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 (mnt->mnt_sb->s_readonly_remount)
314 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
316 return __mnt_is_readonly(mnt);
320 * Most r/o & frozen checks on a fs are for operations that take discrete
321 * amounts of time, like a write() or unlink(). We must keep track of when
322 * those operations start (for permission checks) and when they end, so that we
323 * can determine when writes are able to occur to a filesystem.
326 * __mnt_want_write - get write access to a mount without freeze protection
327 * @m: the mount on which to take a write
329 * This tells the low-level filesystem that a write is about to be performed to
330 * it, and makes sure that writes are allowed (mnt it read-write) before
331 * returning success. This operation does not protect against filesystem being
332 * frozen. When the write operation is finished, __mnt_drop_write() must be
333 * called. This is effectively a refcount.
335 int __mnt_want_write(struct vfsmount *m)
337 struct mount *mnt = real_mount(m);
341 mnt_inc_writers(mnt);
343 * The store to mnt_inc_writers must be visible before we pass
344 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
345 * incremented count after it has set MNT_WRITE_HOLD.
348 might_lock(&mount_lock.lock);
349 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
350 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
354 * This prevents priority inversion, if the task
355 * setting MNT_WRITE_HOLD got preempted on a remote
356 * CPU, and it prevents life lock if the task setting
357 * MNT_WRITE_HOLD has a lower priority and is bound to
358 * the same CPU as the task that is spinning here.
367 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
368 * be set to match its requirements. So we must not load that until
369 * MNT_WRITE_HOLD is cleared.
372 if (mnt_is_readonly(m)) {
373 mnt_dec_writers(mnt);
382 * mnt_want_write - get write access to a mount
383 * @m: the mount on which to take a write
385 * This tells the low-level filesystem that a write is about to be performed to
386 * it, and makes sure that writes are allowed (mount is read-write, filesystem
387 * is not frozen) before returning success. When the write operation is
388 * finished, mnt_drop_write() must be called. This is effectively a refcount.
390 int mnt_want_write(struct vfsmount *m)
394 sb_start_write(m->mnt_sb);
395 ret = __mnt_want_write(m);
397 sb_end_write(m->mnt_sb);
400 EXPORT_SYMBOL_GPL(mnt_want_write);
403 * __mnt_want_write_file - get write access to a file's mount
404 * @file: the file who's mount on which to take a write
406 * This is like __mnt_want_write, but if the file is already open for writing it
407 * skips incrementing mnt_writers (since the open file already has a reference)
408 * and instead only does the check for emergency r/o remounts. This must be
409 * paired with __mnt_drop_write_file.
411 int __mnt_want_write_file(struct file *file)
413 if (file->f_mode & FMODE_WRITER) {
415 * Superblock may have become readonly while there are still
416 * writable fd's, e.g. due to a fs error with errors=remount-ro
418 if (__mnt_is_readonly(file->f_path.mnt))
422 return __mnt_want_write(file->f_path.mnt);
426 * mnt_want_write_file - get write access to a file's mount
427 * @file: the file who's mount on which to take a write
429 * This is like mnt_want_write, but if the file is already open for writing it
430 * skips incrementing mnt_writers (since the open file already has a reference)
431 * and instead only does the freeze protection and the check for emergency r/o
432 * remounts. This must be paired with mnt_drop_write_file.
434 int mnt_want_write_file(struct file *file)
438 sb_start_write(file_inode(file)->i_sb);
439 ret = __mnt_want_write_file(file);
441 sb_end_write(file_inode(file)->i_sb);
444 EXPORT_SYMBOL_GPL(mnt_want_write_file);
447 * __mnt_drop_write - give up write access to a mount
448 * @mnt: the mount on which to give up write access
450 * Tells the low-level filesystem that we are done
451 * performing writes to it. Must be matched with
452 * __mnt_want_write() call above.
454 void __mnt_drop_write(struct vfsmount *mnt)
457 mnt_dec_writers(real_mount(mnt));
462 * mnt_drop_write - give up write access to a mount
463 * @mnt: the mount on which to give up write access
465 * Tells the low-level filesystem that we are done performing writes to it and
466 * also allows filesystem to be frozen again. Must be matched with
467 * mnt_want_write() call above.
469 void mnt_drop_write(struct vfsmount *mnt)
471 __mnt_drop_write(mnt);
472 sb_end_write(mnt->mnt_sb);
474 EXPORT_SYMBOL_GPL(mnt_drop_write);
476 void __mnt_drop_write_file(struct file *file)
478 if (!(file->f_mode & FMODE_WRITER))
479 __mnt_drop_write(file->f_path.mnt);
482 void mnt_drop_write_file(struct file *file)
484 __mnt_drop_write_file(file);
485 sb_end_write(file_inode(file)->i_sb);
487 EXPORT_SYMBOL(mnt_drop_write_file);
490 * mnt_hold_writers - prevent write access to the given mount
491 * @mnt: mnt to prevent write access to
493 * Prevents write access to @mnt if there are no active writers for @mnt.
494 * This function needs to be called and return successfully before changing
495 * properties of @mnt that need to remain stable for callers with write access
498 * After this functions has been called successfully callers must pair it with
499 * a call to mnt_unhold_writers() in order to stop preventing write access to
502 * Context: This function expects lock_mount_hash() to be held serializing
503 * setting MNT_WRITE_HOLD.
504 * Return: On success 0 is returned.
505 * On error, -EBUSY is returned.
507 static inline int mnt_hold_writers(struct mount *mnt)
509 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
511 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
512 * should be visible before we do.
517 * With writers on hold, if this value is zero, then there are
518 * definitely no active writers (although held writers may subsequently
519 * increment the count, they'll have to wait, and decrement it after
520 * seeing MNT_READONLY).
522 * It is OK to have counter incremented on one CPU and decremented on
523 * another: the sum will add up correctly. The danger would be when we
524 * sum up each counter, if we read a counter before it is incremented,
525 * but then read another CPU's count which it has been subsequently
526 * decremented from -- we would see more decrements than we should.
527 * MNT_WRITE_HOLD protects against this scenario, because
528 * mnt_want_write first increments count, then smp_mb, then spins on
529 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
530 * we're counting up here.
532 if (mnt_get_writers(mnt) > 0)
539 * mnt_unhold_writers - stop preventing write access to the given mount
540 * @mnt: mnt to stop preventing write access to
542 * Stop preventing write access to @mnt allowing callers to gain write access
545 * This function can only be called after a successful call to
546 * mnt_hold_writers().
548 * Context: This function expects lock_mount_hash() to be held.
550 static inline void mnt_unhold_writers(struct mount *mnt)
553 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
554 * that become unheld will see MNT_READONLY.
557 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
560 static int mnt_make_readonly(struct mount *mnt)
564 ret = mnt_hold_writers(mnt);
566 mnt->mnt.mnt_flags |= MNT_READONLY;
567 mnt_unhold_writers(mnt);
571 int sb_prepare_remount_readonly(struct super_block *sb)
576 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
577 if (atomic_long_read(&sb->s_remove_count))
581 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
582 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
583 err = mnt_hold_writers(mnt);
588 if (!err && atomic_long_read(&sb->s_remove_count))
592 sb->s_readonly_remount = 1;
595 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
596 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
597 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
604 static void free_vfsmnt(struct mount *mnt)
606 mnt_idmap_put(mnt_idmap(&mnt->mnt));
607 kfree_const(mnt->mnt_devname);
609 free_percpu(mnt->mnt_pcp);
611 kmem_cache_free(mnt_cache, mnt);
614 static void delayed_free_vfsmnt(struct rcu_head *head)
616 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
619 /* call under rcu_read_lock */
620 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
623 if (read_seqretry(&mount_lock, seq))
627 mnt = real_mount(bastard);
628 mnt_add_count(mnt, 1);
629 smp_mb(); // see mntput_no_expire()
630 if (likely(!read_seqretry(&mount_lock, seq)))
632 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
633 mnt_add_count(mnt, -1);
637 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
638 mnt_add_count(mnt, -1);
643 /* caller will mntput() */
647 /* call under rcu_read_lock */
648 static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
650 int res = __legitimize_mnt(bastard, seq);
653 if (unlikely(res < 0)) {
662 * find the first mount at @dentry on vfsmount @mnt.
663 * call under rcu_read_lock()
665 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
667 struct hlist_head *head = m_hash(mnt, dentry);
670 hlist_for_each_entry_rcu(p, head, mnt_hash)
671 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
677 * lookup_mnt - Return the first child mount mounted at path
679 * "First" means first mounted chronologically. If you create the
682 * mount /dev/sda1 /mnt
683 * mount /dev/sda2 /mnt
684 * mount /dev/sda3 /mnt
686 * Then lookup_mnt() on the base /mnt dentry in the root mount will
687 * return successively the root dentry and vfsmount of /dev/sda1, then
688 * /dev/sda2, then /dev/sda3, then NULL.
690 * lookup_mnt takes a reference to the found vfsmount.
692 struct vfsmount *lookup_mnt(const struct path *path)
694 struct mount *child_mnt;
700 seq = read_seqbegin(&mount_lock);
701 child_mnt = __lookup_mnt(path->mnt, path->dentry);
702 m = child_mnt ? &child_mnt->mnt : NULL;
703 } while (!legitimize_mnt(m, seq));
708 static inline void lock_ns_list(struct mnt_namespace *ns)
710 spin_lock(&ns->ns_lock);
713 static inline void unlock_ns_list(struct mnt_namespace *ns)
715 spin_unlock(&ns->ns_lock);
718 static inline bool mnt_is_cursor(struct mount *mnt)
720 return mnt->mnt.mnt_flags & MNT_CURSOR;
724 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
725 * current mount namespace.
727 * The common case is dentries are not mountpoints at all and that
728 * test is handled inline. For the slow case when we are actually
729 * dealing with a mountpoint of some kind, walk through all of the
730 * mounts in the current mount namespace and test to see if the dentry
733 * The mount_hashtable is not usable in the context because we
734 * need to identify all mounts that may be in the current mount
735 * namespace not just a mount that happens to have some specified
738 bool __is_local_mountpoint(struct dentry *dentry)
740 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
742 bool is_covered = false;
744 down_read(&namespace_sem);
746 list_for_each_entry(mnt, &ns->list, mnt_list) {
747 if (mnt_is_cursor(mnt))
749 is_covered = (mnt->mnt_mountpoint == dentry);
754 up_read(&namespace_sem);
759 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
761 struct hlist_head *chain = mp_hash(dentry);
762 struct mountpoint *mp;
764 hlist_for_each_entry(mp, chain, m_hash) {
765 if (mp->m_dentry == dentry) {
773 static struct mountpoint *get_mountpoint(struct dentry *dentry)
775 struct mountpoint *mp, *new = NULL;
778 if (d_mountpoint(dentry)) {
779 /* might be worth a WARN_ON() */
780 if (d_unlinked(dentry))
781 return ERR_PTR(-ENOENT);
783 read_seqlock_excl(&mount_lock);
784 mp = lookup_mountpoint(dentry);
785 read_sequnlock_excl(&mount_lock);
791 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
793 return ERR_PTR(-ENOMEM);
796 /* Exactly one processes may set d_mounted */
797 ret = d_set_mounted(dentry);
799 /* Someone else set d_mounted? */
803 /* The dentry is not available as a mountpoint? */
808 /* Add the new mountpoint to the hash table */
809 read_seqlock_excl(&mount_lock);
810 new->m_dentry = dget(dentry);
812 hlist_add_head(&new->m_hash, mp_hash(dentry));
813 INIT_HLIST_HEAD(&new->m_list);
814 read_sequnlock_excl(&mount_lock);
824 * vfsmount lock must be held. Additionally, the caller is responsible
825 * for serializing calls for given disposal list.
