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 * __lookup_mnt - find first child mount
664 * @dentry: mountpoint
666 * If @mnt has a child mount @c mounted @dentry find and return it.
668 * Note that the child mount @c need not be unique. There are cases
669 * where shadow mounts are created. For example, during mount
670 * propagation when a source mount @mnt whose root got overmounted by a
671 * mount @o after path lookup but before @namespace_sem could be
672 * acquired gets copied and propagated. So @mnt gets copied including
673 * @o. When @mnt is propagated to a destination mount @d that already
674 * has another mount @n mounted at the same mountpoint then the source
675 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
676 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
679 * Return: The first child of @mnt mounted @dentry or NULL.
681 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
683 struct hlist_head *head = m_hash(mnt, dentry);
686 hlist_for_each_entry_rcu(p, head, mnt_hash)
687 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
693 * lookup_mnt - Return the first child mount mounted at path
695 * "First" means first mounted chronologically. If you create the
698 * mount /dev/sda1 /mnt
699 * mount /dev/sda2 /mnt
700 * mount /dev/sda3 /mnt
702 * Then lookup_mnt() on the base /mnt dentry in the root mount will
703 * return successively the root dentry and vfsmount of /dev/sda1, then
704 * /dev/sda2, then /dev/sda3, then NULL.
706 * lookup_mnt takes a reference to the found vfsmount.
708 struct vfsmount *lookup_mnt(const struct path *path)
710 struct mount *child_mnt;
716 seq = read_seqbegin(&mount_lock);
717 child_mnt = __lookup_mnt(path->mnt, path->dentry);
718 m = child_mnt ? &child_mnt->mnt : NULL;
719 } while (!legitimize_mnt(m, seq));
724 static inline void lock_ns_list(struct mnt_namespace *ns)
726 spin_lock(&ns->ns_lock);
729 static inline void unlock_ns_list(struct mnt_namespace *ns)
731 spin_unlock(&ns->ns_lock);
734 static inline bool mnt_is_cursor(struct mount *mnt)
736 return mnt->mnt.mnt_flags & MNT_CURSOR;
740 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
741 * current mount namespace.
743 * The common case is dentries are not mountpoints at all and that
744 * test is handled inline. For the slow case when we are actually
745 * dealing with a mountpoint of some kind, walk through all of the
746 * mounts in the current mount namespace and test to see if the dentry
749 * The mount_hashtable is not usable in the context because we
750 * need to identify all mounts that may be in the current mount
751 * namespace not just a mount that happens to have some specified
754 bool __is_local_mountpoint(struct dentry *dentry)
756 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
758 bool is_covered = false;
760 down_read(&namespace_sem);
762 list_for_each_entry(mnt, &ns->list, mnt_list) {
763 if (mnt_is_cursor(mnt))
765 is_covered = (mnt->mnt_mountpoint == dentry);
770 up_read(&namespace_sem);
775 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
777 struct hlist_head *chain = mp_hash(dentry);
778 struct mountpoint *mp;
780 hlist_for_each_entry(mp, chain, m_hash) {
781 if (mp->m_dentry == dentry) {
789 static struct mountpoint *get_mountpoint(struct dentry *dentry)
791 struct mountpoint *mp, *new = NULL;
794 if (d_mountpoint(dentry)) {
795 /* might be worth a WARN_ON() */
796 if (d_unlinked(dentry))
797 return ERR_PTR(-ENOENT);
799 read_seqlock_excl(&mount_lock);
800 mp = lookup_mountpoint(dentry);
801 read_sequnlock_excl(&mount_lock);
807 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
809 return ERR_PTR(-ENOMEM);
812 /* Exactly one processes may set d_mounted */
813 ret = d_set_mounted(dentry);
815 /* Someone else set d_mounted? */
819 /* The dentry is not available as a mountpoint? */
824 /* Add the new mountpoint to the hash table */
825 read_seqlock_excl(&mount_lock);
826 new->m_dentry = dget(dentry);
828 hlist_add_head(&new->m_hash, mp_hash(dentry));
829 INIT_HLIST_HEAD(&new->m_list);
830 read_sequnlock_excl(&mount_lock);
840 * vfsmount lock must be held. Additionally, the caller is responsible
841 * for serializing calls for given disposal list.
843 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
845 if (!--mp->m_count) {
846 struct dentry *dentry = mp->m_dentry;
847 BUG_ON(!hlist_empty(&mp->m_list));
848 spin_lock(&dentry->d_lock);
849 dentry->d_flags &= ~DCACHE_MOUNTED;
850 spin_unlock(&dentry->d_lock);
851 dput_to_list(dentry, list);
852 hlist_del(&mp->m_hash);
857 /* called with namespace_lock and vfsmount lock */
858 static void put_mountpoint(struct mountpoint *mp)
860 __put_mountpoint(mp, &ex_mountpoints);
863 static inline int check_mnt(struct mount *mnt)
865 return mnt->mnt_ns == current->nsproxy->mnt_ns;
869 * vfsmount lock must be held for write
871 static void touch_mnt_namespace(struct mnt_namespace *ns)
875 wake_up_interruptible(&ns->poll);
880 * vfsmount lock must be held for write
882 static void __touch_mnt_namespace(struct mnt_namespace *ns)
884 if (ns && ns->event != event) {
886 wake_up_interruptible(&ns->poll);
891 * vfsmount lock must be held for write
893 static struct mountpoint *unhash_mnt(struct mount *mnt)
895 struct mountpoint *mp;
896 mnt->mnt_parent = mnt;
897 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
898 list_del_init(&mnt->mnt_child);
899 hlist_del_init_rcu(&mnt->mnt_hash);
900 hlist_del_init(&mnt->mnt_mp_list);
907 * vfsmount lock must be held for write
909 static void umount_mnt(struct mount *mnt)
911 put_mountpoint(unhash_mnt(mnt));
915 * vfsmount lock must be held for write
917 void mnt_set_mountpoint(struct mount *mnt,
918 struct mountpoint *mp,
919 struct mount *child_mnt)
922 mnt_add_count(mnt, 1); /* essentially, that's mntget */
923 child_mnt->mnt_mountpoint = mp->m_dentry;
924 child_mnt->mnt_parent = mnt;
925 child_mnt->mnt_mp = mp;
926 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
929 static void __attach_mnt(struct mount *mnt, struct mount *parent)
931 hlist_add_head_rcu(&mnt->mnt_hash,
932 m_hash(&parent->mnt, mnt->mnt_mountpoint));
933 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
937 * vfsmount lock must be held for write
939 static void attach_mnt(struct mount *mnt,
940 struct mount *parent,
941 struct mountpoint *mp)
943 mnt_set_mountpoint(parent, mp, mnt);
944 __attach_mnt(mnt, parent);
947 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
949 struct mountpoint *old_mp = mnt->mnt_mp;
950 struct mount *old_parent = mnt->mnt_parent;
952 list_del_init(&mnt->mnt_child);
953 hlist_del_init(&mnt->mnt_mp_list);
954 hlist_del_init_rcu(&mnt->mnt_hash);
956 attach_mnt(mnt, parent, mp);
958 put_mountpoint(old_mp);
959 mnt_add_count(old_parent, -1);
963 * vfsmount lock must be held for write
965 static void commit_tree(struct mount *mnt)
967 struct mount *parent = mnt->mnt_parent;
970 struct mnt_namespace *n = parent->mnt_ns;
972 BUG_ON(parent == mnt);
974 list_add_tail(&head, &mnt->mnt_list);
975 list_for_each_entry(m, &head, mnt_list)
978 list_splice(&head, n->list.prev);
980 n->mounts += n->pending_mounts;
981 n->pending_mounts = 0;
983 __attach_mnt(mnt, parent);
984 touch_mnt_namespace(n);
987 static struct mount *next_mnt(struct mount *p, struct mount *root)
989 struct list_head *next = p->mnt_mounts.next;
990 if (next == &p->mnt_mounts) {
994 next = p->mnt_child.next;
995 if (next != &p->mnt_parent->mnt_mounts)
1000 return list_entry(next, struct mount, mnt_child);
1003 static struct mount *skip_mnt_tree(struct mount *p)
1005 struct list_head *prev = p->mnt_mounts.prev;
1006 while (prev != &p->mnt_mounts) {
1007 p = list_entry(prev, struct mount, mnt_child);
1008 prev = p->mnt_mounts.prev;
1014 * vfs_create_mount - Create a mount for a configured superblock
1015 * @fc: The configuration context with the superblock attached
1017 * Create a mount to an already configured superblock. If necessary, the
1018 * caller should invoke vfs_get_tree() before calling this.
1020 * Note that this does not attach the mount to anything.
1022 struct vfsmount *vfs_create_mount(struct fs_context *fc)
1027 return ERR_PTR(-EINVAL);
1029 mnt = alloc_vfsmnt(fc->source ?: "none");
1031 return ERR_PTR(-ENOMEM);
1033 if (fc->sb_flags & SB_KERNMOUNT)
1034 mnt->mnt.mnt_flags = MNT_INTERNAL;
1036 atomic_inc(&fc->root->d_sb->s_active);
1037 mnt->mnt.mnt_sb = fc->root->d_sb;
1038 mnt->mnt.mnt_root = dget(fc->root);
1039 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1040 mnt->mnt_parent = mnt;
1043 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1044 unlock_mount_hash();
1047 EXPORT_SYMBOL(vfs_create_mount);
1049 struct vfsmount *fc_mount(struct fs_context *fc)
1051 int err = vfs_get_tree(fc);
1053 up_write(&fc->root->d_sb->s_umount);
1054 return vfs_create_mount(fc);
1056 return ERR_PTR(err);
1058 EXPORT_SYMBOL(fc_mount);
1060 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1061 int flags, const char *name,
1064 struct fs_context *fc;
1065 struct vfsmount *mnt;
1069 return ERR_PTR(-EINVAL);
1071 fc = fs_context_for_mount(type, flags);
1073 return ERR_CAST(fc);
1076 ret = vfs_parse_fs_string(fc, "source",
1077 name, strlen(name));
1079 ret = parse_monolithic_mount_data(fc, data);
1088 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1091 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1092 const char *name, void *data)
1094 /* Until it is worked out how to pass the user namespace
1095 * through from the parent mount to the submount don't support
1096 * unprivileged mounts with submounts.
