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));
651 static inline void lock_ns_list(struct mnt_namespace *ns)
653 spin_lock(&ns->ns_lock);
656 static inline void unlock_ns_list(struct mnt_namespace *ns)
658 spin_unlock(&ns->ns_lock);
661 static inline bool mnt_is_cursor(struct mount *mnt)
663 return mnt->mnt.mnt_flags & MNT_CURSOR;
667 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
668 * current mount namespace.
670 * The common case is dentries are not mountpoints at all and that
671 * test is handled inline. For the slow case when we are actually
672 * dealing with a mountpoint of some kind, walk through all of the
673 * mounts in the current mount namespace and test to see if the dentry
676 * The mount_hashtable is not usable in the context because we
677 * need to identify all mounts that may be in the current mount
678 * namespace not just a mount that happens to have some specified
681 bool __is_local_mountpoint(struct dentry *dentry)
683 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
685 bool is_covered = false;
687 if (!d_mountpoint(dentry))
690 down_read(&namespace_sem);
692 list_for_each_entry(mnt, &ns->list, mnt_list) {
693 if (mnt_is_cursor(mnt))
695 is_covered = (mnt->mnt_mountpoint == dentry);
700 up_read(&namespace_sem);
705 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
707 struct hlist_head *chain = mp_hash(dentry);
708 struct mountpoint *mp;
710 hlist_for_each_entry(mp, chain, m_hash) {
711 if (mp->m_dentry == dentry) {
719 static struct mountpoint *get_mountpoint(struct dentry *dentry)
721 struct mountpoint *mp, *new = NULL;
724 if (d_mountpoint(dentry)) {
725 /* might be worth a WARN_ON() */
726 if (d_unlinked(dentry))
727 return ERR_PTR(-ENOENT);
729 read_seqlock_excl(&mount_lock);
730 mp = lookup_mountpoint(dentry);
731 read_sequnlock_excl(&mount_lock);
737 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
739 return ERR_PTR(-ENOMEM);
742 /* Exactly one processes may set d_mounted */
743 ret = d_set_mounted(dentry);
745 /* Someone else set d_mounted? */
749 /* The dentry is not available as a mountpoint? */
754 /* Add the new mountpoint to the hash table */
755 read_seqlock_excl(&mount_lock);
756 new->m_dentry = dget(dentry);
758 hlist_add_head(&new->m_hash, mp_hash(dentry));
759 INIT_HLIST_HEAD(&new->m_list);
760 read_sequnlock_excl(&mount_lock);
770 * vfsmount lock must be held. Additionally, the caller is responsible
771 * for serializing calls for given disposal list.
773 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
775 if (!--mp->m_count) {
776 struct dentry *dentry = mp->m_dentry;
777 BUG_ON(!hlist_empty(&mp->m_list));
778 spin_lock(&dentry->d_lock);
779 dentry->d_flags &= ~DCACHE_MOUNTED;
780 spin_unlock(&dentry->d_lock);
781 dput_to_list(dentry, list);
782 hlist_del(&mp->m_hash);
787 /* called with namespace_lock and vfsmount lock */
788 static void put_mountpoint(struct mountpoint *mp)
790 __put_mountpoint(mp, &ex_mountpoints);
793 static inline int check_mnt(struct mount *mnt)
795 return mnt->mnt_ns == current->nsproxy->mnt_ns;
799 * vfsmount lock must be held for write
801 static void touch_mnt_namespace(struct mnt_namespace *ns)
805 wake_up_interruptible(&ns->poll);
810 * vfsmount lock must be held for write
812 static void __touch_mnt_namespace(struct mnt_namespace *ns)
814 if (ns && ns->event != event) {
816 wake_up_interruptible(&ns->poll);
821 * vfsmount lock must be held for write
823 static struct mountpoint *unhash_mnt(struct mount *mnt)
825 struct mountpoint *mp;
826 mnt->mnt_parent = mnt;
827 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
828 list_del_init(&mnt->mnt_child);
829 hlist_del_init_rcu(&mnt->mnt_hash);
830 hlist_del_init(&mnt->mnt_mp_list);
837 * vfsmount lock must be held for write
839 static void umount_mnt(struct mount *mnt)
841 put_mountpoint(unhash_mnt(mnt));
845 * vfsmount lock must be held for write
847 void mnt_set_mountpoint(struct mount *mnt,
848 struct mountpoint *mp,
849 struct mount *child_mnt)
852 mnt_add_count(mnt, 1); /* essentially, that's mntget */
853 child_mnt->mnt_mountpoint = mp->m_dentry;
854 child_mnt->mnt_parent = mnt;
855 child_mnt->mnt_mp = mp;
856 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
859 static void __attach_mnt(struct mount *mnt, struct mount *parent)
861 hlist_add_head_rcu(&mnt->mnt_hash,
862 m_hash(&parent->mnt, mnt->mnt_mountpoint));
863 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
867 * vfsmount lock must be held for write
869 static void attach_mnt(struct mount *mnt,
870 struct mount *parent,
871 struct mountpoint *mp)
873 mnt_set_mountpoint(parent, mp, mnt);
874 __attach_mnt(mnt, parent);
877 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
879 struct mountpoint *old_mp = mnt->mnt_mp;
880 struct mount *old_parent = mnt->mnt_parent;
882 list_del_init(&mnt->mnt_child);
883 hlist_del_init(&mnt->mnt_mp_list);
884 hlist_del_init_rcu(&mnt->mnt_hash);
886 attach_mnt(mnt, parent, mp);
888 put_mountpoint(old_mp);
889 mnt_add_count(old_parent, -1);
893 * vfsmount lock must be held for write
895 static void commit_tree(struct mount *mnt)
897 struct mount *parent = mnt->mnt_parent;
900 struct mnt_namespace *n = parent->mnt_ns;
902 BUG_ON(parent == mnt);
904 list_add_tail(&head, &mnt->mnt_list);
905 list_for_each_entry(m, &head, mnt_list)
908 list_splice(&head, n->list.prev);
910 n->mounts += n->pending_mounts;
911 n->pending_mounts = 0;
913 __attach_mnt(mnt, parent);
914 touch_mnt_namespace(n);
917 static struct mount *next_mnt(struct mount *p, struct mount *root)
919 struct list_head *next = p->mnt_mounts.next;
920 if (next == &p->mnt_mounts) {
924 next = p->mnt_child.next;
925 if (next != &p->mnt_parent->mnt_mounts)
930 return list_entry(next, struct mount, mnt_child);
933 static struct mount *skip_mnt_tree(struct mount *p)
935 struct list_head *prev = p->mnt_mounts.prev;
936 while (prev != &p->mnt_mounts) {
937 p = list_entry(prev, struct mount, mnt_child);
938 prev = p->mnt_mounts.prev;
944 * vfs_create_mount - Create a mount for a configured superblock
945 * @fc: The configuration context with the superblock attached
947 * Create a mount to an already configured superblock. If necessary, the
948 * caller should invoke vfs_get_tree() before calling this.
950 * Note that this does not attach the mount to anything.
952 struct vfsmount *vfs_create_mount(struct fs_context *fc)
957 return ERR_PTR(-EINVAL);
959 mnt = alloc_vfsmnt(fc->source ?: "none");
961 return ERR_PTR(-ENOMEM);
963 if (fc->sb_flags & SB_KERNMOUNT)
964 mnt->mnt.mnt_flags = MNT_INTERNAL;
966 atomic_inc(&fc->root->d_sb->s_active);
967 mnt->mnt.mnt_sb = fc->root->d_sb;
968 mnt->mnt.mnt_root = dget(fc->root);
969 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
970 mnt->mnt_parent = mnt;
973 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
977 EXPORT_SYMBOL(vfs_create_mount);
979 struct vfsmount *fc_mount(struct fs_context *fc)
981 int err = vfs_get_tree(fc);
983 up_write(&fc->root->d_sb->s_umount);
984 return vfs_create_mount(fc);
988 EXPORT_SYMBOL(fc_mount);
990 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
991 int flags, const char *name,
994 struct fs_context *fc;
995 struct vfsmount *mnt;
999 return ERR_PTR(-EINVAL);
1001 fc = fs_context_for_mount(type, flags);
1003 return ERR_CAST(fc);
1006 ret = vfs_parse_fs_string(fc, "source",
1007 name, strlen(name));
1009 ret = parse_monolithic_mount_data(fc, data);
1018 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1021 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1022 const char *name, void *data)
1024 /* Until it is worked out how to pass the user namespace
1025 * through from the parent mount to the submount don't support
1026 * unprivileged mounts with submounts.
