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 __ro_after_init;
43 static unsigned int m_hash_shift __ro_after_init;
44 static unsigned int mp_hash_mask __ro_after_init;
45 static unsigned int mp_hash_shift __ro_after_init;
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 /* Don't allow confusion with old 32bit mount ID */
72 static atomic64_t mnt_id_ctr = ATOMIC64_INIT(1ULL << 32);
74 static struct hlist_head *mount_hashtable __ro_after_init;
75 static struct hlist_head *mountpoint_hashtable __ro_after_init;
76 static struct kmem_cache *mnt_cache __ro_after_init;
77 static DECLARE_RWSEM(namespace_sem);
78 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
79 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
82 unsigned int attr_set;
83 unsigned int attr_clr;
84 unsigned int propagation;
85 unsigned int lookup_flags;
87 struct user_namespace *mnt_userns;
88 struct mnt_idmap *mnt_idmap;
92 struct kobject *fs_kobj __ro_after_init;
93 EXPORT_SYMBOL_GPL(fs_kobj);
96 * vfsmount lock may be taken for read to prevent changes to the
97 * vfsmount hash, ie. during mountpoint lookups or walking back
100 * It should be taken for write in all cases where the vfsmount
101 * tree or hash is modified or when a vfsmount structure is modified.
103 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
105 static inline void lock_mount_hash(void)
107 write_seqlock(&mount_lock);
110 static inline void unlock_mount_hash(void)
112 write_sequnlock(&mount_lock);
115 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
117 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
118 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
119 tmp = tmp + (tmp >> m_hash_shift);
120 return &mount_hashtable[tmp & m_hash_mask];
123 static inline struct hlist_head *mp_hash(struct dentry *dentry)
125 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
126 tmp = tmp + (tmp >> mp_hash_shift);
127 return &mountpoint_hashtable[tmp & mp_hash_mask];
130 static int mnt_alloc_id(struct mount *mnt)
132 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
137 mnt->mnt_id_unique = atomic64_inc_return(&mnt_id_ctr);
141 static void mnt_free_id(struct mount *mnt)
143 ida_free(&mnt_id_ida, mnt->mnt_id);
147 * Allocate a new peer group ID
149 static int mnt_alloc_group_id(struct mount *mnt)
151 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
155 mnt->mnt_group_id = res;
160 * Release a peer group ID
162 void mnt_release_group_id(struct mount *mnt)
164 ida_free(&mnt_group_ida, mnt->mnt_group_id);
165 mnt->mnt_group_id = 0;
169 * vfsmount lock must be held for read
171 static inline void mnt_add_count(struct mount *mnt, int n)
174 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
183 * vfsmount lock must be held for write
185 int mnt_get_count(struct mount *mnt)
191 for_each_possible_cpu(cpu) {
192 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
197 return mnt->mnt_count;
201 static struct mount *alloc_vfsmnt(const char *name)
203 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
207 err = mnt_alloc_id(mnt);
212 mnt->mnt_devname = kstrdup_const(name,
214 if (!mnt->mnt_devname)
219 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 goto out_free_devname;
223 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
226 mnt->mnt_writers = 0;
229 INIT_HLIST_NODE(&mnt->mnt_hash);
230 INIT_LIST_HEAD(&mnt->mnt_child);
231 INIT_LIST_HEAD(&mnt->mnt_mounts);
232 INIT_LIST_HEAD(&mnt->mnt_list);
233 INIT_LIST_HEAD(&mnt->mnt_expire);
234 INIT_LIST_HEAD(&mnt->mnt_share);
235 INIT_LIST_HEAD(&mnt->mnt_slave_list);
236 INIT_LIST_HEAD(&mnt->mnt_slave);
237 INIT_HLIST_NODE(&mnt->mnt_mp_list);
238 INIT_LIST_HEAD(&mnt->mnt_umounting);
239 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
240 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
246 kfree_const(mnt->mnt_devname);
251 kmem_cache_free(mnt_cache, mnt);
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
274 bool __mnt_is_readonly(struct vfsmount *mnt)
276 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
278 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
280 static inline void mnt_inc_writers(struct mount *mnt)
283 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
289 static inline void mnt_dec_writers(struct mount *mnt)
292 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
298 static unsigned int mnt_get_writers(struct mount *mnt)
301 unsigned int count = 0;
304 for_each_possible_cpu(cpu) {
305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 return mnt->mnt_writers;
314 static int mnt_is_readonly(struct vfsmount *mnt)
316 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
319 * The barrier pairs with the barrier in sb_start_ro_state_change()
320 * making sure if we don't see s_readonly_remount set yet, we also will
321 * not see any superblock / mount flag changes done by remount.
322 * It also pairs with the barrier in sb_end_ro_state_change()
323 * assuring that if we see s_readonly_remount already cleared, we will
324 * see the values of superblock / mount flags updated by remount.
327 return __mnt_is_readonly(mnt);
331 * Most r/o & frozen checks on a fs are for operations that take discrete
332 * amounts of time, like a write() or unlink(). We must keep track of when
333 * those operations start (for permission checks) and when they end, so that we
334 * can determine when writes are able to occur to a filesystem.
337 * mnt_get_write_access - get write access to a mount without freeze protection
338 * @m: the mount on which to take a write
340 * This tells the low-level filesystem that a write is about to be performed to
341 * it, and makes sure that writes are allowed (mnt it read-write) before
342 * returning success. This operation does not protect against filesystem being
343 * frozen. When the write operation is finished, mnt_put_write_access() must be
344 * called. This is effectively a refcount.
346 int mnt_get_write_access(struct vfsmount *m)
348 struct mount *mnt = real_mount(m);
352 mnt_inc_writers(mnt);
354 * The store to mnt_inc_writers must be visible before we pass
355 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
356 * incremented count after it has set MNT_WRITE_HOLD.
359 might_lock(&mount_lock.lock);
360 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
361 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
365 * This prevents priority inversion, if the task
366 * setting MNT_WRITE_HOLD got preempted on a remote
367 * CPU, and it prevents life lock if the task setting
368 * MNT_WRITE_HOLD has a lower priority and is bound to
369 * the same CPU as the task that is spinning here.
378 * The barrier pairs with the barrier sb_start_ro_state_change() making
379 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
380 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
381 * mnt_is_readonly() and bail in case we are racing with remount
385 if (mnt_is_readonly(m)) {
386 mnt_dec_writers(mnt);
393 EXPORT_SYMBOL_GPL(mnt_get_write_access);
396 * mnt_want_write - get write access to a mount
397 * @m: the mount on which to take a write
399 * This tells the low-level filesystem that a write is about to be performed to
400 * it, and makes sure that writes are allowed (mount is read-write, filesystem
401 * is not frozen) before returning success. When the write operation is
402 * finished, mnt_drop_write() must be called. This is effectively a refcount.
404 int mnt_want_write(struct vfsmount *m)
408 sb_start_write(m->mnt_sb);
409 ret = mnt_get_write_access(m);
411 sb_end_write(m->mnt_sb);
414 EXPORT_SYMBOL_GPL(mnt_want_write);
417 * mnt_get_write_access_file - get write access to a file's mount
418 * @file: the file who's mount on which to take a write
420 * This is like mnt_get_write_access, but if @file is already open for write it
421 * skips incrementing mnt_writers (since the open file already has a reference)
422 * and instead only does the check for emergency r/o remounts. This must be
423 * paired with mnt_put_write_access_file.
425 int mnt_get_write_access_file(struct file *file)
427 if (file->f_mode & FMODE_WRITER) {
429 * Superblock may have become readonly while there are still
430 * writable fd's, e.g. due to a fs error with errors=remount-ro
432 if (__mnt_is_readonly(file->f_path.mnt))
436 return mnt_get_write_access(file->f_path.mnt);
440 * mnt_want_write_file - get write access to a file's mount
441 * @file: the file who's mount on which to take a write
443 * This is like mnt_want_write, but if the file is already open for writing it
444 * skips incrementing mnt_writers (since the open file already has a reference)
445 * and instead only does the freeze protection and the check for emergency r/o
446 * remounts. This must be paired with mnt_drop_write_file.
448 int mnt_want_write_file(struct file *file)
452 sb_start_write(file_inode(file)->i_sb);
453 ret = mnt_get_write_access_file(file);
455 sb_end_write(file_inode(file)->i_sb);
458 EXPORT_SYMBOL_GPL(mnt_want_write_file);
461 * mnt_put_write_access - give up write access to a mount
462 * @mnt: the mount on which to give up write access
464 * Tells the low-level filesystem that we are done
465 * performing writes to it. Must be matched with
466 * mnt_get_write_access() call above.
468 void mnt_put_write_access(struct vfsmount *mnt)
471 mnt_dec_writers(real_mount(mnt));
474 EXPORT_SYMBOL_GPL(mnt_put_write_access);
477 * mnt_drop_write - give up write access to a mount
478 * @mnt: the mount on which to give up write access
480 * Tells the low-level filesystem that we are done performing writes to it and
481 * also allows filesystem to be frozen again. Must be matched with
482 * mnt_want_write() call above.
484 void mnt_drop_write(struct vfsmount *mnt)
486 mnt_put_write_access(mnt);
487 sb_end_write(mnt->mnt_sb);
489 EXPORT_SYMBOL_GPL(mnt_drop_write);
491 void mnt_put_write_access_file(struct file *file)
493 if (!(file->f_mode & FMODE_WRITER))
494 mnt_put_write_access(file->f_path.mnt);
497 void mnt_drop_write_file(struct file *file)
499 mnt_put_write_access_file(file);
500 sb_end_write(file_inode(file)->i_sb);
502 EXPORT_SYMBOL(mnt_drop_write_file);
505 * mnt_hold_writers - prevent write access to the given mount
506 * @mnt: mnt to prevent write access to
508 * Prevents write access to @mnt if there are no active writers for @mnt.
509 * This function needs to be called and return successfully before changing
510 * properties of @mnt that need to remain stable for callers with write access
513 * After this functions has been called successfully callers must pair it with
514 * a call to mnt_unhold_writers() in order to stop preventing write access to
517 * Context: This function expects lock_mount_hash() to be held serializing
518 * setting MNT_WRITE_HOLD.
519 * Return: On success 0 is returned.
520 * On error, -EBUSY is returned.
522 static inline int mnt_hold_writers(struct mount *mnt)
524 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
526 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
527 * should be visible before we do.
532 * With writers on hold, if this value is zero, then there are
533 * definitely no active writers (although held writers may subsequently
534 * increment the count, they'll have to wait, and decrement it after
535 * seeing MNT_READONLY).
537 * It is OK to have counter incremented on one CPU and decremented on
538 * another: the sum will add up correctly. The danger would be when we
539 * sum up each counter, if we read a counter before it is incremented,
540 * but then read another CPU's count which it has been subsequently
541 * decremented from -- we would see more decrements than we should.
542 * MNT_WRITE_HOLD protects against this scenario, because
543 * mnt_want_write first increments count, then smp_mb, then spins on
544 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
545 * we're counting up here.
547 if (mnt_get_writers(mnt) > 0)
554 * mnt_unhold_writers - stop preventing write access to the given mount
555 * @mnt: mnt to stop preventing write access to
557 * Stop preventing write access to @mnt allowing callers to gain write access
560 * This function can only be called after a successful call to
561 * mnt_hold_writers().
563 * Context: This function expects lock_mount_hash() to be held.
565 static inline void mnt_unhold_writers(struct mount *mnt)
568 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
569 * that become unheld will see MNT_READONLY.
572 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
575 static int mnt_make_readonly(struct mount *mnt)
579 ret = mnt_hold_writers(mnt);
581 mnt->mnt.mnt_flags |= MNT_READONLY;
582 mnt_unhold_writers(mnt);
586 int sb_prepare_remount_readonly(struct super_block *sb)
591 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
592 if (atomic_long_read(&sb->s_remove_count))
596 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
597 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
598 err = mnt_hold_writers(mnt);
603 if (!err && atomic_long_read(&sb->s_remove_count))
607 sb_start_ro_state_change(sb);
608 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
609 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
610 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
617 static void free_vfsmnt(struct mount *mnt)
619 mnt_idmap_put(mnt_idmap(&mnt->mnt));
620 kfree_const(mnt->mnt_devname);
622 free_percpu(mnt->mnt_pcp);
624 kmem_cache_free(mnt_cache, mnt);
627 static void delayed_free_vfsmnt(struct rcu_head *head)
629 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
632 /* call under rcu_read_lock */
633 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
636 if (read_seqretry(&mount_lock, seq))
640 mnt = real_mount(bastard);
641 mnt_add_count(mnt, 1);
642 smp_mb(); // see mntput_no_expire()
643 if (likely(!read_seqretry(&mount_lock, seq)))
645 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
646 mnt_add_count(mnt, -1);
650 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
651 mnt_add_count(mnt, -1);
656 /* caller will mntput() */
660 /* call under rcu_read_lock */
661 static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
663 int res = __legitimize_mnt(bastard, seq);
666 if (unlikely(res < 0)) {
675 * __lookup_mnt - find first child mount
677 * @dentry: mountpoint
679 * If @mnt has a child mount @c mounted @dentry find and return it.
