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/*
2 * linux/fs/namespace.c
3 *
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
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/uaccess.h>
24#include <linux/proc_ns.h>
25#include <linux/magic.h>
26#include <linux/bootmem.h>
27#include <linux/task_work.h>
28#include <linux/sched/task.h>
29
30#include "pnode.h"
31#include "internal.h"
32
33/* Maximum number of mounts in a mount namespace */
34unsigned int sysctl_mount_max __read_mostly = 100000;
35
36static unsigned int m_hash_mask __read_mostly;
37static unsigned int m_hash_shift __read_mostly;
38static unsigned int mp_hash_mask __read_mostly;
39static unsigned int mp_hash_shift __read_mostly;
40
41static __initdata unsigned long mhash_entries;
42static int __init set_mhash_entries(char *str)
43{
44 if (!str)
45 return 0;
46 mhash_entries = simple_strtoul(str, &str, 0);
47 return 1;
48}
49__setup("mhash_entries=", set_mhash_entries);
50
51static __initdata unsigned long mphash_entries;
52static int __init set_mphash_entries(char *str)
53{
54 if (!str)
55 return 0;
56 mphash_entries = simple_strtoul(str, &str, 0);
57 return 1;
58}
59__setup("mphash_entries=", set_mphash_entries);
60
61static u64 event;
62static DEFINE_IDA(mnt_id_ida);
63static DEFINE_IDA(mnt_group_ida);
64static DEFINE_SPINLOCK(mnt_id_lock);
65static int mnt_id_start = 0;
66static int mnt_group_start = 1;
67
68static struct hlist_head *mount_hashtable __read_mostly;
69static struct hlist_head *mountpoint_hashtable __read_mostly;
70static struct kmem_cache *mnt_cache __read_mostly;
71static DECLARE_RWSEM(namespace_sem);
72
73/* /sys/fs */
74struct kobject *fs_kobj;
75EXPORT_SYMBOL_GPL(fs_kobj);
76
77/*
78 * vfsmount lock may be taken for read to prevent changes to the
79 * vfsmount hash, ie. during mountpoint lookups or walking back
80 * up the tree.
81 *
82 * It should be taken for write in all cases where the vfsmount
83 * tree or hash is modified or when a vfsmount structure is modified.
84 */
85__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
86
87static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
88{
89 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
90 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
91 tmp = tmp + (tmp >> m_hash_shift);
92 return &mount_hashtable[tmp & m_hash_mask];
93}
94
95static inline struct hlist_head *mp_hash(struct dentry *dentry)
96{
97 unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
98 tmp = tmp + (tmp >> mp_hash_shift);
99 return &mountpoint_hashtable[tmp & mp_hash_mask];
100}
101
102static int mnt_alloc_id(struct mount *mnt)
103{
104 int res;
105
106retry:
107 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
108 spin_lock(&mnt_id_lock);
109 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
110 if (!res)
111 mnt_id_start = mnt->mnt_id + 1;
112 spin_unlock(&mnt_id_lock);
113 if (res == -EAGAIN)
114 goto retry;
115
116 return res;
117}
118
119static void mnt_free_id(struct mount *mnt)
120{
121 int id = mnt->mnt_id;
122 spin_lock(&mnt_id_lock);
123 ida_remove(&mnt_id_ida, id);
124 if (mnt_id_start > id)
125 mnt_id_start = id;
126 spin_unlock(&mnt_id_lock);
127}
128
129/*
130 * Allocate a new peer group ID
131 *
132 * mnt_group_ida is protected by namespace_sem
133 */
134static int mnt_alloc_group_id(struct mount *mnt)
135{
136 int res;
137
138 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 return -ENOMEM;
140
141 res = ida_get_new_above(&mnt_group_ida,
142 mnt_group_start,
143 &mnt->mnt_group_id);
144 if (!res)
145 mnt_group_start = mnt->mnt_group_id + 1;
146
147 return res;
148}
149
150/*
151 * Release a peer group ID
152 */
153void mnt_release_group_id(struct mount *mnt)
154{
155 int id = mnt->mnt_group_id;
156 ida_remove(&mnt_group_ida, id);
157 if (mnt_group_start > id)
158 mnt_group_start = id;
159 mnt->mnt_group_id = 0;
160}
161
162/*
163 * vfsmount lock must be held for read
164 */
165static inline void mnt_add_count(struct mount *mnt, int n)
166{
167#ifdef CONFIG_SMP
168 this_cpu_add(mnt->mnt_pcp->mnt_count, n);
169#else
170 preempt_disable();
171 mnt->mnt_count += n;
172 preempt_enable();
173#endif
174}
175
176/*
177 * vfsmount lock must be held for write
178 */
179unsigned int mnt_get_count(struct mount *mnt)
180{
181#ifdef CONFIG_SMP
182 unsigned int count = 0;
183 int cpu;
184
185 for_each_possible_cpu(cpu) {
186 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
187 }
188
189 return count;
190#else
191 return mnt->mnt_count;
192#endif
193}
194
195static void drop_mountpoint(struct fs_pin *p)
196{
197 struct mount *m = container_of(p, struct mount, mnt_umount);
198 dput(m->mnt_ex_mountpoint);
199 pin_remove(p);
200 mntput(&m->mnt);
201}
202
203static struct mount *alloc_vfsmnt(const char *name)
204{
205 struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
206 if (mnt) {
207 int err;
208
209 err = mnt_alloc_id(mnt);
210 if (err)
211 goto out_free_cache;
212
213 if (name) {
214 mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
215 if (!mnt->mnt_devname)
216 goto out_free_id;
217 }
218
219#ifdef CONFIG_SMP
220 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 if (!mnt->mnt_pcp)
222 goto out_free_devname;
223
224 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225#else
226 mnt->mnt_count = 1;
227 mnt->mnt_writers = 0;
228#endif
229
230 INIT_HLIST_NODE(&mnt->mnt_hash);
231 INIT_LIST_HEAD(&mnt->mnt_child);
232 INIT_LIST_HEAD(&mnt->mnt_mounts);
233 INIT_LIST_HEAD(&mnt->mnt_list);
234 INIT_LIST_HEAD(&mnt->mnt_expire);
235 INIT_LIST_HEAD(&mnt->mnt_share);
236 INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 INIT_LIST_HEAD(&mnt->mnt_slave);
238 INIT_HLIST_NODE(&mnt->mnt_mp_list);
239 INIT_LIST_HEAD(&mnt->mnt_umounting);
240 init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241 }
242 return mnt;
243
244#ifdef CONFIG_SMP
245out_free_devname:
246 kfree_const(mnt->mnt_devname);
247#endif
248out_free_id:
249 mnt_free_id(mnt);
250out_free_cache:
251 kmem_cache_free(mnt_cache, mnt);
252 return NULL;
253}
254
255/*
256 * Most r/o checks on a fs are for operations that take
257 * discrete amounts of time, like a write() or unlink().
258 * We must keep track of when those operations start
259 * (for permission checks) and when they end, so that
260 * we can determine when writes are able to occur to
261 * a filesystem.
262 */
263/*
264 * __mnt_is_readonly: check whether a mount is read-only
265 * @mnt: the mount to check for its write status
266 *
267 * This shouldn't be used directly ouside of the VFS.
268 * It does not guarantee that the filesystem will stay
269 * r/w, just that it is right *now*. This can not and
270 * should not be used in place of IS_RDONLY(inode).
271 * mnt_want/drop_write() will _keep_ the filesystem
272 * r/w.
273 */
274int __mnt_is_readonly(struct vfsmount *mnt)
275{
276 if (mnt->mnt_flags & MNT_READONLY)
277 return 1;
278 if (sb_rdonly(mnt->mnt_sb))
279 return 1;
280 return 0;
281}
282EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283
284static inline void mnt_inc_writers(struct mount *mnt)
285{
286#ifdef CONFIG_SMP
287 this_cpu_inc(mnt->mnt_pcp->mnt_writers);
288#else
289 mnt->mnt_writers++;
290#endif
291}
292
293static inline void mnt_dec_writers(struct mount *mnt)
294{
295#ifdef CONFIG_SMP
296 this_cpu_dec(mnt->mnt_pcp->mnt_writers);
297#else
298 mnt->mnt_writers--;
299#endif
300}
301
302static unsigned int mnt_get_writers(struct mount *mnt)
303{
304#ifdef CONFIG_SMP
305 unsigned int count = 0;
306 int cpu;
307
308 for_each_possible_cpu(cpu) {
309 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 }
311
312 return count;
313#else
314 return mnt->mnt_writers;
315#endif
316}
317
318static int mnt_is_readonly(struct vfsmount *mnt)
319{
320 if (mnt->mnt_sb->s_readonly_remount)
321 return 1;
322 /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 smp_rmb();
324 return __mnt_is_readonly(mnt);
325}
326
327/*
328 * Most r/o & frozen checks on a fs are for operations that take discrete
329 * amounts of time, like a write() or unlink(). We must keep track of when
330 * those operations start (for permission checks) and when they end, so that we
331 * can determine when writes are able to occur to a filesystem.
332 */
333/**
334 * __mnt_want_write - get write access to a mount without freeze protection
335 * @m: the mount on which to take a write
336 *
337 * This tells the low-level filesystem that a write is about to be performed to
338 * it, and makes sure that writes are allowed (mnt it read-write) before
339 * returning success. This operation does not protect against filesystem being
340 * frozen. When the write operation is finished, __mnt_drop_write() must be
341 * called. This is effectively a refcount.
342 */
343int __mnt_want_write(struct vfsmount *m)
344{
345 struct mount *mnt = real_mount(m);
346 int ret = 0;
347
348 preempt_disable();
349 mnt_inc_writers(mnt);
350 /*
351 * The store to mnt_inc_writers must be visible before we pass
352 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 * incremented count after it has set MNT_WRITE_HOLD.
354 */
355 smp_mb();
356 while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
357 cpu_relax();
358 /*
359 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 * be set to match its requirements. So we must not load that until
361 * MNT_WRITE_HOLD is cleared.
362 */
363 smp_rmb();
364 if (mnt_is_readonly(m)) {
365 mnt_dec_writers(mnt);
366 ret = -EROFS;
367 }
368 preempt_enable();
369
370 return ret;
371}
372
373/**
374 * mnt_want_write - get write access to a mount
375 * @m: the mount on which to take a write
376 *
377 * This tells the low-level filesystem that a write is about to be performed to
378 * it, and makes sure that writes are allowed (mount is read-write, filesystem
379 * is not frozen) before returning success. When the write operation is
380 * finished, mnt_drop_write() must be called. This is effectively a refcount.
381 */
382int mnt_want_write(struct vfsmount *m)
383{
384 int ret;
385
386 sb_start_write(m->mnt_sb);
387 ret = __mnt_want_write(m);
388 if (ret)
389 sb_end_write(m->mnt_sb);
390 return ret;
391}
392EXPORT_SYMBOL_GPL(mnt_want_write);
393
394/**
395 * mnt_clone_write - get write access to a mount
396 * @mnt: the mount on which to take a write
397 *
398 * This is effectively like mnt_want_write, except
399 * it must only be used to take an extra write reference
400 * on a mountpoint that we already know has a write reference
401 * on it. This allows some optimisation.
402 *
403 * After finished, mnt_drop_write must be called as usual to
404 * drop the reference.
405 */
406int mnt_clone_write(struct vfsmount *mnt)
407{
408 /* superblock may be r/o */
409 if (__mnt_is_readonly(mnt))
410 return -EROFS;
411 preempt_disable();
412 mnt_inc_writers(real_mount(mnt));
413 preempt_enable();
414 return 0;
415}
416EXPORT_SYMBOL_GPL(mnt_clone_write);
417
418/**
419 * __mnt_want_write_file - get write access to a file's mount
420 * @file: the file who's mount on which to take a write
421 *
422 * This is like __mnt_want_write, but it takes a file and can
423 * do some optimisations if the file is open for write already
424 */
425int __mnt_want_write_file(struct file *file)
426{
427 if (!(file->f_mode & FMODE_WRITER))
428 return __mnt_want_write(file->f_path.mnt);
429 else
430 return mnt_clone_write(file->f_path.mnt);
431}
432
433/**
434 * mnt_want_write_file_path - get write access to a file's mount
435 * @file: the file who's mount on which to take a write
436 *
437 * This is like mnt_want_write, but it takes a file and can
438 * do some optimisations if the file is open for write already
439 *
440 * Called by the vfs for cases when we have an open file at hand, but will do an
441 * inode operation on it (important distinction for files opened on overlayfs,
442 * since the file operations will come from the real underlying file, while
443 * inode operations come from the overlay).
