void mntns_put(struct ns_common *ns)
3982 put_mnt_ns(to_mnt_ns(ns));
3985 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3987 struct fs_struct *fs = current->fs;
3988 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3992 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3993 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3994 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3997 if (is_anon_ns(mnt_ns))
4004 old_mnt_ns = nsproxy->mnt_ns;
4005 nsproxy->mnt_ns = mnt_ns;
4008 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4009 "/", LOOKUP_DOWN, &root);
4011 /* revert to old namespace */
4012 nsproxy->mnt_ns = old_mnt_ns;
4017 put_mnt_ns(old_mnt_ns);
4019 /* Update the pwd and root */
4020 set_fs_pwd(fs, &root);
4021 set_fs_root(fs, &root);
4027 static struct user_namespace *mntns_owner(struct ns_common *ns)
4029 return to_mnt_ns(ns)->user_ns;
4032 const struct proc_ns_operations mntns_operations = {
4034 .type = CLONE_NEWNS,
4037 .install = mntns_install,
4038 .owner = mntns_owner,