827 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
829 if (!--mp->m_count) {
830 struct dentry *dentry = mp->m_dentry;
831 BUG_ON(!hlist_empty(&mp->m_list));
832 spin_lock(&dentry->d_lock);
833 dentry->d_flags &= ~DCACHE_MOUNTED;
834 spin_unlock(&dentry->d_lock);
835 dput_to_list(dentry, list);
836 hlist_del(&mp->m_hash);
841 /* called with namespace_lock and vfsmount lock */
842 static void put_mountpoint(struct mountpoint *mp)
844 __put_mountpoint(mp, &ex_mountpoints);
847 static inline int check_mnt(struct mount *mnt)
849 return mnt->mnt_ns == current->nsproxy->mnt_ns;
853 * vfsmount lock must be held for write
855 static void touch_mnt_namespace(struct mnt_namespace *ns)
859 wake_up_interruptible(&ns->poll);
864 * vfsmount lock must be held for write
866 static void __touch_mnt_namespace(struct mnt_namespace *ns)
868 if (ns && ns->event != event) {
870 wake_up_interruptible(&ns->poll);
875 * vfsmount lock must be held for write
877 static struct mountpoint *unhash_mnt(struct mount *mnt)
879 struct mountpoint *mp;
880 mnt->mnt_parent = mnt;
881 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
882 list_del_init(&mnt->mnt_child);
883 hlist_del_init_rcu(&mnt->mnt_hash);
884 hlist_del_init(&mnt->mnt_mp_list);
891 * vfsmount lock must be held for write
893 static void umount_mnt(struct mount *mnt)
895 put_mountpoint(unhash_mnt(mnt));
899 * vfsmount lock must be held for write
901 void mnt_set_mountpoint(struct mount *mnt,
902 struct mountpoint *mp,
903 struct mount *child_mnt)
906 mnt_add_count(mnt, 1); /* essentially, that's mntget */
907 child_mnt->mnt_mountpoint = mp->m_dentry;
908 child_mnt->mnt_parent = mnt;
909 child_mnt->mnt_mp = mp;
910 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
913 static void __attach_mnt(struct mount *mnt, struct mount *parent)
915 hlist_add_head_rcu(&mnt->mnt_hash,
916 m_hash(&parent->mnt, mnt->mnt_mountpoint));
917 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
921 * vfsmount lock must be held for write
923 static void attach_mnt(struct mount *mnt,
924 struct mount *parent,
925 struct mountpoint *mp)
927 mnt_set_mountpoint(parent, mp, mnt);
928 __attach_mnt(mnt, parent);
931 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
933 struct mountpoint *old_mp = mnt->mnt_mp;
934 struct mount *old_parent = mnt->mnt_parent;
936 list_del_init(&mnt->mnt_child);
937 hlist_del_init(&mnt->mnt_mp_list);
938 hlist_del_init_rcu(&mnt->mnt_hash);
940 attach_mnt(mnt, parent, mp);
942 put_mountpoint(old_mp);
943 mnt_add_count(old_parent, -1);
947 * vfsmount lock must be held for write
949 static void commit_tree(struct mount *mnt)
951 struct mount *parent = mnt->mnt_parent;
954 struct mnt_namespace *n = parent->mnt_ns;
956 BUG_ON(parent == mnt);
958 list_add_tail(&head, &mnt->mnt_list);
959 list_for_each_entry(m, &head, mnt_list)
962 list_splice(&head, n->list.prev);
964 n->mounts += n->pending_mounts;
965 n->pending_mounts = 0;
967 __attach_mnt(mnt, parent);
968 touch_mnt_namespace(n);
971 static struct mount *next_mnt(struct mount *p, struct mount *root)
973 struct list_head *next = p->mnt_mounts.next;
974 if (next == &p->mnt_mounts) {
978 next = p->mnt_child.next;
979 if (next != &p->mnt_parent->mnt_mounts)
984 return list_entry(next, struct mount, mnt_child);
987 static struct mount *skip_mnt_tree(struct mount *p)
989 struct list_head *prev = p->mnt_mounts.prev;
990 while (prev != &p->mnt_mounts) {
991 p = list_entry(prev, struct mount, mnt_child);
992 prev = p->mnt_mounts.prev;
998 * vfs_create_mount - Create a mount for a configured superblock
999 * @fc: The configuration context with the superblock attached
1001 * Create a mount to an already configured superblock. If necessary, the
1002 * caller should invoke vfs_get_tree() before calling this.
1004 * Note that this does not attach the mount to anything.
1006 struct vfsmount *vfs_create_mount(struct fs_context *fc)
1011 return ERR_PTR(-EINVAL);
1013 mnt = alloc_vfsmnt(fc->source ?: "none");
1015 return ERR_PTR(-ENOMEM);
1017 if (fc->sb_flags & SB_KERNMOUNT)
1018 mnt->mnt.mnt_flags = MNT_INTERNAL;
1020 atomic_inc(&fc->root->d_sb->s_active);
1021 mnt->mnt.mnt_sb = fc->root->d_sb;
1022 mnt->mnt.mnt_root = dget(fc->root);
1023 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1024 mnt->mnt_parent = mnt;
1027 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1028 unlock_mount_hash();
1031 EXPORT_SYMBOL(vfs_create_mount);
1033 struct vfsmount *fc_mount(struct fs_context *fc)
1035 int err = vfs_get_tree(fc);
1037 up_write(&fc->root->d_sb->s_umount);
1038 return vfs_create_mount(fc);
1040 return ERR_PTR(err);
1042 EXPORT_SYMBOL(fc_mount);
1044 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1045 int flags, const char *name,
1048 struct fs_context *fc;
1049 struct vfsmount *mnt;
1053 return ERR_PTR(-EINVAL);
1055 fc = fs_context_for_mount(type, flags);
1057 return ERR_CAST(fc);
1060 ret = vfs_parse_fs_string(fc, "source",
1061 name, strlen(name));
1063 ret = parse_monolithic_mount_data(fc, data);
1072 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1075 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1076 const char *name, void *data)
1078 /* Until it is worked out how to pass the user namespace
1079 * through from the parent mount to the submount don't support
1080 * unprivileged mounts with submounts.
1082 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1083 return ERR_PTR(-EPERM);
1085 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1087 EXPORT_SYMBOL_GPL(vfs_submount);
1089 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1092 struct super_block *sb = old->mnt.mnt_sb;
1096 mnt = alloc_vfsmnt(old->mnt_devname);
1098 return ERR_PTR(-ENOMEM);
1100 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1101 mnt->mnt_group_id = 0; /* not a peer of original */
1103 mnt->mnt_group_id = old->mnt_group_id;
1105 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1106 err = mnt_alloc_group_id(mnt);
1111 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1112 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1114 atomic_inc(&sb->s_active);
1115 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1117 mnt->mnt.mnt_sb = sb;
1118 mnt->mnt.mnt_root = dget(root);
1119 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1120 mnt->mnt_parent = mnt;
1122 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1123 unlock_mount_hash();
1125 if ((flag & CL_SLAVE) ||
1126 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1127 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1128 mnt->mnt_master = old;
1129 CLEAR_MNT_SHARED(mnt);
1130 } else if (!(flag & CL_PRIVATE)) {
1131 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1132 list_add(&mnt->mnt_share, &old->mnt_share);
1133 if (IS_MNT_SLAVE(old))
1134 list_add(&mnt->mnt_slave, &old->mnt_slave);
1135 mnt->mnt_master = old->mnt_master;
1137 CLEAR_MNT_SHARED(mnt);
1139 if (flag & CL_MAKE_SHARED)
1140 set_mnt_shared(mnt);
1142 /* stick the duplicate mount on the same expiry list
1143 * as the original if that was on one */
1144 if (flag & CL_EXPIRE) {
1145 if (!list_empty(&old->mnt_expire))
1146 list_add(&mnt->mnt_expire, &old->mnt_expire);
1154 return ERR_PTR(err);
1157 static void cleanup_mnt(struct mount *mnt)
1159 struct hlist_node *p;
1162 * The warning here probably indicates that somebody messed
1163 * up a mnt_want/drop_write() pair. If this happens, the
1164 * filesystem was probably unable to make r/w->r/o transitions.
1165 * The locking used to deal with mnt_count decrement provides barriers,
1166 * so mnt_get_writers() below is safe.
1168 WARN_ON(mnt_get_writers(mnt));
1169 if (unlikely(mnt->mnt_pins.first))
1171 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1172 hlist_del(&m->mnt_umount);
1175 fsnotify_vfsmount_delete(&mnt->mnt);
1176 dput(mnt->mnt.mnt_root);
1177 deactivate_super(mnt->mnt.mnt_sb);
1179 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1182 static void __cleanup_mnt(struct rcu_head *head)
1184 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1187 static LLIST_HEAD(delayed_mntput_list);
1188 static void delayed_mntput(struct work_struct *unused)
1190 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1191 struct mount *m, *t;
1193 llist_for_each_entry_safe(m, t, node, mnt_llist)
1196 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1198 static void mntput_no_expire(struct mount *mnt)
1204 if (likely(READ_ONCE(mnt->mnt_ns))) {
1206 * Since we don't do lock_mount_hash() here,
1207 * ->mnt_ns can change under us. However, if it's
1208 * non-NULL, then there's a reference that won't
1209 * be dropped until after an RCU delay done after
1210 * turning ->mnt_ns NULL. So if we observe it
1211 * non-NULL under rcu_read_lock(), the reference
1212 * we are dropping is not the final one.
1214 mnt_add_count(mnt, -1);
1220 * make sure that if __legitimize_mnt() has not seen us grab
1221 * mount_lock, we'll see their refcount increment here.
1224 mnt_add_count(mnt, -1);
1225 count = mnt_get_count(mnt);
1229 unlock_mount_hash();
1232 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1234 unlock_mount_hash();
1237 mnt->mnt.mnt_flags |= MNT_DOOMED;
1240 list_del(&mnt->mnt_instance);
1242 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1243 struct mount *p, *tmp;
1244 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1245 __put_mountpoint(unhash_mnt(p), &list);
1246 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1249 unlock_mount_hash();
1250 shrink_dentry_list(&list);
1252 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1253 struct task_struct *task = current;
1254 if (likely(!(task->flags & PF_KTHREAD))) {
1255 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1256 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1259 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1260 schedule_delayed_work(&delayed_mntput_work, 1);
1266 void mntput(struct vfsmount *mnt)
1269 struct mount *m = real_mount(mnt);
1270 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1271 if (unlikely(m->mnt_expiry_mark))
1272 m->mnt_expiry_mark = 0;
1273 mntput_no_expire(m);
1276 EXPORT_SYMBOL(mntput);
1278 struct vfsmount *mntget(struct vfsmount *mnt)
1281 mnt_add_count(real_mount(mnt), 1);
1284 EXPORT_SYMBOL(mntget);
1287 * Make a mount point inaccessible to new lookups.
1288 * Because there may still be current users, the caller MUST WAIT
1289 * for an RCU grace period before destroying the mount point.
1291 void mnt_make_shortterm(struct vfsmount *mnt)
1294 real_mount(mnt)->mnt_ns = NULL;
1298 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1299 * @path: path to check
1301 * d_mountpoint() can only be used reliably to establish if a dentry is
1302 * not mounted in any namespace and that common case is handled inline.
