Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
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/task_work.h>
29#include <linux/sched/task.h>
30#include <uapi/linux/mount.h>
31#include <linux/fs_context.h>
32#include <linux/shmem_fs.h>
33
34#include "pnode.h"
35#include "internal.h"
36
37/* Maximum number of mounts in a mount namespace */
38unsigned int sysctl_mount_max __read_mostly = 100000;
39
40static unsigned int m_hash_mask __read_mostly;
41static unsigned int m_hash_shift __read_mostly;
42static unsigned int mp_hash_mask __read_mostly;
43static unsigned int mp_hash_shift __read_mostly;
44
45static __initdata unsigned long mhash_entries;
46static int __init set_mhash_entries(char *str)
47{
48 if (!str)
49 return 0;
50 mhash_entries = simple_strtoul(str, &str, 0);
51 return 1;
52}
53__setup("mhash_entries=", set_mhash_entries);
54
55static __initdata unsigned long mphash_entries;
56static int __init set_mphash_entries(char *str)
57{
58 if (!str)
59 return 0;
60 mphash_entries = simple_strtoul(str, &str, 0);
61 return 1;
62}
63__setup("mphash_entries=", set_mphash_entries);
64
65static u64 event;
66static DEFINE_IDA(mnt_id_ida);
67static DEFINE_IDA(mnt_group_ida);
68
69static struct hlist_head *mount_hashtable __read_mostly;
70static struct hlist_head *mountpoint_hashtable __read_mostly;
71static struct kmem_cache *mnt_cache __read_mostly;
72static DECLARE_RWSEM(namespace_sem);
73static HLIST_HEAD(unmounted); /* protected by namespace_sem */
74static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
75
76/* /sys/fs */
77struct kobject *fs_kobj;
78EXPORT_SYMBOL_GPL(fs_kobj);
79
80/*
81 * vfsmount lock may be taken for read to prevent changes to the
82 * vfsmount hash, ie. during mountpoint lookups or walking back
83 * up the tree.
84 *
85 * It should be taken for write in all cases where the vfsmount
86 * tree or hash is modified or when a vfsmount structure is modified.
87 */
88__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
89
90static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
91{
92 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
93 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
94 tmp = tmp + (tmp >> m_hash_shift);
95 return &mount_hashtable[tmp & m_hash_mask];
96}
97
98static inline struct hlist_head *mp_hash(struct dentry *dentry)
99{
100 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
101 tmp = tmp + (tmp >> mp_hash_shift);
102 return &mountpoint_hashtable[tmp & mp_hash_mask];
103}
104
105static int mnt_alloc_id(struct mount *mnt)
106{
107 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
108
109 if (res < 0)
110 return res;
111 mnt->mnt_id = res;
112 return 0;
113}
114
115static void mnt_free_id(struct mount *mnt)
116{
117 ida_free(&mnt_id_ida, mnt->mnt_id);
118}
119
120/*
121 * Allocate a new peer group ID
122 */
123static int mnt_alloc_group_id(struct mount *mnt)
124{
125 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
126
127 if (res < 0)
128 return res;
129 mnt->mnt_group_id = res;
130 return 0;
131}
132
133/*
134 * Release a peer group ID
135 */
136void mnt_release_group_id(struct mount *mnt)
137{
138 ida_free(&mnt_group_ida, mnt->mnt_group_id);
139 mnt->mnt_group_id = 0;
140}
141
142/*
143 * vfsmount lock must be held for read
144 */
145static inline void mnt_add_count(struct mount *mnt, int n)
146{
147#ifdef CONFIG_SMP
148 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
149#else
150 preempt_disable();
151 mnt->mnt_count += n;
152 preempt_enable();
153#endif
154}
155
156/*
157 * vfsmount lock must be held for write
158 */
159unsigned int mnt_get_count(struct mount *mnt)
160{
161#ifdef CONFIG_SMP
162 unsigned int count = 0;
163 int cpu;
164
165 for_each_possible_cpu(cpu) {
166 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
167 }
168
169 return count;
170#else
171 return mnt->mnt_count;
172#endif
173}
174
175static struct mount *alloc_vfsmnt(const char *name)
176{
177 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
178 if (mnt) {
179 int err;
180
181 err = mnt_alloc_id(mnt);
182 if (err)
183 goto out_free_cache;
184
185 if (name) {
186 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
187 if (!mnt->mnt_devname)
188 goto out_free_id;
189 }
190
191#ifdef CONFIG_SMP
192 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
193 if (!mnt->mnt_pcp)
194 goto out_free_devname;
195
196 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
197#else
198 mnt->mnt_count = 1;
199 mnt->mnt_writers = 0;
200#endif
201
202 INIT_HLIST_NODE(&mnt->mnt_hash);
203 INIT_LIST_HEAD(&mnt->mnt_child);
204 INIT_LIST_HEAD(&mnt->mnt_mounts);
205 INIT_LIST_HEAD(&mnt->mnt_list);
206 INIT_LIST_HEAD(&mnt->mnt_expire);
207 INIT_LIST_HEAD(&mnt->mnt_share);
208 INIT_LIST_HEAD(&mnt->mnt_slave_list);
209 INIT_LIST_HEAD(&mnt->mnt_slave);
210 INIT_HLIST_NODE(&mnt->mnt_mp_list);
211 INIT_LIST_HEAD(&mnt->mnt_umounting);
212 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
213 }
214 return mnt;
215
216#ifdef CONFIG_SMP
217out_free_devname:
218 kfree_const(mnt->mnt_devname);
219#endif
220out_free_id:
221 mnt_free_id(mnt);
222out_free_cache:
223 kmem_cache_free(mnt_cache, mnt);
224 return NULL;
225}
226
227/*
228 * Most r/o checks on a fs are for operations that take
229 * discrete amounts of time, like a write() or unlink().
230 * We must keep track of when those operations start
231 * (for permission checks) and when they end, so that
232 * we can determine when writes are able to occur to
233 * a filesystem.
234 */
235/*
236 * __mnt_is_readonly: check whether a mount is read-only
237 * @mnt: the mount to check for its write status
238 *
239 * This shouldn't be used directly ouside of the VFS.
240 * It does not guarantee that the filesystem will stay
241 * r/w, just that it is right *now*. This can not and
242 * should not be used in place of IS_RDONLY(inode).
243 * mnt_want/drop_write() will _keep_ the filesystem
244 * r/w.
245 */
246bool __mnt_is_readonly(struct vfsmount *mnt)
247{
248 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
249}
250EXPORT_SYMBOL_GPL(__mnt_is_readonly);
251
252static inline void mnt_inc_writers(struct mount *mnt)
253{
254#ifdef CONFIG_SMP
255 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
256#else
257 mnt->mnt_writers++;
258#endif
259}
260
261static inline void mnt_dec_writers(struct mount *mnt)
262{
263#ifdef CONFIG_SMP
264 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
265#else
266 mnt->mnt_writers--;
267#endif
268}
269
270static unsigned int mnt_get_writers(struct mount *mnt)
271{
272#ifdef CONFIG_SMP
273 unsigned int count = 0;
274 int cpu;
275
276 for_each_possible_cpu(cpu) {
277 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
278 }
279
280 return count;
281#else
282 return mnt->mnt_writers;
283#endif
284}
285
286static int mnt_is_readonly(struct vfsmount *mnt)
287{
288 if (mnt->mnt_sb->s_readonly_remount)
289 return 1;
290 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
291 smp_rmb();
292 return __mnt_is_readonly(mnt);
293}
294
295/*
296 * Most r/o & frozen checks on a fs are for operations that take discrete
297 * amounts of time, like a write() or unlink(). We must keep track of when
298 * those operations start (for permission checks) and when they end, so that we
299 * can determine when writes are able to occur to a filesystem.
300 */
301/**
302 * __mnt_want_write - get write access to a mount without freeze protection
303 * @m: the mount on which to take a write
304 *
305 * This tells the low-level filesystem that a write is about to be performed to
306 * it, and makes sure that writes are allowed (mnt it read-write) before
307 * returning success. This operation does not protect against filesystem being
308 * frozen. When the write operation is finished, __mnt_drop_write() must be
309 * called. This is effectively a refcount.
310 */
311int __mnt_want_write(struct vfsmount *m)
312{
313 struct mount *mnt = real_mount(m);
314 int ret = 0;
315
316 preempt_disable();
317 mnt_inc_writers(mnt);
318 /*
319 * The store to mnt_inc_writers must be visible before we pass
320 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
321 * incremented count after it has set MNT_WRITE_HOLD.
322 */
323 smp_mb();
324 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
325 cpu_relax();
326 /*
327 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
328 * be set to match its requirements. So we must not load that until
329 * MNT_WRITE_HOLD is cleared.
330 */
331 smp_rmb();
332 if (mnt_is_readonly(m)) {
333 mnt_dec_writers(mnt);
334 ret = -EROFS;
335 }
336 preempt_enable();
337
338 return ret;
339}
340
341/**
342 * mnt_want_write - get write access to a mount
343 * @m: the mount on which to take a write
344 *
345 * This tells the low-level filesystem that a write is about to be performed to
346 * it, and makes sure that writes are allowed (mount is read-write, filesystem
347 * is not frozen) before returning success. When the write operation is
348 * finished, mnt_drop_write() must be called. This is effectively a refcount.
349 */
350int mnt_want_write(struct vfsmount *m)
351{
352 int ret;
353
354 sb_start_write(m->mnt_sb);
355 ret = __mnt_want_write(m);
356 if (ret)
357 sb_end_write(m->mnt_sb);
358 return ret;
359}
360EXPORT_SYMBOL_GPL(mnt_want_write);
361
362/**
363 * mnt_clone_write - get write access to a mount
364 * @mnt: the mount on which to take a write
365 *
366 * This is effectively like mnt_want_write, except
367 * it must only be used to take an extra write reference
368 * on a mountpoint that we already know has a write reference
369 * on it. This allows some optimisation.
370 *
371 * After finished, mnt_drop_write must be called as usual to
372 * drop the reference.
373 */
374int mnt_clone_write(struct vfsmount *mnt)
375{
376 /* superblock may be r/o */
377 if (__mnt_is_readonly(mnt))
378 return -EROFS;
379 preempt_disable();
380 mnt_inc_writers(real_mount(mnt));
381 preempt_enable();
382 return 0;
383}
384EXPORT_SYMBOL_GPL(mnt_clone_write);
385
386/**
387 * __mnt_want_write_file - get write access to a file's mount
388 * @file: the file who's mount on which to take a write
389 *
390 * This is like __mnt_want_write, but it takes a file and can
391 * do some optimisations if the file is open for write already
392 */
393int __mnt_want_write_file(struct file *file)
394{
395 if (!(file->f_mode & FMODE_WRITER))
396 return __mnt_want_write(file->f_path.mnt);
397 else
398 return mnt_clone_write(file->f_path.mnt);
399}
400
401/**
402 * mnt_want_write_file - get write access to a file's mount
403 * @file: the file who's mount on which to take a write
404 *
405 * This is like mnt_want_write, but it takes a file and can
406 * do some optimisations if the file is open for write already
407 */
408int mnt_want_write_file(struct file *file)
409{
410 int ret;
411
412 sb_start_write(file_inode(file)->i_sb);
413 ret = __mnt_want_write_file(file);
414 if (ret)
415 sb_end_write(file_inode(file)->i_sb);
416 return ret;
417}
418EXPORT_SYMBOL_GPL(mnt_want_write_file);
419
420/**
421 * __mnt_drop_write - give up write access to a mount
422 * @mnt: the mount on which to give up write access
423 *
424 * Tells the low-level filesystem that we are done
425 * performing writes to it. Must be matched with
426 * __mnt_want_write() call above.
427 */
428void __mnt_drop_write(struct vfsmount *mnt)
429{
430 preempt_disable();
431 mnt_dec_writers(real_mount(mnt));
432 preempt_enable();
433}
434
435/**
436 * mnt_drop_write - give up write access to a mount
437 * @mnt: the mount on which to give up write access
438 *
439 * Tells the low-level filesystem that we are done performing writes to it and
440 * also allows filesystem to be frozen again. Must be matched with
441 * mnt_want_write() call above.
442 */
443void mnt_drop_write(struct vfsmount *mnt)
444{
445 __mnt_drop_write(mnt);
446 sb_end_write(mnt->mnt_sb);
447}
448EXPORT_SYMBOL_GPL(mnt_drop_write);
449
450void __mnt_drop_write_file(struct file *file)
451{
452 __mnt_drop_write(file->f_path.mnt);
453}
454
455void mnt_drop_write_file(struct file *file)
456{
457 __mnt_drop_write_file(file);
458 sb_end_write(file_inode(file)->i_sb);
459}
460EXPORT_SYMBOL(mnt_drop_write_file);
461
462static int mnt_make_readonly(struct mount *mnt)
463{
464 int ret = 0;
465
466 lock_mount_hash();
467 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
468 /*
469 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
470 * should be visible before we do.
471 */
472 smp_mb();
473
474 /*
475 * With writers on hold, if this value is zero, then there are
476 * definitely no active writers (although held writers may subsequently
477 * increment the count, they'll have to wait, and decrement it after
478 * seeing MNT_READONLY).
479 *
480 * It is OK to have counter incremented on one CPU and decremented on
481 * another: the sum will add up correctly. The danger would be when we
482 * sum up each counter, if we read a counter before it is incremented,
483 * but then read another CPU's count which it has been subsequently
484 * decremented from -- we would see more decrements than we should.
485 * MNT_WRITE_HOLD protects against this scenario, because
486 * mnt_want_write first increments count, then smp_mb, then spins on
487 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
488 * we're counting up here.
489 */
490 if (mnt_get_writers(mnt) > 0)
491 ret = -EBUSY;
492 else
493 mnt->mnt.mnt_flags |= MNT_READONLY;
494 /*
495 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
496 * that become unheld will see MNT_READONLY.
497 */
498 smp_wmb();
499 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
500 unlock_mount_hash();
501 return ret;
502}
503
504static int __mnt_unmake_readonly(struct mount *mnt)
505{
506 lock_mount_hash();
507 mnt->mnt.mnt_flags &= ~MNT_READONLY;
508 unlock_mount_hash();
509 return 0;
510}
511
512int sb_prepare_remount_readonly(struct super_block *sb)
513{
514 struct mount *mnt;
515 int err = 0;
516
517 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
518 if (atomic_long_read(&sb->s_remove_count))
519 return -EBUSY;
520
521 lock_mount_hash();
522 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
523 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
524 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
525 smp_mb();
526 if (mnt_get_writers(mnt) > 0) {
527 err = -EBUSY;
528 break;
529 }
530 }
531 }
532 if (!err && atomic_long_read(&sb->s_remove_count))
533 err = -EBUSY;
534
535 if (!err) {
536 sb->s_readonly_remount = 1;
537 smp_wmb();
538 }
539 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
540 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
541 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
542 }
543 unlock_mount_hash();
544
545 return err;
546}
547
548static void free_vfsmnt(struct mount *mnt)
549{
550 kfree_const(mnt->mnt_devname);
551#ifdef CONFIG_SMP
552 free_percpu(mnt->mnt_pcp);
553#endif
554 kmem_cache_free(mnt_cache, mnt);
555}
556
557static void delayed_free_vfsmnt(struct rcu_head *head)
558{
559 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
560}
561
562/* call under rcu_read_lock */
563int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
564{
565 struct mount *mnt;
566 if (read_seqretry(&mount_lock, seq))
567 return 1;
568 if (bastard == NULL)
569 return 0;
570 mnt = real_mount(bastard);
571 mnt_add_count(mnt, 1);
572 smp_mb(); // see mntput_no_expire()
573 if (likely(!read_seqretry(&mount_lock, seq)))
574 return 0;
575 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
576 mnt_add_count(mnt, -1);
577 return 1;
578 }
579 lock_mount_hash();
580 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
581 mnt_add_count(mnt, -1);
582 unlock_mount_hash();
583 return 1;
584 }
585 unlock_mount_hash();
586 /* caller will mntput() */
587 return -1;
588}
589
590/* call under rcu_read_lock */
591bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
592{
593 int res = __legitimize_mnt(bastard, seq);
594 if (likely(!res))
595 return true;
596 if (unlikely(res < 0)) {
597 rcu_read_unlock();
598 mntput(bastard);
599 rcu_read_lock();
600 }
601 return false;
602}
603
604/*
605 * find the first mount at @dentry on vfsmount @mnt.
606 * call under rcu_read_lock()
607 */
608struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
609{
610 struct hlist_head *head = m_hash(mnt, dentry);
611 struct mount *p;
612
613 hlist_for_each_entry_rcu(p, head, mnt_hash)
614 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
615 return p;
616 return NULL;
617}
618
619/*
620 * lookup_mnt - Return the first child mount mounted at path
621 *
622 * "First" means first mounted chronologically. If you create the
623 * following mounts:
624 *
625 * mount /dev/sda1 /mnt
626 * mount /dev/sda2 /mnt
627 * mount /dev/sda3 /mnt
628 *
629 * Then lookup_mnt() on the base /mnt dentry in the root mount will
630 * return successively the root dentry and vfsmount of /dev/sda1, then
631 * /dev/sda2, then /dev/sda3, then NULL.
632 *
633 * lookup_mnt takes a reference to the found vfsmount.
634 */
635struct vfsmount *lookup_mnt(const struct path *path)
636{
637 struct mount *child_mnt;
638 struct vfsmount *m;
639 unsigned seq;
640
641 rcu_read_lock();
642 do {
643 seq = read_seqbegin(&mount_lock);
644 child_mnt = __lookup_mnt(path->mnt, path->dentry);
645 m = child_mnt ? &child_mnt->mnt : NULL;
646 } while (!legitimize_mnt(m, seq));
647 rcu_read_unlock();
648 return m;
649}
650
651/*
652 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
653 * current mount namespace.
654 *
655 * The common case is dentries are not mountpoints at all and that
656 * test is handled inline. For the slow case when we are actually
657 * dealing with a mountpoint of some kind, walk through all of the
658 * mounts in the current mount namespace and test to see if the dentry
659 * is a mountpoint.
660 *
661 * The mount_hashtable is not usable in the context because we
662 * need to identify all mounts that may be in the current mount
663 * namespace not just a mount that happens to have some specified
664 * parent mount.
665 */
666bool __is_local_mountpoint(struct dentry *dentry)
667{
668 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
669 struct mount *mnt;
670 bool is_covered = false;
671
672 if (!d_mountpoint(dentry))
673 goto out;
674
675 down_read(&namespace_sem);
676 list_for_each_entry(mnt, &ns->list, mnt_list) {
677 is_covered = (mnt->mnt_mountpoint == dentry);
678 if (is_covered)
679 break;
680 }
681 up_read(&namespace_sem);
682out:
683 return is_covered;
684}
685
686static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
687{
688 struct hlist_head *chain = mp_hash(dentry);
689 struct mountpoint *mp;
690
691 hlist_for_each_entry(mp, chain, m_hash) {
692 if (mp->m_dentry == dentry) {
693 mp->m_count++;
694 return mp;
695 }
696 }
697 return NULL;
698}
699
700static struct mountpoint *get_mountpoint(struct dentry *dentry)
701{
702 struct mountpoint *mp, *new = NULL;
703 int ret;
704
705 if (d_mountpoint(dentry)) {
706 /* might be worth a WARN_ON() */
707 if (d_unlinked(dentry))
708 return ERR_PTR(-ENOENT);
709mountpoint:
710 read_seqlock_excl(&mount_lock);
711 mp = lookup_mountpoint(dentry);
712 read_sequnlock_excl(&mount_lock);
713 if (mp)
714 goto done;
715 }
716
717 if (!new)
718 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
719 if (!new)
720 return ERR_PTR(-ENOMEM);
721
722
723 /* Exactly one processes may set d_mounted */
724 ret = d_set_mounted(dentry);
725
726 /* Someone else set d_mounted? */
727 if (ret == -EBUSY)
728 goto mountpoint;
729
730 /* The dentry is not available as a mountpoint? */
731 mp = ERR_PTR(ret);
732 if (ret)
733 goto done;
734
735 /* Add the new mountpoint to the hash table */
736 read_seqlock_excl(&mount_lock);
737 new->m_dentry = dget(dentry);
738 new->m_count = 1;
739 hlist_add_head(&new->m_hash, mp_hash(dentry));
740 INIT_HLIST_HEAD(&new->m_list);
741 read_sequnlock_excl(&mount_lock);
742
743 mp = new;
744 new = NULL;
745done:
746 kfree(new);
747 return mp;
748}
749
750/*
751 * vfsmount lock must be held. Additionally, the caller is responsible
752 * for serializing calls for given disposal list.
753 */
754static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
755{
756 if (!--mp->m_count) {
757 struct dentry *dentry = mp->m_dentry;
758 BUG_ON(!hlist_empty(&mp->m_list));
759 spin_lock(&dentry->d_lock);
760 dentry->d_flags &= ~DCACHE_MOUNTED;
761 spin_unlock(&dentry->d_lock);
762 dput_to_list(dentry, list);
763 hlist_del(&mp->m_hash);
764 kfree(mp);
765 }
766}
767
768/* called with namespace_lock and vfsmount lock */
769static void put_mountpoint(struct mountpoint *mp)
770{
771 __put_mountpoint(mp, &ex_mountpoints);
772}
773
774static inline int check_mnt(struct mount *mnt)
775{
776 return mnt->mnt_ns == current->nsproxy->mnt_ns;
777}
778
779/*
780 * vfsmount lock must be held for write
781 */
782static void touch_mnt_namespace(struct mnt_namespace *ns)
783{
784 if (ns) {
785 ns->event = ++event;
786 wake_up_interruptible(&ns->poll);
787 }
788}
789
790/*
791 * vfsmount lock must be held for write
792 */
793static void __touch_mnt_namespace(struct mnt_namespace *ns)
794{
795 if (ns && ns->event != event) {
796 ns->event = event;
797 wake_up_interruptible(&ns->poll);
798 }
799}
800
801/*
802 * vfsmount lock must be held for write
803 */
804static struct mountpoint *unhash_mnt(struct mount *mnt)
805{
806 struct mountpoint *mp;
807 mnt->mnt_parent = mnt;
808 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
809 list_del_init(&mnt->mnt_child);
810 hlist_del_init_rcu(&mnt->mnt_hash);
811 hlist_del_init(&mnt->mnt_mp_list);
812 mp = mnt->mnt_mp;
813 mnt->mnt_mp = NULL;
814 return mp;
815}
816
817/*
818 * vfsmount lock must be held for write
819 */
820static void umount_mnt(struct mount *mnt)
821{
822 put_mountpoint(unhash_mnt(mnt));
823}
824
825/*
826 * vfsmount lock must be held for write
827 */
828void mnt_set_mountpoint(struct mount *mnt,
829 struct mountpoint *mp,
830 struct mount *child_mnt)
831{
832 mp->m_count++;
833 mnt_add_count(mnt, 1); /* essentially, that's mntget */
834 child_mnt->mnt_mountpoint = mp->m_dentry;
835 child_mnt->mnt_parent = mnt;
836 child_mnt->mnt_mp = mp;
837 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
838}
839
840static void __attach_mnt(struct mount *mnt, struct mount *parent)
841{
842 hlist_add_head_rcu(&mnt->mnt_hash,
843 m_hash(&parent->mnt, mnt->mnt_mountpoint));
844 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
845}
846
847/*
848 * vfsmount lock must be held for write
849 */
850static void attach_mnt(struct mount *mnt,
851 struct mount *parent,
852 struct mountpoint *mp)
853{
854 mnt_set_mountpoint(parent, mp, mnt);
855 __attach_mnt(mnt, parent);
856}
857
858void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
859{
860 struct mountpoint *old_mp = mnt->mnt_mp;
861 struct mount *old_parent = mnt->mnt_parent;
862
863 list_del_init(&mnt->mnt_child);
864 hlist_del_init(&mnt->mnt_mp_list);
865 hlist_del_init_rcu(&mnt->mnt_hash);
866
867 attach_mnt(mnt, parent, mp);
868
869 put_mountpoint(old_mp);
870 mnt_add_count(old_parent, -1);
871}
872
873/*
874 * vfsmount lock must be held for write
875 */
876static void commit_tree(struct mount *mnt)
877{
878 struct mount *parent = mnt->mnt_parent;
879 struct mount *m;
880 LIST_HEAD(head);
881 struct mnt_namespace *n = parent->mnt_ns;
882
883 BUG_ON(parent == mnt);
884
885 list_add_tail(&head, &mnt->mnt_list);
886 list_for_each_entry(m, &head, mnt_list)
887 m->mnt_ns = n;
888
889 list_splice(&head, n->list.prev);
890
891 n->mounts += n->pending_mounts;
892 n->pending_mounts = 0;
893
894 __attach_mnt(mnt, parent);
895 touch_mnt_namespace(n);
896}
897
898static struct mount *next_mnt(struct mount *p, struct mount *root)
899{
900 struct list_head *next = p->mnt_mounts.next;
901 if (next == &p->mnt_mounts) {
902 while (1) {
903 if (p == root)
904 return NULL;
905 next = p->mnt_child.next;
906 if (next != &p->mnt_parent->mnt_mounts)
907 break;
908 p = p->mnt_parent;
909 }
910 }
911 return list_entry(next, struct mount, mnt_child);
912}
913
914static struct mount *skip_mnt_tree(struct mount *p)
915{
916 struct list_head *prev = p->mnt_mounts.prev;
917 while (prev != &p->mnt_mounts) {
918 p = list_entry(prev, struct mount, mnt_child);
919 prev = p->mnt_mounts.prev;
920 }
921 return p;
922}
923
924/**
925 * vfs_create_mount - Create a mount for a configured superblock
926 * @fc: The configuration context with the superblock attached
927 *
928 * Create a mount to an already configured superblock. If necessary, the
929 * caller should invoke vfs_get_tree() before calling this.
930 *
931 * Note that this does not attach the mount to anything.
932 */
933struct vfsmount *vfs_create_mount(struct fs_context *fc)
934{
935 struct mount *mnt;
936
937 if (!fc->root)
938 return ERR_PTR(-EINVAL);
939
940 mnt = alloc_vfsmnt(fc->source ?: "none");
941 if (!mnt)
942 return ERR_PTR(-ENOMEM);
943
944 if (fc->sb_flags & SB_KERNMOUNT)
945 mnt->mnt.mnt_flags = MNT_INTERNAL;
946
947 atomic_inc(&fc->root->d_sb->s_active);
948 mnt->mnt.mnt_sb = fc->root->d_sb;
949 mnt->mnt.mnt_root = dget(fc->root);
950 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
951 mnt->mnt_parent = mnt;
952
953 lock_mount_hash();
954 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
955 unlock_mount_hash();
956 return &mnt->mnt;
957}
958EXPORT_SYMBOL(vfs_create_mount);
959
960struct vfsmount *fc_mount(struct fs_context *fc)
961{
962 int err = vfs_get_tree(fc);
963 if (!err) {
964 up_write(&fc->root->d_sb->s_umount);
965 return vfs_create_mount(fc);
966 }
967 return ERR_PTR(err);
968}
969EXPORT_SYMBOL(fc_mount);
970
971struct vfsmount *vfs_kern_mount(struct file_system_type *type,
972 int flags, const char *name,
973 void *data)
974{
975 struct fs_context *fc;
976 struct vfsmount *mnt;
977 int ret = 0;
978
979 if (!type)
980 return ERR_PTR(-EINVAL);
981
982 fc = fs_context_for_mount(type, flags);
983 if (IS_ERR(fc))
984 return ERR_CAST(fc);
985
986 if (name)
987 ret = vfs_parse_fs_string(fc, "source",
988 name, strlen(name));
989 if (!ret)
990 ret = parse_monolithic_mount_data(fc, data);
991 if (!ret)
992 mnt = fc_mount(fc);
993 else
994 mnt = ERR_PTR(ret);
995
996 put_fs_context(fc);
997 return mnt;
998}
999EXPORT_SYMBOL_GPL(vfs_kern_mount);
1000
1001struct vfsmount *
1002vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1003 const char *name, void *data)
1004{
1005 /* Until it is worked out how to pass the user namespace
1006 * through from the parent mount to the submount don't support
1007 * unprivileged mounts with submounts.
1008 */
1009 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1010 return ERR_PTR(-EPERM);
1011
1012 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1013}
1014EXPORT_SYMBOL_GPL(vfs_submount);
1015
1016static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1017 int flag)
1018{
1019 struct super_block *sb = old->mnt.mnt_sb;
1020 struct mount *mnt;
1021 int err;
1022
1023 mnt = alloc_vfsmnt(old->mnt_devname);
1024 if (!mnt)
1025 return ERR_PTR(-ENOMEM);
1026
1027 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1028 mnt->mnt_group_id = 0; /* not a peer of original */
1029 else
1030 mnt->mnt_group_id = old->mnt_group_id;
1031
1032 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1033 err = mnt_alloc_group_id(mnt);
1034 if (err)
1035 goto out_free;
1036 }
1037
1038 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1039 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1040
1041 atomic_inc(&sb->s_active);
1042 mnt->mnt.mnt_sb = sb;
1043 mnt->mnt.mnt_root = dget(root);
1044 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1045 mnt->mnt_parent = mnt;
1046 lock_mount_hash();
1047 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1048 unlock_mount_hash();
1049
1050 if ((flag & CL_SLAVE) ||
1051 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1052 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1053 mnt->mnt_master = old;
1054 CLEAR_MNT_SHARED(mnt);
1055 } else if (!(flag & CL_PRIVATE)) {
1056 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1057 list_add(&mnt->mnt_share, &old->mnt_share);
1058 if (IS_MNT_SLAVE(old))
1059 list_add(&mnt->mnt_slave, &old->mnt_slave);
1060 mnt->mnt_master = old->mnt_master;
1061 } else {
1062 CLEAR_MNT_SHARED(mnt);
1063 }
1064 if (flag & CL_MAKE_SHARED)
1065 set_mnt_shared(mnt);
1066
1067 /* stick the duplicate mount on the same expiry list
1068 * as the original if that was on one */
1069 if (flag & CL_EXPIRE) {
1070 if (!list_empty(&old->mnt_expire))
1071 list_add(&mnt->mnt_expire, &old->mnt_expire);
1072 }
1073
1074 return mnt;
1075
1076 out_free:
1077 mnt_free_id(mnt);
1078 free_vfsmnt(mnt);
1079 return ERR_PTR(err);
1080}
1081
1082static void cleanup_mnt(struct mount *mnt)
1083{
1084 struct hlist_node *p;
1085 struct mount *m;
1086 /*
1087 * The warning here probably indicates that somebody messed
1088 * up a mnt_want/drop_write() pair. If this happens, the
1089 * filesystem was probably unable to make r/w->r/o transitions.
1090 * The locking used to deal with mnt_count decrement provides barriers,
1091 * so mnt_get_writers() below is safe.
1092 */
1093 WARN_ON(mnt_get_writers(mnt));
1094 if (unlikely(mnt->mnt_pins.first))
1095 mnt_pin_kill(mnt);
1096 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1097 hlist_del(&m->mnt_umount);
1098 mntput(&m->mnt);
1099 }
1100 fsnotify_vfsmount_delete(&mnt->mnt);
1101 dput(mnt->mnt.mnt_root);
1102 deactivate_super(mnt->mnt.mnt_sb);
1103 mnt_free_id(mnt);
1104 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1105}
1106
1107static void __cleanup_mnt(struct rcu_head *head)
1108{
1109 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1110}
1111
1112static LLIST_HEAD(delayed_mntput_list);
1113static void delayed_mntput(struct work_struct *unused)
1114{
1115 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1116 struct mount *m, *t;
1117
1118 llist_for_each_entry_safe(m, t, node, mnt_llist)
1119 cleanup_mnt(m);
1120}
1121static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1122
1123static void mntput_no_expire(struct mount *mnt)
1124{
1125 LIST_HEAD(list);
1126
1127 rcu_read_lock();
1128 if (likely(READ_ONCE(mnt->mnt_ns))) {
1129 /*
1130 * Since we don't do lock_mount_hash() here,
1131 * ->mnt_ns can change under us. However, if it's
1132 * non-NULL, then there's a reference that won't
1133 * be dropped until after an RCU delay done after
1134 * turning ->mnt_ns NULL. So if we observe it
1135 * non-NULL under rcu_read_lock(), the reference
1136 * we are dropping is not the final one.
1137 */
1138 mnt_add_count(mnt, -1);
1139 rcu_read_unlock();
1140 return;
1141 }
1142 lock_mount_hash();
1143 /*
1144 * make sure that if __legitimize_mnt() has not seen us grab
1145 * mount_lock, we'll see their refcount increment here.
