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 static struct hlist_head *mount_hashtable __ro_after_init;
72 static struct hlist_head *mountpoint_hashtable __ro_after_init;
73 static struct kmem_cache *mnt_cache __ro_after_init;
74 static DECLARE_RWSEM(namespace_sem);
75 static HLIST_HEAD(unmounted); /* protected by namespace_sem */
76 static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
79 unsigned int attr_set;
80 unsigned int attr_clr;
81 unsigned int propagation;
82 unsigned int lookup_flags;
84 struct user_namespace *mnt_userns;
85 struct mnt_idmap *mnt_idmap;
89 struct kobject *fs_kobj __ro_after_init;
90 EXPORT_SYMBOL_GPL(fs_kobj);
93 * vfsmount lock may be taken for read to prevent changes to the
94 * vfsmount hash, ie. during mountpoint lookups or walking back
97 * It should be taken for write in all cases where the vfsmount
98 * tree or hash is modified or when a vfsmount structure is modified.
100 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
102 static inline void lock_mount_hash(void)
104 write_seqlock(&mount_lock);
107 static inline void unlock_mount_hash(void)
109 write_sequnlock(&mount_lock);
112 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
114 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
115 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
116 tmp = tmp + (tmp >> m_hash_shift);
117 return &mount_hashtable[tmp & m_hash_mask];
120 static inline struct hlist_head *mp_hash(struct dentry *dentry)
122 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
123 tmp = tmp + (tmp >> mp_hash_shift);
124 return &mountpoint_hashtable[tmp & mp_hash_mask];
127 static int mnt_alloc_id(struct mount *mnt)
129 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
137 static void mnt_free_id(struct mount *mnt)
139 ida_free(&mnt_id_ida, mnt->mnt_id);
143 * Allocate a new peer group ID
145 static int mnt_alloc_group_id(struct mount *mnt)
147 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
151 mnt->mnt_group_id = res;
156 * Release a peer group ID
158 void mnt_release_group_id(struct mount *mnt)
160 ida_free(&mnt_group_ida, mnt->mnt_group_id);
161 mnt->mnt_group_id = 0;
165 * vfsmount lock must be held for read
167 static inline void mnt_add_count(struct mount *mnt, int n)
170 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
179 * vfsmount lock must be held for write
181 int mnt_get_count(struct mount *mnt)
187 for_each_possible_cpu(cpu) {
188 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
193 return mnt->mnt_count;
197 static struct mount *alloc_vfsmnt(const char *name)
199 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
203 err = mnt_alloc_id(mnt);
208 mnt->mnt_devname = kstrdup_const(name,
210 if (!mnt->mnt_devname)
215 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
217 goto out_free_devname;
219 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
222 mnt->mnt_writers = 0;
225 INIT_HLIST_NODE(&mnt->mnt_hash);
226 INIT_LIST_HEAD(&mnt->mnt_child);
227 INIT_LIST_HEAD(&mnt->mnt_mounts);
228 INIT_LIST_HEAD(&mnt->mnt_list);
229 INIT_LIST_HEAD(&mnt->mnt_expire);
230 INIT_LIST_HEAD(&mnt->mnt_share);
231 INIT_LIST_HEAD(&mnt->mnt_slave_list);
232 INIT_LIST_HEAD(&mnt->mnt_slave);
233 INIT_HLIST_NODE(&mnt->mnt_mp_list);
234 INIT_LIST_HEAD(&mnt->mnt_umounting);
235 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
236 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
242 kfree_const(mnt->mnt_devname);
247 kmem_cache_free(mnt_cache, mnt);
252 * Most r/o checks on a fs are for operations that take
253 * discrete amounts of time, like a write() or unlink().
254 * We must keep track of when those operations start
255 * (for permission checks) and when they end, so that
256 * we can determine when writes are able to occur to
260 * __mnt_is_readonly: check whether a mount is read-only
261 * @mnt: the mount to check for its write status
263 * This shouldn't be used directly ouside of the VFS.
264 * It does not guarantee that the filesystem will stay
265 * r/w, just that it is right *now*. This can not and
266 * should not be used in place of IS_RDONLY(inode).
267 * mnt_want/drop_write() will _keep_ the filesystem
270 bool __mnt_is_readonly(struct vfsmount *mnt)
272 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
274 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
276 static inline void mnt_inc_writers(struct mount *mnt)
279 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
285 static inline void mnt_dec_writers(struct mount *mnt)
288 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
294 static unsigned int mnt_get_writers(struct mount *mnt)
297 unsigned int count = 0;
300 for_each_possible_cpu(cpu) {
301 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
306 return mnt->mnt_writers;
310 static int mnt_is_readonly(struct vfsmount *mnt)
312 if (READ_ONCE(mnt->mnt_sb->s_readonly_remount))
315 * The barrier pairs with the barrier in sb_start_ro_state_change()
316 * making sure if we don't see s_readonly_remount set yet, we also will
317 * not see any superblock / mount flag changes done by remount.
318 * It also pairs with the barrier in sb_end_ro_state_change()
319 * assuring that if we see s_readonly_remount already cleared, we will
320 * see the values of superblock / mount flags updated by remount.
323 return __mnt_is_readonly(mnt);
327 * Most r/o & frozen checks on a fs are for operations that take discrete
328 * amounts of time, like a write() or unlink(). We must keep track of when
329 * those operations start (for permission checks) and when they end, so that we
330 * can determine when writes are able to occur to a filesystem.
333 * mnt_get_write_access - get write access to a mount without freeze protection
334 * @m: the mount on which to take a write
336 * This tells the low-level filesystem that a write is about to be performed to
337 * it, and makes sure that writes are allowed (mnt it read-write) before
338 * returning success. This operation does not protect against filesystem being
339 * frozen. When the write operation is finished, mnt_put_write_access() must be
340 * called. This is effectively a refcount.
342 int mnt_get_write_access(struct vfsmount *m)
344 struct mount *mnt = real_mount(m);
348 mnt_inc_writers(mnt);
350 * The store to mnt_inc_writers must be visible before we pass
351 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
352 * incremented count after it has set MNT_WRITE_HOLD.
355 might_lock(&mount_lock.lock);
356 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
357 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
361 * This prevents priority inversion, if the task
362 * setting MNT_WRITE_HOLD got preempted on a remote
363 * CPU, and it prevents life lock if the task setting
364 * MNT_WRITE_HOLD has a lower priority and is bound to
365 * the same CPU as the task that is spinning here.
374 * The barrier pairs with the barrier sb_start_ro_state_change() making
375 * sure that if we see MNT_WRITE_HOLD cleared, we will also see
376 * s_readonly_remount set (or even SB_RDONLY / MNT_READONLY flags) in
377 * mnt_is_readonly() and bail in case we are racing with remount
381 if (mnt_is_readonly(m)) {
382 mnt_dec_writers(mnt);
389 EXPORT_SYMBOL_GPL(mnt_get_write_access);
392 * mnt_want_write - get write access to a mount
393 * @m: the mount on which to take a write
395 * This tells the low-level filesystem that a write is about to be performed to
396 * it, and makes sure that writes are allowed (mount is read-write, filesystem
397 * is not frozen) before returning success. When the write operation is
398 * finished, mnt_drop_write() must be called. This is effectively a refcount.
400 int mnt_want_write(struct vfsmount *m)
404 sb_start_write(m->mnt_sb);
405 ret = mnt_get_write_access(m);
407 sb_end_write(m->mnt_sb);
410 EXPORT_SYMBOL_GPL(mnt_want_write);
413 * mnt_get_write_access_file - get write access to a file's mount
414 * @file: the file who's mount on which to take a write
416 * This is like mnt_get_write_access, but if @file is already open for write it
417 * skips incrementing mnt_writers (since the open file already has a reference)
418 * and instead only does the check for emergency r/o remounts. This must be
419 * paired with mnt_put_write_access_file.
421 int mnt_get_write_access_file(struct file *file)
423 if (file->f_mode & FMODE_WRITER) {
425 * Superblock may have become readonly while there are still
426 * writable fd's, e.g. due to a fs error with errors=remount-ro
428 if (__mnt_is_readonly(file->f_path.mnt))
432 return mnt_get_write_access(file->f_path.mnt);
436 * mnt_want_write_file - get write access to a file's mount
437 * @file: the file who's mount on which to take a write
439 * This is like mnt_want_write, but if the file is already open for writing it
440 * skips incrementing mnt_writers (since the open file already has a reference)
441 * and instead only does the freeze protection and the check for emergency r/o
442 * remounts. This must be paired with mnt_drop_write_file.
444 int mnt_want_write_file(struct file *file)
448 sb_start_write(file_inode(file)->i_sb);
449 ret = mnt_get_write_access_file(file);
451 sb_end_write(file_inode(file)->i_sb);
454 EXPORT_SYMBOL_GPL(mnt_want_write_file);
457 * mnt_put_write_access - give up write access to a mount
458 * @mnt: the mount on which to give up write access
460 * Tells the low-level filesystem that we are done
461 * performing writes to it. Must be matched with
462 * mnt_get_write_access() call above.
464 void mnt_put_write_access(struct vfsmount *mnt)
467 mnt_dec_writers(real_mount(mnt));
470 EXPORT_SYMBOL_GPL(mnt_put_write_access);
473 * mnt_drop_write - give up write access to a mount
474 * @mnt: the mount on which to give up write access
476 * Tells the low-level filesystem that we are done performing writes to it and
477 * also allows filesystem to be frozen again. Must be matched with
478 * mnt_want_write() call above.
480 void mnt_drop_write(struct vfsmount *mnt)
482 mnt_put_write_access(mnt);
483 sb_end_write(mnt->mnt_sb);
485 EXPORT_SYMBOL_GPL(mnt_drop_write);
487 void mnt_put_write_access_file(struct file *file)
489 if (!(file->f_mode & FMODE_WRITER))
490 mnt_put_write_access(file->f_path.mnt);
493 void mnt_drop_write_file(struct file *file)
495 mnt_put_write_access_file(file);
496 sb_end_write(file_inode(file)->i_sb);
498 EXPORT_SYMBOL(mnt_drop_write_file);
501 * mnt_hold_writers - prevent write access to the given mount
502 * @mnt: mnt to prevent write access to
504 * Prevents write access to @mnt if there are no active writers for @mnt.
505 * This function needs to be called and return successfully before changing
506 * properties of @mnt that need to remain stable for callers with write access
509 * After this functions has been called successfully callers must pair it with
510 * a call to mnt_unhold_writers() in order to stop preventing write access to
513 * Context: This function expects lock_mount_hash() to be held serializing
514 * setting MNT_WRITE_HOLD.
515 * Return: On success 0 is returned.
516 * On error, -EBUSY is returned.
518 static inline int mnt_hold_writers(struct mount *mnt)
520 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
522 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
523 * should be visible before we do.
528 * With writers on hold, if this value is zero, then there are
529 * definitely no active writers (although held writers may subsequently
530 * increment the count, they'll have to wait, and decrement it after
531 * seeing MNT_READONLY).
533 * It is OK to have counter incremented on one CPU and decremented on
534 * another: the sum will add up correctly. The danger would be when we
535 * sum up each counter, if we read a counter before it is incremented,
536 * but then read another CPU's count which it has been subsequently
537 * decremented from -- we would see more decrements than we should.
538 * MNT_WRITE_HOLD protects against this scenario, because
539 * mnt_want_write first increments count, then smp_mb, then spins on
540 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
541 * we're counting up here.
543 if (mnt_get_writers(mnt) > 0)
550 * mnt_unhold_writers - stop preventing write access to the given mount
551 * @mnt: mnt to stop preventing write access to
553 * Stop preventing write access to @mnt allowing callers to gain write access
556 * This function can only be called after a successful call to
557 * mnt_hold_writers().
559 * Context: This function expects lock_mount_hash() to be held.
561 static inline void mnt_unhold_writers(struct mount *mnt)
564 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
565 * that become unheld will see MNT_READONLY.
568 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
571 static int mnt_make_readonly(struct mount *mnt)
575 ret = mnt_hold_writers(mnt);
577 mnt->mnt.mnt_flags |= MNT_READONLY;
578 mnt_unhold_writers(mnt);
582 int sb_prepare_remount_readonly(struct super_block *sb)
587 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
588 if (atomic_long_read(&sb->s_remove_count))
592 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
593 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
594 err = mnt_hold_writers(mnt);
599 if (!err && atomic_long_read(&sb->s_remove_count))
603 sb_start_ro_state_change(sb);
604 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
605 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
606 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
613 static void free_vfsmnt(struct mount *mnt)
615 mnt_idmap_put(mnt_idmap(&mnt->mnt));
616 kfree_const(mnt->mnt_devname);
618 free_percpu(mnt->mnt_pcp);
620 kmem_cache_free(mnt_cache, mnt);
623 static void delayed_free_vfsmnt(struct rcu_head *head)
625 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
628 /* call under rcu_read_lock */
629 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
632 if (read_seqretry(&mount_lock, seq))
636 mnt = real_mount(bastard);
637 mnt_add_count(mnt, 1);
638 smp_mb(); // see mntput_no_expire()
639 if (likely(!read_seqretry(&mount_lock, seq)))
641 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
642 mnt_add_count(mnt, -1);
646 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
647 mnt_add_count(mnt, -1);
652 /* caller will mntput() */
656 /* call under rcu_read_lock */
657 static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
659 int res = __legitimize_mnt(bastard, seq);
662 if (unlikely(res < 0)) {
671 * __lookup_mnt - find first child mount
673 * @dentry: mountpoint
675 * If @mnt has a child mount @c mounted @dentry find and return it.