1098 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1099 return ERR_PTR(-EPERM);
1101 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1103 EXPORT_SYMBOL_GPL(vfs_submount);
1105 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1108 struct super_block *sb = old->mnt.mnt_sb;
1112 mnt = alloc_vfsmnt(old->mnt_devname);
1114 return ERR_PTR(-ENOMEM);
1116 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1117 mnt->mnt_group_id = 0; /* not a peer of original */
1119 mnt->mnt_group_id = old->mnt_group_id;
1121 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1122 err = mnt_alloc_group_id(mnt);
1127 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1128 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1130 atomic_inc(&sb->s_active);
1131 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1133 mnt->mnt.mnt_sb = sb;
1134 mnt->mnt.mnt_root = dget(root);
1135 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1136 mnt->mnt_parent = mnt;
1138 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1139 unlock_mount_hash();
1141 if ((flag & CL_SLAVE) ||
1142 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1143 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1144 mnt->mnt_master = old;
1145 CLEAR_MNT_SHARED(mnt);
1146 } else if (!(flag & CL_PRIVATE)) {
1147 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1148 list_add(&mnt->mnt_share, &old->mnt_share);
1149 if (IS_MNT_SLAVE(old))
1150 list_add(&mnt->mnt_slave, &old->mnt_slave);
1151 mnt->mnt_master = old->mnt_master;
1153 CLEAR_MNT_SHARED(mnt);
1155 if (flag & CL_MAKE_SHARED)
1156 set_mnt_shared(mnt);
1158 /* stick the duplicate mount on the same expiry list
1159 * as the original if that was on one */
1160 if (flag & CL_EXPIRE) {
1161 if (!list_empty(&old->mnt_expire))
1162 list_add(&mnt->mnt_expire, &old->mnt_expire);
1170 return ERR_PTR(err);
1173 static void cleanup_mnt(struct mount *mnt)
1175 struct hlist_node *p;
1178 * The warning here probably indicates that somebody messed
1179 * up a mnt_want/drop_write() pair. If this happens, the
1180 * filesystem was probably unable to make r/w->r/o transitions.
1181 * The locking used to deal with mnt_count decrement provides barriers,
1182 * so mnt_get_writers() below is safe.
1184 WARN_ON(mnt_get_writers(mnt));
1185 if (unlikely(mnt->mnt_pins.first))
1187 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1188 hlist_del(&m->mnt_umount);
1191 fsnotify_vfsmount_delete(&mnt->mnt);
1192 dput(mnt->mnt.mnt_root);
1193 deactivate_super(mnt->mnt.mnt_sb);
1195 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1198 static void __cleanup_mnt(struct rcu_head *head)
1200 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1203 static LLIST_HEAD(delayed_mntput_list);
1204 static void delayed_mntput(struct work_struct *unused)
1206 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1207 struct mount *m, *t;
1209 llist_for_each_entry_safe(m, t, node, mnt_llist)
1212 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1214 static void mntput_no_expire(struct mount *mnt)
1220 if (likely(READ_ONCE(mnt->mnt_ns))) {
1222 * Since we don't do lock_mount_hash() here,
1223 * ->mnt_ns can change under us. However, if it's
1224 * non-NULL, then there's a reference that won't
1225 * be dropped until after an RCU delay done after
1226 * turning ->mnt_ns NULL. So if we observe it
1227 * non-NULL under rcu_read_lock(), the reference
1228 * we are dropping is not the final one.
1230 mnt_add_count(mnt, -1);
1236 * make sure that if __legitimize_mnt() has not seen us grab
1237 * mount_lock, we'll see their refcount increment here.
1240 mnt_add_count(mnt, -1);
1241 count = mnt_get_count(mnt);
1245 unlock_mount_hash();
1248 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1250 unlock_mount_hash();
1253 mnt->mnt.mnt_flags |= MNT_DOOMED;
1256 list_del(&mnt->mnt_instance);
1258 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1259 struct mount *p, *tmp;
1260 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1261 __put_mountpoint(unhash_mnt(p), &list);
1262 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1265 unlock_mount_hash();
1266 shrink_dentry_list(&list);
1268 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1269 struct task_struct *task = current;
1270 if (likely(!(task->flags & PF_KTHREAD))) {
1271 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1272 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1275 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1276 schedule_delayed_work(&delayed_mntput_work, 1);
1282 void mntput(struct vfsmount *mnt)
1285 struct mount *m = real_mount(mnt);
1286 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1287 if (unlikely(m->mnt_expiry_mark))
1288 m->mnt_expiry_mark = 0;
1289 mntput_no_expire(m);
1292 EXPORT_SYMBOL(mntput);
1294 struct vfsmount *mntget(struct vfsmount *mnt)
1297 mnt_add_count(real_mount(mnt), 1);
1300 EXPORT_SYMBOL(mntget);
1303 * Make a mount point inaccessible to new lookups.
1304 * Because there may still be current users, the caller MUST WAIT
1305 * for an RCU grace period before destroying the mount point.
1307 void mnt_make_shortterm(struct vfsmount *mnt)
1310 real_mount(mnt)->mnt_ns = NULL;
1314 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1315 * @path: path to check
1317 * d_mountpoint() can only be used reliably to establish if a dentry is
1318 * not mounted in any namespace and that common case is handled inline.
1319 * d_mountpoint() isn't aware of the possibility there may be multiple
1320 * mounts using a given dentry in a different namespace. This function
1321 * checks if the passed in path is a mountpoint rather than the dentry
1324 bool path_is_mountpoint(const struct path *path)
1329 if (!d_mountpoint(path->dentry))
1334 seq = read_seqbegin(&mount_lock);
1335 res = __path_is_mountpoint(path);
1336 } while (read_seqretry(&mount_lock, seq));
1341 EXPORT_SYMBOL(path_is_mountpoint);
1343 struct vfsmount *mnt_clone_internal(const struct path *path)
1346 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1349 p->mnt.mnt_flags |= MNT_INTERNAL;
1353 #ifdef CONFIG_PROC_FS
1354 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1355 struct list_head *p)
1357 struct mount *mnt, *ret = NULL;
1360 list_for_each_continue(p, &ns->list) {
1361 mnt = list_entry(p, typeof(*mnt), mnt_list);
1362 if (!mnt_is_cursor(mnt)) {
1372 /* iterator; we want it to have access to namespace_sem, thus here... */
1373 static void *m_start(struct seq_file *m, loff_t *pos)
1375 struct proc_mounts *p = m->private;
1376 struct list_head *prev;
1378 down_read(&namespace_sem);
1380 prev = &p->ns->list;
1382 prev = &p->cursor.mnt_list;
1384 /* Read after we'd reached the end? */
1385 if (list_empty(prev))
1389 return mnt_list_next(p->ns, prev);
1392 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1394 struct proc_mounts *p = m->private;
1395 struct mount *mnt = v;
1398 return mnt_list_next(p->ns, &mnt->mnt_list);
1401 static void m_stop(struct seq_file *m, void *v)
1403 struct proc_mounts *p = m->private;
1404 struct mount *mnt = v;
1406 lock_ns_list(p->ns);
1408 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1410 list_del_init(&p->cursor.mnt_list);
1411 unlock_ns_list(p->ns);
1412 up_read(&namespace_sem);
1415 static int m_show(struct seq_file *m, void *v)
1417 struct proc_mounts *p = m->private;
1418 struct mount *r = v;
1419 return p->show(m, &r->mnt);
1422 const struct seq_operations mounts_op = {
1429 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1431 down_read(&namespace_sem);
1433 list_del(&cursor->mnt_list);
1435 up_read(&namespace_sem);
1437 #endif /* CONFIG_PROC_FS */
1440 * may_umount_tree - check if a mount tree is busy
1441 * @m: root of mount tree
1443 * This is called to check if a tree of mounts has any
1444 * open files, pwds, chroots or sub mounts that are
1447 int may_umount_tree(struct vfsmount *m)
1449 struct mount *mnt = real_mount(m);
1450 int actual_refs = 0;
1451 int minimum_refs = 0;
1455 /* write lock needed for mnt_get_count */
1457 for (p = mnt; p; p = next_mnt(p, mnt)) {
1458 actual_refs += mnt_get_count(p);
1461 unlock_mount_hash();
1463 if (actual_refs > minimum_refs)
1469 EXPORT_SYMBOL(may_umount_tree);
1472 * may_umount - check if a mount point is busy
1473 * @mnt: root of mount
1475 * This is called to check if a mount point has any
1476 * open files, pwds, chroots or sub mounts. If the
1477 * mount has sub mounts this will return busy
1478 * regardless of whether the sub mounts are busy.
1480 * Doesn't take quota and stuff into account. IOW, in some cases it will
1481 * give false negatives. The main reason why it's here is that we need
1482 * a non-destructive way to look for easily umountable filesystems.
1484 int may_umount(struct vfsmount *mnt)
1487 down_read(&namespace_sem);
1489 if (propagate_mount_busy(real_mount(mnt), 2))
1491 unlock_mount_hash();
1492 up_read(&namespace_sem);
1496 EXPORT_SYMBOL(may_umount);
1498 static void namespace_unlock(void)
1500 struct hlist_head head;
1501 struct hlist_node *p;
1505 hlist_move_list(&unmounted, &head);
1506 list_splice_init(&ex_mountpoints, &list);
1508 up_write(&namespace_sem);
1510 shrink_dentry_list(&list);
1512 if (likely(hlist_empty(&head)))
1515 synchronize_rcu_expedited();
1517 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1518 hlist_del(&m->mnt_umount);
1523 static inline void namespace_lock(void)
1525 down_write(&namespace_sem);
1528 enum umount_tree_flags {
1530 UMOUNT_PROPAGATE = 2,
1531 UMOUNT_CONNECTED = 4,
1534 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1536 /* Leaving mounts connected is only valid for lazy umounts */
1537 if (how & UMOUNT_SYNC)
1540 /* A mount without a parent has nothing to be connected to */
1541 if (!mnt_has_parent(mnt))
1544 /* Because the reference counting rules change when mounts are
1545 * unmounted and connected, umounted mounts may not be
1546 * connected to mounted mounts.