1146 */
1147 smp_mb();
1148 mnt_add_count(mnt, -1);
1149 if (mnt_get_count(mnt)) {
1150 rcu_read_unlock();
1151 unlock_mount_hash();
1152 return;
1153 }
1154 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1155 rcu_read_unlock();
1156 unlock_mount_hash();
1157 return;
1158 }
1159 mnt->mnt.mnt_flags |= MNT_DOOMED;
1160 rcu_read_unlock();
1161
1162 list_del(&mnt->mnt_instance);
1163
1164 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1165 struct mount *p, *tmp;
1166 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1167 __put_mountpoint(unhash_mnt(p), &list);
1168 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1169 }
1170 }
1171 unlock_mount_hash();
1172 shrink_dentry_list(&list);
1173
1174 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1175 struct task_struct *task = current;
1176 if (likely(!(task->flags & PF_KTHREAD))) {
1177 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1178 if (!task_work_add(task, &mnt->mnt_rcu, true))
1179 return;
1180 }
1181 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1182 schedule_delayed_work(&delayed_mntput_work, 1);
1183 return;
1184 }
1185 cleanup_mnt(mnt);
1186}
1187
1188void mntput(struct vfsmount *mnt)
1189{
1190 if (mnt) {
1191 struct mount *m = real_mount(mnt);
1192 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1193 if (unlikely(m->mnt_expiry_mark))
1194 m->mnt_expiry_mark = 0;
1195 mntput_no_expire(m);
1196 }
1197}
1198EXPORT_SYMBOL(mntput);
1199
1200struct vfsmount *mntget(struct vfsmount *mnt)
1201{
1202 if (mnt)
1203 mnt_add_count(real_mount(mnt), 1);
1204 return mnt;
1205}
1206EXPORT_SYMBOL(mntget);
1207
1208/* path_is_mountpoint() - Check if path is a mount in the current
1209 * namespace.
1210 *
1211 * d_mountpoint() can only be used reliably to establish if a dentry is
1212 * not mounted in any namespace and that common case is handled inline.
1213 * d_mountpoint() isn't aware of the possibility there may be multiple
1214 * mounts using a given dentry in a different namespace. This function
1215 * checks if the passed in path is a mountpoint rather than the dentry
1216 * alone.
1217 */
1218bool path_is_mountpoint(const struct path *path)
1219{
1220 unsigned seq;
1221 bool res;
1222
1223 if (!d_mountpoint(path->dentry))
1224 return false;
1225
1226 rcu_read_lock();
1227 do {
1228 seq = read_seqbegin(&mount_lock);
1229 res = __path_is_mountpoint(path);
1230 } while (read_seqretry(&mount_lock, seq));
1231 rcu_read_unlock();
1232
1233 return res;
1234}
1235EXPORT_SYMBOL(path_is_mountpoint);
1236
1237struct vfsmount *mnt_clone_internal(const struct path *path)
1238{
1239 struct mount *p;
1240 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1241 if (IS_ERR(p))
1242 return ERR_CAST(p);
1243 p->mnt.mnt_flags |= MNT_INTERNAL;
1244 return &p->mnt;
1245}
1246
1247#ifdef CONFIG_PROC_FS
1248/* iterator; we want it to have access to namespace_sem, thus here... */
1249static void *m_start(struct seq_file *m, loff_t *pos)
1250{
1251 struct proc_mounts *p = m->private;
1252
1253 down_read(&namespace_sem);
1254 if (p->cached_event == p->ns->event) {
1255 void *v = p->cached_mount;
1256 if (*pos == p->cached_index)
1257 return v;
1258 if (*pos == p->cached_index + 1) {
1259 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1260 return p->cached_mount = v;
1261 }
1262 }
1263
1264 p->cached_event = p->ns->event;
1265 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1266 p->cached_index = *pos;
1267 return p->cached_mount;
1268}
1269
1270static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1271{
1272 struct proc_mounts *p = m->private;
1273
1274 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1275 p->cached_index = *pos;
1276 return p->cached_mount;
1277}
1278
1279static void m_stop(struct seq_file *m, void *v)
1280{
1281 up_read(&namespace_sem);
1282}
1283
1284static int m_show(struct seq_file *m, void *v)
1285{
1286 struct proc_mounts *p = m->private;
1287 struct mount *r = list_entry(v, struct mount, mnt_list);
1288 return p->show(m, &r->mnt);
1289}
1290
1291const struct seq_operations mounts_op = {
1292 .start = m_start,
1293 .next = m_next,
1294 .stop = m_stop,
1295 .show = m_show,
1296};
1297#endif /* CONFIG_PROC_FS */
1298
1299/**
1300 * may_umount_tree - check if a mount tree is busy
1301 * @mnt: root of mount tree
1302 *
1303 * This is called to check if a tree of mounts has any
1304 * open files, pwds, chroots or sub mounts that are
1305 * busy.
1306 */
1307int may_umount_tree(struct vfsmount *m)
1308{
1309 struct mount *mnt = real_mount(m);
1310 int actual_refs = 0;
1311 int minimum_refs = 0;
1312 struct mount *p;
1313 BUG_ON(!m);
1314
1315 /* write lock needed for mnt_get_count */
1316 lock_mount_hash();
1317 for (p = mnt; p; p = next_mnt(p, mnt)) {
1318 actual_refs += mnt_get_count(p);
1319 minimum_refs += 2;
1320 }
1321 unlock_mount_hash();
1322
1323 if (actual_refs > minimum_refs)
1324 return 0;
1325
1326 return 1;
1327}
1328
1329EXPORT_SYMBOL(may_umount_tree);
1330
1331/**
1332 * may_umount - check if a mount point is busy
1333 * @mnt: root of mount
1334 *
1335 * This is called to check if a mount point has any
1336 * open files, pwds, chroots or sub mounts. If the
1337 * mount has sub mounts this will return busy
1338 * regardless of whether the sub mounts are busy.
1339 *
1340 * Doesn't take quota and stuff into account. IOW, in some cases it will
1341 * give false negatives. The main reason why it's here is that we need
1342 * a non-destructive way to look for easily umountable filesystems.
1343 */
1344int may_umount(struct vfsmount *mnt)
1345{
1346 int ret = 1;
1347 down_read(&namespace_sem);
1348 lock_mount_hash();
1349 if (propagate_mount_busy(real_mount(mnt), 2))
1350 ret = 0;
1351 unlock_mount_hash();
1352 up_read(&namespace_sem);
1353 return ret;
1354}
1355
1356EXPORT_SYMBOL(may_umount);
1357
1358static void namespace_unlock(void)
1359{
1360 struct hlist_head head;
1361 struct hlist_node *p;
1362 struct mount *m;
1363 LIST_HEAD(list);
1364
1365 hlist_move_list(&unmounted, &head);
1366 list_splice_init(&ex_mountpoints, &list);
1367
1368 up_write(&namespace_sem);
1369
1370 shrink_dentry_list(&list);
1371
1372 if (likely(hlist_empty(&head)))
1373 return;
1374
1375 synchronize_rcu_expedited();
1376
1377 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1378 hlist_del(&m->mnt_umount);
1379 mntput(&m->mnt);
1380 }
1381}
1382
1383static inline void namespace_lock(void)
1384{
1385 down_write(&namespace_sem);
1386}
1387
1388enum umount_tree_flags {
1389 UMOUNT_SYNC = 1,
1390 UMOUNT_PROPAGATE = 2,
1391 UMOUNT_CONNECTED = 4,
1392};
1393
1394static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1395{
1396 /* Leaving mounts connected is only valid for lazy umounts */
1397 if (how & UMOUNT_SYNC)
1398 return true;
1399
1400 /* A mount without a parent has nothing to be connected to */
1401 if (!mnt_has_parent(mnt))
1402 return true;
1403
1404 /* Because the reference counting rules change when mounts are
1405 * unmounted and connected, umounted mounts may not be
1406 * connected to mounted mounts.
1407 */
1408 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1409 return true;
1410
1411 /* Has it been requested that the mount remain connected? */
1412 if (how & UMOUNT_CONNECTED)
1413 return false;
1414
1415 /* Is the mount locked such that it needs to remain connected? */
1416 if (IS_MNT_LOCKED(mnt))
1417 return false;
1418
1419 /* By default disconnect the mount */
1420 return true;
1421}
1422
1423/*
1424 * mount_lock must be held
1425 * namespace_sem must be held for write
1426 */
1427static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1428{
1429 LIST_HEAD(tmp_list);
1430 struct mount *p;
1431
1432 if (how & UMOUNT_PROPAGATE)
1433 propagate_mount_unlock(mnt);
1434
1435 /* Gather the mounts to umount */
1436 for (p = mnt; p; p = next_mnt(p, mnt)) {
1437 p->mnt.mnt_flags |= MNT_UMOUNT;
1438 list_move(&p->mnt_list, &tmp_list);
1439 }
1440
1441 /* Hide the mounts from mnt_mounts */
1442 list_for_each_entry(p, &tmp_list, mnt_list) {
1443 list_del_init(&p->mnt_child);
1444 }
1445
1446 /* Add propogated mounts to the tmp_list */
1447 if (how & UMOUNT_PROPAGATE)
1448 propagate_umount(&tmp_list);
1449
1450 while (!list_empty(&tmp_list)) {
1451 struct mnt_namespace *ns;
1452 bool disconnect;
1453 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1454 list_del_init(&p->mnt_expire);
1455 list_del_init(&p->mnt_list);
1456 ns = p->mnt_ns;
1457 if (ns) {
1458 ns->mounts--;
1459 __touch_mnt_namespace(ns);
1460 }
1461 p->mnt_ns = NULL;
1462 if (how & UMOUNT_SYNC)
1463 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1464
1465 disconnect = disconnect_mount(p, how);
1466 if (mnt_has_parent(p)) {
1467 mnt_add_count(p->mnt_parent, -1);
1468 if (!disconnect) {
1469 /* Don't forget about p */
1470 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1471 } else {
1472 umount_mnt(p);
1473 }
1474 }
1475 change_mnt_propagation(p, MS_PRIVATE);
1476 if (disconnect)
1477 hlist_add_head(&p->mnt_umount, &unmounted);
1478 }
1479}
1480
1481static void shrink_submounts(struct mount *mnt);
1482
1483static int do_umount_root(struct super_block *sb)
1484{
1485 int ret = 0;
1486
1487 down_write(&sb->s_umount);
1488 if (!sb_rdonly(sb)) {
1489 struct fs_context *fc;
1490
1491 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1492 SB_RDONLY);
1493 if (IS_ERR(fc)) {
1494 ret = PTR_ERR(fc);
1495 } else {
1496 ret = parse_monolithic_mount_data(fc, NULL);
1497 if (!ret)
1498 ret = reconfigure_super(fc);
1499 put_fs_context(fc);
1500 }
1501 }
1502 up_write(&sb->s_umount);
1503 return ret;
1504}
1505
1506static int do_umount(struct mount *mnt, int flags)
1507{
1508 struct super_block *sb = mnt->mnt.mnt_sb;
1509 int retval;
1510
1511 retval = security_sb_umount(&mnt->mnt, flags);
1512 if (retval)
1513 return retval;
1514
1515 /*
1516 * Allow userspace to request a mountpoint be expired rather than
1517 * unmounting unconditionally. Unmount only happens if:
1518 * (1) the mark is already set (the mark is cleared by mntput())
1519 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1520 */
1521 if (flags & MNT_EXPIRE) {
1522 if (&mnt->mnt == current->fs->root.mnt ||
1523 flags & (MNT_FORCE | MNT_DETACH))
1524 return -EINVAL;
1525
1526 /*
1527 * probably don't strictly need the lock here if we examined
1528 * all race cases, but it's a slowpath.
1529 */
1530 lock_mount_hash();
1531 if (mnt_get_count(mnt) != 2) {
1532 unlock_mount_hash();
1533 return -EBUSY;
1534 }
1535 unlock_mount_hash();
1536
1537 if (!xchg(&mnt->mnt_expiry_mark, 1))
1538 return -EAGAIN;
1539 }
1540
1541 /*
1542 * If we may have to abort operations to get out of this
1543 * mount, and they will themselves hold resources we must
1544 * allow the fs to do things. In the Unix tradition of
1545 * 'Gee thats tricky lets do it in userspace' the umount_begin
1546 * might fail to complete on the first run through as other tasks
1547 * must return, and the like. Thats for the mount program to worry
1548 * about for the moment.
1549 */
1550
1551 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1552 sb->s_op->umount_begin(sb);
1553 }
1554
1555 /*
1556 * No sense to grab the lock for this test, but test itself looks
1557 * somewhat bogus. Suggestions for better replacement?
1558 * Ho-hum... In principle, we might treat that as umount + switch
1559 * to rootfs. GC would eventually take care of the old vfsmount.
1560 * Actually it makes sense, especially if rootfs would contain a
1561 * /reboot - static binary that would close all descriptors and
1562 * call reboot(9). Then init(8) could umount root and exec /reboot.
1563 */
1564 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1565 /*
1566 * Special case for "unmounting" root ...
1567 * we just try to remount it readonly.
1568 */
1569 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1570 return -EPERM;
1571 return do_umount_root(sb);
1572 }
1573
1574 namespace_lock();
1575 lock_mount_hash();
1576
1577 /* Recheck MNT_LOCKED with the locks held */
1578 retval = -EINVAL;
1579 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1580 goto out;
1581
1582 event++;
1583 if (flags & MNT_DETACH) {
1584 if (!list_empty(&mnt->mnt_list))
1585 umount_tree(mnt, UMOUNT_PROPAGATE);
1586 retval = 0;
1587 } else {
1588 shrink_submounts(mnt);
1589 retval = -EBUSY;
1590 if (!propagate_mount_busy(mnt, 2)) {
1591 if (!list_empty(&mnt->mnt_list))
1592 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1593 retval = 0;
1594 }
1595 }
1596out:
1597 unlock_mount_hash();
1598 namespace_unlock();
1599 return retval;
1600}
1601
1602/*
1603 * __detach_mounts - lazily unmount all mounts on the specified dentry
1604 *
1605 * During unlink, rmdir, and d_drop it is possible to loose the path
1606 * to an existing mountpoint, and wind up leaking the mount.
1607 * detach_mounts allows lazily unmounting those mounts instead of
1608 * leaking them.
1609 *
1610 * The caller may hold dentry->d_inode->i_mutex.
1611 */
1612void __detach_mounts(struct dentry *dentry)
1613{
1614 struct mountpoint *mp;
1615 struct mount *mnt;
1616
1617 namespace_lock();
1618 lock_mount_hash();
1619 mp = lookup_mountpoint(dentry);
1620 if (!mp)
1621 goto out_unlock;
1622
1623 event++;
1624 while (!hlist_empty(&mp->m_list)) {
1625 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1626 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1627 umount_mnt(mnt);
1628 hlist_add_head(&mnt->mnt_umount, &unmounted);
1629 }
1630 else umount_tree(mnt, UMOUNT_CONNECTED);
1631 }
1632 put_mountpoint(mp);
1633out_unlock:
1634 unlock_mount_hash();
1635 namespace_unlock();
1636}
1637
1638/*
1639 * Is the caller allowed to modify his namespace?
1640 */
1641static inline bool may_mount(void)
1642{
1643 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1644}
1645
1646#ifdef CONFIG_MANDATORY_FILE_LOCKING
1647static inline bool may_mandlock(void)
1648{
1649 return capable(CAP_SYS_ADMIN);
1650}
1651#else
1652static inline bool may_mandlock(void)
1653{
1654 pr_warn("VFS: \"mand\" mount option not supported");
1655 return false;
1656}
1657#endif
1658
1659/*
1660 * Now umount can handle mount points as well as block devices.
1661 * This is important for filesystems which use unnamed block devices.
1662 *
1663 * We now support a flag for forced unmount like the other 'big iron'
1664 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1665 */
1666
1667int ksys_umount(char __user *name, int flags)
1668{
1669 struct path path;
1670 struct mount *mnt;
1671 int retval;
1672 int lookup_flags = 0;
1673
1674 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1675 return -EINVAL;
1676
1677 if (!may_mount())
1678 return -EPERM;
1679
1680 if (!(flags & UMOUNT_NOFOLLOW))
1681 lookup_flags |= LOOKUP_FOLLOW;
1682
1683 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1684 if (retval)
1685 goto out;
1686 mnt = real_mount(path.mnt);
1687 retval = -EINVAL;
1688 if (path.dentry != path.mnt->mnt_root)
1689 goto dput_and_out;
1690 if (!check_mnt(mnt))
1691 goto dput_and_out;
1692 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1693 goto dput_and_out;
1694 retval = -EPERM;
1695 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1696 goto dput_and_out;
1697
1698 retval = do_umount(mnt, flags);
1699dput_and_out:
1700 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1701 dput(path.dentry);
1702 mntput_no_expire(mnt);
1703out:
1704 return retval;
1705}
1706
1707SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1708{
1709 return ksys_umount(name, flags);
1710}
1711
1712#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1713
1714/*
1715 * The 2.0 compatible umount. No flags.
1716 */
1717SYSCALL_DEFINE1(oldumount, char __user *, name)
1718{
1719 return ksys_umount(name, 0);
1720}
1721
1722#endif
1723
1724static bool is_mnt_ns_file(struct dentry *dentry)
1725{
1726 /* Is this a proxy for a mount namespace? */
1727 return dentry->d_op == &ns_dentry_operations &&
1728 dentry->d_fsdata == &mntns_operations;
1729}
1730
1731struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1732{
1733 return container_of(ns, struct mnt_namespace, ns);
1734}
1735
1736static bool mnt_ns_loop(struct dentry *dentry)
1737{
1738 /* Could bind mounting the mount namespace inode cause a
1739 * mount namespace loop?
1740 */
1741 struct mnt_namespace *mnt_ns;
1742 if (!is_mnt_ns_file(dentry))
1743 return false;
1744
1745 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1746 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1747}
1748
1749struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1750 int flag)
1751{
1752 struct mount *res, *p, *q, *r, *parent;
1753
1754 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1755 return ERR_PTR(-EINVAL);
1756
1757 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1758 return ERR_PTR(-EINVAL);
1759
1760 res = q = clone_mnt(mnt, dentry, flag);
1761 if (IS_ERR(q))
1762 return q;
1763
1764 q->mnt_mountpoint = mnt->mnt_mountpoint;
1765
1766 p = mnt;
1767 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1768 struct mount *s;
1769 if (!is_subdir(r->mnt_mountpoint, dentry))
1770 continue;
1771
1772 for (s = r; s; s = next_mnt(s, r)) {
1773 if (!(flag & CL_COPY_UNBINDABLE) &&
1774 IS_MNT_UNBINDABLE(s)) {
1775 if (s->mnt.mnt_flags & MNT_LOCKED) {
1776 /* Both unbindable and locked. */
1777 q = ERR_PTR(-EPERM);
1778 goto out;
1779 } else {
1780 s = skip_mnt_tree(s);
1781 continue;
1782 }
1783 }
1784 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1785 is_mnt_ns_file(s->mnt.mnt_root)) {
1786 s = skip_mnt_tree(s);
1787 continue;
1788 }
1789 while (p != s->mnt_parent) {
1790 p = p->mnt_parent;
1791 q = q->mnt_parent;
1792 }
1793 p = s;
1794 parent = q;
1795 q = clone_mnt(p, p->mnt.mnt_root, flag);
1796 if (IS_ERR(q))
1797 goto out;
1798 lock_mount_hash();
1799 list_add_tail(&q->mnt_list, &res->mnt_list);
1800 attach_mnt(q, parent, p->mnt_mp);
1801 unlock_mount_hash();
1802 }
1803 }
1804 return res;
1805out:
1806 if (res) {
1807 lock_mount_hash();
1808 umount_tree(res, UMOUNT_SYNC);
1809 unlock_mount_hash();
1810 }
1811 return q;
1812}
1813
1814/* Caller should check returned pointer for errors */
1815
1816struct vfsmount *collect_mounts(const struct path *path)
1817{
1818 struct mount *tree;
1819 namespace_lock();
1820 if (!check_mnt(real_mount(path->mnt)))
1821 tree = ERR_PTR(-EINVAL);
1822 else
1823 tree = copy_tree(real_mount(path->mnt), path->dentry,
1824 CL_COPY_ALL | CL_PRIVATE);
1825 namespace_unlock();
1826 if (IS_ERR(tree))
1827 return ERR_CAST(tree);
1828 return &tree->mnt;
1829}
1830
1831static void free_mnt_ns(struct mnt_namespace *);
1832static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1833
1834void dissolve_on_fput(struct vfsmount *mnt)
1835{
1836 struct mnt_namespace *ns;
1837 namespace_lock();
1838 lock_mount_hash();
1839 ns = real_mount(mnt)->mnt_ns;
1840 if (ns) {
1841 if (is_anon_ns(ns))
1842 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1843 else
1844 ns = NULL;
1845 }
1846 unlock_mount_hash();
1847 namespace_unlock();
1848 if (ns)
1849 free_mnt_ns(ns);
1850}
1851
1852void drop_collected_mounts(struct vfsmount *mnt)
1853{
1854 namespace_lock();
1855 lock_mount_hash();
1856 umount_tree(real_mount(mnt), 0);
1857 unlock_mount_hash();
1858 namespace_unlock();
1859}
1860
1861/**
1862 * clone_private_mount - create a private clone of a path
1863 *
1864 * This creates a new vfsmount, which will be the clone of @path. The new will
1865 * not be attached anywhere in the namespace and will be private (i.e. changes
1866 * to the originating mount won't be propagated into this).
1867 *
1868 * Release with mntput().
1869 */
1870struct vfsmount *clone_private_mount(const struct path *path)
1871{
1872 struct mount *old_mnt = real_mount(path->mnt);
1873 struct mount *new_mnt;
1874
1875 if (IS_MNT_UNBINDABLE(old_mnt))
1876 return ERR_PTR(-EINVAL);
1877
1878 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1879 if (IS_ERR(new_mnt))
1880 return ERR_CAST(new_mnt);
1881
1882 return &new_mnt->mnt;
1883}
1884EXPORT_SYMBOL_GPL(clone_private_mount);
1885
1886int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1887 struct vfsmount *root)
1888{
1889 struct mount *mnt;
1890 int res = f(root, arg);
1891 if (res)
1892 return res;
1893 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1894 res = f(&mnt->mnt, arg);
1895 if (res)
1896 return res;
1897 }
1898 return 0;
1899}
1900
1901static void lock_mnt_tree(struct mount *mnt)
1902{
1903 struct mount *p;
1904
1905 for (p = mnt; p; p = next_mnt(p, mnt)) {
1906 int flags = p->mnt.mnt_flags;
1907 /* Don't allow unprivileged users to change mount flags */
1908 flags |= MNT_LOCK_ATIME;
1909
1910 if (flags & MNT_READONLY)
1911 flags |= MNT_LOCK_READONLY;
1912
1913 if (flags & MNT_NODEV)
1914 flags |= MNT_LOCK_NODEV;
1915
1916 if (flags & MNT_NOSUID)
1917 flags |= MNT_LOCK_NOSUID;
1918
1919 if (flags & MNT_NOEXEC)
1920 flags |= MNT_LOCK_NOEXEC;
1921 /* Don't allow unprivileged users to reveal what is under a mount */
1922 if (list_empty(&p->mnt_expire))
1923 flags |= MNT_LOCKED;
1924 p->mnt.mnt_flags = flags;
1925 }
1926}
1927
1928static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1929{
1930 struct mount *p;
1931
1932 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1933 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1934 mnt_release_group_id(p);
1935 }
1936}
1937
1938static int invent_group_ids(struct mount *mnt, bool recurse)
1939{
1940 struct mount *p;
1941
1942 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1943 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1944 int err = mnt_alloc_group_id(p);
1945 if (err) {
1946 cleanup_group_ids(mnt, p);
1947 return err;
1948 }
1949 }
1950 }
1951
1952 return 0;
1953}
1954
1955int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1956{
1957 unsigned int max = READ_ONCE(sysctl_mount_max);
1958 unsigned int mounts = 0, old, pending, sum;
1959 struct mount *p;
1960
1961 for (p = mnt; p; p = next_mnt(p, mnt))
1962 mounts++;
1963
1964 old = ns->mounts;
1965 pending = ns->pending_mounts;
1966 sum = old + pending;
1967 if ((old > sum) ||
1968 (pending > sum) ||
1969 (max < sum) ||
1970 (mounts > (max - sum)))
1971 return -ENOSPC;
1972
1973 ns->pending_mounts = pending + mounts;
1974 return 0;
1975}
1976
1977/*
1978 * @source_mnt : mount tree to be attached
1979 * @nd : place the mount tree @source_mnt is attached
1980 * @parent_nd : if non-null, detach the source_mnt from its parent and
1981 * store the parent mount and mountpoint dentry.
1982 * (done when source_mnt is moved)
1983 *
1984 * NOTE: in the table below explains the semantics when a source mount
1985 * of a given type is attached to a destination mount of a given type.
1986 * ---------------------------------------------------------------------------
1987 * | BIND MOUNT OPERATION |
1988 * |**************************************************************************
1989 * | source-->| shared | private | slave | unbindable |
1990 * | dest | | | | |
1991 * | | | | | | |
1992 * | v | | | | |
1993 * |**************************************************************************
1994 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1995 * | | | | | |
1996 * |non-shared| shared (+) | private | slave (*) | invalid |
1997 * ***************************************************************************
1998 * A bind operation clones the source mount and mounts the clone on the
1999 * destination mount.
2000 *
2001 * (++) the cloned mount is propagated to all the mounts in the propagation
2002 * tree of the destination mount and the cloned mount is added to
2003 * the peer group of the source mount.
2004 * (+) the cloned mount is created under the destination mount and is marked
2005 * as shared. The cloned mount is added to the peer group of the source
2006 * mount.
2007 * (+++) the mount is propagated to all the mounts in the propagation tree
2008 * of the destination mount and the cloned mount is made slave
2009 * of the same master as that of the source mount. The cloned mount
2010 * is marked as 'shared and slave'.
2011 * (*) the cloned mount is made a slave of the same master as that of the
2012 * source mount.
2013 *
2014 * ---------------------------------------------------------------------------
2015 * | MOVE MOUNT OPERATION |
2016 * |**************************************************************************
2017 * | source-->| shared | private | slave | unbindable |
2018 * | dest | | | | |
2019 * | | | | | | |
2020 * | v | | | | |
2021 * |**************************************************************************
2022 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2023 * | | | | | |
2024 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2025 * ***************************************************************************
2026 *
2027 * (+) the mount is moved to the destination. And is then propagated to
2028 * all the mounts in the propagation tree of the destination mount.
2029 * (+*) the mount is moved to the destination.
2030 * (+++) the mount is moved to the destination and is then propagated to
2031 * all the mounts belonging to the destination mount's propagation tree.
2032 * the mount is marked as 'shared and slave'.
2033 * (*) the mount continues to be a slave at the new location.
2034 *
2035 * if the source mount is a tree, the operations explained above is
2036 * applied to each mount in the tree.
2037 * Must be called without spinlocks held, since this function can sleep
2038 * in allocations.
2039 */
2040static int attach_recursive_mnt(struct mount *source_mnt,
2041 struct mount *dest_mnt,
2042 struct mountpoint *dest_mp,
2043 bool moving)
2044{
2045 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2046 HLIST_HEAD(tree_list);
2047 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2048 struct mountpoint *smp;
2049 struct mount *child, *p;
2050 struct hlist_node *n;
2051 int err;
2052
2053 /* Preallocate a mountpoint in case the new mounts need
2054 * to be tucked under other mounts.
2055 */
2056 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2057 if (IS_ERR(smp))
2058 return PTR_ERR(smp);
2059
2060 /* Is there space to add these mounts to the mount namespace? */
2061 if (!moving) {
2062 err = count_mounts(ns, source_mnt);
2063 if (err)
2064 goto out;
2065 }
2066
2067 if (IS_MNT_SHARED(dest_mnt)) {
2068 err = invent_group_ids(source_mnt, true);
2069 if (err)
2070 goto out;
2071 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2072 lock_mount_hash();
2073 if (err)
2074 goto out_cleanup_ids;
2075 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2076 set_mnt_shared(p);
2077 } else {
2078 lock_mount_hash();
2079 }
2080 if (moving) {
2081 unhash_mnt(source_mnt);
2082 attach_mnt(source_mnt, dest_mnt, dest_mp);
2083 touch_mnt_namespace(source_mnt->mnt_ns);
2084 } else {
2085 if (source_mnt->mnt_ns) {
2086 /* move from anon - the caller will destroy */
2087 list_del_init(&source_mnt->mnt_ns->list);
2088 }
2089 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2090 commit_tree(source_mnt);
2091 }
2092
2093 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2094 struct mount *q;
2095 hlist_del_init(&child->mnt_hash);
2096 q = __lookup_mnt(&child->mnt_parent->mnt,
2097 child->mnt_mountpoint);
2098 if (q)
2099 mnt_change_mountpoint(child, smp, q);
2100 /* Notice when we are propagating across user namespaces */
2101 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2102 lock_mnt_tree(child);
2103 child->mnt.mnt_flags &= ~MNT_LOCKED;
2104 commit_tree(child);
2105 }
2106 put_mountpoint(smp);
2107 unlock_mount_hash();
2108
2109 return 0;
2110
2111 out_cleanup_ids:
2112 while (!hlist_empty(&tree_list)) {
2113 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2114 child->mnt_parent->mnt_ns->pending_mounts = 0;
2115 umount_tree(child, UMOUNT_SYNC);
2116 }
2117 unlock_mount_hash();
2118 cleanup_group_ids(source_mnt, NULL);
2119 out:
2120 ns->pending_mounts = 0;
2121
2122 read_seqlock_excl(&mount_lock);
2123 put_mountpoint(smp);
2124 read_sequnlock_excl(&mount_lock);
2125
2126 return err;
2127}
2128
2129static struct mountpoint *lock_mount(struct path *path)
2130{
2131 struct vfsmount *mnt;
2132 struct dentry *dentry = path->dentry;
2133retry:
2134 inode_lock(dentry->d_inode);
2135 if (unlikely(cant_mount(dentry))) {
2136 inode_unlock(dentry->d_inode);
2137 return ERR_PTR(-ENOENT);
2138 }
2139 namespace_lock();
2140 mnt = lookup_mnt(path);
2141 if (likely(!mnt)) {
2142 struct mountpoint *mp = get_mountpoint(dentry);
2143 if (IS_ERR(mp)) {
2144 namespace_unlock();
2145 inode_unlock(dentry->d_inode);
2146 return mp;
2147 }
2148 return mp;
2149 }
2150 namespace_unlock();
2151 inode_unlock(path->dentry->d_inode);
2152 path_put(path);
2153 path->mnt = mnt;
2154 dentry = path->dentry = dget(mnt->mnt_root);
2155 goto retry;
2156}
2157
2158static void unlock_mount(struct mountpoint *where)
2159{
2160 struct dentry *dentry = where->m_dentry;
2161
2162 read_seqlock_excl(&mount_lock);
2163 put_mountpoint(where);
2164 read_sequnlock_excl(&mount_lock);
2165
2166 namespace_unlock();
2167 inode_unlock(dentry->d_inode);
2168}
2169
2170static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2171{
2172 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2173 return -EINVAL;
2174
2175 if (d_is_dir(mp->m_dentry) !=
2176 d_is_dir(mnt->mnt.mnt_root))
2177 return -ENOTDIR;
2178
2179 return attach_recursive_mnt(mnt, p, mp, false);
2180}
2181
2182/*
2183 * Sanity check the flags to change_mnt_propagation.
2184 */
2185
2186static int flags_to_propagation_type(int ms_flags)
2187{
2188 int type = ms_flags & ~(MS_REC | MS_SILENT);
2189
2190 /* Fail if any non-propagation flags are set */
2191 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2192 return 0;
2193 /* Only one propagation flag should be set */
2194 if (!is_power_of_2(type))
2195 return 0;
2196 return type;
2197}
2198
2199/*
2200 * recursively change the type of the mountpoint.