681 * Note that the child mount @c need not be unique. There are cases
682 * where shadow mounts are created. For example, during mount
683 * propagation when a source mount @mnt whose root got overmounted by a
684 * mount @o after path lookup but before @namespace_sem could be
685 * acquired gets copied and propagated. So @mnt gets copied including
686 * @o. When @mnt is propagated to a destination mount @d that already
687 * has another mount @n mounted at the same mountpoint then the source
688 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
689 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
692 * Return: The first child of @mnt mounted @dentry or NULL.
694 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
696 struct hlist_head *head = m_hash(mnt, dentry);
699 hlist_for_each_entry_rcu(p, head, mnt_hash)
700 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
706 * lookup_mnt - Return the first child mount mounted at path
708 * "First" means first mounted chronologically. If you create the
711 * mount /dev/sda1 /mnt
712 * mount /dev/sda2 /mnt
713 * mount /dev/sda3 /mnt
715 * Then lookup_mnt() on the base /mnt dentry in the root mount will
716 * return successively the root dentry and vfsmount of /dev/sda1, then
717 * /dev/sda2, then /dev/sda3, then NULL.
719 * lookup_mnt takes a reference to the found vfsmount.
721 struct vfsmount *lookup_mnt(const struct path *path)
723 struct mount *child_mnt;
729 seq = read_seqbegin(&mount_lock);
730 child_mnt = __lookup_mnt(path->mnt, path->dentry);
731 m = child_mnt ? &child_mnt->mnt : NULL;
732 } while (!legitimize_mnt(m, seq));
737 static inline void lock_ns_list(struct mnt_namespace *ns)
739 spin_lock(&ns->ns_lock);
742 static inline void unlock_ns_list(struct mnt_namespace *ns)
744 spin_unlock(&ns->ns_lock);
747 static inline bool mnt_is_cursor(struct mount *mnt)
749 return mnt->mnt.mnt_flags & MNT_CURSOR;
753 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
754 * current mount namespace.
756 * The common case is dentries are not mountpoints at all and that
757 * test is handled inline. For the slow case when we are actually
758 * dealing with a mountpoint of some kind, walk through all of the
759 * mounts in the current mount namespace and test to see if the dentry
762 * The mount_hashtable is not usable in the context because we
763 * need to identify all mounts that may be in the current mount
764 * namespace not just a mount that happens to have some specified
767 bool __is_local_mountpoint(struct dentry *dentry)
769 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
771 bool is_covered = false;
773 down_read(&namespace_sem);
775 list_for_each_entry(mnt, &ns->list, mnt_list) {
776 if (mnt_is_cursor(mnt))
778 is_covered = (mnt->mnt_mountpoint == dentry);
783 up_read(&namespace_sem);
788 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
790 struct hlist_head *chain = mp_hash(dentry);
791 struct mountpoint *mp;
793 hlist_for_each_entry(mp, chain, m_hash) {
794 if (mp->m_dentry == dentry) {
802 static struct mountpoint *get_mountpoint(struct dentry *dentry)
804 struct mountpoint *mp, *new = NULL;
807 if (d_mountpoint(dentry)) {
808 /* might be worth a WARN_ON() */
809 if (d_unlinked(dentry))
810 return ERR_PTR(-ENOENT);
812 read_seqlock_excl(&mount_lock);
813 mp = lookup_mountpoint(dentry);
814 read_sequnlock_excl(&mount_lock);
820 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
822 return ERR_PTR(-ENOMEM);
825 /* Exactly one processes may set d_mounted */
826 ret = d_set_mounted(dentry);
828 /* Someone else set d_mounted? */
832 /* The dentry is not available as a mountpoint? */
837 /* Add the new mountpoint to the hash table */
838 read_seqlock_excl(&mount_lock);
839 new->m_dentry = dget(dentry);
841 hlist_add_head(&new->m_hash, mp_hash(dentry));
842 INIT_HLIST_HEAD(&new->m_list);
843 read_sequnlock_excl(&mount_lock);
853 * vfsmount lock must be held. Additionally, the caller is responsible
854 * for serializing calls for given disposal list.
856 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
858 if (!--mp->m_count) {
859 struct dentry *dentry = mp->m_dentry;
860 BUG_ON(!hlist_empty(&mp->m_list));
861 spin_lock(&dentry->d_lock);
862 dentry->d_flags &= ~DCACHE_MOUNTED;
863 spin_unlock(&dentry->d_lock);
864 dput_to_list(dentry, list);
865 hlist_del(&mp->m_hash);
870 /* called with namespace_lock and vfsmount lock */
871 static void put_mountpoint(struct mountpoint *mp)
873 __put_mountpoint(mp, &ex_mountpoints);
876 static inline int check_mnt(struct mount *mnt)
878 return mnt->mnt_ns == current->nsproxy->mnt_ns;
882 * vfsmount lock must be held for write
884 static void touch_mnt_namespace(struct mnt_namespace *ns)
888 wake_up_interruptible(&ns->poll);
893 * vfsmount lock must be held for write
895 static void __touch_mnt_namespace(struct mnt_namespace *ns)
897 if (ns && ns->event != event) {
899 wake_up_interruptible(&ns->poll);
904 * vfsmount lock must be held for write
906 static struct mountpoint *unhash_mnt(struct mount *mnt)
908 struct mountpoint *mp;
909 mnt->mnt_parent = mnt;
910 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
911 list_del_init(&mnt->mnt_child);
912 hlist_del_init_rcu(&mnt->mnt_hash);
913 hlist_del_init(&mnt->mnt_mp_list);
920 * vfsmount lock must be held for write
922 static void umount_mnt(struct mount *mnt)
924 put_mountpoint(unhash_mnt(mnt));
928 * vfsmount lock must be held for write
930 void mnt_set_mountpoint(struct mount *mnt,
931 struct mountpoint *mp,
932 struct mount *child_mnt)
935 mnt_add_count(mnt, 1); /* essentially, that's mntget */
936 child_mnt->mnt_mountpoint = mp->m_dentry;
937 child_mnt->mnt_parent = mnt;
938 child_mnt->mnt_mp = mp;
939 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
943 * mnt_set_mountpoint_beneath - mount a mount beneath another one
945 * @new_parent: the source mount
946 * @top_mnt: the mount beneath which @new_parent is mounted
947 * @new_mp: the new mountpoint of @top_mnt on @new_parent
949 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
950 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
951 * @new_mp. And mount @new_parent on the old parent and old
952 * mountpoint of @top_mnt.
954 * Context: This function expects namespace_lock() and lock_mount_hash()
955 * to have been acquired in that order.
957 static void mnt_set_mountpoint_beneath(struct mount *new_parent,
958 struct mount *top_mnt,
959 struct mountpoint *new_mp)
961 struct mount *old_top_parent = top_mnt->mnt_parent;
962 struct mountpoint *old_top_mp = top_mnt->mnt_mp;
964 mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
965 mnt_change_mountpoint(new_parent, new_mp, top_mnt);
969 static void __attach_mnt(struct mount *mnt, struct mount *parent)
971 hlist_add_head_rcu(&mnt->mnt_hash,
972 m_hash(&parent->mnt, mnt->mnt_mountpoint));
973 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
977 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
978 * list of child mounts
979 * @parent: the parent
980 * @mnt: the new mount
981 * @mp: the new mountpoint
982 * @beneath: whether to mount @mnt beneath or on top of @parent
984 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
985 * to @parent's child mount list and to @mount_hashtable.
987 * If @beneath is true, remove @mnt from its current parent and
988 * mountpoint and mount it on @mp on @parent, and mount @parent on the
989 * old parent and old mountpoint of @mnt. Finally, attach @parent to
990 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
992 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
993 * to the correct parent.
995 * Context: This function expects namespace_lock() and lock_mount_hash()
996 * to have been acquired in that order.
998 static void attach_mnt(struct mount *mnt, struct mount *parent,
999 struct mountpoint *mp, bool beneath)
1002 mnt_set_mountpoint_beneath(mnt, parent, mp);
1004 mnt_set_mountpoint(parent, mp, mnt);
1006 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
1007 * beneath @parent then @mnt will need to be attached to
1008 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
1009 * isn't the same mount as @parent.
1011 __attach_mnt(mnt, mnt->mnt_parent);
1014 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
1016 struct mountpoint *old_mp = mnt->mnt_mp;
1017 struct mount *old_parent = mnt->mnt_parent;
1019 list_del_init(&mnt->mnt_child);
1020 hlist_del_init(&mnt->mnt_mp_list);
1021 hlist_del_init_rcu(&mnt->mnt_hash);
1023 attach_mnt(mnt, parent, mp, false);
1025 put_mountpoint(old_mp);
1026 mnt_add_count(old_parent, -1);
1030 * vfsmount lock must be held for write
1032 static void commit_tree(struct mount *mnt)
1034 struct mount *parent = mnt->mnt_parent;
1037 struct mnt_namespace *n = parent->mnt_ns;
1039 BUG_ON(parent == mnt);
1041 list_add_tail(&head, &mnt->mnt_list);
1042 list_for_each_entry(m, &head, mnt_list)
1045 list_splice(&head, n->list.prev);
1047 n->mounts += n->pending_mounts;
1048 n->pending_mounts = 0;
1050 __attach_mnt(mnt, parent);
1051 touch_mnt_namespace(n);
1054 static struct mount *next_mnt(struct mount *p, struct mount *root)
1056 struct list_head *next = p->mnt_mounts.next;
1057 if (next == &p->mnt_mounts) {
1061 next = p->mnt_child.next;
1062 if (next != &p->mnt_parent->mnt_mounts)
1067 return list_entry(next, struct mount, mnt_child);
1070 static struct mount *skip_mnt_tree(struct mount *p)
1072 struct list_head *prev = p->mnt_mounts.prev;
1073 while (prev != &p->mnt_mounts) {
1074 p = list_entry(prev, struct mount, mnt_child);
1075 prev = p->mnt_mounts.prev;
1081 * vfs_create_mount - Create a mount for a configured superblock
1082 * @fc: The configuration context with the superblock attached
1084 * Create a mount to an already configured superblock. If necessary, the
1085 * caller should invoke vfs_get_tree() before calling this.
1087 * Note that this does not attach the mount to anything.
1089 struct vfsmount *vfs_create_mount(struct fs_context *fc)
1094 return ERR_PTR(-EINVAL);
1096 mnt = alloc_vfsmnt(fc->source ?: "none");
1098 return ERR_PTR(-ENOMEM);
1100 if (fc->sb_flags & SB_KERNMOUNT)
1101 mnt->mnt.mnt_flags = MNT_INTERNAL;
1103 atomic_inc(&fc->root->d_sb->s_active);
1104 mnt->mnt.mnt_sb = fc->root->d_sb;
1105 mnt->mnt.mnt_root = dget(fc->root);
1106 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1107 mnt->mnt_parent = mnt;
1110 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1111 unlock_mount_hash();
1114 EXPORT_SYMBOL(vfs_create_mount);
1116 struct vfsmount *fc_mount(struct fs_context *fc)
1118 int err = vfs_get_tree(fc);
1120 up_write(&fc->root->d_sb->s_umount);
1121 return vfs_create_mount(fc);
1123 return ERR_PTR(err);
1125 EXPORT_SYMBOL(fc_mount);
1127 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1128 int flags, const char *name,
1131 struct fs_context *fc;
1132 struct vfsmount *mnt;
1136 return ERR_PTR(-EINVAL);
1138 fc = fs_context_for_mount(type, flags);
1140 return ERR_CAST(fc);
1143 ret = vfs_parse_fs_string(fc, "source",
1144 name, strlen(name));
1146 ret = parse_monolithic_mount_data(fc, data);
1155 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1158 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1159 const char *name, void *data)
1161 /* Until it is worked out how to pass the user namespace
1162 * through from the parent mount to the submount don't support
1163 * unprivileged mounts with submounts.
1165 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1166 return ERR_PTR(-EPERM);
1168 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1170 EXPORT_SYMBOL_GPL(vfs_submount);
1172 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1175 struct super_block *sb = old->mnt.mnt_sb;
1179 mnt = alloc_vfsmnt(old->mnt_devname);
1181 return ERR_PTR(-ENOMEM);
1183 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1184 mnt->mnt_group_id = 0; /* not a peer of original */
1186 mnt->mnt_group_id = old->mnt_group_id;
1188 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1189 err = mnt_alloc_group_id(mnt);
1194 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1195 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1197 atomic_inc(&sb->s_active);
1198 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1200 mnt->mnt.mnt_sb = sb;
1201 mnt->mnt.mnt_root = dget(root);
1202 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1203 mnt->mnt_parent = mnt;
1205 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1206 unlock_mount_hash();
1208 if ((flag & CL_SLAVE) ||
1209 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1210 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1211 mnt->mnt_master = old;
1212 CLEAR_MNT_SHARED(mnt);
1213 } else if (!(flag & CL_PRIVATE)) {
1214 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1215 list_add(&mnt->mnt_share, &old->mnt_share);
1216 if (IS_MNT_SLAVE(old))
1217 list_add(&mnt->mnt_slave, &old->mnt_slave);
1218 mnt->mnt_master = old->mnt_master;
1220 CLEAR_MNT_SHARED(mnt);
1222 if (flag & CL_MAKE_SHARED)
1223 set_mnt_shared(mnt);
1225 /* stick the duplicate mount on the same expiry list
1226 * as the original if that was on one */
1227 if (flag & CL_EXPIRE) {
1228 if (!list_empty(&old->mnt_expire))
1229 list_add(&mnt->mnt_expire, &old->mnt_expire);
1237 return ERR_PTR(err);
1240 static void cleanup_mnt(struct mount *mnt)
1242 struct hlist_node *p;
1245 * The warning here probably indicates that somebody messed
1246 * up a mnt_want/drop_write() pair. If this happens, the
1247 * filesystem was probably unable to make r/w->r/o transitions.