1548 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1551 /* Has it been requested that the mount remain connected? */
1552 if (how & UMOUNT_CONNECTED)
1555 /* Is the mount locked such that it needs to remain connected? */
1556 if (IS_MNT_LOCKED(mnt))
1559 /* By default disconnect the mount */
1564 * mount_lock must be held
1565 * namespace_sem must be held for write
1567 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1569 LIST_HEAD(tmp_list);
1572 if (how & UMOUNT_PROPAGATE)
1573 propagate_mount_unlock(mnt);
1575 /* Gather the mounts to umount */
1576 for (p = mnt; p; p = next_mnt(p, mnt)) {
1577 p->mnt.mnt_flags |= MNT_UMOUNT;
1578 list_move(&p->mnt_list, &tmp_list);
1581 /* Hide the mounts from mnt_mounts */
1582 list_for_each_entry(p, &tmp_list, mnt_list) {
1583 list_del_init(&p->mnt_child);
1586 /* Add propogated mounts to the tmp_list */
1587 if (how & UMOUNT_PROPAGATE)
1588 propagate_umount(&tmp_list);
1590 while (!list_empty(&tmp_list)) {
1591 struct mnt_namespace *ns;
1593 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1594 list_del_init(&p->mnt_expire);
1595 list_del_init(&p->mnt_list);
1599 __touch_mnt_namespace(ns);
1602 if (how & UMOUNT_SYNC)
1603 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1605 disconnect = disconnect_mount(p, how);
1606 if (mnt_has_parent(p)) {
1607 mnt_add_count(p->mnt_parent, -1);
1609 /* Don't forget about p */
1610 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1615 change_mnt_propagation(p, MS_PRIVATE);
1617 hlist_add_head(&p->mnt_umount, &unmounted);
1621 static void shrink_submounts(struct mount *mnt);
1623 static int do_umount_root(struct super_block *sb)
1627 down_write(&sb->s_umount);
1628 if (!sb_rdonly(sb)) {
1629 struct fs_context *fc;
1631 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1636 ret = parse_monolithic_mount_data(fc, NULL);
1638 ret = reconfigure_super(fc);
1642 up_write(&sb->s_umount);
1646 static int do_umount(struct mount *mnt, int flags)
1648 struct super_block *sb = mnt->mnt.mnt_sb;
1651 retval = security_sb_umount(&mnt->mnt, flags);
1656 * Allow userspace to request a mountpoint be expired rather than
1657 * unmounting unconditionally. Unmount only happens if:
1658 * (1) the mark is already set (the mark is cleared by mntput())
1659 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1661 if (flags & MNT_EXPIRE) {
1662 if (&mnt->mnt == current->fs->root.mnt ||
1663 flags & (MNT_FORCE | MNT_DETACH))
1667 * probably don't strictly need the lock here if we examined
1668 * all race cases, but it's a slowpath.
1671 if (mnt_get_count(mnt) != 2) {
1672 unlock_mount_hash();
1675 unlock_mount_hash();
1677 if (!xchg(&mnt->mnt_expiry_mark, 1))
1682 * If we may have to abort operations to get out of this
1683 * mount, and they will themselves hold resources we must
1684 * allow the fs to do things. In the Unix tradition of
1685 * 'Gee thats tricky lets do it in userspace' the umount_begin
1686 * might fail to complete on the first run through as other tasks
1687 * must return, and the like. Thats for the mount program to worry
1688 * about for the moment.
1691 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1692 sb->s_op->umount_begin(sb);
1696 * No sense to grab the lock for this test, but test itself looks
1697 * somewhat bogus. Suggestions for better replacement?
1698 * Ho-hum... In principle, we might treat that as umount + switch
1699 * to rootfs. GC would eventually take care of the old vfsmount.
1700 * Actually it makes sense, especially if rootfs would contain a
1701 * /reboot - static binary that would close all descriptors and
1702 * call reboot(9). Then init(8) could umount root and exec /reboot.
1704 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1706 * Special case for "unmounting" root ...
1707 * we just try to remount it readonly.
1709 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1711 return do_umount_root(sb);
1717 /* Recheck MNT_LOCKED with the locks held */
1719 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1723 if (flags & MNT_DETACH) {
1724 if (!list_empty(&mnt->mnt_list))
1725 umount_tree(mnt, UMOUNT_PROPAGATE);
1728 shrink_submounts(mnt);
1730 if (!propagate_mount_busy(mnt, 2)) {
1731 if (!list_empty(&mnt->mnt_list))
1732 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1737 unlock_mount_hash();
1743 * __detach_mounts - lazily unmount all mounts on the specified dentry
1745 * During unlink, rmdir, and d_drop it is possible to loose the path
1746 * to an existing mountpoint, and wind up leaking the mount.
1747 * detach_mounts allows lazily unmounting those mounts instead of
1750 * The caller may hold dentry->d_inode->i_mutex.
1752 void __detach_mounts(struct dentry *dentry)
1754 struct mountpoint *mp;
1759 mp = lookup_mountpoint(dentry);
1764 while (!hlist_empty(&mp->m_list)) {
1765 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1766 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1768 hlist_add_head(&mnt->mnt_umount, &unmounted);
1770 else umount_tree(mnt, UMOUNT_CONNECTED);
1774 unlock_mount_hash();
1779 * Is the caller allowed to modify his namespace?
1781 bool may_mount(void)
1783 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1787 * path_mounted - check whether path is mounted
1788 * @path: path to check
1790 * Determine whether @path refers to the root of a mount.
1792 * Return: true if @path is the root of a mount, false if not.
1794 static inline bool path_mounted(const struct path *path)
1796 return path->mnt->mnt_root == path->dentry;
1799 static void warn_mandlock(void)
1801 pr_warn_once("=======================================================\n"
1802 "WARNING: The mand mount option has been deprecated and\n"
1803 " and is ignored by this kernel. Remove the mand\n"
1804 " option from the mount to silence this warning.\n"
1805 "=======================================================\n");
1808 static int can_umount(const struct path *path, int flags)
1810 struct mount *mnt = real_mount(path->mnt);
1814 if (!path_mounted(path))
1816 if (!check_mnt(mnt))
1818 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1820 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1825 // caller is responsible for flags being sane
1826 int path_umount(struct path *path, int flags)
1828 struct mount *mnt = real_mount(path->mnt);
1831 ret = can_umount(path, flags);
1833 ret = do_umount(mnt, flags);
1835 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1837 mntput_no_expire(mnt);
1841 static int ksys_umount(char __user *name, int flags)
1843 int lookup_flags = LOOKUP_MOUNTPOINT;
1847 // basic validity checks done first
1848 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1851 if (!(flags & UMOUNT_NOFOLLOW))
1852 lookup_flags |= LOOKUP_FOLLOW;
1853 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1856 return path_umount(&path, flags);
1859 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1861 return ksys_umount(name, flags);
1864 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1867 * The 2.0 compatible umount. No flags.
1869 SYSCALL_DEFINE1(oldumount, char __user *, name)
1871 return ksys_umount(name, 0);
1876 static bool is_mnt_ns_file(struct dentry *dentry)
1878 /* Is this a proxy for a mount namespace? */
1879 return dentry->d_op == &ns_dentry_operations &&
1880 dentry->d_fsdata == &mntns_operations;
1883 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1885 return container_of(ns, struct mnt_namespace, ns);
1888 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1893 static bool mnt_ns_loop(struct dentry *dentry)
1895 /* Could bind mounting the mount namespace inode cause a
1896 * mount namespace loop?
1898 struct mnt_namespace *mnt_ns;
1899 if (!is_mnt_ns_file(dentry))
1902 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1903 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1906 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1909 struct mount *res, *p, *q, *r, *parent;
1911 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1912 return ERR_PTR(-EINVAL);
1914 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1915 return ERR_PTR(-EINVAL);
1917 res = q = clone_mnt(mnt, dentry, flag);
1921 q->mnt_mountpoint = mnt->mnt_mountpoint;
1924 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1926 if (!is_subdir(r->mnt_mountpoint, dentry))
1929 for (s = r; s; s = next_mnt(s, r)) {
1930 if (!(flag & CL_COPY_UNBINDABLE) &&
1931 IS_MNT_UNBINDABLE(s)) {
1932 if (s->mnt.mnt_flags & MNT_LOCKED) {
1933 /* Both unbindable and locked. */
1934 q = ERR_PTR(-EPERM);
1937 s = skip_mnt_tree(s);
1941 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1942 is_mnt_ns_file(s->mnt.mnt_root)) {
1943 s = skip_mnt_tree(s);
1946 while (p != s->mnt_parent) {
1952 q = clone_mnt(p, p->mnt.mnt_root, flag);
1956 list_add_tail(&q->mnt_list, &res->mnt_list);
1957 attach_mnt(q, parent, p->mnt_mp);
1958 unlock_mount_hash();
1965 umount_tree(res, UMOUNT_SYNC);
1966 unlock_mount_hash();
1971 /* Caller should check returned pointer for errors */
1973 struct vfsmount *collect_mounts(const struct path *path)
1977 if (!check_mnt(real_mount(path->mnt)))
1978 tree = ERR_PTR(-EINVAL);
1980 tree = copy_tree(real_mount(path->mnt), path->dentry,
1981 CL_COPY_ALL | CL_PRIVATE);
1984 return ERR_CAST(tree);
1988 static void free_mnt_ns(struct mnt_namespace *);
1989 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1991 void dissolve_on_fput(struct vfsmount *mnt)
1993 struct mnt_namespace *ns;
1996 ns = real_mount(mnt)->mnt_ns;
1999 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2003 unlock_mount_hash();
2009 void drop_collected_mounts(struct vfsmount *mnt)
2013 umount_tree(real_mount(mnt), 0);
2014 unlock_mount_hash();
2018 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2020 struct mount *child;
2022 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2023 if (!is_subdir(child->mnt_mountpoint, dentry))
2026 if (child->mnt.mnt_flags & MNT_LOCKED)
2033 * clone_private_mount - create a private clone of a path
2034 * @path: path to clone
2036 * This creates a new vfsmount, which will be the clone of @path. The new mount
2037 * will not be attached anywhere in the namespace and will be private (i.e.
2038 * changes to the originating mount won't be propagated into this).
2040 * Release with mntput().