1303 * d_mountpoint() isn't aware of the possibility there may be multiple
1304 * mounts using a given dentry in a different namespace. This function
1305 * checks if the passed in path is a mountpoint rather than the dentry
1308 bool path_is_mountpoint(const struct path *path)
1313 if (!d_mountpoint(path->dentry))
1318 seq = read_seqbegin(&mount_lock);
1319 res = __path_is_mountpoint(path);
1320 } while (read_seqretry(&mount_lock, seq));
1325 EXPORT_SYMBOL(path_is_mountpoint);
1327 struct vfsmount *mnt_clone_internal(const struct path *path)
1330 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1333 p->mnt.mnt_flags |= MNT_INTERNAL;
1337 #ifdef CONFIG_PROC_FS
1338 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1339 struct list_head *p)
1341 struct mount *mnt, *ret = NULL;
1344 list_for_each_continue(p, &ns->list) {
1345 mnt = list_entry(p, typeof(*mnt), mnt_list);
1346 if (!mnt_is_cursor(mnt)) {
1356 /* iterator; we want it to have access to namespace_sem, thus here... */
1357 static void *m_start(struct seq_file *m, loff_t *pos)
1359 struct proc_mounts *p = m->private;
1360 struct list_head *prev;
1362 down_read(&namespace_sem);
1364 prev = &p->ns->list;
1366 prev = &p->cursor.mnt_list;
1368 /* Read after we'd reached the end? */
1369 if (list_empty(prev))
1373 return mnt_list_next(p->ns, prev);
1376 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1378 struct proc_mounts *p = m->private;
1379 struct mount *mnt = v;
1382 return mnt_list_next(p->ns, &mnt->mnt_list);
1385 static void m_stop(struct seq_file *m, void *v)
1387 struct proc_mounts *p = m->private;
1388 struct mount *mnt = v;
1390 lock_ns_list(p->ns);
1392 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1394 list_del_init(&p->cursor.mnt_list);
1395 unlock_ns_list(p->ns);
1396 up_read(&namespace_sem);
1399 static int m_show(struct seq_file *m, void *v)
1401 struct proc_mounts *p = m->private;
1402 struct mount *r = v;
1403 return p->show(m, &r->mnt);
1406 const struct seq_operations mounts_op = {
1413 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1415 down_read(&namespace_sem);
1417 list_del(&cursor->mnt_list);
1419 up_read(&namespace_sem);
1421 #endif /* CONFIG_PROC_FS */
1424 * may_umount_tree - check if a mount tree is busy
1425 * @m: root of mount tree
1427 * This is called to check if a tree of mounts has any
1428 * open files, pwds, chroots or sub mounts that are
1431 int may_umount_tree(struct vfsmount *m)
1433 struct mount *mnt = real_mount(m);
1434 int actual_refs = 0;
1435 int minimum_refs = 0;
1439 /* write lock needed for mnt_get_count */
1441 for (p = mnt; p; p = next_mnt(p, mnt)) {
1442 actual_refs += mnt_get_count(p);
1445 unlock_mount_hash();
1447 if (actual_refs > minimum_refs)
1453 EXPORT_SYMBOL(may_umount_tree);
1456 * may_umount - check if a mount point is busy
1457 * @mnt: root of mount
1459 * This is called to check if a mount point has any
1460 * open files, pwds, chroots or sub mounts. If the
1461 * mount has sub mounts this will return busy
1462 * regardless of whether the sub mounts are busy.
1464 * Doesn't take quota and stuff into account. IOW, in some cases it will
1465 * give false negatives. The main reason why it's here is that we need
1466 * a non-destructive way to look for easily umountable filesystems.
1468 int may_umount(struct vfsmount *mnt)
1471 down_read(&namespace_sem);
1473 if (propagate_mount_busy(real_mount(mnt), 2))
1475 unlock_mount_hash();
1476 up_read(&namespace_sem);
1480 EXPORT_SYMBOL(may_umount);
1482 static void namespace_unlock(void)
1484 struct hlist_head head;
1485 struct hlist_node *p;
1489 hlist_move_list(&unmounted, &head);
1490 list_splice_init(&ex_mountpoints, &list);
1492 up_write(&namespace_sem);
1494 shrink_dentry_list(&list);
1496 if (likely(hlist_empty(&head)))
1499 synchronize_rcu_expedited();
1501 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1502 hlist_del(&m->mnt_umount);
1507 static inline void namespace_lock(void)
1509 down_write(&namespace_sem);
1512 enum umount_tree_flags {
1514 UMOUNT_PROPAGATE = 2,
1515 UMOUNT_CONNECTED = 4,
1518 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1520 /* Leaving mounts connected is only valid for lazy umounts */
1521 if (how & UMOUNT_SYNC)
1524 /* A mount without a parent has nothing to be connected to */
1525 if (!mnt_has_parent(mnt))
1528 /* Because the reference counting rules change when mounts are
1529 * unmounted and connected, umounted mounts may not be
1530 * connected to mounted mounts.
1532 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1535 /* Has it been requested that the mount remain connected? */
1536 if (how & UMOUNT_CONNECTED)
1539 /* Is the mount locked such that it needs to remain connected? */
1540 if (IS_MNT_LOCKED(mnt))
1543 /* By default disconnect the mount */
1548 * mount_lock must be held
1549 * namespace_sem must be held for write
1551 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1553 LIST_HEAD(tmp_list);
1556 if (how & UMOUNT_PROPAGATE)
1557 propagate_mount_unlock(mnt);
1559 /* Gather the mounts to umount */
1560 for (p = mnt; p; p = next_mnt(p, mnt)) {
1561 p->mnt.mnt_flags |= MNT_UMOUNT;
1562 list_move(&p->mnt_list, &tmp_list);
1565 /* Hide the mounts from mnt_mounts */
1566 list_for_each_entry(p, &tmp_list, mnt_list) {
1567 list_del_init(&p->mnt_child);
1570 /* Add propogated mounts to the tmp_list */
1571 if (how & UMOUNT_PROPAGATE)
1572 propagate_umount(&tmp_list);
1574 while (!list_empty(&tmp_list)) {
1575 struct mnt_namespace *ns;
1577 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1578 list_del_init(&p->mnt_expire);
1579 list_del_init(&p->mnt_list);
1583 __touch_mnt_namespace(ns);
1586 if (how & UMOUNT_SYNC)
1587 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1589 disconnect = disconnect_mount(p, how);
1590 if (mnt_has_parent(p)) {
1591 mnt_add_count(p->mnt_parent, -1);
1593 /* Don't forget about p */
1594 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1599 change_mnt_propagation(p, MS_PRIVATE);
1601 hlist_add_head(&p->mnt_umount, &unmounted);
1605 static void shrink_submounts(struct mount *mnt);
1607 static int do_umount_root(struct super_block *sb)
1611 down_write(&sb->s_umount);
1612 if (!sb_rdonly(sb)) {
1613 struct fs_context *fc;
1615 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1620 ret = parse_monolithic_mount_data(fc, NULL);
1622 ret = reconfigure_super(fc);
1626 up_write(&sb->s_umount);
1630 static int do_umount(struct mount *mnt, int flags)
1632 struct super_block *sb = mnt->mnt.mnt_sb;
1635 retval = security_sb_umount(&mnt->mnt, flags);
1640 * Allow userspace to request a mountpoint be expired rather than
1641 * unmounting unconditionally. Unmount only happens if:
1642 * (1) the mark is already set (the mark is cleared by mntput())
1643 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1645 if (flags & MNT_EXPIRE) {
1646 if (&mnt->mnt == current->fs->root.mnt ||
1647 flags & (MNT_FORCE | MNT_DETACH))
1651 * probably don't strictly need the lock here if we examined
1652 * all race cases, but it's a slowpath.
1655 if (mnt_get_count(mnt) != 2) {
1656 unlock_mount_hash();
1659 unlock_mount_hash();
1661 if (!xchg(&mnt->mnt_expiry_mark, 1))
1666 * If we may have to abort operations to get out of this
1667 * mount, and they will themselves hold resources we must
1668 * allow the fs to do things. In the Unix tradition of
1669 * 'Gee thats tricky lets do it in userspace' the umount_begin
1670 * might fail to complete on the first run through as other tasks
1671 * must return, and the like. Thats for the mount program to worry
1672 * about for the moment.
1675 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1676 sb->s_op->umount_begin(sb);
1680 * No sense to grab the lock for this test, but test itself looks
1681 * somewhat bogus. Suggestions for better replacement?
1682 * Ho-hum... In principle, we might treat that as umount + switch
1683 * to rootfs. GC would eventually take care of the old vfsmount.
1684 * Actually it makes sense, especially if rootfs would contain a
1685 * /reboot - static binary that would close all descriptors and
1686 * call reboot(9). Then init(8) could umount root and exec /reboot.
1688 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1690 * Special case for "unmounting" root ...
1691 * we just try to remount it readonly.
1693 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1695 return do_umount_root(sb);
1701 /* Recheck MNT_LOCKED with the locks held */
1703 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1707 if (flags & MNT_DETACH) {
1708 if (!list_empty(&mnt->mnt_list))
1709 umount_tree(mnt, UMOUNT_PROPAGATE);
1712 shrink_submounts(mnt);
1714 if (!propagate_mount_busy(mnt, 2)) {
1715 if (!list_empty(&mnt->mnt_list))
1716 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1721 unlock_mount_hash();
1727 * __detach_mounts - lazily unmount all mounts on the specified dentry
1729 * During unlink, rmdir, and d_drop it is possible to loose the path
1730 * to an existing mountpoint, and wind up leaking the mount.
1731 * detach_mounts allows lazily unmounting those mounts instead of
1734 * The caller may hold dentry->d_inode->i_mutex.
1736 void __detach_mounts(struct dentry *dentry)
1738 struct mountpoint *mp;
1743 mp = lookup_mountpoint(dentry);
1748 while (!hlist_empty(&mp->m_list)) {
1749 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1750 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1752 hlist_add_head(&mnt->mnt_umount, &unmounted);
1754 else umount_tree(mnt, UMOUNT_CONNECTED);
1758 unlock_mount_hash();
1763 * Is the caller allowed to modify his namespace?
1765 bool may_mount(void)
1767 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1770 static void warn_mandlock(void)
1772 pr_warn_once("=======================================================\n"
1773 "WARNING: The mand mount option has been deprecated and\n"
1774 " and is ignored by this kernel. Remove the mand\n"
1775 " option from the mount to silence this warning.\n"
1776 "=======================================================\n");
1779 static int can_umount(const struct path *path, int flags)
1781 struct mount *mnt = real_mount(path->mnt);
1785 if (path->dentry != path->mnt->mnt_root)
1787 if (!check_mnt(mnt))
1789 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1791 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1796 // caller is responsible for flags being sane
1797 int path_umount(struct path *path, int flags)
1799 struct mount *mnt = real_mount(path->mnt);
1802 ret = can_umount(path, flags);
1804 ret = do_umount(mnt, flags);
1806 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1808 mntput_no_expire(mnt);
1812 static int ksys_umount(char __user *name, int flags)
1814 int lookup_flags = LOOKUP_MOUNTPOINT;
1818 // basic validity checks done first
1819 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1822 if (!(flags & UMOUNT_NOFOLLOW))
1823 lookup_flags |= LOOKUP_FOLLOW;
1824 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1827 return path_umount(&path, flags);
1830 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1832 return ksys_umount(name, flags);
1835 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1838 * The 2.0 compatible umount. No flags.