1248 * The locking used to deal with mnt_count decrement provides barriers,
1249 * so mnt_get_writers() below is safe.
1251 WARN_ON(mnt_get_writers(mnt));
1252 if (unlikely(mnt->mnt_pins.first))
1254 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1255 hlist_del(&m->mnt_umount);
1258 fsnotify_vfsmount_delete(&mnt->mnt);
1259 dput(mnt->mnt.mnt_root);
1260 deactivate_super(mnt->mnt.mnt_sb);
1262 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1265 static void __cleanup_mnt(struct rcu_head *head)
1267 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1270 static LLIST_HEAD(delayed_mntput_list);
1271 static void delayed_mntput(struct work_struct *unused)
1273 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1274 struct mount *m, *t;
1276 llist_for_each_entry_safe(m, t, node, mnt_llist)
1279 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1281 static void mntput_no_expire(struct mount *mnt)
1287 if (likely(READ_ONCE(mnt->mnt_ns))) {
1289 * Since we don't do lock_mount_hash() here,
1290 * ->mnt_ns can change under us. However, if it's
1291 * non-NULL, then there's a reference that won't
1292 * be dropped until after an RCU delay done after
1293 * turning ->mnt_ns NULL. So if we observe it
1294 * non-NULL under rcu_read_lock(), the reference
1295 * we are dropping is not the final one.
1297 mnt_add_count(mnt, -1);
1303 * make sure that if __legitimize_mnt() has not seen us grab
1304 * mount_lock, we'll see their refcount increment here.
1307 mnt_add_count(mnt, -1);
1308 count = mnt_get_count(mnt);
1312 unlock_mount_hash();
1315 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1317 unlock_mount_hash();
1320 mnt->mnt.mnt_flags |= MNT_DOOMED;
1323 list_del(&mnt->mnt_instance);
1325 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1326 struct mount *p, *tmp;
1327 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1328 __put_mountpoint(unhash_mnt(p), &list);
1329 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1332 unlock_mount_hash();
1333 shrink_dentry_list(&list);
1335 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1336 struct task_struct *task = current;
1337 if (likely(!(task->flags & PF_KTHREAD))) {
1338 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1339 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1342 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1343 schedule_delayed_work(&delayed_mntput_work, 1);
1349 void mntput(struct vfsmount *mnt)
1352 struct mount *m = real_mount(mnt);
1353 /* avoid cacheline pingpong */
1354 if (unlikely(m->mnt_expiry_mark))
1355 WRITE_ONCE(m->mnt_expiry_mark, 0);
1356 mntput_no_expire(m);
1359 EXPORT_SYMBOL(mntput);
1361 struct vfsmount *mntget(struct vfsmount *mnt)
1364 mnt_add_count(real_mount(mnt), 1);
1367 EXPORT_SYMBOL(mntget);
1370 * Make a mount point inaccessible to new lookups.
1371 * Because there may still be current users, the caller MUST WAIT
1372 * for an RCU grace period before destroying the mount point.
1374 void mnt_make_shortterm(struct vfsmount *mnt)
1377 real_mount(mnt)->mnt_ns = NULL;
1381 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1382 * @path: path to check
1384 * d_mountpoint() can only be used reliably to establish if a dentry is
1385 * not mounted in any namespace and that common case is handled inline.
1386 * d_mountpoint() isn't aware of the possibility there may be multiple
1387 * mounts using a given dentry in a different namespace. This function
1388 * checks if the passed in path is a mountpoint rather than the dentry
1391 bool path_is_mountpoint(const struct path *path)
1396 if (!d_mountpoint(path->dentry))
1401 seq = read_seqbegin(&mount_lock);
1402 res = __path_is_mountpoint(path);
1403 } while (read_seqretry(&mount_lock, seq));
1408 EXPORT_SYMBOL(path_is_mountpoint);
1410 struct vfsmount *mnt_clone_internal(const struct path *path)
1413 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1416 p->mnt.mnt_flags |= MNT_INTERNAL;
1420 #ifdef CONFIG_PROC_FS
1421 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1422 struct list_head *p)
1424 struct mount *mnt, *ret = NULL;
1427 list_for_each_continue(p, &ns->list) {
1428 mnt = list_entry(p, typeof(*mnt), mnt_list);
1429 if (!mnt_is_cursor(mnt)) {
1439 /* iterator; we want it to have access to namespace_sem, thus here... */
1440 static void *m_start(struct seq_file *m, loff_t *pos)
1442 struct proc_mounts *p = m->private;
1443 struct list_head *prev;
1445 down_read(&namespace_sem);
1447 prev = &p->ns->list;
1449 prev = &p->cursor.mnt_list;
1451 /* Read after we'd reached the end? */
1452 if (list_empty(prev))
1456 return mnt_list_next(p->ns, prev);
1459 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1461 struct proc_mounts *p = m->private;
1462 struct mount *mnt = v;
1465 return mnt_list_next(p->ns, &mnt->mnt_list);
1468 static void m_stop(struct seq_file *m, void *v)
1470 struct proc_mounts *p = m->private;
1471 struct mount *mnt = v;
1473 lock_ns_list(p->ns);
1475 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1477 list_del_init(&p->cursor.mnt_list);
1478 unlock_ns_list(p->ns);
1479 up_read(&namespace_sem);
1482 static int m_show(struct seq_file *m, void *v)
1484 struct proc_mounts *p = m->private;
1485 struct mount *r = v;
1486 return p->show(m, &r->mnt);
1489 const struct seq_operations mounts_op = {
1496 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1498 down_read(&namespace_sem);
1500 list_del(&cursor->mnt_list);
1502 up_read(&namespace_sem);
1504 #endif /* CONFIG_PROC_FS */
1507 * may_umount_tree - check if a mount tree is busy
1508 * @m: root of mount tree
1510 * This is called to check if a tree of mounts has any
1511 * open files, pwds, chroots or sub mounts that are
1514 int may_umount_tree(struct vfsmount *m)
1516 struct mount *mnt = real_mount(m);
1517 int actual_refs = 0;
1518 int minimum_refs = 0;
1522 /* write lock needed for mnt_get_count */
1524 for (p = mnt; p; p = next_mnt(p, mnt)) {
1525 actual_refs += mnt_get_count(p);
1528 unlock_mount_hash();
1530 if (actual_refs > minimum_refs)
1536 EXPORT_SYMBOL(may_umount_tree);
1539 * may_umount - check if a mount point is busy
1540 * @mnt: root of mount
1542 * This is called to check if a mount point has any
1543 * open files, pwds, chroots or sub mounts. If the
1544 * mount has sub mounts this will return busy
1545 * regardless of whether the sub mounts are busy.
1547 * Doesn't take quota and stuff into account. IOW, in some cases it will
1548 * give false negatives. The main reason why it's here is that we need
1549 * a non-destructive way to look for easily umountable filesystems.
1551 int may_umount(struct vfsmount *mnt)
1554 down_read(&namespace_sem);
1556 if (propagate_mount_busy(real_mount(mnt), 2))
1558 unlock_mount_hash();
1559 up_read(&namespace_sem);
1563 EXPORT_SYMBOL(may_umount);
1565 static void namespace_unlock(void)
1567 struct hlist_head head;
1568 struct hlist_node *p;
1572 hlist_move_list(&unmounted, &head);
1573 list_splice_init(&ex_mountpoints, &list);
1575 up_write(&namespace_sem);
1577 shrink_dentry_list(&list);
1579 if (likely(hlist_empty(&head)))
1582 synchronize_rcu_expedited();
1584 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1585 hlist_del(&m->mnt_umount);
1590 static inline void namespace_lock(void)
1592 down_write(&namespace_sem);
1595 enum umount_tree_flags {
1597 UMOUNT_PROPAGATE = 2,
1598 UMOUNT_CONNECTED = 4,
1601 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1603 /* Leaving mounts connected is only valid for lazy umounts */
1604 if (how & UMOUNT_SYNC)
1607 /* A mount without a parent has nothing to be connected to */
1608 if (!mnt_has_parent(mnt))
1611 /* Because the reference counting rules change when mounts are
1612 * unmounted and connected, umounted mounts may not be
1613 * connected to mounted mounts.
1615 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1618 /* Has it been requested that the mount remain connected? */
1619 if (how & UMOUNT_CONNECTED)
1622 /* Is the mount locked such that it needs to remain connected? */
1623 if (IS_MNT_LOCKED(mnt))
1626 /* By default disconnect the mount */
1631 * mount_lock must be held
1632 * namespace_sem must be held for write
1634 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1636 LIST_HEAD(tmp_list);
1639 if (how & UMOUNT_PROPAGATE)
1640 propagate_mount_unlock(mnt);
1642 /* Gather the mounts to umount */
1643 for (p = mnt; p; p = next_mnt(p, mnt)) {
1644 p->mnt.mnt_flags |= MNT_UMOUNT;
1645 list_move(&p->mnt_list, &tmp_list);
1648 /* Hide the mounts from mnt_mounts */
1649 list_for_each_entry(p, &tmp_list, mnt_list) {
1650 list_del_init(&p->mnt_child);
1653 /* Add propogated mounts to the tmp_list */
1654 if (how & UMOUNT_PROPAGATE)
1655 propagate_umount(&tmp_list);
1657 while (!list_empty(&tmp_list)) {
1658 struct mnt_namespace *ns;
1660 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1661 list_del_init(&p->mnt_expire);
1662 list_del_init(&p->mnt_list);
1666 __touch_mnt_namespace(ns);
1669 if (how & UMOUNT_SYNC)
1670 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1672 disconnect = disconnect_mount(p, how);
1673 if (mnt_has_parent(p)) {
1674 mnt_add_count(p->mnt_parent, -1);
1676 /* Don't forget about p */
1677 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1682 change_mnt_propagation(p, MS_PRIVATE);
1684 hlist_add_head(&p->mnt_umount, &unmounted);
1688 static void shrink_submounts(struct mount *mnt);
1690 static int do_umount_root(struct super_block *sb)
1694 down_write(&sb->s_umount);
1695 if (!sb_rdonly(sb)) {
1696 struct fs_context *fc;
1698 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1703 ret = parse_monolithic_mount_data(fc, NULL);
1705 ret = reconfigure_super(fc);
1709 up_write(&sb->s_umount);
1713 static int do_umount(struct mount *mnt, int flags)
1715 struct super_block *sb = mnt->mnt.mnt_sb;
1718 retval = security_sb_umount(&mnt->mnt, flags);
1723 * Allow userspace to request a mountpoint be expired rather than
1724 * unmounting unconditionally. Unmount only happens if:
1725 * (1) the mark is already set (the mark is cleared by mntput())
1726 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1728 if (flags & MNT_EXPIRE) {
1729 if (&mnt->mnt == current->fs->root.mnt ||
1730 flags & (MNT_FORCE | MNT_DETACH))
1734 * probably don't strictly need the lock here if we examined
1735 * all race cases, but it's a slowpath.
1738 if (mnt_get_count(mnt) != 2) {
1739 unlock_mount_hash();
1742 unlock_mount_hash();
1744 if (!xchg(&mnt->mnt_expiry_mark, 1))
1749 * If we may have to abort operations to get out of this
1750 * mount, and they will themselves hold resources we must
1751 * allow the fs to do things. In the Unix tradition of
1752 * 'Gee thats tricky lets do it in userspace' the umount_begin
1753 * might fail to complete on the first run through as other tasks
1754 * must return, and the like. Thats for the mount program to worry
1755 * about for the moment.
1758 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1759 sb->s_op->umount_begin(sb);
1763 * No sense to grab the lock for this test, but test itself looks
1764 * somewhat bogus. Suggestions for better replacement?
1765 * Ho-hum... In principle, we might treat that as umount + switch
1766 * to rootfs. GC would eventually take care of the old vfsmount.
1767 * Actually it makes sense, especially if rootfs would contain a
1768 * /reboot - static binary that would close all descriptors and
1769 * call reboot(9). Then init(8) could umount root and exec /reboot.
1771 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1773 * Special case for "unmounting" root ...
1774 * we just try to remount it readonly.