2201 */
2202static int do_change_type(struct path *path, int ms_flags)
2203{
2204 struct mount *m;
2205 struct mount *mnt = real_mount(path->mnt);
2206 int recurse = ms_flags & MS_REC;
2207 int type;
2208 int err = 0;
2209
2210 if (path->dentry != path->mnt->mnt_root)
2211 return -EINVAL;
2212
2213 type = flags_to_propagation_type(ms_flags);
2214 if (!type)
2215 return -EINVAL;
2216
2217 namespace_lock();
2218 if (type == MS_SHARED) {
2219 err = invent_group_ids(mnt, recurse);
2220 if (err)
2221 goto out_unlock;
2222 }
2223
2224 lock_mount_hash();
2225 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2226 change_mnt_propagation(m, type);
2227 unlock_mount_hash();
2228
2229 out_unlock:
2230 namespace_unlock();
2231 return err;
2232}
2233
2234static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2235{
2236 struct mount *child;
2237 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2238 if (!is_subdir(child->mnt_mountpoint, dentry))
2239 continue;
2240
2241 if (child->mnt.mnt_flags & MNT_LOCKED)
2242 return true;
2243 }
2244 return false;
2245}
2246
2247static struct mount *__do_loopback(struct path *old_path, int recurse)
2248{
2249 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2250
2251 if (IS_MNT_UNBINDABLE(old))
2252 return mnt;
2253
2254 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2255 return mnt;
2256
2257 if (!recurse && has_locked_children(old, old_path->dentry))
2258 return mnt;
2259
2260 if (recurse)
2261 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2262 else
2263 mnt = clone_mnt(old, old_path->dentry, 0);
2264
2265 if (!IS_ERR(mnt))
2266 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2267
2268 return mnt;
2269}
2270
2271/*
2272 * do loopback mount.
2273 */
2274static int do_loopback(struct path *path, const char *old_name,
2275 int recurse)
2276{
2277 struct path old_path;
2278 struct mount *mnt = NULL, *parent;
2279 struct mountpoint *mp;
2280 int err;
2281 if (!old_name || !*old_name)
2282 return -EINVAL;
2283 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2284 if (err)
2285 return err;
2286
2287 err = -EINVAL;
2288 if (mnt_ns_loop(old_path.dentry))
2289 goto out;
2290
2291 mp = lock_mount(path);
2292 if (IS_ERR(mp)) {
2293 err = PTR_ERR(mp);
2294 goto out;
2295 }
2296
2297 parent = real_mount(path->mnt);
2298 if (!check_mnt(parent))
2299 goto out2;
2300
2301 mnt = __do_loopback(&old_path, recurse);
2302 if (IS_ERR(mnt)) {
2303 err = PTR_ERR(mnt);
2304 goto out2;
2305 }
2306
2307 err = graft_tree(mnt, parent, mp);
2308 if (err) {
2309 lock_mount_hash();
2310 umount_tree(mnt, UMOUNT_SYNC);
2311 unlock_mount_hash();
2312 }
2313out2:
2314 unlock_mount(mp);
2315out:
2316 path_put(&old_path);
2317 return err;
2318}
2319
2320static struct file *open_detached_copy(struct path *path, bool recursive)
2321{
2322 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2323 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2324 struct mount *mnt, *p;
2325 struct file *file;
2326
2327 if (IS_ERR(ns))
2328 return ERR_CAST(ns);
2329
2330 namespace_lock();
2331 mnt = __do_loopback(path, recursive);
2332 if (IS_ERR(mnt)) {
2333 namespace_unlock();
2334 free_mnt_ns(ns);
2335 return ERR_CAST(mnt);
2336 }
2337
2338 lock_mount_hash();
2339 for (p = mnt; p; p = next_mnt(p, mnt)) {
2340 p->mnt_ns = ns;
2341 ns->mounts++;
2342 }
2343 ns->root = mnt;
2344 list_add_tail(&ns->list, &mnt->mnt_list);
2345 mntget(&mnt->mnt);
2346 unlock_mount_hash();
2347 namespace_unlock();
2348
2349 mntput(path->mnt);
2350 path->mnt = &mnt->mnt;
2351 file = dentry_open(path, O_PATH, current_cred());
2352 if (IS_ERR(file))
2353 dissolve_on_fput(path->mnt);
2354 else
2355 file->f_mode |= FMODE_NEED_UNMOUNT;
2356 return file;
2357}
2358
2359SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags)
2360{
2361 struct file *file;
2362 struct path path;
2363 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2364 bool detached = flags & OPEN_TREE_CLONE;
2365 int error;
2366 int fd;
2367
2368 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2369
2370 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2371 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2372 OPEN_TREE_CLOEXEC))
2373 return -EINVAL;
2374
2375 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2376 return -EINVAL;
2377
2378 if (flags & AT_NO_AUTOMOUNT)
2379 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2380 if (flags & AT_SYMLINK_NOFOLLOW)
2381 lookup_flags &= ~LOOKUP_FOLLOW;
2382 if (flags & AT_EMPTY_PATH)
2383 lookup_flags |= LOOKUP_EMPTY;
2384
2385 if (detached && !may_mount())
2386 return -EPERM;
2387
2388 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2389 if (fd < 0)
2390 return fd;
2391
2392 error = user_path_at(dfd, filename, lookup_flags, &path);
2393 if (unlikely(error)) {
2394 file = ERR_PTR(error);
2395 } else {
2396 if (detached)
2397 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2398 else
2399 file = dentry_open(&path, O_PATH, current_cred());
2400 path_put(&path);
2401 }
2402 if (IS_ERR(file)) {
2403 put_unused_fd(fd);
2404 return PTR_ERR(file);
2405 }
2406 fd_install(fd, file);
2407 return fd;
2408}
2409
2410/*
2411 * Don't allow locked mount flags to be cleared.
2412 *
2413 * No locks need to be held here while testing the various MNT_LOCK
2414 * flags because those flags can never be cleared once they are set.
2415 */
2416static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2417{
2418 unsigned int fl = mnt->mnt.mnt_flags;
2419
2420 if ((fl & MNT_LOCK_READONLY) &&
2421 !(mnt_flags & MNT_READONLY))
2422 return false;
2423
2424 if ((fl & MNT_LOCK_NODEV) &&
2425 !(mnt_flags & MNT_NODEV))
2426 return false;
2427
2428 if ((fl & MNT_LOCK_NOSUID) &&
2429 !(mnt_flags & MNT_NOSUID))
2430 return false;
2431
2432 if ((fl & MNT_LOCK_NOEXEC) &&
2433 !(mnt_flags & MNT_NOEXEC))
2434 return false;
2435
2436 if ((fl & MNT_LOCK_ATIME) &&
2437 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2438 return false;
2439
2440 return true;
2441}
2442
2443static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2444{
2445 bool readonly_request = (mnt_flags & MNT_READONLY);
2446
2447 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2448 return 0;
2449
2450 if (readonly_request)
2451 return mnt_make_readonly(mnt);
2452
2453 return __mnt_unmake_readonly(mnt);
2454}
2455
2456/*
2457 * Update the user-settable attributes on a mount. The caller must hold
2458 * sb->s_umount for writing.
2459 */
2460static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2461{
2462 lock_mount_hash();
2463 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2464 mnt->mnt.mnt_flags = mnt_flags;
2465 touch_mnt_namespace(mnt->mnt_ns);
2466 unlock_mount_hash();
2467}
2468
2469static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2470{
2471 struct super_block *sb = mnt->mnt_sb;
2472
2473 if (!__mnt_is_readonly(mnt) &&
2474 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2475 char *buf = (char *)__get_free_page(GFP_KERNEL);
2476 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2477 struct tm tm;
2478
2479 time64_to_tm(sb->s_time_max, 0, &tm);
2480
2481 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2482 sb->s_type->name,
2483 is_mounted(mnt) ? "remounted" : "mounted",
2484 mntpath,
2485 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2486
2487 free_page((unsigned long)buf);
2488 }
2489}
2490
2491/*
2492 * Handle reconfiguration of the mountpoint only without alteration of the
2493 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2494 * to mount(2).
2495 */
2496static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2497{
2498 struct super_block *sb = path->mnt->mnt_sb;
2499 struct mount *mnt = real_mount(path->mnt);
2500 int ret;
2501
2502 if (!check_mnt(mnt))
2503 return -EINVAL;
2504
2505 if (path->dentry != mnt->mnt.mnt_root)
2506 return -EINVAL;
2507
2508 if (!can_change_locked_flags(mnt, mnt_flags))
2509 return -EPERM;
2510
2511 down_write(&sb->s_umount);
2512 ret = change_mount_ro_state(mnt, mnt_flags);
2513 if (ret == 0)
2514 set_mount_attributes(mnt, mnt_flags);
2515 up_write(&sb->s_umount);
2516
2517 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2518
2519 return ret;
2520}
2521
2522/*
2523 * change filesystem flags. dir should be a physical root of filesystem.
2524 * If you've mounted a non-root directory somewhere and want to do remount
2525 * on it - tough luck.
2526 */
2527static int do_remount(struct path *path, int ms_flags, int sb_flags,
2528 int mnt_flags, void *data)
2529{
2530 int err;
2531 struct super_block *sb = path->mnt->mnt_sb;
2532 struct mount *mnt = real_mount(path->mnt);
2533 struct fs_context *fc;
2534
2535 if (!check_mnt(mnt))
2536 return -EINVAL;
2537
2538 if (path->dentry != path->mnt->mnt_root)
2539 return -EINVAL;
2540
2541 if (!can_change_locked_flags(mnt, mnt_flags))
2542 return -EPERM;
2543
2544 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2545 if (IS_ERR(fc))
2546 return PTR_ERR(fc);
2547
2548 err = parse_monolithic_mount_data(fc, data);
2549 if (!err) {
2550 down_write(&sb->s_umount);
2551 err = -EPERM;
2552 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2553 err = reconfigure_super(fc);
2554 if (!err)
2555 set_mount_attributes(mnt, mnt_flags);
2556 }
2557 up_write(&sb->s_umount);
2558 }
2559
2560 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2561
2562 put_fs_context(fc);
2563 return err;
2564}
2565
2566static inline int tree_contains_unbindable(struct mount *mnt)
2567{
2568 struct mount *p;
2569 for (p = mnt; p; p = next_mnt(p, mnt)) {
2570 if (IS_MNT_UNBINDABLE(p))
2571 return 1;
2572 }
2573 return 0;
2574}
2575
2576/*
2577 * Check that there aren't references to earlier/same mount namespaces in the
2578 * specified subtree. Such references can act as pins for mount namespaces
2579 * that aren't checked by the mount-cycle checking code, thereby allowing
2580 * cycles to be made.
2581 */
2582static bool check_for_nsfs_mounts(struct mount *subtree)
2583{
2584 struct mount *p;
2585 bool ret = false;
2586
2587 lock_mount_hash();
2588 for (p = subtree; p; p = next_mnt(p, subtree))
2589 if (mnt_ns_loop(p->mnt.mnt_root))
2590 goto out;
2591
2592 ret = true;
2593out:
2594 unlock_mount_hash();
2595 return ret;
2596}
2597
2598static int do_move_mount(struct path *old_path, struct path *new_path)
2599{
2600 struct mnt_namespace *ns;
2601 struct mount *p;
2602 struct mount *old;
2603 struct mount *parent;
2604 struct mountpoint *mp, *old_mp;
2605 int err;
2606 bool attached;
2607
2608 mp = lock_mount(new_path);
2609 if (IS_ERR(mp))
2610 return PTR_ERR(mp);
2611
2612 old = real_mount(old_path->mnt);
2613 p = real_mount(new_path->mnt);
2614 parent = old->mnt_parent;
2615 attached = mnt_has_parent(old);
2616 old_mp = old->mnt_mp;
2617 ns = old->mnt_ns;
2618
2619 err = -EINVAL;
2620 /* The mountpoint must be in our namespace. */
2621 if (!check_mnt(p))
2622 goto out;
2623
2624 /* The thing moved must be mounted... */
2625 if (!is_mounted(&old->mnt))
2626 goto out;
2627
2628 /* ... and either ours or the root of anon namespace */
2629 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2630 goto out;
2631
2632 if (old->mnt.mnt_flags & MNT_LOCKED)
2633 goto out;
2634
2635 if (old_path->dentry != old_path->mnt->mnt_root)
2636 goto out;
2637
2638 if (d_is_dir(new_path->dentry) !=
2639 d_is_dir(old_path->dentry))
2640 goto out;
2641 /*
2642 * Don't move a mount residing in a shared parent.
2643 */
2644 if (attached && IS_MNT_SHARED(parent))
2645 goto out;
2646 /*
2647 * Don't move a mount tree containing unbindable mounts to a destination
2648 * mount which is shared.
2649 */
2650 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2651 goto out;
2652 err = -ELOOP;
2653 if (!check_for_nsfs_mounts(old))
2654 goto out;
2655 for (; mnt_has_parent(p); p = p->mnt_parent)
2656 if (p == old)
2657 goto out;
2658
2659 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2660 attached);
2661 if (err)
2662 goto out;
2663
2664 /* if the mount is moved, it should no longer be expire
2665 * automatically */
2666 list_del_init(&old->mnt_expire);
2667 if (attached)
2668 put_mountpoint(old_mp);
2669out:
2670 unlock_mount(mp);
2671 if (!err) {
2672 if (attached)
2673 mntput_no_expire(parent);
2674 else
2675 free_mnt_ns(ns);
2676 }
2677 return err;
2678}
2679
2680static int do_move_mount_old(struct path *path, const char *old_name)
2681{
2682 struct path old_path;
2683 int err;
2684
2685 if (!old_name || !*old_name)
2686 return -EINVAL;
2687
2688 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2689 if (err)
2690 return err;
2691
2692 err = do_move_mount(&old_path, path);
2693 path_put(&old_path);
2694 return err;
2695}
2696
2697/*
2698 * add a mount into a namespace's mount tree
2699 */
2700static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2701{
2702 struct mountpoint *mp;
2703 struct mount *parent;
2704 int err;
2705
2706 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2707
2708 mp = lock_mount(path);
2709 if (IS_ERR(mp))
2710 return PTR_ERR(mp);
2711
2712 parent = real_mount(path->mnt);
2713 err = -EINVAL;
2714 if (unlikely(!check_mnt(parent))) {
2715 /* that's acceptable only for automounts done in private ns */
2716 if (!(mnt_flags & MNT_SHRINKABLE))
2717 goto unlock;
2718 /* ... and for those we'd better have mountpoint still alive */
2719 if (!parent->mnt_ns)
2720 goto unlock;
2721 }
2722
2723 /* Refuse the same filesystem on the same mount point */
2724 err = -EBUSY;
2725 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2726 path->mnt->mnt_root == path->dentry)
2727 goto unlock;
2728
2729 err = -EINVAL;
2730 if (d_is_symlink(newmnt->mnt.mnt_root))
2731 goto unlock;
2732
2733 newmnt->mnt.mnt_flags = mnt_flags;
2734 err = graft_tree(newmnt, parent, mp);
2735
2736unlock:
2737 unlock_mount(mp);
2738 return err;
2739}
2740
2741static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2742
2743/*
2744 * Create a new mount using a superblock configuration and request it
2745 * be added to the namespace tree.
2746 */
2747static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2748 unsigned int mnt_flags)
2749{
2750 struct vfsmount *mnt;
2751 struct super_block *sb = fc->root->d_sb;
2752 int error;
2753
2754 error = security_sb_kern_mount(sb);
2755 if (!error && mount_too_revealing(sb, &mnt_flags))
2756 error = -EPERM;
2757
2758 if (unlikely(error)) {
2759 fc_drop_locked(fc);
2760 return error;
2761 }
2762
2763 up_write(&sb->s_umount);
2764
2765 mnt = vfs_create_mount(fc);
2766 if (IS_ERR(mnt))
2767 return PTR_ERR(mnt);
2768
2769 mnt_warn_timestamp_expiry(mountpoint, mnt);
2770
2771 error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags);
2772 if (error < 0)
2773 mntput(mnt);
2774 return error;
2775}
2776
2777/*
2778 * create a new mount for userspace and request it to be added into the
2779 * namespace's tree
2780 */
2781static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2782 int mnt_flags, const char *name, void *data)
2783{
2784 struct file_system_type *type;
2785 struct fs_context *fc;
2786 const char *subtype = NULL;
2787 int err = 0;
2788
2789 if (!fstype)
2790 return -EINVAL;
2791
2792 type = get_fs_type(fstype);
2793 if (!type)
2794 return -ENODEV;
2795
2796 if (type->fs_flags & FS_HAS_SUBTYPE) {
2797 subtype = strchr(fstype, '.');
2798 if (subtype) {
2799 subtype++;
2800 if (!*subtype) {
2801 put_filesystem(type);
2802 return -EINVAL;
2803 }
2804 }
2805 }
2806
2807 fc = fs_context_for_mount(type, sb_flags);
2808 put_filesystem(type);
2809 if (IS_ERR(fc))
2810 return PTR_ERR(fc);
2811
2812 if (subtype)
2813 err = vfs_parse_fs_string(fc, "subtype",
2814 subtype, strlen(subtype));
2815 if (!err && name)
2816 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2817 if (!err)
2818 err = parse_monolithic_mount_data(fc, data);
2819 if (!err && !mount_capable(fc))
2820 err = -EPERM;
2821 if (!err)
2822 err = vfs_get_tree(fc);
2823 if (!err)
2824 err = do_new_mount_fc(fc, path, mnt_flags);
2825
2826 put_fs_context(fc);
2827 return err;
2828}
2829
2830int finish_automount(struct vfsmount *m, struct path *path)
2831{
2832 struct mount *mnt = real_mount(m);
2833 int err;
2834 /* The new mount record should have at least 2 refs to prevent it being
2835 * expired before we get a chance to add it
2836 */
2837 BUG_ON(mnt_get_count(mnt) < 2);
2838
2839 if (m->mnt_sb == path->mnt->mnt_sb &&
2840 m->mnt_root == path->dentry) {
2841 err = -ELOOP;
2842 goto fail;
2843 }
2844
2845 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2846 if (!err)
2847 return 0;
2848fail:
2849 /* remove m from any expiration list it may be on */
2850 if (!list_empty(&mnt->mnt_expire)) {
2851 namespace_lock();
2852 list_del_init(&mnt->mnt_expire);
2853 namespace_unlock();
2854 }
2855 mntput(m);
2856 mntput(m);
2857 return err;
2858}
2859
2860/**
2861 * mnt_set_expiry - Put a mount on an expiration list
2862 * @mnt: The mount to list.
2863 * @expiry_list: The list to add the mount to.
2864 */
2865void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2866{
2867 namespace_lock();
2868
2869 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2870
2871 namespace_unlock();
2872}
2873EXPORT_SYMBOL(mnt_set_expiry);
2874
2875/*
2876 * process a list of expirable mountpoints with the intent of discarding any
2877 * mountpoints that aren't in use and haven't been touched since last we came
2878 * here
2879 */
2880void mark_mounts_for_expiry(struct list_head *mounts)
2881{
2882 struct mount *mnt, *next;
2883 LIST_HEAD(graveyard);
2884
2885 if (list_empty(mounts))
2886 return;
2887
2888 namespace_lock();
2889 lock_mount_hash();
2890
2891 /* extract from the expiration list every vfsmount that matches the
2892 * following criteria:
2893 * - only referenced by its parent vfsmount
2894 * - still marked for expiry (marked on the last call here; marks are
2895 * cleared by mntput())
2896 */
2897 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2898 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2899 propagate_mount_busy(mnt, 1))
2900 continue;
2901 list_move(&mnt->mnt_expire, &graveyard);
2902 }
2903 while (!list_empty(&graveyard)) {
2904 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2905 touch_mnt_namespace(mnt->mnt_ns);
2906 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2907 }
2908 unlock_mount_hash();
2909 namespace_unlock();
2910}
2911
2912EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2913
2914/*
2915 * Ripoff of 'select_parent()'
2916 *
2917 * search the list of submounts for a given mountpoint, and move any
2918 * shrinkable submounts to the 'graveyard' list.
2919 */
2920static int select_submounts(struct mount *parent, struct list_head *graveyard)
2921{
2922 struct mount *this_parent = parent;
2923 struct list_head *next;
2924 int found = 0;
2925
2926repeat:
2927 next = this_parent->mnt_mounts.next;
2928resume:
2929 while (next != &this_parent->mnt_mounts) {
2930 struct list_head *tmp = next;
2931 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2932
2933 next = tmp->next;
2934 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2935 continue;
2936 /*
2937 * Descend a level if the d_mounts list is non-empty.
2938 */
2939 if (!list_empty(&mnt->mnt_mounts)) {
2940 this_parent = mnt;
2941 goto repeat;
2942 }
2943
2944 if (!propagate_mount_busy(mnt, 1)) {
2945 list_move_tail(&mnt->mnt_expire, graveyard);
2946 found++;
2947 }
2948 }
2949 /*
2950 * All done at this level ... ascend and resume the search
2951 */
2952 if (this_parent != parent) {
2953 next = this_parent->mnt_child.next;
2954 this_parent = this_parent->mnt_parent;
2955 goto resume;
2956 }
2957 return found;
2958}
2959
2960/*
2961 * process a list of expirable mountpoints with the intent of discarding any
2962 * submounts of a specific parent mountpoint
2963 *
2964 * mount_lock must be held for write
2965 */
2966static void shrink_submounts(struct mount *mnt)
2967{
2968 LIST_HEAD(graveyard);
2969 struct mount *m;
2970
2971 /* extract submounts of 'mountpoint' from the expiration list */
2972 while (select_submounts(mnt, &graveyard)) {
2973 while (!list_empty(&graveyard)) {
2974 m = list_first_entry(&graveyard, struct mount,
2975 mnt_expire);
2976 touch_mnt_namespace(m->mnt_ns);
2977 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2978 }
2979 }
2980}
2981
2982/*
2983 * Some copy_from_user() implementations do not return the exact number of
2984 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2985 * Note that this function differs from copy_from_user() in that it will oops
2986 * on bad values of `to', rather than returning a short copy.
2987 */
2988static long exact_copy_from_user(void *to, const void __user * from,
2989 unsigned long n)
2990{
2991 char *t = to;
2992 const char __user *f = from;
2993 char c;
2994
2995 if (!access_ok(from, n))
2996 return n;
2997
2998 while (n) {
2999 if (__get_user(c, f)) {
3000 memset(t, 0, n);
3001 break;
3002 }
3003 *t++ = c;
3004 f++;
3005 n--;
3006 }
3007 return n;
3008}
3009
3010void *copy_mount_options(const void __user * data)
3011{
3012 int i;
3013 unsigned long size;
3014 char *copy;
3015
3016 if (!data)
3017 return NULL;
3018
3019 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3020 if (!copy)
3021 return ERR_PTR(-ENOMEM);
3022
3023 /* We only care that *some* data at the address the user
3024 * gave us is valid. Just in case, we'll zero
3025 * the remainder of the page.
3026 */
3027 /* copy_from_user cannot cross TASK_SIZE ! */
3028 size = TASK_SIZE - (unsigned long)untagged_addr(data);
3029 if (size > PAGE_SIZE)
3030 size = PAGE_SIZE;
3031
3032 i = size - exact_copy_from_user(copy, data, size);
3033 if (!i) {
3034 kfree(copy);
3035 return ERR_PTR(-EFAULT);
3036 }
3037 if (i != PAGE_SIZE)
3038 memset(copy + i, 0, PAGE_SIZE - i);
3039 return copy;
3040}
3041
3042char *copy_mount_string(const void __user *data)
3043{
3044 return data ? strndup_user(data, PATH_MAX) : NULL;
3045}
3046
3047/*
3048 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3049 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3050 *
3051 * data is a (void *) that can point to any structure up to
3052 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3053 * information (or be NULL).
3054 *
3055 * Pre-0.97 versions of mount() didn't have a flags word.
3056 * When the flags word was introduced its top half was required
3057 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3058 * Therefore, if this magic number is present, it carries no information
3059 * and must be discarded.
3060 */
3061long do_mount(const char *dev_name, const char __user *dir_name,
3062 const char *type_page, unsigned long flags, void *data_page)
3063{
3064 struct path path;
3065 unsigned int mnt_flags = 0, sb_flags;
3066 int retval = 0;
3067
3068 /* Discard magic */
3069 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3070 flags &= ~MS_MGC_MSK;
3071
3072 /* Basic sanity checks */
3073 if (data_page)
3074 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3075
3076 if (flags & MS_NOUSER)
3077 return -EINVAL;
3078
3079 /* ... and get the mountpoint */
3080 retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3081 if (retval)
3082 return retval;
3083
3084 retval = security_sb_mount(dev_name, &path,
3085 type_page, flags, data_page);
3086 if (!retval && !may_mount())
3087 retval = -EPERM;
3088 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
3089 retval = -EPERM;
3090 if (retval)
3091 goto dput_out;
3092
3093 /* Default to relatime unless overriden */
3094 if (!(flags & MS_NOATIME))
3095 mnt_flags |= MNT_RELATIME;
3096
3097 /* Separate the per-mountpoint flags */
3098 if (flags & MS_NOSUID)
3099 mnt_flags |= MNT_NOSUID;
3100 if (flags & MS_NODEV)
3101 mnt_flags |= MNT_NODEV;
3102 if (flags & MS_NOEXEC)
3103 mnt_flags |= MNT_NOEXEC;
3104 if (flags & MS_NOATIME)
3105 mnt_flags |= MNT_NOATIME;
3106 if (flags & MS_NODIRATIME)
3107 mnt_flags |= MNT_NODIRATIME;
3108 if (flags & MS_STRICTATIME)
3109 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3110 if (flags & MS_RDONLY)
3111 mnt_flags |= MNT_READONLY;
3112
3113 /* The default atime for remount is preservation */
3114 if ((flags & MS_REMOUNT) &&
3115 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3116 MS_STRICTATIME)) == 0)) {
3117 mnt_flags &= ~MNT_ATIME_MASK;
3118 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
3119 }
3120
3121 sb_flags = flags & (SB_RDONLY |
3122 SB_SYNCHRONOUS |
3123 SB_MANDLOCK |
3124 SB_DIRSYNC |
3125 SB_SILENT |
3126 SB_POSIXACL |
3127 SB_LAZYTIME |
3128 SB_I_VERSION);
3129
3130 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3131 retval = do_reconfigure_mnt(&path, mnt_flags);
3132 else if (flags & MS_REMOUNT)
3133 retval = do_remount(&path, flags, sb_flags, mnt_flags,
3134 data_page);
3135 else if (flags & MS_BIND)
3136 retval = do_loopback(&path, dev_name, flags & MS_REC);
3137 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3138 retval = do_change_type(&path, flags);
3139 else if (flags & MS_MOVE)
3140 retval = do_move_mount_old(&path, dev_name);
3141 else
3142 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
3143 dev_name, data_page);
3144dput_out:
3145 path_put(&path);
3146 return retval;
3147}
3148
3149static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3150{
3151 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3152}
3153
3154static void dec_mnt_namespaces(struct ucounts *ucounts)
3155{
3156 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3157}
3158
3159static void free_mnt_ns(struct mnt_namespace *ns)
3160{
3161 if (!is_anon_ns(ns))
3162 ns_free_inum(&ns->ns);
3163 dec_mnt_namespaces(ns->ucounts);
3164 put_user_ns(ns->user_ns);
3165 kfree(ns);
3166}
3167
3168/*
3169 * Assign a sequence number so we can detect when we attempt to bind
3170 * mount a reference to an older mount namespace into the current
3171 * mount namespace, preventing reference counting loops. A 64bit
3172 * number incrementing at 10Ghz will take 12,427 years to wrap which
3173 * is effectively never, so we can ignore the possibility.
3174 */
3175static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3176
3177static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3178{
3179 struct mnt_namespace *new_ns;
3180 struct ucounts *ucounts;
3181 int ret;
3182
3183 ucounts = inc_mnt_namespaces(user_ns);
3184 if (!ucounts)
3185 return ERR_PTR(-ENOSPC);
3186
3187 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3188 if (!new_ns) {
3189 dec_mnt_namespaces(ucounts);
3190 return ERR_PTR(-ENOMEM);
3191 }
3192 if (!anon) {
3193 ret = ns_alloc_inum(&new_ns->ns);
3194 if (ret) {
3195 kfree(new_ns);
3196 dec_mnt_namespaces(ucounts);
3197 return ERR_PTR(ret);
3198 }
3199 }
3200 new_ns->ns.ops = &mntns_operations;
3201 if (!anon)
3202 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3203 atomic_set(&new_ns->count, 1);
3204 INIT_LIST_HEAD(&new_ns->list);
3205 init_waitqueue_head(&new_ns->poll);
3206 new_ns->user_ns = get_user_ns(user_ns);
3207 new_ns->ucounts = ucounts;
3208 return new_ns;
3209}
3210
3211__latent_entropy
3212struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3213 struct user_namespace *user_ns, struct fs_struct *new_fs)
3214{
3215 struct mnt_namespace *new_ns;
3216 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3217 struct mount *p, *q;
3218 struct mount *old;
3219 struct mount *new;
3220 int copy_flags;
3221
3222 BUG_ON(!ns);
3223
3224 if (likely(!(flags & CLONE_NEWNS))) {
3225 get_mnt_ns(ns);
3226 return ns;
3227 }
3228
3229 old = ns->root;
3230
3231 new_ns = alloc_mnt_ns(user_ns, false);
3232 if (IS_ERR(new_ns))
3233 return new_ns;
3234
3235 namespace_lock();
3236 /* First pass: copy the tree topology */
3237 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3238 if (user_ns != ns->user_ns)
3239 copy_flags |= CL_SHARED_TO_SLAVE;
3240 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3241 if (IS_ERR(new)) {
3242 namespace_unlock();
3243 free_mnt_ns(new_ns);
3244 return ERR_CAST(new);
3245 }
3246 if (user_ns != ns->user_ns) {
3247 lock_mount_hash();
3248 lock_mnt_tree(new);
3249 unlock_mount_hash();
3250 }
3251 new_ns->root = new;
3252 list_add_tail(&new_ns->list, &new->mnt_list);
3253
3254 /*
3255 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3256 * as belonging to new namespace. We have already acquired a private
3257 * fs_struct, so tsk->fs->lock is not needed.