444 */
445int mnt_want_write_file_path(struct file *file)
446{
447 int ret;
448
449 sb_start_write(file->f_path.mnt->mnt_sb);
450 ret = __mnt_want_write_file(file);
451 if (ret)
452 sb_end_write(file->f_path.mnt->mnt_sb);
453 return ret;
454}
455
456static inline int may_write_real(struct file *file)
457{
458 struct dentry *dentry = file->f_path.dentry;
459 struct dentry *upperdentry;
460
461 /* Writable file? */
462 if (file->f_mode & FMODE_WRITER)
463 return 0;
464
465 /* Not overlayfs? */
466 if (likely(!(dentry->d_flags & DCACHE_OP_REAL)))
467 return 0;
468
469 /* File refers to upper, writable layer? */
470 upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
471 if (upperdentry &&
472 (file_inode(file) == d_inode(upperdentry) ||
473 file_inode(file) == d_inode(dentry)))
474 return 0;
475
476 /* Lower layer: can't write to real file, sorry... */
477 return -EPERM;
478}
479
480/**
481 * mnt_want_write_file - get write access to a file's mount
482 * @file: the file who's mount on which to take a write
483 *
484 * This is like mnt_want_write, but it takes a file and can
485 * do some optimisations if the file is open for write already
486 *
487 * Mostly called by filesystems from their ioctl operation before performing
488 * modification. On overlayfs this needs to check if the file is on a read-only
489 * lower layer and deny access in that case.
490 */
491int mnt_want_write_file(struct file *file)
492{
493 int ret;
494
495 ret = may_write_real(file);
496 if (!ret) {
497 sb_start_write(file_inode(file)->i_sb);
498 ret = __mnt_want_write_file(file);
499 if (ret)
500 sb_end_write(file_inode(file)->i_sb);
501 }
502 return ret;
503}
504EXPORT_SYMBOL_GPL(mnt_want_write_file);
505
506/**
507 * __mnt_drop_write - give up write access to a mount
508 * @mnt: the mount on which to give up write access
509 *
510 * Tells the low-level filesystem that we are done
511 * performing writes to it. Must be matched with
512 * __mnt_want_write() call above.
513 */
514void __mnt_drop_write(struct vfsmount *mnt)
515{
516 preempt_disable();
517 mnt_dec_writers(real_mount(mnt));
518 preempt_enable();
519}
520
521/**
522 * mnt_drop_write - give up write access to a mount
523 * @mnt: the mount on which to give up write access
524 *
525 * Tells the low-level filesystem that we are done performing writes to it and
526 * also allows filesystem to be frozen again. Must be matched with
527 * mnt_want_write() call above.
528 */
529void mnt_drop_write(struct vfsmount *mnt)
530{
531 __mnt_drop_write(mnt);
532 sb_end_write(mnt->mnt_sb);
533}
534EXPORT_SYMBOL_GPL(mnt_drop_write);
535
536void __mnt_drop_write_file(struct file *file)
537{
538 __mnt_drop_write(file->f_path.mnt);
539}
540
541void mnt_drop_write_file_path(struct file *file)
542{
543 mnt_drop_write(file->f_path.mnt);
544}
545
546void mnt_drop_write_file(struct file *file)
547{
548 __mnt_drop_write(file->f_path.mnt);
549 sb_end_write(file_inode(file)->i_sb);
550}
551EXPORT_SYMBOL(mnt_drop_write_file);
552
553static int mnt_make_readonly(struct mount *mnt)
554{
555 int ret = 0;
556
557 lock_mount_hash();
558 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
559 /*
560 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
561 * should be visible before we do.
562 */
563 smp_mb();
564
565 /*
566 * With writers on hold, if this value is zero, then there are
567 * definitely no active writers (although held writers may subsequently
568 * increment the count, they'll have to wait, and decrement it after
569 * seeing MNT_READONLY).
570 *
571 * It is OK to have counter incremented on one CPU and decremented on
572 * another: the sum will add up correctly. The danger would be when we
573 * sum up each counter, if we read a counter before it is incremented,
574 * but then read another CPU's count which it has been subsequently
575 * decremented from -- we would see more decrements than we should.
576 * MNT_WRITE_HOLD protects against this scenario, because
577 * mnt_want_write first increments count, then smp_mb, then spins on
578 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
579 * we're counting up here.
580 */
581 if (mnt_get_writers(mnt) > 0)
582 ret = -EBUSY;
583 else
584 mnt->mnt.mnt_flags |= MNT_READONLY;
585 /*
586 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
587 * that become unheld will see MNT_READONLY.
588 */
589 smp_wmb();
590 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
591 unlock_mount_hash();
592 return ret;
593}
594
595static void __mnt_unmake_readonly(struct mount *mnt)
596{
597 lock_mount_hash();
598 mnt->mnt.mnt_flags &= ~MNT_READONLY;
599 unlock_mount_hash();
600}
601
602int sb_prepare_remount_readonly(struct super_block *sb)
603{
604 struct mount *mnt;
605 int err = 0;
606
607 /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */
608 if (atomic_long_read(&sb->s_remove_count))
609 return -EBUSY;
610
611 lock_mount_hash();
612 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
613 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
614 mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
615 smp_mb();
616 if (mnt_get_writers(mnt) > 0) {
617 err = -EBUSY;
618 break;
619 }
620 }
621 }
622 if (!err && atomic_long_read(&sb->s_remove_count))
623 err = -EBUSY;
624
625 if (!err) {
626 sb->s_readonly_remount = 1;
627 smp_wmb();
628 }
629 list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
630 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
631 mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
632 }
633 unlock_mount_hash();
634
635 return err;
636}
637
638static void free_vfsmnt(struct mount *mnt)
639{
640 kfree_const(mnt->mnt_devname);
641#ifdef CONFIG_SMP
642 free_percpu(mnt->mnt_pcp);
643#endif
644 kmem_cache_free(mnt_cache, mnt);
645}
646
647static void delayed_free_vfsmnt(struct rcu_head *head)
648{
649 free_vfsmnt(container_of(head, struct mount, mnt_rcu));
650}
651
652/* call under rcu_read_lock */
653int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
654{
655 struct mount *mnt;
656 if (read_seqretry(&mount_lock, seq))
657 return 1;
658 if (bastard == NULL)
659 return 0;
660 mnt = real_mount(bastard);
661 mnt_add_count(mnt, 1);
662 if (likely(!read_seqretry(&mount_lock, seq)))
663 return 0;
664 if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
665 mnt_add_count(mnt, -1);
666 return 1;
667 }
668 return -1;
669}
670
671/* call under rcu_read_lock */
672bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
673{
674 int res = __legitimize_mnt(bastard, seq);
675 if (likely(!res))
676 return true;
677 if (unlikely(res < 0)) {
678 rcu_read_unlock();
679 mntput(bastard);
680 rcu_read_lock();
681 }
682 return false;
683}
684
685/*
686 * find the first mount at @dentry on vfsmount @mnt.
687 * call under rcu_read_lock()
688 */
689struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
690{
691 struct hlist_head *head = m_hash(mnt, dentry);
692 struct mount *p;
693
694 hlist_for_each_entry_rcu(p, head, mnt_hash)
695 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
696 return p;
697 return NULL;
698}
699
700/*
701 * lookup_mnt - Return the first child mount mounted at path
702 *
703 * "First" means first mounted chronologically. If you create the
704 * following mounts:
705 *
706 * mount /dev/sda1 /mnt
707 * mount /dev/sda2 /mnt
708 * mount /dev/sda3 /mnt
709 *
710 * Then lookup_mnt() on the base /mnt dentry in the root mount will
711 * return successively the root dentry and vfsmount of /dev/sda1, then
712 * /dev/sda2, then /dev/sda3, then NULL.
713 *
714 * lookup_mnt takes a reference to the found vfsmount.
715 */
716struct vfsmount *lookup_mnt(const struct path *path)
717{
718 struct mount *child_mnt;
719 struct vfsmount *m;
720 unsigned seq;
721
722 rcu_read_lock();
723 do {
724 seq = read_seqbegin(&mount_lock);
725 child_mnt = __lookup_mnt(path->mnt, path->dentry);
726 m = child_mnt ? &child_mnt->mnt : NULL;
727 } while (!legitimize_mnt(m, seq));
728 rcu_read_unlock();
729 return m;
730}
731
732/*
733 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
734 * current mount namespace.
735 *
736 * The common case is dentries are not mountpoints at all and that
737 * test is handled inline. For the slow case when we are actually
738 * dealing with a mountpoint of some kind, walk through all of the
739 * mounts in the current mount namespace and test to see if the dentry
740 * is a mountpoint.
741 *
742 * The mount_hashtable is not usable in the context because we
743 * need to identify all mounts that may be in the current mount
744 * namespace not just a mount that happens to have some specified
745 * parent mount.
746 */
747bool __is_local_mountpoint(struct dentry *dentry)
748{
749 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
750 struct mount *mnt;
751 bool is_covered = false;
752
753 if (!d_mountpoint(dentry))
754 goto out;
755
756 down_read(&namespace_sem);
757 list_for_each_entry(mnt, &ns->list, mnt_list) {
758 is_covered = (mnt->mnt_mountpoint == dentry);
759 if (is_covered)
760 break;
761 }
762 up_read(&namespace_sem);
763out:
764 return is_covered;
765}
766
767static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
768{
769 struct hlist_head *chain = mp_hash(dentry);
770 struct mountpoint *mp;
771
772 hlist_for_each_entry(mp, chain, m_hash) {
773 if (mp->m_dentry == dentry) {
774 /* might be worth a WARN_ON() */
775 if (d_unlinked(dentry))
776 return ERR_PTR(-ENOENT);
777 mp->m_count++;
778 return mp;
779 }
780 }
781 return NULL;
782}
783
784static struct mountpoint *get_mountpoint(struct dentry *dentry)
785{
786 struct mountpoint *mp, *new = NULL;
787 int ret;
788
789 if (d_mountpoint(dentry)) {
790mountpoint:
791 read_seqlock_excl(&mount_lock);
792 mp = lookup_mountpoint(dentry);
793 read_sequnlock_excl(&mount_lock);
794 if (mp)
795 goto done;
796 }
797
798 if (!new)
799 new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
800 if (!new)
801 return ERR_PTR(-ENOMEM);
802
803
804 /* Exactly one processes may set d_mounted */
805 ret = d_set_mounted(dentry);
806
807 /* Someone else set d_mounted? */
808 if (ret == -EBUSY)
809 goto mountpoint;
810
811 /* The dentry is not available as a mountpoint? */
812 mp = ERR_PTR(ret);
813 if (ret)
814 goto done;
815
816 /* Add the new mountpoint to the hash table */
817 read_seqlock_excl(&mount_lock);
818 new->m_dentry = dentry;
819 new->m_count = 1;
820 hlist_add_head(&new->m_hash, mp_hash(dentry));
821 INIT_HLIST_HEAD(&new->m_list);
822 read_sequnlock_excl(&mount_lock);
823
824 mp = new;
825 new = NULL;
826done:
827 kfree(new);
828 return mp;
829}
830
831static void put_mountpoint(struct mountpoint *mp)
832{
833 if (!--mp->m_count) {
834 struct dentry *dentry = mp->m_dentry;
835 BUG_ON(!hlist_empty(&mp->m_list));
836 spin_lock(&dentry->d_lock);
837 dentry->d_flags &= ~DCACHE_MOUNTED;
838 spin_unlock(&dentry->d_lock);
839 hlist_del(&mp->m_hash);
840 kfree(mp);
841 }
842}
843
844static inline int check_mnt(struct mount *mnt)
845{
846 return mnt->mnt_ns == current->nsproxy->mnt_ns;
847}
848
849/*
850 * vfsmount lock must be held for write
851 */
852static void touch_mnt_namespace(struct mnt_namespace *ns)
853{
854 if (ns) {
855 ns->event = ++event;
856 wake_up_interruptible(&ns->poll);
857 }
858}
859
860/*
861 * vfsmount lock must be held for write
862 */
863static void __touch_mnt_namespace(struct mnt_namespace *ns)
864{
865 if (ns && ns->event != event) {
866 ns->event = event;
867 wake_up_interruptible(&ns->poll);
868 }
869}
870
871/*
872 * vfsmount lock must be held for write
873 */
874static void unhash_mnt(struct mount *mnt)
875{
876 mnt->mnt_parent = mnt;
877 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
878 list_del_init(&mnt->mnt_child);
879 hlist_del_init_rcu(&mnt->mnt_hash);
880 hlist_del_init(&mnt->mnt_mp_list);
881 put_mountpoint(mnt->mnt_mp);
882 mnt->mnt_mp = NULL;
883}
884
885/*
886 * vfsmount lock must be held for write
887 */
888static void detach_mnt(struct mount *mnt, struct path *old_path)
889{
890 old_path->dentry = mnt->mnt_mountpoint;
891 old_path->mnt = &mnt->mnt_parent->mnt;
892 unhash_mnt(mnt);
893}
894
895/*
896 * vfsmount lock must be held for write
897 */
898static void umount_mnt(struct mount *mnt)
899{
900 /* old mountpoint will be dropped when we can do that */
901 mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
902 unhash_mnt(mnt);
903}
904
905/*
906 * vfsmount lock must be held for write
907 */
908void mnt_set_mountpoint(struct mount *mnt,
909 struct mountpoint *mp,
910 struct mount *child_mnt)
911{
912 mp->m_count++;
913 mnt_add_count(mnt, 1); /* essentially, that's mntget */
914 child_mnt->mnt_mountpoint = dget(mp->m_dentry);
915 child_mnt->mnt_parent = mnt;
916 child_mnt->mnt_mp = mp;
917 hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
918}
919
920static void __attach_mnt(struct mount *mnt, struct mount *parent)
921{
922 hlist_add_head_rcu(&mnt->mnt_hash,
923 m_hash(&parent->mnt, mnt->mnt_mountpoint));
924 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
925}
926
927/*
928 * vfsmount lock must be held for write
929 */
930static void attach_mnt(struct mount *mnt,
931 struct mount *parent,
932 struct mountpoint *mp)
933{
934 mnt_set_mountpoint(parent, mp, mnt);
935 __attach_mnt(mnt, parent);
936}
937
938void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
939{
940 struct mountpoint *old_mp = mnt->mnt_mp;
941 struct dentry *old_mountpoint = mnt->mnt_mountpoint;
942 struct mount *old_parent = mnt->mnt_parent;
943
944 list_del_init(&mnt->mnt_child);
945 hlist_del_init(&mnt->mnt_mp_list);
946 hlist_del_init_rcu(&mnt->mnt_hash);
947
948 attach_mnt(mnt, parent, mp);
949
950 put_mountpoint(old_mp);
951
952 /*
953 * Safely avoid even the suggestion this code might sleep or
954 * lock the mount hash by taking advantage of the knowledge that
955 * mnt_change_mountpoint will not release the final reference
956 * to a mountpoint.