1776 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1778 return do_umount_root(sb);
1784 /* Recheck MNT_LOCKED with the locks held */
1786 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1790 if (flags & MNT_DETACH) {
1791 if (!list_empty(&mnt->mnt_list))
1792 umount_tree(mnt, UMOUNT_PROPAGATE);
1795 shrink_submounts(mnt);
1797 if (!propagate_mount_busy(mnt, 2)) {
1798 if (!list_empty(&mnt->mnt_list))
1799 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1804 unlock_mount_hash();
1810 * __detach_mounts - lazily unmount all mounts on the specified dentry
1812 * During unlink, rmdir, and d_drop it is possible to loose the path
1813 * to an existing mountpoint, and wind up leaking the mount.
1814 * detach_mounts allows lazily unmounting those mounts instead of
1817 * The caller may hold dentry->d_inode->i_mutex.
1819 void __detach_mounts(struct dentry *dentry)
1821 struct mountpoint *mp;
1826 mp = lookup_mountpoint(dentry);
1831 while (!hlist_empty(&mp->m_list)) {
1832 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1833 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1835 hlist_add_head(&mnt->mnt_umount, &unmounted);
1837 else umount_tree(mnt, UMOUNT_CONNECTED);
1841 unlock_mount_hash();
1846 * Is the caller allowed to modify his namespace?
1848 bool may_mount(void)
1850 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1854 * path_mounted - check whether path is mounted
1855 * @path: path to check
1857 * Determine whether @path refers to the root of a mount.
1859 * Return: true if @path is the root of a mount, false if not.
1861 static inline bool path_mounted(const struct path *path)
1863 return path->mnt->mnt_root == path->dentry;
1866 static void warn_mandlock(void)
1868 pr_warn_once("=======================================================\n"
1869 "WARNING: The mand mount option has been deprecated and\n"
1870 " and is ignored by this kernel. Remove the mand\n"
1871 " option from the mount to silence this warning.\n"
1872 "=======================================================\n");
1875 static int can_umount(const struct path *path, int flags)
1877 struct mount *mnt = real_mount(path->mnt);
1881 if (!path_mounted(path))
1883 if (!check_mnt(mnt))
1885 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1887 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1892 // caller is responsible for flags being sane
1893 int path_umount(struct path *path, int flags)
1895 struct mount *mnt = real_mount(path->mnt);
1898 ret = can_umount(path, flags);
1900 ret = do_umount(mnt, flags);
1902 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1904 mntput_no_expire(mnt);
1908 static int ksys_umount(char __user *name, int flags)
1910 int lookup_flags = LOOKUP_MOUNTPOINT;
1914 // basic validity checks done first
1915 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1918 if (!(flags & UMOUNT_NOFOLLOW))
1919 lookup_flags |= LOOKUP_FOLLOW;
1920 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1923 return path_umount(&path, flags);
1926 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1928 return ksys_umount(name, flags);
1931 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1934 * The 2.0 compatible umount. No flags.
1936 SYSCALL_DEFINE1(oldumount, char __user *, name)
1938 return ksys_umount(name, 0);
1943 static bool is_mnt_ns_file(struct dentry *dentry)
1945 /* Is this a proxy for a mount namespace? */
1946 return dentry->d_op == &ns_dentry_operations &&
1947 dentry->d_fsdata == &mntns_operations;
1950 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1952 return container_of(ns, struct mnt_namespace, ns);
1955 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1960 static bool mnt_ns_loop(struct dentry *dentry)
1962 /* Could bind mounting the mount namespace inode cause a
1963 * mount namespace loop?
1965 struct mnt_namespace *mnt_ns;
1966 if (!is_mnt_ns_file(dentry))
1969 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1970 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1973 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1976 struct mount *res, *p, *q, *r, *parent;
1978 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1979 return ERR_PTR(-EINVAL);
1981 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1982 return ERR_PTR(-EINVAL);
1984 res = q = clone_mnt(mnt, dentry, flag);
1988 q->mnt_mountpoint = mnt->mnt_mountpoint;
1991 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1993 if (!is_subdir(r->mnt_mountpoint, dentry))
1996 for (s = r; s; s = next_mnt(s, r)) {
1997 if (!(flag & CL_COPY_UNBINDABLE) &&
1998 IS_MNT_UNBINDABLE(s)) {
1999 if (s->mnt.mnt_flags & MNT_LOCKED) {
2000 /* Both unbindable and locked. */
2001 q = ERR_PTR(-EPERM);
2004 s = skip_mnt_tree(s);
2008 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2009 is_mnt_ns_file(s->mnt.mnt_root)) {
2010 s = skip_mnt_tree(s);
2013 while (p != s->mnt_parent) {
2019 q = clone_mnt(p, p->mnt.mnt_root, flag);
2023 list_add_tail(&q->mnt_list, &res->mnt_list);
2024 attach_mnt(q, parent, p->mnt_mp, false);
2025 unlock_mount_hash();
2032 umount_tree(res, UMOUNT_SYNC);
2033 unlock_mount_hash();
2038 /* Caller should check returned pointer for errors */
2040 struct vfsmount *collect_mounts(const struct path *path)
2044 if (!check_mnt(real_mount(path->mnt)))
2045 tree = ERR_PTR(-EINVAL);
2047 tree = copy_tree(real_mount(path->mnt), path->dentry,
2048 CL_COPY_ALL | CL_PRIVATE);
2051 return ERR_CAST(tree);
2055 static void free_mnt_ns(struct mnt_namespace *);
2056 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2058 void dissolve_on_fput(struct vfsmount *mnt)
2060 struct mnt_namespace *ns;
2063 ns = real_mount(mnt)->mnt_ns;
2066 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2070 unlock_mount_hash();
2076 void drop_collected_mounts(struct vfsmount *mnt)
2080 umount_tree(real_mount(mnt), 0);
2081 unlock_mount_hash();
2085 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2087 struct mount *child;
2089 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2090 if (!is_subdir(child->mnt_mountpoint, dentry))
2093 if (child->mnt.mnt_flags & MNT_LOCKED)
2100 * clone_private_mount - create a private clone of a path
2101 * @path: path to clone
2103 * This creates a new vfsmount, which will be the clone of @path. The new mount
2104 * will not be attached anywhere in the namespace and will be private (i.e.
2105 * changes to the originating mount won't be propagated into this).
2107 * Release with mntput().
2109 struct vfsmount *clone_private_mount(const struct path *path)
2111 struct mount *old_mnt = real_mount(path->mnt);
2112 struct mount *new_mnt;
2114 down_read(&namespace_sem);
2115 if (IS_MNT_UNBINDABLE(old_mnt))
2118 if (!check_mnt(old_mnt))
2121 if (has_locked_children(old_mnt, path->dentry))
2124 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2125 up_read(&namespace_sem);
2127 if (IS_ERR(new_mnt))
2128 return ERR_CAST(new_mnt);
2130 /* Longterm mount to be removed by kern_unmount*() */
2131 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2133 return &new_mnt->mnt;
2136 up_read(&namespace_sem);
2137 return ERR_PTR(-EINVAL);
2139 EXPORT_SYMBOL_GPL(clone_private_mount);
2141 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2142 struct vfsmount *root)
2145 int res = f(root, arg);
2148 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2149 res = f(&mnt->mnt, arg);
2156 static void lock_mnt_tree(struct mount *mnt)
2160 for (p = mnt; p; p = next_mnt(p, mnt)) {
2161 int flags = p->mnt.mnt_flags;
2162 /* Don't allow unprivileged users to change mount flags */
2163 flags |= MNT_LOCK_ATIME;
2165 if (flags & MNT_READONLY)
2166 flags |= MNT_LOCK_READONLY;
2168 if (flags & MNT_NODEV)
2169 flags |= MNT_LOCK_NODEV;
2171 if (flags & MNT_NOSUID)
2172 flags |= MNT_LOCK_NOSUID;
2174 if (flags & MNT_NOEXEC)
2175 flags |= MNT_LOCK_NOEXEC;
2176 /* Don't allow unprivileged users to reveal what is under a mount */
2177 if (list_empty(&p->mnt_expire))
2178 flags |= MNT_LOCKED;
2179 p->mnt.mnt_flags = flags;
2183 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2187 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2188 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2189 mnt_release_group_id(p);
2193 static int invent_group_ids(struct mount *mnt, bool recurse)
2197 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2198 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2199 int err = mnt_alloc_group_id(p);
2201 cleanup_group_ids(mnt, p);
2210 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2212 unsigned int max = READ_ONCE(sysctl_mount_max);
2213 unsigned int mounts = 0;
2216 if (ns->mounts >= max)
2219 if (ns->pending_mounts >= max)
2221 max -= ns->pending_mounts;
2223 for (p = mnt; p; p = next_mnt(p, mnt))
2229 ns->pending_mounts += mounts;
2233 enum mnt_tree_flags_t {
2234 MNT_TREE_MOVE = BIT(0),
2235 MNT_TREE_BENEATH = BIT(1),
2239 * attach_recursive_mnt - attach a source mount tree
2240 * @source_mnt: mount tree to be attached
2241 * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath
2242 * @dest_mp: the mountpoint @source_mnt will be mounted at
2243 * @flags: modify how @source_mnt is supposed to be attached
2245 * NOTE: in the table below explains the semantics when a source mount
2246 * of a given type is attached to a destination mount of a given type.
2247 * ---------------------------------------------------------------------------
2248 * | BIND MOUNT OPERATION |
2249 * |**************************************************************************
2250 * | source-->| shared | private | slave | unbindable |
2254 * |**************************************************************************
2255 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2257 * |non-shared| shared (+) | private | slave (*) | invalid |
2258 * ***************************************************************************
2259 * A bind operation clones the source mount and mounts the clone on the
2260 * destination mount.
2262 * (++) the cloned mount is propagated to all the mounts in the propagation
2263 * tree of the destination mount and the cloned mount is added to
2264 * the peer group of the source mount.
2265 * (+) the cloned mount is created under the destination mount and is marked
2266 * as shared. The cloned mount is added to the peer group of the source
2268 * (+++) the mount is propagated to all the mounts in the propagation tree
2269 * of the destination mount and the cloned mount is made slave
2270 * of the same master as that of the source mount. The cloned mount
2271 * is marked as 'shared and slave'.
2272 * (*) the cloned mount is made a slave of the same master as that of the
2275 * ---------------------------------------------------------------------------
2276 * | MOVE MOUNT OPERATION |
2277 * |**************************************************************************
2278 * | source-->| shared | private | slave | unbindable |
2282 * |**************************************************************************
2283 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2285 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2286 * ***************************************************************************
2288 * (+) the mount is moved to the destination. And is then propagated to
2289 * all the mounts in the propagation tree of the destination mount.
2290 * (+*) the mount is moved to the destination.
2291 * (+++) the mount is moved to the destination and is then propagated to
2292 * all the mounts belonging to the destination mount's propagation tree.
2293 * the mount is marked as 'shared and slave'.
2294 * (*) the mount continues to be a slave at the new location.
2296 * if the source mount is a tree, the operations explained above is
2297 * applied to each mount in the tree.
2298 * Must be called without spinlocks held, since this function can sleep
2301 * Context: The function expects namespace_lock() to be held.
2302 * Return: If @source_mnt was successfully attached 0 is returned.
2303 * Otherwise a negative error code is returned.
2305 static int attach_recursive_mnt(struct mount *source_mnt,
2306 struct mount *top_mnt,
2307 struct mountpoint *dest_mp,
2308 enum mnt_tree_flags_t flags)
2310 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2311 HLIST_HEAD(tree_list);
2312 struct mnt_namespace *ns = top_mnt->mnt_ns;
2313 struct mountpoint *smp;
2314 struct mount *child, *dest_mnt, *p;
2315 struct hlist_node *n;
2317 bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
2320 * Preallocate a mountpoint in case the new mounts need to be
2321 * mounted beneath mounts on the same mountpoint.
2323 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2325 return PTR_ERR(smp);
2327 /* Is there space to add these mounts to the mount namespace? */
2329 err = count_mounts(ns, source_mnt);
2335 dest_mnt = top_mnt->mnt_parent;
2339 if (IS_MNT_SHARED(dest_mnt)) {
2340 err = invent_group_ids(source_mnt, true);
2343 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2347 goto out_cleanup_ids;
2349 if (IS_MNT_SHARED(dest_mnt)) {
2350 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2357 unhash_mnt(source_mnt);
2358 attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
2359 touch_mnt_namespace(source_mnt->mnt_ns);
2361 if (source_mnt->mnt_ns) {
2362 /* move from anon - the caller will destroy */
2363 list_del_init(&source_mnt->mnt_ns->list);
2366 mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
2368 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2369 commit_tree(source_mnt);
2372 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2374 hlist_del_init(&child->mnt_hash);
2375 q = __lookup_mnt(&child->mnt_parent->mnt,
2376 child->mnt_mountpoint);
2378 mnt_change_mountpoint(child, smp, q);
2379 /* Notice when we are propagating across user namespaces */
2380 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2381 lock_mnt_tree(child);
2382 child->mnt.mnt_flags &= ~MNT_LOCKED;
2385 put_mountpoint(smp);
2386 unlock_mount_hash();
2391 while (!hlist_empty(&tree_list)) {
2392 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2393 child->mnt_parent->mnt_ns->pending_mounts = 0;
2394 umount_tree(child, UMOUNT_SYNC);
2396 unlock_mount_hash();
2397 cleanup_group_ids(source_mnt, NULL);
2399 ns->pending_mounts = 0;
2401 read_seqlock_excl(&mount_lock);
2402 put_mountpoint(smp);
2403 read_sequnlock_excl(&mount_lock);
2409 * do_lock_mount - lock mount and mountpoint
2410 * @path: target path
2411 * @beneath: whether the intention is to mount beneath @path
2413 * Follow the mount stack on @path until the top mount @mnt is found. If
2414 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2415 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2416 * until nothing is stacked on top of it anymore.