1840 SYSCALL_DEFINE1(oldumount, char __user *, name)
1842 return ksys_umount(name, 0);
1847 static bool is_mnt_ns_file(struct dentry *dentry)
1849 /* Is this a proxy for a mount namespace? */
1850 return dentry->d_op == &ns_dentry_operations &&
1851 dentry->d_fsdata == &mntns_operations;
1854 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1856 return container_of(ns, struct mnt_namespace, ns);
1859 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1864 static bool mnt_ns_loop(struct dentry *dentry)
1866 /* Could bind mounting the mount namespace inode cause a
1867 * mount namespace loop?
1869 struct mnt_namespace *mnt_ns;
1870 if (!is_mnt_ns_file(dentry))
1873 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1874 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1877 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1880 struct mount *res, *p, *q, *r, *parent;
1882 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1883 return ERR_PTR(-EINVAL);
1885 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1886 return ERR_PTR(-EINVAL);
1888 res = q = clone_mnt(mnt, dentry, flag);
1892 q->mnt_mountpoint = mnt->mnt_mountpoint;
1895 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1897 if (!is_subdir(r->mnt_mountpoint, dentry))
1900 for (s = r; s; s = next_mnt(s, r)) {
1901 if (!(flag & CL_COPY_UNBINDABLE) &&
1902 IS_MNT_UNBINDABLE(s)) {
1903 if (s->mnt.mnt_flags & MNT_LOCKED) {
1904 /* Both unbindable and locked. */
1905 q = ERR_PTR(-EPERM);
1908 s = skip_mnt_tree(s);
1912 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1913 is_mnt_ns_file(s->mnt.mnt_root)) {
1914 s = skip_mnt_tree(s);
1917 while (p != s->mnt_parent) {
1923 q = clone_mnt(p, p->mnt.mnt_root, flag);
1927 list_add_tail(&q->mnt_list, &res->mnt_list);
1928 attach_mnt(q, parent, p->mnt_mp);
1929 unlock_mount_hash();
1936 umount_tree(res, UMOUNT_SYNC);
1937 unlock_mount_hash();
1942 /* Caller should check returned pointer for errors */
1944 struct vfsmount *collect_mounts(const struct path *path)
1948 if (!check_mnt(real_mount(path->mnt)))
1949 tree = ERR_PTR(-EINVAL);
1951 tree = copy_tree(real_mount(path->mnt), path->dentry,
1952 CL_COPY_ALL | CL_PRIVATE);
1955 return ERR_CAST(tree);
1959 static void free_mnt_ns(struct mnt_namespace *);
1960 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1962 void dissolve_on_fput(struct vfsmount *mnt)
1964 struct mnt_namespace *ns;
1967 ns = real_mount(mnt)->mnt_ns;
1970 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1974 unlock_mount_hash();
1980 void drop_collected_mounts(struct vfsmount *mnt)
1984 umount_tree(real_mount(mnt), 0);
1985 unlock_mount_hash();
1989 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1991 struct mount *child;
1993 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1994 if (!is_subdir(child->mnt_mountpoint, dentry))
1997 if (child->mnt.mnt_flags & MNT_LOCKED)
2004 * clone_private_mount - create a private clone of a path
2005 * @path: path to clone
2007 * This creates a new vfsmount, which will be the clone of @path. The new mount
2008 * will not be attached anywhere in the namespace and will be private (i.e.
2009 * changes to the originating mount won't be propagated into this).
2011 * Release with mntput().
2013 struct vfsmount *clone_private_mount(const struct path *path)
2015 struct mount *old_mnt = real_mount(path->mnt);
2016 struct mount *new_mnt;
2018 down_read(&namespace_sem);
2019 if (IS_MNT_UNBINDABLE(old_mnt))
2022 if (!check_mnt(old_mnt))
2025 if (has_locked_children(old_mnt, path->dentry))
2028 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2029 up_read(&namespace_sem);
2031 if (IS_ERR(new_mnt))
2032 return ERR_CAST(new_mnt);
2034 /* Longterm mount to be removed by kern_unmount*() */
2035 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2037 return &new_mnt->mnt;
2040 up_read(&namespace_sem);
2041 return ERR_PTR(-EINVAL);
2043 EXPORT_SYMBOL_GPL(clone_private_mount);
2045 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2046 struct vfsmount *root)
2049 int res = f(root, arg);
2052 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2053 res = f(&mnt->mnt, arg);
2060 static void lock_mnt_tree(struct mount *mnt)
2064 for (p = mnt; p; p = next_mnt(p, mnt)) {
2065 int flags = p->mnt.mnt_flags;
2066 /* Don't allow unprivileged users to change mount flags */
2067 flags |= MNT_LOCK_ATIME;
2069 if (flags & MNT_READONLY)
2070 flags |= MNT_LOCK_READONLY;
2072 if (flags & MNT_NODEV)
2073 flags |= MNT_LOCK_NODEV;
2075 if (flags & MNT_NOSUID)
2076 flags |= MNT_LOCK_NOSUID;
2078 if (flags & MNT_NOEXEC)
2079 flags |= MNT_LOCK_NOEXEC;
2080 /* Don't allow unprivileged users to reveal what is under a mount */
2081 if (list_empty(&p->mnt_expire))
2082 flags |= MNT_LOCKED;
2083 p->mnt.mnt_flags = flags;
2087 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2091 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2092 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2093 mnt_release_group_id(p);
2097 static int invent_group_ids(struct mount *mnt, bool recurse)
2101 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2102 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2103 int err = mnt_alloc_group_id(p);
2105 cleanup_group_ids(mnt, p);
2114 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2116 unsigned int max = READ_ONCE(sysctl_mount_max);
2117 unsigned int mounts = 0;
2120 if (ns->mounts >= max)
2123 if (ns->pending_mounts >= max)
2125 max -= ns->pending_mounts;
2127 for (p = mnt; p; p = next_mnt(p, mnt))
2133 ns->pending_mounts += mounts;
2138 * @source_mnt : mount tree to be attached
2139 * @nd : place the mount tree @source_mnt is attached
2140 * @parent_nd : if non-null, detach the source_mnt from its parent and
2141 * store the parent mount and mountpoint dentry.
2142 * (done when source_mnt is moved)
2144 * NOTE: in the table below explains the semantics when a source mount
2145 * of a given type is attached to a destination mount of a given type.
2146 * ---------------------------------------------------------------------------
2147 * | BIND MOUNT OPERATION |
2148 * |**************************************************************************
2149 * | source-->| shared | private | slave | unbindable |
2153 * |**************************************************************************
2154 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2156 * |non-shared| shared (+) | private | slave (*) | invalid |
2157 * ***************************************************************************
2158 * A bind operation clones the source mount and mounts the clone on the
2159 * destination mount.
2161 * (++) the cloned mount is propagated to all the mounts in the propagation
2162 * tree of the destination mount and the cloned mount is added to
2163 * the peer group of the source mount.
2164 * (+) the cloned mount is created under the destination mount and is marked
2165 * as shared. The cloned mount is added to the peer group of the source
2167 * (+++) the mount is propagated to all the mounts in the propagation tree
2168 * of the destination mount and the cloned mount is made slave
2169 * of the same master as that of the source mount. The cloned mount
2170 * is marked as 'shared and slave'.
2171 * (*) the cloned mount is made a slave of the same master as that of the
2174 * ---------------------------------------------------------------------------
2175 * | MOVE MOUNT OPERATION |
2176 * |**************************************************************************
2177 * | source-->| shared | private | slave | unbindable |
2181 * |**************************************************************************
2182 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2184 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2185 * ***************************************************************************
2187 * (+) the mount is moved to the destination. And is then propagated to
2188 * all the mounts in the propagation tree of the destination mount.
2189 * (+*) the mount is moved to the destination.
2190 * (+++) the mount is moved to the destination and is then propagated to
2191 * all the mounts belonging to the destination mount's propagation tree.
2192 * the mount is marked as 'shared and slave'.
2193 * (*) the mount continues to be a slave at the new location.
2195 * if the source mount is a tree, the operations explained above is
2196 * applied to each mount in the tree.
2197 * Must be called without spinlocks held, since this function can sleep
2200 static int attach_recursive_mnt(struct mount *source_mnt,
2201 struct mount *dest_mnt,
2202 struct mountpoint *dest_mp,
2205 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2206 HLIST_HEAD(tree_list);
2207 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2208 struct mountpoint *smp;
2209 struct mount *child, *p;
2210 struct hlist_node *n;
2213 /* Preallocate a mountpoint in case the new mounts need
2214 * to be tucked under other mounts.
2216 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2218 return PTR_ERR(smp);
2220 /* Is there space to add these mounts to the mount namespace? */
2222 err = count_mounts(ns, source_mnt);
2227 if (IS_MNT_SHARED(dest_mnt)) {
2228 err = invent_group_ids(source_mnt, true);
2231 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2234 goto out_cleanup_ids;
2235 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2241 unhash_mnt(source_mnt);
2242 attach_mnt(source_mnt, dest_mnt, dest_mp);
2243 touch_mnt_namespace(source_mnt->mnt_ns);
2245 if (source_mnt->mnt_ns) {
2246 /* move from anon - the caller will destroy */
2247 list_del_init(&source_mnt->mnt_ns->list);
2249 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2250 commit_tree(source_mnt);
2253 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2255 hlist_del_init(&child->mnt_hash);
2256 q = __lookup_mnt(&child->mnt_parent->mnt,
2257 child->mnt_mountpoint);
2259 mnt_change_mountpoint(child, smp, q);
2260 /* Notice when we are propagating across user namespaces */
2261 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2262 lock_mnt_tree(child);
2263 child->mnt.mnt_flags &= ~MNT_LOCKED;
2266 put_mountpoint(smp);
2267 unlock_mount_hash();
2272 while (!hlist_empty(&tree_list)) {
2273 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2274 child->mnt_parent->mnt_ns->pending_mounts = 0;
2275 umount_tree(child, UMOUNT_SYNC);
2277 unlock_mount_hash();
2278 cleanup_group_ids(source_mnt, NULL);
2280 ns->pending_mounts = 0;
2282 read_seqlock_excl(&mount_lock);
2283 put_mountpoint(smp);
2284 read_sequnlock_excl(&mount_lock);
2289 static struct mountpoint *lock_mount(struct path *path)
2291 struct vfsmount *mnt;
2292 struct dentry *dentry = path->dentry;
2294 inode_lock(dentry->d_inode);
2295 if (unlikely(cant_mount(dentry))) {
2296 inode_unlock(dentry->d_inode);
2297 return ERR_PTR(-ENOENT);
2300 mnt = lookup_mnt(path);
2302 struct mountpoint *mp = get_mountpoint(dentry);
2305 inode_unlock(dentry->d_inode);
2311 inode_unlock(path->dentry->d_inode);
2314 dentry = path->dentry = dget(mnt->mnt_root);
2318 static void unlock_mount(struct mountpoint *where)
2320 struct dentry *dentry = where->m_dentry;
2322 read_seqlock_excl(&mount_lock);
2323 put_mountpoint(where);
2324 read_sequnlock_excl(&mount_lock);
2327 inode_unlock(dentry->d_inode);
2330 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2332 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2335 if (d_is_dir(mp->m_dentry) !=
2336 d_is_dir(mnt->mnt.mnt_root))
2339 return attach_recursive_mnt(mnt, p, mp, false);
2343 * Sanity check the flags to change_mnt_propagation.