3258 */
3259 p = old;
3260 q = new;
3261 while (p) {
3262 q->mnt_ns = new_ns;
3263 new_ns->mounts++;
3264 if (new_fs) {
3265 if (&p->mnt == new_fs->root.mnt) {
3266 new_fs->root.mnt = mntget(&q->mnt);
3267 rootmnt = &p->mnt;
3268 }
3269 if (&p->mnt == new_fs->pwd.mnt) {
3270 new_fs->pwd.mnt = mntget(&q->mnt);
3271 pwdmnt = &p->mnt;
3272 }
3273 }
3274 p = next_mnt(p, old);
3275 q = next_mnt(q, new);
3276 if (!q)
3277 break;
3278 while (p->mnt.mnt_root != q->mnt.mnt_root)
3279 p = next_mnt(p, old);
3280 }
3281 namespace_unlock();
3282
3283 if (rootmnt)
3284 mntput(rootmnt);
3285 if (pwdmnt)
3286 mntput(pwdmnt);
3287
3288 return new_ns;
3289}
3290
3291struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3292{
3293 struct mount *mnt = real_mount(m);
3294 struct mnt_namespace *ns;
3295 struct super_block *s;
3296 struct path path;
3297 int err;
3298
3299 ns = alloc_mnt_ns(&init_user_ns, true);
3300 if (IS_ERR(ns)) {
3301 mntput(m);
3302 return ERR_CAST(ns);
3303 }
3304 mnt->mnt_ns = ns;
3305 ns->root = mnt;
3306 ns->mounts++;
3307 list_add(&mnt->mnt_list, &ns->list);
3308
3309 err = vfs_path_lookup(m->mnt_root, m,
3310 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3311
3312 put_mnt_ns(ns);
3313
3314 if (err)
3315 return ERR_PTR(err);
3316
3317 /* trade a vfsmount reference for active sb one */
3318 s = path.mnt->mnt_sb;
3319 atomic_inc(&s->s_active);
3320 mntput(path.mnt);
3321 /* lock the sucker */
3322 down_write(&s->s_umount);
3323 /* ... and return the root of (sub)tree on it */
3324 return path.dentry;
3325}
3326EXPORT_SYMBOL(mount_subtree);
3327
3328int ksys_mount(const char __user *dev_name, const char __user *dir_name,
3329 const char __user *type, unsigned long flags, void __user *data)
3330{
3331 int ret;
3332 char *kernel_type;
3333 char *kernel_dev;
3334 void *options;
3335
3336 kernel_type = copy_mount_string(type);
3337 ret = PTR_ERR(kernel_type);
3338 if (IS_ERR(kernel_type))
3339 goto out_type;
3340
3341 kernel_dev = copy_mount_string(dev_name);
3342 ret = PTR_ERR(kernel_dev);
3343 if (IS_ERR(kernel_dev))
3344 goto out_dev;
3345
3346 options = copy_mount_options(data);
3347 ret = PTR_ERR(options);
3348 if (IS_ERR(options))
3349 goto out_data;
3350
3351 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3352
3353 kfree(options);
3354out_data:
3355 kfree(kernel_dev);
3356out_dev:
3357 kfree(kernel_type);
3358out_type:
3359 return ret;
3360}
3361
3362SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3363 char __user *, type, unsigned long, flags, void __user *, data)
3364{
3365 return ksys_mount(dev_name, dir_name, type, flags, data);
3366}
3367
3368/*
3369 * Create a kernel mount representation for a new, prepared superblock
3370 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3371 */
3372SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3373 unsigned int, attr_flags)
3374{
3375 struct mnt_namespace *ns;
3376 struct fs_context *fc;
3377 struct file *file;
3378 struct path newmount;
3379 struct mount *mnt;
3380 struct fd f;
3381 unsigned int mnt_flags = 0;
3382 long ret;
3383
3384 if (!may_mount())
3385 return -EPERM;
3386
3387 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3388 return -EINVAL;
3389
3390 if (attr_flags & ~(MOUNT_ATTR_RDONLY |
3391 MOUNT_ATTR_NOSUID |
3392 MOUNT_ATTR_NODEV |
3393 MOUNT_ATTR_NOEXEC |
3394 MOUNT_ATTR__ATIME |
3395 MOUNT_ATTR_NODIRATIME))
3396 return -EINVAL;
3397
3398 if (attr_flags & MOUNT_ATTR_RDONLY)
3399 mnt_flags |= MNT_READONLY;
3400 if (attr_flags & MOUNT_ATTR_NOSUID)
3401 mnt_flags |= MNT_NOSUID;
3402 if (attr_flags & MOUNT_ATTR_NODEV)
3403 mnt_flags |= MNT_NODEV;
3404 if (attr_flags & MOUNT_ATTR_NOEXEC)
3405 mnt_flags |= MNT_NOEXEC;
3406 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3407 mnt_flags |= MNT_NODIRATIME;
3408
3409 switch (attr_flags & MOUNT_ATTR__ATIME) {
3410 case MOUNT_ATTR_STRICTATIME:
3411 break;
3412 case MOUNT_ATTR_NOATIME:
3413 mnt_flags |= MNT_NOATIME;
3414 break;
3415 case MOUNT_ATTR_RELATIME:
3416 mnt_flags |= MNT_RELATIME;
3417 break;
3418 default:
3419 return -EINVAL;
3420 }
3421
3422 f = fdget(fs_fd);
3423 if (!f.file)
3424 return -EBADF;
3425
3426 ret = -EINVAL;
3427 if (f.file->f_op != &fscontext_fops)
3428 goto err_fsfd;
3429
3430 fc = f.file->private_data;
3431
3432 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3433 if (ret < 0)
3434 goto err_fsfd;
3435
3436 /* There must be a valid superblock or we can't mount it */
3437 ret = -EINVAL;
3438 if (!fc->root)
3439 goto err_unlock;
3440
3441 ret = -EPERM;
3442 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3443 pr_warn("VFS: Mount too revealing\n");
3444 goto err_unlock;
3445 }
3446
3447 ret = -EBUSY;
3448 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3449 goto err_unlock;
3450
3451 ret = -EPERM;
3452 if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3453 goto err_unlock;
3454
3455 newmount.mnt = vfs_create_mount(fc);
3456 if (IS_ERR(newmount.mnt)) {
3457 ret = PTR_ERR(newmount.mnt);
3458 goto err_unlock;
3459 }
3460 newmount.dentry = dget(fc->root);
3461 newmount.mnt->mnt_flags = mnt_flags;
3462
3463 /* We've done the mount bit - now move the file context into more or
3464 * less the same state as if we'd done an fspick(). We don't want to
3465 * do any memory allocation or anything like that at this point as we
3466 * don't want to have to handle any errors incurred.
3467 */
3468 vfs_clean_context(fc);
3469
3470 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3471 if (IS_ERR(ns)) {
3472 ret = PTR_ERR(ns);
3473 goto err_path;
3474 }
3475 mnt = real_mount(newmount.mnt);
3476 mnt->mnt_ns = ns;
3477 ns->root = mnt;
3478 ns->mounts = 1;
3479 list_add(&mnt->mnt_list, &ns->list);
3480 mntget(newmount.mnt);
3481
3482 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3483 * it, not just simply put it.
3484 */
3485 file = dentry_open(&newmount, O_PATH, fc->cred);
3486 if (IS_ERR(file)) {
3487 dissolve_on_fput(newmount.mnt);
3488 ret = PTR_ERR(file);
3489 goto err_path;
3490 }
3491 file->f_mode |= FMODE_NEED_UNMOUNT;
3492
3493 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3494 if (ret >= 0)
3495 fd_install(ret, file);
3496 else
3497 fput(file);
3498
3499err_path:
3500 path_put(&newmount);
3501err_unlock:
3502 mutex_unlock(&fc->uapi_mutex);
3503err_fsfd:
3504 fdput(f);
3505 return ret;
3506}
3507
3508/*
3509 * Move a mount from one place to another. In combination with
3510 * fsopen()/fsmount() this is used to install a new mount and in combination
3511 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3512 * a mount subtree.
3513 *
3514 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3515 */
3516SYSCALL_DEFINE5(move_mount,
3517 int, from_dfd, const char *, from_pathname,
3518 int, to_dfd, const char *, to_pathname,
3519 unsigned int, flags)
3520{
3521 struct path from_path, to_path;
3522 unsigned int lflags;
3523 int ret = 0;
3524
3525 if (!may_mount())
3526 return -EPERM;
3527
3528 if (flags & ~MOVE_MOUNT__MASK)
3529 return -EINVAL;
3530
3531 /* If someone gives a pathname, they aren't permitted to move
3532 * from an fd that requires unmount as we can't get at the flag
3533 * to clear it afterwards.
3534 */
3535 lflags = 0;
3536 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3537 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3538 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3539
3540 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3541 if (ret < 0)
3542 return ret;
3543
3544 lflags = 0;
3545 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3546 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3547 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3548
3549 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3550 if (ret < 0)
3551 goto out_from;
3552
3553 ret = security_move_mount(&from_path, &to_path);
3554 if (ret < 0)
3555 goto out_to;
3556
3557 ret = do_move_mount(&from_path, &to_path);
3558
3559out_to:
3560 path_put(&to_path);
3561out_from:
3562 path_put(&from_path);
3563 return ret;
3564}
3565
3566/*
3567 * Return true if path is reachable from root
3568 *
3569 * namespace_sem or mount_lock is held
3570 */
3571bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3572 const struct path *root)
3573{
3574 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3575 dentry = mnt->mnt_mountpoint;
3576 mnt = mnt->mnt_parent;
3577 }
3578 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3579}
3580
3581bool path_is_under(const struct path *path1, const struct path *path2)
3582{
3583 bool res;
3584 read_seqlock_excl(&mount_lock);
3585 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3586 read_sequnlock_excl(&mount_lock);
3587 return res;
3588}
3589EXPORT_SYMBOL(path_is_under);
3590
3591/*
3592 * pivot_root Semantics:
3593 * Moves the root file system of the current process to the directory put_old,
3594 * makes new_root as the new root file system of the current process, and sets
3595 * root/cwd of all processes which had them on the current root to new_root.
3596 *
3597 * Restrictions:
3598 * The new_root and put_old must be directories, and must not be on the
3599 * same file system as the current process root. The put_old must be
3600 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3601 * pointed to by put_old must yield the same directory as new_root. No other
3602 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3603 *
3604 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3605 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3606 * in this situation.
3607 *
3608 * Notes:
3609 * - we don't move root/cwd if they are not at the root (reason: if something
3610 * cared enough to change them, it's probably wrong to force them elsewhere)
3611 * - it's okay to pick a root that isn't the root of a file system, e.g.
3612 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3613 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3614 * first.
3615 */
3616SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3617 const char __user *, put_old)
3618{
3619 struct path new, old, root;
3620 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3621 struct mountpoint *old_mp, *root_mp;
3622 int error;
3623
3624 if (!may_mount())
3625 return -EPERM;
3626
3627 error = user_path_at(AT_FDCWD, new_root,
3628 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3629 if (error)
3630 goto out0;
3631
3632 error = user_path_at(AT_FDCWD, put_old,
3633 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3634 if (error)
3635 goto out1;
3636
3637 error = security_sb_pivotroot(&old, &new);
3638 if (error)
3639 goto out2;
3640
3641 get_fs_root(current->fs, &root);
3642 old_mp = lock_mount(&old);
3643 error = PTR_ERR(old_mp);
3644 if (IS_ERR(old_mp))
3645 goto out3;
3646
3647 error = -EINVAL;
3648 new_mnt = real_mount(new.mnt);
3649 root_mnt = real_mount(root.mnt);
3650 old_mnt = real_mount(old.mnt);
3651 ex_parent = new_mnt->mnt_parent;
3652 root_parent = root_mnt->mnt_parent;
3653 if (IS_MNT_SHARED(old_mnt) ||
3654 IS_MNT_SHARED(ex_parent) ||
3655 IS_MNT_SHARED(root_parent))
3656 goto out4;
3657 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3658 goto out4;
3659 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3660 goto out4;
3661 error = -ENOENT;
3662 if (d_unlinked(new.dentry))
3663 goto out4;
3664 error = -EBUSY;
3665 if (new_mnt == root_mnt || old_mnt == root_mnt)
3666 goto out4; /* loop, on the same file system */
3667 error = -EINVAL;
3668 if (root.mnt->mnt_root != root.dentry)
3669 goto out4; /* not a mountpoint */
3670 if (!mnt_has_parent(root_mnt))
3671 goto out4; /* not attached */
3672 if (new.mnt->mnt_root != new.dentry)
3673 goto out4; /* not a mountpoint */
3674 if (!mnt_has_parent(new_mnt))
3675 goto out4; /* not attached */
3676 /* make sure we can reach put_old from new_root */
3677 if (!is_path_reachable(old_mnt, old.dentry, &new))
3678 goto out4;
3679 /* make certain new is below the root */
3680 if (!is_path_reachable(new_mnt, new.dentry, &root))
3681 goto out4;
3682 lock_mount_hash();
3683 umount_mnt(new_mnt);
3684 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
3685 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3686 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3687 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3688 }
3689 /* mount old root on put_old */
3690 attach_mnt(root_mnt, old_mnt, old_mp);
3691 /* mount new_root on / */
3692 attach_mnt(new_mnt, root_parent, root_mp);
3693 mnt_add_count(root_parent, -1);
3694 touch_mnt_namespace(current->nsproxy->mnt_ns);
3695 /* A moved mount should not expire automatically */
3696 list_del_init(&new_mnt->mnt_expire);
3697 put_mountpoint(root_mp);
3698 unlock_mount_hash();
3699 chroot_fs_refs(&root, &new);
3700 error = 0;
3701out4:
3702 unlock_mount(old_mp);
3703 if (!error)
3704 mntput_no_expire(ex_parent);
3705out3:
3706 path_put(&root);
3707out2:
3708 path_put(&old);
3709out1:
3710 path_put(&new);
3711out0:
3712 return error;
3713}
3714
3715static void __init init_mount_tree(void)
3716{
3717 struct vfsmount *mnt;
3718 struct mount *m;
3719 struct mnt_namespace *ns;
3720 struct path root;
3721
3722 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
3723 if (IS_ERR(mnt))
3724 panic("Can't create rootfs");
3725
3726 ns = alloc_mnt_ns(&init_user_ns, false);
3727 if (IS_ERR(ns))
3728 panic("Can't allocate initial namespace");
3729 m = real_mount(mnt);
3730 m->mnt_ns = ns;
3731 ns->root = m;
3732 ns->mounts = 1;
3733 list_add(&m->mnt_list, &ns->list);
3734 init_task.nsproxy->mnt_ns = ns;
3735 get_mnt_ns(ns);
3736
3737 root.mnt = mnt;
3738 root.dentry = mnt->mnt_root;
3739 mnt->mnt_flags |= MNT_LOCKED;
3740
3741 set_fs_pwd(current->fs, &root);
3742 set_fs_root(current->fs, &root);
3743}
3744
3745void __init mnt_init(void)
3746{
3747 int err;
3748
3749 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3750 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3751
3752 mount_hashtable = alloc_large_system_hash("Mount-cache",
3753 sizeof(struct hlist_head),
3754 mhash_entries, 19,
3755 HASH_ZERO,
3756 &m_hash_shift, &m_hash_mask, 0, 0);
3757 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3758 sizeof(struct hlist_head),
3759 mphash_entries, 19,
3760 HASH_ZERO,
3761 &mp_hash_shift, &mp_hash_mask, 0, 0);
3762
3763 if (!mount_hashtable || !mountpoint_hashtable)
3764 panic("Failed to allocate mount hash table\n");
3765
3766 kernfs_init();
3767
3768 err = sysfs_init();
3769 if (err)
3770 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3771 __func__, err);
3772 fs_kobj = kobject_create_and_add("fs", NULL);
3773 if (!fs_kobj)
3774 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3775 shmem_init();
3776 init_rootfs();
3777 init_mount_tree();
3778}
3779
3780void put_mnt_ns(struct mnt_namespace *ns)
3781{
3782 if (!atomic_dec_and_test(&ns->count))
3783 return;
3784 drop_collected_mounts(&ns->root->mnt);
3785 free_mnt_ns(ns);
3786}
3787
3788struct vfsmount *kern_mount(struct file_system_type *type)
3789{
3790 struct vfsmount *mnt;
3791 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
3792 if (!IS_ERR(mnt)) {
3793 /*
3794 * it is a longterm mount, don't release mnt until
3795 * we unmount before file sys is unregistered
3796 */
3797 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3798 }
3799 return mnt;
3800}
3801EXPORT_SYMBOL_GPL(kern_mount);
3802
3803void kern_unmount(struct vfsmount *mnt)
3804{
3805 /* release long term mount so mount point can be released */
3806 if (!IS_ERR_OR_NULL(mnt)) {
3807 real_mount(mnt)->mnt_ns = NULL;
3808 synchronize_rcu(); /* yecchhh... */
3809 mntput(mnt);
3810 }
3811}
3812EXPORT_SYMBOL(kern_unmount);
3813
3814bool our_mnt(struct vfsmount *mnt)
3815{
3816 return check_mnt(real_mount(mnt));
3817}
3818
3819bool current_chrooted(void)
3820{
3821 /* Does the current process have a non-standard root */
3822 struct path ns_root;
3823 struct path fs_root;
3824 bool chrooted;
3825
3826 /* Find the namespace root */
3827 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3828 ns_root.dentry = ns_root.mnt->mnt_root;
3829 path_get(&ns_root);
3830 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3831 ;
3832
3833 get_fs_root(current->fs, &fs_root);
3834
3835 chrooted = !path_equal(&fs_root, &ns_root);
3836
3837 path_put(&fs_root);
3838 path_put(&ns_root);
3839
3840 return chrooted;
3841}
3842
3843static bool mnt_already_visible(struct mnt_namespace *ns,
3844 const struct super_block *sb,
3845 int *new_mnt_flags)
3846{
3847 int new_flags = *new_mnt_flags;
3848 struct mount *mnt;
3849 bool visible = false;
3850
3851 down_read(&namespace_sem);
3852 list_for_each_entry(mnt, &ns->list, mnt_list) {
3853 struct mount *child;
3854 int mnt_flags;
3855
3856 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
3857 continue;
3858
3859 /* This mount is not fully visible if it's root directory
3860 * is not the root directory of the filesystem.
3861 */
3862 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3863 continue;
3864
3865 /* A local view of the mount flags */
3866 mnt_flags = mnt->mnt.mnt_flags;
3867
3868 /* Don't miss readonly hidden in the superblock flags */
3869 if (sb_rdonly(mnt->mnt.mnt_sb))
3870 mnt_flags |= MNT_LOCK_READONLY;
3871
3872 /* Verify the mount flags are equal to or more permissive
3873 * than the proposed new mount.
3874 */
3875 if ((mnt_flags & MNT_LOCK_READONLY) &&
3876 !(new_flags & MNT_READONLY))
3877 continue;
3878 if ((mnt_flags & MNT_LOCK_ATIME) &&
3879 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3880 continue;
3881
3882 /* This mount is not fully visible if there are any
3883 * locked child mounts that cover anything except for
3884 * empty directories.
3885 */
3886 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3887 struct inode *inode = child->mnt_mountpoint->d_inode;
3888 /* Only worry about locked mounts */
3889 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3890 continue;
3891 /* Is the directory permanetly empty? */
3892 if (!is_empty_dir_inode(inode))
3893 goto next;
3894 }
3895 /* Preserve the locked attributes */
3896 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3897 MNT_LOCK_ATIME);
3898 visible = true;
3899 goto found;
3900 next: ;
3901 }
3902found:
3903 up_read(&namespace_sem);
3904 return visible;
3905}
3906
3907static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
3908{
3909 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3910 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3911 unsigned long s_iflags;
3912
3913 if (ns->user_ns == &init_user_ns)
3914 return false;
3915
3916 /* Can this filesystem be too revealing? */
3917 s_iflags = sb->s_iflags;
3918 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3919 return false;
3920
3921 if ((s_iflags & required_iflags) != required_iflags) {
3922 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3923 required_iflags);
3924 return true;
3925 }
3926
3927 return !mnt_already_visible(ns, sb, new_mnt_flags);
3928}
3929
3930bool mnt_may_suid(struct vfsmount *mnt)
3931{
3932 /*
3933 * Foreign mounts (accessed via fchdir or through /proc
3934 * symlinks) are always treated as if they are nosuid. This
3935 * prevents namespaces from trusting potentially unsafe
3936 * suid/sgid bits, file caps, or security labels that originate
3937 * in other namespaces.
3938 */
3939 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3940 current_in_userns(mnt->mnt_sb->s_user_ns);
3941}
3942
3943static struct ns_common *mntns_get(struct task_struct *task)
3944{
3945 struct ns_common *ns = NULL;
3946 struct nsproxy *nsproxy;
3947
3948 task_lock(task);
3949 nsproxy = task->nsproxy;
3950 if (nsproxy) {
3951 ns = &nsproxy->mnt_ns->ns;
3952 get_mnt_ns(to_mnt_ns(ns));
3953 }
3954 task_unlock(task);
3955
3956 return ns;
3957}
3958
3959static void mntns_put(struct ns_common *ns)
3960{
3961 put_mnt_ns(to_mnt_ns(ns));
3962}
3963
3964static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3965{
3966 struct fs_struct *fs = current->fs;
3967 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3968 struct path root;
3969 int err;
3970
3971 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3972 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3973 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3974 return -EPERM;
3975
3976 if (is_anon_ns(mnt_ns))
3977 return -EINVAL;
3978
3979 if (fs->users != 1)
3980 return -EINVAL;
3981
3982 get_mnt_ns(mnt_ns);
3983 old_mnt_ns = nsproxy->mnt_ns;
3984 nsproxy->mnt_ns = mnt_ns;
3985
3986 /* Find the root */
3987 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3988 "/", LOOKUP_DOWN, &root);
3989 if (err) {
3990 /* revert to old namespace */
3991 nsproxy->mnt_ns = old_mnt_ns;
3992 put_mnt_ns(mnt_ns);
3993 return err;
3994 }
3995
3996 put_mnt_ns(old_mnt_ns);
3997
3998 /* Update the pwd and root */
3999 set_fs_pwd(fs, &root);
4000 set_fs_root(fs, &root);
4001
4002 path_put(&root);
4003 return 0;
4004}
4005
4006static struct user_namespace *mntns_owner(struct ns_common *ns)
4007{
4008 return to_mnt_ns(ns)->user_ns;
4009}
4010
4011const struct proc_ns_operations mntns_operations = {
4012 .name = "mnt",
4013 .type = CLONE_NEWNS,
4014 .get = mntns_get,
4015 .put = mntns_put,
4016 .install = mntns_install,
4017 .owner = mntns_owner,
4018};
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/namespace.c
4 *
5 * (C) Copyright Al Viro 2000, 2001
6 *
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
10
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>
35
36#include "pnode.h"
37#include "internal.h"
38
39/* Maximum number of mounts in a mount namespace */
40static unsigned int sysctl_mount_max __read_mostly = 100000;
41
42static unsigned int m_hash_mask __read_mostly;
43static unsigned int m_hash_shift __read_mostly;
44static unsigned int mp_hash_mask __read_mostly;
45static unsigned int mp_hash_shift __read_mostly;
46
47static __initdata unsigned long mhash_entries;
48static int __init set_mhash_entries(char *str)
49{
50 if (!str)
51 return 0;
52 mhash_entries = simple_strtoul(str, &str, 0);
53 return 1;
54}
55__setup("mhash_entries=", set_mhash_entries);
56
57static __initdata unsigned long mphash_entries;
58static int __init set_mphash_entries(char *str)
59{
60 if (!str)
61 return 0;
62 mphash_entries = simple_strtoul(str, &str, 0);
63 return 1;
64}
65__setup("mphash_entries=", set_mphash_entries);
66
67static u64 event;
68static DEFINE_IDA(mnt_id_ida);
69static DEFINE_IDA(mnt_group_ida);
70
71static struct hlist_head *mount_hashtable __read_mostly;
72static struct hlist_head *mountpoint_hashtable __read_mostly;
73static struct kmem_cache *mnt_cache __read_mostly;
74static DECLARE_RWSEM(namespace_sem);
75static HLIST_HEAD(unmounted); /* protected by namespace_sem */
76static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
77
78struct mnt_idmap {
79 struct user_namespace *owner;
80 refcount_t count;
81};
82
83/*
84 * Carries the initial idmapping of 0:0:4294967295 which is an identity
85 * mapping. This means that {g,u}id 0 is mapped to {g,u}id 0, {g,u}id 1 is
86 * mapped to {g,u}id 1, [...], {g,u}id 1000 to {g,u}id 1000, [...].
87 */
88struct mnt_idmap nop_mnt_idmap = {
89 .owner = &init_user_ns,
90 .count = REFCOUNT_INIT(1),
91};
92EXPORT_SYMBOL_GPL(nop_mnt_idmap);
93
94struct mount_kattr {
95 unsigned int attr_set;
96 unsigned int attr_clr;
97 unsigned int propagation;
98 unsigned int lookup_flags;
99 bool recurse;
100 struct user_namespace *mnt_userns;
101 struct mnt_idmap *mnt_idmap;
102};
103
104/* /sys/fs */
105struct kobject *fs_kobj;
106EXPORT_SYMBOL_GPL(fs_kobj);
107
108/*
109 * vfsmount lock may be taken for read to prevent changes to the
110 * vfsmount hash, ie. during mountpoint lookups or walking back
111 * up the tree.
112 *
113 * It should be taken for write in all cases where the vfsmount
114 * tree or hash is modified or when a vfsmount structure is modified.
115 */
116__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
117
118static inline void lock_mount_hash(void)
119{
120 write_seqlock(&mount_lock);
121}
122
123static inline void unlock_mount_hash(void)
124{
125 write_sequnlock(&mount_lock);
126}
127
128static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
129{
130 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
131 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
132 tmp = tmp + (tmp >> m_hash_shift);
133 return &mount_hashtable[tmp & m_hash_mask];
134}
135
136static inline struct hlist_head *mp_hash(struct dentry *dentry)
137{
138 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
139 tmp = tmp + (tmp >> mp_hash_shift);
140 return &mountpoint_hashtable[tmp & mp_hash_mask];
141}
142
143static int mnt_alloc_id(struct mount *mnt)
144{
145 int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
146
147 if (res < 0)
148 return res;
149 mnt->mnt_id = res;
150 return 0;
151}
152
153static void mnt_free_id(struct mount *mnt)
154{
155 ida_free(&mnt_id_ida, mnt->mnt_id);
156}
157
158/*
159 * Allocate a new peer group ID
160 */
161static int mnt_alloc_group_id(struct mount *mnt)
162{
163 int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
164
165 if (res < 0)
166 return res;
167 mnt->mnt_group_id = res;
168 return 0;
169}
170
171/*
172 * Release a peer group ID
173 */
174void mnt_release_group_id(struct mount *mnt)
175{
176 ida_free(&mnt_group_ida, mnt->mnt_group_id);
177 mnt->mnt_group_id = 0;
178}
179
180/*
181 * vfsmount lock must be held for read
182 */
183static inline void mnt_add_count(struct mount *mnt, int n)
184{
185#ifdef CONFIG_SMP
186 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
187#else
188 preempt_disable();
189 mnt->mnt_count += n;
190 preempt_enable();
191#endif
192}
193
194/*
195 * vfsmount lock must be held for write
196 */
197int mnt_get_count(struct mount *mnt)
198{
199#ifdef CONFIG_SMP
200 int count = 0;
201 int cpu;
202
203 for_each_possible_cpu(cpu) {
204 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
205 }
206
207 return count;
208#else
209 return mnt->mnt_count;
210#endif
211}
212
213/**
214 * mnt_idmap_owner - retrieve owner of the mount's idmapping
215 * @idmap: mount idmapping
216 *
217 * This helper will go away once the conversion to use struct mnt_idmap
218 * everywhere has finished at which point the helper will be unexported.
219 *
220 * Only code that needs to perform permission checks based on the owner of the
221 * idmapping will get access to it. All other code will solely rely on
222 * idmappings. This will get us type safety so it's impossible to conflate
223 * filesystems idmappings with mount idmappings.
224 *
225 * Return: The owner of the idmapping.
226 */
227struct user_namespace *mnt_idmap_owner(const struct mnt_idmap *idmap)
228{
229 return idmap->owner;
230}
231EXPORT_SYMBOL_GPL(mnt_idmap_owner);
232
233/**
234 * mnt_user_ns - retrieve owner of an idmapped mount
235 * @mnt: the relevant vfsmount
236 *
237 * This helper will go away once the conversion to use struct mnt_idmap
238 * everywhere has finished at which point the helper will be unexported.
239 *
240 * Only code that needs to perform permission checks based on the owner of the
241 * idmapping will get access to it. All other code will solely rely on
242 * idmappings. This will get us type safety so it's impossible to conflate
243 * filesystems idmappings with mount idmappings.
244 *
245 * Return: The owner of the idmapped.
246 */
247struct user_namespace *mnt_user_ns(const struct vfsmount *mnt)
248{
249 struct mnt_idmap *idmap = mnt_idmap(mnt);
250
251 /* Return the actual owner of the filesystem instead of the nop. */
252 if (idmap == &nop_mnt_idmap &&
253 !initial_idmapping(mnt->mnt_sb->s_user_ns))
254 return mnt->mnt_sb->s_user_ns;
255 return mnt_idmap_owner(idmap);
256}
257EXPORT_SYMBOL_GPL(mnt_user_ns);
258
259/**
260 * alloc_mnt_idmap - allocate a new idmapping for the mount
261 * @mnt_userns: owning userns of the idmapping
262 *
263 * Allocate a new struct mnt_idmap which carries the idmapping of the mount.
264 *
265 * Return: On success a new idmap, on error an error pointer is returned.
266 */
267static struct mnt_idmap *alloc_mnt_idmap(struct user_namespace *mnt_userns)
268{
269 struct mnt_idmap *idmap;
270
271 idmap = kzalloc(sizeof(struct mnt_idmap), GFP_KERNEL_ACCOUNT);
272 if (!idmap)
273 return ERR_PTR(-ENOMEM);
274
275 idmap->owner = get_user_ns(mnt_userns);
276 refcount_set(&idmap->count, 1);
277 return idmap;
278}
279
280/**
281 * mnt_idmap_get - get a reference to an idmapping
282 * @idmap: the idmap to bump the reference on
283 *
284 * If @idmap is not the @nop_mnt_idmap bump the reference count.
285 *
286 * Return: @idmap with reference count bumped if @not_mnt_idmap isn't passed.
287 */
288static inline struct mnt_idmap *mnt_idmap_get(struct mnt_idmap *idmap)
289{
290 if (idmap != &nop_mnt_idmap)
291 refcount_inc(&idmap->count);
292
293 return idmap;
294}
295
296/**
297 * mnt_idmap_put - put a reference to an idmapping
298 * @idmap: the idmap to put the reference on
299 *
300 * If this is a non-initial idmapping, put the reference count when a mount is
301 * released and free it if we're the last user.
302 */
303static inline void mnt_idmap_put(struct mnt_idmap *idmap)
304{
305 if (idmap != &nop_mnt_idmap && refcount_dec_and_test(&idmap->count)) {
306 put_user_ns(idmap->owner);
307 kfree(idmap);
308 }
309}
310
311static struct mount *alloc_vfsmnt(const char *name)
312{
313 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
314 if (mnt) {
315 int err;
316
317 err = mnt_alloc_id(mnt);
318 if (err)
319 goto out_free_cache;
320
321 if (name) {
322 mnt->mnt_devname = kstrdup_const(name,
323 GFP_KERNEL_ACCOUNT);
324 if (!mnt->mnt_devname)
325 goto out_free_id;
326 }
327
328#ifdef CONFIG_SMP
329 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
330 if (!mnt->mnt_pcp)
331 goto out_free_devname;
332
333 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
334#else
335 mnt->mnt_count = 1;
336 mnt->mnt_writers = 0;
337#endif
338
339 INIT_HLIST_NODE(&mnt->mnt_hash);
340 INIT_LIST_HEAD(&mnt->mnt_child);
341 INIT_LIST_HEAD(&mnt->mnt_mounts);
342 INIT_LIST_HEAD(&mnt->mnt_list);
343 INIT_LIST_HEAD(&mnt->mnt_expire);
344 INIT_LIST_HEAD(&mnt->mnt_share);
345 INIT_LIST_HEAD(&mnt->mnt_slave_list);
346 INIT_LIST_HEAD(&mnt->mnt_slave);
347 INIT_HLIST_NODE(&mnt->mnt_mp_list);
348 INIT_LIST_HEAD(&mnt->mnt_umounting);
349 INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
350 mnt->mnt.mnt_idmap = &nop_mnt_idmap;
351 }
352 return mnt;
353
354#ifdef CONFIG_SMP
355out_free_devname:
356 kfree_const(mnt->mnt_devname);
357#endif
358out_free_id:
359 mnt_free_id(mnt);
360out_free_cache:
361 kmem_cache_free(mnt_cache, mnt);
362 return NULL;
363}
364
365/*
366 * Most r/o checks on a fs are for operations that take
367 * discrete amounts of time, like a write() or unlink().
368 * We must keep track of when those operations start
369 * (for permission checks) and when they end, so that
370 * we can determine when writes are able to occur to
371 * a filesystem.
372 */
373/*
374 * __mnt_is_readonly: check whether a mount is read-only
375 * @mnt: the mount to check for its write status
376 *
377 * This shouldn't be used directly ouside of the VFS.
378 * It does not guarantee that the filesystem will stay
379 * r/w, just that it is right *now*. This can not and
380 * should not be used in place of IS_RDONLY(inode).
381 * mnt_want/drop_write() will _keep_ the filesystem
382 * r/w.
383 */
384bool __mnt_is_readonly(struct vfsmount *mnt)
385{
386 return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
387}
388EXPORT_SYMBOL_GPL(__mnt_is_readonly);
389
390static inline void mnt_inc_writers(struct mount *mnt)
391{
392#ifdef CONFIG_SMP
393 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
394#else
395 mnt->mnt_writers++;
396#endif
397}
398
399static inline void mnt_dec_writers(struct mount *mnt)
400{
401#ifdef CONFIG_SMP
402 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
403#else
404 mnt->mnt_writers--;
405#endif
406}
407
408static unsigned int mnt_get_writers(struct mount *mnt)
409{
410#ifdef CONFIG_SMP
411 unsigned int count = 0;
412 int cpu;
413
414 for_each_possible_cpu(cpu) {
415 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
416 }
417
418 return count;
419#else
420 return mnt->mnt_writers;
421#endif
422}
423
424static int mnt_is_readonly(struct vfsmount *mnt)
425{
426 if (mnt->mnt_sb->s_readonly_remount)
427 return 1;
428 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
429 smp_rmb();
430 return __mnt_is_readonly(mnt);
431}
432
433/*
434 * Most r/o & frozen checks on a fs are for operations that take discrete
435 * amounts of time, like a write() or unlink(). We must keep track of when
436 * those operations start (for permission checks) and when they end, so that we
437 * can determine when writes are able to occur to a filesystem.
438 */
439/**
440 * __mnt_want_write - get write access to a mount without freeze protection
441 * @m: the mount on which to take a write
442 *
443 * This tells the low-level filesystem that a write is about to be performed to
444 * it, and makes sure that writes are allowed (mnt it read-write) before
445 * returning success. This operation does not protect against filesystem being
446 * frozen. When the write operation is finished, __mnt_drop_write() must be
447 * called. This is effectively a refcount.
