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/proc_fs.h>
29 #include <linux/task_work.h>
30 #include <linux/sched/task.h>
31 #include <uapi/linux/mount.h>
32 #include <linux/fs_context.h>
33 #include <linux/shmem_fs.h>
34 #include <linux/mnt_idmapping.h>
39 /* Maximum number of mounts in a mount namespace */
40 static unsigned int sysctl_mount_max __read_mostly = 100000;
42 static unsigned int m_hash_mask __read_mostly;
43 static unsigned int m_hash_shift __read_mostly;
44 static unsigned int mp_hash_mask __read_mostly;
45 static unsigned int mp_hash_shift __read_mostly;
47 static __initdata unsigned long mhash_entries;
48 static int __init set_mhash_entries(char *str)
52 mhash_entries = simple_strtoul(str, &str, 0);
55 __setup("mhash_entries=", set_mhash_entries);
57 static __initdata unsigned long mphash_entries;
58 static int __init set_mphash_entries(char *str)
62 mphash_entries = simple_strtoul(str, &str, 0);
65 __setup("mphash_entries=", set_mphash_entries);
68 static DEFINE_IDA(mnt_id_ida);
69 static DEFINE_IDA(mnt_group_ida);
71 static struct hlist_head *mount_hashtable __read_mostly;
72 static struct hlist_head *mountpoint_hashtable __read_mostly;
73 static struct kmem_cache *mnt_cache __read_mostly;
74 static DECLARE_RWSEM(namespace_sem);
75 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
76 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
79 unsigned int attr_set;
80 unsigned int attr_clr;
81 unsigned int propagation;
82 unsigned int lookup_flags;
84 struct user_namespace *mnt_userns;
85 struct mnt_idmap *mnt_idmap;
89 struct kobject *fs_kobj;
90 EXPORT_SYMBOL_GPL(fs_kobj);
93 * vfsmount lock may be taken for read to prevent changes to the
94 * vfsmount hash, ie. during mountpoint lookups or walking back
97 * It should be taken for write in all cases where the vfsmount
98 * tree or hash is modified or when a vfsmount structure is modified.
100 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
102 static inline void lock_mount_hash(void)
104 write_seqlock(&mount_lock);
107 static inline void unlock_mount_hash(void)
109 write_sequnlock(&mount_lock);
112 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
114 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
115 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
116 tmp = tmp + (tmp >> m_hash_shift);
117 return &mount_hashtable[tmp & m_hash_mask];
120 static inline struct hlist_head *mp_hash(struct dentry *dentry)
122 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
123 tmp = tmp + (tmp >> mp_hash_shift);
124 return &mountpoint_hashtable[tmp & mp_hash_mask];
127 static int mnt_alloc_id(struct mount *mnt)
129 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
137 static void mnt_free_id(struct mount *mnt)
139 ida_free(&mnt_id_ida, mnt->mnt_id);
143 * Allocate a new peer group ID
145 static int mnt_alloc_group_id(struct mount *mnt)
147 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
151 mnt->mnt_group_id = res;
156 * Release a peer group ID
158 void mnt_release_group_id(struct mount *mnt)
160 ida_free(&mnt_group_ida, mnt->mnt_group_id);
161 mnt->mnt_group_id = 0;
165 * vfsmount lock must be held for read
167 static inline void mnt_add_count(struct mount *mnt, int n)
170 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
179 * vfsmount lock must be held for write
181 int mnt_get_count(struct mount *mnt)
187 for_each_possible_cpu(cpu) {
188 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
193 return mnt->mnt_count;
197 static struct mount *alloc_vfsmnt(const char *name)
199 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
203 err = mnt_alloc_id(mnt);
208 mnt->mnt_devname = kstrdup_const(name,
210 if (!mnt->mnt_devname)
215 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 goto out_free_devname;
219 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
222 mnt->mnt_writers = 0;
225 INIT_HLIST_NODE(&mnt->mnt_hash);
226 INIT_LIST_HEAD(&mnt->mnt_child);
227 INIT_LIST_HEAD(&mnt->mnt_mounts);
228 INIT_LIST_HEAD(&mnt->mnt_list);
229 INIT_LIST_HEAD(&mnt->mnt_expire);
230 INIT_LIST_HEAD(&mnt->mnt_share);
231 INIT_LIST_HEAD(&mnt->mnt_slave_list);
232 INIT_LIST_HEAD(&mnt->mnt_slave);
233 INIT_HLIST_NODE(&mnt->mnt_mp_list);
234 INIT_LIST_HEAD(&mnt->mnt_umounting);
235 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
236 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
242 kfree_const(mnt->mnt_devname);
247 kmem_cache_free(mnt_cache, mnt);
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
270 bool __mnt_is_readonly(struct vfsmount *mnt)
272 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
274 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
276 static inline void mnt_inc_writers(struct mount *mnt)
279 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
285 static inline void mnt_dec_writers(struct mount *mnt)
288 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
294 static unsigned int mnt_get_writers(struct mount *mnt)
297 unsigned int count = 0;
300 for_each_possible_cpu(cpu) {
301 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
306 return mnt->mnt_writers;
310 static int mnt_is_readonly(struct vfsmount *mnt)
312 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
315 * The barrier pairs with the barrier in sb_start_ro_state_change()
316 * making sure if we don't see s_readonly_remount set yet, we also will
317 * not see any superblock / mount flag changes done by remount.
318 * It also pairs with the barrier in sb_end_ro_state_change()
319 * assuring that if we see s_readonly_remount already cleared, we will
320 * see the values of superblock / mount flags updated by remount.
323 return __mnt_is_readonly(mnt);
327 * Most r/o & frozen checks on a fs are for operations that take discrete
328 * amounts of time, like a write() or unlink(). We must keep track of when
329 * those operations start (for permission checks) and when they end, so that we
330 * can determine when writes are able to occur to a filesystem.
333 * __mnt_want_write - get write access to a mount without freeze protection
334 * @m: the mount on which to take a write
336 * This tells the low-level filesystem that a write is about to be performed to
337 * it, and makes sure that writes are allowed (mnt it read-write) before
338 * returning success. This operation does not protect against filesystem being
339 * frozen. When the write operation is finished, __mnt_drop_write() must be
340 * called. This is effectively a refcount.
342 int __mnt_want_write(struct vfsmount *m)
344 struct mount *mnt = real_mount(m);
348 mnt_inc_writers(mnt);
350 * The store to mnt_inc_writers must be visible before we pass
351 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 * incremented count after it has set MNT_WRITE_HOLD.
355 might_lock(&mount_lock.lock);
356 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
357 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
361 * This prevents priority inversion, if the task
362 * setting MNT_WRITE_HOLD got preempted on a remote
363 * CPU, and it prevents life lock if the task setting
364 * MNT_WRITE_HOLD has a lower priority and is bound to
365 * the same CPU as the task that is spinning here.
374 * The barrier pairs with the barrier sb_start_ro_state_change() making
375 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
376 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
377 * mnt_is_readonly() and bail in case we are racing with remount
381 if (mnt_is_readonly(m)) {
382 mnt_dec_writers(mnt);
391 * mnt_want_write - get write access to a mount
392 * @m: the mount on which to take a write
394 * This tells the low-level filesystem that a write is about to be performed to
395 * it, and makes sure that writes are allowed (mount is read-write, filesystem
396 * is not frozen) before returning success. When the write operation is
397 * finished, mnt_drop_write() must be called. This is effectively a refcount.
399 int mnt_want_write(struct vfsmount *m)
403 sb_start_write(m->mnt_sb);
404 ret = __mnt_want_write(m);
406 sb_end_write(m->mnt_sb);
409 EXPORT_SYMBOL_GPL(mnt_want_write);
412 * __mnt_want_write_file - get write access to a file's mount
413 * @file: the file who's mount on which to take a write
415 * This is like __mnt_want_write, but if the file is already open for writing it
416 * skips incrementing mnt_writers (since the open file already has a reference)
417 * and instead only does the check for emergency r/o remounts. This must be
418 * paired with __mnt_drop_write_file.
420 int __mnt_want_write_file(struct file *file)
422 if (file->f_mode & FMODE_WRITER) {
424 * Superblock may have become readonly while there are still
425 * writable fd's, e.g. due to a fs error with errors=remount-ro
427 if (__mnt_is_readonly(file->f_path.mnt))
431 return __mnt_want_write(file->f_path.mnt);
435 * mnt_want_write_file - get write access to a file's mount
436 * @file: the file who's mount on which to take a write
438 * This is like mnt_want_write, but if the file is already open for writing it
439 * skips incrementing mnt_writers (since the open file already has a reference)
440 * and instead only does the freeze protection and the check for emergency r/o
441 * remounts. This must be paired with mnt_drop_write_file.
443 int mnt_want_write_file(struct file *file)
447 sb_start_write(file_inode(file)->i_sb);
448 ret = __mnt_want_write_file(file);
450 sb_end_write(file_inode(file)->i_sb);
453 EXPORT_SYMBOL_GPL(mnt_want_write_file);
456 * __mnt_drop_write - give up write access to a mount
457 * @mnt: the mount on which to give up write access
459 * Tells the low-level filesystem that we are done
460 * performing writes to it. Must be matched with
461 * __mnt_want_write() call above.
463 void __mnt_drop_write(struct vfsmount *mnt)
466 mnt_dec_writers(real_mount(mnt));
471 * mnt_drop_write - give up write access to a mount
472 * @mnt: the mount on which to give up write access
474 * Tells the low-level filesystem that we are done performing writes to it and
475 * also allows filesystem to be frozen again. Must be matched with
476 * mnt_want_write() call above.
478 void mnt_drop_write(struct vfsmount *mnt)
480 __mnt_drop_write(mnt);
481 sb_end_write(mnt->mnt_sb);
483 EXPORT_SYMBOL_GPL(mnt_drop_write);
485 void __mnt_drop_write_file(struct file *file)
487 if (!(file->f_mode & FMODE_WRITER))
488 __mnt_drop_write(file->f_path.mnt);
491 void mnt_drop_write_file(struct file *file)
493 __mnt_drop_write_file(file);
494 sb_end_write(file_inode(file)->i_sb);
496 EXPORT_SYMBOL(mnt_drop_write_file);
499 * mnt_hold_writers - prevent write access to the given mount
500 * @mnt: mnt to prevent write access to
502 * Prevents write access to @mnt if there are no active writers for @mnt.
503 * This function needs to be called and return successfully before changing
504 * properties of @mnt that need to remain stable for callers with write access
507 * After this functions has been called successfully callers must pair it with
508 * a call to mnt_unhold_writers() in order to stop preventing write access to
511 * Context: This function expects lock_mount_hash() to be held serializing
512 * setting MNT_WRITE_HOLD.
513 * Return: On success 0 is returned.
514 * On error, -EBUSY is returned.
516 static inline int mnt_hold_writers(struct mount *mnt)
518 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
520 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
521 * should be visible before we do.
526 * With writers on hold, if this value is zero, then there are
527 * definitely no active writers (although held writers may subsequently
528 * increment the count, they'll have to wait, and decrement it after
529 * seeing MNT_READONLY).
531 * It is OK to have counter incremented on one CPU and decremented on
532 * another: the sum will add up correctly. The danger would be when we
533 * sum up each counter, if we read a counter before it is incremented,
534 * but then read another CPU's count which it has been subsequently
535 * decremented from -- we would see more decrements than we should.
536 * MNT_WRITE_HOLD protects against this scenario, because
537 * mnt_want_write first increments count, then smp_mb, then spins on
538 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
539 * we're counting up here.
541 if (mnt_get_writers(mnt) > 0)
548 * mnt_unhold_writers - stop preventing write access to the given mount
549 * @mnt: mnt to stop preventing write access to
551 * Stop preventing write access to @mnt allowing callers to gain write access
554 * This function can only be called after a successful call to
555 * mnt_hold_writers().
557 * Context: This function expects lock_mount_hash() to be held.
559 static inline void mnt_unhold_writers(struct mount *mnt)
562 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
563 * that become unheld will see MNT_READONLY.