1028 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1029 return ERR_PTR(-EPERM);
1031 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1033 EXPORT_SYMBOL_GPL(vfs_submount);
1035 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1038 struct super_block *sb = old->mnt.mnt_sb;
1042 mnt = alloc_vfsmnt(old->mnt_devname);
1044 return ERR_PTR(-ENOMEM);
1046 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1047 mnt->mnt_group_id = 0; /* not a peer of original */
1049 mnt->mnt_group_id = old->mnt_group_id;
1051 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1052 err = mnt_alloc_group_id(mnt);
1057 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1058 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1060 atomic_inc(&sb->s_active);
1061 mnt->mnt.mnt_sb = sb;
1062 mnt->mnt.mnt_root = dget(root);
1063 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1064 mnt->mnt_parent = mnt;
1066 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1067 unlock_mount_hash();
1069 if ((flag & CL_SLAVE) ||
1070 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1071 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1072 mnt->mnt_master = old;
1073 CLEAR_MNT_SHARED(mnt);
1074 } else if (!(flag & CL_PRIVATE)) {
1075 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1076 list_add(&mnt->mnt_share, &old->mnt_share);
1077 if (IS_MNT_SLAVE(old))
1078 list_add(&mnt->mnt_slave, &old->mnt_slave);
1079 mnt->mnt_master = old->mnt_master;
1081 CLEAR_MNT_SHARED(mnt);
1083 if (flag & CL_MAKE_SHARED)
1084 set_mnt_shared(mnt);
1086 /* stick the duplicate mount on the same expiry list
1087 * as the original if that was on one */
1088 if (flag & CL_EXPIRE) {
1089 if (!list_empty(&old->mnt_expire))
1090 list_add(&mnt->mnt_expire, &old->mnt_expire);
1098 return ERR_PTR(err);
1101 static void cleanup_mnt(struct mount *mnt)
1103 struct hlist_node *p;
1106 * The warning here probably indicates that somebody messed
1107 * up a mnt_want/drop_write() pair. If this happens, the
1108 * filesystem was probably unable to make r/w->r/o transitions.
1109 * The locking used to deal with mnt_count decrement provides barriers,
1110 * so mnt_get_writers() below is safe.
1112 WARN_ON(mnt_get_writers(mnt));
1113 if (unlikely(mnt->mnt_pins.first))
1115 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1116 hlist_del(&m->mnt_umount);
1119 fsnotify_vfsmount_delete(&mnt->mnt);
1120 dput(mnt->mnt.mnt_root);
1121 deactivate_super(mnt->mnt.mnt_sb);
1123 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1126 static void __cleanup_mnt(struct rcu_head *head)
1128 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1131 static LLIST_HEAD(delayed_mntput_list);
1132 static void delayed_mntput(struct work_struct *unused)
1134 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1135 struct mount *m, *t;
1137 llist_for_each_entry_safe(m, t, node, mnt_llist)
1140 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1142 static void mntput_no_expire(struct mount *mnt)
1147 if (likely(READ_ONCE(mnt->mnt_ns))) {
1149 * Since we don't do lock_mount_hash() here,
1150 * ->mnt_ns can change under us. However, if it's
1151 * non-NULL, then there's a reference that won't
1152 * be dropped until after an RCU delay done after
1153 * turning ->mnt_ns NULL. So if we observe it
1154 * non-NULL under rcu_read_lock(), the reference
1155 * we are dropping is not the final one.
1157 mnt_add_count(mnt, -1);
1163 * make sure that if __legitimize_mnt() has not seen us grab
1164 * mount_lock, we'll see their refcount increment here.
1167 mnt_add_count(mnt, -1);
1168 if (mnt_get_count(mnt)) {
1170 unlock_mount_hash();
1173 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1175 unlock_mount_hash();
1178 mnt->mnt.mnt_flags |= MNT_DOOMED;
1181 list_del(&mnt->mnt_instance);
1183 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1184 struct mount *p, *tmp;
1185 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1186 __put_mountpoint(unhash_mnt(p), &list);
1187 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1190 unlock_mount_hash();
1191 shrink_dentry_list(&list);
1193 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1194 struct task_struct *task = current;
1195 if (likely(!(task->flags & PF_KTHREAD))) {
1196 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1197 if (!task_work_add(task, &mnt->mnt_rcu, true))
1200 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1201 schedule_delayed_work(&delayed_mntput_work, 1);
1207 void mntput(struct vfsmount *mnt)
1210 struct mount *m = real_mount(mnt);
1211 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1212 if (unlikely(m->mnt_expiry_mark))
1213 m->mnt_expiry_mark = 0;
1214 mntput_no_expire(m);
1217 EXPORT_SYMBOL(mntput);
1219 struct vfsmount *mntget(struct vfsmount *mnt)
1222 mnt_add_count(real_mount(mnt), 1);
1225 EXPORT_SYMBOL(mntget);
1227 /* path_is_mountpoint() - Check if path is a mount in the current
1230 * d_mountpoint() can only be used reliably to establish if a dentry is
1231 * not mounted in any namespace and that common case is handled inline.
1232 * d_mountpoint() isn't aware of the possibility there may be multiple
1233 * mounts using a given dentry in a different namespace. This function
1234 * checks if the passed in path is a mountpoint rather than the dentry
1237 bool path_is_mountpoint(const struct path *path)
1242 if (!d_mountpoint(path->dentry))
1247 seq = read_seqbegin(&mount_lock);
1248 res = __path_is_mountpoint(path);
1249 } while (read_seqretry(&mount_lock, seq));
1254 EXPORT_SYMBOL(path_is_mountpoint);
1256 struct vfsmount *mnt_clone_internal(const struct path *path)
1259 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1262 p->mnt.mnt_flags |= MNT_INTERNAL;
1266 #ifdef CONFIG_PROC_FS
1267 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1268 struct list_head *p)
1270 struct mount *mnt, *ret = NULL;
1273 list_for_each_continue(p, &ns->list) {
1274 mnt = list_entry(p, typeof(*mnt), mnt_list);
1275 if (!mnt_is_cursor(mnt)) {
1285 /* iterator; we want it to have access to namespace_sem, thus here... */
1286 static void *m_start(struct seq_file *m, loff_t *pos)
1288 struct proc_mounts *p = m->private;
1289 struct list_head *prev;
1291 down_read(&namespace_sem);
1293 prev = &p->ns->list;
1295 prev = &p->cursor.mnt_list;
1297 /* Read after we'd reached the end? */
1298 if (list_empty(prev))
1302 return mnt_list_next(p->ns, prev);
1305 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1307 struct proc_mounts *p = m->private;
1308 struct mount *mnt = v;
1311 return mnt_list_next(p->ns, &mnt->mnt_list);
1314 static void m_stop(struct seq_file *m, void *v)
1316 struct proc_mounts *p = m->private;
1317 struct mount *mnt = v;
1319 lock_ns_list(p->ns);
1321 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1323 list_del_init(&p->cursor.mnt_list);
1324 unlock_ns_list(p->ns);
1325 up_read(&namespace_sem);
1328 static int m_show(struct seq_file *m, void *v)
1330 struct proc_mounts *p = m->private;
1331 struct mount *r = v;
1332 return p->show(m, &r->mnt);
1335 const struct seq_operations mounts_op = {
1342 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1344 down_read(&namespace_sem);
1346 list_del(&cursor->mnt_list);
1348 up_read(&namespace_sem);
1350 #endif /* CONFIG_PROC_FS */
1353 * may_umount_tree - check if a mount tree is busy
1354 * @mnt: root of mount tree
1356 * This is called to check if a tree of mounts has any
1357 * open files, pwds, chroots or sub mounts that are
1360 int may_umount_tree(struct vfsmount *m)
1362 struct mount *mnt = real_mount(m);
1363 int actual_refs = 0;
1364 int minimum_refs = 0;
1368 /* write lock needed for mnt_get_count */
1370 for (p = mnt; p; p = next_mnt(p, mnt)) {
1371 actual_refs += mnt_get_count(p);
1374 unlock_mount_hash();
1376 if (actual_refs > minimum_refs)
1382 EXPORT_SYMBOL(may_umount_tree);
1385 * may_umount - check if a mount point is busy
1386 * @mnt: root of mount
1388 * This is called to check if a mount point has any
1389 * open files, pwds, chroots or sub mounts. If the
1390 * mount has sub mounts this will return busy
1391 * regardless of whether the sub mounts are busy.
1393 * Doesn't take quota and stuff into account. IOW, in some cases it will
1394 * give false negatives. The main reason why it's here is that we need
1395 * a non-destructive way to look for easily umountable filesystems.
1397 int may_umount(struct vfsmount *mnt)
1400 down_read(&namespace_sem);
1402 if (propagate_mount_busy(real_mount(mnt), 2))
1404 unlock_mount_hash();
1405 up_read(&namespace_sem);
1409 EXPORT_SYMBOL(may_umount);
1411 static void namespace_unlock(void)
1413 struct hlist_head head;
1414 struct hlist_node *p;
1418 hlist_move_list(&unmounted, &head);
1419 list_splice_init(&ex_mountpoints, &list);
1421 up_write(&namespace_sem);
1423 shrink_dentry_list(&list);
1425 if (likely(hlist_empty(&head)))
1428 synchronize_rcu_expedited();
1430 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1431 hlist_del(&m->mnt_umount);
1436 static inline void namespace_lock(void)
1438 down_write(&namespace_sem);
1441 enum umount_tree_flags {
1443 UMOUNT_PROPAGATE = 2,
1444 UMOUNT_CONNECTED = 4,
1447 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1449 /* Leaving mounts connected is only valid for lazy umounts */
1450 if (how & UMOUNT_SYNC)
1453 /* A mount without a parent has nothing to be connected to */
1454 if (!mnt_has_parent(mnt))
1457 /* Because the reference counting rules change when mounts are
1458 * unmounted and connected, umounted mounts may not be
1459 * connected to mounted mounts.