2042 struct vfsmount *clone_private_mount(const struct path *path)
2044 struct mount *old_mnt = real_mount(path->mnt);
2045 struct mount *new_mnt;
2047 down_read(&namespace_sem);
2048 if (IS_MNT_UNBINDABLE(old_mnt))
2051 if (!check_mnt(old_mnt))
2054 if (has_locked_children(old_mnt, path->dentry))
2057 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2058 up_read(&namespace_sem);
2060 if (IS_ERR(new_mnt))
2061 return ERR_CAST(new_mnt);
2063 /* Longterm mount to be removed by kern_unmount*() */
2064 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2066 return &new_mnt->mnt;
2069 up_read(&namespace_sem);
2070 return ERR_PTR(-EINVAL);
2072 EXPORT_SYMBOL_GPL(clone_private_mount);
2074 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2075 struct vfsmount *root)
2078 int res = f(root, arg);
2081 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2082 res = f(&mnt->mnt, arg);
2089 static void lock_mnt_tree(struct mount *mnt)
2093 for (p = mnt; p; p = next_mnt(p, mnt)) {
2094 int flags = p->mnt.mnt_flags;
2095 /* Don't allow unprivileged users to change mount flags */
2096 flags |= MNT_LOCK_ATIME;
2098 if (flags & MNT_READONLY)
2099 flags |= MNT_LOCK_READONLY;
2101 if (flags & MNT_NODEV)
2102 flags |= MNT_LOCK_NODEV;
2104 if (flags & MNT_NOSUID)
2105 flags |= MNT_LOCK_NOSUID;
2107 if (flags & MNT_NOEXEC)
2108 flags |= MNT_LOCK_NOEXEC;
2109 /* Don't allow unprivileged users to reveal what is under a mount */
2110 if (list_empty(&p->mnt_expire))
2111 flags |= MNT_LOCKED;
2112 p->mnt.mnt_flags = flags;
2116 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2120 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2121 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2122 mnt_release_group_id(p);
2126 static int invent_group_ids(struct mount *mnt, bool recurse)
2130 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2131 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2132 int err = mnt_alloc_group_id(p);
2134 cleanup_group_ids(mnt, p);
2143 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2145 unsigned int max = READ_ONCE(sysctl_mount_max);
2146 unsigned int mounts = 0;
2149 if (ns->mounts >= max)
2152 if (ns->pending_mounts >= max)
2154 max -= ns->pending_mounts;
2156 for (p = mnt; p; p = next_mnt(p, mnt))
2162 ns->pending_mounts += mounts;
2167 * @source_mnt : mount tree to be attached
2168 * @nd : place the mount tree @source_mnt is attached
2169 * @parent_nd : if non-null, detach the source_mnt from its parent and
2170 * store the parent mount and mountpoint dentry.
2171 * (done when source_mnt is moved)
2173 * NOTE: in the table below explains the semantics when a source mount
2174 * of a given type is attached to a destination mount of a given type.
2175 * ---------------------------------------------------------------------------
2176 * | BIND MOUNT OPERATION |
2177 * |**************************************************************************
2178 * | source-->| shared | private | slave | unbindable |
2182 * |**************************************************************************
2183 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2185 * |non-shared| shared (+) | private | slave (*) | invalid |
2186 * ***************************************************************************
2187 * A bind operation clones the source mount and mounts the clone on the
2188 * destination mount.
2190 * (++) the cloned mount is propagated to all the mounts in the propagation
2191 * tree of the destination mount and the cloned mount is added to
2192 * the peer group of the source mount.
2193 * (+) the cloned mount is created under the destination mount and is marked
2194 * as shared. The cloned mount is added to the peer group of the source
2196 * (+++) the mount is propagated to all the mounts in the propagation tree
2197 * of the destination mount and the cloned mount is made slave
2198 * of the same master as that of the source mount. The cloned mount
2199 * is marked as 'shared and slave'.
2200 * (*) the cloned mount is made a slave of the same master as that of the
2203 * ---------------------------------------------------------------------------
2204 * | MOVE MOUNT OPERATION |
2205 * |**************************************************************************
2206 * | source-->| shared | private | slave | unbindable |
2210 * |**************************************************************************
2211 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2213 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2214 * ***************************************************************************
2216 * (+) the mount is moved to the destination. And is then propagated to
2217 * all the mounts in the propagation tree of the destination mount.
2218 * (+*) the mount is moved to the destination.
2219 * (+++) the mount is moved to the destination and is then propagated to
2220 * all the mounts belonging to the destination mount's propagation tree.
2221 * the mount is marked as 'shared and slave'.
2222 * (*) the mount continues to be a slave at the new location.
2224 * if the source mount is a tree, the operations explained above is
2225 * applied to each mount in the tree.
2226 * Must be called without spinlocks held, since this function can sleep
2229 static int attach_recursive_mnt(struct mount *source_mnt,
2230 struct mount *dest_mnt,
2231 struct mountpoint *dest_mp,
2234 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2235 HLIST_HEAD(tree_list);
2236 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2237 struct mountpoint *smp;
2238 struct mount *child, *p;
2239 struct hlist_node *n;
2242 /* Preallocate a mountpoint in case the new mounts need
2243 * to be tucked under other mounts.
2245 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2247 return PTR_ERR(smp);
2249 /* Is there space to add these mounts to the mount namespace? */
2251 err = count_mounts(ns, source_mnt);
2256 if (IS_MNT_SHARED(dest_mnt)) {
2257 err = invent_group_ids(source_mnt, true);
2260 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2263 goto out_cleanup_ids;
2264 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2270 unhash_mnt(source_mnt);
2271 attach_mnt(source_mnt, dest_mnt, dest_mp);
2272 touch_mnt_namespace(source_mnt->mnt_ns);
2274 if (source_mnt->mnt_ns) {
2275 /* move from anon - the caller will destroy */
2276 list_del_init(&source_mnt->mnt_ns->list);
2278 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2279 commit_tree(source_mnt);
2282 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2284 hlist_del_init(&child->mnt_hash);
2285 q = __lookup_mnt(&child->mnt_parent->mnt,
2286 child->mnt_mountpoint);
2288 mnt_change_mountpoint(child, smp, q);
2289 /* Notice when we are propagating across user namespaces */
2290 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2291 lock_mnt_tree(child);
2292 child->mnt.mnt_flags &= ~MNT_LOCKED;
2295 put_mountpoint(smp);
2296 unlock_mount_hash();
2301 while (!hlist_empty(&tree_list)) {
2302 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2303 child->mnt_parent->mnt_ns->pending_mounts = 0;
2304 umount_tree(child, UMOUNT_SYNC);
2306 unlock_mount_hash();
2307 cleanup_group_ids(source_mnt, NULL);
2309 ns->pending_mounts = 0;
2311 read_seqlock_excl(&mount_lock);
2312 put_mountpoint(smp);
2313 read_sequnlock_excl(&mount_lock);
2318 static struct mountpoint *lock_mount(struct path *path)
2320 struct vfsmount *mnt;
2321 struct dentry *dentry = path->dentry;
2323 inode_lock(dentry->d_inode);
2324 if (unlikely(cant_mount(dentry))) {
2325 inode_unlock(dentry->d_inode);
2326 return ERR_PTR(-ENOENT);
2329 mnt = lookup_mnt(path);
2331 struct mountpoint *mp = get_mountpoint(dentry);
2334 inode_unlock(dentry->d_inode);
2340 inode_unlock(path->dentry->d_inode);
2343 dentry = path->dentry = dget(mnt->mnt_root);
2347 static void unlock_mount(struct mountpoint *where)
2349 struct dentry *dentry = where->m_dentry;
2351 read_seqlock_excl(&mount_lock);
2352 put_mountpoint(where);
2353 read_sequnlock_excl(&mount_lock);
2356 inode_unlock(dentry->d_inode);
2359 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2361 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2364 if (d_is_dir(mp->m_dentry) !=
2365 d_is_dir(mnt->mnt.mnt_root))
2368 return attach_recursive_mnt(mnt, p, mp, false);
2372 * Sanity check the flags to change_mnt_propagation.
2375 static int flags_to_propagation_type(int ms_flags)
2377 int type = ms_flags & ~(MS_REC | MS_SILENT);
2379 /* Fail if any non-propagation flags are set */
2380 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2382 /* Only one propagation flag should be set */
2383 if (!is_power_of_2(type))
2389 * recursively change the type of the mountpoint.
2391 static int do_change_type(struct path *path, int ms_flags)
2394 struct mount *mnt = real_mount(path->mnt);
2395 int recurse = ms_flags & MS_REC;
2399 if (!path_mounted(path))
2402 type = flags_to_propagation_type(ms_flags);
2407 if (type == MS_SHARED) {
2408 err = invent_group_ids(mnt, recurse);
2414 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2415 change_mnt_propagation(m, type);
2416 unlock_mount_hash();
2423 static struct mount *__do_loopback(struct path *old_path, int recurse)
2425 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2427 if (IS_MNT_UNBINDABLE(old))
2430 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2433 if (!recurse && has_locked_children(old, old_path->dentry))
2437 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2439 mnt = clone_mnt(old, old_path->dentry, 0);
2442 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2448 * do loopback mount.
2450 static int do_loopback(struct path *path, const char *old_name,
2453 struct path old_path;
2454 struct mount *mnt = NULL, *parent;
2455 struct mountpoint *mp;
2457 if (!old_name || !*old_name)
2459 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2464 if (mnt_ns_loop(old_path.dentry))
2467 mp = lock_mount(path);
2473 parent = real_mount(path->mnt);
2474 if (!check_mnt(parent))
2477 mnt = __do_loopback(&old_path, recurse);
2483 err = graft_tree(mnt, parent, mp);
2486 umount_tree(mnt, UMOUNT_SYNC);
2487 unlock_mount_hash();
2492 path_put(&old_path);
2496 static struct file *open_detached_copy(struct path *path, bool recursive)
2498 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2499 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2500 struct mount *mnt, *p;
2504 return ERR_CAST(ns);
2507 mnt = __do_loopback(path, recursive);
2511 return ERR_CAST(mnt);
2515 for (p = mnt; p; p = next_mnt(p, mnt)) {
2520 list_add_tail(&ns->list, &mnt->mnt_list);
2522 unlock_mount_hash();
2526 path->mnt = &mnt->mnt;
2527 file = dentry_open(path, O_PATH, current_cred());
2529 dissolve_on_fput(path->mnt);
2531 file->f_mode |= FMODE_NEED_UNMOUNT;
2535 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2539 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2540 bool detached = flags & OPEN_TREE_CLONE;
2544 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2546 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2547 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2551 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2554 if (flags & AT_NO_AUTOMOUNT)
2555 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2556 if (flags & AT_SYMLINK_NOFOLLOW)
2557 lookup_flags &= ~LOOKUP_FOLLOW;
2558 if (flags & AT_EMPTY_PATH)
2559 lookup_flags |= LOOKUP_EMPTY;
2561 if (detached && !may_mount())
2564 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2568 error = user_path_at(dfd, filename, lookup_flags, &path);
2569 if (unlikely(error)) {
2570 file = ERR_PTR(error);
2573 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2575 file = dentry_open(&path, O_PATH, current_cred());
2580 return PTR_ERR(file);
2582 fd_install(fd, file);
2587 * Don't allow locked mount flags to be cleared.
2589 * No locks need to be held here while testing the various MNT_LOCK
2590 * flags because those flags can never be cleared once they are set.