566 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
569 static int mnt_make_readonly(struct mount *mnt)
573 ret = mnt_hold_writers(mnt);
575 mnt->mnt.mnt_flags |= MNT_READONLY;
576 mnt_unhold_writers(mnt);
580 int sb_prepare_remount_readonly(struct super_block *sb)
585 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
586 if (atomic_long_read(&sb->s_remove_count))
590 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
591 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
592 err = mnt_hold_writers(mnt);
597 if (!err && atomic_long_read(&sb->s_remove_count))
601 sb_start_ro_state_change(sb);
602 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
603 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
604 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
611 static void free_vfsmnt(struct mount *mnt)
613 mnt_idmap_put(mnt_idmap(&mnt->mnt));
614 kfree_const(mnt->mnt_devname);
616 free_percpu(mnt->mnt_pcp);
618 kmem_cache_free(mnt_cache, mnt);
621 static void delayed_free_vfsmnt(struct rcu_head *head)
623 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
626 /* call under rcu_read_lock */
627 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
630 if (read_seqretry(&mount_lock, seq))
634 mnt = real_mount(bastard);
635 mnt_add_count(mnt, 1);
636 smp_mb(); // see mntput_no_expire()
637 if (likely(!read_seqretry(&mount_lock, seq)))
639 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
640 mnt_add_count(mnt, -1);
644 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
645 mnt_add_count(mnt, -1);
650 /* caller will mntput() */
654 /* call under rcu_read_lock */
655 static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
657 int res = __legitimize_mnt(bastard, seq);
660 if (unlikely(res < 0)) {
669 * find the first mount at @dentry on vfsmount @mnt.
670 * call under rcu_read_lock()
672 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
674 struct hlist_head *head = m_hash(mnt, dentry);
677 hlist_for_each_entry_rcu(p, head, mnt_hash)
678 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
684 * lookup_mnt - Return the first child mount mounted at path
686 * "First" means first mounted chronologically. If you create the
689 * mount /dev/sda1 /mnt
690 * mount /dev/sda2 /mnt
691 * mount /dev/sda3 /mnt
693 * Then lookup_mnt() on the base /mnt dentry in the root mount will
694 * return successively the root dentry and vfsmount of /dev/sda1, then
695 * /dev/sda2, then /dev/sda3, then NULL.
697 * lookup_mnt takes a reference to the found vfsmount.
699 struct vfsmount *lookup_mnt(const struct path *path)
701 struct mount *child_mnt;
707 seq = read_seqbegin(&mount_lock);
708 child_mnt = __lookup_mnt(path->mnt, path->dentry);
709 m = child_mnt ? &child_mnt->mnt : NULL;
710 } while (!legitimize_mnt(m, seq));
715 static inline void lock_ns_list(struct mnt_namespace *ns)
717 spin_lock(&ns->ns_lock);
720 static inline void unlock_ns_list(struct mnt_namespace *ns)
722 spin_unlock(&ns->ns_lock);
725 static inline bool mnt_is_cursor(struct mount *mnt)
727 return mnt->mnt.mnt_flags & MNT_CURSOR;
731 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
732 * current mount namespace.
734 * The common case is dentries are not mountpoints at all and that
735 * test is handled inline. For the slow case when we are actually
736 * dealing with a mountpoint of some kind, walk through all of the
737 * mounts in the current mount namespace and test to see if the dentry
740 * The mount_hashtable is not usable in the context because we
741 * need to identify all mounts that may be in the current mount
742 * namespace not just a mount that happens to have some specified
745 bool __is_local_mountpoint(struct dentry *dentry)
747 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
749 bool is_covered = false;
751 down_read(&namespace_sem);
753 list_for_each_entry(mnt, &ns->list, mnt_list) {
754 if (mnt_is_cursor(mnt))
756 is_covered = (mnt->mnt_mountpoint == dentry);
761 up_read(&namespace_sem);
766 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
768 struct hlist_head *chain = mp_hash(dentry);
769 struct mountpoint *mp;
771 hlist_for_each_entry(mp, chain, m_hash) {
772 if (mp->m_dentry == dentry) {
780 static struct mountpoint *get_mountpoint(struct dentry *dentry)
782 struct mountpoint *mp, *new = NULL;
785 if (d_mountpoint(dentry)) {
786 /* might be worth a WARN_ON() */
787 if (d_unlinked(dentry))
788 return ERR_PTR(-ENOENT);
790 read_seqlock_excl(&mount_lock);
791 mp = lookup_mountpoint(dentry);
792 read_sequnlock_excl(&mount_lock);
798 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
800 return ERR_PTR(-ENOMEM);
803 /* Exactly one processes may set d_mounted */
804 ret = d_set_mounted(dentry);
806 /* Someone else set d_mounted? */
810 /* The dentry is not available as a mountpoint? */
815 /* Add the new mountpoint to the hash table */
816 read_seqlock_excl(&mount_lock);
817 new->m_dentry = dget(dentry);
819 hlist_add_head(&new->m_hash, mp_hash(dentry));
820 INIT_HLIST_HEAD(&new->m_list);
821 read_sequnlock_excl(&mount_lock);
831 * vfsmount lock must be held. Additionally, the caller is responsible
832 * for serializing calls for given disposal list.
834 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
836 if (!--mp->m_count) {
837 struct dentry *dentry = mp->m_dentry;
838 BUG_ON(!hlist_empty(&mp->m_list));
839 spin_lock(&dentry->d_lock);
840 dentry->d_flags &= ~DCACHE_MOUNTED;
841 spin_unlock(&dentry->d_lock);
842 dput_to_list(dentry, list);
843 hlist_del(&mp->m_hash);
848 /* called with namespace_lock and vfsmount lock */
849 static void put_mountpoint(struct mountpoint *mp)
851 __put_mountpoint(mp, &ex_mountpoints);
854 static inline int check_mnt(struct mount *mnt)
856 return mnt->mnt_ns == current->nsproxy->mnt_ns;
860 * vfsmount lock must be held for write
862 static void touch_mnt_namespace(struct mnt_namespace *ns)
866 wake_up_interruptible(&ns->poll);
871 * vfsmount lock must be held for write
873 static void __touch_mnt_namespace(struct mnt_namespace *ns)
875 if (ns && ns->event != event) {
877 wake_up_interruptible(&ns->poll);
882 * vfsmount lock must be held for write
884 static struct mountpoint *unhash_mnt(struct mount *mnt)
886 struct mountpoint *mp;
887 mnt->mnt_parent = mnt;
888 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
889 list_del_init(&mnt->mnt_child);
890 hlist_del_init_rcu(&mnt->mnt_hash);
891 hlist_del_init(&mnt->mnt_mp_list);
898 * vfsmount lock must be held for write
900 static void umount_mnt(struct mount *mnt)
902 put_mountpoint(unhash_mnt(mnt));
906 * vfsmount lock must be held for write
908 void mnt_set_mountpoint(struct mount *mnt,
909 struct mountpoint *mp,
910 struct mount *child_mnt)
913 mnt_add_count(mnt, 1); /* essentially, that's mntget */
914 child_mnt->mnt_mountpoint = mp->m_dentry;
915 child_mnt->mnt_parent = mnt;
916 child_mnt->mnt_mp = mp;
917 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
920 static void __attach_mnt(struct mount *mnt, struct mount *parent)
922 hlist_add_head_rcu(&mnt->mnt_hash,
923 m_hash(&parent->mnt, mnt->mnt_mountpoint));
924 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
928 * vfsmount lock must be held for write
930 static void attach_mnt(struct mount *mnt,
931 struct mount *parent,
932 struct mountpoint *mp)
934 mnt_set_mountpoint(parent, mp, mnt);
935 __attach_mnt(mnt, parent);
938 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
940 struct mountpoint *old_mp = mnt->mnt_mp;
941 struct mount *old_parent = mnt->mnt_parent;
943 list_del_init(&mnt->mnt_child);
944 hlist_del_init(&mnt->mnt_mp_list);
945 hlist_del_init_rcu(&mnt->mnt_hash);
947 attach_mnt(mnt, parent, mp);
949 put_mountpoint(old_mp);
950 mnt_add_count(old_parent, -1);
954 * vfsmount lock must be held for write
956 static void commit_tree(struct mount *mnt)
958 struct mount *parent = mnt->mnt_parent;
961 struct mnt_namespace *n = parent->mnt_ns;
963 BUG_ON(parent == mnt);
965 list_add_tail(&head, &mnt->mnt_list);
966 list_for_each_entry(m, &head, mnt_list)
969 list_splice(&head, n->list.prev);
971 n->mounts += n->pending_mounts;
972 n->pending_mounts = 0;
974 __attach_mnt(mnt, parent);
975 touch_mnt_namespace(n);
978 static struct mount *next_mnt(struct mount *p, struct mount *root)
980 struct list_head *next = p->mnt_mounts.next;
981 if (next == &p->mnt_mounts) {
985 next = p->mnt_child.next;
986 if (next != &p->mnt_parent->mnt_mounts)
991 return list_entry(next, struct mount, mnt_child);
994 static struct mount *skip_mnt_tree(struct mount *p)
996 struct list_head *prev = p->mnt_mounts.prev;
997 while (prev != &p->mnt_mounts) {
998 p = list_entry(prev, struct mount, mnt_child);
999 prev = p->mnt_mounts.prev;
1005 * vfs_create_mount - Create a mount for a configured superblock
1006 * @fc: The configuration context with the superblock attached
1008 * Create a mount to an already configured superblock. If necessary, the
1009 * caller should invoke vfs_get_tree() before calling this.
1011 * Note that this does not attach the mount to anything.
1013 struct vfsmount *vfs_create_mount(struct fs_context *fc)
1018 return ERR_PTR(-EINVAL);
1020 mnt = alloc_vfsmnt(fc->source ?: "none");
1022 return ERR_PTR(-ENOMEM);
1024 if (fc->sb_flags & SB_KERNMOUNT)
1025 mnt->mnt.mnt_flags = MNT_INTERNAL;
1027 atomic_inc(&fc->root->d_sb->s_active);
1028 mnt->mnt.mnt_sb = fc->root->d_sb;
1029 mnt->mnt.mnt_root = dget(fc->root);
1030 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1031 mnt->mnt_parent = mnt;
1034 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1035 unlock_mount_hash();
1038 EXPORT_SYMBOL(vfs_create_mount);
1040 struct vfsmount *fc_mount(struct fs_context *fc)
1042 int err = vfs_get_tree(fc);
1044 up_write(&fc->root->d_sb->s_umount);
1045 return vfs_create_mount(fc);
1047 return ERR_PTR(err);
1049 EXPORT_SYMBOL(fc_mount);
1051 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1052 int flags, const char *name,
1055 struct fs_context *fc;
1056 struct vfsmount *mnt;
1060 return ERR_PTR(-EINVAL);
1062 fc = fs_context_for_mount(type, flags);
1064 return ERR_CAST(fc);
1067 ret = vfs_parse_fs_string(fc, "source",
1068 name, strlen(name));
1070 ret = parse_monolithic_mount_data(fc, data);
1079 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1082 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1083 const char *name, void *data)
1085 /* Until it is worked out how to pass the user namespace
1086 * through from the parent mount to the submount don't support
1087 * unprivileged mounts with submounts.
1089 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1090 return ERR_PTR(-EPERM);
1092 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1094 EXPORT_SYMBOL_GPL(vfs_submount);
1096 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1099 struct super_block *sb = old->mnt.mnt_sb;
1103 mnt = alloc_vfsmnt(old->mnt_devname);
1105 return ERR_PTR(-ENOMEM);
1107 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1108 mnt->mnt_group_id = 0; /* not a peer of original */
1110 mnt->mnt_group_id = old->mnt_group_id;
1112 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1113 err = mnt_alloc_group_id(mnt);
1118 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1119 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1121 atomic_inc(&sb->s_active);
1122 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1124 mnt->mnt.mnt_sb = sb;
1125 mnt->mnt.mnt_root = dget(root);
1126 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1127 mnt->mnt_parent = mnt;
1129 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1130 unlock_mount_hash();
1132 if ((flag & CL_SLAVE) ||
1133 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1134 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1135 mnt->mnt_master = old;
1136 CLEAR_MNT_SHARED(mnt);
1137 } else if (!(flag & CL_PRIVATE)) {
1138 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1139 list_add(&mnt->mnt_share, &old->mnt_share);
1140 if (IS_MNT_SLAVE(old))
1141 list_add(&mnt->mnt_slave, &old->mnt_slave);
1142 mnt->mnt_master = old->mnt_master;
1144 CLEAR_MNT_SHARED(mnt);
1146 if (flag & CL_MAKE_SHARED)
1147 set_mnt_shared(mnt);
1149 /* stick the duplicate mount on the same expiry list
1150 * as the original if that was on one */
1151 if (flag & CL_EXPIRE) {
1152 if (!list_empty(&old->mnt_expire))
1153 list_add(&mnt->mnt_expire, &old->mnt_expire);
1161 return ERR_PTR(err);
1164 static void cleanup_mnt(struct mount *mnt)
1166 struct hlist_node *p;
1169 * The warning here probably indicates that somebody messed
1170 * up a mnt_want/drop_write() pair. If this happens, the
1171 * filesystem was probably unable to make r/w->r/o transitions.