2418 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2419 * against concurrent removal of the new mountpoint from another mount
2422 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2423 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2424 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2425 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2426 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2427 * on top of it for @beneath.
2429 * In addition, @beneath needs to make sure that @mnt hasn't been
2430 * unmounted or moved from its current mountpoint in between dropping
2431 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2432 * being unmounted would be detected later by e.g., calling
2433 * check_mnt(mnt) in the function it's called from. For the @beneath
2434 * case however, it's useful to detect it directly in do_lock_mount().
2435 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2436 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2437 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2439 * Return: Either the target mountpoint on the top mount or the top
2440 * mount's mountpoint.
2442 static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
2444 struct vfsmount *mnt = path->mnt;
2445 struct dentry *dentry;
2446 struct mountpoint *mp = ERR_PTR(-ENOENT);
2452 m = real_mount(mnt);
2453 read_seqlock_excl(&mount_lock);
2454 dentry = dget(m->mnt_mountpoint);
2455 read_sequnlock_excl(&mount_lock);
2457 dentry = path->dentry;
2460 inode_lock(dentry->d_inode);
2461 if (unlikely(cant_mount(dentry))) {
2462 inode_unlock(dentry->d_inode);
2468 if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) {
2470 inode_unlock(dentry->d_inode);
2474 mnt = lookup_mnt(path);
2479 inode_unlock(dentry->d_inode);
2484 path->dentry = dget(mnt->mnt_root);
2487 mp = get_mountpoint(dentry);
2490 inode_unlock(dentry->d_inode);
2500 static inline struct mountpoint *lock_mount(struct path *path)
2502 return do_lock_mount(path, false);
2505 static void unlock_mount(struct mountpoint *where)
2507 struct dentry *dentry = where->m_dentry;
2509 read_seqlock_excl(&mount_lock);
2510 put_mountpoint(where);
2511 read_sequnlock_excl(&mount_lock);
2514 inode_unlock(dentry->d_inode);
2517 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2519 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2522 if (d_is_dir(mp->m_dentry) !=
2523 d_is_dir(mnt->mnt.mnt_root))
2526 return attach_recursive_mnt(mnt, p, mp, 0);
2530 * Sanity check the flags to change_mnt_propagation.
2533 static int flags_to_propagation_type(int ms_flags)
2535 int type = ms_flags & ~(MS_REC | MS_SILENT);
2537 /* Fail if any non-propagation flags are set */
2538 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2540 /* Only one propagation flag should be set */
2541 if (!is_power_of_2(type))
2547 * recursively change the type of the mountpoint.
2549 static int do_change_type(struct path *path, int ms_flags)
2552 struct mount *mnt = real_mount(path->mnt);
2553 int recurse = ms_flags & MS_REC;
2557 if (!path_mounted(path))
2560 type = flags_to_propagation_type(ms_flags);
2565 if (type == MS_SHARED) {
2566 err = invent_group_ids(mnt, recurse);
2572 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2573 change_mnt_propagation(m, type);
2574 unlock_mount_hash();
2581 static struct mount *__do_loopback(struct path *old_path, int recurse)
2583 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2585 if (IS_MNT_UNBINDABLE(old))
2588 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2591 if (!recurse && has_locked_children(old, old_path->dentry))
2595 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2597 mnt = clone_mnt(old, old_path->dentry, 0);
2600 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2606 * do loopback mount.
2608 static int do_loopback(struct path *path, const char *old_name,
2611 struct path old_path;
2612 struct mount *mnt = NULL, *parent;
2613 struct mountpoint *mp;
2615 if (!old_name || !*old_name)
2617 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2622 if (mnt_ns_loop(old_path.dentry))
2625 mp = lock_mount(path);
2631 parent = real_mount(path->mnt);
2632 if (!check_mnt(parent))
2635 mnt = __do_loopback(&old_path, recurse);
2641 err = graft_tree(mnt, parent, mp);
2644 umount_tree(mnt, UMOUNT_SYNC);
2645 unlock_mount_hash();
2650 path_put(&old_path);
2654 static struct file *open_detached_copy(struct path *path, bool recursive)
2656 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2657 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2658 struct mount *mnt, *p;
2662 return ERR_CAST(ns);
2665 mnt = __do_loopback(path, recursive);
2669 return ERR_CAST(mnt);
2673 for (p = mnt; p; p = next_mnt(p, mnt)) {
2678 list_add_tail(&ns->list, &mnt->mnt_list);
2680 unlock_mount_hash();
2684 path->mnt = &mnt->mnt;
2685 file = dentry_open(path, O_PATH, current_cred());
2687 dissolve_on_fput(path->mnt);
2689 file->f_mode |= FMODE_NEED_UNMOUNT;
2693 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2697 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2698 bool detached = flags & OPEN_TREE_CLONE;
2702 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2704 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2705 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2709 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2712 if (flags & AT_NO_AUTOMOUNT)
2713 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2714 if (flags & AT_SYMLINK_NOFOLLOW)
2715 lookup_flags &= ~LOOKUP_FOLLOW;
2716 if (flags & AT_EMPTY_PATH)
2717 lookup_flags |= LOOKUP_EMPTY;
2719 if (detached && !may_mount())
2722 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2726 error = user_path_at(dfd, filename, lookup_flags, &path);
2727 if (unlikely(error)) {
2728 file = ERR_PTR(error);
2731 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2733 file = dentry_open(&path, O_PATH, current_cred());
2738 return PTR_ERR(file);
2740 fd_install(fd, file);
2745 * Don't allow locked mount flags to be cleared.
2747 * No locks need to be held here while testing the various MNT_LOCK
2748 * flags because those flags can never be cleared once they are set.
2750 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2752 unsigned int fl = mnt->mnt.mnt_flags;
2754 if ((fl & MNT_LOCK_READONLY) &&
2755 !(mnt_flags & MNT_READONLY))
2758 if ((fl & MNT_LOCK_NODEV) &&
2759 !(mnt_flags & MNT_NODEV))
2762 if ((fl & MNT_LOCK_NOSUID) &&
2763 !(mnt_flags & MNT_NOSUID))
2766 if ((fl & MNT_LOCK_NOEXEC) &&
2767 !(mnt_flags & MNT_NOEXEC))
2770 if ((fl & MNT_LOCK_ATIME) &&
2771 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2777 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2779 bool readonly_request = (mnt_flags & MNT_READONLY);
2781 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2784 if (readonly_request)
2785 return mnt_make_readonly(mnt);
2787 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2791 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2793 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2794 mnt->mnt.mnt_flags = mnt_flags;
2795 touch_mnt_namespace(mnt->mnt_ns);
2798 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2800 struct super_block *sb = mnt->mnt_sb;
2802 if (!__mnt_is_readonly(mnt) &&
2803 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2804 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2805 char *buf = (char *)__get_free_page(GFP_KERNEL);
2806 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2808 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2810 is_mounted(mnt) ? "remounted" : "mounted",
2811 mntpath, &sb->s_time_max,
2812 (unsigned long long)sb->s_time_max);
2814 free_page((unsigned long)buf);
2815 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2820 * Handle reconfiguration of the mountpoint only without alteration of the
2821 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2824 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2826 struct super_block *sb = path->mnt->mnt_sb;
2827 struct mount *mnt = real_mount(path->mnt);
2830 if (!check_mnt(mnt))
2833 if (!path_mounted(path))
2836 if (!can_change_locked_flags(mnt, mnt_flags))
2840 * We're only checking whether the superblock is read-only not
2841 * changing it, so only take down_read(&sb->s_umount).
2843 down_read(&sb->s_umount);
2845 ret = change_mount_ro_state(mnt, mnt_flags);
2847 set_mount_attributes(mnt, mnt_flags);
2848 unlock_mount_hash();
2849 up_read(&sb->s_umount);
2851 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2857 * change filesystem flags. dir should be a physical root of filesystem.
2858 * If you've mounted a non-root directory somewhere and want to do remount
2859 * on it - tough luck.
2861 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2862 int mnt_flags, void *data)
2865 struct super_block *sb = path->mnt->mnt_sb;
2866 struct mount *mnt = real_mount(path->mnt);
2867 struct fs_context *fc;
2869 if (!check_mnt(mnt))
2872 if (!path_mounted(path))
2875 if (!can_change_locked_flags(mnt, mnt_flags))
2878 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2883 err = parse_monolithic_mount_data(fc, data);
2885 down_write(&sb->s_umount);
2887 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2888 err = reconfigure_super(fc);
2891 set_mount_attributes(mnt, mnt_flags);
2892 unlock_mount_hash();
2895 up_write(&sb->s_umount);
2898 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2904 static inline int tree_contains_unbindable(struct mount *mnt)
2907 for (p = mnt; p; p = next_mnt(p, mnt)) {
2908 if (IS_MNT_UNBINDABLE(p))
2915 * Check that there aren't references to earlier/same mount namespaces in the
2916 * specified subtree. Such references can act as pins for mount namespaces
2917 * that aren't checked by the mount-cycle checking code, thereby allowing
2918 * cycles to be made.
2920 static bool check_for_nsfs_mounts(struct mount *subtree)
2926 for (p = subtree; p; p = next_mnt(p, subtree))
2927 if (mnt_ns_loop(p->mnt.mnt_root))
2932 unlock_mount_hash();
2936 static int do_set_group(struct path *from_path, struct path *to_path)
2938 struct mount *from, *to;
2941 from = real_mount(from_path->mnt);
2942 to = real_mount(to_path->mnt);
2947 /* To and From must be mounted */
2948 if (!is_mounted(&from->mnt))
2950 if (!is_mounted(&to->mnt))
2954 /* We should be allowed to modify mount namespaces of both mounts */
2955 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2957 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2961 /* To and From paths should be mount roots */
2962 if (!path_mounted(from_path))
2964 if (!path_mounted(to_path))
2967 /* Setting sharing groups is only allowed across same superblock */
2968 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2971 /* From mount root should be wider than To mount root */
2972 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2975 /* From mount should not have locked children in place of To's root */
2976 if (has_locked_children(from, to->mnt.mnt_root))
2979 /* Setting sharing groups is only allowed on private mounts */
2980 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2983 /* From should not be private */
2984 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2987 if (IS_MNT_SLAVE(from)) {
2988 struct mount *m = from->mnt_master;
2990 list_add(&to->mnt_slave, &m->mnt_slave_list);
2994 if (IS_MNT_SHARED(from)) {
2995 to->mnt_group_id = from->mnt_group_id;
2996 list_add(&to->mnt_share, &from->mnt_share);
2999 unlock_mount_hash();
3009 * path_overmounted - check if path is overmounted
3010 * @path: path to check
3012 * Check if path is overmounted, i.e., if there's a mount on top of
3013 * @path->mnt with @path->dentry as mountpoint.
3015 * Context: This function expects namespace_lock() to be held.
3016 * Return: If path is overmounted true is returned, false if not.
3018 static inline bool path_overmounted(const struct path *path)
3021 if (unlikely(__lookup_mnt(path->mnt, path->dentry))) {
3030 * can_move_mount_beneath - check that we can mount beneath the top mount
3031 * @from: mount to mount beneath
3032 * @to: mount under which to mount
3034 * - Make sure that @to->dentry is actually the root of a mount under
3035 * which we can mount another mount.
3036 * - Make sure that nothing can be mounted beneath the caller's current
3037 * root or the rootfs of the namespace.
3038 * - Make sure that the caller can unmount the topmost mount ensuring
3039 * that the caller could reveal the underlying mountpoint.
3040 * - Ensure that nothing has been mounted on top of @from before we
3041 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3042 * - Prevent mounting beneath a mount if the propagation relationship
3043 * between the source mount, parent mount, and top mount would lead to
3044 * nonsensical mount trees.
3046 * Context: This function expects namespace_lock() to be held.
3047 * Return: On success 0, and on error a negative error code is returned.