2346 static int flags_to_propagation_type(int ms_flags)
2348 int type = ms_flags & ~(MS_REC | MS_SILENT);
2350 /* Fail if any non-propagation flags are set */
2351 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2353 /* Only one propagation flag should be set */
2354 if (!is_power_of_2(type))
2360 * recursively change the type of the mountpoint.
2362 static int do_change_type(struct path *path, int ms_flags)
2365 struct mount *mnt = real_mount(path->mnt);
2366 int recurse = ms_flags & MS_REC;
2370 if (path->dentry != path->mnt->mnt_root)
2373 type = flags_to_propagation_type(ms_flags);
2378 if (type == MS_SHARED) {
2379 err = invent_group_ids(mnt, recurse);
2385 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2386 change_mnt_propagation(m, type);
2387 unlock_mount_hash();
2394 static struct mount *__do_loopback(struct path *old_path, int recurse)
2396 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2398 if (IS_MNT_UNBINDABLE(old))
2401 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2404 if (!recurse && has_locked_children(old, old_path->dentry))
2408 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2410 mnt = clone_mnt(old, old_path->dentry, 0);
2413 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2419 * do loopback mount.
2421 static int do_loopback(struct path *path, const char *old_name,
2424 struct path old_path;
2425 struct mount *mnt = NULL, *parent;
2426 struct mountpoint *mp;
2428 if (!old_name || !*old_name)
2430 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2435 if (mnt_ns_loop(old_path.dentry))
2438 mp = lock_mount(path);
2444 parent = real_mount(path->mnt);
2445 if (!check_mnt(parent))
2448 mnt = __do_loopback(&old_path, recurse);
2454 err = graft_tree(mnt, parent, mp);
2457 umount_tree(mnt, UMOUNT_SYNC);
2458 unlock_mount_hash();
2463 path_put(&old_path);
2467 static struct file *open_detached_copy(struct path *path, bool recursive)
2469 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2470 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2471 struct mount *mnt, *p;
2475 return ERR_CAST(ns);
2478 mnt = __do_loopback(path, recursive);
2482 return ERR_CAST(mnt);
2486 for (p = mnt; p; p = next_mnt(p, mnt)) {
2491 list_add_tail(&ns->list, &mnt->mnt_list);
2493 unlock_mount_hash();
2497 path->mnt = &mnt->mnt;
2498 file = dentry_open(path, O_PATH, current_cred());
2500 dissolve_on_fput(path->mnt);
2502 file->f_mode |= FMODE_NEED_UNMOUNT;
2506 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2510 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2511 bool detached = flags & OPEN_TREE_CLONE;
2515 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2517 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2518 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2522 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2525 if (flags & AT_NO_AUTOMOUNT)
2526 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2527 if (flags & AT_SYMLINK_NOFOLLOW)
2528 lookup_flags &= ~LOOKUP_FOLLOW;
2529 if (flags & AT_EMPTY_PATH)
2530 lookup_flags |= LOOKUP_EMPTY;
2532 if (detached && !may_mount())
2535 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2539 error = user_path_at(dfd, filename, lookup_flags, &path);
2540 if (unlikely(error)) {
2541 file = ERR_PTR(error);
2544 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2546 file = dentry_open(&path, O_PATH, current_cred());
2551 return PTR_ERR(file);
2553 fd_install(fd, file);
2558 * Don't allow locked mount flags to be cleared.
2560 * No locks need to be held here while testing the various MNT_LOCK
2561 * flags because those flags can never be cleared once they are set.
2563 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2565 unsigned int fl = mnt->mnt.mnt_flags;
2567 if ((fl & MNT_LOCK_READONLY) &&
2568 !(mnt_flags & MNT_READONLY))
2571 if ((fl & MNT_LOCK_NODEV) &&
2572 !(mnt_flags & MNT_NODEV))
2575 if ((fl & MNT_LOCK_NOSUID) &&
2576 !(mnt_flags & MNT_NOSUID))
2579 if ((fl & MNT_LOCK_NOEXEC) &&
2580 !(mnt_flags & MNT_NOEXEC))
2583 if ((fl & MNT_LOCK_ATIME) &&
2584 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2590 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2592 bool readonly_request = (mnt_flags & MNT_READONLY);
2594 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2597 if (readonly_request)
2598 return mnt_make_readonly(mnt);
2600 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2604 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2606 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2607 mnt->mnt.mnt_flags = mnt_flags;
2608 touch_mnt_namespace(mnt->mnt_ns);
2611 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2613 struct super_block *sb = mnt->mnt_sb;
2615 if (!__mnt_is_readonly(mnt) &&
2616 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2617 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2618 char *buf = (char *)__get_free_page(GFP_KERNEL);
2619 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2622 time64_to_tm(sb->s_time_max, 0, &tm);
2624 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2626 is_mounted(mnt) ? "remounted" : "mounted",
2628 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2630 free_page((unsigned long)buf);
2631 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2636 * Handle reconfiguration of the mountpoint only without alteration of the
2637 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2640 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2642 struct super_block *sb = path->mnt->mnt_sb;
2643 struct mount *mnt = real_mount(path->mnt);
2646 if (!check_mnt(mnt))
2649 if (path->dentry != mnt->mnt.mnt_root)
2652 if (!can_change_locked_flags(mnt, mnt_flags))
2656 * We're only checking whether the superblock is read-only not
2657 * changing it, so only take down_read(&sb->s_umount).
2659 down_read(&sb->s_umount);
2661 ret = change_mount_ro_state(mnt, mnt_flags);
2663 set_mount_attributes(mnt, mnt_flags);
2664 unlock_mount_hash();
2665 up_read(&sb->s_umount);
2667 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2673 * change filesystem flags. dir should be a physical root of filesystem.
2674 * If you've mounted a non-root directory somewhere and want to do remount
2675 * on it - tough luck.
2677 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2678 int mnt_flags, void *data)
2681 struct super_block *sb = path->mnt->mnt_sb;
2682 struct mount *mnt = real_mount(path->mnt);
2683 struct fs_context *fc;
2685 if (!check_mnt(mnt))
2688 if (path->dentry != path->mnt->mnt_root)
2691 if (!can_change_locked_flags(mnt, mnt_flags))
2694 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2699 err = parse_monolithic_mount_data(fc, data);
2701 down_write(&sb->s_umount);
2703 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2704 err = reconfigure_super(fc);
2707 set_mount_attributes(mnt, mnt_flags);
2708 unlock_mount_hash();
2711 up_write(&sb->s_umount);
2714 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2720 static inline int tree_contains_unbindable(struct mount *mnt)
2723 for (p = mnt; p; p = next_mnt(p, mnt)) {
2724 if (IS_MNT_UNBINDABLE(p))
2731 * Check that there aren't references to earlier/same mount namespaces in the
2732 * specified subtree. Such references can act as pins for mount namespaces
2733 * that aren't checked by the mount-cycle checking code, thereby allowing
2734 * cycles to be made.
2736 static bool check_for_nsfs_mounts(struct mount *subtree)
2742 for (p = subtree; p; p = next_mnt(p, subtree))
2743 if (mnt_ns_loop(p->mnt.mnt_root))
2748 unlock_mount_hash();
2752 static int do_set_group(struct path *from_path, struct path *to_path)
2754 struct mount *from, *to;
2757 from = real_mount(from_path->mnt);
2758 to = real_mount(to_path->mnt);
2763 /* To and From must be mounted */
2764 if (!is_mounted(&from->mnt))
2766 if (!is_mounted(&to->mnt))
2770 /* We should be allowed to modify mount namespaces of both mounts */
2771 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2773 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2777 /* To and From paths should be mount roots */
2778 if (from_path->dentry != from_path->mnt->mnt_root)
2780 if (to_path->dentry != to_path->mnt->mnt_root)
2783 /* Setting sharing groups is only allowed across same superblock */
2784 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2787 /* From mount root should be wider than To mount root */
2788 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2791 /* From mount should not have locked children in place of To's root */
2792 if (has_locked_children(from, to->mnt.mnt_root))
2795 /* Setting sharing groups is only allowed on private mounts */
2796 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2799 /* From should not be private */
2800 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2803 if (IS_MNT_SLAVE(from)) {
2804 struct mount *m = from->mnt_master;
2806 list_add(&to->mnt_slave, &m->mnt_slave_list);
2810 if (IS_MNT_SHARED(from)) {
2811 to->mnt_group_id = from->mnt_group_id;
2812 list_add(&to->mnt_share, &from->mnt_share);
2815 unlock_mount_hash();
2824 static int do_move_mount(struct path *old_path, struct path *new_path)
2826 struct mnt_namespace *ns;
2829 struct mount *parent;
2830 struct mountpoint *mp, *old_mp;
2834 mp = lock_mount(new_path);
2838 old = real_mount(old_path->mnt);
2839 p = real_mount(new_path->mnt);
2840 parent = old->mnt_parent;
2841 attached = mnt_has_parent(old);
2842 old_mp = old->mnt_mp;
2846 /* The mountpoint must be in our namespace. */
2850 /* The thing moved must be mounted... */
2851 if (!is_mounted(&old->mnt))
2854 /* ... and either ours or the root of anon namespace */
2855 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2858 if (old->mnt.mnt_flags & MNT_LOCKED)
2861 if (old_path->dentry != old_path->mnt->mnt_root)
2864 if (d_is_dir(new_path->dentry) !=
2865 d_is_dir(old_path->dentry))
2868 * Don't move a mount residing in a shared parent.
2870 if (attached && IS_MNT_SHARED(parent))
2873 * Don't move a mount tree containing unbindable mounts to a destination
2874 * mount which is shared.
2876 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2879 if (!check_for_nsfs_mounts(old))
2881 for (; mnt_has_parent(p); p = p->mnt_parent)
2885 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2890 /* if the mount is moved, it should no longer be expire
2892 list_del_init(&old->mnt_expire);
2894 put_mountpoint(old_mp);
2899 mntput_no_expire(parent);
2906 static int do_move_mount_old(struct path *path, const char *old_name)
2908 struct path old_path;
2911 if (!old_name || !*old_name)
2914 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2918 err = do_move_mount(&old_path, path);
2919 path_put(&old_path);
2924 * add a mount into a namespace's mount tree
2926 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2927 const struct path *path, int mnt_flags)
2929 struct mount *parent = real_mount(path->mnt);
2931 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2933 if (unlikely(!check_mnt(parent))) {
2934 /* that's acceptable only for automounts done in private ns */
2935 if (!(mnt_flags & MNT_SHRINKABLE))
2937 /* ... and for those we'd better have mountpoint still alive */
2938 if (!parent->mnt_ns)
2942 /* Refuse the same filesystem on the same mount point */
2943 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2944 path->mnt->mnt_root == path->dentry)
2947 if (d_is_symlink(newmnt->mnt.mnt_root))
2950 newmnt->mnt.mnt_flags = mnt_flags;
2951 return graft_tree(newmnt, parent, mp);
2954 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2957 * Create a new mount using a superblock configuration and request it
2958 * be added to the namespace tree.