677 * Note that the child mount @c need not be unique. There are cases
678 * where shadow mounts are created. For example, during mount
679 * propagation when a source mount @mnt whose root got overmounted by a
680 * mount @o after path lookup but before @namespace_sem could be
681 * acquired gets copied and propagated. So @mnt gets copied including
682 * @o. When @mnt is propagated to a destination mount @d that already
683 * has another mount @n mounted at the same mountpoint then the source
684 * mount @mnt will be tucked beneath @n, i.e., @n will be mounted on
685 * @mnt and @mnt mounted on @d. Now both @n and @o are mounted at @mnt
688 * Return: The first child of @mnt mounted @dentry or NULL.
690 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
692 struct hlist_head *head = m_hash(mnt, dentry);
695 hlist_for_each_entry_rcu(p, head, mnt_hash)
696 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
702 * lookup_mnt - Return the first child mount mounted at path
704 * "First" means first mounted chronologically. If you create the
707 * mount /dev/sda1 /mnt
708 * mount /dev/sda2 /mnt
709 * mount /dev/sda3 /mnt
711 * Then lookup_mnt() on the base /mnt dentry in the root mount will
712 * return successively the root dentry and vfsmount of /dev/sda1, then
713 * /dev/sda2, then /dev/sda3, then NULL.
715 * lookup_mnt takes a reference to the found vfsmount.
717 struct vfsmount *lookup_mnt(const struct path *path)
719 struct mount *child_mnt;
725 seq = read_seqbegin(&mount_lock);
726 child_mnt = __lookup_mnt(path->mnt, path->dentry);
727 m = child_mnt ? &child_mnt->mnt : NULL;
728 } while (!legitimize_mnt(m, seq));
733 static inline void lock_ns_list(struct mnt_namespace *ns)
735 spin_lock(&ns->ns_lock);
738 static inline void unlock_ns_list(struct mnt_namespace *ns)
740 spin_unlock(&ns->ns_lock);
743 static inline bool mnt_is_cursor(struct mount *mnt)
745 return mnt->mnt.mnt_flags & MNT_CURSOR;
749 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
750 * current mount namespace.
752 * The common case is dentries are not mountpoints at all and that
753 * test is handled inline. For the slow case when we are actually
754 * dealing with a mountpoint of some kind, walk through all of the
755 * mounts in the current mount namespace and test to see if the dentry
758 * The mount_hashtable is not usable in the context because we
759 * need to identify all mounts that may be in the current mount
760 * namespace not just a mount that happens to have some specified
763 bool __is_local_mountpoint(struct dentry *dentry)
765 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
767 bool is_covered = false;
769 down_read(&namespace_sem);
771 list_for_each_entry(mnt, &ns->list, mnt_list) {
772 if (mnt_is_cursor(mnt))
774 is_covered = (mnt->mnt_mountpoint == dentry);
779 up_read(&namespace_sem);
784 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
786 struct hlist_head *chain = mp_hash(dentry);
787 struct mountpoint *mp;
789 hlist_for_each_entry(mp, chain, m_hash) {
790 if (mp->m_dentry == dentry) {
798 static struct mountpoint *get_mountpoint(struct dentry *dentry)
800 struct mountpoint *mp, *new = NULL;
803 if (d_mountpoint(dentry)) {
804 /* might be worth a WARN_ON() */
805 if (d_unlinked(dentry))
806 return ERR_PTR(-ENOENT);
808 read_seqlock_excl(&mount_lock);
809 mp = lookup_mountpoint(dentry);
810 read_sequnlock_excl(&mount_lock);
816 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
818 return ERR_PTR(-ENOMEM);
821 /* Exactly one processes may set d_mounted */
822 ret = d_set_mounted(dentry);
824 /* Someone else set d_mounted? */
828 /* The dentry is not available as a mountpoint? */
833 /* Add the new mountpoint to the hash table */
834 read_seqlock_excl(&mount_lock);
835 new->m_dentry = dget(dentry);
837 hlist_add_head(&new->m_hash, mp_hash(dentry));
838 INIT_HLIST_HEAD(&new->m_list);
839 read_sequnlock_excl(&mount_lock);
849 * vfsmount lock must be held. Additionally, the caller is responsible
850 * for serializing calls for given disposal list.
852 static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
854 if (!--mp->m_count) {
855 struct dentry *dentry = mp->m_dentry;
856 BUG_ON(!hlist_empty(&mp->m_list));
857 spin_lock(&dentry->d_lock);
858 dentry->d_flags &= ~DCACHE_MOUNTED;
859 spin_unlock(&dentry->d_lock);
860 dput_to_list(dentry, list);
861 hlist_del(&mp->m_hash);
866 /* called with namespace_lock and vfsmount lock */
867 static void put_mountpoint(struct mountpoint *mp)
869 __put_mountpoint(mp, &ex_mountpoints);
872 static inline int check_mnt(struct mount *mnt)
874 return mnt->mnt_ns == current->nsproxy->mnt_ns;
878 * vfsmount lock must be held for write
880 static void touch_mnt_namespace(struct mnt_namespace *ns)
884 wake_up_interruptible(&ns->poll);
889 * vfsmount lock must be held for write
891 static void __touch_mnt_namespace(struct mnt_namespace *ns)
893 if (ns && ns->event != event) {
895 wake_up_interruptible(&ns->poll);
900 * vfsmount lock must be held for write
902 static struct mountpoint *unhash_mnt(struct mount *mnt)
904 struct mountpoint *mp;
905 mnt->mnt_parent = mnt;
906 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
907 list_del_init(&mnt->mnt_child);
908 hlist_del_init_rcu(&mnt->mnt_hash);
909 hlist_del_init(&mnt->mnt_mp_list);
916 * vfsmount lock must be held for write
918 static void umount_mnt(struct mount *mnt)
920 put_mountpoint(unhash_mnt(mnt));
924 * vfsmount lock must be held for write
926 void mnt_set_mountpoint(struct mount *mnt,
927 struct mountpoint *mp,
928 struct mount *child_mnt)
931 mnt_add_count(mnt, 1); /* essentially, that's mntget */
932 child_mnt->mnt_mountpoint = mp->m_dentry;
933 child_mnt->mnt_parent = mnt;
934 child_mnt->mnt_mp = mp;
935 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
939 * mnt_set_mountpoint_beneath - mount a mount beneath another one
941 * @new_parent: the source mount
942 * @top_mnt: the mount beneath which @new_parent is mounted
943 * @new_mp: the new mountpoint of @top_mnt on @new_parent
945 * Remove @top_mnt from its current mountpoint @top_mnt->mnt_mp and
946 * parent @top_mnt->mnt_parent and mount it on top of @new_parent at
947 * @new_mp. And mount @new_parent on the old parent and old
948 * mountpoint of @top_mnt.
950 * Context: This function expects namespace_lock() and lock_mount_hash()
951 * to have been acquired in that order.
953 static void mnt_set_mountpoint_beneath(struct mount *new_parent,
954 struct mount *top_mnt,
955 struct mountpoint *new_mp)
957 struct mount *old_top_parent = top_mnt->mnt_parent;
958 struct mountpoint *old_top_mp = top_mnt->mnt_mp;
960 mnt_set_mountpoint(old_top_parent, old_top_mp, new_parent);
961 mnt_change_mountpoint(new_parent, new_mp, top_mnt);
965 static void __attach_mnt(struct mount *mnt, struct mount *parent)
967 hlist_add_head_rcu(&mnt->mnt_hash,
968 m_hash(&parent->mnt, mnt->mnt_mountpoint));
969 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
973 * attach_mnt - mount a mount, attach to @mount_hashtable and parent's
974 * list of child mounts
975 * @parent: the parent
976 * @mnt: the new mount
977 * @mp: the new mountpoint
978 * @beneath: whether to mount @mnt beneath or on top of @parent
980 * If @beneath is false, mount @mnt at @mp on @parent. Then attach @mnt
981 * to @parent's child mount list and to @mount_hashtable.
983 * If @beneath is true, remove @mnt from its current parent and
984 * mountpoint and mount it on @mp on @parent, and mount @parent on the
985 * old parent and old mountpoint of @mnt. Finally, attach @parent to
986 * @mnt_hashtable and @parent->mnt_parent->mnt_mounts.
988 * Note, when __attach_mnt() is called @mnt->mnt_parent already points
989 * to the correct parent.
991 * Context: This function expects namespace_lock() and lock_mount_hash()
992 * to have been acquired in that order.
994 static void attach_mnt(struct mount *mnt, struct mount *parent,
995 struct mountpoint *mp, bool beneath)
998 mnt_set_mountpoint_beneath(mnt, parent, mp);
1000 mnt_set_mountpoint(parent, mp, mnt);
1002 * Note, @mnt->mnt_parent has to be used. If @mnt was mounted
1003 * beneath @parent then @mnt will need to be attached to
1004 * @parent's old parent, not @parent. IOW, @mnt->mnt_parent
1005 * isn't the same mount as @parent.
1007 __attach_mnt(mnt, mnt->mnt_parent);
1010 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
1012 struct mountpoint *old_mp = mnt->mnt_mp;
1013 struct mount *old_parent = mnt->mnt_parent;
1015 list_del_init(&mnt->mnt_child);
1016 hlist_del_init(&mnt->mnt_mp_list);
1017 hlist_del_init_rcu(&mnt->mnt_hash);
1019 attach_mnt(mnt, parent, mp, false);
1021 put_mountpoint(old_mp);
1022 mnt_add_count(old_parent, -1);
1026 * vfsmount lock must be held for write
1028 static void commit_tree(struct mount *mnt)
1030 struct mount *parent = mnt->mnt_parent;
1033 struct mnt_namespace *n = parent->mnt_ns;
1035 BUG_ON(parent == mnt);
1037 list_add_tail(&head, &mnt->mnt_list);
1038 list_for_each_entry(m, &head, mnt_list)
1041 list_splice(&head, n->list.prev);
1043 n->mounts += n->pending_mounts;
1044 n->pending_mounts = 0;
1046 __attach_mnt(mnt, parent);
1047 touch_mnt_namespace(n);
1050 static struct mount *next_mnt(struct mount *p, struct mount *root)
1052 struct list_head *next = p->mnt_mounts.next;
1053 if (next == &p->mnt_mounts) {
1057 next = p->mnt_child.next;
1058 if (next != &p->mnt_parent->mnt_mounts)
1063 return list_entry(next, struct mount, mnt_child);
1066 static struct mount *skip_mnt_tree(struct mount *p)
1068 struct list_head *prev = p->mnt_mounts.prev;
1069 while (prev != &p->mnt_mounts) {
1070 p = list_entry(prev, struct mount, mnt_child);
1071 prev = p->mnt_mounts.prev;
1077 * vfs_create_mount - Create a mount for a configured superblock
1078 * @fc: The configuration context with the superblock attached
1080 * Create a mount to an already configured superblock. If necessary, the
1081 * caller should invoke vfs_get_tree() before calling this.
1083 * Note that this does not attach the mount to anything.
1085 struct vfsmount *vfs_create_mount(struct fs_context *fc)
1090 return ERR_PTR(-EINVAL);
1092 mnt = alloc_vfsmnt(fc->source ?: "none");
1094 return ERR_PTR(-ENOMEM);
1096 if (fc->sb_flags & SB_KERNMOUNT)
1097 mnt->mnt.mnt_flags = MNT_INTERNAL;
1099 atomic_inc(&fc->root->d_sb->s_active);
1100 mnt->mnt.mnt_sb = fc->root->d_sb;
1101 mnt->mnt.mnt_root = dget(fc->root);
1102 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1103 mnt->mnt_parent = mnt;
1106 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1107 unlock_mount_hash();
1110 EXPORT_SYMBOL(vfs_create_mount);
1112 struct vfsmount *fc_mount(struct fs_context *fc)
1114 int err = vfs_get_tree(fc);
1116 up_write(&fc->root->d_sb->s_umount);
1117 return vfs_create_mount(fc);
1119 return ERR_PTR(err);
1121 EXPORT_SYMBOL(fc_mount);
1123 struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1124 int flags, const char *name,
1127 struct fs_context *fc;
1128 struct vfsmount *mnt;
1132 return ERR_PTR(-EINVAL);
1134 fc = fs_context_for_mount(type, flags);
1136 return ERR_CAST(fc);
1139 ret = vfs_parse_fs_string(fc, "source",
1140 name, strlen(name));
1142 ret = parse_monolithic_mount_data(fc, data);
1151 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1154 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1155 const char *name, void *data)
1157 /* Until it is worked out how to pass the user namespace
1158 * through from the parent mount to the submount don't support
1159 * unprivileged mounts with submounts.