957 *
958 * During mounting, the mount passed in as the parent mount will
959 * continue to use the old mountpoint and during unmounting, the
960 * old mountpoint will continue to exist until namespace_unlock,
961 * which happens well after mnt_change_mountpoint.
962 */
963 spin_lock(&old_mountpoint->d_lock);
964 old_mountpoint->d_lockref.count--;
965 spin_unlock(&old_mountpoint->d_lock);
966
967 mnt_add_count(old_parent, -1);
968}
969
970/*
971 * vfsmount lock must be held for write
972 */
973static void commit_tree(struct mount *mnt)
974{
975 struct mount *parent = mnt->mnt_parent;
976 struct mount *m;
977 LIST_HEAD(head);
978 struct mnt_namespace *n = parent->mnt_ns;
979
980 BUG_ON(parent == mnt);
981
982 list_add_tail(&head, &mnt->mnt_list);
983 list_for_each_entry(m, &head, mnt_list)
984 m->mnt_ns = n;
985
986 list_splice(&head, n->list.prev);
987
988 n->mounts += n->pending_mounts;
989 n->pending_mounts = 0;
990
991 __attach_mnt(mnt, parent);
992 touch_mnt_namespace(n);
993}
994
995static struct mount *next_mnt(struct mount *p, struct mount *root)
996{
997 struct list_head *next = p->mnt_mounts.next;
998 if (next == &p->mnt_mounts) {
999 while (1) {
1000 if (p == root)
1001 return NULL;
1002 next = p->mnt_child.next;
1003 if (next != &p->mnt_parent->mnt_mounts)
1004 break;
1005 p = p->mnt_parent;
1006 }
1007 }
1008 return list_entry(next, struct mount, mnt_child);
1009}
1010
1011static struct mount *skip_mnt_tree(struct mount *p)
1012{
1013 struct list_head *prev = p->mnt_mounts.prev;
1014 while (prev != &p->mnt_mounts) {
1015 p = list_entry(prev, struct mount, mnt_child);
1016 prev = p->mnt_mounts.prev;
1017 }
1018 return p;
1019}
1020
1021struct vfsmount *
1022vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
1023{
1024 struct mount *mnt;
1025 struct dentry *root;
1026
1027 if (!type)
1028 return ERR_PTR(-ENODEV);
1029
1030 mnt = alloc_vfsmnt(name);
1031 if (!mnt)
1032 return ERR_PTR(-ENOMEM);
1033
1034 if (flags & SB_KERNMOUNT)
1035 mnt->mnt.mnt_flags = MNT_INTERNAL;
1036
1037 root = mount_fs(type, flags, name, data);
1038 if (IS_ERR(root)) {
1039 mnt_free_id(mnt);
1040 free_vfsmnt(mnt);
1041 return ERR_CAST(root);
1042 }
1043
1044 mnt->mnt.mnt_root = root;
1045 mnt->mnt.mnt_sb = root->d_sb;
1046 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1047 mnt->mnt_parent = mnt;
1048 lock_mount_hash();
1049 list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1050 unlock_mount_hash();
1051 return &mnt->mnt;
1052}
1053EXPORT_SYMBOL_GPL(vfs_kern_mount);
1054
1055struct vfsmount *
1056vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1057 const char *name, void *data)
1058{
1059 /* Until it is worked out how to pass the user namespace
1060 * through from the parent mount to the submount don't support
1061 * unprivileged mounts with submounts.
1062 */
1063 if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1064 return ERR_PTR(-EPERM);
1065
1066 return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1067}
1068EXPORT_SYMBOL_GPL(vfs_submount);
1069
1070static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1071 int flag)
1072{
1073 struct super_block *sb = old->mnt.mnt_sb;
1074 struct mount *mnt;
1075 int err;
1076
1077 mnt = alloc_vfsmnt(old->mnt_devname);
1078 if (!mnt)
1079 return ERR_PTR(-ENOMEM);
1080
1081 if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1082 mnt->mnt_group_id = 0; /* not a peer of original */
1083 else
1084 mnt->mnt_group_id = old->mnt_group_id;
1085
1086 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1087 err = mnt_alloc_group_id(mnt);
1088 if (err)
1089 goto out_free;
1090 }
1091
1092 mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1093 mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1094 /* Don't allow unprivileged users to change mount flags */
1095 if (flag & CL_UNPRIVILEGED) {
1096 mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1097
1098 if (mnt->mnt.mnt_flags & MNT_READONLY)
1099 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1100
1101 if (mnt->mnt.mnt_flags & MNT_NODEV)
1102 mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1103
1104 if (mnt->mnt.mnt_flags & MNT_NOSUID)
1105 mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1106
1107 if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1108 mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1109 }
1110
1111 /* Don't allow unprivileged users to reveal what is under a mount */
1112 if ((flag & CL_UNPRIVILEGED) &&
1113 (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1114 mnt->mnt.mnt_flags |= MNT_LOCKED;
1115
1116 atomic_inc(&sb->s_active);
1117 mnt->mnt.mnt_sb = sb;
1118 mnt->mnt.mnt_root = dget(root);
1119 mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1120 mnt->mnt_parent = mnt;
1121 lock_mount_hash();
1122 list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1123 unlock_mount_hash();
1124
1125 if ((flag & CL_SLAVE) ||
1126 ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1127 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1128 mnt->mnt_master = old;
1129 CLEAR_MNT_SHARED(mnt);
1130 } else if (!(flag & CL_PRIVATE)) {
1131 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1132 list_add(&mnt->mnt_share, &old->mnt_share);
1133 if (IS_MNT_SLAVE(old))
1134 list_add(&mnt->mnt_slave, &old->mnt_slave);
1135 mnt->mnt_master = old->mnt_master;
1136 } else {
1137 CLEAR_MNT_SHARED(mnt);
1138 }
1139 if (flag & CL_MAKE_SHARED)
1140 set_mnt_shared(mnt);
1141
1142 /* stick the duplicate mount on the same expiry list
1143 * as the original if that was on one */
1144 if (flag & CL_EXPIRE) {
1145 if (!list_empty(&old->mnt_expire))
1146 list_add(&mnt->mnt_expire, &old->mnt_expire);
1147 }
1148
1149 return mnt;
1150
1151 out_free:
1152 mnt_free_id(mnt);
1153 free_vfsmnt(mnt);
1154 return ERR_PTR(err);
1155}
1156
1157static void cleanup_mnt(struct mount *mnt)
1158{
1159 /*
1160 * This probably indicates that somebody messed
1161 * up a mnt_want/drop_write() pair. If this
1162 * happens, the filesystem was probably unable
1163 * to make r/w->r/o transitions.
1164 */
1165 /*
1166 * The locking used to deal with mnt_count decrement provides barriers,
1167 * so mnt_get_writers() below is safe.
1168 */
1169 WARN_ON(mnt_get_writers(mnt));
1170 if (unlikely(mnt->mnt_pins.first))
1171 mnt_pin_kill(mnt);
1172 fsnotify_vfsmount_delete(&mnt->mnt);
1173 dput(mnt->mnt.mnt_root);
1174 deactivate_super(mnt->mnt.mnt_sb);
1175 mnt_free_id(mnt);
1176 call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1177}
1178
1179static void __cleanup_mnt(struct rcu_head *head)
1180{
1181 cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1182}
1183
1184static LLIST_HEAD(delayed_mntput_list);
1185static void delayed_mntput(struct work_struct *unused)
1186{
1187 struct llist_node *node = llist_del_all(&delayed_mntput_list);
1188 struct mount *m, *t;
1189
1190 llist_for_each_entry_safe(m, t, node, mnt_llist)
1191 cleanup_mnt(m);
1192}
1193static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1194
1195static void mntput_no_expire(struct mount *mnt)
1196{
1197 rcu_read_lock();
1198 mnt_add_count(mnt, -1);
1199 if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1200 rcu_read_unlock();
1201 return;
1202 }
1203 lock_mount_hash();
1204 if (mnt_get_count(mnt)) {
1205 rcu_read_unlock();
1206 unlock_mount_hash();
1207 return;
1208 }
1209 if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1210 rcu_read_unlock();
1211 unlock_mount_hash();
1212 return;
1213 }
1214 mnt->mnt.mnt_flags |= MNT_DOOMED;
1215 rcu_read_unlock();
1216
1217 list_del(&mnt->mnt_instance);
1218
1219 if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1220 struct mount *p, *tmp;
1221 list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) {
1222 umount_mnt(p);
1223 }
1224 }
1225 unlock_mount_hash();
1226
1227 if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1228 struct task_struct *task = current;
1229 if (likely(!(task->flags & PF_KTHREAD))) {
1230 init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1231 if (!task_work_add(task, &mnt->mnt_rcu, true))
1232 return;
1233 }
1234 if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1235 schedule_delayed_work(&delayed_mntput_work, 1);
1236 return;
1237 }
1238 cleanup_mnt(mnt);
1239}
1240
1241void mntput(struct vfsmount *mnt)
1242{
1243 if (mnt) {
1244 struct mount *m = real_mount(mnt);
1245 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1246 if (unlikely(m->mnt_expiry_mark))
1247 m->mnt_expiry_mark = 0;
1248 mntput_no_expire(m);
1249 }
1250}
1251EXPORT_SYMBOL(mntput);
1252
1253struct vfsmount *mntget(struct vfsmount *mnt)
1254{
1255 if (mnt)
1256 mnt_add_count(real_mount(mnt), 1);
1257 return mnt;
1258}
1259EXPORT_SYMBOL(mntget);
1260
1261/* path_is_mountpoint() - Check if path is a mount in the current
1262 * namespace.
1263 *
1264 * d_mountpoint() can only be used reliably to establish if a dentry is
1265 * not mounted in any namespace and that common case is handled inline.
1266 * d_mountpoint() isn't aware of the possibility there may be multiple
1267 * mounts using a given dentry in a different namespace. This function
1268 * checks if the passed in path is a mountpoint rather than the dentry
1269 * alone.
1270 */
1271bool path_is_mountpoint(const struct path *path)
1272{
1273 unsigned seq;
1274 bool res;
1275
1276 if (!d_mountpoint(path->dentry))
1277 return false;
1278
1279 rcu_read_lock();
1280 do {
1281 seq = read_seqbegin(&mount_lock);
1282 res = __path_is_mountpoint(path);
1283 } while (read_seqretry(&mount_lock, seq));
1284 rcu_read_unlock();
1285
1286 return res;
1287}
1288EXPORT_SYMBOL(path_is_mountpoint);
1289
1290struct vfsmount *mnt_clone_internal(const struct path *path)
1291{
1292 struct mount *p;
1293 p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1294 if (IS_ERR(p))
1295 return ERR_CAST(p);
1296 p->mnt.mnt_flags |= MNT_INTERNAL;
1297 return &p->mnt;
1298}
1299
1300#ifdef CONFIG_PROC_FS
1301/* iterator; we want it to have access to namespace_sem, thus here... */
1302static void *m_start(struct seq_file *m, loff_t *pos)
1303{
1304 struct proc_mounts *p = m->private;
1305
1306 down_read(&namespace_sem);
1307 if (p->cached_event == p->ns->event) {
1308 void *v = p->cached_mount;
1309 if (*pos == p->cached_index)
1310 return v;
1311 if (*pos == p->cached_index + 1) {
1312 v = seq_list_next(v, &p->ns->list, &p->cached_index);
1313 return p->cached_mount = v;
1314 }
1315 }
1316
1317 p->cached_event = p->ns->event;
1318 p->cached_mount = seq_list_start(&p->ns->list, *pos);
1319 p->cached_index = *pos;
1320 return p->cached_mount;
1321}
1322
1323static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1324{
1325 struct proc_mounts *p = m->private;
1326
1327 p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1328 p->cached_index = *pos;
1329 return p->cached_mount;
1330}
1331
1332static void m_stop(struct seq_file *m, void *v)
1333{
1334 up_read(&namespace_sem);
1335}
1336
1337static int m_show(struct seq_file *m, void *v)
1338{
1339 struct proc_mounts *p = m->private;
1340 struct mount *r = list_entry(v, struct mount, mnt_list);
1341 return p->show(m, &r->mnt);
1342}
1343
1344const struct seq_operations mounts_op = {
1345 .start = m_start,
1346 .next = m_next,
1347 .stop = m_stop,
1348 .show = m_show,
1349};
1350#endif /* CONFIG_PROC_FS */
1351
1352/**
1353 * may_umount_tree - check if a mount tree is busy
1354 * @mnt: root of mount tree
1355 *
1356 * This is called to check if a tree of mounts has any
1357 * open files, pwds, chroots or sub mounts that are
1358 * busy.