1461 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1464 /* Has it been requested that the mount remain connected? */
1465 if (how & UMOUNT_CONNECTED)
1468 /* Is the mount locked such that it needs to remain connected? */
1469 if (IS_MNT_LOCKED(mnt))
1472 /* By default disconnect the mount */
1477 * mount_lock must be held
1478 * namespace_sem must be held for write
1480 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1482 LIST_HEAD(tmp_list);
1485 if (how & UMOUNT_PROPAGATE)
1486 propagate_mount_unlock(mnt);
1488 /* Gather the mounts to umount */
1489 for (p = mnt; p; p = next_mnt(p, mnt)) {
1490 p->mnt.mnt_flags |= MNT_UMOUNT;
1491 list_move(&p->mnt_list, &tmp_list);
1494 /* Hide the mounts from mnt_mounts */
1495 list_for_each_entry(p, &tmp_list, mnt_list) {
1496 list_del_init(&p->mnt_child);
1499 /* Add propogated mounts to the tmp_list */
1500 if (how & UMOUNT_PROPAGATE)
1501 propagate_umount(&tmp_list);
1503 while (!list_empty(&tmp_list)) {
1504 struct mnt_namespace *ns;
1506 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1507 list_del_init(&p->mnt_expire);
1508 list_del_init(&p->mnt_list);
1512 __touch_mnt_namespace(ns);
1515 if (how & UMOUNT_SYNC)
1516 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1518 disconnect = disconnect_mount(p, how);
1519 if (mnt_has_parent(p)) {
1520 mnt_add_count(p->mnt_parent, -1);
1522 /* Don't forget about p */
1523 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1528 change_mnt_propagation(p, MS_PRIVATE);
1530 hlist_add_head(&p->mnt_umount, &unmounted);
1534 static void shrink_submounts(struct mount *mnt);
1536 static int do_umount_root(struct super_block *sb)
1540 down_write(&sb->s_umount);
1541 if (!sb_rdonly(sb)) {
1542 struct fs_context *fc;
1544 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1549 ret = parse_monolithic_mount_data(fc, NULL);
1551 ret = reconfigure_super(fc);
1555 up_write(&sb->s_umount);
1559 static int do_umount(struct mount *mnt, int flags)
1561 struct super_block *sb = mnt->mnt.mnt_sb;
1564 retval = security_sb_umount(&mnt->mnt, flags);
1569 * Allow userspace to request a mountpoint be expired rather than
1570 * unmounting unconditionally. Unmount only happens if:
1571 * (1) the mark is already set (the mark is cleared by mntput())
1572 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1574 if (flags & MNT_EXPIRE) {
1575 if (&mnt->mnt == current->fs->root.mnt ||
1576 flags & (MNT_FORCE | MNT_DETACH))
1580 * probably don't strictly need the lock here if we examined
1581 * all race cases, but it's a slowpath.
1584 if (mnt_get_count(mnt) != 2) {
1585 unlock_mount_hash();
1588 unlock_mount_hash();
1590 if (!xchg(&mnt->mnt_expiry_mark, 1))
1595 * If we may have to abort operations to get out of this
1596 * mount, and they will themselves hold resources we must
1597 * allow the fs to do things. In the Unix tradition of
1598 * 'Gee thats tricky lets do it in userspace' the umount_begin
1599 * might fail to complete on the first run through as other tasks
1600 * must return, and the like. Thats for the mount program to worry
1601 * about for the moment.
1604 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1605 sb->s_op->umount_begin(sb);
1609 * No sense to grab the lock for this test, but test itself looks
1610 * somewhat bogus. Suggestions for better replacement?
1611 * Ho-hum... In principle, we might treat that as umount + switch
1612 * to rootfs. GC would eventually take care of the old vfsmount.
1613 * Actually it makes sense, especially if rootfs would contain a
1614 * /reboot - static binary that would close all descriptors and
1615 * call reboot(9). Then init(8) could umount root and exec /reboot.
1617 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1619 * Special case for "unmounting" root ...
1620 * we just try to remount it readonly.
1622 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1624 return do_umount_root(sb);
1630 /* Recheck MNT_LOCKED with the locks held */
1632 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1636 if (flags & MNT_DETACH) {
1637 if (!list_empty(&mnt->mnt_list))
1638 umount_tree(mnt, UMOUNT_PROPAGATE);
1641 shrink_submounts(mnt);
1643 if (!propagate_mount_busy(mnt, 2)) {
1644 if (!list_empty(&mnt->mnt_list))
1645 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1650 unlock_mount_hash();
1656 * __detach_mounts - lazily unmount all mounts on the specified dentry
1658 * During unlink, rmdir, and d_drop it is possible to loose the path
1659 * to an existing mountpoint, and wind up leaking the mount.
1660 * detach_mounts allows lazily unmounting those mounts instead of
1663 * The caller may hold dentry->d_inode->i_mutex.
1665 void __detach_mounts(struct dentry *dentry)
1667 struct mountpoint *mp;
1672 mp = lookup_mountpoint(dentry);
1677 while (!hlist_empty(&mp->m_list)) {
1678 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1679 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1681 hlist_add_head(&mnt->mnt_umount, &unmounted);
1683 else umount_tree(mnt, UMOUNT_CONNECTED);
1687 unlock_mount_hash();
1692 * Is the caller allowed to modify his namespace?
1694 static inline bool may_mount(void)
1696 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1699 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1700 static inline bool may_mandlock(void)
1702 return capable(CAP_SYS_ADMIN);
1705 static inline bool may_mandlock(void)
1707 pr_warn("VFS: \"mand\" mount option not supported");
1713 * Now umount can handle mount points as well as block devices.
1714 * This is important for filesystems which use unnamed block devices.
1716 * We now support a flag for forced unmount like the other 'big iron'
1717 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1720 int ksys_umount(char __user *name, int flags)
1725 int lookup_flags = LOOKUP_MOUNTPOINT;
1727 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1733 if (!(flags & UMOUNT_NOFOLLOW))
1734 lookup_flags |= LOOKUP_FOLLOW;
1736 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1739 mnt = real_mount(path.mnt);
1741 if (path.dentry != path.mnt->mnt_root)
1743 if (!check_mnt(mnt))
1745 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1748 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1751 retval = do_umount(mnt, flags);
1753 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1755 mntput_no_expire(mnt);
1760 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1762 return ksys_umount(name, flags);
1765 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1768 * The 2.0 compatible umount. No flags.
1770 SYSCALL_DEFINE1(oldumount, char __user *, name)
1772 return ksys_umount(name, 0);
1777 static bool is_mnt_ns_file(struct dentry *dentry)
1779 /* Is this a proxy for a mount namespace? */
1780 return dentry->d_op == &ns_dentry_operations &&
1781 dentry->d_fsdata == &mntns_operations;
1784 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1786 return container_of(ns, struct mnt_namespace, ns);
1789 static bool mnt_ns_loop(struct dentry *dentry)
1791 /* Could bind mounting the mount namespace inode cause a
1792 * mount namespace loop?
1794 struct mnt_namespace *mnt_ns;
1795 if (!is_mnt_ns_file(dentry))
1798 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1799 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1802 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1805 struct mount *res, *p, *q, *r, *parent;
1807 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1808 return ERR_PTR(-EINVAL);
1810 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1811 return ERR_PTR(-EINVAL);
1813 res = q = clone_mnt(mnt, dentry, flag);
1817 q->mnt_mountpoint = mnt->mnt_mountpoint;
1820 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1822 if (!is_subdir(r->mnt_mountpoint, dentry))
1825 for (s = r; s; s = next_mnt(s, r)) {
1826 if (!(flag & CL_COPY_UNBINDABLE) &&
1827 IS_MNT_UNBINDABLE(s)) {
1828 if (s->mnt.mnt_flags & MNT_LOCKED) {
1829 /* Both unbindable and locked. */
1830 q = ERR_PTR(-EPERM);
1833 s = skip_mnt_tree(s);
1837 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1838 is_mnt_ns_file(s->mnt.mnt_root)) {
1839 s = skip_mnt_tree(s);
1842 while (p != s->mnt_parent) {
1848 q = clone_mnt(p, p->mnt.mnt_root, flag);
1852 list_add_tail(&q->mnt_list, &res->mnt_list);
1853 attach_mnt(q, parent, p->mnt_mp);
1854 unlock_mount_hash();
1861 umount_tree(res, UMOUNT_SYNC);
1862 unlock_mount_hash();
1867 /* Caller should check returned pointer for errors */
1869 struct vfsmount *collect_mounts(const struct path *path)
1873 if (!check_mnt(real_mount(path->mnt)))
1874 tree = ERR_PTR(-EINVAL);
1876 tree = copy_tree(real_mount(path->mnt), path->dentry,
1877 CL_COPY_ALL | CL_PRIVATE);
1880 return ERR_CAST(tree);
1884 static void free_mnt_ns(struct mnt_namespace *);
1885 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1887 void dissolve_on_fput(struct vfsmount *mnt)
1889 struct mnt_namespace *ns;
1892 ns = real_mount(mnt)->mnt_ns;
1895 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1899 unlock_mount_hash();
1905 void drop_collected_mounts(struct vfsmount *mnt)
1909 umount_tree(real_mount(mnt), 0);
1910 unlock_mount_hash();
1915 * clone_private_mount - create a private clone of a path
1917 * This creates a new vfsmount, which will be the clone of @path. The new will
1918 * not be attached anywhere in the namespace and will be private (i.e. changes
1919 * to the originating mount won't be propagated into this).