1161 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1162 return ERR_PTR(-EPERM);
1164 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1166 EXPORT_SYMBOL_GPL(vfs_submount);
1168 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1171 struct super_block *sb = old->mnt.mnt_sb;
1175 mnt = alloc_vfsmnt(old->mnt_devname);
1177 return ERR_PTR(-ENOMEM);
1179 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1180 mnt->mnt_group_id = 0; /* not a peer of original */
1182 mnt->mnt_group_id = old->mnt_group_id;
1184 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1185 err = mnt_alloc_group_id(mnt);
1190 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1191 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1193 atomic_inc(&sb->s_active);
1194 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1196 mnt->mnt.mnt_sb = sb;
1197 mnt->mnt.mnt_root = dget(root);
1198 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1199 mnt->mnt_parent = mnt;
1201 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1202 unlock_mount_hash();
1204 if ((flag & CL_SLAVE) ||
1205 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1206 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1207 mnt->mnt_master = old;
1208 CLEAR_MNT_SHARED(mnt);
1209 } else if (!(flag & CL_PRIVATE)) {
1210 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1211 list_add(&mnt->mnt_share, &old->mnt_share);
1212 if (IS_MNT_SLAVE(old))
1213 list_add(&mnt->mnt_slave, &old->mnt_slave);
1214 mnt->mnt_master = old->mnt_master;
1216 CLEAR_MNT_SHARED(mnt);
1218 if (flag & CL_MAKE_SHARED)
1219 set_mnt_shared(mnt);
1221 /* stick the duplicate mount on the same expiry list
1222 * as the original if that was on one */
1223 if (flag & CL_EXPIRE) {
1224 if (!list_empty(&old->mnt_expire))
1225 list_add(&mnt->mnt_expire, &old->mnt_expire);
1233 return ERR_PTR(err);
1236 static void cleanup_mnt(struct mount *mnt)
1238 struct hlist_node *p;
1241 * The warning here probably indicates that somebody messed
1242 * up a mnt_want/drop_write() pair. If this happens, the
1243 * filesystem was probably unable to make r/w->r/o transitions.
1244 * The locking used to deal with mnt_count decrement provides barriers,
1245 * so mnt_get_writers() below is safe.
1247 WARN_ON(mnt_get_writers(mnt));
1248 if (unlikely(mnt->mnt_pins.first))
1250 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1251 hlist_del(&m->mnt_umount);
1254 fsnotify_vfsmount_delete(&mnt->mnt);
1255 dput(mnt->mnt.mnt_root);
1256 deactivate_super(mnt->mnt.mnt_sb);
1258 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1261 static void __cleanup_mnt(struct rcu_head *head)
1263 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1266 static LLIST_HEAD(delayed_mntput_list);
1267 static void delayed_mntput(struct work_struct *unused)
1269 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1270 struct mount *m, *t;
1272 llist_for_each_entry_safe(m, t, node, mnt_llist)
1275 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1277 static void mntput_no_expire(struct mount *mnt)
1283 if (likely(READ_ONCE(mnt->mnt_ns))) {
1285 * Since we don't do lock_mount_hash() here,
1286 * ->mnt_ns can change under us. However, if it's
1287 * non-NULL, then there's a reference that won't
1288 * be dropped until after an RCU delay done after
1289 * turning ->mnt_ns NULL. So if we observe it
1290 * non-NULL under rcu_read_lock(), the reference
1291 * we are dropping is not the final one.
1293 mnt_add_count(mnt, -1);
1299 * make sure that if __legitimize_mnt() has not seen us grab
1300 * mount_lock, we'll see their refcount increment here.
1303 mnt_add_count(mnt, -1);
1304 count = mnt_get_count(mnt);
1308 unlock_mount_hash();
1311 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1313 unlock_mount_hash();
1316 mnt->mnt.mnt_flags |= MNT_DOOMED;
1319 list_del(&mnt->mnt_instance);
1321 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1322 struct mount *p, *tmp;
1323 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1324 __put_mountpoint(unhash_mnt(p), &list);
1325 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1328 unlock_mount_hash();
1329 shrink_dentry_list(&list);
1331 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1332 struct task_struct *task = current;
1333 if (likely(!(task->flags & PF_KTHREAD))) {
1334 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1335 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1338 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1339 schedule_delayed_work(&delayed_mntput_work, 1);
1345 void mntput(struct vfsmount *mnt)
1348 struct mount *m = real_mount(mnt);
1349 /* avoid cacheline pingpong */
1350 if (unlikely(m->mnt_expiry_mark))
1351 WRITE_ONCE(m->mnt_expiry_mark, 0);
1352 mntput_no_expire(m);
1355 EXPORT_SYMBOL(mntput);
1357 struct vfsmount *mntget(struct vfsmount *mnt)
1360 mnt_add_count(real_mount(mnt), 1);
1363 EXPORT_SYMBOL(mntget);
1366 * Make a mount point inaccessible to new lookups.
1367 * Because there may still be current users, the caller MUST WAIT
1368 * for an RCU grace period before destroying the mount point.
1370 void mnt_make_shortterm(struct vfsmount *mnt)
1373 real_mount(mnt)->mnt_ns = NULL;
1377 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1378 * @path: path to check
1380 * d_mountpoint() can only be used reliably to establish if a dentry is
1381 * not mounted in any namespace and that common case is handled inline.
1382 * d_mountpoint() isn't aware of the possibility there may be multiple
1383 * mounts using a given dentry in a different namespace. This function
1384 * checks if the passed in path is a mountpoint rather than the dentry
1387 bool path_is_mountpoint(const struct path *path)
1392 if (!d_mountpoint(path->dentry))
1397 seq = read_seqbegin(&mount_lock);
1398 res = __path_is_mountpoint(path);
1399 } while (read_seqretry(&mount_lock, seq));
1404 EXPORT_SYMBOL(path_is_mountpoint);
1406 struct vfsmount *mnt_clone_internal(const struct path *path)
1409 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1412 p->mnt.mnt_flags |= MNT_INTERNAL;
1416 #ifdef CONFIG_PROC_FS
1417 static struct mount *mnt_list_next(struct mnt_namespace *ns,
1418 struct list_head *p)
1420 struct mount *mnt, *ret = NULL;
1423 list_for_each_continue(p, &ns->list) {
1424 mnt = list_entry(p, typeof(*mnt), mnt_list);
1425 if (!mnt_is_cursor(mnt)) {
1435 /* iterator; we want it to have access to namespace_sem, thus here... */
1436 static void *m_start(struct seq_file *m, loff_t *pos)
1438 struct proc_mounts *p = m->private;
1439 struct list_head *prev;
1441 down_read(&namespace_sem);
1443 prev = &p->ns->list;
1445 prev = &p->cursor.mnt_list;
1447 /* Read after we'd reached the end? */
1448 if (list_empty(prev))
1452 return mnt_list_next(p->ns, prev);
1455 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1457 struct proc_mounts *p = m->private;
1458 struct mount *mnt = v;
1461 return mnt_list_next(p->ns, &mnt->mnt_list);
1464 static void m_stop(struct seq_file *m, void *v)
1466 struct proc_mounts *p = m->private;
1467 struct mount *mnt = v;
1469 lock_ns_list(p->ns);
1471 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1473 list_del_init(&p->cursor.mnt_list);
1474 unlock_ns_list(p->ns);
1475 up_read(&namespace_sem);
1478 static int m_show(struct seq_file *m, void *v)
1480 struct proc_mounts *p = m->private;
1481 struct mount *r = v;
1482 return p->show(m, &r->mnt);
1485 const struct seq_operations mounts_op = {
1492 void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1494 down_read(&namespace_sem);
1496 list_del(&cursor->mnt_list);
1498 up_read(&namespace_sem);
1500 #endif /* CONFIG_PROC_FS */
1503 * may_umount_tree - check if a mount tree is busy
1504 * @m: root of mount tree
1506 * This is called to check if a tree of mounts has any
1507 * open files, pwds, chroots or sub mounts that are
1510 int may_umount_tree(struct vfsmount *m)
1512 struct mount *mnt = real_mount(m);
1513 int actual_refs = 0;
1514 int minimum_refs = 0;
1518 /* write lock needed for mnt_get_count */
1520 for (p = mnt; p; p = next_mnt(p, mnt)) {
1521 actual_refs += mnt_get_count(p);
1524 unlock_mount_hash();
1526 if (actual_refs > minimum_refs)
1532 EXPORT_SYMBOL(may_umount_tree);
1535 * may_umount - check if a mount point is busy
1536 * @mnt: root of mount
1538 * This is called to check if a mount point has any
1539 * open files, pwds, chroots or sub mounts. If the
1540 * mount has sub mounts this will return busy
1541 * regardless of whether the sub mounts are busy.
1543 * Doesn't take quota and stuff into account. IOW, in some cases it will
1544 * give false negatives. The main reason why it's here is that we need
1545 * a non-destructive way to look for easily umountable filesystems.
1547 int may_umount(struct vfsmount *mnt)
1550 down_read(&namespace_sem);
1552 if (propagate_mount_busy(real_mount(mnt), 2))
1554 unlock_mount_hash();
1555 up_read(&namespace_sem);
1559 EXPORT_SYMBOL(may_umount);
1561 static void namespace_unlock(void)
1563 struct hlist_head head;
1564 struct hlist_node *p;
1568 hlist_move_list(&unmounted, &head);
1569 list_splice_init(&ex_mountpoints, &list);
1571 up_write(&namespace_sem);
1573 shrink_dentry_list(&list);
1575 if (likely(hlist_empty(&head)))
1578 synchronize_rcu_expedited();
1580 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1581 hlist_del(&m->mnt_umount);
1586 static inline void namespace_lock(void)
1588 down_write(&namespace_sem);
1591 enum umount_tree_flags {
1593 UMOUNT_PROPAGATE = 2,
1594 UMOUNT_CONNECTED = 4,
1597 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1599 /* Leaving mounts connected is only valid for lazy umounts */
1600 if (how & UMOUNT_SYNC)
1603 /* A mount without a parent has nothing to be connected to */
1604 if (!mnt_has_parent(mnt))
1607 /* Because the reference counting rules change when mounts are
1608 * unmounted and connected, umounted mounts may not be
1609 * connected to mounted mounts.
1611 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1614 /* Has it been requested that the mount remain connected? */
1615 if (how & UMOUNT_CONNECTED)
1618 /* Is the mount locked such that it needs to remain connected? */
1619 if (IS_MNT_LOCKED(mnt))
1622 /* By default disconnect the mount */
1627 * mount_lock must be held
1628 * namespace_sem must be held for write
1630 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1632 LIST_HEAD(tmp_list);
1635 if (how & UMOUNT_PROPAGATE)
1636 propagate_mount_unlock(mnt);
1638 /* Gather the mounts to umount */
1639 for (p = mnt; p; p = next_mnt(p, mnt)) {
1640 p->mnt.mnt_flags |= MNT_UMOUNT;
1641 list_move(&p->mnt_list, &tmp_list);
1644 /* Hide the mounts from mnt_mounts */
1645 list_for_each_entry(p, &tmp_list, mnt_list) {
1646 list_del_init(&p->mnt_child);
1649 /* Add propogated mounts to the tmp_list */
1650 if (how & UMOUNT_PROPAGATE)
1651 propagate_umount(&tmp_list);
1653 while (!list_empty(&tmp_list)) {
1654 struct mnt_namespace *ns;
1656 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1657 list_del_init(&p->mnt_expire);
1658 list_del_init(&p->mnt_list);
1662 __touch_mnt_namespace(ns);
1665 if (how & UMOUNT_SYNC)
1666 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1668 disconnect = disconnect_mount(p, how);
1669 if (mnt_has_parent(p)) {
1670 mnt_add_count(p->mnt_parent, -1);
1672 /* Don't forget about p */
1673 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1678 change_mnt_propagation(p, MS_PRIVATE);
1680 hlist_add_head(&p->mnt_umount, &unmounted);
1684 static void shrink_submounts(struct mount *mnt);
1686 static int do_umount_root(struct super_block *sb)
1690 down_write(&sb->s_umount);
1691 if (!sb_rdonly(sb)) {
1692 struct fs_context *fc;
1694 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1699 ret = parse_monolithic_mount_data(fc, NULL);
1701 ret = reconfigure_super(fc);
1705 up_write(&sb->s_umount);
1709 static int do_umount(struct mount *mnt, int flags)
1711 struct super_block *sb = mnt->mnt.mnt_sb;
1714 retval = security_sb_umount(&mnt->mnt, flags);
1719 * Allow userspace to request a mountpoint be expired rather than
1720 * unmounting unconditionally. Unmount only happens if:
1721 * (1) the mark is already set (the mark is cleared by mntput())
1722 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1724 if (flags & MNT_EXPIRE) {
1725 if (&mnt->mnt == current->fs->root.mnt ||
1726 flags & (MNT_FORCE | MNT_DETACH))
1730 * probably don't strictly need the lock here if we examined
1731 * all race cases, but it's a slowpath.
1734 if (mnt_get_count(mnt) != 2) {
1735 unlock_mount_hash();
1738 unlock_mount_hash();
1740 if (!xchg(&mnt->mnt_expiry_mark, 1))
1745 * If we may have to abort operations to get out of this
1746 * mount, and they will themselves hold resources we must
1747 * allow the fs to do things. In the Unix tradition of
1748 * 'Gee thats tricky lets do it in userspace' the umount_begin
1749 * might fail to complete on the first run through as other tasks
1750 * must return, and the like. Thats for the mount program to worry
1751 * about for the moment.
1754 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1755 sb->s_op->umount_begin(sb);
1759 * No sense to grab the lock for this test, but test itself looks
1760 * somewhat bogus. Suggestions for better replacement?
1761 * Ho-hum... In principle, we might treat that as umount + switch
1762 * to rootfs. GC would eventually take care of the old vfsmount.
1763 * Actually it makes sense, especially if rootfs would contain a
1764 * /reboot - static binary that would close all descriptors and
1765 * call reboot(9). Then init(8) could umount root and exec /reboot.
1767 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1769 * Special case for "unmounting" root ...
1770 * we just try to remount it readonly.