2960 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2961 unsigned int mnt_flags)
2963 struct vfsmount *mnt;
2964 struct mountpoint *mp;
2965 struct super_block *sb = fc->root->d_sb;
2968 error = security_sb_kern_mount(sb);
2969 if (!error && mount_too_revealing(sb, &mnt_flags))
2972 if (unlikely(error)) {
2977 up_write(&sb->s_umount);
2979 mnt = vfs_create_mount(fc);
2981 return PTR_ERR(mnt);
2983 mnt_warn_timestamp_expiry(mountpoint, mnt);
2985 mp = lock_mount(mountpoint);
2990 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2998 * create a new mount for userspace and request it to be added into the
3001 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3002 int mnt_flags, const char *name, void *data)
3004 struct file_system_type *type;
3005 struct fs_context *fc;
3006 const char *subtype = NULL;
3012 type = get_fs_type(fstype);
3016 if (type->fs_flags & FS_HAS_SUBTYPE) {
3017 subtype = strchr(fstype, '.');
3021 put_filesystem(type);
3027 fc = fs_context_for_mount(type, sb_flags);
3028 put_filesystem(type);
3033 err = vfs_parse_fs_string(fc, "subtype",
3034 subtype, strlen(subtype));
3036 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3038 err = parse_monolithic_mount_data(fc, data);
3039 if (!err && !mount_capable(fc))
3042 err = vfs_get_tree(fc);
3044 err = do_new_mount_fc(fc, path, mnt_flags);
3050 int finish_automount(struct vfsmount *m, const struct path *path)
3052 struct dentry *dentry = path->dentry;
3053 struct mountpoint *mp;
3062 mnt = real_mount(m);
3063 /* The new mount record should have at least 2 refs to prevent it being
3064 * expired before we get a chance to add it
3066 BUG_ON(mnt_get_count(mnt) < 2);
3068 if (m->mnt_sb == path->mnt->mnt_sb &&
3069 m->mnt_root == dentry) {
3075 * we don't want to use lock_mount() - in this case finding something
3076 * that overmounts our mountpoint to be means "quitely drop what we've
3077 * got", not "try to mount it on top".
3079 inode_lock(dentry->d_inode);
3081 if (unlikely(cant_mount(dentry))) {
3083 goto discard_locked;
3086 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3089 goto discard_locked;
3092 mp = get_mountpoint(dentry);
3095 goto discard_locked;
3098 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3107 inode_unlock(dentry->d_inode);
3109 /* remove m from any expiration list it may be on */
3110 if (!list_empty(&mnt->mnt_expire)) {
3112 list_del_init(&mnt->mnt_expire);
3121 * mnt_set_expiry - Put a mount on an expiration list
3122 * @mnt: The mount to list.
3123 * @expiry_list: The list to add the mount to.
3125 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3129 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3133 EXPORT_SYMBOL(mnt_set_expiry);
3136 * process a list of expirable mountpoints with the intent of discarding any
3137 * mountpoints that aren't in use and haven't been touched since last we came
3140 void mark_mounts_for_expiry(struct list_head *mounts)
3142 struct mount *mnt, *next;
3143 LIST_HEAD(graveyard);
3145 if (list_empty(mounts))
3151 /* extract from the expiration list every vfsmount that matches the
3152 * following criteria:
3153 * - only referenced by its parent vfsmount
3154 * - still marked for expiry (marked on the last call here; marks are
3155 * cleared by mntput())
3157 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3158 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3159 propagate_mount_busy(mnt, 1))
3161 list_move(&mnt->mnt_expire, &graveyard);
3163 while (!list_empty(&graveyard)) {
3164 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3165 touch_mnt_namespace(mnt->mnt_ns);
3166 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3168 unlock_mount_hash();
3172 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3175 * Ripoff of 'select_parent()'
3177 * search the list of submounts for a given mountpoint, and move any
3178 * shrinkable submounts to the 'graveyard' list.
3180 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3182 struct mount *this_parent = parent;
3183 struct list_head *next;
3187 next = this_parent->mnt_mounts.next;
3189 while (next != &this_parent->mnt_mounts) {
3190 struct list_head *tmp = next;
3191 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3194 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3197 * Descend a level if the d_mounts list is non-empty.
3199 if (!list_empty(&mnt->mnt_mounts)) {
3204 if (!propagate_mount_busy(mnt, 1)) {
3205 list_move_tail(&mnt->mnt_expire, graveyard);
3210 * All done at this level ... ascend and resume the search
3212 if (this_parent != parent) {
3213 next = this_parent->mnt_child.next;
3214 this_parent = this_parent->mnt_parent;
3221 * process a list of expirable mountpoints with the intent of discarding any
3222 * submounts of a specific parent mountpoint
3224 * mount_lock must be held for write
3226 static void shrink_submounts(struct mount *mnt)
3228 LIST_HEAD(graveyard);
3231 /* extract submounts of 'mountpoint' from the expiration list */
3232 while (select_submounts(mnt, &graveyard)) {
3233 while (!list_empty(&graveyard)) {
3234 m = list_first_entry(&graveyard, struct mount,
3236 touch_mnt_namespace(m->mnt_ns);
3237 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3242 static void *copy_mount_options(const void __user * data)
3245 unsigned left, offset;
3250 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3252 return ERR_PTR(-ENOMEM);
3254 left = copy_from_user(copy, data, PAGE_SIZE);
3257 * Not all architectures have an exact copy_from_user(). Resort to
3260 offset = PAGE_SIZE - left;
3263 if (get_user(c, (const char __user *)data + offset))
3270 if (left == PAGE_SIZE) {
3272 return ERR_PTR(-EFAULT);
3278 static char *copy_mount_string(const void __user *data)
3280 return data ? strndup_user(data, PATH_MAX) : NULL;
3284 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3285 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3287 * data is a (void *) that can point to any structure up to
3288 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3289 * information (or be NULL).
3291 * Pre-0.97 versions of mount() didn't have a flags word.
3292 * When the flags word was introduced its top half was required
3293 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3294 * Therefore, if this magic number is present, it carries no information
3295 * and must be discarded.
3297 int path_mount(const char *dev_name, struct path *path,
3298 const char *type_page, unsigned long flags, void *data_page)
3300 unsigned int mnt_flags = 0, sb_flags;
3304 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3305 flags &= ~MS_MGC_MSK;
3307 /* Basic sanity checks */
3309 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3311 if (flags & MS_NOUSER)
3314 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3319 if (flags & SB_MANDLOCK)
3322 /* Default to relatime unless overriden */
3323 if (!(flags & MS_NOATIME))
3324 mnt_flags |= MNT_RELATIME;
3326 /* Separate the per-mountpoint flags */
3327 if (flags & MS_NOSUID)
3328 mnt_flags |= MNT_NOSUID;
3329 if (flags & MS_NODEV)
3330 mnt_flags |= MNT_NODEV;
3331 if (flags & MS_NOEXEC)
3332 mnt_flags |= MNT_NOEXEC;
3333 if (flags & MS_NOATIME)
3334 mnt_flags |= MNT_NOATIME;
3335 if (flags & MS_NODIRATIME)
3336 mnt_flags |= MNT_NODIRATIME;
3337 if (flags & MS_STRICTATIME)
3338 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3339 if (flags & MS_RDONLY)
3340 mnt_flags |= MNT_READONLY;
3341 if (flags & MS_NOSYMFOLLOW)
3342 mnt_flags |= MNT_NOSYMFOLLOW;
3344 /* The default atime for remount is preservation */
3345 if ((flags & MS_REMOUNT) &&
3346 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3347 MS_STRICTATIME)) == 0)) {
3348 mnt_flags &= ~MNT_ATIME_MASK;
3349 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3352 sb_flags = flags & (SB_RDONLY |
3361 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3362 return do_reconfigure_mnt(path, mnt_flags);
3363 if (flags & MS_REMOUNT)
3364 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3365 if (flags & MS_BIND)
3366 return do_loopback(path, dev_name, flags & MS_REC);
3367 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3368 return do_change_type(path, flags);
3369 if (flags & MS_MOVE)
3370 return do_move_mount_old(path, dev_name);
3372 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3376 long do_mount(const char *dev_name, const char __user *dir_name,
3377 const char *type_page, unsigned long flags, void *data_page)
3382 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3385 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3390 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3392 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3395 static void dec_mnt_namespaces(struct ucounts *ucounts)
3397 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3400 static void free_mnt_ns(struct mnt_namespace *ns)
3402 if (!is_anon_ns(ns))
3403 ns_free_inum(&ns->ns);
3404 dec_mnt_namespaces(ns->ucounts);
3405 put_user_ns(ns->user_ns);
3410 * Assign a sequence number so we can detect when we attempt to bind
3411 * mount a reference to an older mount namespace into the current
3412 * mount namespace, preventing reference counting loops. A 64bit
3413 * number incrementing at 10Ghz will take 12,427 years to wrap which
3414 * is effectively never, so we can ignore the possibility.
3416 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3418 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3420 struct mnt_namespace *new_ns;
3421 struct ucounts *ucounts;
3424 ucounts = inc_mnt_namespaces(user_ns);
3426 return ERR_PTR(-ENOSPC);
3428 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3430 dec_mnt_namespaces(ucounts);
3431 return ERR_PTR(-ENOMEM);
3434 ret = ns_alloc_inum(&new_ns->ns);
3437 dec_mnt_namespaces(ucounts);
3438 return ERR_PTR(ret);
3441 new_ns->ns.ops = &mntns_operations;
3443 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3444 refcount_set(&new_ns->ns.count, 1);
3445 INIT_LIST_HEAD(&new_ns->list);
3446 init_waitqueue_head(&new_ns->poll);
3447 spin_lock_init(&new_ns->ns_lock);
3448 new_ns->user_ns = get_user_ns(user_ns);
3449 new_ns->ucounts = ucounts;
3454 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3455 struct user_namespace *user_ns, struct fs_struct *new_fs)
3457 struct mnt_namespace *new_ns;
3458 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3459 struct mount *p, *q;
3466 if (likely(!(flags & CLONE_NEWNS))) {
3473 new_ns = alloc_mnt_ns(user_ns, false);
3478 /* First pass: copy the tree topology */
3479 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3480 if (user_ns != ns->user_ns)
3481 copy_flags |= CL_SHARED_TO_SLAVE;
3482 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3485 free_mnt_ns(new_ns);
3486 return ERR_CAST(new);
3488 if (user_ns != ns->user_ns) {
3491 unlock_mount_hash();
3494 list_add_tail(&new_ns->list, &new->mnt_list);
3497 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3498 * as belonging to new namespace. We have already acquired a private
3499 * fs_struct, so tsk->fs->lock is not needed.
3507 if (&p->mnt == new_fs->root.mnt) {
3508 new_fs->root.mnt = mntget(&q->mnt);
3511 if (&p->mnt == new_fs->pwd.mnt) {
3512 new_fs->pwd.mnt = mntget(&q->mnt);
3516 p = next_mnt(p, old);
3517 q = next_mnt(q, new);
3520 // an mntns binding we'd skipped?