1172 * The locking used to deal with mnt_count decrement provides barriers,
1173 * so mnt_get_writers() below is safe.
1175 WARN_ON(mnt_get_writers(mnt));
1176 if (unlikely(mnt->mnt_pins.first))
1178 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1179 hlist_del(&m->mnt_umount);
1182 fsnotify_vfsmount_delete(&mnt->mnt);
1183 dput(mnt->mnt.mnt_root);
1184 deactivate_super(mnt->mnt.mnt_sb);
1186 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1189 static void __cleanup_mnt(struct rcu_head *head)
1191 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1194 static LLIST_HEAD(delayed_mntput_list);
1195 static void delayed_mntput(struct work_struct *unused)
1197 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1198 struct mount *m, *t;
1200 llist_for_each_entry_safe(m, t, node, mnt_llist)
1203 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1205 static void mntput_no_expire(struct mount *mnt)
1211 if (likely(READ_ONCE(mnt->mnt_ns))) {
1213 * Since we don't do lock_mount_hash() here,
1214 * ->mnt_ns can change under us. However, if it's
1215 * non-NULL, then there's a reference that won't
1216 * be dropped until after an RCU delay done after
1217 * turning ->mnt_ns NULL. So if we observe it
1218 * non-NULL under rcu_read_lock(), the reference
1219 * we are dropping is not the final one.
1221 mnt_add_count(mnt, -1);
1227 * make sure that if __legitimize_mnt() has not seen us grab
1228 * mount_lock, we'll see their refcount increment here.
1231 mnt_add_count(mnt, -1);
1232 count = mnt_get_count(mnt);
1236 unlock_mount_hash();
1239 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1241 unlock_mount_hash();
1244 mnt->mnt.mnt_flags |= MNT_DOOMED;
1247 list_del(&mnt->mnt_instance);
1249 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1250 struct mount *p, *tmp;
1251 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1252 __put_mountpoint(unhash_mnt(p), &list);
1253 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1256 unlock_mount_hash();
1257 shrink_dentry_list(&list);
1259 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1260 struct task_struct *task = current;
1261 if (likely(!(task->flags & PF_KTHREAD))) {
1262 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1263 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1266 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1267 schedule_delayed_work(&delayed_mntput_work, 1);
1273 void mntput(struct vfsmount *mnt)
1276 struct mount *m = real_mount(mnt);
1277 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1278 if (unlikely(m->mnt_expiry_mark))
1279 m->mnt_expiry_mark = 0;
1280 mntput_no_expire(m);
1283 EXPORT_SYMBOL(mntput);
1285 struct vfsmount *mntget(struct vfsmount *mnt)
1288 mnt_add_count(real_mount(mnt), 1);
1291 EXPORT_SYMBOL(mntget);
1294 * Make a mount point inaccessible to new lookups.
1295 * Because there may still be current users, the caller MUST WAIT
1296 * for an RCU grace period before destroying the mount point.
1298 void mnt_make_shortterm(struct vfsmount *mnt)
1301 real_mount(mnt)->mnt_ns = NULL;
1305 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1306 * @path: path to check
1308 * d_mountpoint() can only be used reliably to establish if a dentry is
1309 * not mounted in any namespace and that common case is handled inline.
1310 * d_mountpoint() isn't aware of the possibility there may be multiple
1311 * mounts using a given dentry in a different namespace. This function
1312 * checks if the passed in path is a mountpoint rather than the dentry
1315 bool path_is_mountpoint(const struct path *path)
1320 if (!d_mountpoint(path->dentry))
1325 seq = read_seqbegin(&mount_lock);
1326 res = __path_is_mountpoint(path);
1327 } while (read_seqretry(&mount_lock, seq));
1332 EXPORT_SYMBOL(path_is_mountpoint);
1334 struct vfsmount *mnt_clone_internal(const struct path *path)
1337 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1340 p->mnt.mnt_flags |= MNT_INTERNAL;
1344 #ifdef CONFIG_PROC_FS
1345 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1346 struct list_head *p)
1348 struct mount *mnt, *ret = NULL;
1351 list_for_each_continue(p, &ns->list) {
1352 mnt = list_entry(p, typeof(*mnt), mnt_list);
1353 if (!mnt_is_cursor(mnt)) {
1363 /* iterator; we want it to have access to namespace_sem, thus here... */
1364 static void *m_start(struct seq_file *m, loff_t *pos)
1366 struct proc_mounts *p = m->private;
1367 struct list_head *prev;
1369 down_read(&namespace_sem);
1371 prev = &p->ns->list;
1373 prev = &p->cursor.mnt_list;
1375 /* Read after we'd reached the end? */
1376 if (list_empty(prev))
1380 return mnt_list_next(p->ns, prev);
1383 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1385 struct proc_mounts *p = m->private;
1386 struct mount *mnt = v;
1389 return mnt_list_next(p->ns, &mnt->mnt_list);
1392 static void m_stop(struct seq_file *m, void *v)
1394 struct proc_mounts *p = m->private;
1395 struct mount *mnt = v;
1397 lock_ns_list(p->ns);
1399 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1401 list_del_init(&p->cursor.mnt_list);
1402 unlock_ns_list(p->ns);
1403 up_read(&namespace_sem);
1406 static int m_show(struct seq_file *m, void *v)
1408 struct proc_mounts *p = m->private;
1409 struct mount *r = v;
1410 return p->show(m, &r->mnt);
1413 const struct seq_operations mounts_op = {
1420 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1422 down_read(&namespace_sem);
1424 list_del(&cursor->mnt_list);
1426 up_read(&namespace_sem);
1428 #endif /* CONFIG_PROC_FS */
1431 * may_umount_tree - check if a mount tree is busy
1432 * @m: root of mount tree
1434 * This is called to check if a tree of mounts has any
1435 * open files, pwds, chroots or sub mounts that are
1438 int may_umount_tree(struct vfsmount *m)
1440 struct mount *mnt = real_mount(m);
1441 int actual_refs = 0;
1442 int minimum_refs = 0;
1446 /* write lock needed for mnt_get_count */
1448 for (p = mnt; p; p = next_mnt(p, mnt)) {
1449 actual_refs += mnt_get_count(p);
1452 unlock_mount_hash();
1454 if (actual_refs > minimum_refs)
1460 EXPORT_SYMBOL(may_umount_tree);
1463 * may_umount - check if a mount point is busy
1464 * @mnt: root of mount
1466 * This is called to check if a mount point has any
1467 * open files, pwds, chroots or sub mounts. If the
1468 * mount has sub mounts this will return busy
1469 * regardless of whether the sub mounts are busy.
1471 * Doesn't take quota and stuff into account. IOW, in some cases it will
1472 * give false negatives. The main reason why it's here is that we need
1473 * a non-destructive way to look for easily umountable filesystems.
1475 int may_umount(struct vfsmount *mnt)
1478 down_read(&namespace_sem);
1480 if (propagate_mount_busy(real_mount(mnt), 2))
1482 unlock_mount_hash();
1483 up_read(&namespace_sem);
1487 EXPORT_SYMBOL(may_umount);
1489 static void namespace_unlock(void)
1491 struct hlist_head head;
1492 struct hlist_node *p;
1496 hlist_move_list(&unmounted, &head);
1497 list_splice_init(&ex_mountpoints, &list);
1499 up_write(&namespace_sem);
1501 shrink_dentry_list(&list);
1503 if (likely(hlist_empty(&head)))
1506 synchronize_rcu_expedited();
1508 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1509 hlist_del(&m->mnt_umount);
1514 static inline void namespace_lock(void)
1516 down_write(&namespace_sem);
1519 enum umount_tree_flags {
1521 UMOUNT_PROPAGATE = 2,
1522 UMOUNT_CONNECTED = 4,
1525 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1527 /* Leaving mounts connected is only valid for lazy umounts */
1528 if (how & UMOUNT_SYNC)
1531 /* A mount without a parent has nothing to be connected to */
1532 if (!mnt_has_parent(mnt))
1535 /* Because the reference counting rules change when mounts are
1536 * unmounted and connected, umounted mounts may not be
1537 * connected to mounted mounts.
1539 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1542 /* Has it been requested that the mount remain connected? */
1543 if (how & UMOUNT_CONNECTED)
1546 /* Is the mount locked such that it needs to remain connected? */
1547 if (IS_MNT_LOCKED(mnt))
1550 /* By default disconnect the mount */
1555 * mount_lock must be held
1556 * namespace_sem must be held for write
1558 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1560 LIST_HEAD(tmp_list);
1563 if (how & UMOUNT_PROPAGATE)
1564 propagate_mount_unlock(mnt);
1566 /* Gather the mounts to umount */
1567 for (p = mnt; p; p = next_mnt(p, mnt)) {
1568 p->mnt.mnt_flags |= MNT_UMOUNT;
1569 list_move(&p->mnt_list, &tmp_list);
1572 /* Hide the mounts from mnt_mounts */
1573 list_for_each_entry(p, &tmp_list, mnt_list) {
1574 list_del_init(&p->mnt_child);
1577 /* Add propogated mounts to the tmp_list */
1578 if (how & UMOUNT_PROPAGATE)
1579 propagate_umount(&tmp_list);
1581 while (!list_empty(&tmp_list)) {
1582 struct mnt_namespace *ns;
1584 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1585 list_del_init(&p->mnt_expire);
1586 list_del_init(&p->mnt_list);
1590 __touch_mnt_namespace(ns);
1593 if (how & UMOUNT_SYNC)
1594 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1596 disconnect = disconnect_mount(p, how);
1597 if (mnt_has_parent(p)) {
1598 mnt_add_count(p->mnt_parent, -1);
1600 /* Don't forget about p */
1601 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1606 change_mnt_propagation(p, MS_PRIVATE);
1608 hlist_add_head(&p->mnt_umount, &unmounted);
1612 static void shrink_submounts(struct mount *mnt);
1614 static int do_umount_root(struct super_block *sb)
1618 down_write(&sb->s_umount);
1619 if (!sb_rdonly(sb)) {
1620 struct fs_context *fc;
1622 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1627 ret = parse_monolithic_mount_data(fc, NULL);
1629 ret = reconfigure_super(fc);
1633 up_write(&sb->s_umount);
1637 static int do_umount(struct mount *mnt, int flags)
1639 struct super_block *sb = mnt->mnt.mnt_sb;
1642 retval = security_sb_umount(&mnt->mnt, flags);
1647 * Allow userspace to request a mountpoint be expired rather than
1648 * unmounting unconditionally. Unmount only happens if:
1649 * (1) the mark is already set (the mark is cleared by mntput())
1650 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1652 if (flags & MNT_EXPIRE) {
1653 if (&mnt->mnt == current->fs->root.mnt ||
1654 flags & (MNT_FORCE | MNT_DETACH))
1658 * probably don't strictly need the lock here if we examined
1659 * all race cases, but it's a slowpath.
1662 if (mnt_get_count(mnt) != 2) {
1663 unlock_mount_hash();
1666 unlock_mount_hash();
1668 if (!xchg(&mnt->mnt_expiry_mark, 1))
1673 * If we may have to abort operations to get out of this
1674 * mount, and they will themselves hold resources we must
1675 * allow the fs to do things. In the Unix tradition of
1676 * 'Gee thats tricky lets do it in userspace' the umount_begin
1677 * might fail to complete on the first run through as other tasks
1678 * must return, and the like. Thats for the mount program to worry
1679 * about for the moment.