2592 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2594 unsigned int fl = mnt->mnt.mnt_flags;
2596 if ((fl & MNT_LOCK_READONLY) &&
2597 !(mnt_flags & MNT_READONLY))
2600 if ((fl & MNT_LOCK_NODEV) &&
2601 !(mnt_flags & MNT_NODEV))
2604 if ((fl & MNT_LOCK_NOSUID) &&
2605 !(mnt_flags & MNT_NOSUID))
2608 if ((fl & MNT_LOCK_NOEXEC) &&
2609 !(mnt_flags & MNT_NOEXEC))
2612 if ((fl & MNT_LOCK_ATIME) &&
2613 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2619 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2621 bool readonly_request = (mnt_flags & MNT_READONLY);
2623 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2626 if (readonly_request)
2627 return mnt_make_readonly(mnt);
2629 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2633 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2635 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2636 mnt->mnt.mnt_flags = mnt_flags;
2637 touch_mnt_namespace(mnt->mnt_ns);
2640 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2642 struct super_block *sb = mnt->mnt_sb;
2644 if (!__mnt_is_readonly(mnt) &&
2645 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2646 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2647 char *buf = (char *)__get_free_page(GFP_KERNEL);
2648 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2650 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2652 is_mounted(mnt) ? "remounted" : "mounted",
2653 mntpath, &sb->s_time_max,
2654 (unsigned long long)sb->s_time_max);
2656 free_page((unsigned long)buf);
2657 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2662 * Handle reconfiguration of the mountpoint only without alteration of the
2663 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2666 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2668 struct super_block *sb = path->mnt->mnt_sb;
2669 struct mount *mnt = real_mount(path->mnt);
2672 if (!check_mnt(mnt))
2675 if (!path_mounted(path))
2678 if (!can_change_locked_flags(mnt, mnt_flags))
2682 * We're only checking whether the superblock is read-only not
2683 * changing it, so only take down_read(&sb->s_umount).
2685 down_read(&sb->s_umount);
2687 ret = change_mount_ro_state(mnt, mnt_flags);
2689 set_mount_attributes(mnt, mnt_flags);
2690 unlock_mount_hash();
2691 up_read(&sb->s_umount);
2693 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2699 * change filesystem flags. dir should be a physical root of filesystem.
2700 * If you've mounted a non-root directory somewhere and want to do remount
2701 * on it - tough luck.
2703 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2704 int mnt_flags, void *data)
2707 struct super_block *sb = path->mnt->mnt_sb;
2708 struct mount *mnt = real_mount(path->mnt);
2709 struct fs_context *fc;
2711 if (!check_mnt(mnt))
2714 if (!path_mounted(path))
2717 if (!can_change_locked_flags(mnt, mnt_flags))
2720 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2725 err = parse_monolithic_mount_data(fc, data);
2727 down_write(&sb->s_umount);
2729 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2730 err = reconfigure_super(fc);
2733 set_mount_attributes(mnt, mnt_flags);
2734 unlock_mount_hash();
2737 up_write(&sb->s_umount);
2740 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2746 static inline int tree_contains_unbindable(struct mount *mnt)
2749 for (p = mnt; p; p = next_mnt(p, mnt)) {
2750 if (IS_MNT_UNBINDABLE(p))
2757 * Check that there aren't references to earlier/same mount namespaces in the
2758 * specified subtree. Such references can act as pins for mount namespaces
2759 * that aren't checked by the mount-cycle checking code, thereby allowing
2760 * cycles to be made.
2762 static bool check_for_nsfs_mounts(struct mount *subtree)
2768 for (p = subtree; p; p = next_mnt(p, subtree))
2769 if (mnt_ns_loop(p->mnt.mnt_root))
2774 unlock_mount_hash();
2778 static int do_set_group(struct path *from_path, struct path *to_path)
2780 struct mount *from, *to;
2783 from = real_mount(from_path->mnt);
2784 to = real_mount(to_path->mnt);
2789 /* To and From must be mounted */
2790 if (!is_mounted(&from->mnt))
2792 if (!is_mounted(&to->mnt))
2796 /* We should be allowed to modify mount namespaces of both mounts */
2797 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2799 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2803 /* To and From paths should be mount roots */
2804 if (!path_mounted(from_path))
2806 if (!path_mounted(to_path))
2809 /* Setting sharing groups is only allowed across same superblock */
2810 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2813 /* From mount root should be wider than To mount root */
2814 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2817 /* From mount should not have locked children in place of To's root */
2818 if (has_locked_children(from, to->mnt.mnt_root))
2821 /* Setting sharing groups is only allowed on private mounts */
2822 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2825 /* From should not be private */
2826 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2829 if (IS_MNT_SLAVE(from)) {
2830 struct mount *m = from->mnt_master;
2832 list_add(&to->mnt_slave, &m->mnt_slave_list);
2836 if (IS_MNT_SHARED(from)) {
2837 to->mnt_group_id = from->mnt_group_id;
2838 list_add(&to->mnt_share, &from->mnt_share);
2841 unlock_mount_hash();
2850 static int do_move_mount(struct path *old_path, struct path *new_path)
2852 struct mnt_namespace *ns;
2855 struct mount *parent;
2856 struct mountpoint *mp, *old_mp;
2860 mp = lock_mount(new_path);
2864 old = real_mount(old_path->mnt);
2865 p = real_mount(new_path->mnt);
2866 parent = old->mnt_parent;
2867 attached = mnt_has_parent(old);
2868 old_mp = old->mnt_mp;
2872 /* The mountpoint must be in our namespace. */
2876 /* The thing moved must be mounted... */
2877 if (!is_mounted(&old->mnt))
2880 /* ... and either ours or the root of anon namespace */
2881 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2884 if (old->mnt.mnt_flags & MNT_LOCKED)
2887 if (!path_mounted(old_path))
2890 if (d_is_dir(new_path->dentry) !=
2891 d_is_dir(old_path->dentry))
2894 * Don't move a mount residing in a shared parent.
2896 if (attached && IS_MNT_SHARED(parent))
2899 * Don't move a mount tree containing unbindable mounts to a destination
2900 * mount which is shared.
2902 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2905 if (!check_for_nsfs_mounts(old))
2907 for (; mnt_has_parent(p); p = p->mnt_parent)
2911 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2916 /* if the mount is moved, it should no longer be expire
2918 list_del_init(&old->mnt_expire);
2920 put_mountpoint(old_mp);
2925 mntput_no_expire(parent);
2932 static int do_move_mount_old(struct path *path, const char *old_name)
2934 struct path old_path;
2937 if (!old_name || !*old_name)
2940 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2944 err = do_move_mount(&old_path, path);
2945 path_put(&old_path);
2950 * add a mount into a namespace's mount tree
2952 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2953 const struct path *path, int mnt_flags)
2955 struct mount *parent = real_mount(path->mnt);
2957 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2959 if (unlikely(!check_mnt(parent))) {
2960 /* that's acceptable only for automounts done in private ns */
2961 if (!(mnt_flags & MNT_SHRINKABLE))
2963 /* ... and for those we'd better have mountpoint still alive */
2964 if (!parent->mnt_ns)
2968 /* Refuse the same filesystem on the same mount point */
2969 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
2972 if (d_is_symlink(newmnt->mnt.mnt_root))
2975 newmnt->mnt.mnt_flags = mnt_flags;
2976 return graft_tree(newmnt, parent, mp);
2979 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2982 * Create a new mount using a superblock configuration and request it
2983 * be added to the namespace tree.
2985 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2986 unsigned int mnt_flags)
2988 struct vfsmount *mnt;
2989 struct mountpoint *mp;
2990 struct super_block *sb = fc->root->d_sb;
2993 error = security_sb_kern_mount(sb);
2994 if (!error && mount_too_revealing(sb, &mnt_flags))
2997 if (unlikely(error)) {
3002 up_write(&sb->s_umount);
3004 mnt = vfs_create_mount(fc);
3006 return PTR_ERR(mnt);
3008 mnt_warn_timestamp_expiry(mountpoint, mnt);
3010 mp = lock_mount(mountpoint);
3015 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3023 * create a new mount for userspace and request it to be added into the
3026 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3027 int mnt_flags, const char *name, void *data)
3029 struct file_system_type *type;
3030 struct fs_context *fc;
3031 const char *subtype = NULL;
3037 type = get_fs_type(fstype);
3041 if (type->fs_flags & FS_HAS_SUBTYPE) {
3042 subtype = strchr(fstype, '.');
3046 put_filesystem(type);
3052 fc = fs_context_for_mount(type, sb_flags);
3053 put_filesystem(type);
3058 err = vfs_parse_fs_string(fc, "subtype",
3059 subtype, strlen(subtype));
3061 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3063 err = parse_monolithic_mount_data(fc, data);
3064 if (!err && !mount_capable(fc))
3067 err = vfs_get_tree(fc);
3069 err = do_new_mount_fc(fc, path, mnt_flags);
3075 int finish_automount(struct vfsmount *m, const struct path *path)
3077 struct dentry *dentry = path->dentry;
3078 struct mountpoint *mp;
3087 mnt = real_mount(m);
3088 /* The new mount record should have at least 2 refs to prevent it being
3089 * expired before we get a chance to add it
3091 BUG_ON(mnt_get_count(mnt) < 2);
3093 if (m->mnt_sb == path->mnt->mnt_sb &&
3094 m->mnt_root == dentry) {
3100 * we don't want to use lock_mount() - in this case finding something
3101 * that overmounts our mountpoint to be means "quitely drop what we've
3102 * got", not "try to mount it on top".
3104 inode_lock(dentry->d_inode);
3106 if (unlikely(cant_mount(dentry))) {
3108 goto discard_locked;
3111 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3114 goto discard_locked;
3117 mp = get_mountpoint(dentry);
3120 goto discard_locked;
3123 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3132 inode_unlock(dentry->d_inode);
3134 /* remove m from any expiration list it may be on */
3135 if (!list_empty(&mnt->mnt_expire)) {
3137 list_del_init(&mnt->mnt_expire);
3146 * mnt_set_expiry - Put a mount on an expiration list
3147 * @mnt: The mount to list.
3148 * @expiry_list: The list to add the mount to.
3150 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3154 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3158 EXPORT_SYMBOL(mnt_set_expiry);
3161 * process a list of expirable mountpoints with the intent of discarding any
3162 * mountpoints that aren't in use and haven't been touched since last we came
3165 void mark_mounts_for_expiry(struct list_head *mounts)
3167 struct mount *mnt, *next;
3168 LIST_HEAD(graveyard);
3170 if (list_empty(mounts))
3176 /* extract from the expiration list every vfsmount that matches the
3177 * following criteria:
3178 * - only referenced by its parent vfsmount
3179 * - still marked for expiry (marked on the last call here; marks are
3180 * cleared by mntput())
3182 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3183 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3184 propagate_mount_busy(mnt, 1))
3186 list_move(&mnt->mnt_expire, &graveyard);
3188 while (!list_empty(&graveyard)) {
3189 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3190 touch_mnt_namespace(mnt->mnt_ns);
3191 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3193 unlock_mount_hash();
3197 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3200 * Ripoff of 'select_parent()'
3202 * search the list of submounts for a given mountpoint, and move any
3203 * shrinkable submounts to the 'graveyard' list.