1359 */
1360int may_umount_tree(struct vfsmount *m)
1361{
1362 struct mount *mnt = real_mount(m);
1363 int actual_refs = 0;
1364 int minimum_refs = 0;
1365 struct mount *p;
1366 BUG_ON(!m);
1367
1368 /* write lock needed for mnt_get_count */
1369 lock_mount_hash();
1370 for (p = mnt; p; p = next_mnt(p, mnt)) {
1371 actual_refs += mnt_get_count(p);
1372 minimum_refs += 2;
1373 }
1374 unlock_mount_hash();
1375
1376 if (actual_refs > minimum_refs)
1377 return 0;
1378
1379 return 1;
1380}
1381
1382EXPORT_SYMBOL(may_umount_tree);
1383
1384/**
1385 * may_umount - check if a mount point is busy
1386 * @mnt: root of mount
1387 *
1388 * This is called to check if a mount point has any
1389 * open files, pwds, chroots or sub mounts. If the
1390 * mount has sub mounts this will return busy
1391 * regardless of whether the sub mounts are busy.
1392 *
1393 * Doesn't take quota and stuff into account. IOW, in some cases it will
1394 * give false negatives. The main reason why it's here is that we need
1395 * a non-destructive way to look for easily umountable filesystems.
1396 */
1397int may_umount(struct vfsmount *mnt)
1398{
1399 int ret = 1;
1400 down_read(&namespace_sem);
1401 lock_mount_hash();
1402 if (propagate_mount_busy(real_mount(mnt), 2))
1403 ret = 0;
1404 unlock_mount_hash();
1405 up_read(&namespace_sem);
1406 return ret;
1407}
1408
1409EXPORT_SYMBOL(may_umount);
1410
1411static HLIST_HEAD(unmounted); /* protected by namespace_sem */
1412
1413static void namespace_unlock(void)
1414{
1415 struct hlist_head head;
1416
1417 hlist_move_list(&unmounted, &head);
1418
1419 up_write(&namespace_sem);
1420
1421 if (likely(hlist_empty(&head)))
1422 return;
1423
1424 synchronize_rcu();
1425
1426 group_pin_kill(&head);
1427}
1428
1429static inline void namespace_lock(void)
1430{
1431 down_write(&namespace_sem);
1432}
1433
1434enum umount_tree_flags {
1435 UMOUNT_SYNC = 1,
1436 UMOUNT_PROPAGATE = 2,
1437 UMOUNT_CONNECTED = 4,
1438};
1439
1440static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1441{
1442 /* Leaving mounts connected is only valid for lazy umounts */
1443 if (how & UMOUNT_SYNC)
1444 return true;
1445
1446 /* A mount without a parent has nothing to be connected to */
1447 if (!mnt_has_parent(mnt))
1448 return true;
1449
1450 /* Because the reference counting rules change when mounts are
1451 * unmounted and connected, umounted mounts may not be
1452 * connected to mounted mounts.
1453 */
1454 if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1455 return true;
1456
1457 /* Has it been requested that the mount remain connected? */
1458 if (how & UMOUNT_CONNECTED)
1459 return false;
1460
1461 /* Is the mount locked such that it needs to remain connected? */
1462 if (IS_MNT_LOCKED(mnt))
1463 return false;
1464
1465 /* By default disconnect the mount */
1466 return true;
1467}
1468
1469/*
1470 * mount_lock must be held
1471 * namespace_sem must be held for write
1472 */
1473static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1474{
1475 LIST_HEAD(tmp_list);
1476 struct mount *p;
1477
1478 if (how & UMOUNT_PROPAGATE)
1479 propagate_mount_unlock(mnt);
1480
1481 /* Gather the mounts to umount */
1482 for (p = mnt; p; p = next_mnt(p, mnt)) {
1483 p->mnt.mnt_flags |= MNT_UMOUNT;
1484 list_move(&p->mnt_list, &tmp_list);
1485 }
1486
1487 /* Hide the mounts from mnt_mounts */
1488 list_for_each_entry(p, &tmp_list, mnt_list) {
1489 list_del_init(&p->mnt_child);
1490 }
1491
1492 /* Add propogated mounts to the tmp_list */
1493 if (how & UMOUNT_PROPAGATE)
1494 propagate_umount(&tmp_list);
1495
1496 while (!list_empty(&tmp_list)) {
1497 struct mnt_namespace *ns;
1498 bool disconnect;
1499 p = list_first_entry(&tmp_list, struct mount, mnt_list);
1500 list_del_init(&p->mnt_expire);
1501 list_del_init(&p->mnt_list);
1502 ns = p->mnt_ns;
1503 if (ns) {
1504 ns->mounts--;
1505 __touch_mnt_namespace(ns);
1506 }
1507 p->mnt_ns = NULL;
1508 if (how & UMOUNT_SYNC)
1509 p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1510
1511 disconnect = disconnect_mount(p, how);
1512
1513 pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1514 disconnect ? &unmounted : NULL);
1515 if (mnt_has_parent(p)) {
1516 mnt_add_count(p->mnt_parent, -1);
1517 if (!disconnect) {
1518 /* Don't forget about p */
1519 list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1520 } else {
1521 umount_mnt(p);
1522 }
1523 }
1524 change_mnt_propagation(p, MS_PRIVATE);
1525 }
1526}
1527
1528static void shrink_submounts(struct mount *mnt);
1529
1530static int do_umount(struct mount *mnt, int flags)
1531{
1532 struct super_block *sb = mnt->mnt.mnt_sb;
1533 int retval;
1534
1535 retval = security_sb_umount(&mnt->mnt, flags);
1536 if (retval)
1537 return retval;
1538
1539 /*
1540 * Allow userspace to request a mountpoint be expired rather than
1541 * unmounting unconditionally. Unmount only happens if:
1542 * (1) the mark is already set (the mark is cleared by mntput())
1543 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1544 */
1545 if (flags & MNT_EXPIRE) {
1546 if (&mnt->mnt == current->fs->root.mnt ||
1547 flags & (MNT_FORCE | MNT_DETACH))
1548 return -EINVAL;
1549
1550 /*
1551 * probably don't strictly need the lock here if we examined
1552 * all race cases, but it's a slowpath.
1553 */
1554 lock_mount_hash();
1555 if (mnt_get_count(mnt) != 2) {
1556 unlock_mount_hash();
1557 return -EBUSY;
1558 }
1559 unlock_mount_hash();
1560
1561 if (!xchg(&mnt->mnt_expiry_mark, 1))
1562 return -EAGAIN;
1563 }
1564
1565 /*
1566 * If we may have to abort operations to get out of this
1567 * mount, and they will themselves hold resources we must
1568 * allow the fs to do things. In the Unix tradition of
1569 * 'Gee thats tricky lets do it in userspace' the umount_begin
1570 * might fail to complete on the first run through as other tasks
1571 * must return, and the like. Thats for the mount program to worry
1572 * about for the moment.
1573 */
1574
1575 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1576 sb->s_op->umount_begin(sb);
1577 }
1578
1579 /*
1580 * No sense to grab the lock for this test, but test itself looks
1581 * somewhat bogus. Suggestions for better replacement?
1582 * Ho-hum... In principle, we might treat that as umount + switch
1583 * to rootfs. GC would eventually take care of the old vfsmount.
1584 * Actually it makes sense, especially if rootfs would contain a
1585 * /reboot - static binary that would close all descriptors and
1586 * call reboot(9). Then init(8) could umount root and exec /reboot.
1587 */
1588 if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1589 /*
1590 * Special case for "unmounting" root ...
1591 * we just try to remount it readonly.
1592 */
1593 if (!capable(CAP_SYS_ADMIN))
1594 return -EPERM;
1595 down_write(&sb->s_umount);
1596 if (!sb_rdonly(sb))
1597 retval = do_remount_sb(sb, SB_RDONLY, NULL, 0);
1598 up_write(&sb->s_umount);
1599 return retval;
1600 }
1601
1602 namespace_lock();
1603 lock_mount_hash();
1604 event++;
1605
1606 if (flags & MNT_DETACH) {
1607 if (!list_empty(&mnt->mnt_list))
1608 umount_tree(mnt, UMOUNT_PROPAGATE);
1609 retval = 0;
1610 } else {
1611 shrink_submounts(mnt);
1612 retval = -EBUSY;
1613 if (!propagate_mount_busy(mnt, 2)) {
1614 if (!list_empty(&mnt->mnt_list))
1615 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1616 retval = 0;
1617 }
1618 }
1619 unlock_mount_hash();
1620 namespace_unlock();
1621 return retval;
1622}
1623
1624/*
1625 * __detach_mounts - lazily unmount all mounts on the specified dentry
1626 *
1627 * During unlink, rmdir, and d_drop it is possible to loose the path
1628 * to an existing mountpoint, and wind up leaking the mount.
1629 * detach_mounts allows lazily unmounting those mounts instead of
1630 * leaking them.
1631 *
1632 * The caller may hold dentry->d_inode->i_mutex.
1633 */
1634void __detach_mounts(struct dentry *dentry)
1635{
1636 struct mountpoint *mp;
1637 struct mount *mnt;
1638
1639 namespace_lock();
1640 lock_mount_hash();
1641 mp = lookup_mountpoint(dentry);
1642 if (IS_ERR_OR_NULL(mp))
1643 goto out_unlock;
1644
1645 event++;
1646 while (!hlist_empty(&mp->m_list)) {
1647 mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1648 if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1649 hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1650 umount_mnt(mnt);
1651 }
1652 else umount_tree(mnt, UMOUNT_CONNECTED);
1653 }
1654 put_mountpoint(mp);
1655out_unlock:
1656 unlock_mount_hash();
1657 namespace_unlock();
1658}
1659
1660/*
1661 * Is the caller allowed to modify his namespace?
1662 */
1663static inline bool may_mount(void)
1664{
1665 return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1666}
1667
1668static inline bool may_mandlock(void)
1669{
1670#ifndef CONFIG_MANDATORY_FILE_LOCKING
1671 return false;
1672#endif
1673 return capable(CAP_SYS_ADMIN);
1674}
1675
1676/*
1677 * Now umount can handle mount points as well as block devices.
1678 * This is important for filesystems which use unnamed block devices.
1679 *
1680 * We now support a flag for forced unmount like the other 'big iron'
1681 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1682 */
1683
1684int ksys_umount(char __user *name, int flags)
1685{
1686 struct path path;
1687 struct mount *mnt;
1688 int retval;
1689 int lookup_flags = 0;
1690
1691 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1692 return -EINVAL;
1693
1694 if (!may_mount())
1695 return -EPERM;
1696
1697 if (!(flags & UMOUNT_NOFOLLOW))
1698 lookup_flags |= LOOKUP_FOLLOW;
1699
1700 retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1701 if (retval)
1702 goto out;
1703 mnt = real_mount(path.mnt);
1704 retval = -EINVAL;
1705 if (path.dentry != path.mnt->mnt_root)
1706 goto dput_and_out;
1707 if (!check_mnt(mnt))
1708 goto dput_and_out;
1709 if (mnt->mnt.mnt_flags & MNT_LOCKED)
1710 goto dput_and_out;
1711 retval = -EPERM;
1712 if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1713 goto dput_and_out;
1714
1715 retval = do_umount(mnt, flags);
1716dput_and_out:
1717 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1718 dput(path.dentry);
1719 mntput_no_expire(mnt);
1720out:
1721 return retval;
1722}
1723
1724SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1725{
1726 return ksys_umount(name, flags);
1727}
1728
1729#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1730
1731/*
1732 * The 2.0 compatible umount. No flags.
1733 */
1734SYSCALL_DEFINE1(oldumount, char __user *, name)
1735{
1736 return ksys_umount(name, 0);
1737}
1738
1739#endif
1740
1741static bool is_mnt_ns_file(struct dentry *dentry)
1742{
1743 /* Is this a proxy for a mount namespace? */
1744 return dentry->d_op == &ns_dentry_operations &&
1745 dentry->d_fsdata == &mntns_operations;
1746}
1747
1748struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1749{
1750 return container_of(ns, struct mnt_namespace, ns);
1751}
1752
1753static bool mnt_ns_loop(struct dentry *dentry)
1754{
1755 /* Could bind mounting the mount namespace inode cause a
1756 * mount namespace loop?