1682 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1683 sb->s_op->umount_begin(sb);
1687 * No sense to grab the lock for this test, but test itself looks
1688 * somewhat bogus. Suggestions for better replacement?
1689 * Ho-hum... In principle, we might treat that as umount + switch
1690 * to rootfs. GC would eventually take care of the old vfsmount.
1691 * Actually it makes sense, especially if rootfs would contain a
1692 * /reboot - static binary that would close all descriptors and
1693 * call reboot(9). Then init(8) could umount root and exec /reboot.
1695 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1697 * Special case for "unmounting" root ...
1698 * we just try to remount it readonly.
1700 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1702 return do_umount_root(sb);
1708 /* Recheck MNT_LOCKED with the locks held */
1710 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1714 if (flags & MNT_DETACH) {
1715 if (!list_empty(&mnt->mnt_list))
1716 umount_tree(mnt, UMOUNT_PROPAGATE);
1719 shrink_submounts(mnt);
1721 if (!propagate_mount_busy(mnt, 2)) {
1722 if (!list_empty(&mnt->mnt_list))
1723 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1728 unlock_mount_hash();
1734 * __detach_mounts - lazily unmount all mounts on the specified dentry
1736 * During unlink, rmdir, and d_drop it is possible to loose the path
1737 * to an existing mountpoint, and wind up leaking the mount.
1738 * detach_mounts allows lazily unmounting those mounts instead of
1741 * The caller may hold dentry->d_inode->i_mutex.
1743 void __detach_mounts(struct dentry *dentry)
1745 struct mountpoint *mp;
1750 mp = lookup_mountpoint(dentry);
1755 while (!hlist_empty(&mp->m_list)) {
1756 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1757 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1759 hlist_add_head(&mnt->mnt_umount, &unmounted);
1761 else umount_tree(mnt, UMOUNT_CONNECTED);
1765 unlock_mount_hash();
1770 * Is the caller allowed to modify his namespace?
1772 bool may_mount(void)
1774 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1777 static void warn_mandlock(void)
1779 pr_warn_once("=======================================================\n"
1780 "WARNING: The mand mount option has been deprecated and\n"
1781 " and is ignored by this kernel. Remove the mand\n"
1782 " option from the mount to silence this warning.\n"
1783 "=======================================================\n");
1786 static int can_umount(const struct path *path, int flags)
1788 struct mount *mnt = real_mount(path->mnt);
1792 if (path->dentry != path->mnt->mnt_root)
1794 if (!check_mnt(mnt))
1796 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1798 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1803 // caller is responsible for flags being sane
1804 int path_umount(struct path *path, int flags)
1806 struct mount *mnt = real_mount(path->mnt);
1809 ret = can_umount(path, flags);
1811 ret = do_umount(mnt, flags);
1813 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1815 mntput_no_expire(mnt);
1819 static int ksys_umount(char __user *name, int flags)
1821 int lookup_flags = LOOKUP_MOUNTPOINT;
1825 // basic validity checks done first
1826 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1829 if (!(flags & UMOUNT_NOFOLLOW))
1830 lookup_flags |= LOOKUP_FOLLOW;
1831 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1834 return path_umount(&path, flags);
1837 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1839 return ksys_umount(name, flags);
1842 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1845 * The 2.0 compatible umount. No flags.
1847 SYSCALL_DEFINE1(oldumount, char __user *, name)
1849 return ksys_umount(name, 0);
1854 static bool is_mnt_ns_file(struct dentry *dentry)
1856 /* Is this a proxy for a mount namespace? */
1857 return dentry->d_op == &ns_dentry_operations &&
1858 dentry->d_fsdata == &mntns_operations;
1861 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1863 return container_of(ns, struct mnt_namespace, ns);
1866 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1871 static bool mnt_ns_loop(struct dentry *dentry)
1873 /* Could bind mounting the mount namespace inode cause a
1874 * mount namespace loop?
1876 struct mnt_namespace *mnt_ns;
1877 if (!is_mnt_ns_file(dentry))
1880 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1881 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1884 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1887 struct mount *res, *p, *q, *r, *parent;
1889 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1890 return ERR_PTR(-EINVAL);
1892 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1893 return ERR_PTR(-EINVAL);
1895 res = q = clone_mnt(mnt, dentry, flag);
1899 q->mnt_mountpoint = mnt->mnt_mountpoint;
1902 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1904 if (!is_subdir(r->mnt_mountpoint, dentry))
1907 for (s = r; s; s = next_mnt(s, r)) {
1908 if (!(flag & CL_COPY_UNBINDABLE) &&
1909 IS_MNT_UNBINDABLE(s)) {
1910 if (s->mnt.mnt_flags & MNT_LOCKED) {
1911 /* Both unbindable and locked. */
1912 q = ERR_PTR(-EPERM);
1915 s = skip_mnt_tree(s);
1919 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1920 is_mnt_ns_file(s->mnt.mnt_root)) {
1921 s = skip_mnt_tree(s);
1924 while (p != s->mnt_parent) {
1930 q = clone_mnt(p, p->mnt.mnt_root, flag);
1934 list_add_tail(&q->mnt_list, &res->mnt_list);
1935 attach_mnt(q, parent, p->mnt_mp);
1936 unlock_mount_hash();
1943 umount_tree(res, UMOUNT_SYNC);
1944 unlock_mount_hash();
1949 /* Caller should check returned pointer for errors */
1951 struct vfsmount *collect_mounts(const struct path *path)
1955 if (!check_mnt(real_mount(path->mnt)))
1956 tree = ERR_PTR(-EINVAL);
1958 tree = copy_tree(real_mount(path->mnt), path->dentry,
1959 CL_COPY_ALL | CL_PRIVATE);
1962 return ERR_CAST(tree);
1966 static void free_mnt_ns(struct mnt_namespace *);
1967 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1969 void dissolve_on_fput(struct vfsmount *mnt)
1971 struct mnt_namespace *ns;
1974 ns = real_mount(mnt)->mnt_ns;
1977 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1981 unlock_mount_hash();
1987 void drop_collected_mounts(struct vfsmount *mnt)
1991 umount_tree(real_mount(mnt), 0);
1992 unlock_mount_hash();
1996 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1998 struct mount *child;
2000 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2001 if (!is_subdir(child->mnt_mountpoint, dentry))
2004 if (child->mnt.mnt_flags & MNT_LOCKED)
2011 * clone_private_mount - create a private clone of a path
2012 * @path: path to clone
2014 * This creates a new vfsmount, which will be the clone of @path. The new mount
2015 * will not be attached anywhere in the namespace and will be private (i.e.
2016 * changes to the originating mount won't be propagated into this).
2018 * Release with mntput().
2020 struct vfsmount *clone_private_mount(const struct path *path)
2022 struct mount *old_mnt = real_mount(path->mnt);
2023 struct mount *new_mnt;
2025 down_read(&namespace_sem);
2026 if (IS_MNT_UNBINDABLE(old_mnt))
2029 if (!check_mnt(old_mnt))
2032 if (has_locked_children(old_mnt, path->dentry))
2035 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2036 up_read(&namespace_sem);
2038 if (IS_ERR(new_mnt))
2039 return ERR_CAST(new_mnt);
2041 /* Longterm mount to be removed by kern_unmount*() */
2042 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2044 return &new_mnt->mnt;
2047 up_read(&namespace_sem);
2048 return ERR_PTR(-EINVAL);
2050 EXPORT_SYMBOL_GPL(clone_private_mount);
2052 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2053 struct vfsmount *root)
2056 int res = f(root, arg);
2059 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2060 res = f(&mnt->mnt, arg);
2067 static void lock_mnt_tree(struct mount *mnt)
2071 for (p = mnt; p; p = next_mnt(p, mnt)) {
2072 int flags = p->mnt.mnt_flags;
2073 /* Don't allow unprivileged users to change mount flags */
2074 flags |= MNT_LOCK_ATIME;
2076 if (flags & MNT_READONLY)
2077 flags |= MNT_LOCK_READONLY;
2079 if (flags & MNT_NODEV)
2080 flags |= MNT_LOCK_NODEV;
2082 if (flags & MNT_NOSUID)
2083 flags |= MNT_LOCK_NOSUID;
2085 if (flags & MNT_NOEXEC)
2086 flags |= MNT_LOCK_NOEXEC;
2087 /* Don't allow unprivileged users to reveal what is under a mount */
2088 if (list_empty(&p->mnt_expire))
2089 flags |= MNT_LOCKED;
2090 p->mnt.mnt_flags = flags;
2094 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2098 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2099 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2100 mnt_release_group_id(p);
2104 static int invent_group_ids(struct mount *mnt, bool recurse)
2108 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2109 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2110 int err = mnt_alloc_group_id(p);
2112 cleanup_group_ids(mnt, p);
2121 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2123 unsigned int max = READ_ONCE(sysctl_mount_max);
2124 unsigned int mounts = 0;
2127 if (ns->mounts >= max)
2130 if (ns->pending_mounts >= max)
2132 max -= ns->pending_mounts;
2134 for (p = mnt; p; p = next_mnt(p, mnt))
2140 ns->pending_mounts += mounts;
2145 * @source_mnt : mount tree to be attached
2146 * @nd : place the mount tree @source_mnt is attached
2147 * @parent_nd : if non-null, detach the source_mnt from its parent and
2148 * store the parent mount and mountpoint dentry.
2149 * (done when source_mnt is moved)
2151 * NOTE: in the table below explains the semantics when a source mount
2152 * of a given type is attached to a destination mount of a given type.
2153 * ---------------------------------------------------------------------------
2154 * | BIND MOUNT OPERATION |
2155 * |**************************************************************************
2156 * | source-->| shared | private | slave | unbindable |
2160 * |**************************************************************************
2161 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2163 * |non-shared| shared (+) | private | slave (*) | invalid |
2164 * ***************************************************************************
2165 * A bind operation clones the source mount and mounts the clone on the
2166 * destination mount.
2168 * (++) the cloned mount is propagated to all the mounts in the propagation
2169 * tree of the destination mount and the cloned mount is added to
2170 * the peer group of the source mount.
2171 * (+) the cloned mount is created under the destination mount and is marked
2172 * as shared. The cloned mount is added to the peer group of the source
2174 * (+++) the mount is propagated to all the mounts in the propagation tree
2175 * of the destination mount and the cloned mount is made slave
2176 * of the same master as that of the source mount. The cloned mount
2177 * is marked as 'shared and slave'.
2178 * (*) the cloned mount is made a slave of the same master as that of the
2181 * ---------------------------------------------------------------------------
2182 * | MOVE MOUNT OPERATION |
2183 * |**************************************************************************
2184 * | source-->| shared | private | slave | unbindable |
2188 * |**************************************************************************
2189 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2191 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2192 * ***************************************************************************
2194 * (+) the mount is moved to the destination. And is then propagated to
2195 * all the mounts in the propagation tree of the destination mount.
2196 * (+*) the mount is moved to the destination.
2197 * (+++) the mount is moved to the destination and is then propagated to
2198 * all the mounts belonging to the destination mount's propagation tree.
2199 * the mount is marked as 'shared and slave'.
2200 * (*) the mount continues to be a slave at the new location.
2202 * if the source mount is a tree, the operations explained above is
2203 * applied to each mount in the tree.
2204 * Must be called without spinlocks held, since this function can sleep
2207 static int attach_recursive_mnt(struct mount *source_mnt,
2208 struct mount *dest_mnt,
2209 struct mountpoint *dest_mp,
2212 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2213 HLIST_HEAD(tree_list);
2214 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2215 struct mountpoint *smp;
2216 struct mount *child, *p;
2217 struct hlist_node *n;
2220 /* Preallocate a mountpoint in case the new mounts need
2221 * to be tucked under other mounts.