3205 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3207 struct mount *this_parent = parent;
3208 struct list_head *next;
3212 next = this_parent->mnt_mounts.next;
3214 while (next != &this_parent->mnt_mounts) {
3215 struct list_head *tmp = next;
3216 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3219 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3222 * Descend a level if the d_mounts list is non-empty.
3224 if (!list_empty(&mnt->mnt_mounts)) {
3229 if (!propagate_mount_busy(mnt, 1)) {
3230 list_move_tail(&mnt->mnt_expire, graveyard);
3235 * All done at this level ... ascend and resume the search
3237 if (this_parent != parent) {
3238 next = this_parent->mnt_child.next;
3239 this_parent = this_parent->mnt_parent;
3246 * process a list of expirable mountpoints with the intent of discarding any
3247 * submounts of a specific parent mountpoint
3249 * mount_lock must be held for write
3251 static void shrink_submounts(struct mount *mnt)
3253 LIST_HEAD(graveyard);
3256 /* extract submounts of 'mountpoint' from the expiration list */
3257 while (select_submounts(mnt, &graveyard)) {
3258 while (!list_empty(&graveyard)) {
3259 m = list_first_entry(&graveyard, struct mount,
3261 touch_mnt_namespace(m->mnt_ns);
3262 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3267 static void *copy_mount_options(const void __user * data)
3270 unsigned left, offset;
3275 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3277 return ERR_PTR(-ENOMEM);
3279 left = copy_from_user(copy, data, PAGE_SIZE);
3282 * Not all architectures have an exact copy_from_user(). Resort to
3285 offset = PAGE_SIZE - left;
3288 if (get_user(c, (const char __user *)data + offset))
3295 if (left == PAGE_SIZE) {
3297 return ERR_PTR(-EFAULT);
3303 static char *copy_mount_string(const void __user *data)
3305 return data ? strndup_user(data, PATH_MAX) : NULL;
3309 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3310 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3312 * data is a (void *) that can point to any structure up to
3313 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3314 * information (or be NULL).
3316 * Pre-0.97 versions of mount() didn't have a flags word.
3317 * When the flags word was introduced its top half was required
3318 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3319 * Therefore, if this magic number is present, it carries no information
3320 * and must be discarded.
3322 int path_mount(const char *dev_name, struct path *path,
3323 const char *type_page, unsigned long flags, void *data_page)
3325 unsigned int mnt_flags = 0, sb_flags;
3329 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3330 flags &= ~MS_MGC_MSK;
3332 /* Basic sanity checks */
3334 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3336 if (flags & MS_NOUSER)
3339 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3344 if (flags & SB_MANDLOCK)
3347 /* Default to relatime unless overriden */
3348 if (!(flags & MS_NOATIME))
3349 mnt_flags |= MNT_RELATIME;
3351 /* Separate the per-mountpoint flags */
3352 if (flags & MS_NOSUID)
3353 mnt_flags |= MNT_NOSUID;
3354 if (flags & MS_NODEV)
3355 mnt_flags |= MNT_NODEV;
3356 if (flags & MS_NOEXEC)
3357 mnt_flags |= MNT_NOEXEC;
3358 if (flags & MS_NOATIME)
3359 mnt_flags |= MNT_NOATIME;
3360 if (flags & MS_NODIRATIME)
3361 mnt_flags |= MNT_NODIRATIME;
3362 if (flags & MS_STRICTATIME)
3363 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3364 if (flags & MS_RDONLY)
3365 mnt_flags |= MNT_READONLY;
3366 if (flags & MS_NOSYMFOLLOW)
3367 mnt_flags |= MNT_NOSYMFOLLOW;
3369 /* The default atime for remount is preservation */
3370 if ((flags & MS_REMOUNT) &&
3371 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3372 MS_STRICTATIME)) == 0)) {
3373 mnt_flags &= ~MNT_ATIME_MASK;
3374 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3377 sb_flags = flags & (SB_RDONLY |
3386 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3387 return do_reconfigure_mnt(path, mnt_flags);
3388 if (flags & MS_REMOUNT)
3389 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3390 if (flags & MS_BIND)
3391 return do_loopback(path, dev_name, flags & MS_REC);
3392 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3393 return do_change_type(path, flags);
3394 if (flags & MS_MOVE)
3395 return do_move_mount_old(path, dev_name);
3397 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3401 long do_mount(const char *dev_name, const char __user *dir_name,
3402 const char *type_page, unsigned long flags, void *data_page)
3407 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3410 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3415 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3417 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3420 static void dec_mnt_namespaces(struct ucounts *ucounts)
3422 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3425 static void free_mnt_ns(struct mnt_namespace *ns)
3427 if (!is_anon_ns(ns))
3428 ns_free_inum(&ns->ns);
3429 dec_mnt_namespaces(ns->ucounts);
3430 put_user_ns(ns->user_ns);
3435 * Assign a sequence number so we can detect when we attempt to bind
3436 * mount a reference to an older mount namespace into the current
3437 * mount namespace, preventing reference counting loops. A 64bit
3438 * number incrementing at 10Ghz will take 12,427 years to wrap which
3439 * is effectively never, so we can ignore the possibility.
3441 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3443 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3445 struct mnt_namespace *new_ns;
3446 struct ucounts *ucounts;
3449 ucounts = inc_mnt_namespaces(user_ns);
3451 return ERR_PTR(-ENOSPC);
3453 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3455 dec_mnt_namespaces(ucounts);
3456 return ERR_PTR(-ENOMEM);
3459 ret = ns_alloc_inum(&new_ns->ns);
3462 dec_mnt_namespaces(ucounts);
3463 return ERR_PTR(ret);
3466 new_ns->ns.ops = &mntns_operations;
3468 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3469 refcount_set(&new_ns->ns.count, 1);
3470 INIT_LIST_HEAD(&new_ns->list);
3471 init_waitqueue_head(&new_ns->poll);
3472 spin_lock_init(&new_ns->ns_lock);
3473 new_ns->user_ns = get_user_ns(user_ns);
3474 new_ns->ucounts = ucounts;
3479 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3480 struct user_namespace *user_ns, struct fs_struct *new_fs)
3482 struct mnt_namespace *new_ns;
3483 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3484 struct mount *p, *q;
3491 if (likely(!(flags & CLONE_NEWNS))) {
3498 new_ns = alloc_mnt_ns(user_ns, false);
3503 /* First pass: copy the tree topology */
3504 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3505 if (user_ns != ns->user_ns)
3506 copy_flags |= CL_SHARED_TO_SLAVE;
3507 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3510 free_mnt_ns(new_ns);
3511 return ERR_CAST(new);
3513 if (user_ns != ns->user_ns) {
3516 unlock_mount_hash();
3519 list_add_tail(&new_ns->list, &new->mnt_list);
3522 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3523 * as belonging to new namespace. We have already acquired a private
3524 * fs_struct, so tsk->fs->lock is not needed.
3532 if (&p->mnt == new_fs->root.mnt) {
3533 new_fs->root.mnt = mntget(&q->mnt);
3536 if (&p->mnt == new_fs->pwd.mnt) {
3537 new_fs->pwd.mnt = mntget(&q->mnt);
3541 p = next_mnt(p, old);
3542 q = next_mnt(q, new);
3545 // an mntns binding we'd skipped?
3546 while (p->mnt.mnt_root != q->mnt.mnt_root)
3547 p = next_mnt(skip_mnt_tree(p), old);
3559 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3561 struct mount *mnt = real_mount(m);
3562 struct mnt_namespace *ns;
3563 struct super_block *s;
3567 ns = alloc_mnt_ns(&init_user_ns, true);
3570 return ERR_CAST(ns);
3575 list_add(&mnt->mnt_list, &ns->list);
3577 err = vfs_path_lookup(m->mnt_root, m,
3578 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3583 return ERR_PTR(err);
3585 /* trade a vfsmount reference for active sb one */
3586 s = path.mnt->mnt_sb;
3587 atomic_inc(&s->s_active);
3589 /* lock the sucker */
3590 down_write(&s->s_umount);
3591 /* ... and return the root of (sub)tree on it */
3594 EXPORT_SYMBOL(mount_subtree);
3596 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3597 char __user *, type, unsigned long, flags, void __user *, data)
3604 kernel_type = copy_mount_string(type);
3605 ret = PTR_ERR(kernel_type);
3606 if (IS_ERR(kernel_type))
3609 kernel_dev = copy_mount_string(dev_name);
3610 ret = PTR_ERR(kernel_dev);
3611 if (IS_ERR(kernel_dev))
3614 options = copy_mount_options(data);
3615 ret = PTR_ERR(options);
3616 if (IS_ERR(options))
3619 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3630 #define FSMOUNT_VALID_FLAGS \
3631 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3632 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3633 MOUNT_ATTR_NOSYMFOLLOW)
3635 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3637 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3638 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3640 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3642 unsigned int mnt_flags = 0;
3644 if (attr_flags & MOUNT_ATTR_RDONLY)
3645 mnt_flags |= MNT_READONLY;
3646 if (attr_flags & MOUNT_ATTR_NOSUID)
3647 mnt_flags |= MNT_NOSUID;
3648 if (attr_flags & MOUNT_ATTR_NODEV)
3649 mnt_flags |= MNT_NODEV;
3650 if (attr_flags & MOUNT_ATTR_NOEXEC)
3651 mnt_flags |= MNT_NOEXEC;
3652 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3653 mnt_flags |= MNT_NODIRATIME;
3654 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3655 mnt_flags |= MNT_NOSYMFOLLOW;
3661 * Create a kernel mount representation for a new, prepared superblock
3662 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3664 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3665 unsigned int, attr_flags)
3667 struct mnt_namespace *ns;
3668 struct fs_context *fc;
3670 struct path newmount;
3673 unsigned int mnt_flags = 0;
3679 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3682 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3685 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3687 switch (attr_flags & MOUNT_ATTR__ATIME) {
3688 case MOUNT_ATTR_STRICTATIME:
3690 case MOUNT_ATTR_NOATIME:
3691 mnt_flags |= MNT_NOATIME;
3693 case MOUNT_ATTR_RELATIME:
3694 mnt_flags |= MNT_RELATIME;
3705 if (f.file->f_op != &fscontext_fops)
3708 fc = f.file->private_data;
3710 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3714 /* There must be a valid superblock or we can't mount it */
3720 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3721 pr_warn("VFS: Mount too revealing\n");
3726 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3729 if (fc->sb_flags & SB_MANDLOCK)
3732 newmount.mnt = vfs_create_mount(fc);
3733 if (IS_ERR(newmount.mnt)) {
3734 ret = PTR_ERR(newmount.mnt);
3737 newmount.dentry = dget(fc->root);
3738 newmount.mnt->mnt_flags = mnt_flags;
3740 /* We've done the mount bit - now move the file context into more or
3741 * less the same state as if we'd done an fspick(). We don't want to
3742 * do any memory allocation or anything like that at this point as we
3743 * don't want to have to handle any errors incurred.