1772 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1774 return do_umount_root(sb);
1780 /* Recheck MNT_LOCKED with the locks held */
1782 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1786 if (flags & MNT_DETACH) {
1787 if (!list_empty(&mnt->mnt_list))
1788 umount_tree(mnt, UMOUNT_PROPAGATE);
1791 shrink_submounts(mnt);
1793 if (!propagate_mount_busy(mnt, 2)) {
1794 if (!list_empty(&mnt->mnt_list))
1795 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1800 unlock_mount_hash();
1806 * __detach_mounts - lazily unmount all mounts on the specified dentry
1808 * During unlink, rmdir, and d_drop it is possible to loose the path
1809 * to an existing mountpoint, and wind up leaking the mount.
1810 * detach_mounts allows lazily unmounting those mounts instead of
1813 * The caller may hold dentry->d_inode->i_mutex.
1815 void __detach_mounts(struct dentry *dentry)
1817 struct mountpoint *mp;
1822 mp = lookup_mountpoint(dentry);
1827 while (!hlist_empty(&mp->m_list)) {
1828 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1829 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1831 hlist_add_head(&mnt->mnt_umount, &unmounted);
1833 else umount_tree(mnt, UMOUNT_CONNECTED);
1837 unlock_mount_hash();
1842 * Is the caller allowed to modify his namespace?
1844 bool may_mount(void)
1846 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1850 * path_mounted - check whether path is mounted
1851 * @path: path to check
1853 * Determine whether @path refers to the root of a mount.
1855 * Return: true if @path is the root of a mount, false if not.
1857 static inline bool path_mounted(const struct path *path)
1859 return path->mnt->mnt_root == path->dentry;
1862 static void warn_mandlock(void)
1864 pr_warn_once("=======================================================\n"
1865 "WARNING: The mand mount option has been deprecated and\n"
1866 " and is ignored by this kernel. Remove the mand\n"
1867 " option from the mount to silence this warning.\n"
1868 "=======================================================\n");
1871 static int can_umount(const struct path *path, int flags)
1873 struct mount *mnt = real_mount(path->mnt);
1877 if (!path_mounted(path))
1879 if (!check_mnt(mnt))
1881 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1883 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1888 // caller is responsible for flags being sane
1889 int path_umount(struct path *path, int flags)
1891 struct mount *mnt = real_mount(path->mnt);
1894 ret = can_umount(path, flags);
1896 ret = do_umount(mnt, flags);
1898 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1900 mntput_no_expire(mnt);
1904 static int ksys_umount(char __user *name, int flags)
1906 int lookup_flags = LOOKUP_MOUNTPOINT;
1910 // basic validity checks done first
1911 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1914 if (!(flags & UMOUNT_NOFOLLOW))
1915 lookup_flags |= LOOKUP_FOLLOW;
1916 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1919 return path_umount(&path, flags);
1922 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1924 return ksys_umount(name, flags);
1927 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1930 * The 2.0 compatible umount. No flags.
1932 SYSCALL_DEFINE1(oldumount, char __user *, name)
1934 return ksys_umount(name, 0);
1939 static bool is_mnt_ns_file(struct dentry *dentry)
1941 /* Is this a proxy for a mount namespace? */
1942 return dentry->d_op == &ns_dentry_operations &&
1943 dentry->d_fsdata == &mntns_operations;
1946 static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1948 return container_of(ns, struct mnt_namespace, ns);
1951 struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1956 static bool mnt_ns_loop(struct dentry *dentry)
1958 /* Could bind mounting the mount namespace inode cause a
1959 * mount namespace loop?
1961 struct mnt_namespace *mnt_ns;
1962 if (!is_mnt_ns_file(dentry))
1965 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1966 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1969 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1972 struct mount *res, *p, *q, *r, *parent;
1974 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1975 return ERR_PTR(-EINVAL);
1977 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1978 return ERR_PTR(-EINVAL);
1980 res = q = clone_mnt(mnt, dentry, flag);
1984 q->mnt_mountpoint = mnt->mnt_mountpoint;
1987 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1989 if (!is_subdir(r->mnt_mountpoint, dentry))
1992 for (s = r; s; s = next_mnt(s, r)) {
1993 if (!(flag & CL_COPY_UNBINDABLE) &&
1994 IS_MNT_UNBINDABLE(s)) {
1995 if (s->mnt.mnt_flags & MNT_LOCKED) {
1996 /* Both unbindable and locked. */
1997 q = ERR_PTR(-EPERM);
2000 s = skip_mnt_tree(s);
2004 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2005 is_mnt_ns_file(s->mnt.mnt_root)) {
2006 s = skip_mnt_tree(s);
2009 while (p != s->mnt_parent) {
2015 q = clone_mnt(p, p->mnt.mnt_root, flag);
2019 list_add_tail(&q->mnt_list, &res->mnt_list);
2020 attach_mnt(q, parent, p->mnt_mp, false);
2021 unlock_mount_hash();
2028 umount_tree(res, UMOUNT_SYNC);
2029 unlock_mount_hash();
2034 /* Caller should check returned pointer for errors */
2036 struct vfsmount *collect_mounts(const struct path *path)
2040 if (!check_mnt(real_mount(path->mnt)))
2041 tree = ERR_PTR(-EINVAL);
2043 tree = copy_tree(real_mount(path->mnt), path->dentry,
2044 CL_COPY_ALL | CL_PRIVATE);
2047 return ERR_CAST(tree);
2051 static void free_mnt_ns(struct mnt_namespace *);
2052 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2054 void dissolve_on_fput(struct vfsmount *mnt)
2056 struct mnt_namespace *ns;
2059 ns = real_mount(mnt)->mnt_ns;
2062 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2066 unlock_mount_hash();
2072 void drop_collected_mounts(struct vfsmount *mnt)
2076 umount_tree(real_mount(mnt), 0);
2077 unlock_mount_hash();
2081 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2083 struct mount *child;
2085 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2086 if (!is_subdir(child->mnt_mountpoint, dentry))
2089 if (child->mnt.mnt_flags & MNT_LOCKED)
2096 * clone_private_mount - create a private clone of a path
2097 * @path: path to clone
2099 * This creates a new vfsmount, which will be the clone of @path. The new mount
2100 * will not be attached anywhere in the namespace and will be private (i.e.
2101 * changes to the originating mount won't be propagated into this).
2103 * Release with mntput().
2105 struct vfsmount *clone_private_mount(const struct path *path)
2107 struct mount *old_mnt = real_mount(path->mnt);
2108 struct mount *new_mnt;
2110 down_read(&namespace_sem);
2111 if (IS_MNT_UNBINDABLE(old_mnt))
2114 if (!check_mnt(old_mnt))
2117 if (has_locked_children(old_mnt, path->dentry))
2120 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2121 up_read(&namespace_sem);
2123 if (IS_ERR(new_mnt))
2124 return ERR_CAST(new_mnt);
2126 /* Longterm mount to be removed by kern_unmount*() */
2127 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2129 return &new_mnt->mnt;
2132 up_read(&namespace_sem);
2133 return ERR_PTR(-EINVAL);
2135 EXPORT_SYMBOL_GPL(clone_private_mount);
2137 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2138 struct vfsmount *root)
2141 int res = f(root, arg);
2144 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2145 res = f(&mnt->mnt, arg);
2152 static void lock_mnt_tree(struct mount *mnt)
2156 for (p = mnt; p; p = next_mnt(p, mnt)) {
2157 int flags = p->mnt.mnt_flags;
2158 /* Don't allow unprivileged users to change mount flags */
2159 flags |= MNT_LOCK_ATIME;
2161 if (flags & MNT_READONLY)
2162 flags |= MNT_LOCK_READONLY;
2164 if (flags & MNT_NODEV)
2165 flags |= MNT_LOCK_NODEV;
2167 if (flags & MNT_NOSUID)
2168 flags |= MNT_LOCK_NOSUID;
2170 if (flags & MNT_NOEXEC)
2171 flags |= MNT_LOCK_NOEXEC;
2172 /* Don't allow unprivileged users to reveal what is under a mount */
2173 if (list_empty(&p->mnt_expire))
2174 flags |= MNT_LOCKED;
2175 p->mnt.mnt_flags = flags;
2179 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2183 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2184 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2185 mnt_release_group_id(p);
2189 static int invent_group_ids(struct mount *mnt, bool recurse)
2193 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2194 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2195 int err = mnt_alloc_group_id(p);
2197 cleanup_group_ids(mnt, p);
2206 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2208 unsigned int max = READ_ONCE(sysctl_mount_max);
2209 unsigned int mounts = 0;
2212 if (ns->mounts >= max)
2215 if (ns->pending_mounts >= max)
2217 max -= ns->pending_mounts;
2219 for (p = mnt; p; p = next_mnt(p, mnt))
2225 ns->pending_mounts += mounts;
2229 enum mnt_tree_flags_t {
2230 MNT_TREE_MOVE = BIT(0),
2231 MNT_TREE_BENEATH = BIT(1),
2235 * attach_recursive_mnt - attach a source mount tree
2236 * @source_mnt: mount tree to be attached
2237 * @top_mnt: mount that @source_mnt will be mounted on or mounted beneath
2238 * @dest_mp: the mountpoint @source_mnt will be mounted at
2239 * @flags: modify how @source_mnt is supposed to be attached
2241 * NOTE: in the table below explains the semantics when a source mount
2242 * of a given type is attached to a destination mount of a given type.
2243 * ---------------------------------------------------------------------------
2244 * | BIND MOUNT OPERATION |
2245 * |**************************************************************************
2246 * | source-->| shared | private | slave | unbindable |
2250 * |**************************************************************************
2251 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2253 * |non-shared| shared (+) | private | slave (*) | invalid |
2254 * ***************************************************************************
2255 * A bind operation clones the source mount and mounts the clone on the
2256 * destination mount.
2258 * (++) the cloned mount is propagated to all the mounts in the propagation
2259 * tree of the destination mount and the cloned mount is added to
2260 * the peer group of the source mount.
2261 * (+) the cloned mount is created under the destination mount and is marked
2262 * as shared. The cloned mount is added to the peer group of the source
2264 * (+++) the mount is propagated to all the mounts in the propagation tree
2265 * of the destination mount and the cloned mount is made slave
2266 * of the same master as that of the source mount. The cloned mount
2267 * is marked as 'shared and slave'.
2268 * (*) the cloned mount is made a slave of the same master as that of the
2271 * ---------------------------------------------------------------------------
2272 * | MOVE MOUNT OPERATION |
2273 * |**************************************************************************
2274 * | source-->| shared | private | slave | unbindable |
2278 * |**************************************************************************
2279 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2281 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2282 * ***************************************************************************
2284 * (+) the mount is moved to the destination. And is then propagated to
2285 * all the mounts in the propagation tree of the destination mount.
2286 * (+*) the mount is moved to the destination.
2287 * (+++) the mount is moved to the destination and is then propagated to
2288 * all the mounts belonging to the destination mount's propagation tree.
2289 * the mount is marked as 'shared and slave'.
2290 * (*) the mount continues to be a slave at the new location.
2292 * if the source mount is a tree, the operations explained above is
2293 * applied to each mount in the tree.
2294 * Must be called without spinlocks held, since this function can sleep
2297 * Context: The function expects namespace_lock() to be held.
2298 * Return: If @source_mnt was successfully attached 0 is returned.
2299 * Otherwise a negative error code is returned.
2301 static int attach_recursive_mnt(struct mount *source_mnt,
2302 struct mount *top_mnt,
2303 struct mountpoint *dest_mp,
2304 enum mnt_tree_flags_t flags)
2306 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2307 HLIST_HEAD(tree_list);
2308 struct mnt_namespace *ns = top_mnt->mnt_ns;
2309 struct mountpoint *smp;
2310 struct mount *child, *dest_mnt, *p;
2311 struct hlist_node *n;
2313 bool moving = flags & MNT_TREE_MOVE, beneath = flags & MNT_TREE_BENEATH;
2316 * Preallocate a mountpoint in case the new mounts need to be
2317 * mounted beneath mounts on the same mountpoint.
2319 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2321 return PTR_ERR(smp);
2323 /* Is there space to add these mounts to the mount namespace? */
2325 err = count_mounts(ns, source_mnt);
2331 dest_mnt = top_mnt->mnt_parent;
2335 if (IS_MNT_SHARED(dest_mnt)) {
2336 err = invent_group_ids(source_mnt, true);
2339 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2343 goto out_cleanup_ids;
2345 if (IS_MNT_SHARED(dest_mnt)) {
2346 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2353 unhash_mnt(source_mnt);
2354 attach_mnt(source_mnt, top_mnt, dest_mp, beneath);
2355 touch_mnt_namespace(source_mnt->mnt_ns);
2357 if (source_mnt->mnt_ns) {
2358 /* move from anon - the caller will destroy */
2359 list_del_init(&source_mnt->mnt_ns->list);
2362 mnt_set_mountpoint_beneath(source_mnt, top_mnt, smp);
2364 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2365 commit_tree(source_mnt);
2368 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2370 hlist_del_init(&child->mnt_hash);
2371 q = __lookup_mnt(&child->mnt_parent->mnt,
2372 child->mnt_mountpoint);
2374 mnt_change_mountpoint(child, smp, q);
2375 /* Notice when we are propagating across user namespaces */
2376 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2377 lock_mnt_tree(child);
2378 child->mnt.mnt_flags &= ~MNT_LOCKED;
2381 put_mountpoint(smp);
2382 unlock_mount_hash();
2387 while (!hlist_empty(&tree_list)) {
2388 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2389 child->mnt_parent->mnt_ns->pending_mounts = 0;
2390 umount_tree(child, UMOUNT_SYNC);
2392 unlock_mount_hash();
2393 cleanup_group_ids(source_mnt, NULL);
2395 ns->pending_mounts = 0;
2397 read_seqlock_excl(&mount_lock);
2398 put_mountpoint(smp);
2399 read_sequnlock_excl(&mount_lock);
2405 * do_lock_mount - lock mount and mountpoint
2406 * @path: target path
2407 * @beneath: whether the intention is to mount beneath @path
2409 * Follow the mount stack on @path until the top mount @mnt is found. If
2410 * the initial @path->{mnt,dentry} is a mountpoint lookup the first
2411 * mount stacked on top of it. Then simply follow @{mnt,mnt->mnt_root}
2412 * until nothing is stacked on top of it anymore.