2223 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2225 return PTR_ERR(smp);
2227 /* Is there space to add these mounts to the mount namespace? */
2229 err = count_mounts(ns, source_mnt);
2234 if (IS_MNT_SHARED(dest_mnt)) {
2235 err = invent_group_ids(source_mnt, true);
2238 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2241 goto out_cleanup_ids;
2242 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2248 unhash_mnt(source_mnt);
2249 attach_mnt(source_mnt, dest_mnt, dest_mp);
2250 touch_mnt_namespace(source_mnt->mnt_ns);
2252 if (source_mnt->mnt_ns) {
2253 /* move from anon - the caller will destroy */
2254 list_del_init(&source_mnt->mnt_ns->list);
2256 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2257 commit_tree(source_mnt);
2260 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2262 hlist_del_init(&child->mnt_hash);
2263 q = __lookup_mnt(&child->mnt_parent->mnt,
2264 child->mnt_mountpoint);
2266 mnt_change_mountpoint(child, smp, q);
2267 /* Notice when we are propagating across user namespaces */
2268 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2269 lock_mnt_tree(child);
2270 child->mnt.mnt_flags &= ~MNT_LOCKED;
2273 put_mountpoint(smp);
2274 unlock_mount_hash();
2279 while (!hlist_empty(&tree_list)) {
2280 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2281 child->mnt_parent->mnt_ns->pending_mounts = 0;
2282 umount_tree(child, UMOUNT_SYNC);
2284 unlock_mount_hash();
2285 cleanup_group_ids(source_mnt, NULL);
2287 ns->pending_mounts = 0;
2289 read_seqlock_excl(&mount_lock);
2290 put_mountpoint(smp);
2291 read_sequnlock_excl(&mount_lock);
2296 static struct mountpoint *lock_mount(struct path *path)
2298 struct vfsmount *mnt;
2299 struct dentry *dentry = path->dentry;
2301 inode_lock(dentry->d_inode);
2302 if (unlikely(cant_mount(dentry))) {
2303 inode_unlock(dentry->d_inode);
2304 return ERR_PTR(-ENOENT);
2307 mnt = lookup_mnt(path);
2309 struct mountpoint *mp = get_mountpoint(dentry);
2312 inode_unlock(dentry->d_inode);
2318 inode_unlock(path->dentry->d_inode);
2321 dentry = path->dentry = dget(mnt->mnt_root);
2325 static void unlock_mount(struct mountpoint *where)
2327 struct dentry *dentry = where->m_dentry;
2329 read_seqlock_excl(&mount_lock);
2330 put_mountpoint(where);
2331 read_sequnlock_excl(&mount_lock);
2334 inode_unlock(dentry->d_inode);
2337 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2339 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2342 if (d_is_dir(mp->m_dentry) !=
2343 d_is_dir(mnt->mnt.mnt_root))
2346 return attach_recursive_mnt(mnt, p, mp, false);
2350 * Sanity check the flags to change_mnt_propagation.
2353 static int flags_to_propagation_type(int ms_flags)
2355 int type = ms_flags & ~(MS_REC | MS_SILENT);
2357 /* Fail if any non-propagation flags are set */
2358 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2360 /* Only one propagation flag should be set */
2361 if (!is_power_of_2(type))
2367 * recursively change the type of the mountpoint.
2369 static int do_change_type(struct path *path, int ms_flags)
2372 struct mount *mnt = real_mount(path->mnt);
2373 int recurse = ms_flags & MS_REC;
2377 if (path->dentry != path->mnt->mnt_root)
2380 type = flags_to_propagation_type(ms_flags);
2385 if (type == MS_SHARED) {
2386 err = invent_group_ids(mnt, recurse);
2392 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2393 change_mnt_propagation(m, type);
2394 unlock_mount_hash();
2401 static struct mount *__do_loopback(struct path *old_path, int recurse)
2403 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2405 if (IS_MNT_UNBINDABLE(old))
2408 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2411 if (!recurse && has_locked_children(old, old_path->dentry))
2415 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2417 mnt = clone_mnt(old, old_path->dentry, 0);
2420 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2426 * do loopback mount.
2428 static int do_loopback(struct path *path, const char *old_name,
2431 struct path old_path;
2432 struct mount *mnt = NULL, *parent;
2433 struct mountpoint *mp;
2435 if (!old_name || !*old_name)
2437 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2442 if (mnt_ns_loop(old_path.dentry))
2445 mp = lock_mount(path);
2451 parent = real_mount(path->mnt);
2452 if (!check_mnt(parent))
2455 mnt = __do_loopback(&old_path, recurse);
2461 err = graft_tree(mnt, parent, mp);
2464 umount_tree(mnt, UMOUNT_SYNC);
2465 unlock_mount_hash();
2470 path_put(&old_path);
2474 static struct file *open_detached_copy(struct path *path, bool recursive)
2476 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2477 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2478 struct mount *mnt, *p;
2482 return ERR_CAST(ns);
2485 mnt = __do_loopback(path, recursive);
2489 return ERR_CAST(mnt);
2493 for (p = mnt; p; p = next_mnt(p, mnt)) {
2498 list_add_tail(&ns->list, &mnt->mnt_list);
2500 unlock_mount_hash();
2504 path->mnt = &mnt->mnt;
2505 file = dentry_open(path, O_PATH, current_cred());
2507 dissolve_on_fput(path->mnt);
2509 file->f_mode |= FMODE_NEED_UNMOUNT;
2513 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2517 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2518 bool detached = flags & OPEN_TREE_CLONE;
2522 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2524 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2525 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2529 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2532 if (flags & AT_NO_AUTOMOUNT)
2533 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2534 if (flags & AT_SYMLINK_NOFOLLOW)
2535 lookup_flags &= ~LOOKUP_FOLLOW;
2536 if (flags & AT_EMPTY_PATH)
2537 lookup_flags |= LOOKUP_EMPTY;
2539 if (detached && !may_mount())
2542 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2546 error = user_path_at(dfd, filename, lookup_flags, &path);
2547 if (unlikely(error)) {
2548 file = ERR_PTR(error);
2551 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2553 file = dentry_open(&path, O_PATH, current_cred());
2558 return PTR_ERR(file);
2560 fd_install(fd, file);
2565 * Don't allow locked mount flags to be cleared.
2567 * No locks need to be held here while testing the various MNT_LOCK
2568 * flags because those flags can never be cleared once they are set.
2570 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2572 unsigned int fl = mnt->mnt.mnt_flags;
2574 if ((fl & MNT_LOCK_READONLY) &&
2575 !(mnt_flags & MNT_READONLY))
2578 if ((fl & MNT_LOCK_NODEV) &&
2579 !(mnt_flags & MNT_NODEV))
2582 if ((fl & MNT_LOCK_NOSUID) &&
2583 !(mnt_flags & MNT_NOSUID))
2586 if ((fl & MNT_LOCK_NOEXEC) &&
2587 !(mnt_flags & MNT_NOEXEC))
2590 if ((fl & MNT_LOCK_ATIME) &&
2591 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2597 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2599 bool readonly_request = (mnt_flags & MNT_READONLY);
2601 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2604 if (readonly_request)
2605 return mnt_make_readonly(mnt);
2607 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2611 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2613 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2614 mnt->mnt.mnt_flags = mnt_flags;
2615 touch_mnt_namespace(mnt->mnt_ns);
2618 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2620 struct super_block *sb = mnt->mnt_sb;
2622 if (!__mnt_is_readonly(mnt) &&
2623 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2624 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2625 char *buf = (char *)__get_free_page(GFP_KERNEL);
2626 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2628 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2630 is_mounted(mnt) ? "remounted" : "mounted",
2631 mntpath, &sb->s_time_max,
2632 (unsigned long long)sb->s_time_max);
2634 free_page((unsigned long)buf);
2635 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2640 * Handle reconfiguration of the mountpoint only without alteration of the
2641 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2644 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2646 struct super_block *sb = path->mnt->mnt_sb;
2647 struct mount *mnt = real_mount(path->mnt);
2650 if (!check_mnt(mnt))
2653 if (path->dentry != mnt->mnt.mnt_root)
2656 if (!can_change_locked_flags(mnt, mnt_flags))
2660 * We're only checking whether the superblock is read-only not
2661 * changing it, so only take down_read(&sb->s_umount).
2663 down_read(&sb->s_umount);
2665 ret = change_mount_ro_state(mnt, mnt_flags);
2667 set_mount_attributes(mnt, mnt_flags);
2668 unlock_mount_hash();
2669 up_read(&sb->s_umount);
2671 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2677 * change filesystem flags. dir should be a physical root of filesystem.
2678 * If you've mounted a non-root directory somewhere and want to do remount
2679 * on it - tough luck.
2681 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2682 int mnt_flags, void *data)
2685 struct super_block *sb = path->mnt->mnt_sb;
2686 struct mount *mnt = real_mount(path->mnt);
2687 struct fs_context *fc;
2689 if (!check_mnt(mnt))
2692 if (path->dentry != path->mnt->mnt_root)
2695 if (!can_change_locked_flags(mnt, mnt_flags))
2698 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2703 err = parse_monolithic_mount_data(fc, data);
2705 down_write(&sb->s_umount);
2707 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2708 err = reconfigure_super(fc);
2711 set_mount_attributes(mnt, mnt_flags);
2712 unlock_mount_hash();
2715 up_write(&sb->s_umount);
2718 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2724 static inline int tree_contains_unbindable(struct mount *mnt)
2727 for (p = mnt; p; p = next_mnt(p, mnt)) {
2728 if (IS_MNT_UNBINDABLE(p))
2735 * Check that there aren't references to earlier/same mount namespaces in the
2736 * specified subtree. Such references can act as pins for mount namespaces
2737 * that aren't checked by the mount-cycle checking code, thereby allowing
2738 * cycles to be made.
2740 static bool check_for_nsfs_mounts(struct mount *subtree)
2746 for (p = subtree; p; p = next_mnt(p, subtree))
2747 if (mnt_ns_loop(p->mnt.mnt_root))
2752 unlock_mount_hash();
2756 static int do_set_group(struct path *from_path, struct path *to_path)
2758 struct mount *from, *to;
2761 from = real_mount(from_path->mnt);
2762 to = real_mount(to_path->mnt);
2767 /* To and From must be mounted */
2768 if (!is_mounted(&from->mnt))
2770 if (!is_mounted(&to->mnt))
2774 /* We should be allowed to modify mount namespaces of both mounts */
2775 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2777 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2781 /* To and From paths should be mount roots */
2782 if (from_path->dentry != from_path->mnt->mnt_root)
2784 if (to_path->dentry != to_path->mnt->mnt_root)
2787 /* Setting sharing groups is only allowed across same superblock */
2788 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2791 /* From mount root should be wider than To mount root */
2792 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2795 /* From mount should not have locked children in place of To's root */
2796 if (has_locked_children(from, to->mnt.mnt_root))
2799 /* Setting sharing groups is only allowed on private mounts */
2800 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2803 /* From should not be private */
2804 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2807 if (IS_MNT_SLAVE(from)) {
2808 struct mount *m = from->mnt_master;
2810 list_add(&to->mnt_slave, &m->mnt_slave_list);
2814 if (IS_MNT_SHARED(from)) {
2815 to->mnt_group_id = from->mnt_group_id;
2816 list_add(&to->mnt_share, &from->mnt_share);
2819 unlock_mount_hash();
2828 static int do_move_mount(struct path *old_path, struct path *new_path)
2830 struct mnt_namespace *ns;
2833 struct mount *parent;
2834 struct mountpoint *mp, *old_mp;
2838 mp = lock_mount(new_path);
2842 old = real_mount(old_path->mnt);
2843 p = real_mount(new_path->mnt);
2844 parent = old->mnt_parent;
2845 attached = mnt_has_parent(old);
2846 old_mp = old->mnt_mp;
2850 /* The mountpoint must be in our namespace. */
2854 /* The thing moved must be mounted... */
2855 if (!is_mounted(&old->mnt))
2858 /* ... and either ours or the root of anon namespace */
2859 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2862 if (old->mnt.mnt_flags & MNT_LOCKED)
2865 if (old_path->dentry != old_path->mnt->mnt_root)
2868 if (d_is_dir(new_path->dentry) !=
2869 d_is_dir(old_path->dentry))
2872 * Don't move a mount residing in a shared parent.