448 */
449int __mnt_want_write(struct vfsmount *m)
450{
451 struct mount *mnt = real_mount(m);
452 int ret = 0;
453
454 preempt_disable();
455 mnt_inc_writers(mnt);
456 /*
457 * The store to mnt_inc_writers must be visible before we pass
458 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
459 * incremented count after it has set MNT_WRITE_HOLD.
460 */
461 smp_mb();
462 might_lock(&mount_lock.lock);
463 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) {
464 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
465 cpu_relax();
466 } else {
467 /*
468 * This prevents priority inversion, if the task
469 * setting MNT_WRITE_HOLD got preempted on a remote
470 * CPU, and it prevents life lock if the task setting
471 * MNT_WRITE_HOLD has a lower priority and is bound to
472 * the same CPU as the task that is spinning here.
473 */
474 preempt_enable();
475 lock_mount_hash();
476 unlock_mount_hash();
477 preempt_disable();
478 }
479 }
480 /*
481 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
482 * be set to match its requirements. So we must not load that until
483 * MNT_WRITE_HOLD is cleared.
484 */
485 smp_rmb();
486 if (mnt_is_readonly(m)) {
487 mnt_dec_writers(mnt);
488 ret = -EROFS;
489 }
490 preempt_enable();
491
492 return ret;
493}
494
495/**
496 * mnt_want_write - get write access to a mount
497 * @m: the mount on which to take a write
498 *
499 * This tells the low-level filesystem that a write is about to be performed to
500 * it, and makes sure that writes are allowed (mount is read-write, filesystem
501 * is not frozen) before returning success. When the write operation is
502 * finished, mnt_drop_write() must be called. This is effectively a refcount.
503 */
504int mnt_want_write(struct vfsmount *m)
505{
506 int ret;
507
508 sb_start_write(m->mnt_sb);
509 ret = __mnt_want_write(m);
510 if (ret)
511 sb_end_write(m->mnt_sb);
512 return ret;
513}
514EXPORT_SYMBOL_GPL(mnt_want_write);
515
516/**
517 * __mnt_want_write_file - get write access to a file's mount
518 * @file: the file who's mount on which to take a write
519 *
520 * This is like __mnt_want_write, but if the file is already open for writing it
521 * skips incrementing mnt_writers (since the open file already has a reference)
522 * and instead only does the check for emergency r/o remounts. This must be
523 * paired with __mnt_drop_write_file.
524 */
525int __mnt_want_write_file(struct file *file)
526{
527 if (file->f_mode & FMODE_WRITER) {
528 /*
529 * Superblock may have become readonly while there are still
530 * writable fd's, e.g. due to a fs error with errors=remount-ro
531 */
532 if (__mnt_is_readonly(file->f_path.mnt))
533 return -EROFS;
534 return 0;
535 }
536 return __mnt_want_write(file->f_path.mnt);
537}
538
539/**
540 * mnt_want_write_file - get write access to a file's mount
541 * @file: the file who's mount on which to take a write
542 *
543 * This is like mnt_want_write, but if the file is already open for writing it
544 * skips incrementing mnt_writers (since the open file already has a reference)
545 * and instead only does the freeze protection and the check for emergency r/o
546 * remounts. This must be paired with mnt_drop_write_file.
547 */
548int mnt_want_write_file(struct file *file)
549{
550 int ret;
551
552 sb_start_write(file_inode(file)->i_sb);
553 ret = __mnt_want_write_file(file);
554 if (ret)
555 sb_end_write(file_inode(file)->i_sb);
556 return ret;
557}
558EXPORT_SYMBOL_GPL(mnt_want_write_file);
559
560/**
561 * __mnt_drop_write - give up write access to a mount
562 * @mnt: the mount on which to give up write access
563 *
564 * Tells the low-level filesystem that we are done
565 * performing writes to it. Must be matched with
566 * __mnt_want_write() call above.
567 */
568void __mnt_drop_write(struct vfsmount *mnt)
569{
570 preempt_disable();
571 mnt_dec_writers(real_mount(mnt));
572 preempt_enable();
573}
574
575/**
576 * mnt_drop_write - give up write access to a mount
577 * @mnt: the mount on which to give up write access
578 *
579 * Tells the low-level filesystem that we are done performing writes to it and
580 * also allows filesystem to be frozen again. Must be matched with
581 * mnt_want_write() call above.
582 */
583void mnt_drop_write(struct vfsmount *mnt)
584{
585 __mnt_drop_write(mnt);
586 sb_end_write(mnt->mnt_sb);
587}
588EXPORT_SYMBOL_GPL(mnt_drop_write);
589
590void __mnt_drop_write_file(struct file *file)
591{
592 if (!(file->f_mode & FMODE_WRITER))
593 __mnt_drop_write(file->f_path.mnt);
594}
595
596void mnt_drop_write_file(struct file *file)
597{
598 __mnt_drop_write_file(file);
599 sb_end_write(file_inode(file)->i_sb);
600}
601EXPORT_SYMBOL(mnt_drop_write_file);
602
603/**
604 * mnt_hold_writers - prevent write access to the given mount
605 * @mnt: mnt to prevent write access to
606 *
607 * Prevents write access to @mnt if there are no active writers for @mnt.
608 * This function needs to be called and return successfully before changing
609 * properties of @mnt that need to remain stable for callers with write access
610 * to @mnt.
611 *
612 * After this functions has been called successfully callers must pair it with
613 * a call to mnt_unhold_writers() in order to stop preventing write access to
614 * @mnt.
615 *
616 * Context: This function expects lock_mount_hash() to be held serializing
617 * setting MNT_WRITE_HOLD.
618 * Return: On success 0 is returned.
619 * On error, -EBUSY is returned.
620 */
621static inline int mnt_hold_writers(struct mount *mnt)
622{
623 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
624 /*
625 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
626 * should be visible before we do.
627 */
628 smp_mb();
629
630 /*
631 * With writers on hold, if this value is zero, then there are
632 * definitely no active writers (although held writers may subsequently
633 * increment the count, they'll have to wait, and decrement it after
634 * seeing MNT_READONLY).
635 *
636 * It is OK to have counter incremented on one CPU and decremented on
637 * another: the sum will add up correctly. The danger would be when we
638 * sum up each counter, if we read a counter before it is incremented,
639 * but then read another CPU's count which it has been subsequently
640 * decremented from -- we would see more decrements than we should.
641 * MNT_WRITE_HOLD protects against this scenario, because
642 * mnt_want_write first increments count, then smp_mb, then spins on
643 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
644 * we're counting up here.
645 */
646 if (mnt_get_writers(mnt) > 0)
647 return -EBUSY;
648
649 return 0;
650}
651
652/**
653 * mnt_unhold_writers - stop preventing write access to the given mount
654 * @mnt: mnt to stop preventing write access to
655 *
656 * Stop preventing write access to @mnt allowing callers to gain write access
657 * to @mnt again.
658 *
659 * This function can only be called after a successful call to
660 * mnt_hold_writers().
661 *
662 * Context: This function expects lock_mount_hash() to be held.
663 */
664static inline void mnt_unhold_writers(struct mount *mnt)
665{
666 /*
667 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
668 * that become unheld will see MNT_READONLY.
669 */
670 smp_wmb();
671 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
672}
673
674static int mnt_make_readonly(struct mount *mnt)
675{
676 int ret;
677
678 ret = mnt_hold_writers(mnt);
679 if (!ret)
680 mnt->mnt.mnt_flags |= MNT_READONLY;
681 mnt_unhold_writers(mnt);
682 return ret;
683}
684
685int sb_prepare_remount_readonly(struct super_block *sb)
686{
687 struct mount *mnt;
688 int err = 0;
689
690 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
691 if (atomic_long_read(&sb->s_remove_count))
692 return -EBUSY;
693
694 lock_mount_hash();
695 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
696 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
697 err = mnt_hold_writers(mnt);
698 if (err)
699 break;
700 }
701 }
702 if (!err && atomic_long_read(&sb->s_remove_count))
703 err = -EBUSY;
704
705 if (!err) {
706 sb->s_readonly_remount = 1;
707 smp_wmb();
708 }
709 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
710 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
711 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
712 }
713 unlock_mount_hash();
714
715 return err;
716}
717
718static void free_vfsmnt(struct mount *mnt)
719{
720 mnt_idmap_put(mnt_idmap(&mnt->mnt));
721 kfree_const(mnt->mnt_devname);
722#ifdef CONFIG_SMP
723 free_percpu(mnt->mnt_pcp);
724#endif
725 kmem_cache_free(mnt_cache, mnt);
726}
727
728static void delayed_free_vfsmnt(struct rcu_head *head)
729{
730 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
731}
732
733/* call under rcu_read_lock */
734int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
735{
736 struct mount *mnt;
737 if (read_seqretry(&mount_lock, seq))
738 return 1;
739 if (bastard == NULL)
740 return 0;
741 mnt = real_mount(bastard);
742 mnt_add_count(mnt, 1);
743 smp_mb(); // see mntput_no_expire()
744 if (likely(!read_seqretry(&mount_lock, seq)))
745 return 0;
746 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
747 mnt_add_count(mnt, -1);
748 return 1;
749 }
750 lock_mount_hash();
751 if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
752 mnt_add_count(mnt, -1);
753 unlock_mount_hash();
754 return 1;
755 }
756 unlock_mount_hash();
757 /* caller will mntput() */
758 return -1;
759}
760
761/* call under rcu_read_lock */
762static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
763{
764 int res = __legitimize_mnt(bastard, seq);
765 if (likely(!res))
766 return true;
767 if (unlikely(res < 0)) {
768 rcu_read_unlock();
769 mntput(bastard);
770 rcu_read_lock();
771 }
772 return false;
773}
774
775/*
776 * find the first mount at @dentry on vfsmount @mnt.
777 * call under rcu_read_lock()
778 */
779struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
780{
781 struct hlist_head *head = m_hash(mnt, dentry);
782 struct mount *p;
783
784 hlist_for_each_entry_rcu(p, head, mnt_hash)
785 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
786 return p;
787 return NULL;
788}
789
790/*
791 * lookup_mnt - Return the first child mount mounted at path
792 *
793 * "First" means first mounted chronologically. If you create the
794 * following mounts:
795 *
796 * mount /dev/sda1 /mnt
797 * mount /dev/sda2 /mnt
798 * mount /dev/sda3 /mnt
799 *
800 * Then lookup_mnt() on the base /mnt dentry in the root mount will
801 * return successively the root dentry and vfsmount of /dev/sda1, then
802 * /dev/sda2, then /dev/sda3, then NULL.
803 *
804 * lookup_mnt takes a reference to the found vfsmount.
805 */
806struct vfsmount *lookup_mnt(const struct path *path)
807{
808 struct mount *child_mnt;
809 struct vfsmount *m;
810 unsigned seq;
811
812 rcu_read_lock();
813 do {
814 seq = read_seqbegin(&mount_lock);
815 child_mnt = __lookup_mnt(path->mnt, path->dentry);
816 m = child_mnt ? &child_mnt->mnt : NULL;
817 } while (!legitimize_mnt(m, seq));
818 rcu_read_unlock();
819 return m;
820}
821
822static inline void lock_ns_list(struct mnt_namespace *ns)
823{
824 spin_lock(&ns->ns_lock);
825}
826
827static inline void unlock_ns_list(struct mnt_namespace *ns)
828{
829 spin_unlock(&ns->ns_lock);
830}
831
832static inline bool mnt_is_cursor(struct mount *mnt)
833{
834 return mnt->mnt.mnt_flags & MNT_CURSOR;
835}
836
837/*
838 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
839 * current mount namespace.
840 *
841 * The common case is dentries are not mountpoints at all and that
842 * test is handled inline. For the slow case when we are actually
843 * dealing with a mountpoint of some kind, walk through all of the
844 * mounts in the current mount namespace and test to see if the dentry
845 * is a mountpoint.
846 *
847 * The mount_hashtable is not usable in the context because we
848 * need to identify all mounts that may be in the current mount
849 * namespace not just a mount that happens to have some specified
850 * parent mount.
851 */
852bool __is_local_mountpoint(struct dentry *dentry)
853{
854 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
855 struct mount *mnt;
856 bool is_covered = false;
857
858 down_read(&namespace_sem);
859 lock_ns_list(ns);
860 list_for_each_entry(mnt, &ns->list, mnt_list) {
861 if (mnt_is_cursor(mnt))
862 continue;
863 is_covered = (mnt->mnt_mountpoint == dentry);
864 if (is_covered)
865 break;
866 }
867 unlock_ns_list(ns);
868 up_read(&namespace_sem);
869
870 return is_covered;
871}
872
873static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
874{
875 struct hlist_head *chain = mp_hash(dentry);
876 struct mountpoint *mp;
877
878 hlist_for_each_entry(mp, chain, m_hash) {
879 if (mp->m_dentry == dentry) {
880 mp->m_count++;
881 return mp;
882 }
883 }
884 return NULL;
885}
886
887static struct mountpoint *get_mountpoint(struct dentry *dentry)
888{
889 struct mountpoint *mp, *new = NULL;
890 int ret;
891
892 if (d_mountpoint(dentry)) {
893 /* might be worth a WARN_ON() */
894 if (d_unlinked(dentry))
895 return ERR_PTR(-ENOENT);
896mountpoint:
897 read_seqlock_excl(&mount_lock);
898 mp = lookup_mountpoint(dentry);
899 read_sequnlock_excl(&mount_lock);
900 if (mp)
901 goto done;
902 }
903
904 if (!new)
905 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
906 if (!new)
907 return ERR_PTR(-ENOMEM);
908
909
910 /* Exactly one processes may set d_mounted */
911 ret = d_set_mounted(dentry);
912
913 /* Someone else set d_mounted? */
914 if (ret == -EBUSY)
915 goto mountpoint;
916
917 /* The dentry is not available as a mountpoint? */
918 mp = ERR_PTR(ret);
919 if (ret)
920 goto done;
921
922 /* Add the new mountpoint to the hash table */
923 read_seqlock_excl(&mount_lock);
924 new->m_dentry = dget(dentry);
925 new->m_count = 1;
926 hlist_add_head(&new->m_hash, mp_hash(dentry));
927 INIT_HLIST_HEAD(&new->m_list);
928 read_sequnlock_excl(&mount_lock);
929
930 mp = new;
931 new = NULL;
932done:
933 kfree(new);
934 return mp;
935}
936
937/*
938 * vfsmount lock must be held. Additionally, the caller is responsible
939 * for serializing calls for given disposal list.
940 */
941static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
942{
943 if (!--mp->m_count) {
944 struct dentry *dentry = mp->m_dentry;
945 BUG_ON(!hlist_empty(&mp->m_list));
946 spin_lock(&dentry->d_lock);
947 dentry->d_flags &= ~DCACHE_MOUNTED;
948 spin_unlock(&dentry->d_lock);
949 dput_to_list(dentry, list);
950 hlist_del(&mp->m_hash);
951 kfree(mp);
952 }
953}
954
955/* called with namespace_lock and vfsmount lock */
956static void put_mountpoint(struct mountpoint *mp)
957{
958 __put_mountpoint(mp, &ex_mountpoints);
959}
960
961static inline int check_mnt(struct mount *mnt)
962{
963 return mnt->mnt_ns == current->nsproxy->mnt_ns;
964}
965
966/*
967 * vfsmount lock must be held for write
968 */
969static void touch_mnt_namespace(struct mnt_namespace *ns)
970{
971 if (ns) {
972 ns->event = ++event;
973 wake_up_interruptible(&ns->poll);
974 }
975}
976
977/*
978 * vfsmount lock must be held for write
979 */
980static void __touch_mnt_namespace(struct mnt_namespace *ns)
981{
982 if (ns && ns->event != event) {
983 ns->event = event;
984 wake_up_interruptible(&ns->poll);
985 }
986}
987
988/*
989 * vfsmount lock must be held for write
990 */
991static struct mountpoint *unhash_mnt(struct mount *mnt)
992{
993 struct mountpoint *mp;
994 mnt->mnt_parent = mnt;
995 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
996 list_del_init(&mnt->mnt_child);
997 hlist_del_init_rcu(&mnt->mnt_hash);
998 hlist_del_init(&mnt->mnt_mp_list);
999 mp = mnt->mnt_mp;
1000 mnt->mnt_mp = NULL;
1001 return mp;
1002}
1003
1004/*
1005 * vfsmount lock must be held for write
1006 */
1007static void umount_mnt(struct mount *mnt)
1008{
1009 put_mountpoint(unhash_mnt(mnt));
1010}
1011
1012/*
1013 * vfsmount lock must be held for write
1014 */
1015void mnt_set_mountpoint(struct mount *mnt,
1016 struct mountpoint *mp,
1017 struct mount *child_mnt)
1018{
1019 mp->m_count++;
1020 mnt_add_count(mnt, 1); /* essentially, that's mntget */
1021 child_mnt->mnt_mountpoint = mp->m_dentry;
1022 child_mnt->mnt_parent = mnt;
1023 child_mnt->mnt_mp = mp;
1024 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
1025}
1026
1027static void __attach_mnt(struct mount *mnt, struct mount *parent)
1028{
1029 hlist_add_head_rcu(&mnt->mnt_hash,
1030 m_hash(&parent->mnt, mnt->mnt_mountpoint));
1031 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
1032}
1033
1034/*
1035 * vfsmount lock must be held for write
1036 */
1037static void attach_mnt(struct mount *mnt,
1038 struct mount *parent,
1039 struct mountpoint *mp)
1040{
1041 mnt_set_mountpoint(parent, mp, mnt);
1042 __attach_mnt(mnt, parent);
1043}
1044
1045void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
1046{
1047 struct mountpoint *old_mp = mnt->mnt_mp;
1048 struct mount *old_parent = mnt->mnt_parent;
1049
1050 list_del_init(&mnt->mnt_child);
1051 hlist_del_init(&mnt->mnt_mp_list);
1052 hlist_del_init_rcu(&mnt->mnt_hash);
1053
1054 attach_mnt(mnt, parent, mp);
1055
1056 put_mountpoint(old_mp);
1057 mnt_add_count(old_parent, -1);
1058}
1059
1060/*
1061 * vfsmount lock must be held for write
1062 */
1063static void commit_tree(struct mount *mnt)
1064{
1065 struct mount *parent = mnt->mnt_parent;
1066 struct mount *m;
1067 LIST_HEAD(head);
1068 struct mnt_namespace *n = parent->mnt_ns;
1069
1070 BUG_ON(parent == mnt);
1071
1072 list_add_tail(&head, &mnt->mnt_list);
1073 list_for_each_entry(m, &head, mnt_list)
1074 m->mnt_ns = n;
1075
1076 list_splice(&head, n->list.prev);
1077
1078 n->mounts += n->pending_mounts;
1079 n->pending_mounts = 0;
1080
1081 __attach_mnt(mnt, parent);
1082 touch_mnt_namespace(n);
1083}
1084
1085static struct mount *next_mnt(struct mount *p, struct mount *root)
1086{
1087 struct list_head *next = p->mnt_mounts.next;
1088 if (next == &p->mnt_mounts) {
1089 while (1) {
1090 if (p == root)
1091 return NULL;
1092 next = p->mnt_child.next;
1093 if (next != &p->mnt_parent->mnt_mounts)
1094 break;
1095 p = p->mnt_parent;
1096 }
1097 }
1098 return list_entry(next, struct mount, mnt_child);
1099}
1100
1101static struct mount *skip_mnt_tree(struct mount *p)
1102{
1103 struct list_head *prev = p->mnt_mounts.prev;
1104 while (prev != &p->mnt_mounts) {
1105 p = list_entry(prev, struct mount, mnt_child);
1106 prev = p->mnt_mounts.prev;
1107 }
1108 return p;
1109}
1110
1111/**
1112 * vfs_create_mount - Create a mount for a configured superblock
1113 * @fc: The configuration context with the superblock attached
1114 *
1115 * Create a mount to an already configured superblock. If necessary, the
1116 * caller should invoke vfs_get_tree() before calling this.
1117 *
1118 * Note that this does not attach the mount to anything.
1119 */
1120struct vfsmount *vfs_create_mount(struct fs_context *fc)
1121{
1122 struct mount *mnt;
1123
1124 if (!fc->root)
1125 return ERR_PTR(-EINVAL);
1126
1127 mnt = alloc_vfsmnt(fc->source ?: "none");
1128 if (!mnt)
1129 return ERR_PTR(-ENOMEM);
1130
1131 if (fc->sb_flags & SB_KERNMOUNT)
1132 mnt->mnt.mnt_flags = MNT_INTERNAL;
1133
1134 atomic_inc(&fc->root->d_sb->s_active);
1135 mnt->mnt.mnt_sb = fc->root->d_sb;
1136 mnt->mnt.mnt_root = dget(fc->root);
1137 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1138 mnt->mnt_parent = mnt;
1139
1140 lock_mount_hash();
1141 list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1142 unlock_mount_hash();
1143 return &mnt->mnt;
1144}
1145EXPORT_SYMBOL(vfs_create_mount);
1146
1147struct vfsmount *fc_mount(struct fs_context *fc)
1148{
1149 int err = vfs_get_tree(fc);
1150 if (!err) {
1151 up_write(&fc->root->d_sb->s_umount);
1152 return vfs_create_mount(fc);
1153 }
1154 return ERR_PTR(err);
1155}
1156EXPORT_SYMBOL(fc_mount);
1157
1158struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1159 int flags, const char *name,
1160 void *data)
1161{
1162 struct fs_context *fc;
1163 struct vfsmount *mnt;
1164 int ret = 0;
1165
1166 if (!type)
1167 return ERR_PTR(-EINVAL);
1168
1169 fc = fs_context_for_mount(type, flags);
1170 if (IS_ERR(fc))
1171 return ERR_CAST(fc);
1172
1173 if (name)
1174 ret = vfs_parse_fs_string(fc, "source",
1175 name, strlen(name));
1176 if (!ret)
1177 ret = parse_monolithic_mount_data(fc, data);
1178 if (!ret)
1179 mnt = fc_mount(fc);
1180 else
1181 mnt = ERR_PTR(ret);
1182
1183 put_fs_context(fc);
1184 return mnt;
1185}
1186EXPORT_SYMBOL_GPL(vfs_kern_mount);
1187
1188struct vfsmount *
1189vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1190 const char *name, void *data)
1191{
1192 /* Until it is worked out how to pass the user namespace
1193 * through from the parent mount to the submount don't support
1194 * unprivileged mounts with submounts.
1195 */
1196 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1197 return ERR_PTR(-EPERM);
1198
1199 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1200}
1201EXPORT_SYMBOL_GPL(vfs_submount);
1202
1203static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1204 int flag)
1205{
1206 struct super_block *sb = old->mnt.mnt_sb;
1207 struct mount *mnt;
1208 int err;
1209
1210 mnt = alloc_vfsmnt(old->mnt_devname);
1211 if (!mnt)
1212 return ERR_PTR(-ENOMEM);
1213
1214 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1215 mnt->mnt_group_id = 0; /* not a peer of original */
1216 else
1217 mnt->mnt_group_id = old->mnt_group_id;
1218
1219 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1220 err = mnt_alloc_group_id(mnt);
1221 if (err)
1222 goto out_free;
1223 }
1224
1225 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1226 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1227
1228 atomic_inc(&sb->s_active);
1229 mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt));
1230
1231 mnt->mnt.mnt_sb = sb;
1232 mnt->mnt.mnt_root = dget(root);
1233 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1234 mnt->mnt_parent = mnt;
1235 lock_mount_hash();
1236 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1237 unlock_mount_hash();
1238
1239 if ((flag & CL_SLAVE) ||
1240 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1241 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1242 mnt->mnt_master = old;
1243 CLEAR_MNT_SHARED(mnt);
1244 } else if (!(flag & CL_PRIVATE)) {
1245 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1246 list_add(&mnt->mnt_share, &old->mnt_share);
1247 if (IS_MNT_SLAVE(old))
1248 list_add(&mnt->mnt_slave, &old->mnt_slave);
1249 mnt->mnt_master = old->mnt_master;
1250 } else {
1251 CLEAR_MNT_SHARED(mnt);
1252 }
1253 if (flag & CL_MAKE_SHARED)
1254 set_mnt_shared(mnt);
1255
1256 /* stick the duplicate mount on the same expiry list
1257 * as the original if that was on one */
1258 if (flag & CL_EXPIRE) {
1259 if (!list_empty(&old->mnt_expire))
1260 list_add(&mnt->mnt_expire, &old->mnt_expire);
1261 }
1262
1263 return mnt;
1264
1265 out_free:
1266 mnt_free_id(mnt);
1267 free_vfsmnt(mnt);
1268 return ERR_PTR(err);
1269}
1270
1271static void cleanup_mnt(struct mount *mnt)
1272{
1273 struct hlist_node *p;
1274 struct mount *m;
1275 /*
1276 * The warning here probably indicates that somebody messed
1277 * up a mnt_want/drop_write() pair. If this happens, the
1278 * filesystem was probably unable to make r/w->r/o transitions.
1279 * The locking used to deal with mnt_count decrement provides barriers,
1280 * so mnt_get_writers() below is safe.
1281 */
1282 WARN_ON(mnt_get_writers(mnt));
1283 if (unlikely(mnt->mnt_pins.first))
1284 mnt_pin_kill(mnt);
1285 hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1286 hlist_del(&m->mnt_umount);
1287 mntput(&m->mnt);
1288 }
1289 fsnotify_vfsmount_delete(&mnt->mnt);
1290 dput(mnt->mnt.mnt_root);
1291 deactivate_super(mnt->mnt.mnt_sb);
1292 mnt_free_id(mnt);
1293 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1294}
1295
1296static void __cleanup_mnt(struct rcu_head *head)
1297{
1298 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1299}
1300
1301static LLIST_HEAD(delayed_mntput_list);
1302static void delayed_mntput(struct work_struct *unused)
1303{
1304 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1305 struct mount *m, *t;
1306
1307 llist_for_each_entry_safe(m, t, node, mnt_llist)
1308 cleanup_mnt(m);
1309}
1310static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1311
1312static void mntput_no_expire(struct mount *mnt)
1313{
1314 LIST_HEAD(list);
1315 int count;
1316
1317 rcu_read_lock();
1318 if (likely(READ_ONCE(mnt->mnt_ns))) {
1319 /*
1320 * Since we don't do lock_mount_hash() here,
1321 * ->mnt_ns can change under us. However, if it's
1322 * non-NULL, then there's a reference that won't
1323 * be dropped until after an RCU delay done after
1324 * turning ->mnt_ns NULL. So if we observe it
1325 * non-NULL under rcu_read_lock(), the reference
1326 * we are dropping is not the final one.
1327 */
1328 mnt_add_count(mnt, -1);
1329 rcu_read_unlock();
1330 return;
1331 }
1332 lock_mount_hash();
1333 /*
1334 * make sure that if __legitimize_mnt() has not seen us grab
1335 * mount_lock, we'll see their refcount increment here.
1336 */
1337 smp_mb();
1338 mnt_add_count(mnt, -1);
1339 count = mnt_get_count(mnt);
1340 if (count != 0) {
1341 WARN_ON(count < 0);
1342 rcu_read_unlock();
1343 unlock_mount_hash();
1344 return;
1345 }
1346 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1347 rcu_read_unlock();
1348 unlock_mount_hash();
1349 return;
1350 }
1351 mnt->mnt.mnt_flags |= MNT_DOOMED;
1352 rcu_read_unlock();
1353
1354 list_del(&mnt->mnt_instance);
1355
1356 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1357 struct mount *p, *tmp;
1358 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1359 __put_mountpoint(unhash_mnt(p), &list);
1360 hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1361 }
1362 }
1363 unlock_mount_hash();
1364 shrink_dentry_list(&list);
1365
1366 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1367 struct task_struct *task = current;
1368 if (likely(!(task->flags & PF_KTHREAD))) {
1369 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1370 if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1371 return;
1372 }
1373 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1374 schedule_delayed_work(&delayed_mntput_work, 1);
1375 return;
1376 }
1377 cleanup_mnt(mnt);
1378}
1379
1380void mntput(struct vfsmount *mnt)
1381{
1382 if (mnt) {
1383 struct mount *m = real_mount(mnt);
1384 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1385 if (unlikely(m->mnt_expiry_mark))
1386 m->mnt_expiry_mark = 0;
1387 mntput_no_expire(m);
1388 }
1389}
1390EXPORT_SYMBOL(mntput);
1391
1392struct vfsmount *mntget(struct vfsmount *mnt)
1393{
1394 if (mnt)
1395 mnt_add_count(real_mount(mnt), 1);
1396 return mnt;
1397}
1398EXPORT_SYMBOL(mntget);
1399
1400/**
1401 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1402 * @path: path to check
1403 *
1404 * d_mountpoint() can only be used reliably to establish if a dentry is
1405 * not mounted in any namespace and that common case is handled inline.
1406 * d_mountpoint() isn't aware of the possibility there may be multiple
1407 * mounts using a given dentry in a different namespace. This function
1408 * checks if the passed in path is a mountpoint rather than the dentry
1409 * alone.
1410 */
1411bool path_is_mountpoint(const struct path *path)
1412{
1413 unsigned seq;
1414 bool res;
1415
1416 if (!d_mountpoint(path->dentry))
1417 return false;
1418
1419 rcu_read_lock();
1420 do {
1421 seq = read_seqbegin(&mount_lock);
1422 res = __path_is_mountpoint(path);
1423 } while (read_seqretry(&mount_lock, seq));
1424 rcu_read_unlock();
1425
1426 return res;
1427}
1428EXPORT_SYMBOL(path_is_mountpoint);
1429
1430struct vfsmount *mnt_clone_internal(const struct path *path)
1431{
1432 struct mount *p;
1433 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1434 if (IS_ERR(p))
1435 return ERR_CAST(p);
1436 p->mnt.mnt_flags |= MNT_INTERNAL;
1437 return &p->mnt;
1438}
1439
1440#ifdef CONFIG_PROC_FS
1441static struct mount *mnt_list_next(struct mnt_namespace *ns,
1442 struct list_head *p)
1443{
1444 struct mount *mnt, *ret = NULL;
1445
1446 lock_ns_list(ns);
1447 list_for_each_continue(p, &ns->list) {
1448 mnt = list_entry(p, typeof(*mnt), mnt_list);
1449 if (!mnt_is_cursor(mnt)) {
1450 ret = mnt;
1451 break;
1452 }
1453 }
1454 unlock_ns_list(ns);
1455
1456 return ret;
1457}
1458
1459/* iterator; we want it to have access to namespace_sem, thus here... */
1460static void *m_start(struct seq_file *m, loff_t *pos)
1461{
1462 struct proc_mounts *p = m->private;
1463 struct list_head *prev;
1464
1465 down_read(&namespace_sem);
1466 if (!*pos) {
1467 prev = &p->ns->list;
1468 } else {
1469 prev = &p->cursor.mnt_list;
1470
1471 /* Read after we'd reached the end? */
1472 if (list_empty(prev))
1473 return NULL;
1474 }
1475
1476 return mnt_list_next(p->ns, prev);
1477}
1478
1479static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1480{
1481 struct proc_mounts *p = m->private;
1482 struct mount *mnt = v;
1483
1484 ++*pos;
1485 return mnt_list_next(p->ns, &mnt->mnt_list);
1486}
1487
1488static void m_stop(struct seq_file *m, void *v)
1489{
1490 struct proc_mounts *p = m->private;
1491 struct mount *mnt = v;
1492
1493 lock_ns_list(p->ns);
1494 if (mnt)
1495 list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1496 else
1497 list_del_init(&p->cursor.mnt_list);
1498 unlock_ns_list(p->ns);
1499 up_read(&namespace_sem);
1500}
1501
1502static int m_show(struct seq_file *m, void *v)
1503{
1504 struct proc_mounts *p = m->private;
1505 struct mount *r = v;
1506 return p->show(m, &r->mnt);
1507}
1508
1509const struct seq_operations mounts_op = {
1510 .start = m_start,
1511 .next = m_next,
1512 .stop = m_stop,
1513 .show = m_show,
1514};
1515
1516void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1517{
1518 down_read(&namespace_sem);
1519 lock_ns_list(ns);
1520 list_del(&cursor->mnt_list);
1521 unlock_ns_list(ns);
1522 up_read(&namespace_sem);
1523}
1524#endif /* CONFIG_PROC_FS */
1525
1526/**
1527 * may_umount_tree - check if a mount tree is busy
1528 * @m: root of mount tree
1529 *
1530 * This is called to check if a tree of mounts has any
1531 * open files, pwds, chroots or sub mounts that are
1532 * busy.