2414 * Acquire the inode_lock() on the top mount's ->mnt_root to protect
2415 * against concurrent removal of the new mountpoint from another mount
2418 * If @beneath is requested, acquire inode_lock() on @mnt's mountpoint
2419 * @mp on @mnt->mnt_parent must be acquired. This protects against a
2420 * concurrent unlink of @mp->mnt_dentry from another mount namespace
2421 * where @mnt doesn't have a child mount mounted @mp. A concurrent
2422 * removal of @mnt->mnt_root doesn't matter as nothing will be mounted
2423 * on top of it for @beneath.
2425 * In addition, @beneath needs to make sure that @mnt hasn't been
2426 * unmounted or moved from its current mountpoint in between dropping
2427 * @mount_lock and acquiring @namespace_sem. For the !@beneath case @mnt
2428 * being unmounted would be detected later by e.g., calling
2429 * check_mnt(mnt) in the function it's called from. For the @beneath
2430 * case however, it's useful to detect it directly in do_lock_mount().
2431 * If @mnt hasn't been unmounted then @mnt->mnt_mountpoint still points
2432 * to @mnt->mnt_mp->m_dentry. But if @mnt has been unmounted it will
2433 * point to @mnt->mnt_root and @mnt->mnt_mp will be NULL.
2435 * Return: Either the target mountpoint on the top mount or the top
2436 * mount's mountpoint.
2438 static struct mountpoint *do_lock_mount(struct path *path, bool beneath)
2440 struct vfsmount *mnt = path->mnt;
2441 struct dentry *dentry;
2442 struct mountpoint *mp = ERR_PTR(-ENOENT);
2448 m = real_mount(mnt);
2449 read_seqlock_excl(&mount_lock);
2450 dentry = dget(m->mnt_mountpoint);
2451 read_sequnlock_excl(&mount_lock);
2453 dentry = path->dentry;
2456 inode_lock(dentry->d_inode);
2457 if (unlikely(cant_mount(dentry))) {
2458 inode_unlock(dentry->d_inode);
2464 if (beneath && (!is_mounted(mnt) || m->mnt_mountpoint != dentry)) {
2466 inode_unlock(dentry->d_inode);
2470 mnt = lookup_mnt(path);
2475 inode_unlock(dentry->d_inode);
2480 path->dentry = dget(mnt->mnt_root);
2483 mp = get_mountpoint(dentry);
2486 inode_unlock(dentry->d_inode);
2496 static inline struct mountpoint *lock_mount(struct path *path)
2498 return do_lock_mount(path, false);
2501 static void unlock_mount(struct mountpoint *where)
2503 struct dentry *dentry = where->m_dentry;
2505 read_seqlock_excl(&mount_lock);
2506 put_mountpoint(where);
2507 read_sequnlock_excl(&mount_lock);
2510 inode_unlock(dentry->d_inode);
2513 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2515 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2518 if (d_is_dir(mp->m_dentry) !=
2519 d_is_dir(mnt->mnt.mnt_root))
2522 return attach_recursive_mnt(mnt, p, mp, 0);
2526 * Sanity check the flags to change_mnt_propagation.
2529 static int flags_to_propagation_type(int ms_flags)
2531 int type = ms_flags & ~(MS_REC | MS_SILENT);
2533 /* Fail if any non-propagation flags are set */
2534 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2536 /* Only one propagation flag should be set */
2537 if (!is_power_of_2(type))
2543 * recursively change the type of the mountpoint.
2545 static int do_change_type(struct path *path, int ms_flags)
2548 struct mount *mnt = real_mount(path->mnt);
2549 int recurse = ms_flags & MS_REC;
2553 if (!path_mounted(path))
2556 type = flags_to_propagation_type(ms_flags);
2561 if (type == MS_SHARED) {
2562 err = invent_group_ids(mnt, recurse);
2568 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2569 change_mnt_propagation(m, type);
2570 unlock_mount_hash();
2577 static struct mount *__do_loopback(struct path *old_path, int recurse)
2579 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2581 if (IS_MNT_UNBINDABLE(old))
2584 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2587 if (!recurse && has_locked_children(old, old_path->dentry))
2591 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2593 mnt = clone_mnt(old, old_path->dentry, 0);
2596 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2602 * do loopback mount.
2604 static int do_loopback(struct path *path, const char *old_name,
2607 struct path old_path;
2608 struct mount *mnt = NULL, *parent;
2609 struct mountpoint *mp;
2611 if (!old_name || !*old_name)
2613 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2618 if (mnt_ns_loop(old_path.dentry))
2621 mp = lock_mount(path);
2627 parent = real_mount(path->mnt);
2628 if (!check_mnt(parent))
2631 mnt = __do_loopback(&old_path, recurse);
2637 err = graft_tree(mnt, parent, mp);
2640 umount_tree(mnt, UMOUNT_SYNC);
2641 unlock_mount_hash();
2646 path_put(&old_path);
2650 static struct file *open_detached_copy(struct path *path, bool recursive)
2652 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2653 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2654 struct mount *mnt, *p;
2658 return ERR_CAST(ns);
2661 mnt = __do_loopback(path, recursive);
2665 return ERR_CAST(mnt);
2669 for (p = mnt; p; p = next_mnt(p, mnt)) {
2674 list_add_tail(&ns->list, &mnt->mnt_list);
2676 unlock_mount_hash();
2680 path->mnt = &mnt->mnt;
2681 file = dentry_open(path, O_PATH, current_cred());
2683 dissolve_on_fput(path->mnt);
2685 file->f_mode |= FMODE_NEED_UNMOUNT;
2689 SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2693 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2694 bool detached = flags & OPEN_TREE_CLONE;
2698 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2700 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2701 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2705 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2708 if (flags & AT_NO_AUTOMOUNT)
2709 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2710 if (flags & AT_SYMLINK_NOFOLLOW)
2711 lookup_flags &= ~LOOKUP_FOLLOW;
2712 if (flags & AT_EMPTY_PATH)
2713 lookup_flags |= LOOKUP_EMPTY;
2715 if (detached && !may_mount())
2718 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2722 error = user_path_at(dfd, filename, lookup_flags, &path);
2723 if (unlikely(error)) {
2724 file = ERR_PTR(error);
2727 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2729 file = dentry_open(&path, O_PATH, current_cred());
2734 return PTR_ERR(file);
2736 fd_install(fd, file);
2741 * Don't allow locked mount flags to be cleared.
2743 * No locks need to be held here while testing the various MNT_LOCK
2744 * flags because those flags can never be cleared once they are set.
2746 static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2748 unsigned int fl = mnt->mnt.mnt_flags;
2750 if ((fl & MNT_LOCK_READONLY) &&
2751 !(mnt_flags & MNT_READONLY))
2754 if ((fl & MNT_LOCK_NODEV) &&
2755 !(mnt_flags & MNT_NODEV))
2758 if ((fl & MNT_LOCK_NOSUID) &&
2759 !(mnt_flags & MNT_NOSUID))
2762 if ((fl & MNT_LOCK_NOEXEC) &&
2763 !(mnt_flags & MNT_NOEXEC))
2766 if ((fl & MNT_LOCK_ATIME) &&
2767 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2773 static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2775 bool readonly_request = (mnt_flags & MNT_READONLY);
2777 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2780 if (readonly_request)
2781 return mnt_make_readonly(mnt);
2783 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2787 static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2789 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2790 mnt->mnt.mnt_flags = mnt_flags;
2791 touch_mnt_namespace(mnt->mnt_ns);
2794 static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2796 struct super_block *sb = mnt->mnt_sb;
2798 if (!__mnt_is_readonly(mnt) &&
2799 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2800 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2801 char *buf = (char *)__get_free_page(GFP_KERNEL);
2802 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2804 pr_warn("%s filesystem being %s at %s supports timestamps until %ptTd (0x%llx)\n",
2806 is_mounted(mnt) ? "remounted" : "mounted",
2807 mntpath, &sb->s_time_max,
2808 (unsigned long long)sb->s_time_max);
2810 free_page((unsigned long)buf);
2811 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2816 * Handle reconfiguration of the mountpoint only without alteration of the
2817 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2820 static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2822 struct super_block *sb = path->mnt->mnt_sb;
2823 struct mount *mnt = real_mount(path->mnt);
2826 if (!check_mnt(mnt))
2829 if (!path_mounted(path))
2832 if (!can_change_locked_flags(mnt, mnt_flags))
2836 * We're only checking whether the superblock is read-only not
2837 * changing it, so only take down_read(&sb->s_umount).
2839 down_read(&sb->s_umount);
2841 ret = change_mount_ro_state(mnt, mnt_flags);
2843 set_mount_attributes(mnt, mnt_flags);
2844 unlock_mount_hash();
2845 up_read(&sb->s_umount);
2847 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2853 * change filesystem flags. dir should be a physical root of filesystem.
2854 * If you've mounted a non-root directory somewhere and want to do remount
2855 * on it - tough luck.
2857 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2858 int mnt_flags, void *data)
2861 struct super_block *sb = path->mnt->mnt_sb;
2862 struct mount *mnt = real_mount(path->mnt);
2863 struct fs_context *fc;
2865 if (!check_mnt(mnt))
2868 if (!path_mounted(path))
2871 if (!can_change_locked_flags(mnt, mnt_flags))
2874 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2879 err = parse_monolithic_mount_data(fc, data);
2881 down_write(&sb->s_umount);
2883 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2884 err = reconfigure_super(fc);
2887 set_mount_attributes(mnt, mnt_flags);
2888 unlock_mount_hash();
2891 up_write(&sb->s_umount);
2894 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2900 static inline int tree_contains_unbindable(struct mount *mnt)
2903 for (p = mnt; p; p = next_mnt(p, mnt)) {
2904 if (IS_MNT_UNBINDABLE(p))
2911 * Check that there aren't references to earlier/same mount namespaces in the
2912 * specified subtree. Such references can act as pins for mount namespaces
2913 * that aren't checked by the mount-cycle checking code, thereby allowing
2914 * cycles to be made.
2916 static bool check_for_nsfs_mounts(struct mount *subtree)
2922 for (p = subtree; p; p = next_mnt(p, subtree))
2923 if (mnt_ns_loop(p->mnt.mnt_root))
2928 unlock_mount_hash();
2932 static int do_set_group(struct path *from_path, struct path *to_path)
2934 struct mount *from, *to;
2937 from = real_mount(from_path->mnt);
2938 to = real_mount(to_path->mnt);
2943 /* To and From must be mounted */
2944 if (!is_mounted(&from->mnt))
2946 if (!is_mounted(&to->mnt))
2950 /* We should be allowed to modify mount namespaces of both mounts */
2951 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2953 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2957 /* To and From paths should be mount roots */
2958 if (!path_mounted(from_path))
2960 if (!path_mounted(to_path))
2963 /* Setting sharing groups is only allowed across same superblock */
2964 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2967 /* From mount root should be wider than To mount root */
2968 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2971 /* From mount should not have locked children in place of To's root */
2972 if (has_locked_children(from, to->mnt.mnt_root))
2975 /* Setting sharing groups is only allowed on private mounts */
2976 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2979 /* From should not be private */
2980 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2983 if (IS_MNT_SLAVE(from)) {
2984 struct mount *m = from->mnt_master;
2986 list_add(&to->mnt_slave, &m->mnt_slave_list);
2990 if (IS_MNT_SHARED(from)) {
2991 to->mnt_group_id = from->mnt_group_id;
2992 list_add(&to->mnt_share, &from->mnt_share);
2995 unlock_mount_hash();
3005 * path_overmounted - check if path is overmounted
3006 * @path: path to check
3008 * Check if path is overmounted, i.e., if there's a mount on top of
3009 * @path->mnt with @path->dentry as mountpoint.
3011 * Context: This function expects namespace_lock() to be held.
3012 * Return: If path is overmounted true is returned, false if not.
3014 static inline bool path_overmounted(const struct path *path)
3017 if (unlikely(__lookup_mnt(path->mnt, path->dentry))) {
3026 * can_move_mount_beneath - check that we can mount beneath the top mount
3027 * @from: mount to mount beneath
3028 * @to: mount under which to mount
3030 * - Make sure that @to->dentry is actually the root of a mount under
3031 * which we can mount another mount.
3032 * - Make sure that nothing can be mounted beneath the caller's current
3033 * root or the rootfs of the namespace.
3034 * - Make sure that the caller can unmount the topmost mount ensuring
3035 * that the caller could reveal the underlying mountpoint.
3036 * - Ensure that nothing has been mounted on top of @from before we
3037 * grabbed @namespace_sem to avoid creating pointless shadow mounts.
3038 * - Prevent mounting beneath a mount if the propagation relationship
3039 * between the source mount, parent mount, and top mount would lead to
3040 * nonsensical mount trees.
3042 * Context: This function expects namespace_lock() to be held.