1757 */
1758 struct mnt_namespace *mnt_ns;
1759 if (!is_mnt_ns_file(dentry))
1760 return false;
1761
1762 mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1763 return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1764}
1765
1766struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1767 int flag)
1768{
1769 struct mount *res, *p, *q, *r, *parent;
1770
1771 if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1772 return ERR_PTR(-EINVAL);
1773
1774 if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1775 return ERR_PTR(-EINVAL);
1776
1777 res = q = clone_mnt(mnt, dentry, flag);
1778 if (IS_ERR(q))
1779 return q;
1780
1781 q->mnt_mountpoint = mnt->mnt_mountpoint;
1782
1783 p = mnt;
1784 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1785 struct mount *s;
1786 if (!is_subdir(r->mnt_mountpoint, dentry))
1787 continue;
1788
1789 for (s = r; s; s = next_mnt(s, r)) {
1790 if (!(flag & CL_COPY_UNBINDABLE) &&
1791 IS_MNT_UNBINDABLE(s)) {
1792 s = skip_mnt_tree(s);
1793 continue;
1794 }
1795 if (!(flag & CL_COPY_MNT_NS_FILE) &&
1796 is_mnt_ns_file(s->mnt.mnt_root)) {
1797 s = skip_mnt_tree(s);
1798 continue;
1799 }
1800 while (p != s->mnt_parent) {
1801 p = p->mnt_parent;
1802 q = q->mnt_parent;
1803 }
1804 p = s;
1805 parent = q;
1806 q = clone_mnt(p, p->mnt.mnt_root, flag);
1807 if (IS_ERR(q))
1808 goto out;
1809 lock_mount_hash();
1810 list_add_tail(&q->mnt_list, &res->mnt_list);
1811 attach_mnt(q, parent, p->mnt_mp);
1812 unlock_mount_hash();
1813 }
1814 }
1815 return res;
1816out:
1817 if (res) {
1818 lock_mount_hash();
1819 umount_tree(res, UMOUNT_SYNC);
1820 unlock_mount_hash();
1821 }
1822 return q;
1823}
1824
1825/* Caller should check returned pointer for errors */
1826
1827struct vfsmount *collect_mounts(const struct path *path)
1828{
1829 struct mount *tree;
1830 namespace_lock();
1831 if (!check_mnt(real_mount(path->mnt)))
1832 tree = ERR_PTR(-EINVAL);
1833 else
1834 tree = copy_tree(real_mount(path->mnt), path->dentry,
1835 CL_COPY_ALL | CL_PRIVATE);
1836 namespace_unlock();
1837 if (IS_ERR(tree))
1838 return ERR_CAST(tree);
1839 return &tree->mnt;
1840}
1841
1842void drop_collected_mounts(struct vfsmount *mnt)
1843{
1844 namespace_lock();
1845 lock_mount_hash();
1846 umount_tree(real_mount(mnt), UMOUNT_SYNC);
1847 unlock_mount_hash();
1848 namespace_unlock();
1849}
1850
1851/**
1852 * clone_private_mount - create a private clone of a path
1853 *
1854 * This creates a new vfsmount, which will be the clone of @path. The new will
1855 * not be attached anywhere in the namespace and will be private (i.e. changes
1856 * to the originating mount won't be propagated into this).
1857 *
1858 * Release with mntput().
1859 */
1860struct vfsmount *clone_private_mount(const struct path *path)
1861{
1862 struct mount *old_mnt = real_mount(path->mnt);
1863 struct mount *new_mnt;
1864
1865 if (IS_MNT_UNBINDABLE(old_mnt))
1866 return ERR_PTR(-EINVAL);
1867
1868 new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1869 if (IS_ERR(new_mnt))
1870 return ERR_CAST(new_mnt);
1871
1872 return &new_mnt->mnt;
1873}
1874EXPORT_SYMBOL_GPL(clone_private_mount);
1875
1876int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1877 struct vfsmount *root)
1878{
1879 struct mount *mnt;
1880 int res = f(root, arg);
1881 if (res)
1882 return res;
1883 list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1884 res = f(&mnt->mnt, arg);
1885 if (res)
1886 return res;
1887 }
1888 return 0;
1889}
1890
1891static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1892{
1893 struct mount *p;
1894
1895 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1896 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1897 mnt_release_group_id(p);
1898 }
1899}
1900
1901static int invent_group_ids(struct mount *mnt, bool recurse)
1902{
1903 struct mount *p;
1904
1905 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1906 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1907 int err = mnt_alloc_group_id(p);
1908 if (err) {
1909 cleanup_group_ids(mnt, p);
1910 return err;
1911 }
1912 }
1913 }
1914
1915 return 0;
1916}
1917
1918int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1919{
1920 unsigned int max = READ_ONCE(sysctl_mount_max);
1921 unsigned int mounts = 0, old, pending, sum;
1922 struct mount *p;
1923
1924 for (p = mnt; p; p = next_mnt(p, mnt))
1925 mounts++;
1926
1927 old = ns->mounts;
1928 pending = ns->pending_mounts;
1929 sum = old + pending;
1930 if ((old > sum) ||
1931 (pending > sum) ||
1932 (max < sum) ||
1933 (mounts > (max - sum)))
1934 return -ENOSPC;
1935
1936 ns->pending_mounts = pending + mounts;
1937 return 0;
1938}
1939
1940/*
1941 * @source_mnt : mount tree to be attached
1942 * @nd : place the mount tree @source_mnt is attached
1943 * @parent_nd : if non-null, detach the source_mnt from its parent and
1944 * store the parent mount and mountpoint dentry.
1945 * (done when source_mnt is moved)
1946 *
1947 * NOTE: in the table below explains the semantics when a source mount
1948 * of a given type is attached to a destination mount of a given type.
1949 * ---------------------------------------------------------------------------
1950 * | BIND MOUNT OPERATION |
1951 * |**************************************************************************
1952 * | source-->| shared | private | slave | unbindable |
1953 * | dest | | | | |
1954 * | | | | | | |
1955 * | v | | | | |
1956 * |**************************************************************************
1957 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1958 * | | | | | |
1959 * |non-shared| shared (+) | private | slave (*) | invalid |
1960 * ***************************************************************************
1961 * A bind operation clones the source mount and mounts the clone on the
1962 * destination mount.
1963 *
1964 * (++) the cloned mount is propagated to all the mounts in the propagation
1965 * tree of the destination mount and the cloned mount is added to
1966 * the peer group of the source mount.
1967 * (+) the cloned mount is created under the destination mount and is marked
1968 * as shared. The cloned mount is added to the peer group of the source
1969 * mount.
1970 * (+++) the mount is propagated to all the mounts in the propagation tree
1971 * of the destination mount and the cloned mount is made slave
1972 * of the same master as that of the source mount. The cloned mount
1973 * is marked as 'shared and slave'.
1974 * (*) the cloned mount is made a slave of the same master as that of the
1975 * source mount.
1976 *
1977 * ---------------------------------------------------------------------------
1978 * | MOVE MOUNT OPERATION |
1979 * |**************************************************************************
1980 * | source-->| shared | private | slave | unbindable |
1981 * | dest | | | | |
1982 * | | | | | | |
1983 * | v | | | | |
1984 * |**************************************************************************
1985 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1986 * | | | | | |
1987 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1988 * ***************************************************************************
1989 *
1990 * (+) the mount is moved to the destination. And is then propagated to
1991 * all the mounts in the propagation tree of the destination mount.
1992 * (+*) the mount is moved to the destination.
1993 * (+++) the mount is moved to the destination and is then propagated to
1994 * all the mounts belonging to the destination mount's propagation tree.
1995 * the mount is marked as 'shared and slave'.
1996 * (*) the mount continues to be a slave at the new location.
1997 *
1998 * if the source mount is a tree, the operations explained above is
1999 * applied to each mount in the tree.
2000 * Must be called without spinlocks held, since this function can sleep
2001 * in allocations.
2002 */
2003static int attach_recursive_mnt(struct mount *source_mnt,
2004 struct mount *dest_mnt,
2005 struct mountpoint *dest_mp,
2006 struct path *parent_path)
2007{
2008 HLIST_HEAD(tree_list);
2009 struct mnt_namespace *ns = dest_mnt->mnt_ns;
2010 struct mountpoint *smp;
2011 struct mount *child, *p;
2012 struct hlist_node *n;
2013 int err;
2014
2015 /* Preallocate a mountpoint in case the new mounts need
2016 * to be tucked under other mounts.
2017 */
2018 smp = get_mountpoint(source_mnt->mnt.mnt_root);
2019 if (IS_ERR(smp))
2020 return PTR_ERR(smp);
2021
2022 /* Is there space to add these mounts to the mount namespace? */
2023 if (!parent_path) {
2024 err = count_mounts(ns, source_mnt);
2025 if (err)
2026 goto out;
2027 }
2028
2029 if (IS_MNT_SHARED(dest_mnt)) {
2030 err = invent_group_ids(source_mnt, true);
2031 if (err)
2032 goto out;
2033 err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2034 lock_mount_hash();
2035 if (err)
2036 goto out_cleanup_ids;
2037 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2038 set_mnt_shared(p);
2039 } else {
2040 lock_mount_hash();
2041 }
2042 if (parent_path) {
2043 detach_mnt(source_mnt, parent_path);
2044 attach_mnt(source_mnt, dest_mnt, dest_mp);
2045 touch_mnt_namespace(source_mnt->mnt_ns);
2046 } else {
2047 mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2048 commit_tree(source_mnt);
2049 }
2050
2051 hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2052 struct mount *q;
2053 hlist_del_init(&child->mnt_hash);
2054 q = __lookup_mnt(&child->mnt_parent->mnt,
2055 child->mnt_mountpoint);
2056 if (q)
2057 mnt_change_mountpoint(child, smp, q);
2058 commit_tree(child);
2059 }
2060 put_mountpoint(smp);
2061 unlock_mount_hash();
2062
2063 return 0;
2064
2065 out_cleanup_ids:
2066 while (!hlist_empty(&tree_list)) {
2067 child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2068 child->mnt_parent->mnt_ns->pending_mounts = 0;
2069 umount_tree(child, UMOUNT_SYNC);
2070 }
2071 unlock_mount_hash();
2072 cleanup_group_ids(source_mnt, NULL);
2073 out:
2074 ns->pending_mounts = 0;
2075
2076 read_seqlock_excl(&mount_lock);
2077 put_mountpoint(smp);
2078 read_sequnlock_excl(&mount_lock);
2079
2080 return err;
2081}
2082
2083static struct mountpoint *lock_mount(struct path *path)
2084{
2085 struct vfsmount *mnt;
2086 struct dentry *dentry = path->dentry;
2087retry:
2088 inode_lock(dentry->d_inode);
2089 if (unlikely(cant_mount(dentry))) {
2090 inode_unlock(dentry->d_inode);
2091 return ERR_PTR(-ENOENT);
2092 }
2093 namespace_lock();
2094 mnt = lookup_mnt(path);
2095 if (likely(!mnt)) {
2096 struct mountpoint *mp = get_mountpoint(dentry);
2097 if (IS_ERR(mp)) {
2098 namespace_unlock();
2099 inode_unlock(dentry->d_inode);
2100 return mp;
2101 }
2102 return mp;
2103 }
2104 namespace_unlock();
2105 inode_unlock(path->dentry->d_inode);
2106 path_put(path);
2107 path->mnt = mnt;
2108 dentry = path->dentry = dget(mnt->mnt_root);
2109 goto retry;
2110}
2111
2112static void unlock_mount(struct mountpoint *where)
2113{
2114 struct dentry *dentry = where->m_dentry;
2115
2116 read_seqlock_excl(&mount_lock);
2117 put_mountpoint(where);
2118 read_sequnlock_excl(&mount_lock);
2119
2120 namespace_unlock();
2121 inode_unlock(dentry->d_inode);
2122}
2123
2124static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2125{
2126 if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2127 return -EINVAL;
2128
2129 if (d_is_dir(mp->m_dentry) !=
2130 d_is_dir(mnt->mnt.mnt_root))
2131 return -ENOTDIR;
2132
2133 return attach_recursive_mnt(mnt, p, mp, NULL);
2134}
2135
2136/*
2137 * Sanity check the flags to change_mnt_propagation.
2138 */
2139
2140static int flags_to_propagation_type(int ms_flags)
2141{
2142 int type = ms_flags & ~(MS_REC | MS_SILENT);
2143
2144 /* Fail if any non-propagation flags are set */
2145 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2146 return 0;
2147 /* Only one propagation flag should be set */
2148 if (!is_power_of_2(type))
2149 return 0;
2150 return type;
2151}
2152
2153/*
2154 * recursively change the type of the mountpoint.
2155 */
2156static int do_change_type(struct path *path, int ms_flags)
2157{
2158 struct mount *m;
2159 struct mount *mnt = real_mount(path->mnt);
2160 int recurse = ms_flags & MS_REC;
2161 int type;
2162 int err = 0;
2163
2164 if (path->dentry != path->mnt->mnt_root)
2165 return -EINVAL;
2166
2167 type = flags_to_propagation_type(ms_flags);
2168 if (!type)
2169 return -EINVAL;
2170
2171 namespace_lock();
2172 if (type == MS_SHARED) {
2173 err = invent_group_ids(mnt, recurse);
2174 if (err)
2175 goto out_unlock;
2176 }
2177
2178 lock_mount_hash();
2179 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2180 change_mnt_propagation(m, type);
2181 unlock_mount_hash();
2182
2183 out_unlock:
2184 namespace_unlock();
2185 return err;
2186}
2187
2188static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2189{
2190 struct mount *child;
2191 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2192 if (!is_subdir(child->mnt_mountpoint, dentry))
2193 continue;
2194
2195 if (child->mnt.mnt_flags & MNT_LOCKED)
2196 return true;
2197 }
2198 return false;
2199}
2200
2201/*
2202 * do loopback mount.