1921 * Release with mntput().
1923 struct vfsmount *clone_private_mount(const struct path *path)
1925 struct mount *old_mnt = real_mount(path->mnt);
1926 struct mount *new_mnt;
1928 if (IS_MNT_UNBINDABLE(old_mnt))
1929 return ERR_PTR(-EINVAL);
1931 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1932 if (IS_ERR(new_mnt))
1933 return ERR_CAST(new_mnt);
1935 return &new_mnt->mnt;
1937 EXPORT_SYMBOL_GPL(clone_private_mount);
1939 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1940 struct vfsmount *root)
1943 int res = f(root, arg);
1946 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1947 res = f(&mnt->mnt, arg);
1954 static void lock_mnt_tree(struct mount *mnt)
1958 for (p = mnt; p; p = next_mnt(p, mnt)) {
1959 int flags = p->mnt.mnt_flags;
1960 /* Don't allow unprivileged users to change mount flags */
1961 flags |= MNT_LOCK_ATIME;
1963 if (flags & MNT_READONLY)
1964 flags |= MNT_LOCK_READONLY;
1966 if (flags & MNT_NODEV)
1967 flags |= MNT_LOCK_NODEV;
1969 if (flags & MNT_NOSUID)
1970 flags |= MNT_LOCK_NOSUID;
1972 if (flags & MNT_NOEXEC)
1973 flags |= MNT_LOCK_NOEXEC;
1974 /* Don't allow unprivileged users to reveal what is under a mount */
1975 if (list_empty(&p->mnt_expire))
1976 flags |= MNT_LOCKED;
1977 p->mnt.mnt_flags = flags;
1981 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1985 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1986 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1987 mnt_release_group_id(p);
1991 static int invent_group_ids(struct mount *mnt, bool recurse)
1995 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1996 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1997 int err = mnt_alloc_group_id(p);
1999 cleanup_group_ids(mnt, p);
2008 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2010 unsigned int max = READ_ONCE(sysctl_mount_max);
2011 unsigned int mounts = 0, old, pending, sum;
2014 for (p = mnt; p; p = next_mnt(p, mnt))
2018 pending = ns->pending_mounts;
2019 sum = old + pending;
2023 (mounts > (max - sum)))
2026 ns->pending_mounts = pending + mounts;
2031 * @source_mnt : mount tree to be attached
2032 * @nd : place the mount tree @source_mnt is attached
2033 * @parent_nd : if non-null, detach the source_mnt from its parent and
2034 * store the parent mount and mountpoint dentry.
2035 * (done when source_mnt is moved)
2037 * NOTE: in the table below explains the semantics when a source mount
2038 * of a given type is attached to a destination mount of a given type.
2039 * ---------------------------------------------------------------------------
2040 * | BIND MOUNT OPERATION |
2041 * |**************************************************************************
2042 * | source-->| shared | private | slave | unbindable |
2046 * |**************************************************************************
2047 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2049 * |non-shared| shared (+) | private | slave (*) | invalid |
2050 * ***************************************************************************
2051 * A bind operation clones the source mount and mounts the clone on the
2052 * destination mount.
2054 * (++) the cloned mount is propagated to all the mounts in the propagation
2055 * tree of the destination mount and the cloned mount is added to
2056 * the peer group of the source mount.
2057 * (+) the cloned mount is created under the destination mount and is marked
2058 * as shared. The cloned mount is added to the peer group of the source
2060 * (+++) the mount is propagated to all the mounts in the propagation tree
2061 * of the destination mount and the cloned mount is made slave
2062 * of the same master as that of the source mount. The cloned mount
2063 * is marked as 'shared and slave'.
2064 * (*) the cloned mount is made a slave of the same master as that of the
2067 * ---------------------------------------------------------------------------
2068 * | MOVE MOUNT OPERATION |
2069 * |**************************************************************************
2070 * | source-->| shared | private | slave | unbindable |
2074 * |**************************************************************************
2075 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2077 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2078 * ***************************************************************************
2080 * (+) the mount is moved to the destination. And is then propagated to
2081 * all the mounts in the propagation tree of the destination mount.
2082 * (+*) the mount is moved to the destination.
2083 * (+++) the mount is moved to the destination and is then propagated to
2084 * all the mounts belonging to the destination mount's propagation tree.
2085 * the mount is marked as 'shared and slave'.
2086 * (*) the mount continues to be a slave at the new location.
2088 * if the source mount is a tree, the operations explained above is
2089 * applied to each mount in the tree.
2090 * Must be called without spinlocks held, since this function can sleep
2093 static int attach_recursive_mnt(struct mount *source_mnt,
2094 struct mount *dest_mnt,
2095 struct mountpoint *dest_mp,
2098 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2099 HLIST_HEAD(tree_list);
2100 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2101 struct mountpoint *smp;
2102 struct mount *child, *p;
2103 struct hlist_node *n;
2106 /* Preallocate a mountpoint in case the new mounts need
2107 * to be tucked under other mounts.
2109 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2111 return PTR_ERR(smp);
2113 /* Is there space to add these mounts to the mount namespace? */
2115 err = count_mounts(ns, source_mnt);
2120 if (IS_MNT_SHARED(dest_mnt)) {
2121 err = invent_group_ids(source_mnt, true);
2124 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2127 goto out_cleanup_ids;
2128 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2134 unhash_mnt(source_mnt);
2135 attach_mnt(source_mnt, dest_mnt, dest_mp);
2136 touch_mnt_namespace(source_mnt->mnt_ns);
2138 if (source_mnt->mnt_ns) {
2139 /* move from anon - the caller will destroy */
2140 list_del_init(&source_mnt->mnt_ns->list);
2142 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2143 commit_tree(source_mnt);
2146 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2148 hlist_del_init(&child->mnt_hash);
2149 q = __lookup_mnt(&child->mnt_parent->mnt,
2150 child->mnt_mountpoint);
2152 mnt_change_mountpoint(child, smp, q);
2153 /* Notice when we are propagating across user namespaces */
2154 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2155 lock_mnt_tree(child);
2156 child->mnt.mnt_flags &= ~MNT_LOCKED;
2159 put_mountpoint(smp);
2160 unlock_mount_hash();
2165 while (!hlist_empty(&tree_list)) {
2166 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2167 child->mnt_parent->mnt_ns->pending_mounts = 0;
2168 umount_tree(child, UMOUNT_SYNC);
2170 unlock_mount_hash();
2171 cleanup_group_ids(source_mnt, NULL);
2173 ns->pending_mounts = 0;
2175 read_seqlock_excl(&mount_lock);
2176 put_mountpoint(smp);
2177 read_sequnlock_excl(&mount_lock);
2182 static struct mountpoint *lock_mount(struct path *path)
2184 struct vfsmount *mnt;
2185 struct dentry *dentry = path->dentry;
2187 inode_lock(dentry->d_inode);
2188 if (unlikely(cant_mount(dentry))) {
2189 inode_unlock(dentry->d_inode);
2190 return ERR_PTR(-ENOENT);
2193 mnt = lookup_mnt(path);
2195 struct mountpoint *mp = get_mountpoint(dentry);
2198 inode_unlock(dentry->d_inode);
2204 inode_unlock(path->dentry->d_inode);
2207 dentry = path->dentry = dget(mnt->mnt_root);
2211 static void unlock_mount(struct mountpoint *where)
2213 struct dentry *dentry = where->m_dentry;
2215 read_seqlock_excl(&mount_lock);
2216 put_mountpoint(where);
2217 read_sequnlock_excl(&mount_lock);
2220 inode_unlock(dentry->d_inode);
2223 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2225 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2228 if (d_is_dir(mp->m_dentry) !=
2229 d_is_dir(mnt->mnt.mnt_root))
2232 return attach_recursive_mnt(mnt, p, mp, false);
2236 * Sanity check the flags to change_mnt_propagation.