3521 while (p->mnt.mnt_root != q->mnt.mnt_root)
3522 p = next_mnt(skip_mnt_tree(p), old);
3534 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3536 struct mount *mnt = real_mount(m);
3537 struct mnt_namespace *ns;
3538 struct super_block *s;
3542 ns = alloc_mnt_ns(&init_user_ns, true);
3545 return ERR_CAST(ns);
3550 list_add(&mnt->mnt_list, &ns->list);
3552 err = vfs_path_lookup(m->mnt_root, m,
3553 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3558 return ERR_PTR(err);
3560 /* trade a vfsmount reference for active sb one */
3561 s = path.mnt->mnt_sb;
3562 atomic_inc(&s->s_active);
3564 /* lock the sucker */
3565 down_write(&s->s_umount);
3566 /* ... and return the root of (sub)tree on it */
3569 EXPORT_SYMBOL(mount_subtree);
3571 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3572 char __user *, type, unsigned long, flags, void __user *, data)
3579 kernel_type = copy_mount_string(type);
3580 ret = PTR_ERR(kernel_type);
3581 if (IS_ERR(kernel_type))
3584 kernel_dev = copy_mount_string(dev_name);
3585 ret = PTR_ERR(kernel_dev);
3586 if (IS_ERR(kernel_dev))
3589 options = copy_mount_options(data);
3590 ret = PTR_ERR(options);
3591 if (IS_ERR(options))
3594 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3605 #define FSMOUNT_VALID_FLAGS \
3606 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3607 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3608 MOUNT_ATTR_NOSYMFOLLOW)
3610 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3612 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3613 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3615 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3617 unsigned int mnt_flags = 0;
3619 if (attr_flags & MOUNT_ATTR_RDONLY)
3620 mnt_flags |= MNT_READONLY;
3621 if (attr_flags & MOUNT_ATTR_NOSUID)
3622 mnt_flags |= MNT_NOSUID;
3623 if (attr_flags & MOUNT_ATTR_NODEV)
3624 mnt_flags |= MNT_NODEV;
3625 if (attr_flags & MOUNT_ATTR_NOEXEC)
3626 mnt_flags |= MNT_NOEXEC;
3627 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3628 mnt_flags |= MNT_NODIRATIME;
3629 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3630 mnt_flags |= MNT_NOSYMFOLLOW;
3636 * Create a kernel mount representation for a new, prepared superblock
3637 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3639 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3640 unsigned int, attr_flags)
3642 struct mnt_namespace *ns;
3643 struct fs_context *fc;
3645 struct path newmount;
3648 unsigned int mnt_flags = 0;
3654 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3657 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3660 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3662 switch (attr_flags & MOUNT_ATTR__ATIME) {
3663 case MOUNT_ATTR_STRICTATIME:
3665 case MOUNT_ATTR_NOATIME:
3666 mnt_flags |= MNT_NOATIME;
3668 case MOUNT_ATTR_RELATIME:
3669 mnt_flags |= MNT_RELATIME;
3680 if (f.file->f_op != &fscontext_fops)
3683 fc = f.file->private_data;
3685 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3689 /* There must be a valid superblock or we can't mount it */
3695 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3696 pr_warn("VFS: Mount too revealing\n");
3701 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3704 if (fc->sb_flags & SB_MANDLOCK)
3707 newmount.mnt = vfs_create_mount(fc);
3708 if (IS_ERR(newmount.mnt)) {
3709 ret = PTR_ERR(newmount.mnt);
3712 newmount.dentry = dget(fc->root);
3713 newmount.mnt->mnt_flags = mnt_flags;
3715 /* We've done the mount bit - now move the file context into more or
3716 * less the same state as if we'd done an fspick(). We don't want to
3717 * do any memory allocation or anything like that at this point as we
3718 * don't want to have to handle any errors incurred.
3720 vfs_clean_context(fc);
3722 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3727 mnt = real_mount(newmount.mnt);
3731 list_add(&mnt->mnt_list, &ns->list);
3732 mntget(newmount.mnt);
3734 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3735 * it, not just simply put it.
3737 file = dentry_open(&newmount, O_PATH, fc->cred);
3739 dissolve_on_fput(newmount.mnt);
3740 ret = PTR_ERR(file);
3743 file->f_mode |= FMODE_NEED_UNMOUNT;
3745 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3747 fd_install(ret, file);
3752 path_put(&newmount);
3754 mutex_unlock(&fc->uapi_mutex);
3761 * Move a mount from one place to another. In combination with
3762 * fsopen()/fsmount() this is used to install a new mount and in combination
3763 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3766 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3768 SYSCALL_DEFINE5(move_mount,
3769 int, from_dfd, const char __user *, from_pathname,
3770 int, to_dfd, const char __user *, to_pathname,
3771 unsigned int, flags)
3773 struct path from_path, to_path;
3774 unsigned int lflags;
3780 if (flags & ~MOVE_MOUNT__MASK)
3783 /* If someone gives a pathname, they aren't permitted to move
3784 * from an fd that requires unmount as we can't get at the flag
3785 * to clear it afterwards.
3788 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3789 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3790 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3792 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3797 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3798 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3799 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3801 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3805 ret = security_move_mount(&from_path, &to_path);
3809 if (flags & MOVE_MOUNT_SET_GROUP)
3810 ret = do_set_group(&from_path, &to_path);
3812 ret = do_move_mount(&from_path, &to_path);
3817 path_put(&from_path);
3822 * Return true if path is reachable from root
3824 * namespace_sem or mount_lock is held
3826 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3827 const struct path *root)
3829 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3830 dentry = mnt->mnt_mountpoint;
3831 mnt = mnt->mnt_parent;
3833 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3836 bool path_is_under(const struct path *path1, const struct path *path2)
3839 read_seqlock_excl(&mount_lock);
3840 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3841 read_sequnlock_excl(&mount_lock);
3844 EXPORT_SYMBOL(path_is_under);
3847 * pivot_root Semantics:
3848 * Moves the root file system of the current process to the directory put_old,
3849 * makes new_root as the new root file system of the current process, and sets
3850 * root/cwd of all processes which had them on the current root to new_root.
3853 * The new_root and put_old must be directories, and must not be on the
3854 * same file system as the current process root. The put_old must be
3855 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3856 * pointed to by put_old must yield the same directory as new_root. No other
3857 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3859 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3860 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3861 * in this situation.
3864 * - we don't move root/cwd if they are not at the root (reason: if something
3865 * cared enough to change them, it's probably wrong to force them elsewhere)
3866 * - it's okay to pick a root that isn't the root of a file system, e.g.
3867 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3868 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3871 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3872 const char __user *, put_old)
3874 struct path new, old, root;
3875 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3876 struct mountpoint *old_mp, *root_mp;
3882 error = user_path_at(AT_FDCWD, new_root,
3883 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3887 error = user_path_at(AT_FDCWD, put_old,
3888 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3892 error = security_sb_pivotroot(&old, &new);
3896 get_fs_root(current->fs, &root);
3897 old_mp = lock_mount(&old);
3898 error = PTR_ERR(old_mp);
3903 new_mnt = real_mount(new.mnt);
3904 root_mnt = real_mount(root.mnt);
3905 old_mnt = real_mount(old.mnt);
3906 ex_parent = new_mnt->mnt_parent;
3907 root_parent = root_mnt->mnt_parent;
3908 if (IS_MNT_SHARED(old_mnt) ||
3909 IS_MNT_SHARED(ex_parent) ||
3910 IS_MNT_SHARED(root_parent))
3912 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3914 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3917 if (d_unlinked(new.dentry))
3920 if (new_mnt == root_mnt || old_mnt == root_mnt)
3921 goto out4; /* loop, on the same file system */
3923 if (root.mnt->mnt_root != root.dentry)
3924 goto out4; /* not a mountpoint */
3925 if (!mnt_has_parent(root_mnt))
3926 goto out4; /* not attached */
3927 if (new.mnt->mnt_root != new.dentry)
3928 goto out4; /* not a mountpoint */
3929 if (!mnt_has_parent(new_mnt))
3930 goto out4; /* not attached */
3931 /* make sure we can reach put_old from new_root */
3932 if (!is_path_reachable(old_mnt, old.dentry, &new))
3934 /* make certain new is below the root */
3935 if (!is_path_reachable(new_mnt, new.dentry, &root))
3938 umount_mnt(new_mnt);
3939 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3940 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3941 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3942 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3944 /* mount old root on put_old */
3945 attach_mnt(root_mnt, old_mnt, old_mp);
3946 /* mount new_root on / */
3947 attach_mnt(new_mnt, root_parent, root_mp);
3948 mnt_add_count(root_parent, -1);
3949 touch_mnt_namespace(current->nsproxy->mnt_ns);
3950 /* A moved mount should not expire automatically */
3951 list_del_init(&new_mnt->mnt_expire);
3952 put_mountpoint(root_mp);
3953 unlock_mount_hash();
3954 chroot_fs_refs(&root, &new);
3957 unlock_mount(old_mp);
3959 mntput_no_expire(ex_parent);
3970 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3972 unsigned int flags = mnt->mnt.mnt_flags;
3974 /* flags to clear */
3975 flags &= ~kattr->attr_clr;
3976 /* flags to raise */
3977 flags |= kattr->attr_set;
3982 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3984 struct vfsmount *m = &mnt->mnt;
3985 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
3987 if (!kattr->mnt_idmap)
3991 * Creating an idmapped mount with the filesystem wide idmapping
3992 * doesn't make sense so block that. We don't allow mushy semantics.
3994 if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb))
3998 * Once a mount has been idmapped we don't allow it to change its
3999 * mapping. It makes things simpler and callers can just create
4000 * another bind-mount they can idmap if they want to.
4002 if (is_idmapped_mnt(m))
4005 /* The underlying filesystem doesn't support idmapped mounts yet. */
4006 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4009 /* We're not controlling the superblock. */
4010 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4013 /* Mount has already been visible in the filesystem hierarchy. */
4014 if (!is_anon_ns(mnt->mnt_ns))
4021 * mnt_allow_writers() - check whether the attribute change allows writers
4022 * @kattr: the new mount attributes
4023 * @mnt: the mount to which @kattr will be applied
4025 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4027 * Return: true if writers need to be held, false if not
4029 static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4030 const struct mount *mnt)
4032 return (!(kattr->attr_set & MNT_READONLY) ||
4033 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4037 static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4042 for (m = mnt; m; m = next_mnt(m, mnt)) {
4043 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4048 err = can_idmap_mount(kattr, m);
4052 if (!mnt_allow_writers(kattr, m)) {
4053 err = mnt_hold_writers(m);
4058 if (!kattr->recurse)
4066 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4067 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4068 * mounts and needs to take care to include the first mount.
4070 for (p = mnt; p; p = next_mnt(p, mnt)) {
4071 /* If we had to hold writers unblock them. */
4072 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4073 mnt_unhold_writers(p);
4076 * We're done once the first mount we changed got
4077 * MNT_WRITE_HOLD unset.
4086 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4088 if (!kattr->mnt_idmap)
4092 * Pairs with smp_load_acquire() in mnt_idmap().
4094 * Since we only allow a mount to change the idmapping once and
4095 * verified this in can_idmap_mount() we know that the mount has
4096 * @nop_mnt_idmap attached to it. So there's no need to drop any
4099 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4102 static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4106 for (m = mnt; m; m = next_mnt(m, mnt)) {
4109 do_idmap_mount(kattr, m);
4110 flags = recalc_flags(kattr, m);
4111 WRITE_ONCE(m->mnt.mnt_flags, flags);
4113 /* If we had to hold writers unblock them. */
4114 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4115 mnt_unhold_writers(m);
4117 if (kattr->propagation)
4118 change_mnt_propagation(m, kattr->propagation);
4119 if (!kattr->recurse)
4122 touch_mnt_namespace(mnt->mnt_ns);
4125 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4127 struct mount *mnt = real_mount(path->mnt);
4130 if (path->dentry != mnt->mnt.mnt_root)
4133 if (kattr->mnt_userns) {
4134 struct mnt_idmap *mnt_idmap;
4136 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4137 if (IS_ERR(mnt_idmap))
4138 return PTR_ERR(mnt_idmap);
4139 kattr->mnt_idmap = mnt_idmap;
4142 if (kattr->propagation) {
4144 * Only take namespace_lock() if we're actually changing
4148 if (kattr->propagation == MS_SHARED) {
4149 err = invent_group_ids(mnt, kattr->recurse);
4160 /* Ensure that this isn't anything purely vfs internal. */
4161 if (!is_mounted(&mnt->mnt))
4165 * If this is an attached mount make sure it's located in the callers
4166 * mount namespace. If it's not don't let the caller interact with it.