2203 */
2204static int do_loopback(struct path *path, const char *old_name,
2205 int recurse)
2206{
2207 struct path old_path;
2208 struct mount *mnt = NULL, *old, *parent;
2209 struct mountpoint *mp;
2210 int err;
2211 if (!old_name || !*old_name)
2212 return -EINVAL;
2213 err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2214 if (err)
2215 return err;
2216
2217 err = -EINVAL;
2218 if (mnt_ns_loop(old_path.dentry))
2219 goto out;
2220
2221 mp = lock_mount(path);
2222 err = PTR_ERR(mp);
2223 if (IS_ERR(mp))
2224 goto out;
2225
2226 old = real_mount(old_path.mnt);
2227 parent = real_mount(path->mnt);
2228
2229 err = -EINVAL;
2230 if (IS_MNT_UNBINDABLE(old))
2231 goto out2;
2232
2233 if (!check_mnt(parent))
2234 goto out2;
2235
2236 if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2237 goto out2;
2238
2239 if (!recurse && has_locked_children(old, old_path.dentry))
2240 goto out2;
2241
2242 if (recurse)
2243 mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2244 else
2245 mnt = clone_mnt(old, old_path.dentry, 0);
2246
2247 if (IS_ERR(mnt)) {
2248 err = PTR_ERR(mnt);
2249 goto out2;
2250 }
2251
2252 mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2253
2254 err = graft_tree(mnt, parent, mp);
2255 if (err) {
2256 lock_mount_hash();
2257 umount_tree(mnt, UMOUNT_SYNC);
2258 unlock_mount_hash();
2259 }
2260out2:
2261 unlock_mount(mp);
2262out:
2263 path_put(&old_path);
2264 return err;
2265}
2266
2267static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2268{
2269 int error = 0;
2270 int readonly_request = 0;
2271
2272 if (ms_flags & MS_RDONLY)
2273 readonly_request = 1;
2274 if (readonly_request == __mnt_is_readonly(mnt))
2275 return 0;
2276
2277 if (readonly_request)
2278 error = mnt_make_readonly(real_mount(mnt));
2279 else
2280 __mnt_unmake_readonly(real_mount(mnt));
2281 return error;
2282}
2283
2284/*
2285 * change filesystem flags. dir should be a physical root of filesystem.
2286 * If you've mounted a non-root directory somewhere and want to do remount
2287 * on it - tough luck.
2288 */
2289static int do_remount(struct path *path, int ms_flags, int sb_flags,
2290 int mnt_flags, void *data)
2291{
2292 int err;
2293 struct super_block *sb = path->mnt->mnt_sb;
2294 struct mount *mnt = real_mount(path->mnt);
2295
2296 if (!check_mnt(mnt))
2297 return -EINVAL;
2298
2299 if (path->dentry != path->mnt->mnt_root)
2300 return -EINVAL;
2301
2302 /* Don't allow changing of locked mnt flags.
2303 *
2304 * No locks need to be held here while testing the various
2305 * MNT_LOCK flags because those flags can never be cleared
2306 * once they are set.
2307 */
2308 if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2309 !(mnt_flags & MNT_READONLY)) {
2310 return -EPERM;
2311 }
2312 if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2313 !(mnt_flags & MNT_NODEV)) {
2314 return -EPERM;
2315 }
2316 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2317 !(mnt_flags & MNT_NOSUID)) {
2318 return -EPERM;
2319 }
2320 if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2321 !(mnt_flags & MNT_NOEXEC)) {
2322 return -EPERM;
2323 }
2324 if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2325 ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2326 return -EPERM;
2327 }
2328
2329 err = security_sb_remount(sb, data);
2330 if (err)
2331 return err;
2332
2333 down_write(&sb->s_umount);
2334 if (ms_flags & MS_BIND)
2335 err = change_mount_flags(path->mnt, ms_flags);
2336 else if (!capable(CAP_SYS_ADMIN))
2337 err = -EPERM;
2338 else
2339 err = do_remount_sb(sb, sb_flags, data, 0);
2340 if (!err) {
2341 lock_mount_hash();
2342 mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2343 mnt->mnt.mnt_flags = mnt_flags;
2344 touch_mnt_namespace(mnt->mnt_ns);
2345 unlock_mount_hash();
2346 }
2347 up_write(&sb->s_umount);
2348 return err;
2349}
2350
2351static inline int tree_contains_unbindable(struct mount *mnt)
2352{
2353 struct mount *p;
2354 for (p = mnt; p; p = next_mnt(p, mnt)) {
2355 if (IS_MNT_UNBINDABLE(p))
2356 return 1;
2357 }
2358 return 0;
2359}
2360
2361static int do_move_mount(struct path *path, const char *old_name)
2362{
2363 struct path old_path, parent_path;
2364 struct mount *p;
2365 struct mount *old;
2366 struct mountpoint *mp;
2367 int err;
2368 if (!old_name || !*old_name)
2369 return -EINVAL;
2370 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2371 if (err)
2372 return err;
2373
2374 mp = lock_mount(path);
2375 err = PTR_ERR(mp);
2376 if (IS_ERR(mp))
2377 goto out;
2378
2379 old = real_mount(old_path.mnt);
2380 p = real_mount(path->mnt);
2381
2382 err = -EINVAL;
2383 if (!check_mnt(p) || !check_mnt(old))
2384 goto out1;
2385
2386 if (old->mnt.mnt_flags & MNT_LOCKED)
2387 goto out1;
2388
2389 err = -EINVAL;
2390 if (old_path.dentry != old_path.mnt->mnt_root)
2391 goto out1;
2392
2393 if (!mnt_has_parent(old))
2394 goto out1;
2395
2396 if (d_is_dir(path->dentry) !=
2397 d_is_dir(old_path.dentry))
2398 goto out1;
2399 /*
2400 * Don't move a mount residing in a shared parent.
2401 */
2402 if (IS_MNT_SHARED(old->mnt_parent))
2403 goto out1;
2404 /*
2405 * Don't move a mount tree containing unbindable mounts to a destination
2406 * mount which is shared.
2407 */
2408 if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2409 goto out1;
2410 err = -ELOOP;
2411 for (; mnt_has_parent(p); p = p->mnt_parent)
2412 if (p == old)
2413 goto out1;
2414
2415 err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2416 if (err)
2417 goto out1;
2418
2419 /* if the mount is moved, it should no longer be expire
2420 * automatically */
2421 list_del_init(&old->mnt_expire);
2422out1:
2423 unlock_mount(mp);
2424out:
2425 if (!err)
2426 path_put(&parent_path);
2427 path_put(&old_path);
2428 return err;
2429}
2430
2431static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2432{
2433 int err;
2434 const char *subtype = strchr(fstype, '.');
2435 if (subtype) {
2436 subtype++;
2437 err = -EINVAL;
2438 if (!subtype[0])
2439 goto err;
2440 } else
2441 subtype = "";
2442
2443 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2444 err = -ENOMEM;
2445 if (!mnt->mnt_sb->s_subtype)
2446 goto err;
2447 return mnt;
2448
2449 err:
2450 mntput(mnt);
2451 return ERR_PTR(err);
2452}
2453
2454/*
2455 * add a mount into a namespace's mount tree
2456 */
2457static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2458{
2459 struct mountpoint *mp;
2460 struct mount *parent;
2461 int err;
2462
2463 mnt_flags &= ~MNT_INTERNAL_FLAGS;
2464
2465 mp = lock_mount(path);
2466 if (IS_ERR(mp))
2467 return PTR_ERR(mp);
2468
2469 parent = real_mount(path->mnt);
2470 err = -EINVAL;
2471 if (unlikely(!check_mnt(parent))) {
2472 /* that's acceptable only for automounts done in private ns */
2473 if (!(mnt_flags & MNT_SHRINKABLE))
2474 goto unlock;
2475 /* ... and for those we'd better have mountpoint still alive */
2476 if (!parent->mnt_ns)
2477 goto unlock;
2478 }
2479
2480 /* Refuse the same filesystem on the same mount point */
2481 err = -EBUSY;
2482 if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2483 path->mnt->mnt_root == path->dentry)
2484 goto unlock;
2485
2486 err = -EINVAL;
2487 if (d_is_symlink(newmnt->mnt.mnt_root))
2488 goto unlock;
2489
2490 newmnt->mnt.mnt_flags = mnt_flags;
2491 err = graft_tree(newmnt, parent, mp);
2492
2493unlock:
2494 unlock_mount(mp);
2495 return err;
2496}
2497
2498static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2499
2500/*
2501 * create a new mount for userspace and request it to be added into the
2502 * namespace's tree
2503 */
2504static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2505 int mnt_flags, const char *name, void *data)
2506{
2507 struct file_system_type *type;
2508 struct vfsmount *mnt;
2509 int err;
2510
2511 if (!fstype)
2512 return -EINVAL;
2513
2514 type = get_fs_type(fstype);
2515 if (!type)
2516 return -ENODEV;
2517
2518 mnt = vfs_kern_mount(type, sb_flags, name, data);
2519 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2520 !mnt->mnt_sb->s_subtype)
2521 mnt = fs_set_subtype(mnt, fstype);
2522
2523 put_filesystem(type);
2524 if (IS_ERR(mnt))
2525 return PTR_ERR(mnt);
2526
2527 if (mount_too_revealing(mnt, &mnt_flags)) {
2528 mntput(mnt);
2529 return -EPERM;
2530 }
2531
2532 err = do_add_mount(real_mount(mnt), path, mnt_flags);
2533 if (err)
2534 mntput(mnt);
2535 return err;
2536}
2537
2538int finish_automount(struct vfsmount *m, struct path *path)
2539{
2540 struct mount *mnt = real_mount(m);
2541 int err;
2542 /* The new mount record should have at least 2 refs to prevent it being
2543 * expired before we get a chance to add it
2544 */
2545 BUG_ON(mnt_get_count(mnt) < 2);
2546
2547 if (m->mnt_sb == path->mnt->mnt_sb &&
2548 m->mnt_root == path->dentry) {
2549 err = -ELOOP;
2550 goto fail;
2551 }
2552
2553 err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2554 if (!err)
2555 return 0;
2556fail:
2557 /* remove m from any expiration list it may be on */
2558 if (!list_empty(&mnt->mnt_expire)) {
2559 namespace_lock();
2560 list_del_init(&mnt->mnt_expire);
2561 namespace_unlock();
2562 }
2563 mntput(m);
2564 mntput(m);
2565 return err;
2566}
2567
2568/**
2569 * mnt_set_expiry - Put a mount on an expiration list
2570 * @mnt: The mount to list.
2571 * @expiry_list: The list to add the mount to.
2572 */
2573void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2574{
2575 namespace_lock();
2576
2577 list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2578
2579 namespace_unlock();
2580}
2581EXPORT_SYMBOL(mnt_set_expiry);
2582
2583/*
2584 * process a list of expirable mountpoints with the intent of discarding any
2585 * mountpoints that aren't in use and haven't been touched since last we came
2586 * here
2587 */
2588void mark_mounts_for_expiry(struct list_head *mounts)
2589{
2590 struct mount *mnt, *next;
2591 LIST_HEAD(graveyard);
2592
2593 if (list_empty(mounts))
2594 return;
2595
2596 namespace_lock();
2597 lock_mount_hash();
2598
2599 /* extract from the expiration list every vfsmount that matches the
2600 * following criteria:
2601 * - only referenced by its parent vfsmount
2602 * - still marked for expiry (marked on the last call here; marks are
2603 * cleared by mntput())
2604 */
2605 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2606 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2607 propagate_mount_busy(mnt, 1))
2608 continue;
2609 list_move(&mnt->mnt_expire, &graveyard);
2610 }
2611 while (!list_empty(&graveyard)) {
2612 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2613 touch_mnt_namespace(mnt->mnt_ns);
2614 umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2615 }
2616 unlock_mount_hash();
2617 namespace_unlock();
2618}
2619
2620EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2621
2622/*
2623 * Ripoff of 'select_parent()'
2624 *
2625 * search the list of submounts for a given mountpoint, and move any
2626 * shrinkable submounts to the 'graveyard' list.
2627 */
2628static int select_submounts(struct mount *parent, struct list_head *graveyard)
2629{
2630 struct mount *this_parent = parent;
2631 struct list_head *next;
2632 int found = 0;
2633
2634repeat:
2635 next = this_parent->mnt_mounts.next;
2636resume:
2637 while (next != &this_parent->mnt_mounts) {
2638 struct list_head *tmp = next;
2639 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2640
2641 next = tmp->next;
2642 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2643 continue;
2644 /*
2645 * Descend a level if the d_mounts list is non-empty.