3049 static int can_move_mount_beneath(const struct path *from,
3050 const struct path *to,
3051 const struct mountpoint *mp)
3053 struct mount *mnt_from = real_mount(from->mnt),
3054 *mnt_to = real_mount(to->mnt),
3055 *parent_mnt_to = mnt_to->mnt_parent;
3057 if (!mnt_has_parent(mnt_to))
3060 if (!path_mounted(to))
3063 if (IS_MNT_LOCKED(mnt_to))
3066 /* Avoid creating shadow mounts during mount propagation. */
3067 if (path_overmounted(from))
3071 * Mounting beneath the rootfs only makes sense when the
3072 * semantics of pivot_root(".", ".") are used.
3074 if (&mnt_to->mnt == current->fs->root.mnt)
3076 if (parent_mnt_to == current->nsproxy->mnt_ns->root)
3079 for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
3084 * If the parent mount propagates to the child mount this would
3085 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3086 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3087 * defeats the whole purpose of mounting beneath another mount.
3089 if (propagation_would_overmount(parent_mnt_to, mnt_to, mp))
3093 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3094 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3095 * Afterwards @mnt_from would be mounted on top of
3096 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3097 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3098 * already mounted on @mnt_from, @mnt_to would ultimately be
3099 * remounted on top of @c. Afterwards, @mnt_from would be
3100 * covered by a copy @c of @mnt_from and @c would be covered by
3101 * @mnt_from itself. This defeats the whole purpose of mounting
3102 * @mnt_from beneath @mnt_to.
3104 if (propagation_would_overmount(parent_mnt_to, mnt_from, mp))
3110 static int do_move_mount(struct path *old_path, struct path *new_path,
3113 struct mnt_namespace *ns;
3116 struct mount *parent;
3117 struct mountpoint *mp, *old_mp;
3120 enum mnt_tree_flags_t flags = 0;
3122 mp = do_lock_mount(new_path, beneath);
3126 old = real_mount(old_path->mnt);
3127 p = real_mount(new_path->mnt);
3128 parent = old->mnt_parent;
3129 attached = mnt_has_parent(old);
3131 flags |= MNT_TREE_MOVE;
3132 old_mp = old->mnt_mp;
3136 /* The mountpoint must be in our namespace. */
3140 /* The thing moved must be mounted... */
3141 if (!is_mounted(&old->mnt))
3144 /* ... and either ours or the root of anon namespace */
3145 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
3148 if (old->mnt.mnt_flags & MNT_LOCKED)
3151 if (!path_mounted(old_path))
3154 if (d_is_dir(new_path->dentry) !=
3155 d_is_dir(old_path->dentry))
3158 * Don't move a mount residing in a shared parent.
3160 if (attached && IS_MNT_SHARED(parent))
3164 err = can_move_mount_beneath(old_path, new_path, mp);
3170 flags |= MNT_TREE_BENEATH;
3174 * Don't move a mount tree containing unbindable mounts to a destination
3175 * mount which is shared.
3177 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
3180 if (!check_for_nsfs_mounts(old))
3182 for (; mnt_has_parent(p); p = p->mnt_parent)
3186 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
3190 /* if the mount is moved, it should no longer be expire
3192 list_del_init(&old->mnt_expire);
3194 put_mountpoint(old_mp);
3199 mntput_no_expire(parent);
3206 static int do_move_mount_old(struct path *path, const char *old_name)
3208 struct path old_path;
3211 if (!old_name || !*old_name)
3214 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3218 err = do_move_mount(&old_path, path, false);
3219 path_put(&old_path);
3224 * add a mount into a namespace's mount tree
3226 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
3227 const struct path *path, int mnt_flags)
3229 struct mount *parent = real_mount(path->mnt);
3231 mnt_flags &= ~MNT_INTERNAL_FLAGS;
3233 if (unlikely(!check_mnt(parent))) {
3234 /* that's acceptable only for automounts done in private ns */
3235 if (!(mnt_flags & MNT_SHRINKABLE))
3237 /* ... and for those we'd better have mountpoint still alive */
3238 if (!parent->mnt_ns)
3242 /* Refuse the same filesystem on the same mount point */
3243 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
3246 if (d_is_symlink(newmnt->mnt.mnt_root))
3249 newmnt->mnt.mnt_flags = mnt_flags;
3250 return graft_tree(newmnt, parent, mp);
3253 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3256 * Create a new mount using a superblock configuration and request it
3257 * be added to the namespace tree.
3259 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3260 unsigned int mnt_flags)
3262 struct vfsmount *mnt;
3263 struct mountpoint *mp;
3264 struct super_block *sb = fc->root->d_sb;
3267 error = security_sb_kern_mount(sb);
3268 if (!error && mount_too_revealing(sb, &mnt_flags))
3271 if (unlikely(error)) {
3276 up_write(&sb->s_umount);
3278 mnt = vfs_create_mount(fc);
3280 return PTR_ERR(mnt);
3282 mnt_warn_timestamp_expiry(mountpoint, mnt);
3284 mp = lock_mount(mountpoint);
3289 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3297 * create a new mount for userspace and request it to be added into the
3300 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3301 int mnt_flags, const char *name, void *data)
3303 struct file_system_type *type;
3304 struct fs_context *fc;
3305 const char *subtype = NULL;
3311 type = get_fs_type(fstype);
3315 if (type->fs_flags & FS_HAS_SUBTYPE) {
3316 subtype = strchr(fstype, '.');
3320 put_filesystem(type);
3326 fc = fs_context_for_mount(type, sb_flags);
3327 put_filesystem(type);
3332 err = vfs_parse_fs_string(fc, "subtype",
3333 subtype, strlen(subtype));
3335 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3337 err = parse_monolithic_mount_data(fc, data);
3338 if (!err && !mount_capable(fc))
3341 err = vfs_get_tree(fc);
3343 err = do_new_mount_fc(fc, path, mnt_flags);
3349 int finish_automount(struct vfsmount *m, const struct path *path)
3351 struct dentry *dentry = path->dentry;
3352 struct mountpoint *mp;
3361 mnt = real_mount(m);
3362 /* The new mount record should have at least 2 refs to prevent it being
3363 * expired before we get a chance to add it
3365 BUG_ON(mnt_get_count(mnt) < 2);
3367 if (m->mnt_sb == path->mnt->mnt_sb &&
3368 m->mnt_root == dentry) {
3374 * we don't want to use lock_mount() - in this case finding something
3375 * that overmounts our mountpoint to be means "quitely drop what we've
3376 * got", not "try to mount it on top".
3378 inode_lock(dentry->d_inode);
3380 if (unlikely(cant_mount(dentry))) {
3382 goto discard_locked;
3384 if (path_overmounted(path)) {
3386 goto discard_locked;
3388 mp = get_mountpoint(dentry);
3391 goto discard_locked;
3394 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3403 inode_unlock(dentry->d_inode);
3405 /* remove m from any expiration list it may be on */
3406 if (!list_empty(&mnt->mnt_expire)) {
3408 list_del_init(&mnt->mnt_expire);
3417 * mnt_set_expiry - Put a mount on an expiration list
3418 * @mnt: The mount to list.
3419 * @expiry_list: The list to add the mount to.
3421 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3425 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3429 EXPORT_SYMBOL(mnt_set_expiry);
3432 * process a list of expirable mountpoints with the intent of discarding any
3433 * mountpoints that aren't in use and haven't been touched since last we came
3436 void mark_mounts_for_expiry(struct list_head *mounts)
3438 struct mount *mnt, *next;
3439 LIST_HEAD(graveyard);
3441 if (list_empty(mounts))
3447 /* extract from the expiration list every vfsmount that matches the
3448 * following criteria:
3449 * - only referenced by its parent vfsmount
3450 * - still marked for expiry (marked on the last call here; marks are
3451 * cleared by mntput())
3453 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3454 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3455 propagate_mount_busy(mnt, 1))
3457 list_move(&mnt->mnt_expire, &graveyard);
3459 while (!list_empty(&graveyard)) {
3460 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3461 touch_mnt_namespace(mnt->mnt_ns);
3462 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3464 unlock_mount_hash();
3468 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3471 * Ripoff of 'select_parent()'
3473 * search the list of submounts for a given mountpoint, and move any
3474 * shrinkable submounts to the 'graveyard' list.
3476 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3478 struct mount *this_parent = parent;
3479 struct list_head *next;
3483 next = this_parent->mnt_mounts.next;
3485 while (next != &this_parent->mnt_mounts) {
3486 struct list_head *tmp = next;
3487 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3490 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3493 * Descend a level if the d_mounts list is non-empty.
3495 if (!list_empty(&mnt->mnt_mounts)) {
3500 if (!propagate_mount_busy(mnt, 1)) {
3501 list_move_tail(&mnt->mnt_expire, graveyard);
3506 * All done at this level ... ascend and resume the search
3508 if (this_parent != parent) {
3509 next = this_parent->mnt_child.next;
3510 this_parent = this_parent->mnt_parent;
3517 * process a list of expirable mountpoints with the intent of discarding any
3518 * submounts of a specific parent mountpoint
3520 * mount_lock must be held for write
3522 static void shrink_submounts(struct mount *mnt)
3524 LIST_HEAD(graveyard);
3527 /* extract submounts of 'mountpoint' from the expiration list */
3528 while (select_submounts(mnt, &graveyard)) {
3529 while (!list_empty(&graveyard)) {
3530 m = list_first_entry(&graveyard, struct mount,
3532 touch_mnt_namespace(m->mnt_ns);
3533 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3538 static void *copy_mount_options(const void __user * data)
3541 unsigned left, offset;
3546 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3548 return ERR_PTR(-ENOMEM);
3550 left = copy_from_user(copy, data, PAGE_SIZE);
3553 * Not all architectures have an exact copy_from_user(). Resort to
3556 offset = PAGE_SIZE - left;
3559 if (get_user(c, (const char __user *)data + offset))
3566 if (left == PAGE_SIZE) {
3568 return ERR_PTR(-EFAULT);
3574 static char *copy_mount_string(const void __user *data)
3576 return data ? strndup_user(data, PATH_MAX) : NULL;
3580 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3581 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3583 * data is a (void *) that can point to any structure up to
3584 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3585 * information (or be NULL).
3587 * Pre-0.97 versions of mount() didn't have a flags word.
3588 * When the flags word was introduced its top half was required
3589 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3590 * Therefore, if this magic number is present, it carries no information
3591 * and must be discarded.
3593 int path_mount(const char *dev_name, struct path *path,
3594 const char *type_page, unsigned long flags, void *data_page)
3596 unsigned int mnt_flags = 0, sb_flags;
3600 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3601 flags &= ~MS_MGC_MSK;
3603 /* Basic sanity checks */
3605 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3607 if (flags & MS_NOUSER)
3610 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3615 if (flags & SB_MANDLOCK)
3618 /* Default to relatime unless overriden */
3619 if (!(flags & MS_NOATIME))
3620 mnt_flags |= MNT_RELATIME;
3622 /* Separate the per-mountpoint flags */
3623 if (flags & MS_NOSUID)
3624 mnt_flags |= MNT_NOSUID;
3625 if (flags & MS_NODEV)
3626 mnt_flags |= MNT_NODEV;
3627 if (flags & MS_NOEXEC)
3628 mnt_flags |= MNT_NOEXEC;
3629 if (flags & MS_NOATIME)
3630 mnt_flags |= MNT_NOATIME;
3631 if (flags & MS_NODIRATIME)
3632 mnt_flags |= MNT_NODIRATIME;
3633 if (flags & MS_STRICTATIME)
3634 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3635 if (flags & MS_RDONLY)
3636 mnt_flags |= MNT_READONLY;
3637 if (flags & MS_NOSYMFOLLOW)
3638 mnt_flags |= MNT_NOSYMFOLLOW;
3640 /* The default atime for remount is preservation */
3641 if ((flags & MS_REMOUNT) &&
3642 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3643 MS_STRICTATIME)) == 0)) {
3644 mnt_flags &= ~MNT_ATIME_MASK;
3645 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3648 sb_flags = flags & (SB_RDONLY |
3657 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3658 return do_reconfigure_mnt(path, mnt_flags);
3659 if (flags & MS_REMOUNT)
3660 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3661 if (flags & MS_BIND)
3662 return do_loopback(path, dev_name, flags & MS_REC);
3663 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3664 return do_change_type(path, flags);
3665 if (flags & MS_MOVE)
3666 return do_move_mount_old(path, dev_name);
3668 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3672 long do_mount(const char *dev_name, const char __user *dir_name,
3673 const char *type_page, unsigned long flags, void *data_page)
3678 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3681 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3686 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3688 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3691 static void dec_mnt_namespaces(struct ucounts *ucounts)
3693 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3696 static void free_mnt_ns(struct mnt_namespace *ns)
3698 if (!is_anon_ns(ns))
3699 ns_free_inum(&ns->ns);
3700 dec_mnt_namespaces(ns->ucounts);
3701 put_user_ns(ns->user_ns);
3706 * Assign a sequence number so we can detect when we attempt to bind
3707 * mount a reference to an older mount namespace into the current
3708 * mount namespace, preventing reference counting loops. A 64bit
3709 * number incrementing at 10Ghz will take 12,427 years to wrap which
3710 * is effectively never, so we can ignore the possibility.