4167 * If this is a detached mount make sure it has an anonymous mount
4168 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4170 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4174 * First, we get the mount tree in a shape where we can change mount
4175 * properties without failure. If we succeeded to do so we commit all
4176 * changes and if we failed we clean up.
4178 err = mount_setattr_prepare(kattr, mnt);
4180 mount_setattr_commit(kattr, mnt);
4183 unlock_mount_hash();
4185 if (kattr->propagation) {
4188 cleanup_group_ids(mnt, NULL);
4194 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4195 struct mount_kattr *kattr, unsigned int flags)
4198 struct ns_common *ns;
4199 struct user_namespace *mnt_userns;
4202 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4206 * We currently do not support clearing an idmapped mount. If this ever
4207 * is a use-case we can revisit this but for now let's keep it simple
4210 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4213 if (attr->userns_fd > INT_MAX)
4216 file = fget(attr->userns_fd);
4220 if (!proc_ns_file(file)) {
4225 ns = get_proc_ns(file_inode(file));
4226 if (ns->ops->type != CLONE_NEWUSER) {
4232 * The initial idmapping cannot be used to create an idmapped
4233 * mount. We use the initial idmapping as an indicator of a mount
4234 * that is not idmapped. It can simply be passed into helpers that
4235 * are aware of idmapped mounts as a convenient shortcut. A user
4236 * can just create a dedicated identity mapping to achieve the same
4239 mnt_userns = container_of(ns, struct user_namespace, ns);
4240 if (mnt_userns == &init_user_ns) {
4245 /* We're not controlling the target namespace. */
4246 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4251 kattr->mnt_userns = get_user_ns(mnt_userns);
4258 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4259 struct mount_kattr *kattr, unsigned int flags)
4261 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4263 if (flags & AT_NO_AUTOMOUNT)
4264 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4265 if (flags & AT_SYMLINK_NOFOLLOW)
4266 lookup_flags &= ~LOOKUP_FOLLOW;
4267 if (flags & AT_EMPTY_PATH)
4268 lookup_flags |= LOOKUP_EMPTY;
4270 *kattr = (struct mount_kattr) {
4271 .lookup_flags = lookup_flags,
4272 .recurse = !!(flags & AT_RECURSIVE),
4275 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4277 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4279 kattr->propagation = attr->propagation;
4281 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4284 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4285 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4288 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4289 * users wanting to transition to a different atime setting cannot
4290 * simply specify the atime setting in @attr_set, but must also
4291 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4292 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4293 * @attr_clr and that @attr_set can't have any atime bits set if
4294 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4296 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4297 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4301 * Clear all previous time settings as they are mutually
4304 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4305 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4306 case MOUNT_ATTR_RELATIME:
4307 kattr->attr_set |= MNT_RELATIME;
4309 case MOUNT_ATTR_NOATIME:
4310 kattr->attr_set |= MNT_NOATIME;
4312 case MOUNT_ATTR_STRICTATIME:
4318 if (attr->attr_set & MOUNT_ATTR__ATIME)
4322 return build_mount_idmapped(attr, usize, kattr, flags);
4325 static void finish_mount_kattr(struct mount_kattr *kattr)
4327 put_user_ns(kattr->mnt_userns);
4328 kattr->mnt_userns = NULL;
4330 if (kattr->mnt_idmap)
4331 mnt_idmap_put(kattr->mnt_idmap);
4334 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4335 unsigned int, flags, struct mount_attr __user *, uattr,
4340 struct mount_attr attr;
4341 struct mount_kattr kattr;
4343 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4345 if (flags & ~(AT_EMPTY_PATH |
4347 AT_SYMLINK_NOFOLLOW |
4351 if (unlikely(usize > PAGE_SIZE))
4353 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4359 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4363 /* Don't bother walking through the mounts if this is a nop. */
4364 if (attr.attr_set == 0 &&
4365 attr.attr_clr == 0 &&
4366 attr.propagation == 0)
4369 err = build_mount_kattr(&attr, usize, &kattr, flags);
4373 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4375 err = do_mount_setattr(&target, &kattr);
4378 finish_mount_kattr(&kattr);
4382 static void __init init_mount_tree(void)
4384 struct vfsmount *mnt;
4386 struct mnt_namespace *ns;
4389 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4391 panic("Can't create rootfs");
4393 ns = alloc_mnt_ns(&init_user_ns, false);
4395 panic("Can't allocate initial namespace");
4396 m = real_mount(mnt);
4400 list_add(&m->mnt_list, &ns->list);
4401 init_task.nsproxy->mnt_ns = ns;
4405 root.dentry = mnt->mnt_root;
4406 mnt->mnt_flags |= MNT_LOCKED;
4408 set_fs_pwd(current->fs, &root);
4409 set_fs_root(current->fs, &root);
4412 void __init mnt_init(void)
4416 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4417 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4419 mount_hashtable = alloc_large_system_hash("Mount-cache",
4420 sizeof(struct hlist_head),
4423 &m_hash_shift, &m_hash_mask, 0, 0);
4424 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4425 sizeof(struct hlist_head),
4428 &mp_hash_shift, &mp_hash_mask, 0, 0);
4430 if (!mount_hashtable || !mountpoint_hashtable)
4431 panic("Failed to allocate mount hash table\n");
4437 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4439 fs_kobj = kobject_create_and_add("fs", NULL);
4441 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4447 void put_mnt_ns(struct mnt_namespace *ns)
4449 if (!refcount_dec_and_test(&ns->ns.count))
4451 drop_collected_mounts(&ns->root->mnt);
4455 struct vfsmount *kern_mount(struct file_system_type *type)
4457 struct vfsmount *mnt;
4458 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4461 * it is a longterm mount, don't release mnt until
4462 * we unmount before file sys is unregistered
4464 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4468 EXPORT_SYMBOL_GPL(kern_mount);
4470 void kern_unmount(struct vfsmount *mnt)
4472 /* release long term mount so mount point can be released */
4474 mnt_make_shortterm(mnt);
4475 synchronize_rcu(); /* yecchhh... */
4479 EXPORT_SYMBOL(kern_unmount);
4481 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4485 for (i = 0; i < num; i++)
4486 mnt_make_shortterm(mnt[i]);
4487 synchronize_rcu_expedited();
4488 for (i = 0; i < num; i++)
4491 EXPORT_SYMBOL(kern_unmount_array);
4493 bool our_mnt(struct vfsmount *mnt)
4495 return check_mnt(real_mount(mnt));
4498 bool current_chrooted(void)
4500 /* Does the current process have a non-standard root */
4501 struct path ns_root;
4502 struct path fs_root;
4505 /* Find the namespace root */
4506 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4507 ns_root.dentry = ns_root.mnt->mnt_root;
4509 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4512 get_fs_root(current->fs, &fs_root);
4514 chrooted = !path_equal(&fs_root, &ns_root);
4522 static bool mnt_already_visible(struct mnt_namespace *ns,
4523 const struct super_block *sb,
4526 int new_flags = *new_mnt_flags;
4528 bool visible = false;
4530 down_read(&namespace_sem);
4532 list_for_each_entry(mnt, &ns->list, mnt_list) {
4533 struct mount *child;
4536 if (mnt_is_cursor(mnt))
4539 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4542 /* This mount is not fully visible if it's root directory
4543 * is not the root directory of the filesystem.
4545 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4548 /* A local view of the mount flags */
4549 mnt_flags = mnt->mnt.mnt_flags;
4551 /* Don't miss readonly hidden in the superblock flags */
4552 if (sb_rdonly(mnt->mnt.mnt_sb))
4553 mnt_flags |= MNT_LOCK_READONLY;
4555 /* Verify the mount flags are equal to or more permissive
4556 * than the proposed new mount.
4558 if ((mnt_flags & MNT_LOCK_READONLY) &&
4559 !(new_flags & MNT_READONLY))
4561 if ((mnt_flags & MNT_LOCK_ATIME) &&
4562 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4565 /* This mount is not fully visible if there are any
4566 * locked child mounts that cover anything except for
4567 * empty directories.
4569 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4570 struct inode *inode = child->mnt_mountpoint->d_inode;
4571 /* Only worry about locked mounts */
4572 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4574 /* Is the directory permanetly empty? */
4575 if (!is_empty_dir_inode(inode))
4578 /* Preserve the locked attributes */
4579 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4587 up_read(&namespace_sem);
4591 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4593 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4594 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4595 unsigned long s_iflags;
4597 if (ns->user_ns == &init_user_ns)
4600 /* Can this filesystem be too revealing? */
4601 s_iflags = sb->s_iflags;
4602 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4605 if ((s_iflags & required_iflags) != required_iflags) {
4606 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4611 return !mnt_already_visible(ns, sb, new_mnt_flags);
4614 bool mnt_may_suid(struct vfsmount *mnt)
4617 * Foreign mounts (accessed via fchdir or through /proc
4618 * symlinks) are always treated as if they are nosuid. This
4619 * prevents namespaces from trusting potentially unsafe
4620 * suid/sgid bits, file caps, or security labels that originate
4621 * in other namespaces.
4623 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4624 current_in_userns(mnt->mnt_sb->s_user_ns);
4627 static struct ns_common *mntns_get(struct task_struct *task)
4629 struct ns_common *ns = NULL;
4630 struct nsproxy *nsproxy;
4633 nsproxy = task->nsproxy;
4635 ns = &nsproxy->mnt_ns->ns;
4636 get_mnt_ns(to_mnt_ns(ns));
4643 static void mntns_put(struct ns_common *ns)
4645 put_mnt_ns(to_mnt_ns(ns));
4648 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4650 struct nsproxy *nsproxy = nsset->nsproxy;
4651 struct fs_struct *fs = nsset->fs;
4652 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4653 struct user_namespace *user_ns = nsset->cred->user_ns;
4657 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4658 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4659 !ns_capable(user_ns, CAP_SYS_ADMIN))
4662 if (is_anon_ns(mnt_ns))
4669 old_mnt_ns = nsproxy->mnt_ns;
4670 nsproxy->mnt_ns = mnt_ns;
4673 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4674 "/", LOOKUP_DOWN, &root);
4676 /* revert to old namespace */
4677 nsproxy->mnt_ns = old_mnt_ns;
4682 put_mnt_ns(old_mnt_ns);
4684 /* Update the pwd and root */
4685 set_fs_pwd(fs, &root);
4686 set_fs_root(fs, &root);
4692 static struct user_namespace *mntns_owner(struct ns_common *ns)
4694 return to_mnt_ns(ns)->user_ns;
4697 const struct proc_ns_operations mntns_operations = {
4699 .type = CLONE_NEWNS,
4702 .install = mntns_install,
4703 .owner = mntns_owner,
4706 #ifdef CONFIG_SYSCTL
4707 static struct ctl_table fs_namespace_sysctls[] = {
4709 .procname = "mount-max",
4710 .data = &sysctl_mount_max,
4711 .maxlen = sizeof(unsigned int),
4713 .proc_handler = proc_dointvec_minmax,
4714 .extra1 = SYSCTL_ONE,
4719 static int __init init_fs_namespace_sysctls(void)
4721 register_sysctl_init("fs", fs_namespace_sysctls);
4724 fs_initcall(init_fs_namespace_sysctls);
4726 #endif /* CONFIG_SYSCTL */