2239 static int flags_to_propagation_type(int ms_flags)
2241 int type = ms_flags & ~(MS_REC | MS_SILENT);
2243 /* Fail if any non-propagation flags are set */
2244 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2246 /* Only one propagation flag should be set */
2247 if (!is_power_of_2(type))
2253 * recursively change the type of the mountpoint.
2255 static int do_change_type(struct path *path, int ms_flags)
2258 struct mount *mnt = real_mount(path->mnt);
2259 int recurse = ms_flags & MS_REC;
2263 if (path->dentry != path->mnt->mnt_root)
2266 type = flags_to_propagation_type(ms_flags);
2271 if (type == MS_SHARED) {
2272 err = invent_group_ids(mnt, recurse);
2278 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2279 change_mnt_propagation(m, type);
2280 unlock_mount_hash();
2287 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2289 struct mount *child;
2290 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2291 if (!is_subdir(child->mnt_mountpoint, dentry))
2294 if (child->mnt.mnt_flags & MNT_LOCKED)
2300 static struct mount *__do_loopback(struct path *old_path, int recurse)
2302 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2304 if (IS_MNT_UNBINDABLE(old))
2307 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2310 if (!recurse && has_locked_children(old, old_path->dentry))
2314 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2316 mnt = clone_mnt(old, old_path->dentry, 0);
2319 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2325 * do loopback mount.
2327 static int do_loopback(struct path *path, const char *old_name,
2330 struct path old_path;
2331 struct mount *mnt = NULL, *parent;
2332 struct mountpoint *mp;
2334 if (!old_name || !*old_name)
2336 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2341 if (mnt_ns_loop(old_path.dentry))
2344 mp = lock_mount(path);
2350 parent = real_mount(path->mnt);
2351 if (!check_mnt(parent))
2354 mnt = __do_loopback(&old_path, recurse);
2360 err = graft_tree(mnt, parent, mp);
2363 umount_tree(mnt, UMOUNT_SYNC);
2364 unlock_mount_hash();
2369 path_put(&old_path);
2373 static struct file *open_detached_copy(struct path *path, bool recursive)
2375 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2376 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2377 struct mount *mnt, *p;
2381 return ERR_CAST(ns);
2384 mnt = __do_loopback(path, recursive);
2388 return ERR_CAST(mnt);
2392 for (p = mnt; p; p = next_mnt(p, mnt)) {
2397 list_add_tail(&ns->list, &mnt->mnt_list);
2399 unlock_mount_hash();
2403 path->mnt = &mnt->mnt;
2404 file = dentry_open(path, O_PATH, current_cred());
2406 dissolve_on_fput(path->mnt);
2408 file->f_mode |= FMODE_NEED_UNMOUNT;
2412 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2416 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2417 bool detached = flags & OPEN_TREE_CLONE;
2421 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2423 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2424 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2428 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2431 if (flags & AT_NO_AUTOMOUNT)
2432 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2433 if (flags & AT_SYMLINK_NOFOLLOW)
2434 lookup_flags &= ~LOOKUP_FOLLOW;
2435 if (flags & AT_EMPTY_PATH)
2436 lookup_flags |= LOOKUP_EMPTY;
2438 if (detached && !may_mount())
2441 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2445 error = user_path_at(dfd, filename, lookup_flags, &path);
2446 if (unlikely(error)) {
2447 file = ERR_PTR(error);
2450 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2452 file = dentry_open(&path, O_PATH, current_cred());
2457 return PTR_ERR(file);
2459 fd_install(fd, file);
2464 * Don't allow locked mount flags to be cleared.
2466 * No locks need to be held here while testing the various MNT_LOCK
2467 * flags because those flags can never be cleared once they are set.
2469 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2471 unsigned int fl = mnt->mnt.mnt_flags;
2473 if ((fl & MNT_LOCK_READONLY) &&
2474 !(mnt_flags & MNT_READONLY))
2477 if ((fl & MNT_LOCK_NODEV) &&
2478 !(mnt_flags & MNT_NODEV))
2481 if ((fl & MNT_LOCK_NOSUID) &&
2482 !(mnt_flags & MNT_NOSUID))
2485 if ((fl & MNT_LOCK_NOEXEC) &&
2486 !(mnt_flags & MNT_NOEXEC))
2489 if ((fl & MNT_LOCK_ATIME) &&
2490 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2496 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2498 bool readonly_request = (mnt_flags & MNT_READONLY);
2500 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2503 if (readonly_request)
2504 return mnt_make_readonly(mnt);
2506 return __mnt_unmake_readonly(mnt);
2510 * Update the user-settable attributes on a mount. The caller must hold
2511 * sb->s_umount for writing.
2513 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2516 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2517 mnt->mnt.mnt_flags = mnt_flags;
2518 touch_mnt_namespace(mnt->mnt_ns);
2519 unlock_mount_hash();
2522 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2524 struct super_block *sb = mnt->mnt_sb;
2526 if (!__mnt_is_readonly(mnt) &&
2527 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2528 char *buf = (char *)__get_free_page(GFP_KERNEL);
2529 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2532 time64_to_tm(sb->s_time_max, 0, &tm);
2534 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2536 is_mounted(mnt) ? "remounted" : "mounted",
2538 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2540 free_page((unsigned long)buf);
2545 * Handle reconfiguration of the mountpoint only without alteration of the
2546 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2549 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2551 struct super_block *sb = path->mnt->mnt_sb;
2552 struct mount *mnt = real_mount(path->mnt);
2555 if (!check_mnt(mnt))
2558 if (path->dentry != mnt->mnt.mnt_root)
2561 if (!can_change_locked_flags(mnt, mnt_flags))
2564 down_write(&sb->s_umount);
2565 ret = change_mount_ro_state(mnt, mnt_flags);
2567 set_mount_attributes(mnt, mnt_flags);
2568 up_write(&sb->s_umount);
2570 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2576 * change filesystem flags. dir should be a physical root of filesystem.
2577 * If you've mounted a non-root directory somewhere and want to do remount
2578 * on it - tough luck.
2580 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2581 int mnt_flags, void *data)
2584 struct super_block *sb = path->mnt->mnt_sb;
2585 struct mount *mnt = real_mount(path->mnt);
2586 struct fs_context *fc;
2588 if (!check_mnt(mnt))
2591 if (path->dentry != path->mnt->mnt_root)
2594 if (!can_change_locked_flags(mnt, mnt_flags))
2597 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2601 err = parse_monolithic_mount_data(fc, data);
2603 down_write(&sb->s_umount);
2605 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2606 err = reconfigure_super(fc);
2608 set_mount_attributes(mnt, mnt_flags);
2610 up_write(&sb->s_umount);
2613 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2619 static inline int tree_contains_unbindable(struct mount *mnt)
2622 for (p = mnt; p; p = next_mnt(p, mnt)) {
2623 if (IS_MNT_UNBINDABLE(p))
2630 * Check that there aren't references to earlier/same mount namespaces in the
2631 * specified subtree. Such references can act as pins for mount namespaces
2632 * that aren't checked by the mount-cycle checking code, thereby allowing
2633 * cycles to be made.
2635 static bool check_for_nsfs_mounts(struct mount *subtree)
2641 for (p = subtree; p; p = next_mnt(p, subtree))
2642 if (mnt_ns_loop(p->mnt.mnt_root))
2647 unlock_mount_hash();
2651 static int do_move_mount(struct path *old_path, struct path *new_path)
2653 struct mnt_namespace *ns;
2656 struct mount *parent;
2657 struct mountpoint *mp, *old_mp;
2661 mp = lock_mount(new_path);
2665 old = real_mount(old_path->mnt);
2666 p = real_mount(new_path->mnt);
2667 parent = old->mnt_parent;
2668 attached = mnt_has_parent(old);
2669 old_mp = old->mnt_mp;
2673 /* The mountpoint must be in our namespace. */
2677 /* The thing moved must be mounted... */
2678 if (!is_mounted(&old->mnt))
2681 /* ... and either ours or the root of anon namespace */
2682 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2685 if (old->mnt.mnt_flags & MNT_LOCKED)
2688 if (old_path->dentry != old_path->mnt->mnt_root)
2691 if (d_is_dir(new_path->dentry) !=
2692 d_is_dir(old_path->dentry))
2695 * Don't move a mount residing in a shared parent.
2697 if (attached && IS_MNT_SHARED(parent))
2700 * Don't move a mount tree containing unbindable mounts to a destination
2701 * mount which is shared.