3043 * Return: On success 0, and on error a negative error code is returned.
3045 static int can_move_mount_beneath(const struct path *from,
3046 const struct path *to,
3047 const struct mountpoint *mp)
3049 struct mount *mnt_from = real_mount(from->mnt),
3050 *mnt_to = real_mount(to->mnt),
3051 *parent_mnt_to = mnt_to->mnt_parent;
3053 if (!mnt_has_parent(mnt_to))
3056 if (!path_mounted(to))
3059 if (IS_MNT_LOCKED(mnt_to))
3062 /* Avoid creating shadow mounts during mount propagation. */
3063 if (path_overmounted(from))
3067 * Mounting beneath the rootfs only makes sense when the
3068 * semantics of pivot_root(".", ".") are used.
3070 if (&mnt_to->mnt == current->fs->root.mnt)
3072 if (parent_mnt_to == current->nsproxy->mnt_ns->root)
3075 for (struct mount *p = mnt_from; mnt_has_parent(p); p = p->mnt_parent)
3080 * If the parent mount propagates to the child mount this would
3081 * mean mounting @mnt_from on @mnt_to->mnt_parent and then
3082 * propagating a copy @c of @mnt_from on top of @mnt_to. This
3083 * defeats the whole purpose of mounting beneath another mount.
3085 if (propagation_would_overmount(parent_mnt_to, mnt_to, mp))
3089 * If @mnt_to->mnt_parent propagates to @mnt_from this would
3090 * mean propagating a copy @c of @mnt_from on top of @mnt_from.
3091 * Afterwards @mnt_from would be mounted on top of
3092 * @mnt_to->mnt_parent and @mnt_to would be unmounted from
3093 * @mnt->mnt_parent and remounted on @mnt_from. But since @c is
3094 * already mounted on @mnt_from, @mnt_to would ultimately be
3095 * remounted on top of @c. Afterwards, @mnt_from would be
3096 * covered by a copy @c of @mnt_from and @c would be covered by
3097 * @mnt_from itself. This defeats the whole purpose of mounting
3098 * @mnt_from beneath @mnt_to.
3100 if (propagation_would_overmount(parent_mnt_to, mnt_from, mp))
3106 static int do_move_mount(struct path *old_path, struct path *new_path,
3109 struct mnt_namespace *ns;
3112 struct mount *parent;
3113 struct mountpoint *mp, *old_mp;
3116 enum mnt_tree_flags_t flags = 0;
3118 mp = do_lock_mount(new_path, beneath);
3122 old = real_mount(old_path->mnt);
3123 p = real_mount(new_path->mnt);
3124 parent = old->mnt_parent;
3125 attached = mnt_has_parent(old);
3127 flags |= MNT_TREE_MOVE;
3128 old_mp = old->mnt_mp;
3132 /* The mountpoint must be in our namespace. */
3136 /* The thing moved must be mounted... */
3137 if (!is_mounted(&old->mnt))
3140 /* ... and either ours or the root of anon namespace */
3141 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
3144 if (old->mnt.mnt_flags & MNT_LOCKED)
3147 if (!path_mounted(old_path))
3150 if (d_is_dir(new_path->dentry) !=
3151 d_is_dir(old_path->dentry))
3154 * Don't move a mount residing in a shared parent.
3156 if (attached && IS_MNT_SHARED(parent))
3160 err = can_move_mount_beneath(old_path, new_path, mp);
3166 flags |= MNT_TREE_BENEATH;
3170 * Don't move a mount tree containing unbindable mounts to a destination
3171 * mount which is shared.
3173 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
3176 if (!check_for_nsfs_mounts(old))
3178 for (; mnt_has_parent(p); p = p->mnt_parent)
3182 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, flags);
3186 /* if the mount is moved, it should no longer be expire
3188 list_del_init(&old->mnt_expire);
3190 put_mountpoint(old_mp);
3195 mntput_no_expire(parent);
3202 static int do_move_mount_old(struct path *path, const char *old_name)
3204 struct path old_path;
3207 if (!old_name || !*old_name)
3210 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3214 err = do_move_mount(&old_path, path, false);
3215 path_put(&old_path);
3220 * add a mount into a namespace's mount tree
3222 static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
3223 const struct path *path, int mnt_flags)
3225 struct mount *parent = real_mount(path->mnt);
3227 mnt_flags &= ~MNT_INTERNAL_FLAGS;
3229 if (unlikely(!check_mnt(parent))) {
3230 /* that's acceptable only for automounts done in private ns */
3231 if (!(mnt_flags & MNT_SHRINKABLE))
3233 /* ... and for those we'd better have mountpoint still alive */
3234 if (!parent->mnt_ns)
3238 /* Refuse the same filesystem on the same mount point */
3239 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && path_mounted(path))
3242 if (d_is_symlink(newmnt->mnt.mnt_root))
3245 newmnt->mnt.mnt_flags = mnt_flags;
3246 return graft_tree(newmnt, parent, mp);
3249 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3252 * Create a new mount using a superblock configuration and request it
3253 * be added to the namespace tree.
3255 static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3256 unsigned int mnt_flags)
3258 struct vfsmount *mnt;
3259 struct mountpoint *mp;
3260 struct super_block *sb = fc->root->d_sb;
3263 error = security_sb_kern_mount(sb);
3264 if (!error && mount_too_revealing(sb, &mnt_flags))
3267 if (unlikely(error)) {
3272 up_write(&sb->s_umount);
3274 mnt = vfs_create_mount(fc);
3276 return PTR_ERR(mnt);
3278 mnt_warn_timestamp_expiry(mountpoint, mnt);
3280 mp = lock_mount(mountpoint);
3285 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3293 * create a new mount for userspace and request it to be added into the
3296 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3297 int mnt_flags, const char *name, void *data)
3299 struct file_system_type *type;
3300 struct fs_context *fc;
3301 const char *subtype = NULL;
3307 type = get_fs_type(fstype);
3311 if (type->fs_flags & FS_HAS_SUBTYPE) {
3312 subtype = strchr(fstype, '.');
3316 put_filesystem(type);
3322 fc = fs_context_for_mount(type, sb_flags);
3323 put_filesystem(type);
3328 err = vfs_parse_fs_string(fc, "subtype",
3329 subtype, strlen(subtype));
3331 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3333 err = parse_monolithic_mount_data(fc, data);
3334 if (!err && !mount_capable(fc))
3337 err = vfs_get_tree(fc);
3339 err = do_new_mount_fc(fc, path, mnt_flags);
3345 int finish_automount(struct vfsmount *m, const struct path *path)
3347 struct dentry *dentry = path->dentry;
3348 struct mountpoint *mp;
3357 mnt = real_mount(m);
3358 /* The new mount record should have at least 2 refs to prevent it being
3359 * expired before we get a chance to add it
3361 BUG_ON(mnt_get_count(mnt) < 2);
3363 if (m->mnt_sb == path->mnt->mnt_sb &&
3364 m->mnt_root == dentry) {
3370 * we don't want to use lock_mount() - in this case finding something
3371 * that overmounts our mountpoint to be means "quitely drop what we've
3372 * got", not "try to mount it on top".
3374 inode_lock(dentry->d_inode);
3376 if (unlikely(cant_mount(dentry))) {
3378 goto discard_locked;
3380 if (path_overmounted(path)) {
3382 goto discard_locked;
3384 mp = get_mountpoint(dentry);
3387 goto discard_locked;
3390 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3399 inode_unlock(dentry->d_inode);
3401 /* remove m from any expiration list it may be on */
3402 if (!list_empty(&mnt->mnt_expire)) {
3404 list_del_init(&mnt->mnt_expire);
3413 * mnt_set_expiry - Put a mount on an expiration list
3414 * @mnt: The mount to list.
3415 * @expiry_list: The list to add the mount to.
3417 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3421 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3425 EXPORT_SYMBOL(mnt_set_expiry);
3428 * process a list of expirable mountpoints with the intent of discarding any
3429 * mountpoints that aren't in use and haven't been touched since last we came
3432 void mark_mounts_for_expiry(struct list_head *mounts)
3434 struct mount *mnt, *next;
3435 LIST_HEAD(graveyard);
3437 if (list_empty(mounts))
3443 /* extract from the expiration list every vfsmount that matches the
3444 * following criteria:
3445 * - only referenced by its parent vfsmount
3446 * - still marked for expiry (marked on the last call here; marks are
3447 * cleared by mntput())
3449 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3450 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3451 propagate_mount_busy(mnt, 1))
3453 list_move(&mnt->mnt_expire, &graveyard);
3455 while (!list_empty(&graveyard)) {
3456 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3457 touch_mnt_namespace(mnt->mnt_ns);
3458 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3460 unlock_mount_hash();
3464 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3467 * Ripoff of 'select_parent()'
3469 * search the list of submounts for a given mountpoint, and move any
3470 * shrinkable submounts to the 'graveyard' list.
3472 static int select_submounts(struct mount *parent, struct list_head *graveyard)
3474 struct mount *this_parent = parent;
3475 struct list_head *next;
3479 next = this_parent->mnt_mounts.next;
3481 while (next != &this_parent->mnt_mounts) {
3482 struct list_head *tmp = next;
3483 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3486 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3489 * Descend a level if the d_mounts list is non-empty.
3491 if (!list_empty(&mnt->mnt_mounts)) {
3496 if (!propagate_mount_busy(mnt, 1)) {
3497 list_move_tail(&mnt->mnt_expire, graveyard);
3502 * All done at this level ... ascend and resume the search
3504 if (this_parent != parent) {
3505 next = this_parent->mnt_child.next;
3506 this_parent = this_parent->mnt_parent;
3513 * process a list of expirable mountpoints with the intent of discarding any
3514 * submounts of a specific parent mountpoint
3516 * mount_lock must be held for write
3518 static void shrink_submounts(struct mount *mnt)
3520 LIST_HEAD(graveyard);
3523 /* extract submounts of 'mountpoint' from the expiration list */
3524 while (select_submounts(mnt, &graveyard)) {
3525 while (!list_empty(&graveyard)) {
3526 m = list_first_entry(&graveyard, struct mount,
3528 touch_mnt_namespace(m->mnt_ns);
3529 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3534 static void *copy_mount_options(const void __user * data)
3537 unsigned left, offset;
3542 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3544 return ERR_PTR(-ENOMEM);
3546 left = copy_from_user(copy, data, PAGE_SIZE);
3549 * Not all architectures have an exact copy_from_user(). Resort to
3552 offset = PAGE_SIZE - left;
3555 if (get_user(c, (const char __user *)data + offset))
3562 if (left == PAGE_SIZE) {
3564 return ERR_PTR(-EFAULT);
3570 static char *copy_mount_string(const void __user *data)
3572 return data ? strndup_user(data, PATH_MAX) : NULL;
3576 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3577 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3579 * data is a (void *) that can point to any structure up to
3580 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3581 * information (or be NULL).
3583 * Pre-0.97 versions of mount() didn't have a flags word.
3584 * When the flags word was introduced its top half was required
3585 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3586 * Therefore, if this magic number is present, it carries no information
3587 * and must be discarded.
3589 int path_mount(const char *dev_name, struct path *path,
3590 const char *type_page, unsigned long flags, void *data_page)
3592 unsigned int mnt_flags = 0, sb_flags;
3596 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3597 flags &= ~MS_MGC_MSK;
3599 /* Basic sanity checks */
3601 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3603 if (flags & MS_NOUSER)
3606 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3611 if (flags & SB_MANDLOCK)
3614 /* Default to relatime unless overriden */
3615 if (!(flags & MS_NOATIME))
3616 mnt_flags |= MNT_RELATIME;
3618 /* Separate the per-mountpoint flags */
3619 if (flags & MS_NOSUID)
3620 mnt_flags |= MNT_NOSUID;
3621 if (flags & MS_NODEV)
3622 mnt_flags |= MNT_NODEV;
3623 if (flags & MS_NOEXEC)
3624 mnt_flags |= MNT_NOEXEC;
3625 if (flags & MS_NOATIME)
3626 mnt_flags |= MNT_NOATIME;
3627 if (flags & MS_NODIRATIME)
3628 mnt_flags |= MNT_NODIRATIME;
3629 if (flags & MS_STRICTATIME)
3630 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3631 if (flags & MS_RDONLY)
3632 mnt_flags |= MNT_READONLY;
3633 if (flags & MS_NOSYMFOLLOW)
3634 mnt_flags |= MNT_NOSYMFOLLOW;
3636 /* The default atime for remount is preservation */
3637 if ((flags & MS_REMOUNT) &&
3638 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3639 MS_STRICTATIME)) == 0)) {
3640 mnt_flags &= ~MNT_ATIME_MASK;
3641 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3644 sb_flags = flags & (SB_RDONLY |
3653 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3654 return do_reconfigure_mnt(path, mnt_flags);
3655 if (flags & MS_REMOUNT)
3656 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3657 if (flags & MS_BIND)
3658 return do_loopback(path, dev_name, flags & MS_REC);
3659 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3660 return do_change_type(path, flags);
3661 if (flags & MS_MOVE)
3662 return do_move_mount_old(path, dev_name);
3664 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3668 long do_mount(const char *dev_name, const char __user *dir_name,
3669 const char *type_page, unsigned long flags, void *data_page)
3674 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3677 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3682 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3684 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3687 static void dec_mnt_namespaces(struct ucounts *ucounts)
3689 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3692 static void free_mnt_ns(struct mnt_namespace *ns)
3694 if (!is_anon_ns(ns))
3695 ns_free_inum(&ns->ns);
3696 dec_mnt_namespaces(ns->ucounts);
3697 put_user_ns(ns->user_ns);
3702 * Assign a sequence number so we can detect when we attempt to bind
3703 * mount a reference to an older mount namespace into the current
3704 * mount namespace, preventing reference counting loops. A 64bit
3705 * number incrementing at 10Ghz will take 12,427 years to wrap which
3706 * is effectively never, so we can ignore the possibility.