3712 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3714 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3716 struct mnt_namespace *new_ns;
3717 struct ucounts *ucounts;
3720 ucounts = inc_mnt_namespaces(user_ns);
3722 return ERR_PTR(-ENOSPC);
3724 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3726 dec_mnt_namespaces(ucounts);
3727 return ERR_PTR(-ENOMEM);
3730 ret = ns_alloc_inum(&new_ns->ns);
3733 dec_mnt_namespaces(ucounts);
3734 return ERR_PTR(ret);
3737 new_ns->ns.ops = &mntns_operations;
3739 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3740 refcount_set(&new_ns->ns.count, 1);
3741 INIT_LIST_HEAD(&new_ns->list);
3742 init_waitqueue_head(&new_ns->poll);
3743 spin_lock_init(&new_ns->ns_lock);
3744 new_ns->user_ns = get_user_ns(user_ns);
3745 new_ns->ucounts = ucounts;
3750 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3751 struct user_namespace *user_ns, struct fs_struct *new_fs)
3753 struct mnt_namespace *new_ns;
3754 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3755 struct mount *p, *q;
3762 if (likely(!(flags & CLONE_NEWNS))) {
3769 new_ns = alloc_mnt_ns(user_ns, false);
3774 /* First pass: copy the tree topology */
3775 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3776 if (user_ns != ns->user_ns)
3777 copy_flags |= CL_SHARED_TO_SLAVE;
3778 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3781 free_mnt_ns(new_ns);
3782 return ERR_CAST(new);
3784 if (user_ns != ns->user_ns) {
3787 unlock_mount_hash();
3790 list_add_tail(&new_ns->list, &new->mnt_list);
3793 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3794 * as belonging to new namespace. We have already acquired a private
3795 * fs_struct, so tsk->fs->lock is not needed.
3803 if (&p->mnt == new_fs->root.mnt) {
3804 new_fs->root.mnt = mntget(&q->mnt);
3807 if (&p->mnt == new_fs->pwd.mnt) {
3808 new_fs->pwd.mnt = mntget(&q->mnt);
3812 p = next_mnt(p, old);
3813 q = next_mnt(q, new);
3816 // an mntns binding we'd skipped?
3817 while (p->mnt.mnt_root != q->mnt.mnt_root)
3818 p = next_mnt(skip_mnt_tree(p), old);
3830 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3832 struct mount *mnt = real_mount(m);
3833 struct mnt_namespace *ns;
3834 struct super_block *s;
3838 ns = alloc_mnt_ns(&init_user_ns, true);
3841 return ERR_CAST(ns);
3846 list_add(&mnt->mnt_list, &ns->list);
3848 err = vfs_path_lookup(m->mnt_root, m,
3849 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3854 return ERR_PTR(err);
3856 /* trade a vfsmount reference for active sb one */
3857 s = path.mnt->mnt_sb;
3858 atomic_inc(&s->s_active);
3860 /* lock the sucker */
3861 down_write(&s->s_umount);
3862 /* ... and return the root of (sub)tree on it */
3865 EXPORT_SYMBOL(mount_subtree);
3867 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3868 char __user *, type, unsigned long, flags, void __user *, data)
3875 kernel_type = copy_mount_string(type);
3876 ret = PTR_ERR(kernel_type);
3877 if (IS_ERR(kernel_type))
3880 kernel_dev = copy_mount_string(dev_name);
3881 ret = PTR_ERR(kernel_dev);
3882 if (IS_ERR(kernel_dev))
3885 options = copy_mount_options(data);
3886 ret = PTR_ERR(options);
3887 if (IS_ERR(options))
3890 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3901 #define FSMOUNT_VALID_FLAGS \
3902 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3903 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3904 MOUNT_ATTR_NOSYMFOLLOW)
3906 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3908 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3909 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3911 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3913 unsigned int mnt_flags = 0;
3915 if (attr_flags & MOUNT_ATTR_RDONLY)
3916 mnt_flags |= MNT_READONLY;
3917 if (attr_flags & MOUNT_ATTR_NOSUID)
3918 mnt_flags |= MNT_NOSUID;
3919 if (attr_flags & MOUNT_ATTR_NODEV)
3920 mnt_flags |= MNT_NODEV;
3921 if (attr_flags & MOUNT_ATTR_NOEXEC)
3922 mnt_flags |= MNT_NOEXEC;
3923 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3924 mnt_flags |= MNT_NODIRATIME;
3925 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3926 mnt_flags |= MNT_NOSYMFOLLOW;
3932 * Create a kernel mount representation for a new, prepared superblock
3933 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3935 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3936 unsigned int, attr_flags)
3938 struct mnt_namespace *ns;
3939 struct fs_context *fc;
3941 struct path newmount;
3944 unsigned int mnt_flags = 0;
3950 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3953 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3956 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3958 switch (attr_flags & MOUNT_ATTR__ATIME) {
3959 case MOUNT_ATTR_STRICTATIME:
3961 case MOUNT_ATTR_NOATIME:
3962 mnt_flags |= MNT_NOATIME;
3964 case MOUNT_ATTR_RELATIME:
3965 mnt_flags |= MNT_RELATIME;
3976 if (f.file->f_op != &fscontext_fops)
3979 fc = f.file->private_data;
3981 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3985 /* There must be a valid superblock or we can't mount it */
3991 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3992 pr_warn("VFS: Mount too revealing\n");
3997 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
4000 if (fc->sb_flags & SB_MANDLOCK)
4003 newmount.mnt = vfs_create_mount(fc);
4004 if (IS_ERR(newmount.mnt)) {
4005 ret = PTR_ERR(newmount.mnt);
4008 newmount.dentry = dget(fc->root);
4009 newmount.mnt->mnt_flags = mnt_flags;
4011 /* We've done the mount bit - now move the file context into more or
4012 * less the same state as if we'd done an fspick(). We don't want to
4013 * do any memory allocation or anything like that at this point as we
4014 * don't want to have to handle any errors incurred.
4016 vfs_clean_context(fc);
4018 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
4023 mnt = real_mount(newmount.mnt);
4027 list_add(&mnt->mnt_list, &ns->list);
4028 mntget(newmount.mnt);
4030 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4031 * it, not just simply put it.
4033 file = dentry_open(&newmount, O_PATH, fc->cred);
4035 dissolve_on_fput(newmount.mnt);
4036 ret = PTR_ERR(file);
4039 file->f_mode |= FMODE_NEED_UNMOUNT;
4041 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
4043 fd_install(ret, file);
4048 path_put(&newmount);
4050 mutex_unlock(&fc->uapi_mutex);
4057 * Move a mount from one place to another. In combination with
4058 * fsopen()/fsmount() this is used to install a new mount and in combination
4059 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4062 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4064 SYSCALL_DEFINE5(move_mount,
4065 int, from_dfd, const char __user *, from_pathname,
4066 int, to_dfd, const char __user *, to_pathname,
4067 unsigned int, flags)
4069 struct path from_path, to_path;
4070 unsigned int lflags;
4076 if (flags & ~MOVE_MOUNT__MASK)
4079 if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) ==
4080 (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP))
4083 /* If someone gives a pathname, they aren't permitted to move
4084 * from an fd that requires unmount as we can't get at the flag
4085 * to clear it afterwards.
4088 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4089 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4090 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4092 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
4097 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4098 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4099 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4101 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
4105 ret = security_move_mount(&from_path, &to_path);
4109 if (flags & MOVE_MOUNT_SET_GROUP)
4110 ret = do_set_group(&from_path, &to_path);
4112 ret = do_move_mount(&from_path, &to_path,
4113 (flags & MOVE_MOUNT_BENEATH));
4118 path_put(&from_path);
4123 * Return true if path is reachable from root
4125 * namespace_sem or mount_lock is held
4127 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
4128 const struct path *root)
4130 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
4131 dentry = mnt->mnt_mountpoint;
4132 mnt = mnt->mnt_parent;
4134 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
4137 bool path_is_under(const struct path *path1, const struct path *path2)
4140 read_seqlock_excl(&mount_lock);
4141 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
4142 read_sequnlock_excl(&mount_lock);
4145 EXPORT_SYMBOL(path_is_under);
4148 * pivot_root Semantics:
4149 * Moves the root file system of the current process to the directory put_old,
4150 * makes new_root as the new root file system of the current process, and sets
4151 * root/cwd of all processes which had them on the current root to new_root.
4154 * The new_root and put_old must be directories, and must not be on the
4155 * same file system as the current process root. The put_old must be
4156 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4157 * pointed to by put_old must yield the same directory as new_root. No other
4158 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4160 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
4161 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4162 * in this situation.
4165 * - we don't move root/cwd if they are not at the root (reason: if something
4166 * cared enough to change them, it's probably wrong to force them elsewhere)
4167 * - it's okay to pick a root that isn't the root of a file system, e.g.
4168 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4169 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4172 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
4173 const char __user *, put_old)
4175 struct path new, old, root;
4176 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
4177 struct mountpoint *old_mp, *root_mp;
4183 error = user_path_at(AT_FDCWD, new_root,
4184 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
4188 error = user_path_at(AT_FDCWD, put_old,
4189 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
4193 error = security_sb_pivotroot(&old, &new);
4197 get_fs_root(current->fs, &root);
4198 old_mp = lock_mount(&old);
4199 error = PTR_ERR(old_mp);
4204 new_mnt = real_mount(new.mnt);
4205 root_mnt = real_mount(root.mnt);
4206 old_mnt = real_mount(old.mnt);
4207 ex_parent = new_mnt->mnt_parent;
4208 root_parent = root_mnt->mnt_parent;
4209 if (IS_MNT_SHARED(old_mnt) ||
4210 IS_MNT_SHARED(ex_parent) ||
4211 IS_MNT_SHARED(root_parent))
4213 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
4215 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4218 if (d_unlinked(new.dentry))
4221 if (new_mnt == root_mnt || old_mnt == root_mnt)
4222 goto out4; /* loop, on the same file system */
4224 if (!path_mounted(&root))
4225 goto out4; /* not a mountpoint */
4226 if (!mnt_has_parent(root_mnt))
4227 goto out4; /* not attached */
4228 if (!path_mounted(&new))
4229 goto out4; /* not a mountpoint */
4230 if (!mnt_has_parent(new_mnt))
4231 goto out4; /* not attached */
4232 /* make sure we can reach put_old from new_root */
4233 if (!is_path_reachable(old_mnt, old.dentry, &new))
4235 /* make certain new is below the root */
4236 if (!is_path_reachable(new_mnt, new.dentry, &root))
4239 umount_mnt(new_mnt);
4240 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
4241 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4242 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4243 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4245 /* mount old root on put_old */
4246 attach_mnt(root_mnt, old_mnt, old_mp, false);
4247 /* mount new_root on / */
4248 attach_mnt(new_mnt, root_parent, root_mp, false);
4249 mnt_add_count(root_parent, -1);
4250 touch_mnt_namespace(current->nsproxy->mnt_ns);
4251 /* A moved mount should not expire automatically */
4252 list_del_init(&new_mnt->mnt_expire);
4253 put_mountpoint(root_mp);
4254 unlock_mount_hash();
4255 chroot_fs_refs(&root, &new);
4258 unlock_mount(old_mp);
4260 mntput_no_expire(ex_parent);
4271 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4273 unsigned int flags = mnt->mnt.mnt_flags;
4275 /* flags to clear */
4276 flags &= ~kattr->attr_clr;
4277 /* flags to raise */
4278 flags |= kattr->attr_set;
4283 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4285 struct vfsmount *m = &mnt->mnt;
4286 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4288 if (!kattr->mnt_idmap)
4292 * Creating an idmapped mount with the filesystem wide idmapping
4293 * doesn't make sense so block that. We don't allow mushy semantics.
4295 if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb))
4299 * Once a mount has been idmapped we don't allow it to change its
4300 * mapping. It makes things simpler and callers can just create
4301 * another bind-mount they can idmap if they want to.
4303 if (is_idmapped_mnt(m))
4306 /* The underlying filesystem doesn't support idmapped mounts yet. */
4307 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4310 /* We're not controlling the superblock. */
4311 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4314 /* Mount has already been visible in the filesystem hierarchy. */
4315 if (!is_anon_ns(mnt->mnt_ns))
4322 * mnt_allow_writers() - check whether the attribute change allows writers
4323 * @kattr: the new mount attributes
4324 * @mnt: the mount to which @kattr will be applied
4326 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4328 * Return: true if writers need to be held, false if not
4330 static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4331 const struct mount *mnt)
4333 return (!(kattr->attr_set & MNT_READONLY) ||
4334 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4338 static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4343 for (m = mnt; m; m = next_mnt(m, mnt)) {
4344 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4349 err = can_idmap_mount(kattr, m);
4353 if (!mnt_allow_writers(kattr, m)) {
4354 err = mnt_hold_writers(m);
4359 if (!kattr->recurse)
4367 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4368 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4369 * mounts and needs to take care to include the first mount.
4371 for (p = mnt; p; p = next_mnt(p, mnt)) {
4372 /* If we had to hold writers unblock them. */
4373 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4374 mnt_unhold_writers(p);
4377 * We're done once the first mount we changed got
4378 * MNT_WRITE_HOLD unset.