2703 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2706 if (!check_for_nsfs_mounts(old))
2708 for (; mnt_has_parent(p); p = p->mnt_parent)
2712 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2717 /* if the mount is moved, it should no longer be expire
2719 list_del_init(&old->mnt_expire);
2721 put_mountpoint(old_mp);
2726 mntput_no_expire(parent);
2733 static int do_move_mount_old(struct path *path, const char *old_name)
2735 struct path old_path;
2738 if (!old_name || !*old_name)
2741 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2745 err = do_move_mount(&old_path, path);
2746 path_put(&old_path);
2751 * add a mount into a namespace's mount tree
2753 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2754 struct path *path, int mnt_flags)
2756 struct mount *parent = real_mount(path->mnt);
2758 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2760 if (unlikely(!check_mnt(parent))) {
2761 /* that's acceptable only for automounts done in private ns */
2762 if (!(mnt_flags & MNT_SHRINKABLE))
2764 /* ... and for those we'd better have mountpoint still alive */
2765 if (!parent->mnt_ns)
2769 /* Refuse the same filesystem on the same mount point */
2770 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2771 path->mnt->mnt_root == path->dentry)
2774 if (d_is_symlink(newmnt->mnt.mnt_root))
2777 newmnt->mnt.mnt_flags = mnt_flags;
2778 return graft_tree(newmnt, parent, mp);
2781 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2784 * Create a new mount using a superblock configuration and request it
2785 * be added to the namespace tree.
2787 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2788 unsigned int mnt_flags)
2790 struct vfsmount *mnt;
2791 struct mountpoint *mp;
2792 struct super_block *sb = fc->root->d_sb;
2795 error = security_sb_kern_mount(sb);
2796 if (!error && mount_too_revealing(sb, &mnt_flags))
2799 if (unlikely(error)) {
2804 up_write(&sb->s_umount);
2806 mnt = vfs_create_mount(fc);
2808 return PTR_ERR(mnt);
2810 mnt_warn_timestamp_expiry(mountpoint, mnt);
2812 mp = lock_mount(mountpoint);
2817 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2825 * create a new mount for userspace and request it to be added into the
2828 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2829 int mnt_flags, const char *name, void *data)
2831 struct file_system_type *type;
2832 struct fs_context *fc;
2833 const char *subtype = NULL;
2839 type = get_fs_type(fstype);
2843 if (type->fs_flags & FS_HAS_SUBTYPE) {
2844 subtype = strchr(fstype, '.');
2848 put_filesystem(type);
2854 fc = fs_context_for_mount(type, sb_flags);
2855 put_filesystem(type);
2860 err = vfs_parse_fs_string(fc, "subtype",
2861 subtype, strlen(subtype));
2863 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2865 err = parse_monolithic_mount_data(fc, data);
2866 if (!err && !mount_capable(fc))
2869 err = vfs_get_tree(fc);
2871 err = do_new_mount_fc(fc, path, mnt_flags);
2877 int finish_automount(struct vfsmount *m, struct path *path)
2879 struct dentry *dentry = path->dentry;
2880 struct mountpoint *mp;
2889 mnt = real_mount(m);
2890 /* The new mount record should have at least 2 refs to prevent it being
2891 * expired before we get a chance to add it
2893 BUG_ON(mnt_get_count(mnt) < 2);
2895 if (m->mnt_sb == path->mnt->mnt_sb &&
2896 m->mnt_root == dentry) {
2902 * we don't want to use lock_mount() - in this case finding something
2903 * that overmounts our mountpoint to be means "quitely drop what we've
2904 * got", not "try to mount it on top".
2906 inode_lock(dentry->d_inode);
2908 if (unlikely(cant_mount(dentry))) {
2910 goto discard_locked;
2913 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2916 goto discard_locked;
2919 mp = get_mountpoint(dentry);
2922 goto discard_locked;
2925 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2934 inode_unlock(dentry->d_inode);
2936 /* remove m from any expiration list it may be on */
2937 if (!list_empty(&mnt->mnt_expire)) {
2939 list_del_init(&mnt->mnt_expire);
2948 * mnt_set_expiry - Put a mount on an expiration list
2949 * @mnt: The mount to list.
2950 * @expiry_list: The list to add the mount to.
2952 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2956 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2960 EXPORT_SYMBOL(mnt_set_expiry);
2963 * process a list of expirable mountpoints with the intent of discarding any
2964 * mountpoints that aren't in use and haven't been touched since last we came
2967 void mark_mounts_for_expiry(struct list_head *mounts)
2969 struct mount *mnt, *next;
2970 LIST_HEAD(graveyard);
2972 if (list_empty(mounts))
2978 /* extract from the expiration list every vfsmount that matches the
2979 * following criteria:
2980 * - only referenced by its parent vfsmount
2981 * - still marked for expiry (marked on the last call here; marks are
2982 * cleared by mntput())
2984 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2985 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2986 propagate_mount_busy(mnt, 1))
2988 list_move(&mnt->mnt_expire, &graveyard);
2990 while (!list_empty(&graveyard)) {
2991 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2992 touch_mnt_namespace(mnt->mnt_ns);
2993 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2995 unlock_mount_hash();
2999 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3002 * Ripoff of 'select_parent()'
3004 * search the list of submounts for a given mountpoint, and move any
3005 * shrinkable submounts to the 'graveyard' list.
3007 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3009 struct mount *this_parent = parent;
3010 struct list_head *next;
3014 next = this_parent->mnt_mounts.next;
3016 while (next != &this_parent->mnt_mounts) {
3017 struct list_head *tmp = next;
3018 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3021 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3024 * Descend a level if the d_mounts list is non-empty.
3026 if (!list_empty(&mnt->mnt_mounts)) {
3031 if (!propagate_mount_busy(mnt, 1)) {
3032 list_move_tail(&mnt->mnt_expire, graveyard);
3037 * All done at this level ... ascend and resume the search
3039 if (this_parent != parent) {
3040 next = this_parent->mnt_child.next;
3041 this_parent = this_parent->mnt_parent;
3048 * process a list of expirable mountpoints with the intent of discarding any
3049 * submounts of a specific parent mountpoint
3051 * mount_lock must be held for write
3053 static void shrink_submounts(struct mount *mnt)
3055 LIST_HEAD(graveyard);
3058 /* extract submounts of 'mountpoint' from the expiration list */
3059 while (select_submounts(mnt, &graveyard)) {
3060 while (!list_empty(&graveyard)) {
3061 m = list_first_entry(&graveyard, struct mount,
3063 touch_mnt_namespace(m->mnt_ns);
3064 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3069 void *copy_mount_options(const void __user * data)
3077 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3079 return ERR_PTR(-ENOMEM);
3081 size = PAGE_SIZE - offset_in_page(data);
3083 if (copy_from_user(copy, data, size)) {
3085 return ERR_PTR(-EFAULT);
3087 if (size != PAGE_SIZE) {
3088 if (copy_from_user(copy + size, data + size, PAGE_SIZE - size))
3089 memset(copy + size, 0, PAGE_SIZE - size);
3094 char *copy_mount_string(const void __user *data)
3096 return data ? strndup_user(data, PATH_MAX) : NULL;
3100 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3101 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3103 * data is a (void *) that can point to any structure up to
3104 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3105 * information (or be NULL).
3107 * Pre-0.97 versions of mount() didn't have a flags word.
3108 * When the flags word was introduced its top half was required
3109 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3110 * Therefore, if this magic number is present, it carries no information
3111 * and must be discarded.
3113 long do_mount(const char *dev_name, const char __user *dir_name,
3114 const char *type_page, unsigned long flags, void *data_page)
3117 unsigned int mnt_flags = 0, sb_flags;
3121 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3122 flags &= ~MS_MGC_MSK;
3124 /* Basic sanity checks */
3126 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3128 if (flags & MS_NOUSER)
3131 /* ... and get the mountpoint */
3132 retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3136 retval = security_sb_mount(dev_name, &path,
3137 type_page, flags, data_page);
3138 if (!retval && !may_mount())
3140 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
3145 /* Default to relatime unless overriden */
3146 if (!(flags & MS_NOATIME))
3147 mnt_flags |= MNT_RELATIME;
3149 /* Separate the per-mountpoint flags */
3150 if (flags & MS_NOSUID)
3151 mnt_flags |= MNT_NOSUID;
3152 if (flags & MS_NODEV)
3153 mnt_flags |= MNT_NODEV;
3154 if (flags & MS_NOEXEC)
3155 mnt_flags |= MNT_NOEXEC;
3156 if (flags & MS_NOATIME)
3157 mnt_flags |= MNT_NOATIME;
3158 if (flags & MS_NODIRATIME)
3159 mnt_flags |= MNT_NODIRATIME;
3160 if (flags & MS_STRICTATIME)
3161 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3162 if (flags & MS_RDONLY)
3163 mnt_flags |= MNT_READONLY;
3165 /* The default atime for remount is preservation */
3166 if ((flags & MS_REMOUNT) &&
3167 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3168 MS_STRICTATIME)) == 0)) {
3169 mnt_flags &= ~MNT_ATIME_MASK;
3170 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
3173 sb_flags = flags & (SB_RDONLY |
3182 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3183 retval = do_reconfigure_mnt(&path, mnt_flags);
3184 else if (flags & MS_REMOUNT)
3185 retval = do_remount(&path, flags, sb_flags, mnt_flags,
3187 else if (flags & MS_BIND)
3188 retval = do_loopback(&path, dev_name, flags & MS_REC);
3189 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3190 retval = do_change_type(&path, flags);
3191 else if (flags & MS_MOVE)
3192 retval = do_move_mount_old(&path, dev_name);
3194 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
3195 dev_name, data_page);
3201 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3203 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3206 static void dec_mnt_namespaces(struct ucounts *ucounts)
3208 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3211 static void free_mnt_ns(struct mnt_namespace *ns)
3213 if (!is_anon_ns(ns))
3214 ns_free_inum(&ns->ns);
3215 dec_mnt_namespaces(ns->ucounts);
3216 put_user_ns(ns->user_ns);
3221 * Assign a sequence number so we can detect when we attempt to bind
3222 * mount a reference to an older mount namespace into the current
3223 * mount namespace, preventing reference counting loops. A 64bit
3224 * number incrementing at 10Ghz will take 12,427 years to wrap which
3225 * is effectively never, so we can ignore the possibility.