2646 */
2647 if (!list_empty(&mnt->mnt_mounts)) {
2648 this_parent = mnt;
2649 goto repeat;
2650 }
2651
2652 if (!propagate_mount_busy(mnt, 1)) {
2653 list_move_tail(&mnt->mnt_expire, graveyard);
2654 found++;
2655 }
2656 }
2657 /*
2658 * All done at this level ... ascend and resume the search
2659 */
2660 if (this_parent != parent) {
2661 next = this_parent->mnt_child.next;
2662 this_parent = this_parent->mnt_parent;
2663 goto resume;
2664 }
2665 return found;
2666}
2667
2668/*
2669 * process a list of expirable mountpoints with the intent of discarding any
2670 * submounts of a specific parent mountpoint
2671 *
2672 * mount_lock must be held for write
2673 */
2674static void shrink_submounts(struct mount *mnt)
2675{
2676 LIST_HEAD(graveyard);
2677 struct mount *m;
2678
2679 /* extract submounts of 'mountpoint' from the expiration list */
2680 while (select_submounts(mnt, &graveyard)) {
2681 while (!list_empty(&graveyard)) {
2682 m = list_first_entry(&graveyard, struct mount,
2683 mnt_expire);
2684 touch_mnt_namespace(m->mnt_ns);
2685 umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2686 }
2687 }
2688}
2689
2690/*
2691 * Some copy_from_user() implementations do not return the exact number of
2692 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
2693 * Note that this function differs from copy_from_user() in that it will oops
2694 * on bad values of `to', rather than returning a short copy.
2695 */
2696static long exact_copy_from_user(void *to, const void __user * from,
2697 unsigned long n)
2698{
2699 char *t = to;
2700 const char __user *f = from;
2701 char c;
2702
2703 if (!access_ok(VERIFY_READ, from, n))
2704 return n;
2705
2706 while (n) {
2707 if (__get_user(c, f)) {
2708 memset(t, 0, n);
2709 break;
2710 }
2711 *t++ = c;
2712 f++;
2713 n--;
2714 }
2715 return n;
2716}
2717
2718void *copy_mount_options(const void __user * data)
2719{
2720 int i;
2721 unsigned long size;
2722 char *copy;
2723
2724 if (!data)
2725 return NULL;
2726
2727 copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2728 if (!copy)
2729 return ERR_PTR(-ENOMEM);
2730
2731 /* We only care that *some* data at the address the user
2732 * gave us is valid. Just in case, we'll zero
2733 * the remainder of the page.
2734 */
2735 /* copy_from_user cannot cross TASK_SIZE ! */
2736 size = TASK_SIZE - (unsigned long)data;
2737 if (size > PAGE_SIZE)
2738 size = PAGE_SIZE;
2739
2740 i = size - exact_copy_from_user(copy, data, size);
2741 if (!i) {
2742 kfree(copy);
2743 return ERR_PTR(-EFAULT);
2744 }
2745 if (i != PAGE_SIZE)
2746 memset(copy + i, 0, PAGE_SIZE - i);
2747 return copy;
2748}
2749
2750char *copy_mount_string(const void __user *data)
2751{
2752 return data ? strndup_user(data, PAGE_SIZE) : NULL;
2753}
2754
2755/*
2756 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2757 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2758 *
2759 * data is a (void *) that can point to any structure up to
2760 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2761 * information (or be NULL).
2762 *
2763 * Pre-0.97 versions of mount() didn't have a flags word.
2764 * When the flags word was introduced its top half was required
2765 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2766 * Therefore, if this magic number is present, it carries no information
2767 * and must be discarded.
2768 */
2769long do_mount(const char *dev_name, const char __user *dir_name,
2770 const char *type_page, unsigned long flags, void *data_page)
2771{
2772 struct path path;
2773 unsigned int mnt_flags = 0, sb_flags;
2774 int retval = 0;
2775
2776 /* Discard magic */
2777 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2778 flags &= ~MS_MGC_MSK;
2779
2780 /* Basic sanity checks */
2781 if (data_page)
2782 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2783
2784 if (flags & MS_NOUSER)
2785 return -EINVAL;
2786
2787 /* ... and get the mountpoint */
2788 retval = user_path(dir_name, &path);
2789 if (retval)
2790 return retval;
2791
2792 retval = security_sb_mount(dev_name, &path,
2793 type_page, flags, data_page);
2794 if (!retval && !may_mount())
2795 retval = -EPERM;
2796 if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
2797 retval = -EPERM;
2798 if (retval)
2799 goto dput_out;
2800
2801 /* Default to relatime unless overriden */
2802 if (!(flags & MS_NOATIME))
2803 mnt_flags |= MNT_RELATIME;
2804
2805 /* Separate the per-mountpoint flags */
2806 if (flags & MS_NOSUID)
2807 mnt_flags |= MNT_NOSUID;
2808 if (flags & MS_NODEV)
2809 mnt_flags |= MNT_NODEV;
2810 if (flags & MS_NOEXEC)
2811 mnt_flags |= MNT_NOEXEC;
2812 if (flags & MS_NOATIME)
2813 mnt_flags |= MNT_NOATIME;
2814 if (flags & MS_NODIRATIME)
2815 mnt_flags |= MNT_NODIRATIME;
2816 if (flags & MS_STRICTATIME)
2817 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2818 if (flags & MS_RDONLY)
2819 mnt_flags |= MNT_READONLY;
2820
2821 /* The default atime for remount is preservation */
2822 if ((flags & MS_REMOUNT) &&
2823 ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2824 MS_STRICTATIME)) == 0)) {
2825 mnt_flags &= ~MNT_ATIME_MASK;
2826 mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2827 }
2828
2829 sb_flags = flags & (SB_RDONLY |
2830 SB_SYNCHRONOUS |
2831 SB_MANDLOCK |
2832 SB_DIRSYNC |
2833 SB_SILENT |
2834 SB_POSIXACL |
2835 SB_LAZYTIME |
2836 SB_I_VERSION);
2837
2838 if (flags & MS_REMOUNT)
2839 retval = do_remount(&path, flags, sb_flags, mnt_flags,
2840 data_page);
2841 else if (flags & MS_BIND)
2842 retval = do_loopback(&path, dev_name, flags & MS_REC);
2843 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2844 retval = do_change_type(&path, flags);
2845 else if (flags & MS_MOVE)
2846 retval = do_move_mount(&path, dev_name);
2847 else
2848 retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2849 dev_name, data_page);
2850dput_out:
2851 path_put(&path);
2852 return retval;
2853}
2854
2855static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2856{
2857 return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2858}
2859
2860static void dec_mnt_namespaces(struct ucounts *ucounts)
2861{
2862 dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2863}
2864
2865static void free_mnt_ns(struct mnt_namespace *ns)
2866{
2867 ns_free_inum(&ns->ns);
2868 dec_mnt_namespaces(ns->ucounts);
2869 put_user_ns(ns->user_ns);
2870 kfree(ns);
2871}
2872
2873/*
2874 * Assign a sequence number so we can detect when we attempt to bind
2875 * mount a reference to an older mount namespace into the current
2876 * mount namespace, preventing reference counting loops. A 64bit
2877 * number incrementing at 10Ghz will take 12,427 years to wrap which
2878 * is effectively never, so we can ignore the possibility.
2879 */
2880static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2881
2882static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2883{
2884 struct mnt_namespace *new_ns;
2885 struct ucounts *ucounts;
2886 int ret;
2887
2888 ucounts = inc_mnt_namespaces(user_ns);
2889 if (!ucounts)
2890 return ERR_PTR(-ENOSPC);
2891
2892 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2893 if (!new_ns) {
2894 dec_mnt_namespaces(ucounts);
2895 return ERR_PTR(-ENOMEM);
2896 }
2897 ret = ns_alloc_inum(&new_ns->ns);
2898 if (ret) {
2899 kfree(new_ns);
2900 dec_mnt_namespaces(ucounts);
2901 return ERR_PTR(ret);
2902 }
2903 new_ns->ns.ops = &mntns_operations;
2904 new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2905 atomic_set(&new_ns->count, 1);
2906 new_ns->root = NULL;
2907 INIT_LIST_HEAD(&new_ns->list);
2908 init_waitqueue_head(&new_ns->poll);
2909 new_ns->event = 0;
2910 new_ns->user_ns = get_user_ns(user_ns);
2911 new_ns->ucounts = ucounts;
2912 new_ns->mounts = 0;
2913 new_ns->pending_mounts = 0;
2914 return new_ns;
2915}
2916
2917__latent_entropy
2918struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2919 struct user_namespace *user_ns, struct fs_struct *new_fs)
2920{
2921 struct mnt_namespace *new_ns;
2922 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2923 struct mount *p, *q;
2924 struct mount *old;
2925 struct mount *new;
2926 int copy_flags;
2927
2928 BUG_ON(!ns);
2929
2930 if (likely(!(flags & CLONE_NEWNS))) {
2931 get_mnt_ns(ns);
2932 return ns;
2933 }
2934
2935 old = ns->root;
2936
2937 new_ns = alloc_mnt_ns(user_ns);
2938 if (IS_ERR(new_ns))
2939 return new_ns;
2940
2941 namespace_lock();
2942 /* First pass: copy the tree topology */
2943 copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2944 if (user_ns != ns->user_ns)
2945 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2946 new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2947 if (IS_ERR(new)) {
2948 namespace_unlock();
2949 free_mnt_ns(new_ns);
2950 return ERR_CAST(new);
2951 }
2952 new_ns->root = new;
2953 list_add_tail(&new_ns->list, &new->mnt_list);
2954
2955 /*
2956 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2957 * as belonging to new namespace. We have already acquired a private
2958 * fs_struct, so tsk->fs->lock is not needed.
2959 */
2960 p = old;
2961 q = new;
2962 while (p) {
2963 q->mnt_ns = new_ns;
2964 new_ns->mounts++;
2965 if (new_fs) {
2966 if (&p->mnt == new_fs->root.mnt) {
2967 new_fs->root.mnt = mntget(&q->mnt);
2968 rootmnt = &p->mnt;
2969 }
2970 if (&p->mnt == new_fs->pwd.mnt) {
2971 new_fs->pwd.mnt = mntget(&q->mnt);
2972 pwdmnt = &p->mnt;
2973 }
2974 }
2975 p = next_mnt(p, old);
2976 q = next_mnt(q, new);
2977 if (!q)
2978 break;
2979 while (p->mnt.mnt_root != q->mnt.mnt_root)
2980 p = next_mnt(p, old);
2981 }
2982 namespace_unlock();
2983
2984 if (rootmnt)
2985 mntput(rootmnt);
2986 if (pwdmnt)
2987 mntput(pwdmnt);
2988
2989 return new_ns;
2990}
2991
2992/**
2993 * create_mnt_ns - creates a private namespace and adds a root filesystem
2994 * @mnt: pointer to the new root filesystem mountpoint
2995 */
2996static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2997{
2998 struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2999 if (!IS_ERR(new_ns)) {
3000 struct mount *mnt = real_mount(m);
3001 mnt->mnt_ns = new_ns;
3002 new_ns->root = mnt;
3003 new_ns->mounts++;
3004 list_add(&mnt->mnt_list, &new_ns->list);
3005 } else {
3006 mntput(m);
3007 }
3008 return new_ns;
3009}
3010
3011struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3012{
3013 struct mnt_namespace *ns;
3014 struct super_block *s;
3015 struct path path;
3016 int err;
3017
3018 ns = create_mnt_ns(mnt);
3019 if (IS_ERR(ns))
3020 return ERR_CAST(ns);
3021
3022 err = vfs_path_lookup(mnt->mnt_root, mnt,
3023 name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3024
3025 put_mnt_ns(ns);
3026
3027 if (err)
3028 return ERR_PTR(err);
3029
3030 /* trade a vfsmount reference for active sb one */
3031 s = path.mnt->mnt_sb;
3032 atomic_inc(&s->s_active);
3033 mntput(path.mnt);
3034 /* lock the sucker */
3035 down_write(&s->s_umount);
3036 /* ... and return the root of (sub)tree on it */
3037 return path.dentry;
3038}
3039EXPORT_SYMBOL(mount_subtree);
3040
3041int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type,
3042 unsigned long flags, void __user *data)
3043{
3044 int ret;
3045 char *kernel_type;
3046 char *kernel_dev;
3047 void *options;
3048
3049 kernel_type = copy_mount_string(type);
3050 ret = PTR_ERR(kernel_type);
3051 if (IS_ERR(kernel_type))
3052 goto out_type;
3053
3054 kernel_dev = copy_mount_string(dev_name);
3055 ret = PTR_ERR(kernel_dev);
3056 if (IS_ERR(kernel_dev))
3057 goto out_dev;
3058
3059 options = copy_mount_options(data);
3060 ret = PTR_ERR(options);
3061 if (IS_ERR(options))
3062 goto out_data;
3063
3064 ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3065
3066 kfree(options);
3067out_data:
3068 kfree(kernel_dev);
3069out_dev:
3070 kfree(kernel_type);
3071out_type:
3072 return ret;
3073}
3074
3075SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3076 char __user *, type, unsigned long, flags, void __user *, data)
3077{
3078 return ksys_mount(dev_name, dir_name, type, flags, data);
3079}
3080
3081/*
3082 * Return true if path is reachable from root
3083 *
3084 * namespace_sem or mount_lock is held
3085 */
3086bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3087 const struct path *root)
3088{
3089 while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3090 dentry = mnt->mnt_mountpoint;
3091 mnt = mnt->mnt_parent;
3092 }
3093 return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3094}
3095
3096bool path_is_under(const struct path *path1, const struct path *path2)
3097{
3098 bool res;
3099 read_seqlock_excl(&mount_lock);
3100 res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3101 read_sequnlock_excl(&mount_lock);
3102 return res;
3103}
3104EXPORT_SYMBOL(path_is_under);
3105
3106/*
3107 * pivot_root Semantics:
3108 * Moves the root file system of the current process to the directory put_old,
3109 * makes new_root as the new root file system of the current process, and sets
3110 * root/cwd of all processes which had them on the current root to new_root.