1533 */
1534int may_umount_tree(struct vfsmount *m)
1535{
1536 struct mount *mnt = real_mount(m);
1537 int actual_refs = 0;
1538 int minimum_refs = 0;
1539 struct mount *p;
1540 BUG_ON(!m);
1541
1542 /* write lock needed for mnt_get_count */
1543 lock_mount_hash();
1544 for (p = mnt; p; p = next_mnt(p, mnt)) {
1545 actual_refs += mnt_get_count(p);
1546 minimum_refs += 2;
1547 }
1548 unlock_mount_hash();
1549
1550 if (actual_refs > minimum_refs)
1551 return 0;
1552
1553 return 1;
1554}
1555
1556EXPORT_SYMBOL(may_umount_tree);
1557
1558/**
1559 * may_umount - check if a mount point is busy
1560 * @mnt: root of mount
1561 *
1562 * This is called to check if a mount point has any
1563 * open files, pwds, chroots or sub mounts. If the
1564 * mount has sub mounts this will return busy
1565 * regardless of whether the sub mounts are busy.
1566 *
1567 * Doesn't take quota and stuff into account. IOW, in some cases it will
1568 * give false negatives. The main reason why it's here is that we need
1569 * a non-destructive way to look for easily umountable filesystems.
1570 */
1571int may_umount(struct vfsmount *mnt)
1572{
1573 int ret = 1;
1574 down_read(&namespace_sem);
1575 lock_mount_hash();
1576 if (propagate_mount_busy(real_mount(mnt), 2))
1577 ret = 0;
1578 unlock_mount_hash();
1579 up_read(&namespace_sem);
1580 return ret;
1581}
1582
1583EXPORT_SYMBOL(may_umount);
1584
1585static void namespace_unlock(void)
1586{
1587 struct hlist_head head;
1588 struct hlist_node *p;
1589 struct mount *m;
1590 LIST_HEAD(list);
1591
1592 hlist_move_list(&unmounted, &head);
1593 list_splice_init(&ex_mountpoints, &list);
1594
1595 up_write(&namespace_sem);
1596
1597 shrink_dentry_list(&list);
1598
1599 if (likely(hlist_empty(&head)))
1600 return;
1601
1602 synchronize_rcu_expedited();
1603
1604 hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1605 hlist_del(&m->mnt_umount);
1606 mntput(&m->mnt);
1607 }
1608}
1609
1610static inline void namespace_lock(void)
1611{
1612 down_write(&namespace_sem);
1613}
1614
1615enum umount_tree_flags {
1616 UMOUNT_SYNC = 1,
1617 UMOUNT_PROPAGATE = 2,
1618 UMOUNT_CONNECTED = 4,
1619};
1620
1621static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1622{
1623 /* Leaving mounts connected is only valid for lazy umounts */
1624 if (how & UMOUNT_SYNC)
1625 return true;
1626
1627 /* A mount without a parent has nothing to be connected to */
1628 if (!mnt_has_parent(mnt))
1629 return true;
1630
1631 /* Because the reference counting rules change when mounts are
1632 * unmounted and connected, umounted mounts may not be
1633 * connected to mounted mounts.
1634 */
1635 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1636 return true;
1637
1638 /* Has it been requested that the mount remain connected? */
1639 if (how & UMOUNT_CONNECTED)
1640 return false;
1641
1642 /* Is the mount locked such that it needs to remain connected? */
1643 if (IS_MNT_LOCKED(mnt))
1644 return false;
1645
1646 /* By default disconnect the mount */
1647 return true;
1648}
1649
1650/*
1651 * mount_lock must be held
1652 * namespace_sem must be held for write
1653 */
1654static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1655{
1656 LIST_HEAD(tmp_list);
1657 struct mount *p;
1658
1659 if (how & UMOUNT_PROPAGATE)
1660 propagate_mount_unlock(mnt);
1661
1662 /* Gather the mounts to umount */
1663 for (p = mnt; p; p = next_mnt(p, mnt)) {
1664 p->mnt.mnt_flags |= MNT_UMOUNT;
1665 list_move(&p->mnt_list, &tmp_list);
1666 }
1667
1668 /* Hide the mounts from mnt_mounts */
1669 list_for_each_entry(p, &tmp_list, mnt_list) {
1670 list_del_init(&p->mnt_child);
1671 }
1672
1673 /* Add propogated mounts to the tmp_list */
1674 if (how & UMOUNT_PROPAGATE)
1675 propagate_umount(&tmp_list);
1676
1677 while (!list_empty(&tmp_list)) {
1678 struct mnt_namespace *ns;
1679 bool disconnect;
1680 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1681 list_del_init(&p->mnt_expire);
1682 list_del_init(&p->mnt_list);
1683 ns = p->mnt_ns;
1684 if (ns) {
1685 ns->mounts--;
1686 __touch_mnt_namespace(ns);
1687 }
1688 p->mnt_ns = NULL;
1689 if (how & UMOUNT_SYNC)
1690 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1691
1692 disconnect = disconnect_mount(p, how);
1693 if (mnt_has_parent(p)) {
1694 mnt_add_count(p->mnt_parent, -1);
1695 if (!disconnect) {
1696 /* Don't forget about p */
1697 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1698 } else {
1699 umount_mnt(p);
1700 }
1701 }
1702 change_mnt_propagation(p, MS_PRIVATE);
1703 if (disconnect)
1704 hlist_add_head(&p->mnt_umount, &unmounted);
1705 }
1706}
1707
1708static void shrink_submounts(struct mount *mnt);
1709
1710static int do_umount_root(struct super_block *sb)
1711{
1712 int ret = 0;
1713
1714 down_write(&sb->s_umount);
1715 if (!sb_rdonly(sb)) {
1716 struct fs_context *fc;
1717
1718 fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1719 SB_RDONLY);
1720 if (IS_ERR(fc)) {
1721 ret = PTR_ERR(fc);
1722 } else {
1723 ret = parse_monolithic_mount_data(fc, NULL);
1724 if (!ret)
1725 ret = reconfigure_super(fc);
1726 put_fs_context(fc);
1727 }
1728 }
1729 up_write(&sb->s_umount);
1730 return ret;
1731}
1732
1733static int do_umount(struct mount *mnt, int flags)
1734{
1735 struct super_block *sb = mnt->mnt.mnt_sb;
1736 int retval;
1737
1738 retval = security_sb_umount(&mnt->mnt, flags);
1739 if (retval)
1740 return retval;
1741
1742 /*
1743 * Allow userspace to request a mountpoint be expired rather than
1744 * unmounting unconditionally. Unmount only happens if:
1745 * (1) the mark is already set (the mark is cleared by mntput())
1746 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1747 */
1748 if (flags & MNT_EXPIRE) {
1749 if (&mnt->mnt == current->fs->root.mnt ||
1750 flags & (MNT_FORCE | MNT_DETACH))
1751 return -EINVAL;
1752
1753 /*
1754 * probably don't strictly need the lock here if we examined
1755 * all race cases, but it's a slowpath.
1756 */
1757 lock_mount_hash();
1758 if (mnt_get_count(mnt) != 2) {
1759 unlock_mount_hash();
1760 return -EBUSY;
1761 }
1762 unlock_mount_hash();
1763
1764 if (!xchg(&mnt->mnt_expiry_mark, 1))
1765 return -EAGAIN;
1766 }
1767
1768 /*
1769 * If we may have to abort operations to get out of this
1770 * mount, and they will themselves hold resources we must
1771 * allow the fs to do things. In the Unix tradition of
1772 * 'Gee thats tricky lets do it in userspace' the umount_begin
1773 * might fail to complete on the first run through as other tasks
1774 * must return, and the like. Thats for the mount program to worry
1775 * about for the moment.
1776 */
1777
1778 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1779 sb->s_op->umount_begin(sb);
1780 }
1781
1782 /*
1783 * No sense to grab the lock for this test, but test itself looks
1784 * somewhat bogus. Suggestions for better replacement?
1785 * Ho-hum... In principle, we might treat that as umount + switch
1786 * to rootfs. GC would eventually take care of the old vfsmount.
1787 * Actually it makes sense, especially if rootfs would contain a
1788 * /reboot - static binary that would close all descriptors and
1789 * call reboot(9). Then init(8) could umount root and exec /reboot.
1790 */
1791 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1792 /*
1793 * Special case for "unmounting" root ...
1794 * we just try to remount it readonly.
1795 */
1796 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1797 return -EPERM;
1798 return do_umount_root(sb);
1799 }
1800
1801 namespace_lock();
1802 lock_mount_hash();
1803
1804 /* Recheck MNT_LOCKED with the locks held */
1805 retval = -EINVAL;
1806 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1807 goto out;
1808
1809 event++;
1810 if (flags & MNT_DETACH) {
1811 if (!list_empty(&mnt->mnt_list))
1812 umount_tree(mnt, UMOUNT_PROPAGATE);
1813 retval = 0;
1814 } else {
1815 shrink_submounts(mnt);
1816 retval = -EBUSY;
1817 if (!propagate_mount_busy(mnt, 2)) {
1818 if (!list_empty(&mnt->mnt_list))
1819 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1820 retval = 0;
1821 }
1822 }
1823out:
1824 unlock_mount_hash();
1825 namespace_unlock();
1826 return retval;
1827}
1828
1829/*
1830 * __detach_mounts - lazily unmount all mounts on the specified dentry
1831 *
1832 * During unlink, rmdir, and d_drop it is possible to loose the path
1833 * to an existing mountpoint, and wind up leaking the mount.
1834 * detach_mounts allows lazily unmounting those mounts instead of
1835 * leaking them.
1836 *
1837 * The caller may hold dentry->d_inode->i_mutex.
1838 */
1839void __detach_mounts(struct dentry *dentry)
1840{
1841 struct mountpoint *mp;
1842 struct mount *mnt;
1843
1844 namespace_lock();
1845 lock_mount_hash();
1846 mp = lookup_mountpoint(dentry);
1847 if (!mp)
1848 goto out_unlock;
1849
1850 event++;
1851 while (!hlist_empty(&mp->m_list)) {
1852 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1853 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1854 umount_mnt(mnt);
1855 hlist_add_head(&mnt->mnt_umount, &unmounted);
1856 }
1857 else umount_tree(mnt, UMOUNT_CONNECTED);
1858 }
1859 put_mountpoint(mp);
1860out_unlock:
1861 unlock_mount_hash();
1862 namespace_unlock();
1863}
1864
1865/*
1866 * Is the caller allowed to modify his namespace?
1867 */
1868bool may_mount(void)
1869{
1870 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1871}
1872
1873static void warn_mandlock(void)
1874{
1875 pr_warn_once("=======================================================\n"
1876 "WARNING: The mand mount option has been deprecated and\n"
1877 " and is ignored by this kernel. Remove the mand\n"
1878 " option from the mount to silence this warning.\n"
1879 "=======================================================\n");
1880}
1881
1882static int can_umount(const struct path *path, int flags)
1883{
1884 struct mount *mnt = real_mount(path->mnt);
1885
1886 if (!may_mount())
1887 return -EPERM;
1888 if (path->dentry != path->mnt->mnt_root)
1889 return -EINVAL;
1890 if (!check_mnt(mnt))
1891 return -EINVAL;
1892 if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1893 return -EINVAL;
1894 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1895 return -EPERM;
1896 return 0;
1897}
1898
1899// caller is responsible for flags being sane
1900int path_umount(struct path *path, int flags)
1901{
1902 struct mount *mnt = real_mount(path->mnt);
1903 int ret;
1904
1905 ret = can_umount(path, flags);
1906 if (!ret)
1907 ret = do_umount(mnt, flags);
1908
1909 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1910 dput(path->dentry);
1911 mntput_no_expire(mnt);
1912 return ret;
1913}
1914
1915static int ksys_umount(char __user *name, int flags)
1916{
1917 int lookup_flags = LOOKUP_MOUNTPOINT;
1918 struct path path;
1919 int ret;
1920
1921 // basic validity checks done first
1922 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1923 return -EINVAL;
1924
1925 if (!(flags & UMOUNT_NOFOLLOW))
1926 lookup_flags |= LOOKUP_FOLLOW;
1927 ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1928 if (ret)
1929 return ret;
1930 return path_umount(&path, flags);
1931}
1932
1933SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1934{
1935 return ksys_umount(name, flags);
1936}
1937
1938#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1939
1940/*
1941 * The 2.0 compatible umount. No flags.
1942 */
1943SYSCALL_DEFINE1(oldumount, char __user *, name)
1944{
1945 return ksys_umount(name, 0);
1946}
1947
1948#endif
1949
1950static bool is_mnt_ns_file(struct dentry *dentry)
1951{
1952 /* Is this a proxy for a mount namespace? */
1953 return dentry->d_op == &ns_dentry_operations &&
1954 dentry->d_fsdata == &mntns_operations;
1955}
1956
1957static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1958{
1959 return container_of(ns, struct mnt_namespace, ns);
1960}
1961
1962struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1963{
1964 return &mnt->ns;
1965}
1966
1967static bool mnt_ns_loop(struct dentry *dentry)
1968{
1969 /* Could bind mounting the mount namespace inode cause a
1970 * mount namespace loop?
1971 */
1972 struct mnt_namespace *mnt_ns;
1973 if (!is_mnt_ns_file(dentry))
1974 return false;
1975
1976 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1977 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1978}
1979
1980struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1981 int flag)
1982{
1983 struct mount *res, *p, *q, *r, *parent;
1984
1985 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1986 return ERR_PTR(-EINVAL);
1987
1988 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1989 return ERR_PTR(-EINVAL);
1990
1991 res = q = clone_mnt(mnt, dentry, flag);
1992 if (IS_ERR(q))
1993 return q;
1994
1995 q->mnt_mountpoint = mnt->mnt_mountpoint;
1996
1997 p = mnt;
1998 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1999 struct mount *s;
2000 if (!is_subdir(r->mnt_mountpoint, dentry))
2001 continue;
2002
2003 for (s = r; s; s = next_mnt(s, r)) {
2004 if (!(flag & CL_COPY_UNBINDABLE) &&
2005 IS_MNT_UNBINDABLE(s)) {
2006 if (s->mnt.mnt_flags & MNT_LOCKED) {
2007 /* Both unbindable and locked. */
2008 q = ERR_PTR(-EPERM);
2009 goto out;
2010 } else {
2011 s = skip_mnt_tree(s);
2012 continue;
2013 }
2014 }
2015 if (!(flag & CL_COPY_MNT_NS_FILE) &&
2016 is_mnt_ns_file(s->mnt.mnt_root)) {
2017 s = skip_mnt_tree(s);
2018 continue;
2019 }
2020 while (p != s->mnt_parent) {
2021 p = p->mnt_parent;
2022 q = q->mnt_parent;
2023 }
2024 p = s;
2025 parent = q;
2026 q = clone_mnt(p, p->mnt.mnt_root, flag);
2027 if (IS_ERR(q))
2028 goto out;
2029 lock_mount_hash();
2030 list_add_tail(&q->mnt_list, &res->mnt_list);
2031 attach_mnt(q, parent, p->mnt_mp);
2032 unlock_mount_hash();
2033 }
2034 }
2035 return res;
2036out:
2037 if (res) {
2038 lock_mount_hash();
2039 umount_tree(res, UMOUNT_SYNC);
2040 unlock_mount_hash();
2041 }
2042 return q;
2043}
2044
2045/* Caller should check returned pointer for errors */
2046
2047struct vfsmount *collect_mounts(const struct path *path)
2048{
2049 struct mount *tree;
2050 namespace_lock();
2051 if (!check_mnt(real_mount(path->mnt)))
2052 tree = ERR_PTR(-EINVAL);
2053 else
2054 tree = copy_tree(real_mount(path->mnt), path->dentry,
2055 CL_COPY_ALL | CL_PRIVATE);
2056 namespace_unlock();
2057 if (IS_ERR(tree))
2058 return ERR_CAST(tree);
2059 return &tree->mnt;
2060}
2061
2062static void free_mnt_ns(struct mnt_namespace *);
2063static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
2064
2065void dissolve_on_fput(struct vfsmount *mnt)
2066{
2067 struct mnt_namespace *ns;
2068 namespace_lock();
2069 lock_mount_hash();
2070 ns = real_mount(mnt)->mnt_ns;
2071 if (ns) {
2072 if (is_anon_ns(ns))
2073 umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
2074 else
2075 ns = NULL;
2076 }
2077 unlock_mount_hash();
2078 namespace_unlock();
2079 if (ns)
2080 free_mnt_ns(ns);
2081}
2082
2083void drop_collected_mounts(struct vfsmount *mnt)
2084{
2085 namespace_lock();
2086 lock_mount_hash();
2087 umount_tree(real_mount(mnt), 0);
2088 unlock_mount_hash();
2089 namespace_unlock();
2090}
2091
2092static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2093{
2094 struct mount *child;
2095
2096 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2097 if (!is_subdir(child->mnt_mountpoint, dentry))
2098 continue;
2099
2100 if (child->mnt.mnt_flags & MNT_LOCKED)
2101 return true;
2102 }
2103 return false;
2104}
2105
2106/**
2107 * clone_private_mount - create a private clone of a path
2108 * @path: path to clone
2109 *
2110 * This creates a new vfsmount, which will be the clone of @path. The new mount
2111 * will not be attached anywhere in the namespace and will be private (i.e.
2112 * changes to the originating mount won't be propagated into this).
2113 *
2114 * Release with mntput().
2115 */
2116struct vfsmount *clone_private_mount(const struct path *path)
2117{
2118 struct mount *old_mnt = real_mount(path->mnt);
2119 struct mount *new_mnt;
2120
2121 down_read(&namespace_sem);
2122 if (IS_MNT_UNBINDABLE(old_mnt))
2123 goto invalid;
2124
2125 if (!check_mnt(old_mnt))
2126 goto invalid;
2127
2128 if (has_locked_children(old_mnt, path->dentry))
2129 goto invalid;
2130
2131 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2132 up_read(&namespace_sem);
2133
2134 if (IS_ERR(new_mnt))
2135 return ERR_CAST(new_mnt);
2136
2137 /* Longterm mount to be removed by kern_unmount*() */
2138 new_mnt->mnt_ns = MNT_NS_INTERNAL;
2139
2140 return &new_mnt->mnt;
2141
2142invalid:
2143 up_read(&namespace_sem);
2144 return ERR_PTR(-EINVAL);
2145}
2146EXPORT_SYMBOL_GPL(clone_private_mount);
2147
2148int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2149 struct vfsmount *root)
2150{
2151 struct mount *mnt;
2152 int res = f(root, arg);
2153 if (res)
2154 return res;
2155 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2156 res = f(&mnt->mnt, arg);
2157 if (res)
2158 return res;
2159 }
2160 return 0;
2161}
2162
2163static void lock_mnt_tree(struct mount *mnt)
2164{
2165 struct mount *p;
2166
2167 for (p = mnt; p; p = next_mnt(p, mnt)) {
2168 int flags = p->mnt.mnt_flags;
2169 /* Don't allow unprivileged users to change mount flags */
2170 flags |= MNT_LOCK_ATIME;
2171
2172 if (flags & MNT_READONLY)
2173 flags |= MNT_LOCK_READONLY;
2174
2175 if (flags & MNT_NODEV)
2176 flags |= MNT_LOCK_NODEV;
2177
2178 if (flags & MNT_NOSUID)
2179 flags |= MNT_LOCK_NOSUID;
2180
2181 if (flags & MNT_NOEXEC)
2182 flags |= MNT_LOCK_NOEXEC;
2183 /* Don't allow unprivileged users to reveal what is under a mount */
2184 if (list_empty(&p->mnt_expire))
2185 flags |= MNT_LOCKED;
2186 p->mnt.mnt_flags = flags;
2187 }
2188}
2189
2190static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2191{
2192 struct mount *p;
2193
2194 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2195 if (p->mnt_group_id && !IS_MNT_SHARED(p))
2196 mnt_release_group_id(p);
2197 }
2198}
2199
2200static int invent_group_ids(struct mount *mnt, bool recurse)
2201{
2202 struct mount *p;
2203
2204 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2205 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2206 int err = mnt_alloc_group_id(p);
2207 if (err) {
2208 cleanup_group_ids(mnt, p);
2209 return err;
2210 }
2211 }
2212 }
2213
2214 return 0;
2215}
2216
2217int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2218{
2219 unsigned int max = READ_ONCE(sysctl_mount_max);
2220 unsigned int mounts = 0;
2221 struct mount *p;
2222
2223 if (ns->mounts >= max)
2224 return -ENOSPC;
2225 max -= ns->mounts;
2226 if (ns->pending_mounts >= max)
2227 return -ENOSPC;
2228 max -= ns->pending_mounts;
2229
2230 for (p = mnt; p; p = next_mnt(p, mnt))
2231 mounts++;
2232
2233 if (mounts > max)
2234 return -ENOSPC;
2235
2236 ns->pending_mounts += mounts;
2237 return 0;
2238}
2239
2240/*
2241 * @source_mnt : mount tree to be attached
2242 * @nd : place the mount tree @source_mnt is attached
2243 * @parent_nd : if non-null, detach the source_mnt from its parent and
2244 * store the parent mount and mountpoint dentry.
2245 * (done when source_mnt is moved)
2246 *
2247 * NOTE: in the table below explains the semantics when a source mount
2248 * of a given type is attached to a destination mount of a given type.
2249 * ---------------------------------------------------------------------------
2250 * | BIND MOUNT OPERATION |
2251 * |**************************************************************************
2252 * | source-->| shared | private | slave | unbindable |
2253 * | dest | | | | |
2254 * | | | | | | |
2255 * | v | | | | |
2256 * |**************************************************************************
2257 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
2258 * | | | | | |
2259 * |non-shared| shared (+) | private | slave (*) | invalid |
2260 * ***************************************************************************
2261 * A bind operation clones the source mount and mounts the clone on the
2262 * destination mount.
2263 *
2264 * (++) the cloned mount is propagated to all the mounts in the propagation
2265 * tree of the destination mount and the cloned mount is added to
2266 * the peer group of the source mount.
2267 * (+) the cloned mount is created under the destination mount and is marked
2268 * as shared. The cloned mount is added to the peer group of the source
2269 * mount.
2270 * (+++) the mount is propagated to all the mounts in the propagation tree
2271 * of the destination mount and the cloned mount is made slave
2272 * of the same master as that of the source mount. The cloned mount
2273 * is marked as 'shared and slave'.
2274 * (*) the cloned mount is made a slave of the same master as that of the
2275 * source mount.
2276 *
2277 * ---------------------------------------------------------------------------
2278 * | MOVE MOUNT OPERATION |
2279 * |**************************************************************************
2280 * | source-->| shared | private | slave | unbindable |
2281 * | dest | | | | |
2282 * | | | | | | |
2283 * | v | | | | |
2284 * |**************************************************************************
2285 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
2286 * | | | | | |
2287 * |non-shared| shared (+*) | private | slave (*) | unbindable |
2288 * ***************************************************************************
2289 *
2290 * (+) the mount is moved to the destination. And is then propagated to
2291 * all the mounts in the propagation tree of the destination mount.
2292 * (+*) the mount is moved to the destination.
2293 * (+++) the mount is moved to the destination and is then propagated to
2294 * all the mounts belonging to the destination mount's propagation tree.
2295 * the mount is marked as 'shared and slave'.
2296 * (*) the mount continues to be a slave at the new location.
2297 *
2298 * if the source mount is a tree, the operations explained above is
2299 * applied to each mount in the tree.
2300 * Must be called without spinlocks held, since this function can sleep
2301 * in allocations.
2302 */
2303static int attach_recursive_mnt(struct mount *source_mnt,
2304 struct mount *dest_mnt,
2305 struct mountpoint *dest_mp,
2306 bool moving)
2307{
2308 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2309 HLIST_HEAD(tree_list);
2310 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2311 struct mountpoint *smp;
2312 struct mount *child, *p;
2313 struct hlist_node *n;
2314 int err;
2315
2316 /* Preallocate a mountpoint in case the new mounts need
2317 * to be tucked under other mounts.
2318 */
2319 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2320 if (IS_ERR(smp))
2321 return PTR_ERR(smp);
2322
2323 /* Is there space to add these mounts to the mount namespace? */
2324 if (!moving) {
2325 err = count_mounts(ns, source_mnt);
2326 if (err)
2327 goto out;
2328 }
2329
2330 if (IS_MNT_SHARED(dest_mnt)) {
2331 err = invent_group_ids(source_mnt, true);
2332 if (err)
2333 goto out;
2334 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2335 lock_mount_hash();
2336 if (err)
2337 goto out_cleanup_ids;
2338 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2339 set_mnt_shared(p);
2340 } else {
2341 lock_mount_hash();
2342 }
2343 if (moving) {
2344 unhash_mnt(source_mnt);
2345 attach_mnt(source_mnt, dest_mnt, dest_mp);
2346 touch_mnt_namespace(source_mnt->mnt_ns);
2347 } else {
2348 if (source_mnt->mnt_ns) {
2349 /* move from anon - the caller will destroy */
2350 list_del_init(&source_mnt->mnt_ns->list);
2351 }
2352 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2353 commit_tree(source_mnt);
2354 }
2355
2356 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2357 struct mount *q;
2358 hlist_del_init(&child->mnt_hash);
2359 q = __lookup_mnt(&child->mnt_parent->mnt,
2360 child->mnt_mountpoint);
2361 if (q)
2362 mnt_change_mountpoint(child, smp, q);
2363 /* Notice when we are propagating across user namespaces */
2364 if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2365 lock_mnt_tree(child);
2366 child->mnt.mnt_flags &= ~MNT_LOCKED;
2367 commit_tree(child);
2368 }
2369 put_mountpoint(smp);
2370 unlock_mount_hash();
2371
2372 return 0;
2373
2374 out_cleanup_ids:
2375 while (!hlist_empty(&tree_list)) {
2376 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2377 child->mnt_parent->mnt_ns->pending_mounts = 0;
2378 umount_tree(child, UMOUNT_SYNC);
2379 }
2380 unlock_mount_hash();
2381 cleanup_group_ids(source_mnt, NULL);
2382 out:
2383 ns->pending_mounts = 0;
2384
2385 read_seqlock_excl(&mount_lock);
2386 put_mountpoint(smp);
2387 read_sequnlock_excl(&mount_lock);
2388
2389 return err;
2390}
2391
2392static struct mountpoint *lock_mount(struct path *path)
2393{
2394 struct vfsmount *mnt;
2395 struct dentry *dentry = path->dentry;
2396retry:
2397 inode_lock(dentry->d_inode);
2398 if (unlikely(cant_mount(dentry))) {
2399 inode_unlock(dentry->d_inode);
2400 return ERR_PTR(-ENOENT);
2401 }
2402 namespace_lock();
2403 mnt = lookup_mnt(path);
2404 if (likely(!mnt)) {
2405 struct mountpoint *mp = get_mountpoint(dentry);
2406 if (IS_ERR(mp)) {
2407 namespace_unlock();
2408 inode_unlock(dentry->d_inode);
2409 return mp;
2410 }
2411 return mp;
2412 }
2413 namespace_unlock();
2414 inode_unlock(path->dentry->d_inode);
2415 path_put(path);
2416 path->mnt = mnt;
2417 dentry = path->dentry = dget(mnt->mnt_root);
2418 goto retry;
2419}
2420
2421static void unlock_mount(struct mountpoint *where)
2422{
2423 struct dentry *dentry = where->m_dentry;
2424
2425 read_seqlock_excl(&mount_lock);
2426 put_mountpoint(where);
2427 read_sequnlock_excl(&mount_lock);
2428
2429 namespace_unlock();
2430 inode_unlock(dentry->d_inode);
2431}
2432
2433static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2434{
2435 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2436 return -EINVAL;
2437
2438 if (d_is_dir(mp->m_dentry) !=
2439 d_is_dir(mnt->mnt.mnt_root))
2440 return -ENOTDIR;
2441
2442 return attach_recursive_mnt(mnt, p, mp, false);
2443}
2444
2445/*
2446 * Sanity check the flags to change_mnt_propagation.
2447 */
2448
2449static int flags_to_propagation_type(int ms_flags)
2450{
2451 int type = ms_flags & ~(MS_REC | MS_SILENT);
2452
2453 /* Fail if any non-propagation flags are set */
2454 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2455 return 0;
2456 /* Only one propagation flag should be set */
2457 if (!is_power_of_2(type))
2458 return 0;
2459 return type;
2460}
2461
2462/*
2463 * recursively change the type of the mountpoint.
2464 */
2465static int do_change_type(struct path *path, int ms_flags)
2466{
2467 struct mount *m;
2468 struct mount *mnt = real_mount(path->mnt);
2469 int recurse = ms_flags & MS_REC;
2470 int type;
2471 int err = 0;
2472
2473 if (path->dentry != path->mnt->mnt_root)
2474 return -EINVAL;
2475
2476 type = flags_to_propagation_type(ms_flags);
2477 if (!type)
2478 return -EINVAL;
2479
2480 namespace_lock();
2481 if (type == MS_SHARED) {
2482 err = invent_group_ids(mnt, recurse);
2483 if (err)
2484 goto out_unlock;
2485 }
2486
2487 lock_mount_hash();
2488 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2489 change_mnt_propagation(m, type);
2490 unlock_mount_hash();
2491
2492 out_unlock:
2493 namespace_unlock();
2494 return err;
2495}
2496
2497static struct mount *__do_loopback(struct path *old_path, int recurse)
2498{
2499 struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2500
2501 if (IS_MNT_UNBINDABLE(old))
2502 return mnt;
2503
2504 if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2505 return mnt;
2506
2507 if (!recurse && has_locked_children(old, old_path->dentry))
2508 return mnt;
2509
2510 if (recurse)
2511 mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2512 else
2513 mnt = clone_mnt(old, old_path->dentry, 0);
2514
2515 if (!IS_ERR(mnt))
2516 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2517
2518 return mnt;
2519}
2520
2521/*
2522 * do loopback mount.
2523 */
2524static int do_loopback(struct path *path, const char *old_name,
2525 int recurse)
2526{
2527 struct path old_path;
2528 struct mount *mnt = NULL, *parent;
2529 struct mountpoint *mp;
2530 int err;
2531 if (!old_name || !*old_name)
2532 return -EINVAL;
2533 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2534 if (err)
2535 return err;
2536
2537 err = -EINVAL;
2538 if (mnt_ns_loop(old_path.dentry))
2539 goto out;
2540
2541 mp = lock_mount(path);
2542 if (IS_ERR(mp)) {
2543 err = PTR_ERR(mp);
2544 goto out;
2545 }
2546
2547 parent = real_mount(path->mnt);
2548 if (!check_mnt(parent))
2549 goto out2;
2550
2551 mnt = __do_loopback(&old_path, recurse);
2552 if (IS_ERR(mnt)) {
2553 err = PTR_ERR(mnt);
2554 goto out2;
2555 }
2556
2557 err = graft_tree(mnt, parent, mp);
2558 if (err) {
2559 lock_mount_hash();
2560 umount_tree(mnt, UMOUNT_SYNC);
2561 unlock_mount_hash();
2562 }
2563out2:
2564 unlock_mount(mp);
2565out:
2566 path_put(&old_path);
2567 return err;
2568}
2569
2570static struct file *open_detached_copy(struct path *path, bool recursive)
2571{
2572 struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2573 struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2574 struct mount *mnt, *p;
2575 struct file *file;
2576
2577 if (IS_ERR(ns))
2578 return ERR_CAST(ns);
2579
2580 namespace_lock();
2581 mnt = __do_loopback(path, recursive);
2582 if (IS_ERR(mnt)) {
2583 namespace_unlock();
2584 free_mnt_ns(ns);
2585 return ERR_CAST(mnt);
2586 }
2587
2588 lock_mount_hash();
2589 for (p = mnt; p; p = next_mnt(p, mnt)) {
2590 p->mnt_ns = ns;
2591 ns->mounts++;
2592 }
2593 ns->root = mnt;
2594 list_add_tail(&ns->list, &mnt->mnt_list);
2595 mntget(&mnt->mnt);
2596 unlock_mount_hash();
2597 namespace_unlock();
2598
2599 mntput(path->mnt);
2600 path->mnt = &mnt->mnt;
2601 file = dentry_open(path, O_PATH, current_cred());
2602 if (IS_ERR(file))
2603 dissolve_on_fput(path->mnt);
2604 else
2605 file->f_mode |= FMODE_NEED_UNMOUNT;
2606 return file;
2607}
2608
2609SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2610{
2611 struct file *file;
2612 struct path path;
2613 int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2614 bool detached = flags & OPEN_TREE_CLONE;
2615 int error;
2616 int fd;
2617
2618 BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2619
2620 if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2621 AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2622 OPEN_TREE_CLOEXEC))
2623 return -EINVAL;
2624
2625 if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2626 return -EINVAL;
2627
2628 if (flags & AT_NO_AUTOMOUNT)
2629 lookup_flags &= ~LOOKUP_AUTOMOUNT;
2630 if (flags & AT_SYMLINK_NOFOLLOW)
2631 lookup_flags &= ~LOOKUP_FOLLOW;
2632 if (flags & AT_EMPTY_PATH)
2633 lookup_flags |= LOOKUP_EMPTY;
2634
2635 if (detached && !may_mount())
2636 return -EPERM;
2637
2638 fd = get_unused_fd_flags(flags & O_CLOEXEC);
2639 if (fd < 0)
2640 return fd;
2641
2642 error = user_path_at(dfd, filename, lookup_flags, &path);
2643 if (unlikely(error)) {
2644 file = ERR_PTR(error);
2645 } else {
2646 if (detached)
2647 file = open_detached_copy(&path, flags & AT_RECURSIVE);
2648 else
2649 file = dentry_open(&path, O_PATH, current_cred());
2650 path_put(&path);
2651 }
2652 if (IS_ERR(file)) {
2653 put_unused_fd(fd);
2654 return PTR_ERR(file);
2655 }
2656 fd_install(fd, file);
2657 return fd;
2658}
2659
2660/*
2661 * Don't allow locked mount flags to be cleared.