2874 if (attached && IS_MNT_SHARED(parent))
2877 * Don't move a mount tree containing unbindable mounts to a destination
2878 * mount which is shared.
2880 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2883 if (!check_for_nsfs_mounts(old))
2885 for (; mnt_has_parent(p); p = p->mnt_parent)
2889 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2894 /* if the mount is moved, it should no longer be expire
2896 list_del_init(&old->mnt_expire);
2898 put_mountpoint(old_mp);
2903 mntput_no_expire(parent);
2910 static int do_move_mount_old(struct path *path, const char *old_name)
2912 struct path old_path;
2915 if (!old_name || !*old_name)
2918 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2922 err = do_move_mount(&old_path, path);
2923 path_put(&old_path);
2928 * add a mount into a namespace's mount tree
2930 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2931 const struct path *path, int mnt_flags)
2933 struct mount *parent = real_mount(path->mnt);
2935 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2937 if (unlikely(!check_mnt(parent))) {
2938 /* that's acceptable only for automounts done in private ns */
2939 if (!(mnt_flags & MNT_SHRINKABLE))
2941 /* ... and for those we'd better have mountpoint still alive */
2942 if (!parent->mnt_ns)
2946 /* Refuse the same filesystem on the same mount point */
2947 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2948 path->mnt->mnt_root == path->dentry)
2951 if (d_is_symlink(newmnt->mnt.mnt_root))
2954 newmnt->mnt.mnt_flags = mnt_flags;
2955 return graft_tree(newmnt, parent, mp);
2958 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2961 * Create a new mount using a superblock configuration and request it
2962 * be added to the namespace tree.
2964 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2965 unsigned int mnt_flags)
2967 struct vfsmount *mnt;
2968 struct mountpoint *mp;
2969 struct super_block *sb = fc->root->d_sb;
2972 error = security_sb_kern_mount(sb);
2973 if (!error && mount_too_revealing(sb, &mnt_flags))
2976 if (unlikely(error)) {
2981 up_write(&sb->s_umount);
2983 mnt = vfs_create_mount(fc);
2985 return PTR_ERR(mnt);
2987 mnt_warn_timestamp_expiry(mountpoint, mnt);
2989 mp = lock_mount(mountpoint);
2994 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3002 * create a new mount for userspace and request it to be added into the
3005 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3006 int mnt_flags, const char *name, void *data)
3008 struct file_system_type *type;
3009 struct fs_context *fc;
3010 const char *subtype = NULL;
3016 type = get_fs_type(fstype);
3020 if (type->fs_flags & FS_HAS_SUBTYPE) {
3021 subtype = strchr(fstype, '.');
3025 put_filesystem(type);
3031 fc = fs_context_for_mount(type, sb_flags);
3032 put_filesystem(type);
3037 err = vfs_parse_fs_string(fc, "subtype",
3038 subtype, strlen(subtype));
3040 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3042 err = parse_monolithic_mount_data(fc, data);
3043 if (!err && !mount_capable(fc))
3046 err = vfs_get_tree(fc);
3048 err = do_new_mount_fc(fc, path, mnt_flags);
3054 int finish_automount(struct vfsmount *m, const struct path *path)
3056 struct dentry *dentry = path->dentry;
3057 struct mountpoint *mp;
3066 mnt = real_mount(m);
3067 /* The new mount record should have at least 2 refs to prevent it being
3068 * expired before we get a chance to add it
3070 BUG_ON(mnt_get_count(mnt) < 2);
3072 if (m->mnt_sb == path->mnt->mnt_sb &&
3073 m->mnt_root == dentry) {
3079 * we don't want to use lock_mount() - in this case finding something
3080 * that overmounts our mountpoint to be means "quitely drop what we've
3081 * got", not "try to mount it on top".
3083 inode_lock(dentry->d_inode);
3085 if (unlikely(cant_mount(dentry))) {
3087 goto discard_locked;
3090 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3093 goto discard_locked;
3096 mp = get_mountpoint(dentry);
3099 goto discard_locked;
3102 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3111 inode_unlock(dentry->d_inode);
3113 /* remove m from any expiration list it may be on */
3114 if (!list_empty(&mnt->mnt_expire)) {
3116 list_del_init(&mnt->mnt_expire);
3125 * mnt_set_expiry - Put a mount on an expiration list
3126 * @mnt: The mount to list.
3127 * @expiry_list: The list to add the mount to.
3129 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3133 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3137 EXPORT_SYMBOL(mnt_set_expiry);
3140 * process a list of expirable mountpoints with the intent of discarding any
3141 * mountpoints that aren't in use and haven't been touched since last we came
3144 void mark_mounts_for_expiry(struct list_head *mounts)
3146 struct mount *mnt, *next;
3147 LIST_HEAD(graveyard);
3149 if (list_empty(mounts))
3155 /* extract from the expiration list every vfsmount that matches the
3156 * following criteria:
3157 * - only referenced by its parent vfsmount
3158 * - still marked for expiry (marked on the last call here; marks are
3159 * cleared by mntput())
3161 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3162 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3163 propagate_mount_busy(mnt, 1))
3165 list_move(&mnt->mnt_expire, &graveyard);
3167 while (!list_empty(&graveyard)) {
3168 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3169 touch_mnt_namespace(mnt->mnt_ns);
3170 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3172 unlock_mount_hash();
3176 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3179 * Ripoff of 'select_parent()'
3181 * search the list of submounts for a given mountpoint, and move any
3182 * shrinkable submounts to the 'graveyard' list.
3184 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3186 struct mount *this_parent = parent;
3187 struct list_head *next;
3191 next = this_parent->mnt_mounts.next;
3193 while (next != &this_parent->mnt_mounts) {
3194 struct list_head *tmp = next;
3195 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3198 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3201 * Descend a level if the d_mounts list is non-empty.
3203 if (!list_empty(&mnt->mnt_mounts)) {
3208 if (!propagate_mount_busy(mnt, 1)) {
3209 list_move_tail(&mnt->mnt_expire, graveyard);
3214 * All done at this level ... ascend and resume the search
3216 if (this_parent != parent) {
3217 next = this_parent->mnt_child.next;
3218 this_parent = this_parent->mnt_parent;
3225 * process a list of expirable mountpoints with the intent of discarding any
3226 * submounts of a specific parent mountpoint
3228 * mount_lock must be held for write
3230 static void shrink_submounts(struct mount *mnt)
3232 LIST_HEAD(graveyard);
3235 /* extract submounts of 'mountpoint' from the expiration list */
3236 while (select_submounts(mnt, &graveyard)) {
3237 while (!list_empty(&graveyard)) {
3238 m = list_first_entry(&graveyard, struct mount,
3240 touch_mnt_namespace(m->mnt_ns);
3241 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3246 static void *copy_mount_options(const void __user * data)
3249 unsigned left, offset;
3254 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3256 return ERR_PTR(-ENOMEM);
3258 left = copy_from_user(copy, data, PAGE_SIZE);
3261 * Not all architectures have an exact copy_from_user(). Resort to
3264 offset = PAGE_SIZE - left;
3267 if (get_user(c, (const char __user *)data + offset))
3274 if (left == PAGE_SIZE) {
3276 return ERR_PTR(-EFAULT);
3282 static char *copy_mount_string(const void __user *data)
3284 return data ? strndup_user(data, PATH_MAX) : NULL;
3288 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3289 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3291 * data is a (void *) that can point to any structure up to
3292 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3293 * information (or be NULL).
3295 * Pre-0.97 versions of mount() didn't have a flags word.
3296 * When the flags word was introduced its top half was required
3297 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3298 * Therefore, if this magic number is present, it carries no information
3299 * and must be discarded.
3301 int path_mount(const char *dev_name, struct path *path,
3302 const char *type_page, unsigned long flags, void *data_page)
3304 unsigned int mnt_flags = 0, sb_flags;
3308 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3309 flags &= ~MS_MGC_MSK;
3311 /* Basic sanity checks */
3313 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3315 if (flags & MS_NOUSER)
3318 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3323 if (flags & SB_MANDLOCK)
3326 /* Default to relatime unless overriden */
3327 if (!(flags & MS_NOATIME))
3328 mnt_flags |= MNT_RELATIME;
3330 /* Separate the per-mountpoint flags */
3331 if (flags & MS_NOSUID)
3332 mnt_flags |= MNT_NOSUID;
3333 if (flags & MS_NODEV)
3334 mnt_flags |= MNT_NODEV;
3335 if (flags & MS_NOEXEC)
3336 mnt_flags |= MNT_NOEXEC;
3337 if (flags & MS_NOATIME)
3338 mnt_flags |= MNT_NOATIME;
3339 if (flags & MS_NODIRATIME)
3340 mnt_flags |= MNT_NODIRATIME;
3341 if (flags & MS_STRICTATIME)
3342 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3343 if (flags & MS_RDONLY)
3344 mnt_flags |= MNT_READONLY;
3345 if (flags & MS_NOSYMFOLLOW)
3346 mnt_flags |= MNT_NOSYMFOLLOW;
3348 /* The default atime for remount is preservation */
3349 if ((flags & MS_REMOUNT) &&
3350 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3351 MS_STRICTATIME)) == 0)) {
3352 mnt_flags &= ~MNT_ATIME_MASK;
3353 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3356 sb_flags = flags & (SB_RDONLY |
3365 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3366 return do_reconfigure_mnt(path, mnt_flags);
3367 if (flags & MS_REMOUNT)
3368 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3369 if (flags & MS_BIND)
3370 return do_loopback(path, dev_name, flags & MS_REC);
3371 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3372 return do_change_type(path, flags);
3373 if (flags & MS_MOVE)
3374 return do_move_mount_old(path, dev_name);
3376 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3380 long do_mount(const char *dev_name, const char __user *dir_name,
3381 const char *type_page, unsigned long flags, void *data_page)
3386 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3389 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3394 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3396 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3399 static void dec_mnt_namespaces(struct ucounts *ucounts)
3401 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3404 static void free_mnt_ns(struct mnt_namespace *ns)
3406 if (!is_anon_ns(ns))
3407 ns_free_inum(&ns->ns);
3408 dec_mnt_namespaces(ns->ucounts);
3409 put_user_ns(ns->user_ns);
3414 * Assign a sequence number so we can detect when we attempt to bind
3415 * mount a reference to an older mount namespace into the current
3416 * mount namespace, preventing reference counting loops. A 64bit
3417 * number incrementing at 10Ghz will take 12,427 years to wrap which
3418 * is effectively never, so we can ignore the possibility.
3420 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3422 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3424 struct mnt_namespace *new_ns;
3425 struct ucounts *ucounts;
3428 ucounts = inc_mnt_namespaces(user_ns);
3430 return ERR_PTR(-ENOSPC);
3432 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3434 dec_mnt_namespaces(ucounts);
3435 return ERR_PTR(-ENOMEM);
3438 ret = ns_alloc_inum(&new_ns->ns);
3441 dec_mnt_namespaces(ucounts);
3442 return ERR_PTR(ret);
3445 new_ns->ns.ops = &mntns_operations;
3447 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3448 refcount_set(&new_ns->ns.count, 1);
3449 INIT_LIST_HEAD(&new_ns->list);
3450 init_waitqueue_head(&new_ns->poll);
3451 spin_lock_init(&new_ns->ns_lock);
3452 new_ns->user_ns = get_user_ns(user_ns);
3453 new_ns->ucounts = ucounts;
3458 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3459 struct user_namespace *user_ns, struct fs_struct *new_fs)
3461 struct mnt_namespace *new_ns;
3462 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3463 struct mount *p, *q;
3470 if (likely(!(flags & CLONE_NEWNS))) {
3477 new_ns = alloc_mnt_ns(user_ns, false);
3482 /* First pass: copy the tree topology */
3483 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3484 if (user_ns != ns->user_ns)
3485 copy_flags |= CL_SHARED_TO_SLAVE;
3486 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3489 free_mnt_ns(new_ns);
3490 return ERR_CAST(new);
3492 if (user_ns != ns->user_ns) {
3495 unlock_mount_hash();
3498 list_add_tail(&new_ns->list, &new->mnt_list);
3501 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3502 * as belonging to new namespace. We have already acquired a private
3503 * fs_struct, so tsk->fs->lock is not needed.