3708 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3710 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3712 struct mnt_namespace *new_ns;
3713 struct ucounts *ucounts;
3716 ucounts = inc_mnt_namespaces(user_ns);
3718 return ERR_PTR(-ENOSPC);
3720 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3722 dec_mnt_namespaces(ucounts);
3723 return ERR_PTR(-ENOMEM);
3726 ret = ns_alloc_inum(&new_ns->ns);
3729 dec_mnt_namespaces(ucounts);
3730 return ERR_PTR(ret);
3733 new_ns->ns.ops = &mntns_operations;
3735 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3736 refcount_set(&new_ns->ns.count, 1);
3737 INIT_LIST_HEAD(&new_ns->list);
3738 init_waitqueue_head(&new_ns->poll);
3739 spin_lock_init(&new_ns->ns_lock);
3740 new_ns->user_ns = get_user_ns(user_ns);
3741 new_ns->ucounts = ucounts;
3746 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3747 struct user_namespace *user_ns, struct fs_struct *new_fs)
3749 struct mnt_namespace *new_ns;
3750 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3751 struct mount *p, *q;
3758 if (likely(!(flags & CLONE_NEWNS))) {
3765 new_ns = alloc_mnt_ns(user_ns, false);
3770 /* First pass: copy the tree topology */
3771 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3772 if (user_ns != ns->user_ns)
3773 copy_flags |= CL_SHARED_TO_SLAVE;
3774 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3777 free_mnt_ns(new_ns);
3778 return ERR_CAST(new);
3780 if (user_ns != ns->user_ns) {
3783 unlock_mount_hash();
3786 list_add_tail(&new_ns->list, &new->mnt_list);
3789 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3790 * as belonging to new namespace. We have already acquired a private
3791 * fs_struct, so tsk->fs->lock is not needed.
3799 if (&p->mnt == new_fs->root.mnt) {
3800 new_fs->root.mnt = mntget(&q->mnt);
3803 if (&p->mnt == new_fs->pwd.mnt) {
3804 new_fs->pwd.mnt = mntget(&q->mnt);
3808 p = next_mnt(p, old);
3809 q = next_mnt(q, new);
3812 // an mntns binding we'd skipped?
3813 while (p->mnt.mnt_root != q->mnt.mnt_root)
3814 p = next_mnt(skip_mnt_tree(p), old);
3826 struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3828 struct mount *mnt = real_mount(m);
3829 struct mnt_namespace *ns;
3830 struct super_block *s;
3834 ns = alloc_mnt_ns(&init_user_ns, true);
3837 return ERR_CAST(ns);
3842 list_add(&mnt->mnt_list, &ns->list);
3844 err = vfs_path_lookup(m->mnt_root, m,
3845 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3850 return ERR_PTR(err);
3852 /* trade a vfsmount reference for active sb one */
3853 s = path.mnt->mnt_sb;
3854 atomic_inc(&s->s_active);
3856 /* lock the sucker */
3857 down_write(&s->s_umount);
3858 /* ... and return the root of (sub)tree on it */
3861 EXPORT_SYMBOL(mount_subtree);
3863 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3864 char __user *, type, unsigned long, flags, void __user *, data)
3871 kernel_type = copy_mount_string(type);
3872 ret = PTR_ERR(kernel_type);
3873 if (IS_ERR(kernel_type))
3876 kernel_dev = copy_mount_string(dev_name);
3877 ret = PTR_ERR(kernel_dev);
3878 if (IS_ERR(kernel_dev))
3881 options = copy_mount_options(data);
3882 ret = PTR_ERR(options);
3883 if (IS_ERR(options))
3886 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3897 #define FSMOUNT_VALID_FLAGS \
3898 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3899 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3900 MOUNT_ATTR_NOSYMFOLLOW)
3902 #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3904 #define MOUNT_SETATTR_PROPAGATION_FLAGS \
3905 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3907 static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3909 unsigned int mnt_flags = 0;
3911 if (attr_flags & MOUNT_ATTR_RDONLY)
3912 mnt_flags |= MNT_READONLY;
3913 if (attr_flags & MOUNT_ATTR_NOSUID)
3914 mnt_flags |= MNT_NOSUID;
3915 if (attr_flags & MOUNT_ATTR_NODEV)
3916 mnt_flags |= MNT_NODEV;
3917 if (attr_flags & MOUNT_ATTR_NOEXEC)
3918 mnt_flags |= MNT_NOEXEC;
3919 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3920 mnt_flags |= MNT_NODIRATIME;
3921 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3922 mnt_flags |= MNT_NOSYMFOLLOW;
3928 * Create a kernel mount representation for a new, prepared superblock
3929 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3931 SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3932 unsigned int, attr_flags)
3934 struct mnt_namespace *ns;
3935 struct fs_context *fc;
3937 struct path newmount;
3940 unsigned int mnt_flags = 0;
3946 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3949 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3952 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3954 switch (attr_flags & MOUNT_ATTR__ATIME) {
3955 case MOUNT_ATTR_STRICTATIME:
3957 case MOUNT_ATTR_NOATIME:
3958 mnt_flags |= MNT_NOATIME;
3960 case MOUNT_ATTR_RELATIME:
3961 mnt_flags |= MNT_RELATIME;
3972 if (f.file->f_op != &fscontext_fops)
3975 fc = f.file->private_data;
3977 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3981 /* There must be a valid superblock or we can't mount it */
3987 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3988 pr_warn("VFS: Mount too revealing\n");
3993 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3996 if (fc->sb_flags & SB_MANDLOCK)
3999 newmount.mnt = vfs_create_mount(fc);
4000 if (IS_ERR(newmount.mnt)) {
4001 ret = PTR_ERR(newmount.mnt);
4004 newmount.dentry = dget(fc->root);
4005 newmount.mnt->mnt_flags = mnt_flags;
4007 /* We've done the mount bit - now move the file context into more or
4008 * less the same state as if we'd done an fspick(). We don't want to
4009 * do any memory allocation or anything like that at this point as we
4010 * don't want to have to handle any errors incurred.
4012 vfs_clean_context(fc);
4014 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
4019 mnt = real_mount(newmount.mnt);
4023 list_add(&mnt->mnt_list, &ns->list);
4024 mntget(newmount.mnt);
4026 /* Attach to an apparent O_PATH fd with a note that we need to unmount
4027 * it, not just simply put it.
4029 file = dentry_open(&newmount, O_PATH, fc->cred);
4031 dissolve_on_fput(newmount.mnt);
4032 ret = PTR_ERR(file);
4035 file->f_mode |= FMODE_NEED_UNMOUNT;
4037 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
4039 fd_install(ret, file);
4044 path_put(&newmount);
4046 mutex_unlock(&fc->uapi_mutex);
4053 * Move a mount from one place to another. In combination with
4054 * fsopen()/fsmount() this is used to install a new mount and in combination
4055 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
4058 * Note the flags value is a combination of MOVE_MOUNT_* flags.
4060 SYSCALL_DEFINE5(move_mount,
4061 int, from_dfd, const char __user *, from_pathname,
4062 int, to_dfd, const char __user *, to_pathname,
4063 unsigned int, flags)
4065 struct path from_path, to_path;
4066 unsigned int lflags;
4072 if (flags & ~MOVE_MOUNT__MASK)
4075 if ((flags & (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP)) ==
4076 (MOVE_MOUNT_BENEATH | MOVE_MOUNT_SET_GROUP))
4079 /* If someone gives a pathname, they aren't permitted to move
4080 * from an fd that requires unmount as we can't get at the flag
4081 * to clear it afterwards.
4084 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4085 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4086 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4088 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
4093 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
4094 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
4095 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
4097 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
4101 ret = security_move_mount(&from_path, &to_path);
4105 if (flags & MOVE_MOUNT_SET_GROUP)
4106 ret = do_set_group(&from_path, &to_path);
4108 ret = do_move_mount(&from_path, &to_path,
4109 (flags & MOVE_MOUNT_BENEATH));
4114 path_put(&from_path);
4119 * Return true if path is reachable from root
4121 * namespace_sem or mount_lock is held
4123 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
4124 const struct path *root)
4126 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
4127 dentry = mnt->mnt_mountpoint;
4128 mnt = mnt->mnt_parent;
4130 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
4133 bool path_is_under(const struct path *path1, const struct path *path2)
4136 read_seqlock_excl(&mount_lock);
4137 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
4138 read_sequnlock_excl(&mount_lock);
4141 EXPORT_SYMBOL(path_is_under);
4144 * pivot_root Semantics:
4145 * Moves the root file system of the current process to the directory put_old,
4146 * makes new_root as the new root file system of the current process, and sets
4147 * root/cwd of all processes which had them on the current root to new_root.
4150 * The new_root and put_old must be directories, and must not be on the
4151 * same file system as the current process root. The put_old must be
4152 * underneath new_root, i.e. adding a non-zero number of /.. to the string
4153 * pointed to by put_old must yield the same directory as new_root. No other
4154 * file system may be mounted on put_old. After all, new_root is a mountpoint.
4156 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
4157 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
4158 * in this situation.
4161 * - we don't move root/cwd if they are not at the root (reason: if something
4162 * cared enough to change them, it's probably wrong to force them elsewhere)
4163 * - it's okay to pick a root that isn't the root of a file system, e.g.
4164 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
4165 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
4168 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
4169 const char __user *, put_old)
4171 struct path new, old, root;
4172 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
4173 struct mountpoint *old_mp, *root_mp;
4179 error = user_path_at(AT_FDCWD, new_root,
4180 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
4184 error = user_path_at(AT_FDCWD, put_old,
4185 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
4189 error = security_sb_pivotroot(&old, &new);
4193 get_fs_root(current->fs, &root);
4194 old_mp = lock_mount(&old);
4195 error = PTR_ERR(old_mp);
4200 new_mnt = real_mount(new.mnt);
4201 root_mnt = real_mount(root.mnt);
4202 old_mnt = real_mount(old.mnt);
4203 ex_parent = new_mnt->mnt_parent;
4204 root_parent = root_mnt->mnt_parent;
4205 if (IS_MNT_SHARED(old_mnt) ||
4206 IS_MNT_SHARED(ex_parent) ||
4207 IS_MNT_SHARED(root_parent))
4209 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
4211 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4214 if (d_unlinked(new.dentry))
4217 if (new_mnt == root_mnt || old_mnt == root_mnt)
4218 goto out4; /* loop, on the same file system */
4220 if (!path_mounted(&root))
4221 goto out4; /* not a mountpoint */
4222 if (!mnt_has_parent(root_mnt))
4223 goto out4; /* not attached */
4224 if (!path_mounted(&new))
4225 goto out4; /* not a mountpoint */
4226 if (!mnt_has_parent(new_mnt))
4227 goto out4; /* not attached */
4228 /* make sure we can reach put_old from new_root */
4229 if (!is_path_reachable(old_mnt, old.dentry, &new))
4231 /* make certain new is below the root */
4232 if (!is_path_reachable(new_mnt, new.dentry, &root))
4235 umount_mnt(new_mnt);
4236 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
4237 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4238 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4239 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4241 /* mount old root on put_old */
4242 attach_mnt(root_mnt, old_mnt, old_mp, false);
4243 /* mount new_root on / */
4244 attach_mnt(new_mnt, root_parent, root_mp, false);
4245 mnt_add_count(root_parent, -1);
4246 touch_mnt_namespace(current->nsproxy->mnt_ns);
4247 /* A moved mount should not expire automatically */
4248 list_del_init(&new_mnt->mnt_expire);
4249 put_mountpoint(root_mp);
4250 unlock_mount_hash();
4251 chroot_fs_refs(&root, &new);
4254 unlock_mount(old_mp);
4256 mntput_no_expire(ex_parent);
4267 static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4269 unsigned int flags = mnt->mnt.mnt_flags;
4271 /* flags to clear */
4272 flags &= ~kattr->attr_clr;
4273 /* flags to raise */
4274 flags |= kattr->attr_set;
4279 static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4281 struct vfsmount *m = &mnt->mnt;
4282 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4284 if (!kattr->mnt_idmap)
4288 * Creating an idmapped mount with the filesystem wide idmapping
4289 * doesn't make sense so block that. We don't allow mushy semantics.
4291 if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb))
4295 * Once a mount has been idmapped we don't allow it to change its
4296 * mapping. It makes things simpler and callers can just create
4297 * another bind-mount they can idmap if they want to.
4299 if (is_idmapped_mnt(m))
4302 /* The underlying filesystem doesn't support idmapped mounts yet. */
4303 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4306 /* We're not controlling the superblock. */
4307 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4310 /* Mount has already been visible in the filesystem hierarchy. */
4311 if (!is_anon_ns(mnt->mnt_ns))
4318 * mnt_allow_writers() - check whether the attribute change allows writers
4319 * @kattr: the new mount attributes
4320 * @mnt: the mount to which @kattr will be applied
4322 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4324 * Return: true if writers need to be held, false if not
4326 static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4327 const struct mount *mnt)
4329 return (!(kattr->attr_set & MNT_READONLY) ||
4330 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4334 static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4339 for (m = mnt; m; m = next_mnt(m, mnt)) {
4340 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4345 err = can_idmap_mount(kattr, m);
4349 if (!mnt_allow_writers(kattr, m)) {
4350 err = mnt_hold_writers(m);
4355 if (!kattr->recurse)
4363 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4364 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4365 * mounts and needs to take care to include the first mount.