2662 *
2663 * No locks need to be held here while testing the various MNT_LOCK
2664 * flags because those flags can never be cleared once they are set.
2665 */
2666static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2667{
2668 unsigned int fl = mnt->mnt.mnt_flags;
2669
2670 if ((fl & MNT_LOCK_READONLY) &&
2671 !(mnt_flags & MNT_READONLY))
2672 return false;
2673
2674 if ((fl & MNT_LOCK_NODEV) &&
2675 !(mnt_flags & MNT_NODEV))
2676 return false;
2677
2678 if ((fl & MNT_LOCK_NOSUID) &&
2679 !(mnt_flags & MNT_NOSUID))
2680 return false;
2681
2682 if ((fl & MNT_LOCK_NOEXEC) &&
2683 !(mnt_flags & MNT_NOEXEC))
2684 return false;
2685
2686 if ((fl & MNT_LOCK_ATIME) &&
2687 ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2688 return false;
2689
2690 return true;
2691}
2692
2693static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2694{
2695 bool readonly_request = (mnt_flags & MNT_READONLY);
2696
2697 if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2698 return 0;
2699
2700 if (readonly_request)
2701 return mnt_make_readonly(mnt);
2702
2703 mnt->mnt.mnt_flags &= ~MNT_READONLY;
2704 return 0;
2705}
2706
2707static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2708{
2709 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2710 mnt->mnt.mnt_flags = mnt_flags;
2711 touch_mnt_namespace(mnt->mnt_ns);
2712}
2713
2714static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2715{
2716 struct super_block *sb = mnt->mnt_sb;
2717
2718 if (!__mnt_is_readonly(mnt) &&
2719 (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) &&
2720 (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2721 char *buf = (char *)__get_free_page(GFP_KERNEL);
2722 char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2723 struct tm tm;
2724
2725 time64_to_tm(sb->s_time_max, 0, &tm);
2726
2727 pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2728 sb->s_type->name,
2729 is_mounted(mnt) ? "remounted" : "mounted",
2730 mntpath,
2731 tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2732
2733 free_page((unsigned long)buf);
2734 sb->s_iflags |= SB_I_TS_EXPIRY_WARNED;
2735 }
2736}
2737
2738/*
2739 * Handle reconfiguration of the mountpoint only without alteration of the
2740 * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND
2741 * to mount(2).
2742 */
2743static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2744{
2745 struct super_block *sb = path->mnt->mnt_sb;
2746 struct mount *mnt = real_mount(path->mnt);
2747 int ret;
2748
2749 if (!check_mnt(mnt))
2750 return -EINVAL;
2751
2752 if (path->dentry != mnt->mnt.mnt_root)
2753 return -EINVAL;
2754
2755 if (!can_change_locked_flags(mnt, mnt_flags))
2756 return -EPERM;
2757
2758 /*
2759 * We're only checking whether the superblock is read-only not
2760 * changing it, so only take down_read(&sb->s_umount).
2761 */
2762 down_read(&sb->s_umount);
2763 lock_mount_hash();
2764 ret = change_mount_ro_state(mnt, mnt_flags);
2765 if (ret == 0)
2766 set_mount_attributes(mnt, mnt_flags);
2767 unlock_mount_hash();
2768 up_read(&sb->s_umount);
2769
2770 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2771
2772 return ret;
2773}
2774
2775/*
2776 * change filesystem flags. dir should be a physical root of filesystem.
2777 * If you've mounted a non-root directory somewhere and want to do remount
2778 * on it - tough luck.
2779 */
2780static int do_remount(struct path *path, int ms_flags, int sb_flags,
2781 int mnt_flags, void *data)
2782{
2783 int err;
2784 struct super_block *sb = path->mnt->mnt_sb;
2785 struct mount *mnt = real_mount(path->mnt);
2786 struct fs_context *fc;
2787
2788 if (!check_mnt(mnt))
2789 return -EINVAL;
2790
2791 if (path->dentry != path->mnt->mnt_root)
2792 return -EINVAL;
2793
2794 if (!can_change_locked_flags(mnt, mnt_flags))
2795 return -EPERM;
2796
2797 fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2798 if (IS_ERR(fc))
2799 return PTR_ERR(fc);
2800
2801 fc->oldapi = true;
2802 err = parse_monolithic_mount_data(fc, data);
2803 if (!err) {
2804 down_write(&sb->s_umount);
2805 err = -EPERM;
2806 if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2807 err = reconfigure_super(fc);
2808 if (!err) {
2809 lock_mount_hash();
2810 set_mount_attributes(mnt, mnt_flags);
2811 unlock_mount_hash();
2812 }
2813 }
2814 up_write(&sb->s_umount);
2815 }
2816
2817 mnt_warn_timestamp_expiry(path, &mnt->mnt);
2818
2819 put_fs_context(fc);
2820 return err;
2821}
2822
2823static inline int tree_contains_unbindable(struct mount *mnt)
2824{
2825 struct mount *p;
2826 for (p = mnt; p; p = next_mnt(p, mnt)) {
2827 if (IS_MNT_UNBINDABLE(p))
2828 return 1;
2829 }
2830 return 0;
2831}
2832
2833/*
2834 * Check that there aren't references to earlier/same mount namespaces in the
2835 * specified subtree. Such references can act as pins for mount namespaces
2836 * that aren't checked by the mount-cycle checking code, thereby allowing
2837 * cycles to be made.
2838 */
2839static bool check_for_nsfs_mounts(struct mount *subtree)
2840{
2841 struct mount *p;
2842 bool ret = false;
2843
2844 lock_mount_hash();
2845 for (p = subtree; p; p = next_mnt(p, subtree))
2846 if (mnt_ns_loop(p->mnt.mnt_root))
2847 goto out;
2848
2849 ret = true;
2850out:
2851 unlock_mount_hash();
2852 return ret;
2853}
2854
2855static int do_set_group(struct path *from_path, struct path *to_path)
2856{
2857 struct mount *from, *to;
2858 int err;
2859
2860 from = real_mount(from_path->mnt);
2861 to = real_mount(to_path->mnt);
2862
2863 namespace_lock();
2864
2865 err = -EINVAL;
2866 /* To and From must be mounted */
2867 if (!is_mounted(&from->mnt))
2868 goto out;
2869 if (!is_mounted(&to->mnt))
2870 goto out;
2871
2872 err = -EPERM;
2873 /* We should be allowed to modify mount namespaces of both mounts */
2874 if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2875 goto out;
2876 if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2877 goto out;
2878
2879 err = -EINVAL;
2880 /* To and From paths should be mount roots */
2881 if (from_path->dentry != from_path->mnt->mnt_root)
2882 goto out;
2883 if (to_path->dentry != to_path->mnt->mnt_root)
2884 goto out;
2885
2886 /* Setting sharing groups is only allowed across same superblock */
2887 if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2888 goto out;
2889
2890 /* From mount root should be wider than To mount root */
2891 if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2892 goto out;
2893
2894 /* From mount should not have locked children in place of To's root */
2895 if (has_locked_children(from, to->mnt.mnt_root))
2896 goto out;
2897
2898 /* Setting sharing groups is only allowed on private mounts */
2899 if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2900 goto out;
2901
2902 /* From should not be private */
2903 if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2904 goto out;
2905
2906 if (IS_MNT_SLAVE(from)) {
2907 struct mount *m = from->mnt_master;
2908
2909 list_add(&to->mnt_slave, &m->mnt_slave_list);
2910 to->mnt_master = m;
2911 }
2912
2913 if (IS_MNT_SHARED(from)) {
2914 to->mnt_group_id = from->mnt_group_id;
2915 list_add(&to->mnt_share, &from->mnt_share);
2916 lock_mount_hash();
2917 set_mnt_shared(to);
2918 unlock_mount_hash();
2919 }
2920
2921 err = 0;
2922out:
2923 namespace_unlock();
2924 return err;
2925}
2926
2927static int do_move_mount(struct path *old_path, struct path *new_path)
2928{
2929 struct mnt_namespace *ns;
2930 struct mount *p;
2931 struct mount *old;
2932 struct mount *parent;
2933 struct mountpoint *mp, *old_mp;
2934 int err;
2935 bool attached;
2936
2937 mp = lock_mount(new_path);
2938 if (IS_ERR(mp))
2939 return PTR_ERR(mp);
2940
2941 old = real_mount(old_path->mnt);
2942 p = real_mount(new_path->mnt);
2943 parent = old->mnt_parent;
2944 attached = mnt_has_parent(old);
2945 old_mp = old->mnt_mp;
2946 ns = old->mnt_ns;
2947
2948 err = -EINVAL;
2949 /* The mountpoint must be in our namespace. */
2950 if (!check_mnt(p))
2951 goto out;
2952
2953 /* The thing moved must be mounted... */
2954 if (!is_mounted(&old->mnt))
2955 goto out;
2956
2957 /* ... and either ours or the root of anon namespace */
2958 if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2959 goto out;
2960
2961 if (old->mnt.mnt_flags & MNT_LOCKED)
2962 goto out;
2963
2964 if (old_path->dentry != old_path->mnt->mnt_root)
2965 goto out;
2966
2967 if (d_is_dir(new_path->dentry) !=
2968 d_is_dir(old_path->dentry))
2969 goto out;
2970 /*
2971 * Don't move a mount residing in a shared parent.
2972 */
2973 if (attached && IS_MNT_SHARED(parent))
2974 goto out;
2975 /*
2976 * Don't move a mount tree containing unbindable mounts to a destination
2977 * mount which is shared.
2978 */
2979 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2980 goto out;
2981 err = -ELOOP;
2982 if (!check_for_nsfs_mounts(old))
2983 goto out;
2984 for (; mnt_has_parent(p); p = p->mnt_parent)
2985 if (p == old)
2986 goto out;
2987
2988 err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2989 attached);
2990 if (err)
2991 goto out;
2992
2993 /* if the mount is moved, it should no longer be expire
2994 * automatically */
2995 list_del_init(&old->mnt_expire);
2996 if (attached)
2997 put_mountpoint(old_mp);
2998out:
2999 unlock_mount(mp);
3000 if (!err) {
3001 if (attached)
3002 mntput_no_expire(parent);
3003 else
3004 free_mnt_ns(ns);
3005 }
3006 return err;
3007}
3008
3009static int do_move_mount_old(struct path *path, const char *old_name)
3010{
3011 struct path old_path;
3012 int err;
3013
3014 if (!old_name || !*old_name)
3015 return -EINVAL;
3016
3017 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
3018 if (err)
3019 return err;
3020
3021 err = do_move_mount(&old_path, path);
3022 path_put(&old_path);
3023 return err;
3024}
3025
3026/*
3027 * add a mount into a namespace's mount tree
3028 */
3029static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
3030 const struct path *path, int mnt_flags)
3031{
3032 struct mount *parent = real_mount(path->mnt);
3033
3034 mnt_flags &= ~MNT_INTERNAL_FLAGS;
3035
3036 if (unlikely(!check_mnt(parent))) {
3037 /* that's acceptable only for automounts done in private ns */
3038 if (!(mnt_flags & MNT_SHRINKABLE))
3039 return -EINVAL;
3040 /* ... and for those we'd better have mountpoint still alive */
3041 if (!parent->mnt_ns)
3042 return -EINVAL;
3043 }
3044
3045 /* Refuse the same filesystem on the same mount point */
3046 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
3047 path->mnt->mnt_root == path->dentry)
3048 return -EBUSY;
3049
3050 if (d_is_symlink(newmnt->mnt.mnt_root))
3051 return -EINVAL;
3052
3053 newmnt->mnt.mnt_flags = mnt_flags;
3054 return graft_tree(newmnt, parent, mp);
3055}
3056
3057static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
3058
3059/*
3060 * Create a new mount using a superblock configuration and request it
3061 * be added to the namespace tree.
3062 */
3063static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
3064 unsigned int mnt_flags)
3065{
3066 struct vfsmount *mnt;
3067 struct mountpoint *mp;
3068 struct super_block *sb = fc->root->d_sb;
3069 int error;
3070
3071 error = security_sb_kern_mount(sb);
3072 if (!error && mount_too_revealing(sb, &mnt_flags))
3073 error = -EPERM;
3074
3075 if (unlikely(error)) {
3076 fc_drop_locked(fc);
3077 return error;
3078 }
3079
3080 up_write(&sb->s_umount);
3081
3082 mnt = vfs_create_mount(fc);
3083 if (IS_ERR(mnt))
3084 return PTR_ERR(mnt);
3085
3086 mnt_warn_timestamp_expiry(mountpoint, mnt);
3087
3088 mp = lock_mount(mountpoint);
3089 if (IS_ERR(mp)) {
3090 mntput(mnt);
3091 return PTR_ERR(mp);
3092 }
3093 error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
3094 unlock_mount(mp);
3095 if (error < 0)
3096 mntput(mnt);
3097 return error;
3098}
3099
3100/*
3101 * create a new mount for userspace and request it to be added into the
3102 * namespace's tree
3103 */
3104static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
3105 int mnt_flags, const char *name, void *data)
3106{
3107 struct file_system_type *type;
3108 struct fs_context *fc;
3109 const char *subtype = NULL;
3110 int err = 0;
3111
3112 if (!fstype)
3113 return -EINVAL;
3114
3115 type = get_fs_type(fstype);
3116 if (!type)
3117 return -ENODEV;
3118
3119 if (type->fs_flags & FS_HAS_SUBTYPE) {
3120 subtype = strchr(fstype, '.');
3121 if (subtype) {
3122 subtype++;
3123 if (!*subtype) {
3124 put_filesystem(type);
3125 return -EINVAL;
3126 }
3127 }
3128 }
3129
3130 fc = fs_context_for_mount(type, sb_flags);
3131 put_filesystem(type);
3132 if (IS_ERR(fc))
3133 return PTR_ERR(fc);
3134
3135 if (subtype)
3136 err = vfs_parse_fs_string(fc, "subtype",
3137 subtype, strlen(subtype));
3138 if (!err && name)
3139 err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3140 if (!err)
3141 err = parse_monolithic_mount_data(fc, data);
3142 if (!err && !mount_capable(fc))
3143 err = -EPERM;
3144 if (!err)
3145 err = vfs_get_tree(fc);
3146 if (!err)
3147 err = do_new_mount_fc(fc, path, mnt_flags);
3148
3149 put_fs_context(fc);
3150 return err;
3151}
3152
3153int finish_automount(struct vfsmount *m, const struct path *path)
3154{
3155 struct dentry *dentry = path->dentry;
3156 struct mountpoint *mp;
3157 struct mount *mnt;
3158 int err;
3159
3160 if (!m)
3161 return 0;
3162 if (IS_ERR(m))
3163 return PTR_ERR(m);
3164
3165 mnt = real_mount(m);
3166 /* The new mount record should have at least 2 refs to prevent it being
3167 * expired before we get a chance to add it
3168 */
3169 BUG_ON(mnt_get_count(mnt) < 2);
3170
3171 if (m->mnt_sb == path->mnt->mnt_sb &&
3172 m->mnt_root == dentry) {
3173 err = -ELOOP;
3174 goto discard;
3175 }
3176
3177 /*
3178 * we don't want to use lock_mount() - in this case finding something
3179 * that overmounts our mountpoint to be means "quitely drop what we've
3180 * got", not "try to mount it on top".
3181 */
3182 inode_lock(dentry->d_inode);
3183 namespace_lock();
3184 if (unlikely(cant_mount(dentry))) {
3185 err = -ENOENT;
3186 goto discard_locked;
3187 }
3188 rcu_read_lock();
3189 if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3190 rcu_read_unlock();
3191 err = 0;
3192 goto discard_locked;
3193 }
3194 rcu_read_unlock();
3195 mp = get_mountpoint(dentry);
3196 if (IS_ERR(mp)) {
3197 err = PTR_ERR(mp);
3198 goto discard_locked;
3199 }
3200
3201 err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3202 unlock_mount(mp);
3203 if (unlikely(err))
3204 goto discard;
3205 mntput(m);
3206 return 0;
3207
3208discard_locked:
3209 namespace_unlock();
3210 inode_unlock(dentry->d_inode);
3211discard:
3212 /* remove m from any expiration list it may be on */
3213 if (!list_empty(&mnt->mnt_expire)) {
3214 namespace_lock();
3215 list_del_init(&mnt->mnt_expire);
3216 namespace_unlock();
3217 }
3218 mntput(m);
3219 mntput(m);
3220 return err;
3221}
3222
3223/**
3224 * mnt_set_expiry - Put a mount on an expiration list
3225 * @mnt: The mount to list.
3226 * @expiry_list: The list to add the mount to.
3227 */
3228void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3229{
3230 namespace_lock();
3231
3232 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3233
3234 namespace_unlock();
3235}
3236EXPORT_SYMBOL(mnt_set_expiry);
3237
3238/*
3239 * process a list of expirable mountpoints with the intent of discarding any
3240 * mountpoints that aren't in use and haven't been touched since last we came
3241 * here
3242 */
3243void mark_mounts_for_expiry(struct list_head *mounts)
3244{
3245 struct mount *mnt, *next;
3246 LIST_HEAD(graveyard);
3247
3248 if (list_empty(mounts))
3249 return;
3250
3251 namespace_lock();
3252 lock_mount_hash();
3253
3254 /* extract from the expiration list every vfsmount that matches the
3255 * following criteria:
3256 * - only referenced by its parent vfsmount
3257 * - still marked for expiry (marked on the last call here; marks are
3258 * cleared by mntput())
3259 */
3260 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3261 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3262 propagate_mount_busy(mnt, 1))
3263 continue;
3264 list_move(&mnt->mnt_expire, &graveyard);
3265 }
3266 while (!list_empty(&graveyard)) {
3267 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3268 touch_mnt_namespace(mnt->mnt_ns);
3269 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3270 }
3271 unlock_mount_hash();
3272 namespace_unlock();
3273}
3274
3275EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3276
3277/*
3278 * Ripoff of 'select_parent()'
3279 *
3280 * search the list of submounts for a given mountpoint, and move any
3281 * shrinkable submounts to the 'graveyard' list.
3282 */
3283static int select_submounts(struct mount *parent, struct list_head *graveyard)
3284{
3285 struct mount *this_parent = parent;
3286 struct list_head *next;
3287 int found = 0;
3288
3289repeat:
3290 next = this_parent->mnt_mounts.next;
3291resume:
3292 while (next != &this_parent->mnt_mounts) {
3293 struct list_head *tmp = next;
3294 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3295
3296 next = tmp->next;
3297 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3298 continue;
3299 /*
3300 * Descend a level if the d_mounts list is non-empty.
3301 */
3302 if (!list_empty(&mnt->mnt_mounts)) {
3303 this_parent = mnt;
3304 goto repeat;
3305 }
3306
3307 if (!propagate_mount_busy(mnt, 1)) {
3308 list_move_tail(&mnt->mnt_expire, graveyard);
3309 found++;
3310 }
3311 }
3312 /*
3313 * All done at this level ... ascend and resume the search
3314 */
3315 if (this_parent != parent) {
3316 next = this_parent->mnt_child.next;
3317 this_parent = this_parent->mnt_parent;
3318 goto resume;
3319 }
3320 return found;
3321}
3322
3323/*
3324 * process a list of expirable mountpoints with the intent of discarding any
3325 * submounts of a specific parent mountpoint
3326 *
3327 * mount_lock must be held for write
3328 */
3329static void shrink_submounts(struct mount *mnt)
3330{
3331 LIST_HEAD(graveyard);
3332 struct mount *m;
3333
3334 /* extract submounts of 'mountpoint' from the expiration list */
3335 while (select_submounts(mnt, &graveyard)) {
3336 while (!list_empty(&graveyard)) {
3337 m = list_first_entry(&graveyard, struct mount,
3338 mnt_expire);
3339 touch_mnt_namespace(m->mnt_ns);
3340 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3341 }
3342 }
3343}
3344
3345static void *copy_mount_options(const void __user * data)
3346{
3347 char *copy;
3348 unsigned left, offset;
3349
3350 if (!data)
3351 return NULL;
3352
3353 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3354 if (!copy)
3355 return ERR_PTR(-ENOMEM);
3356
3357 left = copy_from_user(copy, data, PAGE_SIZE);
3358
3359 /*
3360 * Not all architectures have an exact copy_from_user(). Resort to
3361 * byte at a time.
3362 */
3363 offset = PAGE_SIZE - left;
3364 while (left) {
3365 char c;
3366 if (get_user(c, (const char __user *)data + offset))
3367 break;
3368 copy[offset] = c;
3369 left--;
3370 offset++;
3371 }
3372
3373 if (left == PAGE_SIZE) {
3374 kfree(copy);
3375 return ERR_PTR(-EFAULT);
3376 }
3377
3378 return copy;
3379}
3380
3381static char *copy_mount_string(const void __user *data)
3382{
3383 return data ? strndup_user(data, PATH_MAX) : NULL;
3384}
3385
3386/*
3387 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3388 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3389 *
3390 * data is a (void *) that can point to any structure up to
3391 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3392 * information (or be NULL).
3393 *
3394 * Pre-0.97 versions of mount() didn't have a flags word.
3395 * When the flags word was introduced its top half was required
3396 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3397 * Therefore, if this magic number is present, it carries no information
3398 * and must be discarded.
3399 */
3400int path_mount(const char *dev_name, struct path *path,
3401 const char *type_page, unsigned long flags, void *data_page)
3402{
3403 unsigned int mnt_flags = 0, sb_flags;
3404 int ret;
3405
3406 /* Discard magic */
3407 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3408 flags &= ~MS_MGC_MSK;
3409
3410 /* Basic sanity checks */
3411 if (data_page)
3412 ((char *)data_page)[PAGE_SIZE - 1] = 0;
3413
3414 if (flags & MS_NOUSER)
3415 return -EINVAL;
3416
3417 ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3418 if (ret)
3419 return ret;
3420 if (!may_mount())
3421 return -EPERM;
3422 if (flags & SB_MANDLOCK)
3423 warn_mandlock();
3424
3425 /* Default to relatime unless overriden */
3426 if (!(flags & MS_NOATIME))
3427 mnt_flags |= MNT_RELATIME;
3428
3429 /* Separate the per-mountpoint flags */
3430 if (flags & MS_NOSUID)
3431 mnt_flags |= MNT_NOSUID;
3432 if (flags & MS_NODEV)
3433 mnt_flags |= MNT_NODEV;
3434 if (flags & MS_NOEXEC)
3435 mnt_flags |= MNT_NOEXEC;
3436 if (flags & MS_NOATIME)
3437 mnt_flags |= MNT_NOATIME;
3438 if (flags & MS_NODIRATIME)
3439 mnt_flags |= MNT_NODIRATIME;
3440 if (flags & MS_STRICTATIME)
3441 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3442 if (flags & MS_RDONLY)
3443 mnt_flags |= MNT_READONLY;
3444 if (flags & MS_NOSYMFOLLOW)
3445 mnt_flags |= MNT_NOSYMFOLLOW;
3446
3447 /* The default atime for remount is preservation */
3448 if ((flags & MS_REMOUNT) &&
3449 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3450 MS_STRICTATIME)) == 0)) {
3451 mnt_flags &= ~MNT_ATIME_MASK;
3452 mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3453 }
3454
3455 sb_flags = flags & (SB_RDONLY |
3456 SB_SYNCHRONOUS |
3457 SB_MANDLOCK |
3458 SB_DIRSYNC |
3459 SB_SILENT |
3460 SB_POSIXACL |
3461 SB_LAZYTIME |
3462 SB_I_VERSION);
3463
3464 if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3465 return do_reconfigure_mnt(path, mnt_flags);
3466 if (flags & MS_REMOUNT)
3467 return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3468 if (flags & MS_BIND)
3469 return do_loopback(path, dev_name, flags & MS_REC);
3470 if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3471 return do_change_type(path, flags);
3472 if (flags & MS_MOVE)
3473 return do_move_mount_old(path, dev_name);
3474
3475 return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3476 data_page);
3477}
3478
3479long do_mount(const char *dev_name, const char __user *dir_name,
3480 const char *type_page, unsigned long flags, void *data_page)
3481{
3482 struct path path;
3483 int ret;
3484
3485 ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3486 if (ret)
3487 return ret;
3488 ret = path_mount(dev_name, &path, type_page, flags, data_page);
3489 path_put(&path);
3490 return ret;
3491}
3492
3493static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3494{
3495 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3496}
3497
3498static void dec_mnt_namespaces(struct ucounts *ucounts)
3499{
3500 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3501}
3502
3503static void free_mnt_ns(struct mnt_namespace *ns)
3504{
3505 if (!is_anon_ns(ns))
3506 ns_free_inum(&ns->ns);
3507 dec_mnt_namespaces(ns->ucounts);
3508 put_user_ns(ns->user_ns);
3509 kfree(ns);
3510}
3511
3512/*
3513 * Assign a sequence number so we can detect when we attempt to bind
3514 * mount a reference to an older mount namespace into the current
3515 * mount namespace, preventing reference counting loops. A 64bit
3516 * number incrementing at 10Ghz will take 12,427 years to wrap which
3517 * is effectively never, so we can ignore the possibility.
3518 */
3519static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3520
3521static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3522{
3523 struct mnt_namespace *new_ns;
3524 struct ucounts *ucounts;
3525 int ret;
3526
3527 ucounts = inc_mnt_namespaces(user_ns);
3528 if (!ucounts)
3529 return ERR_PTR(-ENOSPC);
3530
3531 new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3532 if (!new_ns) {
3533 dec_mnt_namespaces(ucounts);
3534 return ERR_PTR(-ENOMEM);
3535 }
3536 if (!anon) {
3537 ret = ns_alloc_inum(&new_ns->ns);
3538 if (ret) {
3539 kfree(new_ns);
3540 dec_mnt_namespaces(ucounts);
3541 return ERR_PTR(ret);
3542 }
3543 }
3544 new_ns->ns.ops = &mntns_operations;
3545 if (!anon)
3546 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3547 refcount_set(&new_ns->ns.count, 1);
3548 INIT_LIST_HEAD(&new_ns->list);
3549 init_waitqueue_head(&new_ns->poll);
3550 spin_lock_init(&new_ns->ns_lock);
3551 new_ns->user_ns = get_user_ns(user_ns);
3552 new_ns->ucounts = ucounts;
3553 return new_ns;
3554}
3555
3556__latent_entropy
3557struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3558 struct user_namespace *user_ns, struct fs_struct *new_fs)
3559{
3560 struct mnt_namespace *new_ns;
3561 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3562 struct mount *p, *q;
3563 struct mount *old;
3564 struct mount *new;
3565 int copy_flags;
3566
3567 BUG_ON(!ns);
3568
3569 if (likely(!(flags & CLONE_NEWNS))) {
3570 get_mnt_ns(ns);
3571 return ns;
3572 }
3573
3574 old = ns->root;
3575
3576 new_ns = alloc_mnt_ns(user_ns, false);
3577 if (IS_ERR(new_ns))
3578 return new_ns;
3579
3580 namespace_lock();
3581 /* First pass: copy the tree topology */
3582 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3583 if (user_ns != ns->user_ns)
3584 copy_flags |= CL_SHARED_TO_SLAVE;
3585 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3586 if (IS_ERR(new)) {
3587 namespace_unlock();
3588 free_mnt_ns(new_ns);
3589 return ERR_CAST(new);
3590 }
3591 if (user_ns != ns->user_ns) {
3592 lock_mount_hash();
3593 lock_mnt_tree(new);
3594 unlock_mount_hash();
3595 }
3596 new_ns->root = new;
3597 list_add_tail(&new_ns->list, &new->mnt_list);
3598
3599 /*
3600 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3601 * as belonging to new namespace. We have already acquired a private
3602 * fs_struct, so tsk->fs->lock is not needed.
3603 */
3604 p = old;
3605 q = new;
3606 while (p) {
3607 q->mnt_ns = new_ns;
3608 new_ns->mounts++;
3609 if (new_fs) {
3610 if (&p->mnt == new_fs->root.mnt) {
3611 new_fs->root.mnt = mntget(&q->mnt);
3612 rootmnt = &p->mnt;
3613 }
3614 if (&p->mnt == new_fs->pwd.mnt) {
3615 new_fs->pwd.mnt = mntget(&q->mnt);
3616 pwdmnt = &p->mnt;
3617 }
3618 }
3619 p = next_mnt(p, old);
3620 q = next_mnt(q, new);
3621 if (!q)
3622 break;
3623 // an mntns binding we'd skipped?
3624 while (p->mnt.mnt_root != q->mnt.mnt_root)
3625 p = next_mnt(skip_mnt_tree(p), old);
3626 }
3627 namespace_unlock();
3628
3629 if (rootmnt)
3630 mntput(rootmnt);
3631 if (pwdmnt)
3632 mntput(pwdmnt);
3633
3634 return new_ns;
3635}
3636
3637struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3638{
3639 struct mount *mnt = real_mount(m);
3640 struct mnt_namespace *ns;
3641 struct super_block *s;
3642 struct path path;
3643 int err;
3644
3645 ns = alloc_mnt_ns(&init_user_ns, true);
3646 if (IS_ERR(ns)) {
3647 mntput(m);
3648 return ERR_CAST(ns);
3649 }
3650 mnt->mnt_ns = ns;
3651 ns->root = mnt;
3652 ns->mounts++;
3653 list_add(&mnt->mnt_list, &ns->list);
3654
3655 err = vfs_path_lookup(m->mnt_root, m,
3656 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3657
3658 put_mnt_ns(ns);
3659
3660 if (err)
3661 return ERR_PTR(err);
3662
3663 /* trade a vfsmount reference for active sb one */
3664 s = path.mnt->mnt_sb;
3665 atomic_inc(&s->s_active);
3666 mntput(path.mnt);
3667 /* lock the sucker */
3668 down_write(&s->s_umount);
3669 /* ... and return the root of (sub)tree on it */
3670 return path.dentry;
3671}
3672EXPORT_SYMBOL(mount_subtree);
3673
3674SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3675 char __user *, type, unsigned long, flags, void __user *, data)
3676{
3677 int ret;
3678 char *kernel_type;
3679 char *kernel_dev;
3680 void *options;
3681
3682 kernel_type = copy_mount_string(type);
3683 ret = PTR_ERR(kernel_type);
3684 if (IS_ERR(kernel_type))
3685 goto out_type;
3686
3687 kernel_dev = copy_mount_string(dev_name);
3688 ret = PTR_ERR(kernel_dev);
3689 if (IS_ERR(kernel_dev))
3690 goto out_dev;
3691
3692 options = copy_mount_options(data);
3693 ret = PTR_ERR(options);
3694 if (IS_ERR(options))
3695 goto out_data;
3696
3697 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3698
3699 kfree(options);
3700out_data:
3701 kfree(kernel_dev);
3702out_dev:
3703 kfree(kernel_type);
3704out_type:
3705 return ret;
3706}
3707
3708#define FSMOUNT_VALID_FLAGS \
3709 (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3710 MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \
3711 MOUNT_ATTR_NOSYMFOLLOW)
3712
3713#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3714
3715#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3716 (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3717
3718static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3719{
3720 unsigned int mnt_flags = 0;
3721
3722 if (attr_flags & MOUNT_ATTR_RDONLY)
3723 mnt_flags |= MNT_READONLY;
3724 if (attr_flags & MOUNT_ATTR_NOSUID)
3725 mnt_flags |= MNT_NOSUID;
3726 if (attr_flags & MOUNT_ATTR_NODEV)
3727 mnt_flags |= MNT_NODEV;
3728 if (attr_flags & MOUNT_ATTR_NOEXEC)
3729 mnt_flags |= MNT_NOEXEC;
3730 if (attr_flags & MOUNT_ATTR_NODIRATIME)
3731 mnt_flags |= MNT_NODIRATIME;
3732 if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3733 mnt_flags |= MNT_NOSYMFOLLOW;
3734
3735 return mnt_flags;
3736}
3737
3738/*
3739 * Create a kernel mount representation for a new, prepared superblock
3740 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3741 */
3742SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3743 unsigned int, attr_flags)
3744{
3745 struct mnt_namespace *ns;
3746 struct fs_context *fc;
3747 struct file *file;
3748 struct path newmount;
3749 struct mount *mnt;
3750 struct fd f;
3751 unsigned int mnt_flags = 0;
3752 long ret;
3753
3754 if (!may_mount())
3755 return -EPERM;
3756
3757 if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3758 return -EINVAL;
3759
3760 if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3761 return -EINVAL;
3762
3763 mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3764
3765 switch (attr_flags & MOUNT_ATTR__ATIME) {
3766 case MOUNT_ATTR_STRICTATIME:
3767 break;
3768 case MOUNT_ATTR_NOATIME:
3769 mnt_flags |= MNT_NOATIME;
3770 break;
3771 case MOUNT_ATTR_RELATIME:
3772 mnt_flags |= MNT_RELATIME;
3773 break;
3774 default:
3775 return -EINVAL;
3776 }
3777
3778 f = fdget(fs_fd);
3779 if (!f.file)
3780 return -EBADF;
3781
3782 ret = -EINVAL;
3783 if (f.file->f_op != &fscontext_fops)
3784 goto err_fsfd;
3785
3786 fc = f.file->private_data;
3787
3788 ret = mutex_lock_interruptible(&fc->uapi_mutex);
3789 if (ret < 0)
3790 goto err_fsfd;
3791
3792 /* There must be a valid superblock or we can't mount it */
3793 ret = -EINVAL;
3794 if (!fc->root)
3795 goto err_unlock;
3796
3797 ret = -EPERM;
3798 if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3799 pr_warn("VFS: Mount too revealing\n");
3800 goto err_unlock;
3801 }
3802
3803 ret = -EBUSY;
3804 if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3805 goto err_unlock;
3806
3807 if (fc->sb_flags & SB_MANDLOCK)
3808 warn_mandlock();
3809
3810 newmount.mnt = vfs_create_mount(fc);
3811 if (IS_ERR(newmount.mnt)) {
3812 ret = PTR_ERR(newmount.mnt);
3813 goto err_unlock;
3814 }
3815 newmount.dentry = dget(fc->root);
3816 newmount.mnt->mnt_flags = mnt_flags;
3817
3818 /* We've done the mount bit - now move the file context into more or
3819 * less the same state as if we'd done an fspick(). We don't want to
3820 * do any memory allocation or anything like that at this point as we
3821 * don't want to have to handle any errors incurred.