3511 if (&p->mnt == new_fs->root.mnt) {
3512 new_fs->root.mnt = mntget(&q->mnt);
3515 if (&p->mnt == new_fs->pwd.mnt) {
3516 new_fs->pwd.mnt = mntget(&q->mnt);
3520 p = next_mnt(p, old);
3521 q = next_mnt(q, new);
3524 // an mntns binding we'd skipped?
3525 while (p->mnt.mnt_root != q->mnt.mnt_root)
3526 p = next_mnt(skip_mnt_tree(p), old);
3538 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3540 struct mount *mnt = real_mount(m);
3541 struct mnt_namespace *ns;
3542 struct super_block *s;
3546 ns = alloc_mnt_ns(&init_user_ns, true);
3549 return ERR_CAST(ns);
3554 list_add(&mnt->mnt_list, &ns->list);
3556 err = vfs_path_lookup(m->mnt_root, m,
3557 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3562 return ERR_PTR(err);
3564 /* trade a vfsmount reference for active sb one */
3565 s = path.mnt->mnt_sb;
3566 atomic_inc(&s->s_active);
3568 /* lock the sucker */
3569 down_write(&s->s_umount);
3570 /* ... and return the root of (sub)tree on it */
3573 EXPORT_SYMBOL(mount_subtree);
3575 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3576 char __user *, type, unsigned long, flags, void __user *, data)
3583 kernel_type = copy_mount_string(type);
3584 ret = PTR_ERR(kernel_type);
3585 if (IS_ERR(kernel_type))
3588 kernel_dev = copy_mount_string(dev_name);
3589 ret = PTR_ERR(kernel_dev);
3590 if (IS_ERR(kernel_dev))
3593 options = copy_mount_options(data);
3594 ret = PTR_ERR(options);
3595 if (IS_ERR(options))
3598 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3609 #define FSMOUNT_VALID_FLAGS \
3610 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3611 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3612 MOUNT_ATTR_NOSYMFOLLOW)
3614 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3616 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3617 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3619 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3621 unsigned int mnt_flags = 0;
3623 if (attr_flags & MOUNT_ATTR_RDONLY)
3624 mnt_flags |= MNT_READONLY;
3625 if (attr_flags & MOUNT_ATTR_NOSUID)
3626 mnt_flags |= MNT_NOSUID;
3627 if (attr_flags & MOUNT_ATTR_NODEV)
3628 mnt_flags |= MNT_NODEV;
3629 if (attr_flags & MOUNT_ATTR_NOEXEC)
3630 mnt_flags |= MNT_NOEXEC;
3631 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3632 mnt_flags |= MNT_NODIRATIME;
3633 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3634 mnt_flags |= MNT_NOSYMFOLLOW;
3640 * Create a kernel mount representation for a new, prepared superblock
3641 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3643 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3644 unsigned int, attr_flags)
3646 struct mnt_namespace *ns;
3647 struct fs_context *fc;
3649 struct path newmount;
3652 unsigned int mnt_flags = 0;
3658 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3661 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3664 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3666 switch (attr_flags & MOUNT_ATTR__ATIME) {
3667 case MOUNT_ATTR_STRICTATIME:
3669 case MOUNT_ATTR_NOATIME:
3670 mnt_flags |= MNT_NOATIME;
3672 case MOUNT_ATTR_RELATIME:
3673 mnt_flags |= MNT_RELATIME;
3684 if (f.file->f_op != &fscontext_fops)
3687 fc = f.file->private_data;
3689 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3693 /* There must be a valid superblock or we can't mount it */
3699 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3700 pr_warn("VFS: Mount too revealing\n");
3705 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3708 if (fc->sb_flags & SB_MANDLOCK)
3711 newmount.mnt = vfs_create_mount(fc);
3712 if (IS_ERR(newmount.mnt)) {
3713 ret = PTR_ERR(newmount.mnt);
3716 newmount.dentry = dget(fc->root);
3717 newmount.mnt->mnt_flags = mnt_flags;
3719 /* We've done the mount bit - now move the file context into more or
3720 * less the same state as if we'd done an fspick(). We don't want to
3721 * do any memory allocation or anything like that at this point as we
3722 * don't want to have to handle any errors incurred.
3724 vfs_clean_context(fc);
3726 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3731 mnt = real_mount(newmount.mnt);
3735 list_add(&mnt->mnt_list, &ns->list);
3736 mntget(newmount.mnt);
3738 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3739 * it, not just simply put it.
3741 file = dentry_open(&newmount, O_PATH, fc->cred);
3743 dissolve_on_fput(newmount.mnt);
3744 ret = PTR_ERR(file);
3747 file->f_mode |= FMODE_NEED_UNMOUNT;
3749 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3751 fd_install(ret, file);
3756 path_put(&newmount);
3758 mutex_unlock(&fc->uapi_mutex);
3765 * Move a mount from one place to another. In combination with
3766 * fsopen()/fsmount() this is used to install a new mount and in combination
3767 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3770 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3772 SYSCALL_DEFINE5(move_mount,
3773 int, from_dfd, const char __user *, from_pathname,
3774 int, to_dfd, const char __user *, to_pathname,
3775 unsigned int, flags)
3777 struct path from_path, to_path;
3778 unsigned int lflags;
3784 if (flags & ~MOVE_MOUNT__MASK)
3787 /* If someone gives a pathname, they aren't permitted to move
3788 * from an fd that requires unmount as we can't get at the flag
3789 * to clear it afterwards.
3792 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3793 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3794 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3796 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3801 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3802 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3803 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3805 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3809 ret = security_move_mount(&from_path, &to_path);
3813 if (flags & MOVE_MOUNT_SET_GROUP)
3814 ret = do_set_group(&from_path, &to_path);
3816 ret = do_move_mount(&from_path, &to_path);
3821 path_put(&from_path);
3826 * Return true if path is reachable from root
3828 * namespace_sem or mount_lock is held
3830 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3831 const struct path *root)
3833 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3834 dentry = mnt->mnt_mountpoint;
3835 mnt = mnt->mnt_parent;
3837 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3840 bool path_is_under(const struct path *path1, const struct path *path2)
3843 read_seqlock_excl(&mount_lock);
3844 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3845 read_sequnlock_excl(&mount_lock);
3848 EXPORT_SYMBOL(path_is_under);
3851 * pivot_root Semantics:
3852 * Moves the root file system of the current process to the directory put_old,
3853 * makes new_root as the new root file system of the current process, and sets
3854 * root/cwd of all processes which had them on the current root to new_root.
3857 * The new_root and put_old must be directories, and must not be on the
3858 * same file system as the current process root. The put_old must be
3859 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3860 * pointed to by put_old must yield the same directory as new_root. No other
3861 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3863 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3864 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3865 * in this situation.
3868 * - we don't move root/cwd if they are not at the root (reason: if something
3869 * cared enough to change them, it's probably wrong to force them elsewhere)
3870 * - it's okay to pick a root that isn't the root of a file system, e.g.
3871 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3872 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3875 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3876 const char __user *, put_old)
3878 struct path new, old, root;
3879 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3880 struct mountpoint *old_mp, *root_mp;
3886 error = user_path_at(AT_FDCWD, new_root,
3887 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3891 error = user_path_at(AT_FDCWD, put_old,
3892 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3896 error = security_sb_pivotroot(&old, &new);
3900 get_fs_root(current->fs, &root);
3901 old_mp = lock_mount(&old);
3902 error = PTR_ERR(old_mp);
3907 new_mnt = real_mount(new.mnt);
3908 root_mnt = real_mount(root.mnt);
3909 old_mnt = real_mount(old.mnt);
3910 ex_parent = new_mnt->mnt_parent;
3911 root_parent = root_mnt->mnt_parent;
3912 if (IS_MNT_SHARED(old_mnt) ||
3913 IS_MNT_SHARED(ex_parent) ||
3914 IS_MNT_SHARED(root_parent))
3916 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3918 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3921 if (d_unlinked(new.dentry))
3924 if (new_mnt == root_mnt || old_mnt == root_mnt)
3925 goto out4; /* loop, on the same file system */
3927 if (root.mnt->mnt_root != root.dentry)
3928 goto out4; /* not a mountpoint */
3929 if (!mnt_has_parent(root_mnt))
3930 goto out4; /* not attached */
3931 if (new.mnt->mnt_root != new.dentry)
3932 goto out4; /* not a mountpoint */
3933 if (!mnt_has_parent(new_mnt))
3934 goto out4; /* not attached */
3935 /* make sure we can reach put_old from new_root */
3936 if (!is_path_reachable(old_mnt, old.dentry, &new))
3938 /* make certain new is below the root */
3939 if (!is_path_reachable(new_mnt, new.dentry, &root))
3942 umount_mnt(new_mnt);
3943 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3944 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3945 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3946 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3948 /* mount old root on put_old */
3949 attach_mnt(root_mnt, old_mnt, old_mp);
3950 /* mount new_root on / */
3951 attach_mnt(new_mnt, root_parent, root_mp);
3952 mnt_add_count(root_parent, -1);
3953 touch_mnt_namespace(current->nsproxy->mnt_ns);
3954 /* A moved mount should not expire automatically */
3955 list_del_init(&new_mnt->mnt_expire);
3956 put_mountpoint(root_mp);
3957 unlock_mount_hash();
3958 chroot_fs_refs(&root, &new);
3961 unlock_mount(old_mp);
3963 mntput_no_expire(ex_parent);
3974 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3976 unsigned int flags = mnt->mnt.mnt_flags;
3978 /* flags to clear */
3979 flags &= ~kattr->attr_clr;
3980 /* flags to raise */
3981 flags |= kattr->attr_set;
3986 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3988 struct vfsmount *m = &mnt->mnt;
3989 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
3991 if (!kattr->mnt_idmap)
3995 * Creating an idmapped mount with the filesystem wide idmapping
3996 * doesn't make sense so block that. We don't allow mushy semantics.
3998 if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb))
4002 * Once a mount has been idmapped we don't allow it to change its
4003 * mapping. It makes things simpler and callers can just create
4004 * another bind-mount they can idmap if they want to.
4006 if (is_idmapped_mnt(m))
4009 /* The underlying filesystem doesn't support idmapped mounts yet. */
4010 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4013 /* We're not controlling the superblock. */
4014 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4017 /* Mount has already been visible in the filesystem hierarchy. */
4018 if (!is_anon_ns(mnt->mnt_ns))
4025 * mnt_allow_writers() - check whether the attribute change allows writers
4026 * @kattr: the new mount attributes
4027 * @mnt: the mount to which @kattr will be applied
4029 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4031 * Return: true if writers need to be held, false if not
4033 static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4034 const struct mount *mnt)
4036 return (!(kattr->attr_set & MNT_READONLY) ||
4037 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4041 static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4046 for (m = mnt; m; m = next_mnt(m, mnt)) {
4047 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4052 err = can_idmap_mount(kattr, m);
4056 if (!mnt_allow_writers(kattr, m)) {
4057 err = mnt_hold_writers(m);
4062 if (!kattr->recurse)
4070 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4071 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4072 * mounts and needs to take care to include the first mount.
4074 for (p = mnt; p; p = next_mnt(p, mnt)) {
4075 /* If we had to hold writers unblock them. */
4076 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4077 mnt_unhold_writers(p);
4080 * We're done once the first mount we changed got
4081 * MNT_WRITE_HOLD unset.
4090 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4092 if (!kattr->mnt_idmap)
4096 * Pairs with smp_load_acquire() in mnt_idmap().