4367 for (p = mnt; p; p = next_mnt(p, mnt)) {
4368 /* If we had to hold writers unblock them. */
4369 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4370 mnt_unhold_writers(p);
4373 * We're done once the first mount we changed got
4374 * MNT_WRITE_HOLD unset.
4383 static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4385 if (!kattr->mnt_idmap)
4389 * Pairs with smp_load_acquire() in mnt_idmap().
4391 * Since we only allow a mount to change the idmapping once and
4392 * verified this in can_idmap_mount() we know that the mount has
4393 * @nop_mnt_idmap attached to it. So there's no need to drop any
4396 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4399 static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4403 for (m = mnt; m; m = next_mnt(m, mnt)) {
4406 do_idmap_mount(kattr, m);
4407 flags = recalc_flags(kattr, m);
4408 WRITE_ONCE(m->mnt.mnt_flags, flags);
4410 /* If we had to hold writers unblock them. */
4411 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4412 mnt_unhold_writers(m);
4414 if (kattr->propagation)
4415 change_mnt_propagation(m, kattr->propagation);
4416 if (!kattr->recurse)
4419 touch_mnt_namespace(mnt->mnt_ns);
4422 static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4424 struct mount *mnt = real_mount(path->mnt);
4427 if (!path_mounted(path))
4430 if (kattr->mnt_userns) {
4431 struct mnt_idmap *mnt_idmap;
4433 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4434 if (IS_ERR(mnt_idmap))
4435 return PTR_ERR(mnt_idmap);
4436 kattr->mnt_idmap = mnt_idmap;
4439 if (kattr->propagation) {
4441 * Only take namespace_lock() if we're actually changing
4445 if (kattr->propagation == MS_SHARED) {
4446 err = invent_group_ids(mnt, kattr->recurse);
4457 /* Ensure that this isn't anything purely vfs internal. */
4458 if (!is_mounted(&mnt->mnt))
4462 * If this is an attached mount make sure it's located in the callers
4463 * mount namespace. If it's not don't let the caller interact with it.
4464 * If this is a detached mount make sure it has an anonymous mount
4465 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4467 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4471 * First, we get the mount tree in a shape where we can change mount
4472 * properties without failure. If we succeeded to do so we commit all
4473 * changes and if we failed we clean up.
4475 err = mount_setattr_prepare(kattr, mnt);
4477 mount_setattr_commit(kattr, mnt);
4480 unlock_mount_hash();
4482 if (kattr->propagation) {
4484 cleanup_group_ids(mnt, NULL);
4491 static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4492 struct mount_kattr *kattr, unsigned int flags)
4495 struct ns_common *ns;
4496 struct user_namespace *mnt_userns;
4499 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4503 * We currently do not support clearing an idmapped mount. If this ever
4504 * is a use-case we can revisit this but for now let's keep it simple
4507 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4510 if (attr->userns_fd > INT_MAX)
4513 f = fdget(attr->userns_fd);
4517 if (!proc_ns_file(f.file)) {
4522 ns = get_proc_ns(file_inode(f.file));
4523 if (ns->ops->type != CLONE_NEWUSER) {
4529 * The initial idmapping cannot be used to create an idmapped
4530 * mount. We use the initial idmapping as an indicator of a mount
4531 * that is not idmapped. It can simply be passed into helpers that
4532 * are aware of idmapped mounts as a convenient shortcut. A user
4533 * can just create a dedicated identity mapping to achieve the same
4536 mnt_userns = container_of(ns, struct user_namespace, ns);
4537 if (mnt_userns == &init_user_ns) {
4542 /* We're not controlling the target namespace. */
4543 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4548 kattr->mnt_userns = get_user_ns(mnt_userns);
4555 static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4556 struct mount_kattr *kattr, unsigned int flags)
4558 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4560 if (flags & AT_NO_AUTOMOUNT)
4561 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4562 if (flags & AT_SYMLINK_NOFOLLOW)
4563 lookup_flags &= ~LOOKUP_FOLLOW;
4564 if (flags & AT_EMPTY_PATH)
4565 lookup_flags |= LOOKUP_EMPTY;
4567 *kattr = (struct mount_kattr) {
4568 .lookup_flags = lookup_flags,
4569 .recurse = !!(flags & AT_RECURSIVE),
4572 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4574 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4576 kattr->propagation = attr->propagation;
4578 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4581 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4582 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4585 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4586 * users wanting to transition to a different atime setting cannot
4587 * simply specify the atime setting in @attr_set, but must also
4588 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4589 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4590 * @attr_clr and that @attr_set can't have any atime bits set if
4591 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4593 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4594 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4598 * Clear all previous time settings as they are mutually
4601 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4602 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4603 case MOUNT_ATTR_RELATIME:
4604 kattr->attr_set |= MNT_RELATIME;
4606 case MOUNT_ATTR_NOATIME:
4607 kattr->attr_set |= MNT_NOATIME;
4609 case MOUNT_ATTR_STRICTATIME:
4615 if (attr->attr_set & MOUNT_ATTR__ATIME)
4619 return build_mount_idmapped(attr, usize, kattr, flags);
4622 static void finish_mount_kattr(struct mount_kattr *kattr)
4624 put_user_ns(kattr->mnt_userns);
4625 kattr->mnt_userns = NULL;
4627 if (kattr->mnt_idmap)
4628 mnt_idmap_put(kattr->mnt_idmap);
4631 SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4632 unsigned int, flags, struct mount_attr __user *, uattr,
4637 struct mount_attr attr;
4638 struct mount_kattr kattr;
4640 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4642 if (flags & ~(AT_EMPTY_PATH |
4644 AT_SYMLINK_NOFOLLOW |
4648 if (unlikely(usize > PAGE_SIZE))
4650 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4656 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4660 /* Don't bother walking through the mounts if this is a nop. */
4661 if (attr.attr_set == 0 &&
4662 attr.attr_clr == 0 &&
4663 attr.propagation == 0)
4666 err = build_mount_kattr(&attr, usize, &kattr, flags);
4670 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4672 err = do_mount_setattr(&target, &kattr);
4675 finish_mount_kattr(&kattr);
4679 static void __init init_mount_tree(void)
4681 struct vfsmount *mnt;
4683 struct mnt_namespace *ns;
4686 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4688 panic("Can't create rootfs");
4690 ns = alloc_mnt_ns(&init_user_ns, false);
4692 panic("Can't allocate initial namespace");
4693 m = real_mount(mnt);
4697 list_add(&m->mnt_list, &ns->list);
4698 init_task.nsproxy->mnt_ns = ns;
4702 root.dentry = mnt->mnt_root;
4703 mnt->mnt_flags |= MNT_LOCKED;
4705 set_fs_pwd(current->fs, &root);
4706 set_fs_root(current->fs, &root);
4709 void __init mnt_init(void)
4713 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4714 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4716 mount_hashtable = alloc_large_system_hash("Mount-cache",
4717 sizeof(struct hlist_head),
4720 &m_hash_shift, &m_hash_mask, 0, 0);
4721 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4722 sizeof(struct hlist_head),
4725 &mp_hash_shift, &mp_hash_mask, 0, 0);
4727 if (!mount_hashtable || !mountpoint_hashtable)
4728 panic("Failed to allocate mount hash table\n");
4734 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4736 fs_kobj = kobject_create_and_add("fs", NULL);
4738 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4744 void put_mnt_ns(struct mnt_namespace *ns)
4746 if (!refcount_dec_and_test(&ns->ns.count))
4748 drop_collected_mounts(&ns->root->mnt);
4752 struct vfsmount *kern_mount(struct file_system_type *type)
4754 struct vfsmount *mnt;
4755 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4758 * it is a longterm mount, don't release mnt until
4759 * we unmount before file sys is unregistered
4761 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4765 EXPORT_SYMBOL_GPL(kern_mount);
4767 void kern_unmount(struct vfsmount *mnt)
4769 /* release long term mount so mount point can be released */
4771 mnt_make_shortterm(mnt);
4772 synchronize_rcu(); /* yecchhh... */
4776 EXPORT_SYMBOL(kern_unmount);
4778 void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4782 for (i = 0; i < num; i++)
4783 mnt_make_shortterm(mnt[i]);
4784 synchronize_rcu_expedited();
4785 for (i = 0; i < num; i++)
4788 EXPORT_SYMBOL(kern_unmount_array);
4790 bool our_mnt(struct vfsmount *mnt)
4792 return check_mnt(real_mount(mnt));
4795 bool current_chrooted(void)
4797 /* Does the current process have a non-standard root */
4798 struct path ns_root;
4799 struct path fs_root;
4802 /* Find the namespace root */
4803 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4804 ns_root.dentry = ns_root.mnt->mnt_root;
4806 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4809 get_fs_root(current->fs, &fs_root);
4811 chrooted = !path_equal(&fs_root, &ns_root);
4819 static bool mnt_already_visible(struct mnt_namespace *ns,
4820 const struct super_block *sb,
4823 int new_flags = *new_mnt_flags;
4825 bool visible = false;
4827 down_read(&namespace_sem);
4829 list_for_each_entry(mnt, &ns->list, mnt_list) {
4830 struct mount *child;
4833 if (mnt_is_cursor(mnt))
4836 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4839 /* This mount is not fully visible if it's root directory
4840 * is not the root directory of the filesystem.
4842 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4845 /* A local view of the mount flags */
4846 mnt_flags = mnt->mnt.mnt_flags;
4848 /* Don't miss readonly hidden in the superblock flags */
4849 if (sb_rdonly(mnt->mnt.mnt_sb))
4850 mnt_flags |= MNT_LOCK_READONLY;
4852 /* Verify the mount flags are equal to or more permissive
4853 * than the proposed new mount.
4855 if ((mnt_flags & MNT_LOCK_READONLY) &&
4856 !(new_flags & MNT_READONLY))
4858 if ((mnt_flags & MNT_LOCK_ATIME) &&
4859 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4862 /* This mount is not fully visible if there are any
4863 * locked child mounts that cover anything except for
4864 * empty directories.
4866 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4867 struct inode *inode = child->mnt_mountpoint->d_inode;
4868 /* Only worry about locked mounts */
4869 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4871 /* Is the directory permanetly empty? */
4872 if (!is_empty_dir_inode(inode))
4875 /* Preserve the locked attributes */
4876 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4884 up_read(&namespace_sem);
4888 static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4890 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4891 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4892 unsigned long s_iflags;
4894 if (ns->user_ns == &init_user_ns)
4897 /* Can this filesystem be too revealing? */
4898 s_iflags = sb->s_iflags;
4899 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4902 if ((s_iflags & required_iflags) != required_iflags) {
4903 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4908 return !mnt_already_visible(ns, sb, new_mnt_flags);
4911 bool mnt_may_suid(struct vfsmount *mnt)
4914 * Foreign mounts (accessed via fchdir or through /proc
4915 * symlinks) are always treated as if they are nosuid. This
4916 * prevents namespaces from trusting potentially unsafe
4917 * suid/sgid bits, file caps, or security labels that originate
4918 * in other namespaces.
4920 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4921 current_in_userns(mnt->mnt_sb->s_user_ns);
4924 static struct ns_common *mntns_get(struct task_struct *task)
4926 struct ns_common *ns = NULL;
4927 struct nsproxy *nsproxy;
4930 nsproxy = task->nsproxy;
4932 ns = &nsproxy->mnt_ns->ns;
4933 get_mnt_ns(to_mnt_ns(ns));
4940 static void mntns_put(struct ns_common *ns)
4942 put_mnt_ns(to_mnt_ns(ns));
4945 static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4947 struct nsproxy *nsproxy = nsset->nsproxy;
4948 struct fs_struct *fs = nsset->fs;
4949 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4950 struct user_namespace *user_ns = nsset->cred->user_ns;
4954 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4955 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4956 !ns_capable(user_ns, CAP_SYS_ADMIN))
4959 if (is_anon_ns(mnt_ns))
4966 old_mnt_ns = nsproxy->mnt_ns;
4967 nsproxy->mnt_ns = mnt_ns;
4970 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4971 "/", LOOKUP_DOWN, &root);
4973 /* revert to old namespace */
4974 nsproxy->mnt_ns = old_mnt_ns;
4979 put_mnt_ns(old_mnt_ns);
4981 /* Update the pwd and root */
4982 set_fs_pwd(fs, &root);
4983 set_fs_root(fs, &root);
4989 static struct user_namespace *mntns_owner(struct ns_common *ns)
4991 return to_mnt_ns(ns)->user_ns;
4994 const struct proc_ns_operations mntns_operations = {
4996 .type = CLONE_NEWNS,
4999 .install = mntns_install,
5000 .owner = mntns_owner,
5003 #ifdef CONFIG_SYSCTL
5004 static struct ctl_table fs_namespace_sysctls[] = {
5006 .procname = "mount-max",
5007 .data = &sysctl_mount_max,
5008 .maxlen = sizeof(unsigned int),
5010 .proc_handler = proc_dointvec_minmax,
5011 .extra1 = SYSCTL_ONE,
5016 static int __init init_fs_namespace_sysctls(void)
5018 register_sysctl_init("fs", fs_namespace_sysctls);
5021 fs_initcall(init_fs_namespace_sysctls);
5023 #endif /* CONFIG_SYSCTL */