3822 */
3823 vfs_clean_context(fc);
3824
3825 ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3826 if (IS_ERR(ns)) {
3827 ret = PTR_ERR(ns);
3828 goto err_path;
3829 }
3830 mnt = real_mount(newmount.mnt);
3831 mnt->mnt_ns = ns;
3832 ns->root = mnt;
3833 ns->mounts = 1;
3834 list_add(&mnt->mnt_list, &ns->list);
3835 mntget(newmount.mnt);
3836
3837 /* Attach to an apparent O_PATH fd with a note that we need to unmount
3838 * it, not just simply put it.
3839 */
3840 file = dentry_open(&newmount, O_PATH, fc->cred);
3841 if (IS_ERR(file)) {
3842 dissolve_on_fput(newmount.mnt);
3843 ret = PTR_ERR(file);
3844 goto err_path;
3845 }
3846 file->f_mode |= FMODE_NEED_UNMOUNT;
3847
3848 ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3849 if (ret >= 0)
3850 fd_install(ret, file);
3851 else
3852 fput(file);
3853
3854err_path:
3855 path_put(&newmount);
3856err_unlock:
3857 mutex_unlock(&fc->uapi_mutex);
3858err_fsfd:
3859 fdput(f);
3860 return ret;
3861}
3862
3863/*
3864 * Move a mount from one place to another. In combination with
3865 * fsopen()/fsmount() this is used to install a new mount and in combination
3866 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3867 * a mount subtree.
3868 *
3869 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3870 */
3871SYSCALL_DEFINE5(move_mount,
3872 int, from_dfd, const char __user *, from_pathname,
3873 int, to_dfd, const char __user *, to_pathname,
3874 unsigned int, flags)
3875{
3876 struct path from_path, to_path;
3877 unsigned int lflags;
3878 int ret = 0;
3879
3880 if (!may_mount())
3881 return -EPERM;
3882
3883 if (flags & ~MOVE_MOUNT__MASK)
3884 return -EINVAL;
3885
3886 /* If someone gives a pathname, they aren't permitted to move
3887 * from an fd that requires unmount as we can't get at the flag
3888 * to clear it afterwards.
3889 */
3890 lflags = 0;
3891 if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3892 if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3893 if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3894
3895 ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3896 if (ret < 0)
3897 return ret;
3898
3899 lflags = 0;
3900 if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW;
3901 if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT;
3902 if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY;
3903
3904 ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3905 if (ret < 0)
3906 goto out_from;
3907
3908 ret = security_move_mount(&from_path, &to_path);
3909 if (ret < 0)
3910 goto out_to;
3911
3912 if (flags & MOVE_MOUNT_SET_GROUP)
3913 ret = do_set_group(&from_path, &to_path);
3914 else
3915 ret = do_move_mount(&from_path, &to_path);
3916
3917out_to:
3918 path_put(&to_path);
3919out_from:
3920 path_put(&from_path);
3921 return ret;
3922}
3923
3924/*
3925 * Return true if path is reachable from root
3926 *
3927 * namespace_sem or mount_lock is held
3928 */
3929bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3930 const struct path *root)
3931{
3932 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3933 dentry = mnt->mnt_mountpoint;
3934 mnt = mnt->mnt_parent;
3935 }
3936 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3937}
3938
3939bool path_is_under(const struct path *path1, const struct path *path2)
3940{
3941 bool res;
3942 read_seqlock_excl(&mount_lock);
3943 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3944 read_sequnlock_excl(&mount_lock);
3945 return res;
3946}
3947EXPORT_SYMBOL(path_is_under);
3948
3949/*
3950 * pivot_root Semantics:
3951 * Moves the root file system of the current process to the directory put_old,
3952 * makes new_root as the new root file system of the current process, and sets
3953 * root/cwd of all processes which had them on the current root to new_root.
3954 *
3955 * Restrictions:
3956 * The new_root and put_old must be directories, and must not be on the
3957 * same file system as the current process root. The put_old must be
3958 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3959 * pointed to by put_old must yield the same directory as new_root. No other
3960 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3961 *
3962 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3963 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3964 * in this situation.
3965 *
3966 * Notes:
3967 * - we don't move root/cwd if they are not at the root (reason: if something
3968 * cared enough to change them, it's probably wrong to force them elsewhere)
3969 * - it's okay to pick a root that isn't the root of a file system, e.g.
3970 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3971 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3972 * first.
3973 */
3974SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3975 const char __user *, put_old)
3976{
3977 struct path new, old, root;
3978 struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3979 struct mountpoint *old_mp, *root_mp;
3980 int error;
3981
3982 if (!may_mount())
3983 return -EPERM;
3984
3985 error = user_path_at(AT_FDCWD, new_root,
3986 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3987 if (error)
3988 goto out0;
3989
3990 error = user_path_at(AT_FDCWD, put_old,
3991 LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3992 if (error)
3993 goto out1;
3994
3995 error = security_sb_pivotroot(&old, &new);
3996 if (error)
3997 goto out2;
3998
3999 get_fs_root(current->fs, &root);
4000 old_mp = lock_mount(&old);
4001 error = PTR_ERR(old_mp);
4002 if (IS_ERR(old_mp))
4003 goto out3;
4004
4005 error = -EINVAL;
4006 new_mnt = real_mount(new.mnt);
4007 root_mnt = real_mount(root.mnt);
4008 old_mnt = real_mount(old.mnt);
4009 ex_parent = new_mnt->mnt_parent;
4010 root_parent = root_mnt->mnt_parent;
4011 if (IS_MNT_SHARED(old_mnt) ||
4012 IS_MNT_SHARED(ex_parent) ||
4013 IS_MNT_SHARED(root_parent))
4014 goto out4;
4015 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
4016 goto out4;
4017 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
4018 goto out4;
4019 error = -ENOENT;
4020 if (d_unlinked(new.dentry))
4021 goto out4;
4022 error = -EBUSY;
4023 if (new_mnt == root_mnt || old_mnt == root_mnt)
4024 goto out4; /* loop, on the same file system */
4025 error = -EINVAL;
4026 if (root.mnt->mnt_root != root.dentry)
4027 goto out4; /* not a mountpoint */
4028 if (!mnt_has_parent(root_mnt))
4029 goto out4; /* not attached */
4030 if (new.mnt->mnt_root != new.dentry)
4031 goto out4; /* not a mountpoint */
4032 if (!mnt_has_parent(new_mnt))
4033 goto out4; /* not attached */
4034 /* make sure we can reach put_old from new_root */
4035 if (!is_path_reachable(old_mnt, old.dentry, &new))
4036 goto out4;
4037 /* make certain new is below the root */
4038 if (!is_path_reachable(new_mnt, new.dentry, &root))
4039 goto out4;
4040 lock_mount_hash();
4041 umount_mnt(new_mnt);
4042 root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */
4043 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
4044 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
4045 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
4046 }
4047 /* mount old root on put_old */
4048 attach_mnt(root_mnt, old_mnt, old_mp);
4049 /* mount new_root on / */
4050 attach_mnt(new_mnt, root_parent, root_mp);
4051 mnt_add_count(root_parent, -1);
4052 touch_mnt_namespace(current->nsproxy->mnt_ns);
4053 /* A moved mount should not expire automatically */
4054 list_del_init(&new_mnt->mnt_expire);
4055 put_mountpoint(root_mp);
4056 unlock_mount_hash();
4057 chroot_fs_refs(&root, &new);
4058 error = 0;
4059out4:
4060 unlock_mount(old_mp);
4061 if (!error)
4062 mntput_no_expire(ex_parent);
4063out3:
4064 path_put(&root);
4065out2:
4066 path_put(&old);
4067out1:
4068 path_put(&new);
4069out0:
4070 return error;
4071}
4072
4073static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
4074{
4075 unsigned int flags = mnt->mnt.mnt_flags;
4076
4077 /* flags to clear */
4078 flags &= ~kattr->attr_clr;
4079 /* flags to raise */
4080 flags |= kattr->attr_set;
4081
4082 return flags;
4083}
4084
4085static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4086{
4087 struct vfsmount *m = &mnt->mnt;
4088 struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
4089
4090 if (!kattr->mnt_idmap)
4091 return 0;
4092
4093 /*
4094 * Creating an idmapped mount with the filesystem wide idmapping
4095 * doesn't make sense so block that. We don't allow mushy semantics.
4096 */
4097 if (mnt_idmap_owner(kattr->mnt_idmap) == fs_userns)
4098 return -EINVAL;
4099
4100 /*
4101 * Once a mount has been idmapped we don't allow it to change its
4102 * mapping. It makes things simpler and callers can just create
4103 * another bind-mount they can idmap if they want to.
4104 */
4105 if (is_idmapped_mnt(m))
4106 return -EPERM;
4107
4108 /* The underlying filesystem doesn't support idmapped mounts yet. */
4109 if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
4110 return -EINVAL;
4111
4112 /* We're not controlling the superblock. */
4113 if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
4114 return -EPERM;
4115
4116 /* Mount has already been visible in the filesystem hierarchy. */
4117 if (!is_anon_ns(mnt->mnt_ns))
4118 return -EINVAL;
4119
4120 return 0;
4121}
4122
4123/**
4124 * mnt_allow_writers() - check whether the attribute change allows writers
4125 * @kattr: the new mount attributes
4126 * @mnt: the mount to which @kattr will be applied
4127 *
4128 * Check whether thew new mount attributes in @kattr allow concurrent writers.
4129 *
4130 * Return: true if writers need to be held, false if not
4131 */
4132static inline bool mnt_allow_writers(const struct mount_kattr *kattr,
4133 const struct mount *mnt)
4134{
4135 return (!(kattr->attr_set & MNT_READONLY) ||
4136 (mnt->mnt.mnt_flags & MNT_READONLY)) &&
4137 !kattr->mnt_idmap;
4138}
4139
4140static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt)
4141{
4142 struct mount *m;
4143 int err;
4144
4145 for (m = mnt; m; m = next_mnt(m, mnt)) {
4146 if (!can_change_locked_flags(m, recalc_flags(kattr, m))) {
4147 err = -EPERM;
4148 break;
4149 }
4150
4151 err = can_idmap_mount(kattr, m);
4152 if (err)
4153 break;
4154
4155 if (!mnt_allow_writers(kattr, m)) {
4156 err = mnt_hold_writers(m);
4157 if (err)
4158 break;
4159 }
4160
4161 if (!kattr->recurse)
4162 return 0;
4163 }
4164
4165 if (err) {
4166 struct mount *p;
4167
4168 /*
4169 * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will
4170 * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all
4171 * mounts and needs to take care to include the first mount.
4172 */
4173 for (p = mnt; p; p = next_mnt(p, mnt)) {
4174 /* If we had to hold writers unblock them. */
4175 if (p->mnt.mnt_flags & MNT_WRITE_HOLD)
4176 mnt_unhold_writers(p);
4177
4178 /*
4179 * We're done once the first mount we changed got
4180 * MNT_WRITE_HOLD unset.
4181 */
4182 if (p == m)
4183 break;
4184 }
4185 }
4186 return err;
4187}
4188
4189static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4190{
4191 if (!kattr->mnt_idmap)
4192 return;
4193
4194 /*
4195 * Pairs with smp_load_acquire() in mnt_idmap().
4196 *
4197 * Since we only allow a mount to change the idmapping once and
4198 * verified this in can_idmap_mount() we know that the mount has
4199 * @nop_mnt_idmap attached to it. So there's no need to drop any
4200 * references.
4201 */
4202 smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap));
4203}
4204
4205static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt)
4206{
4207 struct mount *m;
4208
4209 for (m = mnt; m; m = next_mnt(m, mnt)) {
4210 unsigned int flags;
4211
4212 do_idmap_mount(kattr, m);
4213 flags = recalc_flags(kattr, m);
4214 WRITE_ONCE(m->mnt.mnt_flags, flags);
4215
4216 /* If we had to hold writers unblock them. */
4217 if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
4218 mnt_unhold_writers(m);
4219
4220 if (kattr->propagation)
4221 change_mnt_propagation(m, kattr->propagation);
4222 if (!kattr->recurse)
4223 break;
4224 }
4225 touch_mnt_namespace(mnt->mnt_ns);
4226}
4227
4228static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4229{
4230 struct mount *mnt = real_mount(path->mnt);
4231 int err = 0;
4232
4233 if (path->dentry != mnt->mnt.mnt_root)
4234 return -EINVAL;
4235
4236 if (kattr->mnt_userns) {
4237 struct mnt_idmap *mnt_idmap;
4238
4239 mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns);
4240 if (IS_ERR(mnt_idmap))
4241 return PTR_ERR(mnt_idmap);
4242 kattr->mnt_idmap = mnt_idmap;
4243 }
4244
4245 if (kattr->propagation) {
4246 /*
4247 * Only take namespace_lock() if we're actually changing
4248 * propagation.
4249 */
4250 namespace_lock();
4251 if (kattr->propagation == MS_SHARED) {
4252 err = invent_group_ids(mnt, kattr->recurse);
4253 if (err) {
4254 namespace_unlock();
4255 return err;
4256 }
4257 }
4258 }
4259
4260 err = -EINVAL;
4261 lock_mount_hash();
4262
4263 /* Ensure that this isn't anything purely vfs internal. */
4264 if (!is_mounted(&mnt->mnt))
4265 goto out;
4266
4267 /*
4268 * If this is an attached mount make sure it's located in the callers
4269 * mount namespace. If it's not don't let the caller interact with it.
4270 * If this is a detached mount make sure it has an anonymous mount
4271 * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE.
4272 */
4273 if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns)))
4274 goto out;
4275
4276 /*
4277 * First, we get the mount tree in a shape where we can change mount
4278 * properties without failure. If we succeeded to do so we commit all
4279 * changes and if we failed we clean up.
4280 */
4281 err = mount_setattr_prepare(kattr, mnt);
4282 if (!err)
4283 mount_setattr_commit(kattr, mnt);
4284
4285out:
4286 unlock_mount_hash();
4287
4288 if (kattr->propagation) {
4289 namespace_unlock();
4290 if (err)
4291 cleanup_group_ids(mnt, NULL);
4292 }
4293
4294 return err;
4295}
4296
4297static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4298 struct mount_kattr *kattr, unsigned int flags)
4299{
4300 int err = 0;
4301 struct ns_common *ns;
4302 struct user_namespace *mnt_userns;
4303 struct file *file;
4304
4305 if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4306 return 0;
4307
4308 /*
4309 * We currently do not support clearing an idmapped mount. If this ever
4310 * is a use-case we can revisit this but for now let's keep it simple
4311 * and not allow it.
4312 */
4313 if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4314 return -EINVAL;
4315
4316 if (attr->userns_fd > INT_MAX)
4317 return -EINVAL;
4318
4319 file = fget(attr->userns_fd);
4320 if (!file)
4321 return -EBADF;
4322
4323 if (!proc_ns_file(file)) {
4324 err = -EINVAL;
4325 goto out_fput;
4326 }
4327
4328 ns = get_proc_ns(file_inode(file));
4329 if (ns->ops->type != CLONE_NEWUSER) {
4330 err = -EINVAL;
4331 goto out_fput;
4332 }
4333
4334 /*
4335 * The initial idmapping cannot be used to create an idmapped
4336 * mount. We use the initial idmapping as an indicator of a mount
4337 * that is not idmapped. It can simply be passed into helpers that
4338 * are aware of idmapped mounts as a convenient shortcut. A user
4339 * can just create a dedicated identity mapping to achieve the same
4340 * result.
4341 */
4342 mnt_userns = container_of(ns, struct user_namespace, ns);
4343 if (initial_idmapping(mnt_userns)) {
4344 err = -EPERM;
4345 goto out_fput;
4346 }
4347
4348 /* We're not controlling the target namespace. */
4349 if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) {
4350 err = -EPERM;
4351 goto out_fput;
4352 }
4353
4354 kattr->mnt_userns = get_user_ns(mnt_userns);
4355
4356out_fput:
4357 fput(file);
4358 return err;
4359}
4360
4361static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4362 struct mount_kattr *kattr, unsigned int flags)
4363{
4364 unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4365
4366 if (flags & AT_NO_AUTOMOUNT)
4367 lookup_flags &= ~LOOKUP_AUTOMOUNT;
4368 if (flags & AT_SYMLINK_NOFOLLOW)
4369 lookup_flags &= ~LOOKUP_FOLLOW;
4370 if (flags & AT_EMPTY_PATH)
4371 lookup_flags |= LOOKUP_EMPTY;
4372
4373 *kattr = (struct mount_kattr) {
4374 .lookup_flags = lookup_flags,
4375 .recurse = !!(flags & AT_RECURSIVE),
4376 };
4377
4378 if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4379 return -EINVAL;
4380 if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4381 return -EINVAL;
4382 kattr->propagation = attr->propagation;
4383
4384 if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4385 return -EINVAL;
4386
4387 kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4388 kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4389
4390 /*
4391 * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4392 * users wanting to transition to a different atime setting cannot
4393 * simply specify the atime setting in @attr_set, but must also
4394 * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4395 * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4396 * @attr_clr and that @attr_set can't have any atime bits set if
4397 * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4398 */
4399 if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4400 if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4401 return -EINVAL;
4402
4403 /*
4404 * Clear all previous time settings as they are mutually
4405 * exclusive.
4406 */
4407 kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4408 switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4409 case MOUNT_ATTR_RELATIME:
4410 kattr->attr_set |= MNT_RELATIME;
4411 break;
4412 case MOUNT_ATTR_NOATIME:
4413 kattr->attr_set |= MNT_NOATIME;
4414 break;
4415 case MOUNT_ATTR_STRICTATIME:
4416 break;
4417 default:
4418 return -EINVAL;
4419 }
4420 } else {
4421 if (attr->attr_set & MOUNT_ATTR__ATIME)
4422 return -EINVAL;
4423 }
4424
4425 return build_mount_idmapped(attr, usize, kattr, flags);
4426}
4427
4428static void finish_mount_kattr(struct mount_kattr *kattr)
4429{
4430 put_user_ns(kattr->mnt_userns);
4431 kattr->mnt_userns = NULL;
4432
4433 if (kattr->mnt_idmap)
4434 mnt_idmap_put(kattr->mnt_idmap);
4435}
4436
4437SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4438 unsigned int, flags, struct mount_attr __user *, uattr,
4439 size_t, usize)
4440{
4441 int err;
4442 struct path target;
4443 struct mount_attr attr;
4444 struct mount_kattr kattr;
4445
4446 BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4447
4448 if (flags & ~(AT_EMPTY_PATH |
4449 AT_RECURSIVE |
4450 AT_SYMLINK_NOFOLLOW |
4451 AT_NO_AUTOMOUNT))
4452 return -EINVAL;
4453
4454 if (unlikely(usize > PAGE_SIZE))
4455 return -E2BIG;
4456 if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4457 return -EINVAL;
4458
4459 if (!may_mount())
4460 return -EPERM;
4461
4462 err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4463 if (err)
4464 return err;
4465
4466 /* Don't bother walking through the mounts if this is a nop. */
4467 if (attr.attr_set == 0 &&
4468 attr.attr_clr == 0 &&
4469 attr.propagation == 0)
4470 return 0;
4471
4472 err = build_mount_kattr(&attr, usize, &kattr, flags);
4473 if (err)
4474 return err;
4475
4476 err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4477 if (!err) {
4478 err = do_mount_setattr(&target, &kattr);
4479 path_put(&target);
4480 }
4481 finish_mount_kattr(&kattr);
4482 return err;
4483}
4484
4485static void __init init_mount_tree(void)
4486{
4487 struct vfsmount *mnt;
4488 struct mount *m;
4489 struct mnt_namespace *ns;
4490 struct path root;
4491
4492 mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4493 if (IS_ERR(mnt))
4494 panic("Can't create rootfs");
4495
4496 ns = alloc_mnt_ns(&init_user_ns, false);
4497 if (IS_ERR(ns))
4498 panic("Can't allocate initial namespace");
4499 m = real_mount(mnt);
4500 m->mnt_ns = ns;
4501 ns->root = m;
4502 ns->mounts = 1;
4503 list_add(&m->mnt_list, &ns->list);
4504 init_task.nsproxy->mnt_ns = ns;
4505 get_mnt_ns(ns);
4506
4507 root.mnt = mnt;
4508 root.dentry = mnt->mnt_root;
4509 mnt->mnt_flags |= MNT_LOCKED;
4510
4511 set_fs_pwd(current->fs, &root);
4512 set_fs_root(current->fs, &root);
4513}
4514
4515void __init mnt_init(void)
4516{
4517 int err;
4518
4519 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4520 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4521
4522 mount_hashtable = alloc_large_system_hash("Mount-cache",
4523 sizeof(struct hlist_head),
4524 mhash_entries, 19,
4525 HASH_ZERO,
4526 &m_hash_shift, &m_hash_mask, 0, 0);
4527 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4528 sizeof(struct hlist_head),
4529 mphash_entries, 19,
4530 HASH_ZERO,
4531 &mp_hash_shift, &mp_hash_mask, 0, 0);
4532
4533 if (!mount_hashtable || !mountpoint_hashtable)
4534 panic("Failed to allocate mount hash table\n");
4535
4536 kernfs_init();
4537
4538 err = sysfs_init();
4539 if (err)
4540 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4541 __func__, err);
4542 fs_kobj = kobject_create_and_add("fs", NULL);
4543 if (!fs_kobj)
4544 printk(KERN_WARNING "%s: kobj create error\n", __func__);
4545 shmem_init();
4546 init_rootfs();
4547 init_mount_tree();
4548}
4549
4550void put_mnt_ns(struct mnt_namespace *ns)
4551{
4552 if (!refcount_dec_and_test(&ns->ns.count))
4553 return;
4554 drop_collected_mounts(&ns->root->mnt);
4555 free_mnt_ns(ns);
4556}
4557
4558struct vfsmount *kern_mount(struct file_system_type *type)
4559{
4560 struct vfsmount *mnt;
4561 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4562 if (!IS_ERR(mnt)) {
4563 /*
4564 * it is a longterm mount, don't release mnt until
4565 * we unmount before file sys is unregistered
4566 */
4567 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4568 }
4569 return mnt;
4570}
4571EXPORT_SYMBOL_GPL(kern_mount);
4572
4573void kern_unmount(struct vfsmount *mnt)
4574{
4575 /* release long term mount so mount point can be released */
4576 if (!IS_ERR_OR_NULL(mnt)) {
4577 real_mount(mnt)->mnt_ns = NULL;
4578 synchronize_rcu(); /* yecchhh... */
4579 mntput(mnt);
4580 }
4581}
4582EXPORT_SYMBOL(kern_unmount);
4583
4584void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4585{
4586 unsigned int i;
4587
4588 for (i = 0; i < num; i++)
4589 if (mnt[i])
4590 real_mount(mnt[i])->mnt_ns = NULL;
4591 synchronize_rcu_expedited();
4592 for (i = 0; i < num; i++)
4593 mntput(mnt[i]);
4594}
4595EXPORT_SYMBOL(kern_unmount_array);
4596
4597bool our_mnt(struct vfsmount *mnt)
4598{
4599 return check_mnt(real_mount(mnt));
4600}
4601
4602bool current_chrooted(void)
4603{
4604 /* Does the current process have a non-standard root */
4605 struct path ns_root;
4606 struct path fs_root;
4607 bool chrooted;
4608
4609 /* Find the namespace root */
4610 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
4611 ns_root.dentry = ns_root.mnt->mnt_root;
4612 path_get(&ns_root);
4613 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4614 ;
4615
4616 get_fs_root(current->fs, &fs_root);
4617
4618 chrooted = !path_equal(&fs_root, &ns_root);
4619
4620 path_put(&fs_root);
4621 path_put(&ns_root);
4622
4623 return chrooted;
4624}
4625
4626static bool mnt_already_visible(struct mnt_namespace *ns,
4627 const struct super_block *sb,
4628 int *new_mnt_flags)
4629{
4630 int new_flags = *new_mnt_flags;
4631 struct mount *mnt;
4632 bool visible = false;
4633
4634 down_read(&namespace_sem);
4635 lock_ns_list(ns);
4636 list_for_each_entry(mnt, &ns->list, mnt_list) {
4637 struct mount *child;
4638 int mnt_flags;
4639
4640 if (mnt_is_cursor(mnt))
4641 continue;
4642
4643 if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4644 continue;
4645
4646 /* This mount is not fully visible if it's root directory
4647 * is not the root directory of the filesystem.
4648 */
4649 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4650 continue;
4651
4652 /* A local view of the mount flags */
4653 mnt_flags = mnt->mnt.mnt_flags;
4654
4655 /* Don't miss readonly hidden in the superblock flags */
4656 if (sb_rdonly(mnt->mnt.mnt_sb))
4657 mnt_flags |= MNT_LOCK_READONLY;
4658
4659 /* Verify the mount flags are equal to or more permissive
4660 * than the proposed new mount.
4661 */
4662 if ((mnt_flags & MNT_LOCK_READONLY) &&
4663 !(new_flags & MNT_READONLY))
4664 continue;
4665 if ((mnt_flags & MNT_LOCK_ATIME) &&
4666 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4667 continue;
4668
4669 /* This mount is not fully visible if there are any
4670 * locked child mounts that cover anything except for
4671 * empty directories.
4672 */
4673 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4674 struct inode *inode = child->mnt_mountpoint->d_inode;
4675 /* Only worry about locked mounts */
4676 if (!(child->mnt.mnt_flags & MNT_LOCKED))
4677 continue;
4678 /* Is the directory permanetly empty? */
4679 if (!is_empty_dir_inode(inode))
4680 goto next;
4681 }
4682 /* Preserve the locked attributes */
4683 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4684 MNT_LOCK_ATIME);
4685 visible = true;
4686 goto found;
4687 next: ;
4688 }
4689found:
4690 unlock_ns_list(ns);
4691 up_read(&namespace_sem);
4692 return visible;
4693}
4694
4695static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4696{
4697 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4698 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4699 unsigned long s_iflags;
4700
4701 if (ns->user_ns == &init_user_ns)
4702 return false;
4703
4704 /* Can this filesystem be too revealing? */
4705 s_iflags = sb->s_iflags;
4706 if (!(s_iflags & SB_I_USERNS_VISIBLE))
4707 return false;
4708
4709 if ((s_iflags & required_iflags) != required_iflags) {
4710 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4711 required_iflags);
4712 return true;
4713 }
4714
4715 return !mnt_already_visible(ns, sb, new_mnt_flags);
4716}
4717
4718bool mnt_may_suid(struct vfsmount *mnt)
4719{
4720 /*
4721 * Foreign mounts (accessed via fchdir or through /proc
4722 * symlinks) are always treated as if they are nosuid. This
4723 * prevents namespaces from trusting potentially unsafe
4724 * suid/sgid bits, file caps, or security labels that originate
4725 * in other namespaces.
4726 */
4727 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4728 current_in_userns(mnt->mnt_sb->s_user_ns);
4729}
4730
4731static struct ns_common *mntns_get(struct task_struct *task)
4732{
4733 struct ns_common *ns = NULL;
4734 struct nsproxy *nsproxy;
4735
4736 task_lock(task);
4737 nsproxy = task->nsproxy;
4738 if (nsproxy) {
4739 ns = &nsproxy->mnt_ns->ns;
4740 get_mnt_ns(to_mnt_ns(ns));
4741 }
4742 task_unlock(task);
4743
4744 return ns;
4745}
4746
4747static void mntns_put(struct ns_common *ns)
4748{
4749 put_mnt_ns(to_mnt_ns(ns));
4750}
4751
4752static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4753{
4754 struct nsproxy *nsproxy = nsset->nsproxy;
4755 struct fs_struct *fs = nsset->fs;
4756 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4757 struct user_namespace *user_ns = nsset->cred->user_ns;
4758 struct path root;
4759 int err;
4760
4761 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4762 !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4763 !ns_capable(user_ns, CAP_SYS_ADMIN))
4764 return -EPERM;
4765
4766 if (is_anon_ns(mnt_ns))
4767 return -EINVAL;
4768
4769 if (fs->users != 1)
4770 return -EINVAL;
4771
4772 get_mnt_ns(mnt_ns);
4773 old_mnt_ns = nsproxy->mnt_ns;
4774 nsproxy->mnt_ns = mnt_ns;
4775
4776 /* Find the root */
4777 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4778 "/", LOOKUP_DOWN, &root);
4779 if (err) {
4780 /* revert to old namespace */
4781 nsproxy->mnt_ns = old_mnt_ns;
4782 put_mnt_ns(mnt_ns);
4783 return err;
4784 }
4785
4786 put_mnt_ns(old_mnt_ns);
4787
4788 /* Update the pwd and root */
4789 set_fs_pwd(fs, &root);
4790 set_fs_root(fs, &root);
4791
4792 path_put(&root);
4793 return 0;
4794}
4795
4796static struct user_namespace *mntns_owner(struct ns_common *ns)
4797{
4798 return to_mnt_ns(ns)->user_ns;
4799}
4800
4801const struct proc_ns_operations mntns_operations = {
4802 .name = "mnt",
4803 .type = CLONE_NEWNS,
4804 .get = mntns_get,
4805 .put = mntns_put,
4806 .install = mntns_install,
4807 .owner = mntns_owner,
4808};
4809
4810#ifdef CONFIG_SYSCTL
4811static struct ctl_table fs_namespace_sysctls[] = {
4812 {
4813 .procname = "mount-max",
4814 .data = &sysctl_mount_max,
4815 .maxlen = sizeof(unsigned int),
4816 .mode = 0644,
4817 .proc_handler = proc_dointvec_minmax,
4818 .extra1 = SYSCTL_ONE,
4819 },
4820 { }
4821};
4822
4823static int __init init_fs_namespace_sysctls(void)
4824{
4825 register_sysctl_init("fs", fs_namespace_sysctls);
4826 return 0;
4827}
4828fs_initcall(init_fs_namespace_sysctls);
4829
4830#endif /* CONFIG_SYSCTL */