3745 vfs_clean_context(fc);
3747 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3752 mnt = real_mount(newmount.mnt);
3756 list_add(&mnt->mnt_list, &ns->list);
3757 mntget(newmount.mnt);
3759 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3760 * it, not just simply put it.
3762 file = dentry_open(&newmount, O_PATH, fc->cred);
3764 dissolve_on_fput(newmount.mnt);
3765 ret = PTR_ERR(file);
3768 file->f_mode |= FMODE_NEED_UNMOUNT;
3770 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3772 fd_install(ret, file);
3777 path_put(&newmount);
3779 mutex_unlock(&fc->uapi_mutex);
3786 * Move a mount from one place to another. In combination with
3787 * fsopen()/fsmount() this is used to install a new mount and in combination
3788 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3791 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3793 SYSCALL_DEFINE5(move_mount,
3794 int, from_dfd, const char __user *, from_pathname,
3795 int, to_dfd, const char __user *, to_pathname,
3796 unsigned int, flags)
3798 struct path from_path, to_path;
3799 unsigned int lflags;
3805 if (flags & ~MOVE_MOUNT__MASK)
3808 /* If someone gives a pathname, they aren't permitted to move
3809 * from an fd that requires unmount as we can't get at the flag
3810 * to clear it afterwards.
3813 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3814 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3815 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3817 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3822 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3823 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3824 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3826 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3830 ret = security_move_mount(&from_path, &to_path);
3834 if (flags & MOVE_MOUNT_SET_GROUP)
3835 ret = do_set_group(&from_path, &to_path);
3837 ret = do_move_mount(&from_path, &to_path);
3842 path_put(&from_path);
3847 * Return true if path is reachable from root
3849 * namespace_sem or mount_lock is held
3851 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3852 const struct path *root)
3854 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3855 dentry = mnt->mnt_mountpoint;
3856 mnt = mnt->mnt_parent;
3858 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3861 bool path_is_under(const struct path *path1, const struct path *path2)
3864 read_seqlock_excl(&mount_lock);
3865 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3866 read_sequnlock_excl(&mount_lock);
3869 EXPORT_SYMBOL(path_is_under);
3872 * pivot_root Semantics:
3873 * Moves the root file system of the current process to the directory put_old,
3874 * makes new_root as the new root file system of the current process, and sets
3875 * root/cwd of all processes which had them on the current root to new_root.
3878 * The new_root and put_old must be directories, and must not be on the
3879 * same file system as the current process root. The put_old must be
3880 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3881 * pointed to by put_old must yield the same directory as new_root. No other
3882 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3884 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3885 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3886 * in this situation.
3889 * - we don't move root/cwd if they are not at the root (reason: if something
3890 * cared enough to change them, it's probably wrong to force them elsewhere)
3891 * - it's okay to pick a root that isn't the root of a file system, e.g.
3892 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3893 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3896 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3897 const char __user *, put_old)
3899 struct path new, old, root;
3900 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3901 struct mountpoint *old_mp, *root_mp;
3907 error = user_path_at(AT_FDCWD, new_root,
3908 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3912 error = user_path_at(AT_FDCWD, put_old,
3913 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3917 error = security_sb_pivotroot(&old, &new);
3921 get_fs_root(current->fs, &root);
3922 old_mp = lock_mount(&old);
3923 error = PTR_ERR(old_mp);
3928 new_mnt = real_mount(new.mnt);
3929 root_mnt = real_mount(root.mnt);
3930 old_mnt = real_mount(old.mnt);
3931 ex_parent = new_mnt->mnt_parent;
3932 root_parent = root_mnt->mnt_parent;
3933 if (IS_MNT_SHARED(old_mnt) ||
3934 IS_MNT_SHARED(ex_parent) ||
3935 IS_MNT_SHARED(root_parent))
3937 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3939 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3942 if (d_unlinked(new.dentry))
3945 if (new_mnt == root_mnt || old_mnt == root_mnt)
3946 goto out4; /* loop, on the same file system */
3948 if (!path_mounted(&root))
3949 goto out4; /* not a mountpoint */
3950 if (!mnt_has_parent(root_mnt))
3951 goto out4; /* not attached */
3952 if (!path_mounted(&new))
3953 goto out4; /* not a mountpoint */
3954 if (!mnt_has_parent(new_mnt))
3955 goto out4; /* not attached */
3956 /* make sure we can reach put_old from new_root */
3957 if (!is_path_reachable(old_mnt, old.dentry, &new))
3959 /* make certain new is below the root */
3960 if (!is_path_reachable(new_mnt, new.dentry, &root))
3963 umount_mnt(new_mnt);
3964 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3965 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3966 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3967 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3969 /* mount old root on put_old */
3970 attach_mnt(root_mnt, old_mnt, old_mp);
3971 /* mount new_root on / */
3972 attach_mnt(new_mnt, root_parent, root_mp);
3973 mnt_add_count(root_parent, -1);
3974 touch_mnt_namespace(current->nsproxy->mnt_ns);
3975 /* A moved mount should not expire automatically */
3976 list_del_init(&new_mnt->mnt_expire);
3977 put_mountpoint(root_mp);
3978 unlock_mount_hash();
3979 chroot_fs_refs(&root, &new);
3982 unlock_mount(old_mp);
3984 mntput_no_expire(ex_parent);
3995 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3997 unsigned int flags = mnt->mnt.mnt_flags;
3999 /* flags to clear */
4000 flags &= ~kattr->attr_clr;
4001 /* flags to raise */
4002 flags |= kattr->attr_set;
4007 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4009 struct vfsmount *m = &mnt->mnt;
4010 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4012 if (!kattr->mnt_idmap)
4016 * Creating an idmapped mount with the filesystem wide idmapping
4017 * doesn't make sense so block that. We don't allow mushy semantics.
4019 if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb))
4023 * Once a mount has been idmapped we don't allow it to change its
4024 * mapping. It makes things simpler and callers can just create
4025 * another bind-mount they can idmap if they want to.
4027 if (is_idmapped_mnt(m))
4030 /* The underlying filesystem doesn't support idmapped mounts yet. */
4031 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4034 /* We're not controlling the superblock. */
4035 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4038 /* Mount has already been visible in the filesystem hierarchy. */
4039 if (!is_anon_ns(mnt->mnt_ns))
4046 * mnt_allow_writers() - check whether the attribute change allows writers
4047 * @kattr: the new mount attributes
4048 * @mnt: the mount to which @kattr will be applied
4050 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4052 * Return: true if writers need to be held, false if not
4054 static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4055 const struct mount *mnt)
4057 return (!(kattr->attr_set & MNT_READONLY) ||
4058 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4062 static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4067 for (m = mnt; m; m = next_mnt(m, mnt)) {
4068 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4073 err = can_idmap_mount(kattr, m);
4077 if (!mnt_allow_writers(kattr, m)) {
4078 err = mnt_hold_writers(m);
4083 if (!kattr->recurse)
4091 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4092 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4093 * mounts and needs to take care to include the first mount.
4095 for (p = mnt; p; p = next_mnt(p, mnt)) {
4096 /* If we had to hold writers unblock them. */
4097 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4098 mnt_unhold_writers(p);
4101 * We're done once the first mount we changed got
4102 * MNT_WRITE_HOLD unset.
4111 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4113 if (!kattr->mnt_idmap)
4117 * Pairs with smp_load_acquire() in mnt_idmap().
4119 * Since we only allow a mount to change the idmapping once and
4120 * verified this in can_idmap_mount() we know that the mount has
4121 * @nop_mnt_idmap attached to it. So there's no need to drop any
4124 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4127 static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4131 for (m = mnt; m; m = next_mnt(m, mnt)) {
4134 do_idmap_mount(kattr, m);
4135 flags = recalc_flags(kattr, m);
4136 WRITE_ONCE(m->mnt.mnt_flags, flags);
4138 /* If we had to hold writers unblock them. */
4139 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4140 mnt_unhold_writers(m);
4142 if (kattr->propagation)
4143 change_mnt_propagation(m, kattr->propagation);
4144 if (!kattr->recurse)
4147 touch_mnt_namespace(mnt->mnt_ns);
4150 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4152 struct mount *mnt = real_mount(path->mnt);
4155 if (!path_mounted(path))
4158 if (kattr->mnt_userns) {
4159 struct mnt_idmap *mnt_idmap;
4161 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4162 if (IS_ERR(mnt_idmap))
4163 return PTR_ERR(mnt_idmap);
4164 kattr->mnt_idmap = mnt_idmap;
4167 if (kattr->propagation) {
4169 * Only take namespace_lock() if we're actually changing
4173 if (kattr->propagation == MS_SHARED) {
4174 err = invent_group_ids(mnt, kattr->recurse);
4185 /* Ensure that this isn't anything purely vfs internal. */
4186 if (!is_mounted(&mnt->mnt))
4190 * If this is an attached mount make sure it's located in the callers
4191 * mount namespace. If it's not don't let the caller interact with it.
4192 * If this is a detached mount make sure it has an anonymous mount
4193 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4195 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4199 * First, we get the mount tree in a shape where we can change mount
4200 * properties without failure. If we succeeded to do so we commit all
4201 * changes and if we failed we clean up.