3227 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3229 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3231 struct mnt_namespace *new_ns;
3232 struct ucounts *ucounts;
3235 ucounts = inc_mnt_namespaces(user_ns);
3237 return ERR_PTR(-ENOSPC);
3239 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3241 dec_mnt_namespaces(ucounts);
3242 return ERR_PTR(-ENOMEM);
3245 ret = ns_alloc_inum(&new_ns->ns);
3248 dec_mnt_namespaces(ucounts);
3249 return ERR_PTR(ret);
3252 new_ns->ns.ops = &mntns_operations;
3254 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3255 atomic_set(&new_ns->count, 1);
3256 INIT_LIST_HEAD(&new_ns->list);
3257 init_waitqueue_head(&new_ns->poll);
3258 spin_lock_init(&new_ns->ns_lock);
3259 new_ns->user_ns = get_user_ns(user_ns);
3260 new_ns->ucounts = ucounts;
3265 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3266 struct user_namespace *user_ns, struct fs_struct *new_fs)
3268 struct mnt_namespace *new_ns;
3269 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3270 struct mount *p, *q;
3277 if (likely(!(flags & CLONE_NEWNS))) {
3284 new_ns = alloc_mnt_ns(user_ns, false);
3289 /* First pass: copy the tree topology */
3290 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3291 if (user_ns != ns->user_ns)
3292 copy_flags |= CL_SHARED_TO_SLAVE;
3293 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3296 free_mnt_ns(new_ns);
3297 return ERR_CAST(new);
3299 if (user_ns != ns->user_ns) {
3302 unlock_mount_hash();
3305 list_add_tail(&new_ns->list, &new->mnt_list);
3308 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3309 * as belonging to new namespace. We have already acquired a private
3310 * fs_struct, so tsk->fs->lock is not needed.
3318 if (&p->mnt == new_fs->root.mnt) {
3319 new_fs->root.mnt = mntget(&q->mnt);
3322 if (&p->mnt == new_fs->pwd.mnt) {
3323 new_fs->pwd.mnt = mntget(&q->mnt);
3327 p = next_mnt(p, old);
3328 q = next_mnt(q, new);
3331 while (p->mnt.mnt_root != q->mnt.mnt_root)
3332 p = next_mnt(p, old);
3344 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3346 struct mount *mnt = real_mount(m);
3347 struct mnt_namespace *ns;
3348 struct super_block *s;
3352 ns = alloc_mnt_ns(&init_user_ns, true);
3355 return ERR_CAST(ns);
3360 list_add(&mnt->mnt_list, &ns->list);
3362 err = vfs_path_lookup(m->mnt_root, m,
3363 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3368 return ERR_PTR(err);
3370 /* trade a vfsmount reference for active sb one */
3371 s = path.mnt->mnt_sb;
3372 atomic_inc(&s->s_active);
3374 /* lock the sucker */
3375 down_write(&s->s_umount);
3376 /* ... and return the root of (sub)tree on it */
3379 EXPORT_SYMBOL(mount_subtree);
3381 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3382 char __user *, type, unsigned long, flags, void __user *, data)
3389 kernel_type = copy_mount_string(type);
3390 ret = PTR_ERR(kernel_type);
3391 if (IS_ERR(kernel_type))
3394 kernel_dev = copy_mount_string(dev_name);
3395 ret = PTR_ERR(kernel_dev);
3396 if (IS_ERR(kernel_dev))
3399 options = copy_mount_options(data);
3400 ret = PTR_ERR(options);
3401 if (IS_ERR(options))
3404 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3416 * Create a kernel mount representation for a new, prepared superblock
3417 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3419 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3420 unsigned int, attr_flags)
3422 struct mnt_namespace *ns;
3423 struct fs_context *fc;
3425 struct path newmount;
3428 unsigned int mnt_flags = 0;
3434 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3437 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3442 MOUNT_ATTR_NODIRATIME))
3445 if (attr_flags & MOUNT_ATTR_RDONLY)
3446 mnt_flags |= MNT_READONLY;
3447 if (attr_flags & MOUNT_ATTR_NOSUID)
3448 mnt_flags |= MNT_NOSUID;
3449 if (attr_flags & MOUNT_ATTR_NODEV)
3450 mnt_flags |= MNT_NODEV;
3451 if (attr_flags & MOUNT_ATTR_NOEXEC)
3452 mnt_flags |= MNT_NOEXEC;
3453 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3454 mnt_flags |= MNT_NODIRATIME;
3456 switch (attr_flags & MOUNT_ATTR__ATIME) {
3457 case MOUNT_ATTR_STRICTATIME:
3459 case MOUNT_ATTR_NOATIME:
3460 mnt_flags |= MNT_NOATIME;
3462 case MOUNT_ATTR_RELATIME:
3463 mnt_flags |= MNT_RELATIME;
3474 if (f.file->f_op != &fscontext_fops)
3477 fc = f.file->private_data;
3479 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3483 /* There must be a valid superblock or we can't mount it */
3489 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3490 pr_warn("VFS: Mount too revealing\n");
3495 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3499 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3502 newmount.mnt = vfs_create_mount(fc);
3503 if (IS_ERR(newmount.mnt)) {
3504 ret = PTR_ERR(newmount.mnt);
3507 newmount.dentry = dget(fc->root);
3508 newmount.mnt->mnt_flags = mnt_flags;
3510 /* We've done the mount bit - now move the file context into more or
3511 * less the same state as if we'd done an fspick(). We don't want to
3512 * do any memory allocation or anything like that at this point as we
3513 * don't want to have to handle any errors incurred.
3515 vfs_clean_context(fc);
3517 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3522 mnt = real_mount(newmount.mnt);
3526 list_add(&mnt->mnt_list, &ns->list);
3527 mntget(newmount.mnt);
3529 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3530 * it, not just simply put it.
3532 file = dentry_open(&newmount, O_PATH, fc->cred);
3534 dissolve_on_fput(newmount.mnt);
3535 ret = PTR_ERR(file);
3538 file->f_mode |= FMODE_NEED_UNMOUNT;
3540 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3542 fd_install(ret, file);
3547 path_put(&newmount);
3549 mutex_unlock(&fc->uapi_mutex);
3556 * Move a mount from one place to another. In combination with
3557 * fsopen()/fsmount() this is used to install a new mount and in combination
3558 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3561 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3563 SYSCALL_DEFINE5(move_mount,
3564 int, from_dfd, const char __user *, from_pathname,
3565 int, to_dfd, const char __user *, to_pathname,
3566 unsigned int, flags)
3568 struct path from_path, to_path;
3569 unsigned int lflags;
3575 if (flags & ~MOVE_MOUNT__MASK)
3578 /* If someone gives a pathname, they aren't permitted to move
3579 * from an fd that requires unmount as we can't get at the flag
3580 * to clear it afterwards.
3583 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3584 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3585 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3587 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3592 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3593 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3594 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3596 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3600 ret = security_move_mount(&from_path, &to_path);
3604 ret = do_move_mount(&from_path, &to_path);
3609 path_put(&from_path);
3614 * Return true if path is reachable from root
3616 * namespace_sem or mount_lock is held
3618 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3619 const struct path *root)
3621 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3622 dentry = mnt->mnt_mountpoint;
3623 mnt = mnt->mnt_parent;
3625 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3628 bool path_is_under(const struct path *path1, const struct path *path2)
3631 read_seqlock_excl(&mount_lock);
3632 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3633 read_sequnlock_excl(&mount_lock);
3636 EXPORT_SYMBOL(path_is_under);
3639 * pivot_root Semantics:
3640 * Moves the root file system of the current process to the directory put_old,
3641 * makes new_root as the new root file system of the current process, and sets
3642 * root/cwd of all processes which had them on the current root to new_root.
3645 * The new_root and put_old must be directories, and must not be on the
3646 * same file system as the current process root. The put_old must be
3647 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3648 * pointed to by put_old must yield the same directory as new_root. No other
3649 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3651 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3652 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3653 * in this situation.