4387 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4389 if (!kattr->mnt_idmap)
4393 * Pairs with smp_load_acquire() in mnt_idmap().
4395 * Since we only allow a mount to change the idmapping once and
4396 * verified this in can_idmap_mount() we know that the mount has
4397 * @nop_mnt_idmap attached to it. So there's no need to drop any
4400 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4403 static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4407 for (m = mnt; m; m = next_mnt(m, mnt)) {
4410 do_idmap_mount(kattr, m);
4411 flags = recalc_flags(kattr, m);
4412 WRITE_ONCE(m->mnt.mnt_flags, flags);
4414 /* If we had to hold writers unblock them. */
4415 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4416 mnt_unhold_writers(m);
4418 if (kattr->propagation)
4419 change_mnt_propagation(m, kattr->propagation);
4420 if (!kattr->recurse)
4423 touch_mnt_namespace(mnt->mnt_ns);
4426 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4428 struct mount *mnt = real_mount(path->mnt);
4431 if (!path_mounted(path))
4434 if (kattr->mnt_userns) {
4435 struct mnt_idmap *mnt_idmap;
4437 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4438 if (IS_ERR(mnt_idmap))
4439 return PTR_ERR(mnt_idmap);
4440 kattr->mnt_idmap = mnt_idmap;
4443 if (kattr->propagation) {
4445 * Only take namespace_lock() if we're actually changing
4449 if (kattr->propagation == MS_SHARED) {
4450 err = invent_group_ids(mnt, kattr->recurse);
4461 /* Ensure that this isn't anything purely vfs internal. */
4462 if (!is_mounted(&mnt->mnt))
4466 * If this is an attached mount make sure it's located in the callers
4467 * mount namespace. If it's not don't let the caller interact with it.
4468 * If this is a detached mount make sure it has an anonymous mount
4469 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4471 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4475 * First, we get the mount tree in a shape where we can change mount
4476 * properties without failure. If we succeeded to do so we commit all
4477 * changes and if we failed we clean up.
4479 err = mount_setattr_prepare(kattr, mnt);
4481 mount_setattr_commit(kattr, mnt);
4484 unlock_mount_hash();
4486 if (kattr->propagation) {
4488 cleanup_group_ids(mnt, NULL);
4495 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4496 struct mount_kattr *kattr, unsigned int flags)
4499 struct ns_common *ns;
4500 struct user_namespace *mnt_userns;
4503 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4507 * We currently do not support clearing an idmapped mount. If this ever
4508 * is a use-case we can revisit this but for now let's keep it simple
4511 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4514 if (attr->userns_fd > INT_MAX)
4517 f = fdget(attr->userns_fd);
4521 if (!proc_ns_file(f.file)) {
4526 ns = get_proc_ns(file_inode(f.file));
4527 if (ns->ops->type != CLONE_NEWUSER) {
4533 * The initial idmapping cannot be used to create an idmapped
4534 * mount. We use the initial idmapping as an indicator of a mount
4535 * that is not idmapped. It can simply be passed into helpers that
4536 * are aware of idmapped mounts as a convenient shortcut. A user
4537 * can just create a dedicated identity mapping to achieve the same
4540 mnt_userns = container_of(ns, struct user_namespace, ns);
4541 if (mnt_userns == &init_user_ns) {
4546 /* We're not controlling the target namespace. */
4547 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4552 kattr->mnt_userns = get_user_ns(mnt_userns);
4559 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4560 struct mount_kattr *kattr, unsigned int flags)
4562 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4564 if (flags & AT_NO_AUTOMOUNT)
4565 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4566 if (flags & AT_SYMLINK_NOFOLLOW)
4567 lookup_flags &= ~LOOKUP_FOLLOW;
4568 if (flags & AT_EMPTY_PATH)
4569 lookup_flags |= LOOKUP_EMPTY;
4571 *kattr = (struct mount_kattr) {
4572 .lookup_flags = lookup_flags,
4573 .recurse = !!(flags & AT_RECURSIVE),
4576 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4578 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4580 kattr->propagation = attr->propagation;
4582 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4585 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4586 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4589 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4590 * users wanting to transition to a different atime setting cannot
4591 * simply specify the atime setting in @attr_set, but must also
4592 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4593 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4594 * @attr_clr and that @attr_set can't have any atime bits set if
4595 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4597 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4598 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4602 * Clear all previous time settings as they are mutually
4605 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4606 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4607 case MOUNT_ATTR_RELATIME:
4608 kattr->attr_set |= MNT_RELATIME;
4610 case MOUNT_ATTR_NOATIME:
4611 kattr->attr_set |= MNT_NOATIME;
4613 case MOUNT_ATTR_STRICTATIME:
4619 if (attr->attr_set & MOUNT_ATTR__ATIME)
4623 return build_mount_idmapped(attr, usize, kattr, flags);
4626 static void finish_mount_kattr(struct mount_kattr *kattr)
4628 put_user_ns(kattr->mnt_userns);
4629 kattr->mnt_userns = NULL;
4631 if (kattr->mnt_idmap)
4632 mnt_idmap_put(kattr->mnt_idmap);
4635 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4636 unsigned int, flags, struct mount_attr __user *, uattr,
4641 struct mount_attr attr;
4642 struct mount_kattr kattr;
4644 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4646 if (flags & ~(AT_EMPTY_PATH |
4648 AT_SYMLINK_NOFOLLOW |
4652 if (unlikely(usize > PAGE_SIZE))
4654 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4660 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4664 /* Don't bother walking through the mounts if this is a nop. */
4665 if (attr.attr_set == 0 &&
4666 attr.attr_clr == 0 &&
4667 attr.propagation == 0)
4670 err = build_mount_kattr(&attr, usize, &kattr, flags);
4674 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4676 err = do_mount_setattr(&target, &kattr);
4679 finish_mount_kattr(&kattr);
4683 static void __init init_mount_tree(void)
4685 struct vfsmount *mnt;
4687 struct mnt_namespace *ns;
4690 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4692 panic("Can't create rootfs");
4694 ns = alloc_mnt_ns(&init_user_ns, false);
4696 panic("Can't allocate initial namespace");
4697 m = real_mount(mnt);
4701 list_add(&m->mnt_list, &ns->list);
4702 init_task.nsproxy->mnt_ns = ns;
4706 root.dentry = mnt->mnt_root;
4707 mnt->mnt_flags |= MNT_LOCKED;
4709 set_fs_pwd(current->fs, &root);
4710 set_fs_root(current->fs, &root);
4713 void __init mnt_init(void)
4717 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4718 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4720 mount_hashtable = alloc_large_system_hash("Mount-cache",
4721 sizeof(struct hlist_head),
4724 &m_hash_shift, &m_hash_mask, 0, 0);
4725 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4726 sizeof(struct hlist_head),
4729 &mp_hash_shift, &mp_hash_mask, 0, 0);
4731 if (!mount_hashtable || !mountpoint_hashtable)
4732 panic("Failed to allocate mount hash table\n");
4738 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4740 fs_kobj = kobject_create_and_add("fs", NULL);
4742 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4748 void put_mnt_ns(struct mnt_namespace *ns)
4750 if (!refcount_dec_and_test(&ns->ns.count))
4752 drop_collected_mounts(&ns->root->mnt);
4756 struct vfsmount *kern_mount(struct file_system_type *type)
4758 struct vfsmount *mnt;
4759 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4762 * it is a longterm mount, don't release mnt until
4763 * we unmount before file sys is unregistered
4765 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4769 EXPORT_SYMBOL_GPL(kern_mount);
4771 void kern_unmount(struct vfsmount *mnt)
4773 /* release long term mount so mount point can be released */
4775 mnt_make_shortterm(mnt);
4776 synchronize_rcu(); /* yecchhh... */
4780 EXPORT_SYMBOL(kern_unmount);
4782 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4786 for (i = 0; i < num; i++)
4787 mnt_make_shortterm(mnt[i]);
4788 synchronize_rcu_expedited();
4789 for (i = 0; i < num; i++)
4792 EXPORT_SYMBOL(kern_unmount_array);
4794 bool our_mnt(struct vfsmount *mnt)
4796 return check_mnt(real_mount(mnt));
4799 bool current_chrooted(void)
4801 /* Does the current process have a non-standard root */
4802 struct path ns_root;
4803 struct path fs_root;
4806 /* Find the namespace root */
4807 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4808 ns_root.dentry = ns_root.mnt->mnt_root;
4810 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4813 get_fs_root(current->fs, &fs_root);
4815 chrooted = !path_equal(&fs_root, &ns_root);
4823 static bool mnt_already_visible(struct mnt_namespace *ns,
4824 const struct super_block *sb,
4827 int new_flags = *new_mnt_flags;
4829 bool visible = false;
4831 down_read(&namespace_sem);
4833 list_for_each_entry(mnt, &ns->list, mnt_list) {
4834 struct mount *child;
4837 if (mnt_is_cursor(mnt))
4840 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4843 /* This mount is not fully visible if it's root directory
4844 * is not the root directory of the filesystem.
4846 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4849 /* A local view of the mount flags */
4850 mnt_flags = mnt->mnt.mnt_flags;
4852 /* Don't miss readonly hidden in the superblock flags */
4853 if (sb_rdonly(mnt->mnt.mnt_sb))
4854 mnt_flags |= MNT_LOCK_READONLY;
4856 /* Verify the mount flags are equal to or more permissive
4857 * than the proposed new mount.
4859 if ((mnt_flags & MNT_LOCK_READONLY) &&
4860 !(new_flags & MNT_READONLY))
4862 if ((mnt_flags & MNT_LOCK_ATIME) &&
4863 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4866 /* This mount is not fully visible if there are any
4867 * locked child mounts that cover anything except for
4868 * empty directories.
4870 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4871 struct inode *inode = child->mnt_mountpoint->d_inode;
4872 /* Only worry about locked mounts */
4873 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4875 /* Is the directory permanetly empty? */
4876 if (!is_empty_dir_inode(inode))
4879 /* Preserve the locked attributes */
4880 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4888 up_read(&namespace_sem);
4892 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4894 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4895 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4896 unsigned long s_iflags;
4898 if (ns->user_ns == &init_user_ns)
4901 /* Can this filesystem be too revealing? */
4902 s_iflags = sb->s_iflags;
4903 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4906 if ((s_iflags & required_iflags) != required_iflags) {
4907 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4912 return !mnt_already_visible(ns, sb, new_mnt_flags);
4915 bool mnt_may_suid(struct vfsmount *mnt)
4918 * Foreign mounts (accessed via fchdir or through /proc
4919 * symlinks) are always treated as if they are nosuid. This
4920 * prevents namespaces from trusting potentially unsafe
4921 * suid/sgid bits, file caps, or security labels that originate
4922 * in other namespaces.
4924 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4925 current_in_userns(mnt->mnt_sb->s_user_ns);
4928 static struct ns_common *mntns_get(struct task_struct *task)
4930 struct ns_common *ns = NULL;
4931 struct nsproxy *nsproxy;
4934 nsproxy = task->nsproxy;
4936 ns = &nsproxy->mnt_ns->ns;
4937 get_mnt_ns(to_mnt_ns(ns));
4944 static void mntns_put(struct ns_common *ns)
4946 put_mnt_ns(to_mnt_ns(ns));
4949 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4951 struct nsproxy *nsproxy = nsset->nsproxy;
4952 struct fs_struct *fs = nsset->fs;
4953 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4954 struct user_namespace *user_ns = nsset->cred->user_ns;
4958 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4959 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4960 !ns_capable(user_ns, CAP_SYS_ADMIN))
4963 if (is_anon_ns(mnt_ns))
4970 old_mnt_ns = nsproxy->mnt_ns;
4971 nsproxy->mnt_ns = mnt_ns;
4974 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4975 "/", LOOKUP_DOWN, &root);
4977 /* revert to old namespace */
4978 nsproxy->mnt_ns = old_mnt_ns;
4983 put_mnt_ns(old_mnt_ns);
4985 /* Update the pwd and root */
4986 set_fs_pwd(fs, &root);
4987 set_fs_root(fs, &root);
4993 static struct user_namespace *mntns_owner(struct ns_common *ns)
4995 return to_mnt_ns(ns)->user_ns;
4998 const struct proc_ns_operations mntns_operations = {
5000 .type = CLONE_NEWNS,
5003 .install = mntns_install,
5004 .owner = mntns_owner,
5007 #ifdef CONFIG_SYSCTL
5008 static struct ctl_table fs_namespace_sysctls[] = {
5010 .procname = "mount-max",
5011 .data = &sysctl_mount_max,
5012 .maxlen = sizeof(unsigned int),
5014 .proc_handler = proc_dointvec_minmax,
5015 .extra1 = SYSCTL_ONE,
5020 static int __init init_fs_namespace_sysctls(void)
5022 register_sysctl_init("fs", fs_namespace_sysctls);
5025 fs_initcall(init_fs_namespace_sysctls);
5027 #endif /* CONFIG_SYSCTL */