3111 *
3112 * Restrictions:
3113 * The new_root and put_old must be directories, and must not be on the
3114 * same file system as the current process root. The put_old must be
3115 * underneath new_root, i.e. adding a non-zero number of /.. to the string
3116 * pointed to by put_old must yield the same directory as new_root. No other
3117 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3118 *
3119 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3120 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3121 * in this situation.
3122 *
3123 * Notes:
3124 * - we don't move root/cwd if they are not at the root (reason: if something
3125 * cared enough to change them, it's probably wrong to force them elsewhere)
3126 * - it's okay to pick a root that isn't the root of a file system, e.g.
3127 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3128 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3129 * first.
3130 */
3131SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3132 const char __user *, put_old)
3133{
3134 struct path new, old, parent_path, root_parent, root;
3135 struct mount *new_mnt, *root_mnt, *old_mnt;
3136 struct mountpoint *old_mp, *root_mp;
3137 int error;
3138
3139 if (!may_mount())
3140 return -EPERM;
3141
3142 error = user_path_dir(new_root, &new);
3143 if (error)
3144 goto out0;
3145
3146 error = user_path_dir(put_old, &old);
3147 if (error)
3148 goto out1;
3149
3150 error = security_sb_pivotroot(&old, &new);
3151 if (error)
3152 goto out2;
3153
3154 get_fs_root(current->fs, &root);
3155 old_mp = lock_mount(&old);
3156 error = PTR_ERR(old_mp);
3157 if (IS_ERR(old_mp))
3158 goto out3;
3159
3160 error = -EINVAL;
3161 new_mnt = real_mount(new.mnt);
3162 root_mnt = real_mount(root.mnt);
3163 old_mnt = real_mount(old.mnt);
3164 if (IS_MNT_SHARED(old_mnt) ||
3165 IS_MNT_SHARED(new_mnt->mnt_parent) ||
3166 IS_MNT_SHARED(root_mnt->mnt_parent))
3167 goto out4;
3168 if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3169 goto out4;
3170 if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3171 goto out4;
3172 error = -ENOENT;
3173 if (d_unlinked(new.dentry))
3174 goto out4;
3175 error = -EBUSY;
3176 if (new_mnt == root_mnt || old_mnt == root_mnt)
3177 goto out4; /* loop, on the same file system */
3178 error = -EINVAL;
3179 if (root.mnt->mnt_root != root.dentry)
3180 goto out4; /* not a mountpoint */
3181 if (!mnt_has_parent(root_mnt))
3182 goto out4; /* not attached */
3183 root_mp = root_mnt->mnt_mp;
3184 if (new.mnt->mnt_root != new.dentry)
3185 goto out4; /* not a mountpoint */
3186 if (!mnt_has_parent(new_mnt))
3187 goto out4; /* not attached */
3188 /* make sure we can reach put_old from new_root */
3189 if (!is_path_reachable(old_mnt, old.dentry, &new))
3190 goto out4;
3191 /* make certain new is below the root */
3192 if (!is_path_reachable(new_mnt, new.dentry, &root))
3193 goto out4;
3194 root_mp->m_count++; /* pin it so it won't go away */
3195 lock_mount_hash();
3196 detach_mnt(new_mnt, &parent_path);
3197 detach_mnt(root_mnt, &root_parent);
3198 if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3199 new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3200 root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3201 }
3202 /* mount old root on put_old */
3203 attach_mnt(root_mnt, old_mnt, old_mp);
3204 /* mount new_root on / */
3205 attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3206 touch_mnt_namespace(current->nsproxy->mnt_ns);
3207 /* A moved mount should not expire automatically */
3208 list_del_init(&new_mnt->mnt_expire);
3209 put_mountpoint(root_mp);
3210 unlock_mount_hash();
3211 chroot_fs_refs(&root, &new);
3212 error = 0;
3213out4:
3214 unlock_mount(old_mp);
3215 if (!error) {
3216 path_put(&root_parent);
3217 path_put(&parent_path);
3218 }
3219out3:
3220 path_put(&root);
3221out2:
3222 path_put(&old);
3223out1:
3224 path_put(&new);
3225out0:
3226 return error;
3227}
3228
3229static void __init init_mount_tree(void)
3230{
3231 struct vfsmount *mnt;
3232 struct mnt_namespace *ns;
3233 struct path root;
3234 struct file_system_type *type;
3235
3236 type = get_fs_type("rootfs");
3237 if (!type)
3238 panic("Can't find rootfs type");
3239 mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3240 put_filesystem(type);
3241 if (IS_ERR(mnt))
3242 panic("Can't create rootfs");
3243
3244 ns = create_mnt_ns(mnt);
3245 if (IS_ERR(ns))
3246 panic("Can't allocate initial namespace");
3247
3248 init_task.nsproxy->mnt_ns = ns;
3249 get_mnt_ns(ns);
3250
3251 root.mnt = mnt;
3252 root.dentry = mnt->mnt_root;
3253 mnt->mnt_flags |= MNT_LOCKED;
3254
3255 set_fs_pwd(current->fs, &root);
3256 set_fs_root(current->fs, &root);
3257}
3258
3259void __init mnt_init(void)
3260{
3261 int err;
3262
3263 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3264 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3265
3266 mount_hashtable = alloc_large_system_hash("Mount-cache",
3267 sizeof(struct hlist_head),
3268 mhash_entries, 19,
3269 HASH_ZERO,
3270 &m_hash_shift, &m_hash_mask, 0, 0);
3271 mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3272 sizeof(struct hlist_head),
3273 mphash_entries, 19,
3274 HASH_ZERO,
3275 &mp_hash_shift, &mp_hash_mask, 0, 0);
3276
3277 if (!mount_hashtable || !mountpoint_hashtable)
3278 panic("Failed to allocate mount hash table\n");
3279
3280 kernfs_init();
3281
3282 err = sysfs_init();
3283 if (err)
3284 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3285 __func__, err);
3286 fs_kobj = kobject_create_and_add("fs", NULL);
3287 if (!fs_kobj)
3288 printk(KERN_WARNING "%s: kobj create error\n", __func__);
3289 init_rootfs();
3290 init_mount_tree();
3291}
3292
3293void put_mnt_ns(struct mnt_namespace *ns)
3294{
3295 if (!atomic_dec_and_test(&ns->count))
3296 return;
3297 drop_collected_mounts(&ns->root->mnt);
3298 free_mnt_ns(ns);
3299}
3300
3301struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3302{
3303 struct vfsmount *mnt;
3304 mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data);
3305 if (!IS_ERR(mnt)) {
3306 /*
3307 * it is a longterm mount, don't release mnt until
3308 * we unmount before file sys is unregistered
3309 */
3310 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3311 }
3312 return mnt;
3313}
3314EXPORT_SYMBOL_GPL(kern_mount_data);
3315
3316void kern_unmount(struct vfsmount *mnt)
3317{
3318 /* release long term mount so mount point can be released */
3319 if (!IS_ERR_OR_NULL(mnt)) {
3320 real_mount(mnt)->mnt_ns = NULL;
3321 synchronize_rcu(); /* yecchhh... */
3322 mntput(mnt);
3323 }
3324}
3325EXPORT_SYMBOL(kern_unmount);
3326
3327bool our_mnt(struct vfsmount *mnt)
3328{
3329 return check_mnt(real_mount(mnt));
3330}
3331
3332bool current_chrooted(void)
3333{
3334 /* Does the current process have a non-standard root */
3335 struct path ns_root;
3336 struct path fs_root;
3337 bool chrooted;
3338
3339 /* Find the namespace root */
3340 ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt;
3341 ns_root.dentry = ns_root.mnt->mnt_root;
3342 path_get(&ns_root);
3343 while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3344 ;
3345
3346 get_fs_root(current->fs, &fs_root);
3347
3348 chrooted = !path_equal(&fs_root, &ns_root);
3349
3350 path_put(&fs_root);
3351 path_put(&ns_root);
3352
3353 return chrooted;
3354}
3355
3356static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3357 int *new_mnt_flags)
3358{
3359 int new_flags = *new_mnt_flags;
3360 struct mount *mnt;
3361 bool visible = false;
3362
3363 down_read(&namespace_sem);
3364 list_for_each_entry(mnt, &ns->list, mnt_list) {
3365 struct mount *child;
3366 int mnt_flags;
3367
3368 if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3369 continue;
3370
3371 /* This mount is not fully visible if it's root directory
3372 * is not the root directory of the filesystem.
3373 */
3374 if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3375 continue;
3376
3377 /* A local view of the mount flags */
3378 mnt_flags = mnt->mnt.mnt_flags;
3379
3380 /* Don't miss readonly hidden in the superblock flags */
3381 if (sb_rdonly(mnt->mnt.mnt_sb))
3382 mnt_flags |= MNT_LOCK_READONLY;
3383
3384 /* Verify the mount flags are equal to or more permissive
3385 * than the proposed new mount.
3386 */
3387 if ((mnt_flags & MNT_LOCK_READONLY) &&
3388 !(new_flags & MNT_READONLY))
3389 continue;
3390 if ((mnt_flags & MNT_LOCK_ATIME) &&
3391 ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3392 continue;
3393
3394 /* This mount is not fully visible if there are any
3395 * locked child mounts that cover anything except for
3396 * empty directories.
3397 */
3398 list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3399 struct inode *inode = child->mnt_mountpoint->d_inode;
3400 /* Only worry about locked mounts */
3401 if (!(child->mnt.mnt_flags & MNT_LOCKED))
3402 continue;
3403 /* Is the directory permanetly empty? */
3404 if (!is_empty_dir_inode(inode))
3405 goto next;
3406 }
3407 /* Preserve the locked attributes */
3408 *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3409 MNT_LOCK_ATIME);
3410 visible = true;
3411 goto found;
3412 next: ;
3413 }
3414found:
3415 up_read(&namespace_sem);
3416 return visible;
3417}
3418
3419static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3420{
3421 const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3422 struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3423 unsigned long s_iflags;
3424
3425 if (ns->user_ns == &init_user_ns)
3426 return false;
3427
3428 /* Can this filesystem be too revealing? */
3429 s_iflags = mnt->mnt_sb->s_iflags;
3430 if (!(s_iflags & SB_I_USERNS_VISIBLE))
3431 return false;
3432
3433 if ((s_iflags & required_iflags) != required_iflags) {
3434 WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3435 required_iflags);
3436 return true;
3437 }
3438
3439 return !mnt_already_visible(ns, mnt, new_mnt_flags);
3440}
3441
3442bool mnt_may_suid(struct vfsmount *mnt)
3443{
3444 /*
3445 * Foreign mounts (accessed via fchdir or through /proc
3446 * symlinks) are always treated as if they are nosuid. This
3447 * prevents namespaces from trusting potentially unsafe
3448 * suid/sgid bits, file caps, or security labels that originate
3449 * in other namespaces.
3450 */
3451 return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3452 current_in_userns(mnt->mnt_sb->s_user_ns);
3453}
3454
3455static struct ns_common *mntns_get(struct task_struct *task)
3456{
3457 struct ns_common *ns = NULL;
3458 struct nsproxy *nsproxy;
3459
3460 task_lock(task);
3461 nsproxy = task->nsproxy;
3462 if (nsproxy) {
3463 ns = &nsproxy->mnt_ns->ns;
3464 get_mnt_ns(to_mnt_ns(ns));
3465 }
3466 task_unlock(task);
3467
3468 return ns;
3469}
3470
3471static void mntns_put(struct ns_common *ns)
3472{
3473 put_mnt_ns(to_mnt_ns(ns));
3474}
3475
3476static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3477{
3478 struct fs_struct *fs = current->fs;
3479 struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3480 struct path root;
3481 int err;
3482
3483 if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3484 !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3485 !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3486 return -EPERM;
3487
3488 if (fs->users != 1)
3489 return -EINVAL;
3490
3491 get_mnt_ns(mnt_ns);
3492 old_mnt_ns = nsproxy->mnt_ns;
3493 nsproxy->mnt_ns = mnt_ns;
3494
3495 /* Find the root */
3496 err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3497 "/", LOOKUP_DOWN, &root);
3498 if (err) {
3499 /* revert to old namespace */
3500 nsproxy->mnt_ns = old_mnt_ns;
3501 put_mnt_ns(mnt_ns);
3502 return err;
3503 }
3504
3505 put_mnt_ns(old_mnt_ns);
3506
3507 /* Update the pwd and root */
3508 set_fs_pwd(fs, &root);
3509 set_fs_root(fs, &root);
3510
3511 path_put(&root);
3512 return 0;
3513}
3514
3515static struct user_namespace *mntns_owner(struct ns_common *ns)
3516{
3517 return to_mnt_ns(ns)->user_ns;
3518}
3519
3520const struct proc_ns_operations mntns_operations = {
3521 .name = "mnt",
3522 .type = CLONE_NEWNS,
3523 .get = mntns_get,
3524 .put = mntns_put,
3525 .install = mntns_install,
3526 .owner = mntns_owner,
3527};