3656 * - we don't move root/cwd if they are not at the root (reason: if something
3657 * cared enough to change them, it's probably wrong to force them elsewhere)
3658 * - it's okay to pick a root that isn't the root of a file system, e.g.
3659 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3660 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3663 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3664 const char __user *, put_old)
3666 struct path new, old, root;
3667 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3668 struct mountpoint *old_mp, *root_mp;
3674 error = user_path_at(AT_FDCWD, new_root,
3675 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3679 error = user_path_at(AT_FDCWD, put_old,
3680 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3684 error = security_sb_pivotroot(&old, &new);
3688 get_fs_root(current->fs, &root);
3689 old_mp = lock_mount(&old);
3690 error = PTR_ERR(old_mp);
3695 new_mnt = real_mount(new.mnt);
3696 root_mnt = real_mount(root.mnt);
3697 old_mnt = real_mount(old.mnt);
3698 ex_parent = new_mnt->mnt_parent;
3699 root_parent = root_mnt->mnt_parent;
3700 if (IS_MNT_SHARED(old_mnt) ||
3701 IS_MNT_SHARED(ex_parent) ||
3702 IS_MNT_SHARED(root_parent))
3704 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3706 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3709 if (d_unlinked(new.dentry))
3712 if (new_mnt == root_mnt || old_mnt == root_mnt)
3713 goto out4; /* loop, on the same file system */
3715 if (root.mnt->mnt_root != root.dentry)
3716 goto out4; /* not a mountpoint */
3717 if (!mnt_has_parent(root_mnt))
3718 goto out4; /* not attached */
3719 if (new.mnt->mnt_root != new.dentry)
3720 goto out4; /* not a mountpoint */
3721 if (!mnt_has_parent(new_mnt))
3722 goto out4; /* not attached */
3723 /* make sure we can reach put_old from new_root */
3724 if (!is_path_reachable(old_mnt, old.dentry, &new))
3726 /* make certain new is below the root */
3727 if (!is_path_reachable(new_mnt, new.dentry, &root))
3730 umount_mnt(new_mnt);
3731 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3732 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3733 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3734 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3736 /* mount old root on put_old */
3737 attach_mnt(root_mnt, old_mnt, old_mp);
3738 /* mount new_root on / */
3739 attach_mnt(new_mnt, root_parent, root_mp);
3740 mnt_add_count(root_parent, -1);
3741 touch_mnt_namespace(current->nsproxy->mnt_ns);
3742 /* A moved mount should not expire automatically */
3743 list_del_init(&new_mnt->mnt_expire);
3744 put_mountpoint(root_mp);
3745 unlock_mount_hash();
3746 chroot_fs_refs(&root, &new);
3749 unlock_mount(old_mp);
3751 mntput_no_expire(ex_parent);
3762 static void __init init_mount_tree(void)
3764 struct vfsmount *mnt;
3766 struct mnt_namespace *ns;
3769 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3771 panic("Can't create rootfs");
3773 ns = alloc_mnt_ns(&init_user_ns, false);
3775 panic("Can't allocate initial namespace");
3776 m = real_mount(mnt);
3780 list_add(&m->mnt_list, &ns->list);
3781 init_task.nsproxy->mnt_ns = ns;
3785 root.dentry = mnt->mnt_root;
3786 mnt->mnt_flags |= MNT_LOCKED;
3788 set_fs_pwd(current->fs, &root);
3789 set_fs_root(current->fs, &root);
3792 void __init mnt_init(void)
3796 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3797 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3799 mount_hashtable = alloc_large_system_hash("Mount-cache",
3800 sizeof(struct hlist_head),
3803 &m_hash_shift, &m_hash_mask, 0, 0);
3804 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3805 sizeof(struct hlist_head),
3808 &mp_hash_shift, &mp_hash_mask, 0, 0);
3810 if (!mount_hashtable || !mountpoint_hashtable)
3811 panic("Failed to allocate mount hash table\n");
3817 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3819 fs_kobj = kobject_create_and_add("fs", NULL);
3821 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3827 void put_mnt_ns(struct mnt_namespace *ns)
3829 if (!atomic_dec_and_test(&ns->count))
3831 drop_collected_mounts(&ns->root->mnt);
3835 struct vfsmount *kern_mount(struct file_system_type *type)
3837 struct vfsmount *mnt;
3838 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3841 * it is a longterm mount, don't release mnt until
3842 * we unmount before file sys is unregistered
3844 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3848 EXPORT_SYMBOL_GPL(kern_mount);
3850 void kern_unmount(struct vfsmount *mnt)
3852 /* release long term mount so mount point can be released */
3853 if (!IS_ERR_OR_NULL(mnt)) {
3854 real_mount(mnt)->mnt_ns = NULL;
3855 synchronize_rcu(); /* yecchhh... */
3859 EXPORT_SYMBOL(kern_unmount);
3861 bool our_mnt(struct vfsmount *mnt)
3863 return check_mnt(real_mount(mnt));
3866 bool current_chrooted(void)
3868 /* Does the current process have a non-standard root */
3869 struct path ns_root;
3870 struct path fs_root;
3873 /* Find the namespace root */
3874 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3875 ns_root.dentry = ns_root.mnt->mnt_root;
3877 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3880 get_fs_root(current->fs, &fs_root);
3882 chrooted = !path_equal(&fs_root, &ns_root);
3890 static bool mnt_already_visible(struct mnt_namespace *ns,
3891 const struct super_block *sb,
3894 int new_flags = *new_mnt_flags;
3896 bool visible = false;
3898 down_read(&namespace_sem);
3900 list_for_each_entry(mnt, &ns->list, mnt_list) {
3901 struct mount *child;
3904 if (mnt_is_cursor(mnt))
3907 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3910 /* This mount is not fully visible if it's root directory
3911 * is not the root directory of the filesystem.
3913 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3916 /* A local view of the mount flags */
3917 mnt_flags = mnt->mnt.mnt_flags;
3919 /* Don't miss readonly hidden in the superblock flags */
3920 if (sb_rdonly(mnt->mnt.mnt_sb))
3921 mnt_flags |= MNT_LOCK_READONLY;
3923 /* Verify the mount flags are equal to or more permissive
3924 * than the proposed new mount.
3926 if ((mnt_flags & MNT_LOCK_READONLY) &&
3927 !(new_flags & MNT_READONLY))
3929 if ((mnt_flags & MNT_LOCK_ATIME) &&
3930 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3933 /* This mount is not fully visible if there are any
3934 * locked child mounts that cover anything except for
3935 * empty directories.
3937 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3938 struct inode *inode = child->mnt_mountpoint->d_inode;
3939 /* Only worry about locked mounts */
3940 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3942 /* Is the directory permanetly empty? */
3943 if (!is_empty_dir_inode(inode))
3946 /* Preserve the locked attributes */
3947 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3955 up_read(&namespace_sem);
3959 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3961 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3962 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3963 unsigned long s_iflags;
3965 if (ns->user_ns == &init_user_ns)
3968 /* Can this filesystem be too revealing? */
3969 s_iflags = sb->s_iflags;
3970 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3973 if ((s_iflags & required_iflags) != required_iflags) {
3974 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3979 return !mnt_already_visible(ns, sb, new_mnt_flags);
3982 bool mnt_may_suid(struct vfsmount *mnt)
3985 * Foreign mounts (accessed via fchdir or through /proc
3986 * symlinks) are always treated as if they are nosuid. This
3987 * prevents namespaces from trusting potentially unsafe
3988 * suid/sgid bits, file caps, or security labels that originate
3989 * in other namespaces.
3991 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3992 current_in_userns(mnt->mnt_sb->s_user_ns);
3995 static struct ns_common *mntns_get(struct task_struct *task)
3997 struct ns_common *ns = NULL;
3998 struct nsproxy *nsproxy;
4001 nsproxy = task->nsproxy;
4003 ns = &nsproxy->mnt_ns->ns;
4004 get_mnt_ns(to_mnt_ns(ns));
4011 static void mntns_put(struct ns_common *ns)
4013 put_mnt_ns(to_mnt_ns(ns));
4016 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
4018 struct fs_struct *fs = current->fs;
4019 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4023 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4024 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
4025 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
4028 if (is_anon_ns(mnt_ns))
4035 old_mnt_ns = nsproxy->mnt_ns;
4036 nsproxy->mnt_ns = mnt_ns;
4039 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4040 "/", LOOKUP_DOWN, &root);
4042 /* revert to old namespace */
4043 nsproxy->mnt_ns = old_mnt_ns;
4048 put_mnt_ns(old_mnt_ns);
4050 /* Update the pwd and root */
4051 set_fs_pwd(fs, &root);
4052 set_fs_root(fs, &root);
4058 static struct user_namespace *mntns_owner(struct ns_common *ns)
4060 return to_mnt_ns(ns)->user_ns;
4063 const struct proc_ns_operations mntns_operations = {
4065 .type = CLONE_NEWNS,
4068 .install = mntns_install,
4069 .owner = mntns_owner,