4098 * Since we only allow a mount to change the idmapping once and
4099 * verified this in can_idmap_mount() we know that the mount has
4100 * @nop_mnt_idmap attached to it. So there's no need to drop any
4103 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4106 static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4110 for (m = mnt; m; m = next_mnt(m, mnt)) {
4113 do_idmap_mount(kattr, m);
4114 flags = recalc_flags(kattr, m);
4115 WRITE_ONCE(m->mnt.mnt_flags, flags);
4117 /* If we had to hold writers unblock them. */
4118 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4119 mnt_unhold_writers(m);
4121 if (kattr->propagation)
4122 change_mnt_propagation(m, kattr->propagation);
4123 if (!kattr->recurse)
4126 touch_mnt_namespace(mnt->mnt_ns);
4129 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4131 struct mount *mnt = real_mount(path->mnt);
4134 if (path->dentry != mnt->mnt.mnt_root)
4137 if (kattr->mnt_userns) {
4138 struct mnt_idmap *mnt_idmap;
4140 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4141 if (IS_ERR(mnt_idmap))
4142 return PTR_ERR(mnt_idmap);
4143 kattr->mnt_idmap = mnt_idmap;
4146 if (kattr->propagation) {
4148 * Only take namespace_lock() if we're actually changing
4152 if (kattr->propagation == MS_SHARED) {
4153 err = invent_group_ids(mnt, kattr->recurse);
4164 /* Ensure that this isn't anything purely vfs internal. */
4165 if (!is_mounted(&mnt->mnt))
4169 * If this is an attached mount make sure it's located in the callers
4170 * mount namespace. If it's not don't let the caller interact with it.
4171 * If this is a detached mount make sure it has an anonymous mount
4172 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4174 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4178 * First, we get the mount tree in a shape where we can change mount
4179 * properties without failure. If we succeeded to do so we commit all
4180 * changes and if we failed we clean up.
4182 err = mount_setattr_prepare(kattr, mnt);
4184 mount_setattr_commit(kattr, mnt);
4187 unlock_mount_hash();
4189 if (kattr->propagation) {
4191 cleanup_group_ids(mnt, NULL);
4198 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4199 struct mount_kattr *kattr, unsigned int flags)
4202 struct ns_common *ns;
4203 struct user_namespace *mnt_userns;
4206 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4210 * We currently do not support clearing an idmapped mount. If this ever
4211 * is a use-case we can revisit this but for now let's keep it simple
4214 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4217 if (attr->userns_fd > INT_MAX)
4220 f = fdget(attr->userns_fd);
4224 if (!proc_ns_file(f.file)) {
4229 ns = get_proc_ns(file_inode(f.file));
4230 if (ns->ops->type != CLONE_NEWUSER) {
4236 * The initial idmapping cannot be used to create an idmapped
4237 * mount. We use the initial idmapping as an indicator of a mount
4238 * that is not idmapped. It can simply be passed into helpers that
4239 * are aware of idmapped mounts as a convenient shortcut. A user
4240 * can just create a dedicated identity mapping to achieve the same
4243 mnt_userns = container_of(ns, struct user_namespace, ns);
4244 if (mnt_userns == &init_user_ns) {
4249 /* We're not controlling the target namespace. */
4250 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4255 kattr->mnt_userns = get_user_ns(mnt_userns);
4262 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4263 struct mount_kattr *kattr, unsigned int flags)
4265 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4267 if (flags & AT_NO_AUTOMOUNT)
4268 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4269 if (flags & AT_SYMLINK_NOFOLLOW)
4270 lookup_flags &= ~LOOKUP_FOLLOW;
4271 if (flags & AT_EMPTY_PATH)
4272 lookup_flags |= LOOKUP_EMPTY;
4274 *kattr = (struct mount_kattr) {
4275 .lookup_flags = lookup_flags,
4276 .recurse = !!(flags & AT_RECURSIVE),
4279 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4281 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4283 kattr->propagation = attr->propagation;
4285 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4288 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4289 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4292 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4293 * users wanting to transition to a different atime setting cannot
4294 * simply specify the atime setting in @attr_set, but must also
4295 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4296 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4297 * @attr_clr and that @attr_set can't have any atime bits set if
4298 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4300 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4301 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4305 * Clear all previous time settings as they are mutually
4308 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4309 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4310 case MOUNT_ATTR_RELATIME:
4311 kattr->attr_set |= MNT_RELATIME;
4313 case MOUNT_ATTR_NOATIME:
4314 kattr->attr_set |= MNT_NOATIME;
4316 case MOUNT_ATTR_STRICTATIME:
4322 if (attr->attr_set & MOUNT_ATTR__ATIME)
4326 return build_mount_idmapped(attr, usize, kattr, flags);
4329 static void finish_mount_kattr(struct mount_kattr *kattr)
4331 put_user_ns(kattr->mnt_userns);
4332 kattr->mnt_userns = NULL;
4334 if (kattr->mnt_idmap)
4335 mnt_idmap_put(kattr->mnt_idmap);
4338 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4339 unsigned int, flags, struct mount_attr __user *, uattr,
4344 struct mount_attr attr;
4345 struct mount_kattr kattr;
4347 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4349 if (flags & ~(AT_EMPTY_PATH |
4351 AT_SYMLINK_NOFOLLOW |
4355 if (unlikely(usize > PAGE_SIZE))
4357 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4363 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4367 /* Don't bother walking through the mounts if this is a nop. */
4368 if (attr.attr_set == 0 &&
4369 attr.attr_clr == 0 &&
4370 attr.propagation == 0)
4373 err = build_mount_kattr(&attr, usize, &kattr, flags);
4377 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4379 err = do_mount_setattr(&target, &kattr);
4382 finish_mount_kattr(&kattr);
4386 static void __init init_mount_tree(void)
4388 struct vfsmount *mnt;
4390 struct mnt_namespace *ns;
4393 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4395 panic("Can't create rootfs");
4397 ns = alloc_mnt_ns(&init_user_ns, false);
4399 panic("Can't allocate initial namespace");
4400 m = real_mount(mnt);
4404 list_add(&m->mnt_list, &ns->list);
4405 init_task.nsproxy->mnt_ns = ns;
4409 root.dentry = mnt->mnt_root;
4410 mnt->mnt_flags |= MNT_LOCKED;
4412 set_fs_pwd(current->fs, &root);
4413 set_fs_root(current->fs, &root);
4416 void __init mnt_init(void)
4420 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4421 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4423 mount_hashtable = alloc_large_system_hash("Mount-cache",
4424 sizeof(struct hlist_head),
4427 &m_hash_shift, &m_hash_mask, 0, 0);
4428 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4429 sizeof(struct hlist_head),
4432 &mp_hash_shift, &mp_hash_mask, 0, 0);
4434 if (!mount_hashtable || !mountpoint_hashtable)
4435 panic("Failed to allocate mount hash table\n");
4441 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4443 fs_kobj = kobject_create_and_add("fs", NULL);
4445 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4451 void put_mnt_ns(struct mnt_namespace *ns)
4453 if (!refcount_dec_and_test(&ns->ns.count))
4455 drop_collected_mounts(&ns->root->mnt);
4459 struct vfsmount *kern_mount(struct file_system_type *type)
4461 struct vfsmount *mnt;
4462 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4465 * it is a longterm mount, don't release mnt until
4466 * we unmount before file sys is unregistered
4468 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4472 EXPORT_SYMBOL_GPL(kern_mount);
4474 void kern_unmount(struct vfsmount *mnt)
4476 /* release long term mount so mount point can be released */
4478 mnt_make_shortterm(mnt);
4479 synchronize_rcu(); /* yecchhh... */
4483 EXPORT_SYMBOL(kern_unmount);
4485 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4489 for (i = 0; i < num; i++)
4490 mnt_make_shortterm(mnt[i]);
4491 synchronize_rcu_expedited();
4492 for (i = 0; i < num; i++)
4495 EXPORT_SYMBOL(kern_unmount_array);
4497 bool our_mnt(struct vfsmount *mnt)
4499 return check_mnt(real_mount(mnt));
4502 bool current_chrooted(void)
4504 /* Does the current process have a non-standard root */
4505 struct path ns_root;
4506 struct path fs_root;
4509 /* Find the namespace root */
4510 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4511 ns_root.dentry = ns_root.mnt->mnt_root;
4513 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4516 get_fs_root(current->fs, &fs_root);
4518 chrooted = !path_equal(&fs_root, &ns_root);
4526 static bool mnt_already_visible(struct mnt_namespace *ns,
4527 const struct super_block *sb,
4530 int new_flags = *new_mnt_flags;
4532 bool visible = false;
4534 down_read(&namespace_sem);
4536 list_for_each_entry(mnt, &ns->list, mnt_list) {
4537 struct mount *child;
4540 if (mnt_is_cursor(mnt))
4543 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4546 /* This mount is not fully visible if it's root directory
4547 * is not the root directory of the filesystem.
4549 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4552 /* A local view of the mount flags */
4553 mnt_flags = mnt->mnt.mnt_flags;
4555 /* Don't miss readonly hidden in the superblock flags */
4556 if (sb_rdonly(mnt->mnt.mnt_sb))
4557 mnt_flags |= MNT_LOCK_READONLY;
4559 /* Verify the mount flags are equal to or more permissive
4560 * than the proposed new mount.
4562 if ((mnt_flags & MNT_LOCK_READONLY) &&
4563 !(new_flags & MNT_READONLY))
4565 if ((mnt_flags & MNT_LOCK_ATIME) &&
4566 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4569 /* This mount is not fully visible if there are any
4570 * locked child mounts that cover anything except for
4571 * empty directories.
4573 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4574 struct inode *inode = child->mnt_mountpoint->d_inode;
4575 /* Only worry about locked mounts */
4576 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4578 /* Is the directory permanetly empty? */
4579 if (!is_empty_dir_inode(inode))
4582 /* Preserve the locked attributes */
4583 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4591 up_read(&namespace_sem);
4595 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4597 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4598 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4599 unsigned long s_iflags;
4601 if (ns->user_ns == &init_user_ns)
4604 /* Can this filesystem be too revealing? */
4605 s_iflags = sb->s_iflags;
4606 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4609 if ((s_iflags & required_iflags) != required_iflags) {
4610 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4615 return !mnt_already_visible(ns, sb, new_mnt_flags);
4618 bool mnt_may_suid(struct vfsmount *mnt)
4621 * Foreign mounts (accessed via fchdir or through /proc
4622 * symlinks) are always treated as if they are nosuid. This
4623 * prevents namespaces from trusting potentially unsafe
4624 * suid/sgid bits, file caps, or security labels that originate
4625 * in other namespaces.
4627 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4628 current_in_userns(mnt->mnt_sb->s_user_ns);
4631 static struct ns_common *mntns_get(struct task_struct *task)
4633 struct ns_common *ns = NULL;
4634 struct nsproxy *nsproxy;
4637 nsproxy = task->nsproxy;
4639 ns = &nsproxy->mnt_ns->ns;
4640 get_mnt_ns(to_mnt_ns(ns));
4647 static void mntns_put(struct ns_common *ns)
4649 put_mnt_ns(to_mnt_ns(ns));
4652 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4654 struct nsproxy *nsproxy = nsset->nsproxy;
4655 struct fs_struct *fs = nsset->fs;
4656 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4657 struct user_namespace *user_ns = nsset->cred->user_ns;
4661 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4662 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4663 !ns_capable(user_ns, CAP_SYS_ADMIN))
4666 if (is_anon_ns(mnt_ns))
4673 old_mnt_ns = nsproxy->mnt_ns;
4674 nsproxy->mnt_ns = mnt_ns;
4677 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4678 "/", LOOKUP_DOWN, &root);
4680 /* revert to old namespace */
4681 nsproxy->mnt_ns = old_mnt_ns;
4686 put_mnt_ns(old_mnt_ns);
4688 /* Update the pwd and root */
4689 set_fs_pwd(fs, &root);
4690 set_fs_root(fs, &root);
4696 static struct user_namespace *mntns_owner(struct ns_common *ns)
4698 return to_mnt_ns(ns)->user_ns;
4701 const struct proc_ns_operations mntns_operations = {
4703 .type = CLONE_NEWNS,
4706 .install = mntns_install,
4707 .owner = mntns_owner,
4710 #ifdef CONFIG_SYSCTL
4711 static struct ctl_table fs_namespace_sysctls[] = {
4713 .procname = "mount-max",
4714 .data = &sysctl_mount_max,
4715 .maxlen = sizeof(unsigned int),
4717 .proc_handler = proc_dointvec_minmax,
4718 .extra1 = SYSCTL_ONE,
4723 static int __init init_fs_namespace_sysctls(void)
4725 register_sysctl_init("fs", fs_namespace_sysctls);
4728 fs_initcall(init_fs_namespace_sysctls);
4730 #endif /* CONFIG_SYSCTL */