4203 err = mount_setattr_prepare(kattr, mnt);
4205 mount_setattr_commit(kattr, mnt);
4208 unlock_mount_hash();
4210 if (kattr->propagation) {
4212 cleanup_group_ids(mnt, NULL);
4219 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4220 struct mount_kattr *kattr, unsigned int flags)
4223 struct ns_common *ns;
4224 struct user_namespace *mnt_userns;
4227 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4231 * We currently do not support clearing an idmapped mount. If this ever
4232 * is a use-case we can revisit this but for now let's keep it simple
4235 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4238 if (attr->userns_fd > INT_MAX)
4241 f = fdget(attr->userns_fd);
4245 if (!proc_ns_file(f.file)) {
4250 ns = get_proc_ns(file_inode(f.file));
4251 if (ns->ops->type != CLONE_NEWUSER) {
4257 * The initial idmapping cannot be used to create an idmapped
4258 * mount. We use the initial idmapping as an indicator of a mount
4259 * that is not idmapped. It can simply be passed into helpers that
4260 * are aware of idmapped mounts as a convenient shortcut. A user
4261 * can just create a dedicated identity mapping to achieve the same
4264 mnt_userns = container_of(ns, struct user_namespace, ns);
4265 if (mnt_userns == &init_user_ns) {
4270 /* We're not controlling the target namespace. */
4271 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4276 kattr->mnt_userns = get_user_ns(mnt_userns);
4283 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4284 struct mount_kattr *kattr, unsigned int flags)
4286 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4288 if (flags & AT_NO_AUTOMOUNT)
4289 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4290 if (flags & AT_SYMLINK_NOFOLLOW)
4291 lookup_flags &= ~LOOKUP_FOLLOW;
4292 if (flags & AT_EMPTY_PATH)
4293 lookup_flags |= LOOKUP_EMPTY;
4295 *kattr = (struct mount_kattr) {
4296 .lookup_flags = lookup_flags,
4297 .recurse = !!(flags & AT_RECURSIVE),
4300 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4302 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4304 kattr->propagation = attr->propagation;
4306 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4309 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4310 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4313 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4314 * users wanting to transition to a different atime setting cannot
4315 * simply specify the atime setting in @attr_set, but must also
4316 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4317 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4318 * @attr_clr and that @attr_set can't have any atime bits set if
4319 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4321 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4322 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4326 * Clear all previous time settings as they are mutually
4329 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4330 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4331 case MOUNT_ATTR_RELATIME:
4332 kattr->attr_set |= MNT_RELATIME;
4334 case MOUNT_ATTR_NOATIME:
4335 kattr->attr_set |= MNT_NOATIME;
4337 case MOUNT_ATTR_STRICTATIME:
4343 if (attr->attr_set & MOUNT_ATTR__ATIME)
4347 return build_mount_idmapped(attr, usize, kattr, flags);
4350 static void finish_mount_kattr(struct mount_kattr *kattr)
4352 put_user_ns(kattr->mnt_userns);
4353 kattr->mnt_userns = NULL;
4355 if (kattr->mnt_idmap)
4356 mnt_idmap_put(kattr->mnt_idmap);
4359 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4360 unsigned int, flags, struct mount_attr __user *, uattr,
4365 struct mount_attr attr;
4366 struct mount_kattr kattr;
4368 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4370 if (flags & ~(AT_EMPTY_PATH |
4372 AT_SYMLINK_NOFOLLOW |
4376 if (unlikely(usize > PAGE_SIZE))
4378 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4384 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4388 /* Don't bother walking through the mounts if this is a nop. */
4389 if (attr.attr_set == 0 &&
4390 attr.attr_clr == 0 &&
4391 attr.propagation == 0)
4394 err = build_mount_kattr(&attr, usize, &kattr, flags);
4398 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4400 err = do_mount_setattr(&target, &kattr);
4403 finish_mount_kattr(&kattr);
4407 static void __init init_mount_tree(void)
4409 struct vfsmount *mnt;
4411 struct mnt_namespace *ns;
4414 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4416 panic("Can't create rootfs");
4418 ns = alloc_mnt_ns(&init_user_ns, false);
4420 panic("Can't allocate initial namespace");
4421 m = real_mount(mnt);
4425 list_add(&m->mnt_list, &ns->list);
4426 init_task.nsproxy->mnt_ns = ns;
4430 root.dentry = mnt->mnt_root;
4431 mnt->mnt_flags |= MNT_LOCKED;
4433 set_fs_pwd(current->fs, &root);
4434 set_fs_root(current->fs, &root);
4437 void __init mnt_init(void)
4441 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4442 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4444 mount_hashtable = alloc_large_system_hash("Mount-cache",
4445 sizeof(struct hlist_head),
4448 &m_hash_shift, &m_hash_mask, 0, 0);
4449 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4450 sizeof(struct hlist_head),
4453 &mp_hash_shift, &mp_hash_mask, 0, 0);
4455 if (!mount_hashtable || !mountpoint_hashtable)
4456 panic("Failed to allocate mount hash table\n");
4462 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4464 fs_kobj = kobject_create_and_add("fs", NULL);
4466 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4472 void put_mnt_ns(struct mnt_namespace *ns)
4474 if (!refcount_dec_and_test(&ns->ns.count))
4476 drop_collected_mounts(&ns->root->mnt);
4480 struct vfsmount *kern_mount(struct file_system_type *type)
4482 struct vfsmount *mnt;
4483 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4486 * it is a longterm mount, don't release mnt until
4487 * we unmount before file sys is unregistered
4489 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4493 EXPORT_SYMBOL_GPL(kern_mount);
4495 void kern_unmount(struct vfsmount *mnt)
4497 /* release long term mount so mount point can be released */
4499 mnt_make_shortterm(mnt);
4500 synchronize_rcu(); /* yecchhh... */
4504 EXPORT_SYMBOL(kern_unmount);
4506 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4510 for (i = 0; i < num; i++)
4511 mnt_make_shortterm(mnt[i]);
4512 synchronize_rcu_expedited();
4513 for (i = 0; i < num; i++)
4516 EXPORT_SYMBOL(kern_unmount_array);
4518 bool our_mnt(struct vfsmount *mnt)
4520 return check_mnt(real_mount(mnt));
4523 bool current_chrooted(void)
4525 /* Does the current process have a non-standard root */
4526 struct path ns_root;
4527 struct path fs_root;
4530 /* Find the namespace root */
4531 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4532 ns_root.dentry = ns_root.mnt->mnt_root;
4534 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4537 get_fs_root(current->fs, &fs_root);
4539 chrooted = !path_equal(&fs_root, &ns_root);
4547 static bool mnt_already_visible(struct mnt_namespace *ns,
4548 const struct super_block *sb,
4551 int new_flags = *new_mnt_flags;
4553 bool visible = false;
4555 down_read(&namespace_sem);
4557 list_for_each_entry(mnt, &ns->list, mnt_list) {
4558 struct mount *child;
4561 if (mnt_is_cursor(mnt))
4564 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4567 /* This mount is not fully visible if it's root directory
4568 * is not the root directory of the filesystem.
4570 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4573 /* A local view of the mount flags */
4574 mnt_flags = mnt->mnt.mnt_flags;
4576 /* Don't miss readonly hidden in the superblock flags */
4577 if (sb_rdonly(mnt->mnt.mnt_sb))
4578 mnt_flags |= MNT_LOCK_READONLY;
4580 /* Verify the mount flags are equal to or more permissive
4581 * than the proposed new mount.
4583 if ((mnt_flags & MNT_LOCK_READONLY) &&
4584 !(new_flags & MNT_READONLY))
4586 if ((mnt_flags & MNT_LOCK_ATIME) &&
4587 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4590 /* This mount is not fully visible if there are any
4591 * locked child mounts that cover anything except for
4592 * empty directories.
4594 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4595 struct inode *inode = child->mnt_mountpoint->d_inode;
4596 /* Only worry about locked mounts */
4597 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4599 /* Is the directory permanetly empty? */
4600 if (!is_empty_dir_inode(inode))
4603 /* Preserve the locked attributes */
4604 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4612 up_read(&namespace_sem);
4616 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4618 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4619 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4620 unsigned long s_iflags;
4622 if (ns->user_ns == &init_user_ns)
4625 /* Can this filesystem be too revealing? */
4626 s_iflags = sb->s_iflags;
4627 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4630 if ((s_iflags & required_iflags) != required_iflags) {
4631 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4636 return !mnt_already_visible(ns, sb, new_mnt_flags);
4639 bool mnt_may_suid(struct vfsmount *mnt)
4642 * Foreign mounts (accessed via fchdir or through /proc
4643 * symlinks) are always treated as if they are nosuid. This
4644 * prevents namespaces from trusting potentially unsafe
4645 * suid/sgid bits, file caps, or security labels that originate
4646 * in other namespaces.
4648 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4649 current_in_userns(mnt->mnt_sb->s_user_ns);
4652 static struct ns_common *mntns_get(struct task_struct *task)
4654 struct ns_common *ns = NULL;
4655 struct nsproxy *nsproxy;
4658 nsproxy = task->nsproxy;
4660 ns = &nsproxy->mnt_ns->ns;
4661 get_mnt_ns(to_mnt_ns(ns));
4668 static void mntns_put(struct ns_common *ns)
4670 put_mnt_ns(to_mnt_ns(ns));
4673 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4675 struct nsproxy *nsproxy = nsset->nsproxy;
4676 struct fs_struct *fs = nsset->fs;
4677 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4678 struct user_namespace *user_ns = nsset->cred->user_ns;
4682 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4683 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4684 !ns_capable(user_ns, CAP_SYS_ADMIN))
4687 if (is_anon_ns(mnt_ns))
4694 old_mnt_ns = nsproxy->mnt_ns;
4695 nsproxy->mnt_ns = mnt_ns;
4698 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4699 "/", LOOKUP_DOWN, &root);
4701 /* revert to old namespace */
4702 nsproxy->mnt_ns = old_mnt_ns;
4707 put_mnt_ns(old_mnt_ns);
4709 /* Update the pwd and root */
4710 set_fs_pwd(fs, &root);
4711 set_fs_root(fs, &root);
4717 static struct user_namespace *mntns_owner(struct ns_common *ns)
4719 return to_mnt_ns(ns)->user_ns;
4722 const struct proc_ns_operations mntns_operations = {
4724 .type = CLONE_NEWNS,
4727 .install = mntns_install,
4728 .owner = mntns_owner,
4731 #ifdef CONFIG_SYSCTL
4732 static struct ctl_table fs_namespace_sysctls[] = {
4734 .procname = "mount-max",
4735 .data = &sysctl_mount_max,
4736 .maxlen = sizeof(unsigned int),
4738 .proc_handler = proc_dointvec_minmax,
4739 .extra1 = SYSCTL_ONE,
4744 static int __init init_fs_namespace_sysctls(void)
4746 register_sysctl_init("fs", fs_namespace_sysctls);
4749 fs_initcall(init_fs_namespace_sysctls);
4751 #endif /* CONFIG_SYSCTL */